Nucl - Phys.B v.594

Nuclear Physics B 594 (2001) 322
www.elsevier.nl/locate/npe
The b s decay in SUSY models

with non-universal A-terms
E. Gabrielli a, , S. Khalil a,b , E. Torrente-Lujan a
a Departmento de Fsica Terica C-XI, Universidad Autnoma de Madrid, Cantoblanco,
28049 Madrid, Spain

b Ain Shams University, Faculty of Science, Cairo 11566, Egypt
Received 16 June 2000; accepted 28 September 2000
Abstract
We analyse the predictions for the inclusive branching ratio for B Xs in a class of stringinspired SUSY models with non-universal soft-breaking A-terms. These models are particularly
interesting since the non-universality of the A-terms plays an important role in providing new
significant contributions to CP violation effects in kaon physics while respecting the severe bounds
on electric dipole moments. We show that b s do not severely constrain the non-universality of
these models. In particular, at low tan (tan ' 2), our predictions are close to the universal case.
For large tan (tan & 15), the effect of non-universality is enhanced and stronger constraints hold.
We find that the parameter regions which are important for generating sizeable contribution to 0 /,
of order 2 103 , are not excluded. 2001 Elsevier Science B.V. All rights reserved.
PACS: 13.25.Hw; 12.60.Jv
1. Introduction
Recently there has been a growing interest concerning supersymmetric models with
non-universal soft-breaking terms [17]. The main theoretical reason is that some stringinspired models naturally favour SUSY models with non-universality in the soft-breaking
sector [810,12]. Within this class of string-inspired models particularly interesting are
that ones with non-universal A-terms [5,6]. This is mainly due to the relevant role they can
play in solving the SUSY CP problem, especially in the light of the recent experimental
results on the direct CP violation parameter 0 /.
* Corresponding author.
E-mail addresses: emidio.gabrielli@cern.ch (E. Gabrielli), shaaban.khalil@uam.es (S. Khalil),

e.torrente@cern.ch (E. Torrente-Lujan).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 9 7 - 6
E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322
The new measurements of 0 / at KTeV [13] and NA48 [14] lead to a world average of
Re 0 / = (21.4 4.0) 104 [15]. This result is higher than the Standard Model (SM)
predictions [1620], opening the way to the interpretation that it may be a signal of new
physics beyond the SM. Clearly it is still premature to claim that this is a genuine new
physics effect, mainly due to the large theoretical uncertainties in the non-perturbative
hadronic sector which affect the SM predictions for 0 / [2124]. However if one accepts
the point of view that new CP sources are needed in order to obtain large values for 0 /,
one may wonder if the minimal supersymmetric extension of the SM (MSSM) can help in
solving this problem. The answer is no, mainly due to the assumption of universal boundary
conditions of the soft-breaking terms [5,2529]. Moreover we stress that without new
flavor structure beyond the usual Yukawa couplings, general SUSY models with phases
of the soft terms of order O(1) (but with a vanishing CKM phase CKM = 0) can not give
a sizeable contribution to the CP violating processes [4,5,2830].
The impact of the new flavor structure in the non-universality of the A-terms have
been studied in Refs. [17]. In these works it was emphasized that the non-degenerate
A-terms can generate the experimentally observed CP violation and 0 / even with a
vanishing CKM , i.e., fully supersymmetric CP violation in the kaon system is possible
in a class of models with non-universal A-terms. This effect can be simply understood by
making use of the mass-insertion approximation [31]: the non-degenerate A-terms enhance
the gluino contributions to and 0 / through enhancing the imaginary parts of the LR
d
d
)12 and Im(RL
)12 [5].
mass insertions Im(LR
It is well known that the experimental b s constraints cause a dramatic reduction
of the allowed parameter space in case of universal soft terms [3247]. Hence one may
worry if these constraints would be even more severe in the case of the non-degenerate
A-terms. However a complete analysis of the b s constraints has not been considered
in Refs. [57]. In particular in these works it was roughly checked, by using the results
d ) 6 1.6 102
of Ref. [31], that the b s constraints are satisfied, namely (LR
23
d
and (LL )23 6 8.2. Note that these constraints are obtained by assuming that the gluino
amplitude is the dominant contribution to b s (dominant also with respect to the SM).
Although the gluino contribution to b s decay is usually very small in the
d ) , we could expect that this
universal case, being proportional to the mass insertions (LR
23
contribution may be enhanced by the non-universality of the A-terms. However the nonuniversal A-terms could also give large contributions to the chargino amplitude through
u
)23 . Even though the non-degenerate A-terms enhance the gluino contribution,
the (LR
one can not a priori expect that it will be the dominant effect. Therefore a careful analysis
of the b s predictions, including the full SUSY contributions, is necessary in this
scenario.
The purpose of this work is to perform a complete analysis of the b s constraints
for the SUSY models with non-universal A-terms studied in Refs. [57]. Indeed we will
show that our results are different from the naive expectations based only on the gluinodominance approximation.
The paper is organized as follows. In Section 2 we present two models with nonuniversal A-terms that have recently been considered for solving the SUSY CP problem.
These two models are based on weakly coupled heterotic string and type I string theories,
respectively. In Section 3 we present formulas for the total branching ratio taking into
account the QCD corrections and we discuss the different SUSY contributions to the
b s decay in these models. Our numerical results for the predictions of the b
s branching ratio are presented in Section 4. The last section is devoted to conclusions.
Finally, various formulas are summarized in Appendix A.
2. String inspired models with non-degenerate A-terms

In this work we consider the class of string inspired model which has been recently
studied in Refs. [57]. In this class of models, the trilinear A-terms of the soft SUSY
breaking are non-universal. It was shown that this non-universality among the A-terms
plays an important role on CP violating processes. In particular, it has been shown that
non-degenerate A-parameters can generate the experimentally observed CP violation
and 0 / even with a vanishing CKM .
Here we consider two models for non-degenerate A-terms. The first model (model A)
is based on weakly coupled heterotic strings, where the dilaton and the moduli fields
contribute to SUSY breaking [10,11]. The second model (model B) is based on type I
string theory where the gauge group SU(3) U (1)Y is originated from the 9 brane and the
gauge group SU(2) is originated from one of the 5 branes [12].
In order to fix the conventions, the following Lagrangian LSB for the soft-breaking terms
is assumed
1
1
1
1
j
(1)
LSB = hij k i j k + (B)ij i j + (m2 )i ?i j + Ma a a + h.c.,
6
2
2
2
where the i are the scalar parts of the chiral superfields i and a are the gaugino fields. In
the notation for the trilinear couplings, the A-terms are defined as hij k = Yij k Aij k (indices
not summed) where Yij k are the corresponding Yukawa couplings.
2.1. Model A
We start with the weakly coupled string-inspired supergravity theory. In this class of
models, it is assumed that the superpotential of the dilaton (S) and moduli (T ) fields is
generated by some non-perturbative mechanism and the F -terms of S and T contribute to
the SUSY breaking. Then one can parametrize the F -terms as [10]
F T = m3/2(T + T ) cos .
(2)
F S = 3 m3/2(S + S ) sin ,
Here m3/2 is the gravitino mass, ni is the modular weight and tan corresponds to the
ratio between the F -terms of S and T . In this framework, the soft scalar masses mi and the
gaugino masses Ma are given by [10]
m2i = m23/2 (1 + ni cos2 ),
Ma = 3 m3/2 sin .
The
Au,d -terms
are written as
(3)
(4)
(Au,d )ij = 3 m3/2 sin m3/2 cos (3 + ni + nj + nHu,d ),
(5)
where ni,j,k are the modular weights of the fields that are coupled by this A-term. As
shown in Eqs. (3)(5), the values of the soft SUSY breaking parameters depend on the
modular weight of the matter fields. These modular weights ni are negative integers, and
their natural values (in case of ZN orbifolds) are 1, 2, and 3 [8]. If we assign ni =
1 for the third family and ni = 2 for the first and second families (we also assume
that nH1 = 1 and nH2 = 2). Note that with this choice of modular weights we have
m2H2 < m2H1 which is favored for the electroweak breaking (EW) and all the squark mass
matrices are equal. Also we find the following texture for the A-parameter matrix at the
string scale
xu,d xu,d yu,d

(6)
Au,d = xu,d xu,d yu,d ,
yu,d yu,d zu,d
where

xu = m3/2 3 sin + 3 cos ,

xd = yu = m3/2 3 sin + 2 cos ,

yd = zu = m3/2 3 sin + cos ,
zd = 3 m3/2 sin .
(7)
(8)
(9)
(10)
By fixing the value of tan we can determine the values of and B from the radiatively
EW breaking conditions. Then all the SUSY particle spectrum is completely determined
in terms of m3/2 and . The non-universality of this model is parameterized by the angle
and the value = /2 corresponds to the universal limit for the soft terms. In order to
avoid negative mass squared in the scalar masses we restrict ourselves to the case with
cos2 < 1/2. Such restriction on makes the non-universality in the whole soft SUSY
breaking terms very limited. However, as shown in [5,6], this small range of variation for
the non-universality is enough to generate sizeable SUSY CP violations in K system.
We emphasize that choosing modular weights different from those assigned above, the
allowed range of the soft SUSY breaking terms, which are relevant for the b s decay,
do not essentially change. This can be easily understood as follows. The magnitude of the
non universality between the relevant entries of the A-terms to the b s process can be
parametrized by the difference yu,d xu,d = m3/2 (n3 n2 ) cos , where n3 and n2 are the
modular weights of the third family and the first/second families respectively. Therefore,
it appears that increasing the values of the modular weight of the third generation/or of
the first generation (since we are interested in the relative magnitude) it may increase this
difference. However, the allowed values of the angle become more restricted, hence the
range of the relevant non-universality is slightly effected. For instance, if we assign ni =
3 for the first family instead of ni = 2 we find that the angle is more restricted than
before (cos2 < 1/3 instead of 1/2, respectively), which roughly gives the same allowed
range for the A-terms.
2.2. Model B
As mentioned in the introduction, this model is based on type I string theory. Like
model A, this is a good candidate for generating sizeable SUSY CP violations. Recently,
there has been considerable interest in studying the phenomenological implications of this
class of models [4855]. In type I string theory, non-universality in the scalar masses, Aterms and gaugino masses can be naturally obtained [12]. Type I models contain either
9 branes and three types of 5i (i = 1, 2, 3) branes or 7i branes and 3 branes. From the
phenomenological point of view there is no difference between these two scenarios. Here
we consider the same model used in Ref. [7], where the gauge group SU(3)C U (1)Y is
associated with 9 brane while SU(2)L is associated with 51 brane.
If SUSY breaking is analysed, as in model A, in terms of the vevs of the dilaton and
moduli fields [12]

F Ti = m3/2 Ti + Ti i cos ,
(11)
F S = 3 m3/2(S + S ) sin ,
P
where the angle and the parameter i with i |i |2 = 1, just parametrize the direction
of the goldstino in the S and Ti fields space. Within this framework, the gaugino masses
are [12]
(12)
M1 = M3 = 3 m3/2 sin ,
(13)
M2 = 3 m3/2 1 cos .
In this case the quark doublets and the Higgs fields are assigned to the open string which
spans between the 51 and 9 branes. While the quark singlets correspond to the open
string which starts and ends on the 9 brane, such open string includes three sectors which
correspond to the three complex compact dimensions. If we assign the quark singlets
to different sectors we obtain non-universal A-terms. It turns out that in this model the
trilinear couplings Au and Ad are given by [7,12]
x y z
u
d
A = A = x y z ,
(14)
x y z
where

x = 3 m3/2 sin + (1 3 ) cos ,

y = 3 m3/2 sin + (1 2 ) cos ,
z = 3 m3/2 sin .
(15)
(16)
(17)
The soft scalar masses for quark-doublets and Higgs fields (m2L ), and the quark-singlets
(m2Ri ) are given by

3
(18)
m2L = m23/2 1 1 12 cos2 ,
2

(19)
m2Ri = m23/2 1 3i2 cos2 ,
where i refers to the three families. For i = 1/ 3 the A-terms and the scalar masses are
universal while the gaugino masses could be non-universal. The universal gaugino masses
are obtained at = /6.
It is worth mentioning that in these models (A and B) the gaugino masses, the A-terms,
and the -term are in general complex. However, by using R-rotation we can make the
gaugino masses real and we end up, in addition to the phase of , with the phases of the
A-terms. The phase of is severely constrained by the electric dipole moment (EDM)
of the electron and the neutron [30], while the phases of the A-terms are essentially
unconstrained. Thus one can set the phase of to zero, as done in Refs. [57]. Moreover
it has been shown that the phases of the A-terms can lead to sizeable supersymmetric
contribution to CP observables, in particular on the direct CP violation 0 /. However,
since the total branching ratio b s decay is a CP conserving observable, this should
not be very effective in constraining the phases of the SUSY soft-breaking terms. For this
reason we have made in our analysis the simplifying assumption to set to zero all the
phases.
2.3. Yukawa textures
As emphasized in Refs. [57], in models with non-degenerate A-terms we have to fix
the Yukawa matrices to completely specify the model. In fact, with universal A-terms
the textures of the Yukawa matrices at GUT scale affect the physics at EW scale only
through the quark masses and usual CKM matrix, since the extra parameters contained in
the Yukawa matrices can be eliminated by unitary fields transformations. This is no longer
true with non-degenerate A-terms since in the scalar potential the A-terms enter through
the tensorial product (YqA )ij = (Yq )ij (Aq )ij . Thus the diagonalization of Yq cannot be done
simultaneously with YqA (unlike the universal case). Thus in the models with non-universal
A-terms, some extra degrees of freedom (in addition to the quark masses and CKM matrix)
contained in the Yukawa matrices become observable. Hence, the analysis of the nondegenerate A-terms could shed some light on the favoured Yukawa textures. For instance
in Ref. [6], using a symmetric Yukawa texture with a symmetric A-terms, an accidental
cancellation between the different SUSY contributions to 0 / was found, cancellation
which leads to a very small value for 0 /. On the contrary, by using asymmetric Yukawa
matrices with symmetric A-terms, this cancellation does not occur and it is found that 0 /
can be easily of the order of the KTeV result.
Here we show two realistic examples of Yukawa matrix textures that have already been
used in Refs. [57]. In the first one we have the following symmetric Yukawa matrices
ms
0
0 V13
0
V12
V13
mb
mc
ms
Y u = yt 0
(20)
Y d = y b V12 ms
,
0 ,
V
23
mt
mb
mb
0
1
V13
V13
V23
1
where y b,t are the Yukawa couplings of the bottom and top respectively, and V is the
CKM matrix. The second example is based on the assumption that the CKM mixing matrix
originates from the down Yukawa couplings and that the Yukawa matrices are hermitian.
1
diag(mu , mc , mt ),
v cos
Yu =
Yd =
1
V diag(md , ms , mb ) V .
v sin
(21)
Although the analysis of the CP violation is quite sensitive to the specific Yukawa matrix,
we found that the branching ratio of b s does not essentially depend on it. We checked
this property by using different Yukawa textures (the two examples presented here and
others). We will only present through all the paper the results concerning the second
example just as a representative case.
Now we present the general expressions for the squark mass matrices in the nonuniversal case in the SCKM basis, in this basis the unitary matrices S UR,L and S DR,L
(obtained by a superfield rotation) are chosen to diagonalize the up- and down-Yukawa
couplings Y u,d
T
v sin
mU = SUR Y u,d SU L ,
2
mD =
T
v cos
SDR Y u,d SD
,
L
2
(22)
where T stands for the transpose and mU,D are the diagonal up- and down-quark mass
matrices respectively. In this basis the up and down squark mass matrices at low energy
(respectively, Mu2 and M 2 ) are given by [56]
2
=
Mu,
d
2
Mu,
d
2
Mu,
d

LL
2
Mu,
d
RL
2
Mu,
d

LR
(23)
RR
where for the up-sector

Mu2
Mu2

LL

RR

2
Mu
LR

m2Z
3 4 sin2 W cos 2,
6

2m2Z 2
T
sin W cos 2,
= SUR Mu2 c SU R + m2U +
3

v sin
= Mu2 RL = mU cot + SUL YuA? SU R ,
2
2
2
= SUL MQ
SUL + mU +
(24)
and for the down-sector

Md2
Md2

LL

RR

2
Md
LR

m2Z
3 2 sin2 W cos 2,
6

m2Z 2
T
2
sin W cos 2,
= SDR Md2 c SD
+
m
D
R
3

v cos
= Md2 RL = mD tan + SDL YdA? SD

R
2
2
2
= SDL MQ
SDL + mD
(25)
where M 2 and M 2c c are the soft-breaking (3 3) mass matrices for the squark doublet
Q
u ,d
and singlets respectively. Our convention for the sign of the -term is opposite to the same
one of Ref. [56]. Note that the matrices SU,D , unlike in the universal case, can not be
re-absorbed in the definition of diagonal Yukawa couplings.
10
3. The b s decay in SUSY models

In this section we analyse the b s decay in SUSY models with non-universal Aterms. As pointed out previously, these models are particularly interesting because they can
give sizeable contributions to the CP violating processes through their large contributions
u,d
u,d
)ij or (RL
)ij . For this reason one may expect that large effects
to the mass insertions (LR
can be also induced in the processes mediated by the dipole-magnetic operators, such as
u,d
are large enough, then the SUSY contributions
the rare decay b s . Indeed, if the LR
to these operators are enhanced since the typical chiral suppression is removed by the
insertion of the internal gaugino mass, thus allowing for a competition with the chiralsuppressed SM amplitude.
Let us start with the experimental results. The most recent result reported by CLEO
collaboration for the total (inclusive) B meson branching ratio B Xs is [57]
BR(B Xs ) = (3.15 0.35 0.32 0.26) 104 ,
(26)
where the first error is statistical, the second systematic, and the third one accounts for
model dependence. From this result the following bounds (each of them at 95% C.L.) are
obtained
2.0 104 < BR(B Xs ) < 4.5 104 .
(27)
In addition the ALEPH collaboration at LEP reported a compatible measurement of the

corresponding branching ratio for b hadrons at the Z resonance [58].
The starting point for the theoretical study of b s decay is given by the effective
Hamiltonian
8
X
4GF ?
V33
Ci (b )Qi (b ),
Heff = V32
2
i=1
(28)
where the complete basis of operators in the SM can be found in Ref. [5962]. Recently the
main theoretical uncertainties present in the previous leading order (LO) SM calculations
have been reduced by including the NLO corrections to the b s decay, through the
calculation of the three-loop anomalous dimension matrix of the effective theory [5962].
The relevant SUSY contributions to the effective Hamiltonian in Eq. (28) affect only the
Q7 and Q8 operators, the expression for these operators are given (in the usual notation)
by

e
mb sL bR F ,
Q7 =
2
16

gs
mb sL T a bR Ga .
(29)
Q8 =
2
16
The Wilson coefficients Ci () are evaluated at the renormalization scale b ' O(mb ) by
including the NLO corrections [5962]. They can be formally decomposed as follows
(0)
Ci () = Ci () +

s () (1)
Ci () + O s2 .
4
(30)
11
where Ci(0) and Ci(1) stand for the LO and NLO order respectively. Finally, the branching
ratio BR(B Xs ), conventionally normalized to the semileptonic branching ratio
BRexp (B Xc e) = (10.4 0.4)% [63], is given by [5962]

|V V33 |2
8 s (mb )
6e
1
BRNLO (B Xs ) = BRexp (B Xc e) 32 2
g(z)k(z)
3
|V23|

2
(31)
|D| + A (1 + np ),
with

!
8
X
s (b )
mb
(1)
(0)
(0)
Ci (b ) ri (z) + i7 log
C7 (b ) +
,
D =
4
b
i=1
2

A = es (b ) log (7+2 log )/3 1 C7(0) (b )
(0)
C7 (b ) +
8
s (b ) X (0)
Ci (b )Cj(0) (b )fij (),
i6j =1
where z = m2c /m2b . The expressions for Ci(0) , Ci(1) , and the anomalous dimension matrix
, together with the functions g(z), k(z), ri (z) and fij (), can be found in Ref. [59]. The
term np (of order a few percent) includes the non-perturbative 1/mb [64] and 1/mc [65
68] corrections. From the formula above we obtain the theoretical result for BR(B Xs )
in the SM which is given by
BRNLO (B Xs ) = (3.29 0.33) 104 ,
(32)
where the main theoretical uncertainty comes from uncertainties in the SM input
parameters, namely mt , s (MZ ), em , mc /mb , mb , Vij , and the small residual scale
dependence. 1 The central value in Eq. (32) corresponds to the following central values
pole
pole
pole
= 4.8 GeV, mc = 1.3 GeV,
for the SM parameters mt ' mMS
t (mZ ) ' 174 GeV, mb
b = mb , s (mZ ) = 0.118, e1 (mZ ) = 128, sin2 W = 0.23 and a photon energy
resolution corresponding to = 0.9 is assumed. Note that in Eq. (31) the (small) 1/mc
corrections have not been included.
(0,1)
are obtained by calculating the
The SUSY contributions to the Wilson coefficients C7,8
b s and b sg amplitudes at EW scale respectively. The LO contributions to these
amplitudes are given by the 1-loop magnetic-dipole and chromomagnetic dipole penguin
diagrams respectively, mediated by charged Higgs boson, chargino, gluino, and neutralino
exchanges. The corresponding results for these amplitudes can be found in Ref. [47]. It is
known that the charged Higgs contribution always interferes with the SM contribution [32
47]. The chargino contribution could give rise to a substantial destructive interference with
SM and charged Higgs amplitudes, depending on the sign of , the value of tan , and the
mass difference between the stop masses [3247].
1 Recently in Ref. [69] the current method of extracting the inclusive rate for b s from the currently
published CLEO data has been criticized arguing that the theoretical uncertainties have so far been underestimated
and only a precise measurement of the photon spectrum would be help in reducing these uncertainties.
12
We point out that the SUSY models with non-universal A-terms may induce none7,8 which have opposite chirality with
negligible contributions to the dipole operators Q
respect to Q7,8 . It is worth mentioning that these operators are also induced in the
SM and in the MSSM with supergravity scenario, but their contributions are negligible
being suppressed by terms of order O(ms /mb ). In particular in MSSM, due to the
e7,8 turn out to
universality of the A-terms, the gluino and chargino contributions to Q
be of order O(ms /mb ). This argument does not hold in the models with non-universal
A-terms and in particular in our case. It can be simply understood by using the mass
e7 operators
insertion method [31]. For instance, the gluino contributions to Q7 and Q
d
d
A?
2
A
)23 /mq2 ,
are proportional to (LR )23 ' (SDL Yd SDR )23 /mq and (RL )23 ' (SDR Yd SD
L
D
respectively. Since the AD matrix is symmetric in model A and AD
ij ' Aj i in model B,
d ) ' ( d ) . Then in our case we should consistently take into account the
then (LR
23
RL 23
e8 .
e7 in b s . Analogous considerations hold for the operator Q
SUSY contributions to Q
e
By taking into account the above considerations regarding the operators Q7,8 , the new
physics effects in b s can be parametrized in a model independent way by introducing
e7,8 parameters defined at EW scale as
the so called R7,8 and R
(0)
(0)SM
e(0)
C7,8
C
C7,8
7,8
e
,
R7,8 = (0)SM ,
(33)
R7,8 =
(0)SM
C7,8
C7,8
SM contains only the SM ones. Note that
where C7,8 include the total contribution while C7,8
e7,8 respectively, we have set
e7,8 , which are the corresponding Wilson coefficients for Q
in C
to zero the SM contribution. In Ref. [47] only the expressions for the R7,8 are given, for
e7,8 in the appendix.
completeness we report the corresponding expressions for R
Inserting these definitions into the BR(B Xs ) formula in Eq. (31) yields a general
parametrization of the branching ratio in terms of the new physics contributions [70,71] 2

e72
BR(B Xs ) = (3.29 0.33) 104 1 + 0.622R7 + 0.090 R72 + R

e7 R
e82 , (34)
e8 + 0.002 R82 + R
+ 0.066R8 + 0.019 R7 R8 + R
where the overall SM uncertainty has been factorized outside. We have checked explicitly
that the result in Eq. (34) is in agreement with the corresponding one used in Ref. [72].
Recently the leading EW corrections in the SM have been included in the C7,8
coefficients [73]. However, we have not included them since they should affect the results
in a few percent which is within the theoretical uncertainties present in the SUSY sector.
In particular, in order to be consistent with the NLO calculations, one should also include
(1)
the corresponding two-loop QCD corrections to the SUSY amplitudes, namely C7,8 . In
this respect recent works [7477] have been done in this direction. They calculated the
two-loop s corrections to the chargino and gluino amplitudes. However these corrections
can be consistently applied only in a restricted SUSY scenario with low tan and large
gluino masses, since on the contrary other two-loop diagrams with no QCD interactions
2 Note that the SM central value for BR(B X ) in Refs. [70,71] slightly differs from the result in Eq. (32),
s
since in Refs. [70,71] the non-perturbative /mc corrections have been included.
13
at all could become relevant. In our work we are going to explore SUSY scenarios in
the full range of tan (2 6 tan 6 mt /mb ) and gluino mass, so we do not include these
corrections since they would not change our main conclusions.
4. Numerical results and discussions

In this section we present our results for the total branching ratio BR(B Xs ) as a
function of the fundamental parameters of the soft-breaking sector of models A and B. We
start our discussion by analysing the constraints on these models set by the condition of
vacuum stability and the experimental bounds on the SUSY particle spectrum.
We calculate the low energy SUSY spectrum by running the soft-breaking terms and
the other SUSY parameters (by means of the renormalization group equations (RGE) of
MSSM generalized to the non-universal soft-breaking terms [47]) from the GUT scale
(' 2 1016 GeV) to the EW scale (' MZ ), by using the boundary conditions given in
Section 2. For fixed tan and sign of , we restrict the parameter space by imposing
the present experimental bounds on the SUSY spectra [63]. Since the spectrum of these
models is given in terms of few parameters (m3/2 , and/or i ) we find that by requiring
the lightest chargino mass to be m > 90 GeV implies that all the other SUSY particle
and the lightest Higgs masses are above their experimental bounds.
In order to avoid vacuum instabilities and color-charge breaking, we require that all the
square scalar masses in Eqs. (3), (19) should be positive. All these constraints set strong
restrictions on the parameter
space of both models. As pointed out in Section 2, in model A
mass are
this leads to sin > 1/ 2 ' 0.7. The bounds on m3/2 from the lightest chargino
as follows: for low tan (tan = 2) m3/2 > 80 (60) GeV for sin ' 1/ 2 (1). For large
tan these bounds are increased by 20 GeV in both cases.
In model
B the non-universality is parameterized by the angle and the i s. The values
i = 1/ 3 give universal A terms and scalar masses, the gaugino masses are universal
only at = /6. To avoid negative scalar masses in this model one needs to impose the
constraints
i cos2 < 1/3 (i = 1, 2, 3),

1 12 cos2 < 2/3.
(35)
In the limit of universal A terms these two constraints are satisfied for any value of .
Further constraints on the parameter space of model B are obtained from the gaugino
sector. As shown in Eq. (13), for a not very large m3/2 , the limit of ' /2 cannot be
reached since M2 would approach zero and the mass of the lightest chargino would be
too small. Moreover, the lower bound on the gluino mass and the condition of having the
EW breaking at the correct scale, lead also to a lower bound on the angle . In the case
of universal A-terms we find that > 0.1, while in the non-universal case Eqs. (35) set
more severe constraints on . For example, for 1 ' 1 we obtain that 0.9 6 < /2 and
for close to /2 the gravitino mass m3/2 should be quite heavy (of order TeV) to make
14
Fig. 1. Top: R7 for chargino contribution versus sin in model A, for > 0, m3/2 = 150 GeV, and
for tan = 2, 15, 40. Bottom: The same as before, but for < 0.
the lightest chargino higher than the experimental bound. We find that these constraints
together strongly reduce the allowed parameter space of model B.
S mixing measurements do not set
We have checked that, in both models, the BB
further constraints on the allowed ranges of the parameter space. This mixing, being a
d ) ( d )
1B = 2 process, is proportional to (AB
13
CD 13 where A, B, C, D = (L, R). We
have found that, in our case, the values of these mass insertions satisfy the constraints
given in Ref. [31].
Our results for the partial SUSY amplitude contributions are presented in Figs. 13.
The total branching ratio BR(B Xs ) is shown in Figs. 4, 5 and 6 for model A and
B respectively. In Figs. 13 we show, for model A, the individual SUSY contributions to
the R7 variable, (see Eq. (33) for its definition) versus sin . We see that the chargino and
charged Higgs contributions give the dominant effect in all the range of , while the gluino
is sub-dominant, such as in the universal case. For model B, as expected from its tightly
15
Fig. 2. Top: R7 for the gluino contribution versus sin in model A, for > 0, m3/2 = 150 GeV, and
for tan = 2, 15, 40. Bottom: The same as before, but for < 0.
Fig. 3. R7 for charged Higgs contribution versus sin in model A, m3/2 = 150 GeV, and for
tan = 2, 15, 40.
16
Fig. 4. Top: The BR(B Xs ) versus sin in model A, for > 0, m3/2 = 150 GeV and tan = 2,
15, 40. Bottom: The same as before, but for < 0.
constrained parameter space, we have found that the results of the separate amplitude
contributions do not differ from the corresponding ones in the universal scenario, and will
not be presented here.
We can understand this behavior by using the mass insertion method. The gluino
amplitude gets two leading contributions: one is proportional to the single-mass-insertion
d
d
d
)23 and the another one to the double-mass-insertion, namely (LR
)22 (LL
)23 . In the
(LR
low and intermediate tan regions these two mass insertions are comparable, so a possible
destructive or constructive interference between them may appear depending on the sign
d )
of . In the large tan region the double mass insertion becomes dominant, since (LR
22
is proportional to tan and the amplitude (normalized to the SM one) is
2
MbL sL
Ag
s
tan m2W Mg
,
ASM
W V32
m6q
(36)
17
Fig. 5. Top: The BR(B Xs ) versus sin in model A, for > 0, m3/2 = 300 GeV and tan = 2,
15, 40. Bottom: The same as before, but for < 0.
where mq is the average squark mass in the down sector and M 2

is the off diagonal
bL sL
element of the down-squark mass matrix defined in Eqs. (23), (25). In this way, the
sensitivity of the gluino amplitude to the sign of for large tan can be understood.
The chargino amplitude, like in the gluino case, is enhanced by tan and the dominant
contributions to it are the (Higgsino) Ah and (WinoHiggsino) AW h amplitudes, given
by
Mt2 t
Ah
tan (mt ) 2 L R2 ,
ASM
m m
tL
(37)
tR
Mt2 t
AW h
1
tan (M2 ) 2 L L2 .
ASM
V32
m m
tL
(38)
tR
We see that these contributions are large with respect to the gluino one, such as in the
universal case of MSSM, mainly because the mass insertions get a large enhancement of
18
Fig. 6. Top: The branching ratio BR(B Xs ) versus tan in model B, for > 0, m3/2 = 150 GeV,
and for some values of (1 , 2 ) = (1/ 3, 1/ 3), (0.9, 0.2), (0.6, 0.2), corresponding to the
continuos, dashed, and dot-dashed lines respectively. Bottom: The same as before, but for < 0.
the light-stop mass m2t in the denominator, unlike in the gluino case where the 1/m6q
L
suppression is effective.
e7,8 operators to the total branching ratio, we
Regarding the contribution of the Q
checked that their effect is negligible, almost an order of magnitude smaller than the total
contribution to Q7,8 . This can be explained by observing that the dominant effect to these
operators comes from the gluino amplitude which is much smaller than the chargino or
charged Higgs one.
In Figs. 4, 5 we plot the results for the branching ratio BR(B Xs ), in model A,
versus sin for different values of tan , the sign of and for two representative values
of gravitino mass m3/2 , namely m3/2 = 150, 300 GeV. The main message arising from
these results is that the sensitivity of BR(B Xs ) respect to sin increases with tan .
In particular for the low tan region the b s result does not differ significantly from
19
the universal case. In the large tan region, tan = 1540, the CLEO measurement of
b s set severe constraints on the angle for low gravitino masses. For < 0 almost
the whole range of parameter space is excluded as shown in Figs. 4, 5.
Comparing Top and Bottom plots in Figs. 4, 5, we see that that for positive (negative)
) decreases (increase) when the departure
sign of the branching ratio BR(B Xs
from the universality increases (sin 1/ 2 ). Clearly, due to decoupling effects, the
deviations from universality tend to be reduced for large gravitino masses, as can be seen by
comparing the plots at m3/2 = 150 GeV with the corresponding ones at m3/2 = 300 GeV.
We here recall that the results of Figs. 4, 5 are obtained for a particular choice of the
modular weights ni which is relevant for enhancing the value of 0 . However, we stress
that these results hold a more general validity than for the particular choice which is made.
As we explained in Section 2, that the non universality of the A-terms is not essentially
affected by changing the modular weights, one expects that the b s decay should not
be very sentitive to the modular weights. Indeed we have checked that the variation of the
b s branching ratio is less than 10% when the modular weights are varied within their
natural ranges ni = 1, 2, 3.
Now we discuss results in model B. In Fig. 6 we plot the branching ratio BR(B
Xs ) versus tan for three different values of 1 , 2 (see the figure caption) which are
representative examples for universal and highly non-universal cases. From these figures it
is clear that BR(B Xs ) is not very sensitive to the values of i s parameters, even at
very large tan , unlike model A. The constraints from CLEO measurement are almost the
same in the universal and non-universal cases. For > 0 the branching ratio is constrained
from the lower bound of CLEO only at very large tan , while for < 0 the branching
ratio is almost excluded except at low tan .
5. Conclusions
The recent CP violation measurements of 0 / indicate a large deviation from the SM
predictions which might be interpreted like a signal of new CP violation sources beyond
the SM. It is unlikely that the minimal supersymmetric standard model with universal
boundary conditions at GUT scale can explain these large enhancements in the direct
CP violations. On the contrary, non-minimal SUSY scenarios with non-universal A-terms,
derived from some string inspired models, have been found effective in explaining large
values for 0 / while keeping the electric dipole moments below the experimental bounds.
In this paper we have considered two models based on weakly coupled heterotic string
(model A) and type I string theories (model B). In this framework we carefully analysed
the constraints set by the b s decay on these two models by taking into account the
relevant set of SUSY diagrams. In the calculation of the total branching ratio we take
into account the NLO QCD corrections to the SM. We found that in the model based on
the weakly coupled heterotic string, the b s branching ratio is more sensitive to the
non-universality at large tan and that the dominant SUSY contribution comes from the
chargino amplitude for any value of tan . For type I string-derived model, we found that
20
the sensitivity of the b s branching ratio to the non-universality parameters and i

is quite weak. The main reason for this weakness is because in this model the allowed
ranges for these parameters are strongly constrained by the vacuum stability bounds and
the experimental limits on the lightest chargino mass.
We conclude that the recent CLEO measurements on the total inclusive B meson
branching ratio BR(B Xs ) do not set severe constraints on the non-universality of
these models. Moreover the constraints set on tan and gravitino mass are almost the
same as in the universal case. In this respect we have found that the parameter regions
which are important for generating sizeable contributions to 0 / [57], in particular the
low tan regions, are not excluded by b s decay.
Acknowledgements
We would like to thank C. Muoz for useful discussions. S.K. acknowledges the
financial support of a Spanish Ministerio de Educacion y Cultura research grant. E.G.
acknowledges the financial support of the TMR network, project Physics beyond the
standard model, FMRX-CT96-0090. The work of E.T. was supported by DGICYT grant
AEN97-1678.
Appendix A
e7,8 defined in
Here we give the expressions for the dominant SUSY contributions to R
Eq. (33), namely, the chargino and gluino ones, in the approximation O(ms /mb ) = 0:
e + R
eg ,
e7,8 = R
R
7,8
7,8
e =
R
7
2
? V x F (x )
3V32
33 t W 7 t W
6
2 X
X
xW u k
I =1 k=1
e =
R
8
eg =
R
7
XIL k3
?
XIR k2

2

mI
F3 xI u k + F4 xI u k ,
mb
3
6
2 X
X
xW u k
? V x F (x )
3V32
33 t W 1 t W
I =1 k=1
XIL

k3
XIR
? mI

F4 xI u k ,
k2 m
b
X

? mg

16S
xW dk DL k3 DR k2
F4 xg dk ,
?
27W V32 V33 xt W F7 (xt W )
mb
6
k=1
eg
R
8
2S
=
? V x F (x )
9W V32
33 t W 1 t W
6
X
k=1
with
xW dk DL

k3
DR
? mg

9F3 xg dk + F4 xg dk ,
k2 m
b
(39)
XIL
XIR

ki

ki
= VI 1 UL V
1
=
2 mW cos

1
VI 2 UR MU V ki ,
+
2 mW sin

UI 2 UL V MD ki ,
21
ki
(40)
where xij m2i /m2j , F7 (x) = 23 F1 (x) + F2 (x), U and V are the 2 2 diagonalization
matrices for the chargino mass matrix defined as in Ref. [47], and the 6 6 matrices
(DL ,UL ) (DR ,UR )
, 63
, respectively, diagonalize the Up- and Down-squark mass
(U,D) 63
matrices in Eqs. (23), (25). The expressions for the functions Fi (x) can be found in
Ref. [47].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
M. Brhlik, L. Everett, G.L. Kane, S.F. King, O. Lebedev, Phys. Rev. Lett. 84 (2000) 3041.
R. Barbieri, R. Contino, A. Strumia, hep-ph/9908255.
K. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. D 61 (2000) 091701.
S. Abel, J. Frere, Phys. Rev. D 55 (1997) 1623.
S. Khalil, T. Kobayashi, A. Masiero, Phys. Rev. D 60 (1999) 075003.
S. Khalil, T. Kobayashi, Phys. Lett. B 460 (1999) 341.
S. Khalil, T. Kobayashi, O. Vives, hep-ph/0003086.
L.E. Ibaez, D. Lust, Nucl. Phys. B 382 (1992) 305.
V.S. Kaplunovsky, J. Louis, Phys. Lett. B 306 (1993) 269.
A. Brignole, L.E. Ibaez, C. Muoz, Nucl. Phys. B 422 (1994) 125.
A. Brignole, L.E. Ibaez, C. Muoz, Nucl. Phys. B 436 (1995) 747, Erratum.
L.E. Ibaez, C. Muoz, S. Rigolin, Nucl. Phys. B 553 (1999) 43.
A. AlaviHarati et al., KTeV Collaboration, Phys. Rev. Lett. 83 (1999) 22.
V. Fanti et al., NA48 Collaboration, Phys. Lett. B 465 (1999) 335.
G. DAgostini, hep-ex/9910036.
A. Buras, M. Jamin, M.E. Lautenbacher, Nucl. Phys. B 408 (1993) 209.
M. Ciuchini, E. Franco, G. Martinelli, L. Reina, Nucl. Phys. B 415 (1994) 403.
S. Bosh, A.J. Buras, M. Gorbahn, S. Jager, M. Jamin, M.E. Lautenbacher, L. Silvestrini, hepph/9904408.
M. Ciuchini, E. Franco, L. Giusti, V. Lubicz, G. Martinelli, hep-ph/9910237.
M. Jamin, hep-ph/9911390.
S. Bertolini, M. Fabbrichesi, J.O. Eeg, hep-ph/9802405.
T. Hambye, G.O. Kohler, E.A. Paschos, P.H. Soldan, hep-ph/9906434.
J. Bijnens, J. Prades, JHEP 01 (1999) 023.
E. Pallante, A. Pich, hep-ph/9911233.
E. Gabrielli, G.F. Giudice, Nucl. Phys. B 433 (1995) 3.
E. Gabrielli, G.F. Giudice, Nucl. Phys. B 507 (1997) 549, Erratum.
A. Masiero, H. Murayama, Phys. Rev. Lett. 83 (1999) 907.
D.A. Demir, A. Masiero, O. Vives, hep-ph/9911337.
D.A. Demir, A. Masiero, O. Vives, Phys. Rev. D 61 (2000) 075009.
S. Barr, S. Khalil, Phys. Rev. D 61 (2000) 035005.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
W.S. Hou, R.S. Willey, Phys. Lett. B 202 (1988) 591.
T.G. Rizzo, Phys. Rev. D 38 (1988) 820.
V. Barger, M.S. Berger, R.J.N. Phillips, Phys. Rev. Lett. 70 (1993) 1368.
22
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
J.L. Hewett, Phys. Rev. Lett. 70 (1993) 1045.

R. Barbieri, G.F. Giudice, Phys. Lett. B 309 (1993) 86.
J.L. Lopez, D.V. Nanopoulos, G.T. Park, Phys. Rev. D 48 (1993) 974.
Y. Okada, Phys. Lett. B 315 (1993) 119.
R. Garisto, J.N. Ng, Phys. Lett. B 315 (1993) 372.
M.A. Diaz, Phys. Lett. B 322 (1994) 207.
F.M. Borzumati, Z. Phys. C 63 (1994) 291.
P. Nath, R. Arnowitt, Phys. Lett. B 336 (1994) 395.
S. Bertolini, F. Vissani, Z. Phys. C 67 (1995) 513.
N.G. Deshpande, B. Dutta, S. Oh, Phys. Rev. D 56 (1997) 519.
S. Khalil, A. Masiero, Q. Shafi, Phys. Rev. D 56 (1997) 5754.
T. Blazek, S. Raby, Phys. Rev. D 59 (1999) 095002.
S. Bertolini, F. Borzumati, A. Masiero, G. Ridolfi, Nucl. Phys. B 353 (1991) 591.
L. Everett, G.L. Kane, S.F. King, hep-ph/0005204.
E. Accomando, R. Arnowitt, B. Dutta, Phys. Rev. D 61 (2000) 075010.
M. Brhlik, L. Everett, G.L. Kane, L. Lykken, hep-ph/9908326.
L.E. Ibaez, F. Quevedo, JHEP 9910 (1999) 001.
M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Fermilab, Phys. Rev. Lett. 83 (1999) 2124.
A. Corsetti, P. Nath, hep-ph/0003186.
S. Khalil, hep-ph/9910408.
T. Ibrahim, P. Nath, Phys. Rev. D 61 (2000) 093004.
P. Cho, M. Misiak, D. Wyler, Phys. Rev. D 54 (1996) 3329.
S. Ahmed et al., CLEO Collaboration, CLEO-CONF-99-10, hep-ex/9908022.
R. Barate et al., ALEPH Collaboration, Phys. Lett. B 429 (1998) 169.
K. Chetyrkin, M. Misiak, M. Munz, Phys. Lett. B 400 (1997) 206.
A.J. Buras, A. Kwiatkowski, N. Pott, Phys. Lett. B 414 (1997) 157.
C. Greub, T. Hurth, Phys. Rev. D 54 (1996) 3350.
C. Greub, T. Hurth, Phys. Rev. D 56 (1997) 2934.
Review of Particle Physics, Eur. Phys. J. C 3 (1998) 1
A.F. Falk, M. Luke, M. Savage, Phys. Rev. D 49 (1994) 3367.
M.B. Voloshin, Phys. Lett. B 397 (1997) 275.
A. Khodjamirian, R. Ruckl, G. Stoll, D. Wyler, Phys. Lett. B 402 (1997) 167.
Z. Ligeti, L. Randall, M.B. Wise, Phys. Lett. B 402 (1997) 178.
A.K. Grant, A.G. Morgan, S. Nussinov, R.D. Peccei, Phys. Rev. D 56 (1997) 3151.
A.L. Kagan, M. Neubert, Eur. Phys. J. C 7 (1999) 5.
E. Gabrielli, U. Sarid, Phys. Rev. D 58 (1998) 115003.
E. Gabrielli, U. Sarid, Phys. Rev. Lett. 79 (1997) 4752.
M.A. Diaz, E. Torrente-Lujan, J.W.F. Valle, Nucl. Phys. B 551 (1999) 78.
A. Czarnecki, W.J. Marciano, Phys. Rev. Lett. 81 (1998) 277.
M. Ciuchini, G. Degrassi, P. Gambino, G.F. Giudice, Nucl. Phys. B 527 (1998) 21.
M. Ciuchini, G. Degrassi, P. Gambino, G.F. Giudice, Nucl. Phys. B 534 (1998) 3.
F. Borzumati, C. Greub, Phys. Rev. D 58 (1998) 074004.
C. Bobeth, M. Misiak, J. Urban, Nucl. Phys. B 567 (2000) 153.

From prototype SU(5) to realistic SU(7)

SUSY GUT
J.L. Chkareuli a , C.D. Froggatt b, , I.G. Gogoladze a,c , A.B. Kobakhidze a,d
a Institute of Physics, Georgian Academy of Sciences, 380077 Tbilisi, Georgia
b Department of Physics and Astronomy, Glasgow University, Glasgow G12 8QQ, Scotland, UK
c International Centre for Theoretical Physics, 34100 Trieste, Italy
d Department of Physics, University of Helsinki, 00014 Helsinki, Finland
Received 24 March 2000; accepted 2 November 2000
Abstract
We construct a realistic SU(7) model which provides answers to many questions presently facing
the prototype SU(5) SUSY GUT. Among them are a solution to the doublettriplet splitting problem,
string scale unification, proton decay, the hierarchy of baryon vs lepton number violation and neutrino
masses. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Presently, leaving aside the non-supersymmetric Grand Unified Theories which certainly
contradict experiment unless some special extension of the particle spectrum at intermediate scale(s) is made [1], even the commonly accepted SU(5), SO(10) and E(6) SUSY
GUTs are far from perfect. The problems, as they appear for the prototype supersymmetric
SU(5) model (with a minimal matter, Higgs and gauge boson content) [2], can conventionally be classified as phenomenological and conceptual.
The phenomenological problems include:
(1) The large value of the strong coupling s (MZ ) predicted, s (MZ ) > 0.126 for the
effective SUSY scale MSUSY < 1 TeV [3], in contrast to the world average value [4]
s (MZ ) = 0.119 0.002;
E-mail address: c.froggatt@physics.gla.ac.uk (C.D. Froggatt).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 6 0 - X
24
J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345
Sc ) exchange [2], is largely

(2) The predicted proton decay rate, due to colour-triplet Hc (H
excluded by a combination of the detailed renormalization group (RG) analysis for the
gauge couplings [5] and the improved lower limits on the proton decay mode p K
+
from Super-Kamiokande and on the superparticle masses from LEP2 [6];
(3) The absence of any sizeable neutrino masses m > 102 eV in the model is in conflict
with the atmospheric neutrino deficit data reported in [7].
Furthermore, the present status of the minimal SU(5) SUSY GUT appears to be
inadequate for the following conceptual reasons:
(4) The first one is of course the doublettriplet splitting problem for the Weinberg
Salam Higgs field, taken in the fundamental representation of SU(5), which underlies the
gauge hierarchy phenomenon in SUSY GUT [2];
(5) Then a low unification scale MU is obtained for the MSSM gauge coupling constants,
whose value lies one order of magnitude below that of the typical string unification scale
Mstr ' 5.3 1017 GeV [8];
(6) And lastly, the gravitational smearing of its principal predictions (particularly for
s (MZ )), due to the uncontrollable high-dimension operators induced by gravity in the
kinetic terms of the basic SM gauge bosons [9], makes the ordinary SU(5) model largely
untestable.
The above mentioned three plus three problems seem to be generic for all the presently
popular GUTs. One could imagine that they are all related, in one way or another, with
the still inexplicable gauge hierarchy phenomenon underlying the Grand Unification of
quarks and leptons. It even seems possible that the true solution to the doublettriplet
problem would itself specify the complete framework for Grand Unification, including the
fundamental starting gauge symmetry of the GUT involved.
A number of interesting solutions to the doublettriplet splitting problem have
been proposed, including the sliding singlet mechanism [10], the missing partner
mechanism [11], a special SO(10) based solution [12], the Higgs as pseudo-Goldstone
mechanism [13] and others. However all of these GUTs still have a low unification scale
MU and their further renormalisation group evolution to the string scale Mstr causes a
serious problem. The point is that all these solutions to the doublettriplet problem require
the introduction of a lot of new extended multiplets into the models, which make the
corresponding grand unified theories strongly non-asymptotically free. This invalidates the
perturbative description in these GUTs 1 at energies a little way above MU and below Mstr .
Hence one problem proves to be converted into another one.
In this connection we would like to call the readers attention to a novel possibility
which opens up [15] in some SU(N) GUTs beyond the prototype SU(5) model: namely,
the existence of missing VEV vacuum configurations for an adjoint scalar, which give a
natural solution to the doublettriplet problem together with string scale unification.
1 A good example of this kind is the missing partner SU(5) model [11], in which the extended multiplets
50 + 50 and 75 are introduced. While this model partially improves the situation regarding proton stability [14]
and the predicted large value of s (MZ ) [3], it suffers from a low unification scale.
25
According to the missing VEV mechanism, formulated some time ago [16], the basic
symmetry breaking adjoint scalar ji (i, j = 1, . . . , N ) does not develop a VEV in some of
the directions in the SU(N) space and, thereby, splits the masses of a pair of Higgs fields
S in a hierarchical way through its coupling with them. The hierarchy is supposed
H and H
to be such as to give light weak doublets which break the electroweak gauge symmetry
and give masses to up and down type quarks, on the one hand, and superheavy colour
triplets mediating proton decay, on the other. Interestingly enough, while this possibility
is unrealizable in the standard one-adjoint case [16], the situation radically changes when
two adjoint scalar fields are used or, equally, when the non-renormalisable higher-order
terms are included [15]. As a result, all the above questions appear to be, at least partially,
answered. We show in Section 2 that the missing VEV configurations, which ensure the
survival of the MSSM at low energies, only emerge in extended SU(N) GUTs with N > 7.
We consider the minimal SU(7) model in detail in the main Section 3 of this paper.
Inasmuch as the missing VEV vacuum must be strictly protected from any large
influence coming from the extra symmetry breaking (SU(7) SU(5)), we introduce an
anomalous U (1)A symmetry, supposedly inherited from superstrings [17], as a custodial
symmetry in the model. The separation of the adjoint scalar and extra symmetry breaking
scalar sectors, provided by the U (1)A symmetry, in fact leads to an increase in the global
symmetry of the model. This global symmetry is partially broken, when the SU(7) gauge
symmetry is spontaneously broken, thus producing a set of pseudo-Goldstone states of the
type
5 + 5 + SU(5)-singlets,
(1)
which gain a mass at the TeV scale due to SUSY breaking. With this exception, which
should be considered as the most generic prediction of the model, the spectrum at low
energies looks just as if one had the prototype SU(5) as a starting symmetry. All other
SU(7) inherited states, with the chosen assignment of matter and Higgs superfields, aquire
GUT scale masses during the symmetry breaking, thus completely decoupling from lowenergy physics.
Further, we construct a general R-parity violating superpotential where the effective
lepton number violating couplings immediately evolve at the GUT scale, while the
baryon number non-conserving ones are safely projected out by the missing VEV vacuum
configuration involved. Remarkably, the anomalous U (1)A introduced purely as a missing
VEV protecting symmetry is found to naturally act also as a family symmetry, which can
lead to the observed pattern of quark and lepton masses and mixings.
Another distinctive feature of the SU(7) model considered concerns the relatively low
2/3 1/3
mass scale MP MSUSY [18] of the adjoint moduli fields surviving after SU(7) breaks
and, more importantly, their mass-splitting which inevitably appears in the missing VEV
generating superpotential. This leads to a very different unification picture in SU(7).
String scale gauge coupling unification in the SU(7) model is explicitly demonstrated in
Section 4, for both small and large tan values.
Finally, our conclusions are summarized in Section 5.
26
2. Motivations for SU(7) GUT

It is well known that a missing VEV solution for an adjoint scalar is absent in the
prototype SU(5). Furthermore, even in the general SU(N) GUT, it cannot appear in the
standard one-adjoint case. The main obstacle to this is the presence of a cubic term 3 in
the Higgs superpotential W . This cubic term leads to the impracticable trace condition
Tr 2 = 0 for the missing VEV vacuum configuration, unless there is a special finetuned cancellation between Tr 2 and driving terms stemming from other parts of the
superpotential W [16].
On its own the elimination of the 3 term leads to the trivial unbroken symmetry case.
However the inclusion of higher even-order terms in the effective superpotential, or the
introduction of another adjoint scalar with only renormalizable couplings appearing in
W , leads to an all-order missing VEV solution, as was shown in recent papers [15].
Let us consider briefly the high-order term case first. The superpotential for an adjoint
scalar field conditioned by the gauge Z2 reflection symmetry ( )
1
1
2
4 +
22 +
WA = m 2 +
2
4MP
4MP
(2)
contains, in general, all possible even-order terms scaled by inverse powers of the
(conventionally reduced) Planck mass MP = (8GN )1/2 ' 2.4 1018 GeV. As one can
readily confirm, the necessary condition for any missing VEV solution to appear in the
SU(N) Z2 invariant superpotential WA is the tracelessness of all the odd-order terms
Tr 2s+1 = 0 ,
s = 0, 1, 2, . . . .
(3)
This condition uniquely leads to a missing VEV pattern of the type

Nk
k/2
k/2
z }| { z }| { z }| {
hi = Diag( 0, . . . , 0, 1, . . . , 1, 1, . . . , 1 ),
(4)
where the VEV value is calculated using the polynomial taken in WA (2). The
vacuum configuration (4) gives rise to a particular breaking channel of the starting SU(N)
symmetry
SU(N) SU(N k) SU(k/2) SU(k/2) U (I )1 U (I )2 ,
(5)
which we will discuss in some detail a little later. For the moment one may conclude
from Eqs. (4), (5) that a missing VEV solution, which retains the ordinary MSSM gauge
symmetry SU(3)C SU(2)W U (1)Y at low energies, could actually exist if the SU(5)
GUT were properly extended.
The superpotential (2) could be viewed as an effective one, following from an ordinary
renormalizable two-adjoint superpotential with the second heavy adjoint scalar integrated
out. Hereafter, although both approaches are closely related, we deal for simplicity with
the two-adjoint case. Towards this end, let us consider in more detail a general SU(N)
invariant renormalizable superpotential for two adjoint scalars and , satisfying also
the gauge-type Z2 reflection symmetry ( , ) supposedly inherited from
superstrings
27
1
1
1
1
WA = m 2 + MP 2 + h 2 + 3 .
(6)
2
2
2
3
Here the second adjoint can be considered as a state originating from a massive
string mode with the Planck mass MP . The basic adjoint may be taken at another
2/3 1/3
well motivated scale m MP MSUSY O(1013) GeV [18] where, according to many
string models, the adjoint moduli states (1c , 1w ), (1c , 3w ) and (8c , 1w ) (in a self-evident
SU(3)C SU(2)W notation) appear. In the present context these states can be identified
as just the non-Goldstone remnants 0, 3 and 8 of the relatively light adjoint which
breaks SU(N) in some way. However, all our conclusions remain valid for any reasonable
value of m, which is the only mass parameter (apart from MP ) in the model considered. In
fact we vary m between 1012 and 1016 GeV.
As a general analysis of the superpotential WA (6) shows [15], there are just four possible
VEV patterns for the adjoint scalars and :
(i) the trivial unbroken symmetry case, = = 0;
(ii) the single-adjoint condensation, = 0, 6= 0;
(iii) the parallel vacuum configurations, and
(iv) the orthogonal vacuum configurations, Tr() = 0.
The Planck-mass mode , having a cubic term in WA , in all non-trivial cases develops a
standard single-breaking VEV pattern
k
Nk
}|
{
z }| { z
Nk
,
.
.
.
,
,
hi = Diag 0, . . . , 0, Nk
k
k
(7)
which breaks the starting SU(N) symmetry to

SU(N) SU(N k) SU(k) U (I ).
(8)
However, in case (iv), the basic adjoint develops the radically new missing VEV
vacuum configuration (4), thus giving a double breaking of SU(N) to (5). Using the
approximation h MmP , which is satisfied for any reasonable values of the couplings h
and in the generic superpotential WA (6), the VEV values are given by

2N 1/2 p
k m
,
=
mMP / h,
(9)
=
N k h
N k
respectively. Remarkably, just the light adjoint develops the largest VEV in the model
which, for a properly chosen adjoint mass m and coupling constant h, can easily come up
to the string scale Mstr (see Section 4).
Furthermore, as has already been intimated above, in order to have the standard
gauge symmetry SU(3)C SU(2)W U (1)Y remaining after the breaking (5), one must
go beyond the prototype SU(5). As is easily seen from Eqs. (4), (5), there are two
principal possibilities: the weak-component and colour-component missing VEV solutions
respectively. If it is granted that the missing VEV subgroup SU(N k) in (5) is just the
weak symmetry group SU(2)W , as is traditionally argued [16], one is led to SU(8) as the
GUT symmetry group (N k = 2, k/2 = 3) [15]. Another, and in fact minimal, possibility
is to identify SU(N k) with the colour symmetry group SU(3)C in the framework
28
of an SU(7) GUT symmetry (N k = 3, k/2 = 2); this SU(7) model is considered

further here. The higher SU(N) GUT solutions, if considered, are based on the same two
principal possibilities: the weak-component or colour-component missing VEV vacuum
configurations, respectively.
Let us see now how this missing VEV mechanism works to solve the doublettriplet
splitting problem in both SU(8) and SU(7) GUTs. It is supposed that there is a reflectioninvariant coupling of the ordinary MSSM Higgs-boson containing supermultiplets H and
S with the basic adjoint in the superpotential WH :
H
SH
WH = f0 H
SH H
SH ).
( , H
(10)
S with the second adjoint is then forbidden by the above Z2

The coupling of H and H
reflection symmetry ( , ) underlying the basic superpotential WA (6). The
S do not develop VEVs during the first stage of the symmetry breaking.
superfields H and H
S determined by the missing VEV pattern
Thus the WH turns into a mass term for H and H
(4). This vacuum, while giving generally heavy masses (of the order of the GUT scale
S, leaves their weak components strictly massless. To be certain of this,
MU ) to H and H
S for both the weak-component and the
we must specify the multiplet structure of H and H
colour-component missing VEV vacuum configurations, that is in SU(8) and SU(7) GUTs,
S are fundamental octets whose weak components
respectively. In the SU(8) case, H and H
(ordinary Higgs doublets) do not get masses from the basic coupling (10). In the SU(7)
S are 2-index antisymmetric 21-plets which (after projecting out the extra
case, H and H
heavy states, see Section 3.3) contain just a pair of massless Higgs doublets. Thus, there
certainly is a natural doublettriplet splitting in both cases and we also have a vanishing
term at this stage. However, one can readily show that the right order term always
appears as a result of radiative corrections at the next stage when SUSY breaks [15].
We consider below the minimal SU(7) GUT in some detail.
3. Basics of the SU(7) model

3.1. Matter and Higgs superfields
By analogy with the standard SU(5) model, we consider the simplest anomalyfree
combination of the fundamental and antisymmetric 2-index representations of the SU(7)
gauge group

A
S A + [AB]
(11)
+ A +
i
(where A, B = 1, . . . , 7 are SU(7) indices) for each of the three quarklepton families
S A (7) +
or generations (i = 1, 2, 3). The quarklepton states reside in the multiplets
A
A
[AB] (21), while the extra fundamental multiplets (7) and (7) are specially
introduced in (11) for anomaly cancellation. There is also a set of Higgs superfields, among
which are the two already mentioned adjoint Higgs multiplets BA (48) and BA (48). They
are responsible here for the GUT scale breaking of SU(7)
SU(7) SU(3)C SU(2)W SU(2)E U (I )1 U (I )2
(12)
29
according to the VEVs (4), (7) and (9), which now become
hi = Diag[0, 0, 0, 1, 1, 1, 1],
(13)

3
3
3
3
(14)
hi = Diag 1, 1, 1, 4 , 4 , 4 , 4 ,
where = 14mMP /3/ h, = 4m/3h and the colour (C), weak (W ) and extra (E)
components are indicated. In addition there are a pair of Higgs multiplets H[AB] (21) and
S [AB] (21) in conjugate representations, which contain the ordinary electroweak doublets.
H
One can easily check that, due to their basic coupling (10) with the adjoint , which
develops the missing VEV configuration (13), all the states in the multiplets H[AB] and
S [AB] become superheavy with mass of order MU , except for one pair of colour triplets
H
and two pairs of weak doublets. Thus, in order to have just the standard pair of MSSM
electroweak doublets, the other pair of weak doublets should be projected out from the
massless state spectrum together with the colour triplet states. This is accomplished by
S [AB] with specially introduced heavy scalar supermultiplets,
the mixing of H[AB] and H
S [ABC] , which contain just the
the 3-index antisymmetric 35-plets of SU(7) [ABC] and
required states (see Section 3.3). And, finally, there are two fundamental scalar superfields
A (7) and A (7) and their conjugates A (7) and A (7) which break the extra symmetry
at the GUT scale. We consider this key question first.
3.2. Extra symmetry breaking
Inasmuch as the extra symmetry should also be broken
SU(7) SU(5)
(15)
at the GUT scale, in order not to spoil gauge coupling unification, a question arises: how
can the adjoint missing VEV configuration (13) survive so as to be subjected to at most
a shift of order the electroweak scale? This requires, in general, that the superpotential
WA (6) be strictly protected from any large influence of the scalars and , which
provide the extra symmetry breaking (15). Technically, such a custodial symmetry may be
a superstring-inherited anomalous U (1)A [17], which induces a high-scale extra symmetry
breaking (15) through the FayetIliopoulos (FI) D-term [19]:
X

2
Tr QA 2 2
QnA F n ,
=
g M .
(16)
DA = +
192 2 str P
Here the sum runs over all charged scalar fields in the theory, including those which do
not develop VEVs and only contribute to Tr QA . For realistic or semi-realistic models,
Tr QA has turned out to be quite large, Tr QA = O(100) (see the recent discussion
in [20]). So, the spontaneous breaking scale of the U (1)A symmetry and the related extra
symmetry (15) is naturally located at the string scale. The protecting anomalous U (1)A
symmetry is supposed to keep the scalars and essentially decoupled from the basic
adjoint superpotential WA (6), so as not to strongly influence its missing VEV vacuum
configuration (13) through the appearance of potentially dangerous couplings of the type
and .
30
Anyway, once a separation of the adjoint scalar and extra symmetry breaking scalar
sectors takes place in the supersymmetric SU(7) U (1)A theory considered, an accidental
global symmetry SU(7) U (7) appears. This global symmetry is radiatively
broken and one or two families of pseudo-Goldstone (PG) states of type (1) are produced
at a TeV scale, where SUSY softly breaks [15]. The two-family case corresponds to the
most degenerate Higgs potential, where the scalars and are only allowed to appear
through the basic SU(7) and U (1)A D-terms and, thereby, increase their global symmetry
to U (7) U (7) . This two-family case occurs when the U (1)A charges of the bilinears
,
,

and
are all positive (or all negative), so that they can not appear in the
SU(7) U (1)A invariant superpotential in any order.
However, remarkably, it is possible for the adjoint and fundamental scalar sectors in
the superpotential to overlap without disturbing the adjoint missing VEV configuration.
This naturally occurs when the scalars and are conditioned by the U (1)A symmetry to
develop orthogonal VEVs along the extra directions
A = A6 V1,
A = A7 V2 .
(17)
The simplest choice of such safe mixing terms is given by dimension-5 operators, invariant
under the reflection symmetry , ,
,
of the type 2
WH1 =

1
a S + b
.
MP
(18)
The dimensionless coupling constants a and b are both of order O(1) and S is some new
singlet superfield which gets its VEV through the FI D-term (16), just as the scalars and
do. One can consider the field S as a basic carrier of unit U (1)A charge in the model.
In terms of its charge, the charges of the bilinears in WH1 are determined to be +1 for
and 1 for ,
while the charges of the bilinears
and
are not yet determined
from the couplings (18). The latter charges can be taken to be positive, as follows when
we consider the other coupling terms in Sections 3.33.6. This implies that any terms
containing and scalars in the superpotential must also include the bilinear ,
so as to
properly compensate the U (1)A charges. However, for a vacuum configuration where the
orthogonality condition
= 0 naturally arises, this gives an all-order solution excluding
the dangerous
and
terms. In fact this orthogonality condition is precisely one

of the conditions satisfied at the SUSY invariant global minimum of the Higgs potential,
as follows from the vanishing F -terms of the superfields involved in (18):
a
(19)
= 0,
= S.
b
Here the VEV value, appearing on the extra symmetry components of the adjoint (13),
has been used. One can now readily see that a non-diagonal mass-term appears for the PG
states related with the multiplets and and their conjugates
00
a
=
S( + I ),
(20)
M
WH1
MP
2 Enlarging the scalar sector properly one can write a renormalizable superpotential as well.
31
where I is the 7 7 unit matrix. This explicitly shows that one superposition of the two
PG 5 + 5 families (1) becomes heavy, while the other is always left massless. In fact this
result is a general consequence of the symmetry breaking pattern involved. The point is
that neither of the other mass-terms M
, M
and M
can be allowed by the U (1)A
symmetry for any generalization of the superpotential WH1 (18); otherwise the dangerous
and
couplings inevitably appear as well. Similarly, in the general SU(N) case,

it can be shown [21] that at least one family of PG states of type (1) always exists. Together
with the ordinary quarks and leptons and their superpartners these PG states, both bosons
and fermions, determine the particle spectrum at low energies. In most of what follows the
existence of just one family of PG states at the sub-TeV scale will be assumed.
3.3. Heavy states
We demonstrate below that, when the extra symmetry is broken (15), all the states
in the SU(7) model, beyond the ordinary MSSM particle spectrum plus one family of
pseudogoldstones (1), acquire masses of order the GUT scale. An exception can be made
for the sterile states (the states in the matter multiplets (11) having the extra symmetry
charges only) whose fermionic components might be referred to as sterile neutrinos (see
Section 3.4).
First of all let us consider the Higgs sector and show that all the states in the basic Higgs
S [AB] become superheavy, except for one pair of weak doublets.
multiplets H[AB] and H
One can readily check that, when the colour-component missing VEV solution (13) is
substituted into the superpotential WH (10), superheavy masses are generated for most of
S multiplets. However, the following states (weak, colour
the components of the H and H
and extra symmetry components are explicitly indicated)
Hw6 ,
Sw6 ,
H
Hw7 ,
Sw7 ,
H
H[cc0 ] ,
S [cc0 ]
H
(21)
remain massless at this stage of SU(7) symmetry breaking. Therefore one of the two pairs
of weak doublets in (21), as well as the colour triplets, must also become heavy in order to
obtain the ordinary picture of MSSM at low energies. This happens as a result of mixing
S with the specially introduced (see Section 3.1) superheavy scalar supermultiplets
H and H
S [ABC] in the basic Higgs superpotential
[ABC] and
S + f H
S + y S .
S
WH 2 = f H
(22)
Here f , f and y are dimensionless coupling constants. When the scalars and get their
VEVs, thus breaking the extra symmetry, the required mixing terms are generated. It is
S
S mixings in WH2
worth noting that the presence of the conjugated H
and H
could allow the dangerous
and
terms, destroying the missing VEV solution,

S has a nonzero U (1)A charge. Therefore, this term appears
unless the bilinear term
in WH2 together with the singlet scalar superfield S the basic U (1)A charge carrier
introduced earlier in WH1 (18).
It is easy to see now that the WH2 couplings (22) will rearrange the mass spectrum of
the states (21), so as to leave just a standard pair of MSSM electroweak doublets massless.
32
Considering the mixing of the colour triplet states first, one can see from the 2 2 mass
S [cc0 ] and the double-coloured components [cc0 6] and
matrix for the states H[cc0 ] and H
0
S [cc 6] that, when properly diagonalized, the colour components in (21) obtain a mass M
of order
f f hihi
MU .
(23)
M
y
S
The combination of primary coupling constants f, f and y in (23), can be taken O(1)
in general. For the weak doublet case, there is a 3 3 mass matrix corresponding to the
Sw6 , H
Sw7 and
S [w67] ,
mixing of the states Hw6 , Hw7 and [w67] , and their conjugates H
respectively. After diagonalization this matrix leaves, as can readily be checked, just one
Sw6 strictly massless, while the other pair Hw7 and H
Sw7
pair of weak-doublets Hw6 and H
acquires a mass of order M (23).
S, which get masses from
It seems reasonable to assume that the components of H and H
their direct coupling (10) with the basic adjoint , should have a mass of order MU , while
the states in (21), which develop a mass M from their mixing with the heavy and
S multiplets (22), could naturally be relatively light, M = O(102 1)MU . In this case,
Sc6 , which are partners of the light

the proton decay inducing colour-triplet states Hc6 and H
w6
S
weak-doublets Hw6 and H , are located at the GUT scale MU , while the double-coloured
S [cc0 ] (21), which can not induce proton decay, are relatively light. Thus
states H[cc0 ] and H
the problem of an unacceptably fast proton decay, due to dimension-5 operators, would
be naturally solved in the SU(7) model considered. At the same time the double-coloured
S [cc0 ] , if taken relatively light, would properly contribute to the running
states H[cc0 ] and H
of the gauge coupling constants (see Section 4).
Next we consider the additional matter multiplets, two anti-septets A and A for each of
the three generations, which were introduced (Section 3.1) for SU(7) anomaly cancellation.
They are assumed to form their masses by combining with the basic 2-index multiplet
[AB] of their own generation
WY 1 = g A [A6] 6 + g A [A7] 7 .
(24)
Here the SU(7) index A and the extra symmetry subgroup indices 6 and 7, along which the
VEVs (17) develop, are explicitly indicated (g, are coupling constants). So, again one
can see that any state in the multiplets and (for all three generations) carrying colour
and/or electric charge acquires a mass of order MU .
3.4. Sterile neutrinos
The states in the matter multiplets (11) of each generation, which only have charges
under the extra symmetry
[67] ,
S6 ,
S 7,
6 ,
7 ,
6 ,
(25)
are of particular interest as possible SU(7) candidates for sterile neutrinos. One can see
hi
MU from
that two superpositions of [67] , 7 and 6 acquire masses of order hi
the couplings (24). As to the other states in (25), their masses depend on the U (1)A charges
33
Table 1
The U (1)A (eiQA ) and Z2 (einZ ) charges, which allow just the couplings appearing in the total
superpotential WT (34)
F
S
H
QA
nZ
0
1
0
0
3
0
1
0
2
0
2
1
0
1
0
0
2
0
3
1
2
0
1
1
5
0
4
0
1
0
assigned to the matter and Higgs superfields involved. If one takes the simple set of charges
presented in Table 1, all of them in fact can get masses from the allowed high-order terms
(inherited from superstrings or induced by gravitational corrections) of the type

S
S 2
S 3
S
S + a2
S
S
+ a3
+ a4

MP WY 2 = a1
MP
MP
MP

S 2
+ ,
(26)
+ a5 + a6
MP
where the higher order terms indicated by dots are ignored (overall SU(7) index contraction
is implied and the dimensionless coupling constants a1 , . . . , a6 are assumed to be all of
O(1)). It follows, from the couplings in WY 1 (24) and WY 2 (26), that for each generation
the symmetric 7 7 mass matrix of the sterile states (25) has all its eigenvalues nonzero in
general. The smallest eigenvalue is just given by the order of the highest term kept in the
superpotential WY 2 and is thus of order MU5 /MP4 .
Some of these sterile states should be considered as candidates for right-handed
neutrinos. The traditionally used Planck or GUT mass scale right-handed neutrino leads,
via the well-known see-saw mechanism [24], to ordinary neutrino masses in the range
m = 105 103 eV or lower, which cannot explain the recent Super-Kamiokande
atmospheric neutrino data [7]. By contrast, the sterile states discussed here 3 could naturally
lead to neutrino mass(es) in just the required region m = 102 1 eV.
Interestingly, with the choice of U (1)A charges taken in Table 1, the direct Higgs-matter
mixing terms such as
2
3
S S + b3 S +
WY 3 = b1 S + b2
MP
MP2
(27)
can also evolve, provided that R-parity symmetry is not assumed. They will induce the
condensation of some of the heavy sterile sneutrinos, just like takes place for ordinary
sneutrinos in the bilinear Higgs-lepton mixing model [23].
At the same time, it is well to bear in mind that some other choice of the U (1)A charges,
different from those in Table 1, could (partially or completely) forbid the couplings (26)
and (27), thus leading to certain light and even strictly massless sterile neutrinos. As is well
known, they are not ruled out by experiment [7,23].
3 Other potentially interesting right-handed neutrino candidates, also well motivated in the SU(7) model, are
2/3 1/3
the fermionic superpartners of the adjoint moduli states 0 and 3 at a scale of MP MSUSY stemming from
superstrings [18].
34
3.5. Lepton number violation

One would think that there is no fundamental reason for exact R-parity (RP) symmetry
in the framework of supersymmetric GUTs, where not only fermions but also their scalar
superpartners are the natural carriers of lepton and baryon numbers. Accordingly, we
suppose that all the generalized Yukawa couplings, the RP-conserving (ordinary up and
down fermion Yukawas), as well as the RP-violating ones allowed by SU(7) U (1)A
symmetry, are given by a similar set of dimension-5 operators. We note that the usual
dimension-4 trilinear Yukawa couplings are forbidden by the underlying gauge invariance.
The dimension-5 operators take the form (i, j, k = 1, 2, 3 are the generation indices)
up
Oij =
Guij
MP
(i j )(H ),
(28)
Gdij
S),
Si j )(H
(
(29)
MP
Gij k
rpv
Si j )(
Sk ),
(
(30)
Oij k =
MP
and are collected in the Yukawa part of the superpotential WY . As specified above
(Section 3.1), each of the three generations of quarks and leptons lies in two multiplets
S A + [AB] of SU(7). Also the ordinary electroweak doublets are contained in the
S [AB] , while the scalars and and their conjugates, which break
SU(7) to SU(5), are fundamental septets and anti-septets, respectively.
Now, on substituting VEVs for the scalars (13), and (17) in the basic operators
(28)(30), one obtains at low energies the effective renormalisable Yukawa and lepton
number violating (LNV) interactions with coupling constants
Oijdown =
hi
hi
hi
,
Yijd = Gdij
,
ij k = Gij k
.
(31)
MP
MP
MP
At the same time the baryon number violating (BNV) couplings prove to be completely
eliminated. The key element here turns out to be that the adjoint field , involved
in the effective couplings (30), develops a VEV configuration with strictly zero colour
components (13) in the SUSY limit. However, at the next stage when SUSY breaks, the
radiative corrections will shift the missing VEV components of to some nonzero values
of order MSUSY . In this way the ordinary -term of the MSSM, on the one hand, and
baryon number violating couplings with hierarchically small coupling constants of the
order MSUSY /MU , on the other, are induced. So, our missing VEV solution to the gauge
hierarchy problem leads in fact to a similar hierarchy of baryon vs lepton number violation.
The O-operators (28)(30) can be viewed as effective interactions generated through
the exchange of some superheavy states, which we can interpret as massive string modes.
When they are generated by the exchange of the same superheavy multiplet (formed
from a pair of fundamental septets 7 + 7), the operators (29) and (30) appear with the
dimensionless effective coupling constants aligned in flavour space [22]:
Yiju = Guij
ij k = Yijd k .
(32)
35
Here the parameters k (k = 1, 2, 3) include some known combination of the primary

dimensionless coupling constants and a ratio of the VEVs of the scalars and . This strict
relation, between general RP-violating and down fermion Yukawa coupling constants, is
then split into the ones for charged lepton (cl) and down quark (dq) LNV couplings,
respectively,
ij k = Yijcl k ,
0ij k = Yij k ,
dq
(33)
when it is evolved from the GUT scale down to low energies.

Therefore, the postulated common origin of all the generalized Yukawa couplings,
both RP-conserving and RP-violating, at the GUT scale results in some minimal form
of lepton number violation, with the proviso that appropriate mediating superheavy-states
exist. As a result, all significant physical manifestations of LNV reduce to those of the
S and LQD
S aligned, both in magnitude and orientation
effective trilinear couplings 4 LLE
in flavour space, with the down fermion (charged lepton and down quark) effective Yukawa
couplings. However the effective bilinear terms i Li H appear to be generically suppressed
by the custodial U (1)A symmetry involved. Detailed phenomenology of this model related
to the flavor-changing processes both in quark and lepton sectors, radiatively induced
neutrino masses and decays of the LSP can be found in a recent paper [22].
3.6. Masses and mixings of quarks and leptons
So far we have used an anomalous U (1)A symmetry purely as a missing VEV protecting
symmetry in the SU(7) model. In Table 1 we list the U (1)A (eiQA ) and Z2 (einZ ) charges,
which allow just the couplings appearing in the total superpotential WT :
WT = WA + WH + WH1 + WH2 + WY + WY 1 + WY 2 + WY 3
(34)
(see Eqs. (6), (10), (18), (22), (24), (26)(30)). The superpotential WT in turn completely
fixes all the QA charges in terms of the charge of the singlet scalar superfield S, which we
take to be unity QSA = 1. Remarkably, as one can see from Table 1, the QA charges of the
scalar bilinears
and ,
which break the extra symmetry (SU(7) SU(5)), happen to
be positive, while the adjoint scalar has QA = 0. Hence the potentially dangerous
and
couplings are forbidden, thus providing an all-order solution to the doublettriplet

splitting problem in the SU(7) model.
However, together with its protecting function, it is tempting to treat the anomalous
U (1)A also as a family symmetry [25]. Then the zero charges QA presented in Table 1 for
S A would only be assigned to the third family of quarks
the matter multiplets [AB] and
and leptons, while the other two families would be assigned some nonzero QA charges.
The observed hierarchy of quarklepton masses and mixings might then be generated via
the FroggattNielsen mechanism [26]. In the case at hand this hierarchy is described by
hSi
str
M
the natural expansion parameter = M
MP (' 0.2). One can readily see that a simple
P
4 Here L and Q denote lepton and quark SU(2) doublet superfields, while E
S and D
S denote lepton and down
quark SU(2) singlet superfields.
36
S A (as explicitly indicated in

choice of QA charges for the matter multiplets [AB]i and
i
S
brackets in units of QA ) leads to Yukawa matrices of the type
(3)
6
1(3)

Y u 2(2) 5
(0)
3
3
(2)
2
5
4
2
(0)
3
2
1
(35)
and
(3)
Yd
S (3)
1
S (2)
2
S (0)
(2)
5
3
3
4
2
2
(0)
2
1 .
1
(36)
These matrices yield a quark and charged lepton mass hierarchy similar to the observed
pattern, as well as the right orders of magnitude for the CKM matrix elements (see
also [27]). Phenomenologically, in the case of Y u (35) the constant of proportionality is
of order unity. However, in the case of Y d (36), the constant of proportionality is only of
order unity for large values of tan . In the case of small tan it is necessary to assume that
the underlying fundamental couplings give a constant of proportionality of order 0.01. 5
There is another important aspect of the model related with masses of quarks and leptons
which is worthy of special note. The SU(7) GUT considered, as well as the prototype
SU(5) model, predicts (see Eq. (29)) the equality of the down quark and charged lepton
masses at the grand unification scale. However, due to the string scale unification in the
SU(7) model, the influence of the gravitational corrections on the Yukawa couplings is
much more important than in the prototype SU(5) [30]. Actually, in general, one must
include SU(7) U (1)A Z2 invariant corrections to the down fermion Yukawa couplings
(29), (31) and (36) of the type
down
Oij,cor
|q i +qj | S 2 S
i j H ,
MP3
(37)
Si ) and qj = QA (j ) are the U (1)A charges for the matter multiplets

where q i = QA (
given in Eq. (36). These gravitational corrections break the equality between the down
quark and charged lepton Yukawa couplings at the unification scale. They give rise to
effective dimensionless coupling constants Yij0 of order hi2 /MP2 times the right-hand
side of Eq. (36). Thus, taking hi to be of order the string scale, the corrections Yij0
0
turn out to be of the same order as the physical Yukawa couplings Yijd (e.g., Y33
Yb, 0.01) in the small tan case. Thereby, under the circumstances considered (stringscale unification plus small tan ), gravitational corrections are expected not only to spoil
5 It must be admitted that the origin of this small factor is not understood, just as the origin of a large value of
tan is not understood. Nonetheless, it is possible to generate this hierarchical bottom to top quark mass ratio,
S1 ,
S2 and
S3 . The Y d
by uniformly subtracting two units of U (1)A charge from all down fermion multiplets
matrix is then multiplied by an overall factor of order O( 2 ). However, we do not make use of this option here.
37
the SU(5) quarklepton mass predictions for the d and s quarks, as argued in [30], but
also for the b quark. At the same time these corrections will not significantly disturb the
hierarchical structure of the mass matrices (35) and (36), which is essentially protected
by the U (1)A gauge symmetry. In the large tan case the gravitational corrections are
quite small even for a string-scale unification and so the b Yukawa unification, properly
corrected by supersymmetric loop contributions (see further discussion in Section 4), is
actually predicted as in the SU(5) model.
4. Gauge coupling unification

We now consider gauge coupling unification in the above SU(7) model. At the lowenergy scale, in addition to the three standard families of quarks and leptons (and squarks
and sleptons), there is just one family of PG bosons (1) and their superpartners which will
modify the running of the gauge and Yukawa couplings in the model. However, as we will
see, the contribution of the split adjoint moduli 3 and 8 taken at their natural scale
2/3 1/3
MP MSUSY (for earlier works see [18,28]) turns out to be much more important.
We start with the reminder that, to one-loop order, gauge coupling unification is given by
the three renormalization group (RG) equations, relating the values of the gauge couplings
at the Z-peak i (MZ ) (i = 1, 2, 3) and the common gauge coupling U at the unification
scale MU [1]:
1
+
i1 = U
X bp MU
i
ln
.
2
Mp
p
(38)
Here bi are the three beta function coefficients, corresponding to the SU(7) subgroups
U (1)Y , SU(2)W and SU(3)C , respectively, for the particle labeled by p. The sum extends
over all the contributing particles in the model, and Mp is the mass threshold at which
each decouples. All of the SM particles, and also the second electroweak doublet of
MSSM, are taken to be present at the starting scale MZ . The next contribution enters at the
supersymmetric threshold, associated with the decoupling of the supersymmetric particles
at some single effective scale MSUSY [3]. We propose to use relatively low values for this
scale, MSUSY MZ , so as to keep sparticle masses typically in the few hundred GeV
region. Furthermore, the PG states (1) are also taken at a sub-TeV scale. As to the heavy
states, there are basic thresholds relating to the adjoint moduli 8 and 3 , with masses M8
and M3 respectively, and the thresholds caused by certain components in the SU(7) Higgs
S [AB] ,
Hw7 ,
Sw7 ,
H
H[cc0 ] ,
S [cc0 ] ,
H
(39)
developing masses of order M (23), which we argued in Section 3.3 could lie somewhat
lower than the GUT scale MU . We refer to the latter as the H -states. All other states in
the Higgs, matter and gauge multiplets, including the superheavy gauge bosons and their
superpartners, do not contribute to Eq. (38), for they are assumed to lie at the GUT scale
MU , above which all particles fill complete SU(7) multiplets.
38
Now, the matching equation for the gauge couplings (38) reads as follows:

3
MU
M3 19 MSUSY
MU
+ ln
7 ln
2 ln
ln
1221 731 511 =
.
2
M
M3
M8
6
MZ
(40)
This can be viewed as the basis for understanding qualitatively the constraint on the value
of 3 = s (MZ ) from grand unification, and its dependence on the present precision
electroweak measurements of 1 and 2 [4] and all the thresholds involved. One can
see from Eq. (40) that s (MZ ) increases with MU /M3 and decreases with MU /M ,
MSUSY /MZ and, especially, with M3 /M8 (the term with the largest coefficient in front of
the logarithm). Paradoxically enough, in the absence of all threshold effects (M = M3 =
M8 = MU and MSUSY = MZ ), Eq. (40) leads to a phenomenologically acceptable value of
s (MZ )
7em
= 0.117,
(41)
s =
15 sin2 W 3
1 = 127.9 and sin2 = 0.2313 taken at M [4]. Unfortunately, this
using the values em
W
Z
value increases unacceptably when 2-loop order corrections are included. In the standard
SU(5) case [2], with degenerate adjoint moduli 3 and 8 (M3 = M8 at the GUT scale),
an unacceptably high value of s is obtained even if all the possible threshold effects are
taken into account [3,5,6].
However, a drastically different unification picture appears when a generic mass splitting
between 3 and 8 , which is a consequence of the missing VEV inducing superpotential
(6), is taken into account. 6 Actually, after the adjoint scalars and develop their VEVs
(13), the physical masses of the surviving adjoint moduli 3 and 8 turn out to be fixed.
Diagonalization of the common mass matrix
00
(42)
Mab = Wab hi,hi
(where the indices a and b stand for the corresponding components of and ) leads to
M3 =
14m
,
3
M8 =
7m
,
3
M3
= 2,
M8
(43)
in contrast to M3 /M8 = 1 in the standard SU(5) model [2].

So, with the above observations, we are now ready to carry out the standard two-loop
analysis (with conversion from the MS scheme to the DR one included) [1,29] for the
gauge (1 , 2 , 3 ) and Yukawa (t , b and ) coupling evolution. Here we are using the
2 /(4)
notation Yt,b,
t,b, for the top- and bottom-quarks and the tau-lepton. We include
the standard supersymmetric threshold corrections at low energies, taken at a single scale
MSUSY = MZ , and those related with the PG states (1) taken also at a sub-TeV scale,
namely at 300 GeV. The heavy threshold corrections due to the H states (39) with mass
6 This two-adjoint superpotential was recently applied to the SU(5) model [28]. Although there can not be a
missing VEV solution in the framework of SU(5), a generic mass-splitting between the adjoint moduli 3 and
8 appears (M3 /M8 = 4) that leads to a natural string-scale unification without any extension of the matter or
Higgs sectors in the model.
39
Fig. 1. The SU(7) predictions of s (MZ ) as a function of the grand unification scale MU . Curves
are shown for small tan values (the dotted line) with top-Yukawa coupling t (MU ) = 0.1 and the
H -state mass threshold M = 1016 GeV, and for large tan values (the solid line) corresponding to
t (MU ) = 0.02, b (MU ) = 0.1 and M = 1016.8 GeV. The unification mass MU varies from the
0 = 1016.3 GeV) to the string scale (M = M = 1017.8 GeV for the
MSSM unification point (MU
str
U
KacMoody level k = 2), while the proton decay inducing colour-triplet mass is assumed to be at the
unification scale in all cases. The all-shaded areas on the left are generally disallowed by the present
bound [4] on nucleon stability for both cases (small tan , dark) and (large tan , light), respectively.
M are taken close to the grand unification scale MU . As to the heavy adjoint moduli,
their masses were treated differently in the two parts of our calculation. When making
predictions of s (MZ ) as a function of MU (see Fig. 1), they were varied from the MSSM
unification point MU0 ' 2 1016 GeV (M3 = MU0 , M8 = 12 MU0 ) down to the intermediate
value m = O(1013) GeV [18], thus pushing MU up to the string scale Mstr . 7 However the
main focus of this paper is on the second part of our calculation the study of string-scale
unification. In the string-scale unification case MU = Mstr (see Tables 2 and 3), the adjoint
moduli masses M3 and M8 can be predicted and they, in fact, turned out to be at their
2/3 1/3
natural scale MP MSUSY stemming from superstrings [18]. The mass splitting between
the weak triplet 3 and the colour octet 8 , which is fixed at the unification scale MU
according to (43), noticeably decreases as M3 and M8 run down from MU to the lower
energies according to their own two-loop RG evolution. This effect was included in the
analysis.
As to the Yukawa coupling evolution, we have considered both the cases of small and
large values of tan . The first case corresponds to t having a large enough value at the
7 It is worth mentioning that, while the SU(7) theory is non-asymptotically free above M , its unified coupling
U
constant U remains perturbative up to Mstr in the allowed regions (see Fig. 1) for the large and small tan cases;
however only marginally so in the latter case.
40
Table 2
The 2-loop order string-scale unification for the small tan case is presented. All masses are given
in GeV. The top quark Yukawa coupling t (MU ) and
H -state mass threshold M (MU ) are taken as
basic string scale input parameters (MU = Mstr = 8U 5.27 1017 for the KacMoody level
1
k = 2), while low scale input parameters include em
= 127.9 0.1, sin2 W = 0.2313 0.0002 and
the top quark pole mass mt = 175 6. The values of s (MZ ), the SU(7) unified coupling constant
0 (M )
U , tan , the bare (uncorrected by gravitational contributions) bottom-tau mass ratio Rb
Z
1
and the adjoint moduli triplet 3 mass M (MU ) (the adjoint octet 8 mass is 2 M (MU )) are then
predicted
t (MU )
M (MU )
M (MU )
s (MZ )
tan
0 (M )
Rb
Z
0.1
0.08
0.04
0.02
0.01
1016
1015.8
1015.4
1015.2
1015
1013.3
1013.1
1012.9
1012.8
1012.8
0.080
0.082
0.087
0.090
0.092
0.120
0.119
0.119
0.119
0.119
1.6
1.6
1.8
2.2
4.8
2.2
2.3
2.4
2.6
2.7
Table 3
The 2-loop order string-scale unification for the large tan case is presented. All masses are given
in GeV. The Yukawa couplings t,b, (MU ) and H -state mass threshold M (MU ) are taken as basic
string scale input parameters (MU = Mstr = 8U 5.27 1017 for the KacMoody level k = 2),
1
while low scale input parameters include em = 127.9 0.1 and sin2 W = 0.2313 0.0002. The
values of s (MZ ), the SU(7) unified coupling constant U , tan , the top quark pole mass mt , the
0 (M ) and the adjoint
bare (uncorrected by SUSY loop contributions) bottom-tau mass ratio Rb
Z
1
moduli triplet 3 mass M (MU ) (the adjoint octet 8 mass is 2 M (MU )) are then predicted
t (MU )
b, (MU )
M (MU )
M (MU )
s (MZ )
mt
tan
0 (M )
Rb
Z
0.025
0.02
0.01
0.01
0.01
0.3
0.1
0.01
0.004
0.001
1017.8
1016.8
1015.2
1015.2
1015.2
1014
1013.5
1012.7
1012.8
1012.9
0.066
0.073
0.089
0.089
0.090
0.120
0.119
0.118
0.119
0.120
176
175
172
175
179
55.1
52.8
38.9
30.3
17.9
1.9
1.9
2.3
2.4
2.6
unification scale MU (0.1 > t (MU ) > 0.01) that it evolves towards its infrared fixed point,
while b (MU ) and (MU ) are significantly smaller (b, (MU ) . 104 ). By requiring the
RG evolved value of t (mt ) to reproduce the observed value of the top quark pole mass,
mt = 175 6 GeV, the values of tan in Table 2 were determined. These values naturally
satisfy the usual RG infrared fixed point bound tan > 1.5, which is just consistent with
the present experimental lower limit 8 on the lightest MSSM Higgs mass [4].
8 The limits on the lightest CP-even Higgs mass coming from the latest LEP2 runs [31] tend to disfavour values
of tan close to the infrared fixed point. They suggest values of tan > 2, from which the first three points in
Table 2 seem to be disfavoured.
41
At the same time the direct b unification prediction at the string scale, Rb (MU ) = 1,
for the bottom quark to tau lepton mass ratio does not work well, when evolved down to
low energies (see Table 2) and compared to the experimental value [4]
exp
Rb (MZ ) = 1.6 0.2.
(44)
However, as we argued in Section 3.6, the SU(5)-like equalities of down quark and
charged lepton masses, including mb (MU ) = m (MU ), are highly unstable under gravity
corrections for small tan in the string-scale SU(7) GUT. So they are not expected to work
well in our model.
The case of large tan , where all the couplings t,b, are relatively large and can
approach a fixed-point, was found to have acceptable solutions over the whole of the
following range of starting values: 0.01 < t (MU ) < 0.03 and 0.001 < b, (MU ) < 0.3.
Here, however, the predictions for the top quark pole mass mt , the ratio Rb (MZ ) and
tan depend on detailed information about the superparticle mass spectrum. Generally, in
the large tan case, large supersymmetric loop contributions to the bottom quark mass mb
are expected, which make the top mass prediction uncertain as well. Since, fortunately, the
direct SUSY loop contributions to the t quark and lepton masses are quite small [32], one
can adopt the following calculational strategy: calculate the value of tan using (m )
from the RG equations and the observed tau lepton mass m = 1.777 GeV and use this
value of tan to obtain the mt value. Further, for the given top mass mt one can calculate
the size of the SUSY loop corrections to the bottom mass mb required to bring the
0 (M ) values given in Table 3 into agreement with experiment
predicted bare Rb
Z
exp
0
(MZ )[1 + ] = Rb (MZ )
Rb (MZ ) = Rb
(45)
exp
Rb (MZ )
above). Interestingly, as one can see from Table 3, the required values of
(see
turn out to be in the range = (0.3)(0.1). The SUSY loop corrections in this range
are readily obtained for most of SUSY parameter space [32] with the values of s (MZ )
and mt in Table 3. So, presently, the bottom-tau unification prediction on its own does not
seem to be a critical test of the SU(7) model for either large or small tan .
Our results, obtained by numerical integration of all the RG equations listed above, are
summarized in Figs. 1, 2 and Tables 2, 3. The predicted s (MZ ) values are in a good
agreement with the world average value (see Fig. 1), in sharp contrast to the standard
SUSY SU(5) model predictions taken under the same conditions. 9 Examples of stringscale unification are presented in Tables 2 and 3 for the small and large tan cases,
respectively. It is worth noting that the large tan examples actually include medium
values of tan down to tan 20, in contrast to the prototype SU(5) model. Remarkably,
string-scale unification appears to work both in the case of a large bottom Yukawa coupling
constant (the first line in Table 3), and in the case of topbottom unification with a small
common Yukawa coupling constant (the third line).
9 It is worth mentioning here that, by introducing non-universal terms in the soft SUSY breaking sector, it
is possible to obtain a small gluino mass and, as a result, an improvement in the unification conditions for the
presently observed value of the QCD coupling constant s (MZ ).
42
Fig. 2. The unification of the gauge coupling constants 1 , 2 and 3 at the string scale is shown.
The breaks on the curves correspond to the heavy thresholds associated with the adjoint moduli 3
and 8 (on the left) and the H states (on the right).
Thus we have seen how the missing VEV generating superpotential WA (6) opens the
way to a natural string-scale grand unification in supersymmetric SU(7), prescribed at low
energies by the gauge coupling values and the standard MSSM particle content plus one
family of PG states (1). This is explicitly demonstrated in Fig. 2.
At the same time, due to the reflection symmetry ( , ) of the
superpotential WA , the Planck scale inherited smearing operators, which induce a
dependence into the kinetic terms of the SM gauge bosons, must have dimension 6 and
higher,
c
(46)
L = 2 Tr(GG 2 ) + .
MP
Here G is the gauge field-strength matrix of the SU(7) model and c is some dimensionless
constant of order 1. Thus, the influence of gravitational corrections on our gauge coupling
predictions seems to be negligible, in contrast to the standard SU(5) model predictions
0
which can largely be smeared out by the dimension-5 operator Mc P Tr(GG) [9].
5. Conclusions
We have shown that a missing VEV vacuum configuration, which solves the doublet
triplet splitting problem and ensures the survival of the MSSM at low energies,
only emerges in extended SU(N) SUSY GUTs with N > 7. Furthermore, a realistic
supersymmetric SU(7) model was constructed which provides answers to many questions
presently facing the prototype SUSY SU(5) model: the doublettriplet splitting problem,
string scale unification, the hierarchy of baryon vs lepton number violation, quark and
lepton (particularly neutrino) masses and mixings, etc.
43
With the chosen assignment of matter and Higgs superfields in our SU(7) model,
the situation at low energies looks as if one had just the prototype SU(5) as a starting
symmetry, except that one family of PG states of type (1) appears when a missing VEV
vacuum configuration develops in SU(7). Apart from this exception, all other extra SU(7)
inherited states in matter and Higgs multiplets acquire GUT scale masses during symmetry
breaking, thus completely decoupling from low-energy physics. Another distinctive feature
of our model concerns the relatively low mass scale of the adjoint moduli 0, 3
and 8 surviving after the SU(7) breaks and, more importantly, their mass-splitting
which inevitably appears in the missing VEV generating superpotential (6). The threshold
corrections due to these states lead to a very different unification picture in SU(7). String
scale unification is obtained with a QCD coupling constant in agreement with the world
average value, s (MZ ) = 0.119 0.002, for both small and large tan .
There is no fundamental reason for exact R-parity (RP) symmetry in the framework
of supersymmetric GUTs, where not only fermions but also their scalar superpartners are
the natural carriers of lepton and baryon numbers. Accordingly, we constructed a general
(RP-violating) superpotential where the effective lepton number violating couplings
immediately evolve from the GUT scale. However the baryon number non-conserving
ones are safely projected out by the missing VEV vacuum configuration which breaks
the SU(7) symmetry down to that of the MSSM. At the next stage when SUSY breaks,
the radiative corrections shift the missing VEV components to some nonzero values of
order MSUSY , thereby inducing the ordinary Higgs doublet mass, on the one hand, and tiny
baryon number violation, on the other. So, a missing VEV solution to the gauge hierarchy
problem leads in fact to a similar hierarchy of baryon vs lepton number violation. Finally,
an additional anomalous U (1)A symmetry introduced into the model purely as a missing
VEV protecting symmetry was found to naturally act also as a family symmetry, which can
lead to the observed pattern of quark and lepton masses and mixings.
To conclude the most crucial prediction of the presented SU(7) model must surely be
the very existence of the PG states and their superpartners of type (1). While at present not
producing any unacceptable experimental predictions, they may include a few very longlived states, depending on the details of their mixing pattern with the ordinary MSSM
Higgs states. The colour-triplet states among them may gather together with ordinary
quarks to give a new series of heavy hadrons (mesons and baryons), the lightest of which
could be almost stable [33]. These PG states would clearly strongly influence particle
phenomenology at the TeV scale.
Acknowledgements
We would like to thank many of our colleagues, especially Riccardo Barbieri, Zurab
Berezhiani, Grahame Blair, Gia Dvali, Mike Green, David Hutchcroft, Gordon Moorhouse,
S. Randjbar-Daemi, Alexei Smirnov and David Sutherland, for stimulating discussions and
useful remarks. Financial support by INTAS Grants No. RFBR 95-567 and 96-155 , and a
Joint Project grant from the Royal Society are also gratefully acknowledged.
44
References
[1] W. de Boer, Prog. Part. Nucl. Phys. 33 (1994) 201.
[2] S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150;
N. Sakai, Z. Phys. C 11 (1981) 153;
S. Weinberg, Phys. Rev. D 26 (1982) 287;
N. Sakai, T. Yanagida, Nucl. Phys. B 197 (1982) 533.
[3] J. Bagger, K. Matchev, D. Pierce, Phys. Lett. B 348 (1995) 443;
P. Langacker, N. Polonsky, Phys. Rev. D 52 (1995) 3081.
[4] Particle Data Group, Eur. Phys. J. C 3 (1998) 1.
[5] H. Hisano, T. Moroi, K. Tobe, T. Yanagida, Mod. Phys. Lett. A 10 (1995) 2267.
[6] H. Murayama, in: Z. Adjuk, A.K. Wroblewski (Eds.), Proceedings of the 28th International
Conference on High Energy Physics, Warsaw, 1996, World Scientific, 1997, p. 1377, hepph/9610419.
[7] Super-Kamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1562.
[8] For a discussion and extensive references, see: K. Dienes, Phys. Rep. 287 (1997) 447.
[9] C.T. Hill, Phys. Lett. B 135 (1984) 47;
L.J. Hall, U. Sarid, Phys. Rev. Lett. 70 (1993) 2673;
S. Urano, D. Ring, R. Arnowitt, Phys. Rev. Lett. 76 (1996) 3663.
[10] E. Witten, Phys. Lett. B 105 (1981) 267.
[11] A. Masiero, D.V. Nanopoulos, K. Tamvakis, T. Yanagida, Phys. Lett. B 115 (1982) 380;
B. Grinstein, Nucl. Phys. B 206 (1982) 387.
[12] S. Dimopoulos, F. Wilczek, in: Erice Summer Lectures, Plenum, New York, 1981;
K.S. Babu, S.M. Barr, Phys. Rev. D 50 (1994) 3259.
[13] K. Inoue, A. Kakuto, H. Takano, Prog. Theor. Phys. 75 (1986) 664;
A. Anselm, A. Johansen, Phys. Lett. B 200 (1988) 331;
Z. Berezhiani, G. Dvali, Sov. Phys. Lebedev Inst. Reports 5 (1989) 55.
[14] J. Hisano, T. Moroi, K. Tobe, T. Yanagida, Phys. Lett. B 342 (1995) 138;
J. Hisano, Progr. Theor. Phys. Suppl. 123 (1996) 301.
[15] J. Chkareuli, Talk given at Trieste Conference on Quarks and Leptons: Masses and Mixings,
Trieste, 1996, hep-ph/9706280;
J.L. Chkareuli, A.B. Kobakhidze, Phys. Lett. B 407 (1997) 234;
J.L. Chkareuli, I.G. Gogoladze, A.B. Kobakhidze, Phys. Rev. Lett. 80 (1998) 912;
J.L. Chkareuli, in: A. Astbury, D. Axen, J. Robinson (Eds.), Proceedings of 29th International
Conference on High Energy Physics, Vancouver, 1998, World Scientific, 1999, Vol. 2, p. 1669,
hep-ph/9809464.
[16] S. Dimopoulos, F. Wilczek, see Ref. [12].
[17] M. Green, J. Schwarz, Phys. Lett. B 149 (1998) 117.
[18] C. Bachas, C. Fabre, T. Yanagida, Phys. Lett. B 370 (1996) 49;
M. Bastero-Gil, B. Brahmachari, Phys. Lett. B 403 (1997) 51.
[19] M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 587.
[20] Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 396 (1997) 150.
[21] J.L. Chkareuli, C.D. Froggatt, Phys. Lett. B 484 (2000) 87.
[22] J.L. Chkareuli, I.G. Gogoladze, M.G. Green, D.E. Hutchcroft, A.B. Kobakhidze, Phys. Rev.
D 62 (2000) 015014.
[23] H.K. Dreiner, in: G.L. Kane (Ed.), Perspectives in Supersymmetry, World Scientific, 1998,
p. 462, hep-ph/9707435;
G. Bhattacharyya, hep-ph/9709395;
R. Barbier et al., hep-ph/9810232.
[24] M. Gell-Mann, P. Ramond, R. Slansky, in: P. von Nienvenhuizen, D.Z. Friedman (Eds.),
Supergravity, North-Holland, 1979;
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
45
T. Yanagida, in: A. Sawada, H. Sugawara (Eds.), Proc. Workshop on Unified Theory and Baryon
Number in the Universe, KEK, Tsukuba-Gu, Ibaraki-ken, 1979;
C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 164 (1979) 114;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
L.E. Ibanez, G.G. Ross, Phys. Lett. B 332 (1994) 100;
J.K. Elwood, N. Irges, P. Ramond, Phys. Lett. B 413 (1997) 322;
N. Irges, S. Lavignac, P. Ramond, Phys. Rev. D 58 (1998) 035003.
C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
Q. Shafi, Z. Tavartkiladze, Phys. Lett. B 451 (1999) 129.
J.L. Chkareuli, I.G. Gogoladze, Phys. Rev. D 58 (1998) 055011.
V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 47 (1993) 1093;
P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2282.
J. Ellis, M.K. Gaillard, Phys. Lett. B 88 (1979) 315.
A. Sopczak, hep-ph/0004015.
L.J. Hall, R. Rattazzi, U. Sarid, Phys. Rev. D 50 (1994) 7048;
M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 426 (1994) 269.
R. Barbieri, G. Dvali, A. Strumia, Nucl. Phys. B 391 (1993) 487.

Evolution of truncated moments

of singlet parton distributions
Stefano Forte a,,1 , Lorenzo Magnea b , Andrea Piccione c,d ,
Giovanni Ridolfi d
a INFN, Sezione di Roma Tre, Via della Vasca Navale 84, I-00146 Rome, Italy
b Dipartimento di Fisica Teorica, Universit di Torino and INFN, Sezione di Torino,
Via P. Giuria 1, I-10125 Torino, Italy

c Dipartimento di Fisica, Universit di Genova, Via Dodecaneso 33, I-16146 Genova, Italy
d INFN, Sezione di Genova, Via Dodecaneso 33, I-16146 Genova, Italy
Received 4 July 2000; accepted 2 November 2000
Abstract
We define truncated Mellin moments of parton distributions by restricting the integration range
over the Bjorken variable to the experimentally accessible subset x0 6 x 6 1 of the allowed kinematic
range 0 6 x 6 1. We derive the evolution equations satisfied by truncated moments in the general
(singlet) case in terms of an infinite triangular matrix of anomalous dimensions which couple each
truncated moment to all higher moments with orders differing by integers. We show that the evolution
of any moment can be determined to arbitrarily good accuracy by truncating the system of coupled
moments to a sufficiently large but finite size, and show how the equations can be solved in a
way suitable for numerical applications. We discuss in detail the accuracy of the method in view
of applications to precision phenomenology. 2001 Elsevier Science B.V. All rights reserved.
1. The use of truncated moments

The needs of accurate phenomenology at current and future hadron colliders have
recently led to the development of more refined tools for the QCD analysis of collider
processes [1]. A detailed understanding of parton distributions and their scaling violations
is in particular an essential ingredient of such phenomenology [1,2]. Scaling violations
of parton distributions are described by renormalization group equations for matrix
elements of leading twist operators, whose anomalous dimensions are currently fully
known to next-to-leading order [3], and to next-to-next-to-leading order for a handful of
E-mail address: forte@fis.uniroma3.it (S. Forte).

1 On leave from INFN, Sezione di Torino, Italy.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 7 0 - 2
S. Forte et al. / Nuclear Physics B 594 (2001) 4670
47
operators [4]. Moments of parton distributions are related to moments of deep-inelastic

structure functions by Wilson coefficients, which have also been computed up to next-tonext-to-leading order [5]. Moments of structure functions, however, cannot be measured
even indirectly, since they are defined as integrals over all kinematically allowed values
of x, and thus require knowledge of the structure function for arbitrarily small x, i.e.,
arbitrarily large energy.
There is of course a well-known solution to this problem, which consists of using the
AltarelliParisi equation [6] to evolve parton distributions directly: the scale dependence of
any parton distribution at x0 is then determined by knowledge of parton distributions for all
x > x0 , i.e., parton evolution is causal. In fact, through a judicious choice of factorization
scheme [8,10] all parton distributions can be identified with physical observables, and it
is then possible to use the AltarelliParisi equations to express the scaling violations of
structure functions entirely in terms of physically observable quantities. It is, however,
hard to measure local scaling violations of structure functions in all the relevant processes:
in practice, a detailed comparison with the data requires the solution of the evolution
equations.
What is usually done instead is to introduce a parametrization of parton distributions,
and then solve the evolution equations in terms of this parametrization. The idea is that
a parametrization fitted to the data will reproduce them in the experimentally accessible
region, but it will also provide an extrapolation, so that the evolution equations can
be solved easily, for instance taking Mellin moments. The results in the measured
region should be independent of this extrapolation since, by the AltarelliParisi equation,
measured scaling violations are independent of it. It has however become increasingly
clear that in practice this procedure introduces a potentially large theoretical bias, whose
exact size is very hard to assess [2,9]. First, the very fact of adopting a specific functional
form constrains not only the extrapolation in the unmeasured region, but also the allowed
behavior at the boundary of the measured region. Especially when data are not very precise,
it can be seen explicitly [7] that rather different results are obtained simply by changing
the functional form used to parametrize parton distributions. Furthermore, it is very hard to
assess the uncertainty on the best-fit functional form of the parton distributions, essentially
because of the very nonlinear and indirect relation between the data and the quantity which
is parametrized. Hence, the need to go through such a parametrization makes it very hard
to assess the uncertainty on the desired result.
Various methods to overcome these problem have been discussed in the literature.
One possibility is to minimize the bias introduced by the parton parametrization, by
projecting parton distributions on an optimized basis of functions, such as suitable families
of orthogonal polynomials [11]. A more ambitious proposal is to construct the probability
functional for parton distributions through Bayesian inference applied on a Monte Carlo
sampling of the relevant space of functions [9]. The probability functional then summarizes
in an unbiased way all the available experimental information, and can be used to determine
the mean value and error on any physical observable. For many applications, however, there
is a simpler, although more limited option, which consists of dealing directly with the
experimentally accessible quantities. Indeed, consider two typical problems in the study
48
of scaling violations: the determination of s , and the determination of a moment of the

gluon distribution, for instance the first moment of the polarized gluon distribution [7],
which gives the gluon spin fraction. Manifestly, in the latter case only the contribution
to the moment from the measured region x0 6 x 6 1 is accessible experimentally, and
it would be useful to be able to separate in a clean way the measured quantity from the
extrapolation to the unmeasured region, which is necessarily based on assumptions. It is
then natural to study the scaling violation of these measurable contributions to Mellin
moments, i.e., truncated moments. Similarly, s can be determined from the evolution
equation of truncated moments [12], without any reference to the behavior of the structure
function in the region in which it is not measured, and in principle without the need to
resort to a specific functional parametrization.
It turns out [12] that scaling violations of truncated moments are described by a
triangular matrix of anomalous dimensions which couple the nth truncated moment to
all truncated moments of order n + k, where k runs over positive integers. This means
that truncated moments share with full moments the property that evolution equations
are ordinary first-order differential equations, and not integro-differential equations, as for
the parton distributions themselves. Unlike full moments, however, truncated moments
can be measured without extrapolations. The price to pay for this is that their evolution
equations do not decouple (unlike the case of full moments); however, the causal nature of
the evolution implies that the evolution of each moment is only affected by higher order
moments. Furthermore, the series of couplings to higher moments converges, and thus it
can be truncated to any desired accuracy. The problem then reduces to the solution of
a system of ordinary differential equations coupled by a triangular matrix, whose initial
conditions are (measurable) truncated moments.
This gives a simple solution to both problems mentioned above: s and the truncated
moment of the gluon can be determined directly from the observed scaling violations,
without having to go through an intermediate parton parametrization. A model independent
error analysis can then be performed, provided only that data for the truncated moments
and their errors are available. Of course, these could be extracted directly by summing
over experimental bins, if a sufficiently abundant data set were available. In practice,
however, it is more convenient to manipulate a smooth interpolation of the data. It turns
out to be possible to do this without invoking an explicit functional parametrization for
the measured structure functions, by constructing neural networks which are trained to
simulate all available experimental information, including statistical and systematic errors
and correlations. By means of these neural networks it is then easy to compute the
observed truncated moments and their errors, which can be further used to perform a
phenomenological analysis of scaling violations, free of theoretical bias. The application
of the method of neural networks to the parametrization of structure functions is currently
under way and will be presented in a separate publication.
Evolution equations for truncated moments were presented in Ref. [12] in the simplest
(nonsinglet) case, along with a preliminary study of the viability of the method. It is the
purpose of this work to present a full treatment of the method of truncated moments,
suitable for future phenomenological applications. In particular, in Section 2 we will
49
derive evolution equations for truncated moments in the general (singlet) case, and we
will discuss their solution in a form which is suitable for numerical implementation. In
Section 3 we will consider numerical solutions of the evolution equations with typical
quark and gluon distributions, and discuss the accuracy of the truncation of the infinite
system of coupled evolution equations. Problems and techniques of interest for future
phenomenological applications are discussed in Section 4. The appendices collect various
technical results which are needed for the actual implementation of the methods discussed
here in an analysis code: in particular, we give explicit expressions for the NLO singlet
splitting functions in the DIS scheme, and we list all the integrals needed to compute their
truncated moments.
2. Evolution equations for truncated moments and their solutions

The Q2 dependence of parton distributions q(x, Q2 ) is governed by the AltarelliParisi
(AP) equations [6]
S (Q2 )
d
q x, Q2 =
dt
2
Z1

x
dy
P
; S Q2 q y, Q2 ,
y
y
(2.1)
In the non-singlet case, q is simply one of the flavour non-singlet

where t =
combinations of quark distributions, and P the corresponding splitting function. In the
singlet case, q is a vector, whose components are the flavour-singlet combination of quark
distributions,
log Q2 /2 .
x, Q
nf
X
qi x, Q2
(2.2)
i=1
and the gluon distribution g(x, Q2 ). Correspondingly, P is a 2 2 matrix of splitting

functions, given as an expansion in powers of S .
As is well known, upon taking ordinary Mellin moments convolutions turn into ordinary
products and evolution equations become ordinary first-order differential equations. By
contrast, we are interested in the evolution of truncated moments, defined for a generic
function f (x) by
Z1
fn (x0 ) =
dx x n1 f (x).
(2.3)
x0
The corresponding evolution equations in the nonsinglet case were derived in Ref. [12],
which we will follow in presenting the generalization to the singlet case. One finds
immediately that the truncated moments of q(x, Q2 ) obey the equation
S (Q2 )
d
qn x0 , Q2 =
dt
2
Z1
dy y
x0
n1
Gn

x0
q y, Q2 ,
y
(2.4)
50
where
Z1
Gn (x) =
dz zn1 P (z)
(2.5)
is perturbatively calculable as a power series in s .

Expanding Gn (x0 /y) in powers of y around y = 1,
p
X

gpn (x0 )
x0
x0
(y 1)p ,
gpn (x0 ) =
G
,
=
Gn
n
p
y
p!
y
y
y=1
(2.6)
p=0
one obtains
p
X
(1)k+p gpn (x0 )
S (Q2 ) X

d
2
qn x0 , Q =
qn+k x0 , Q2 .
dt
2
k!(p k)!
(2.7)
p=0 k=0
The key step in the derivation of Eq. (2.7) is the term-by-term integration of the series
expansion. This is allowed, despite the fact that the radius of convergence of the series in
Eq. (2.6) is 1 x0 , because the singularity of Gn (x0 /y) at y = x0 is integrable (this can
be proven [13] using the Lebesgue definition of the integral). One can then express each
power of (y 1) using the binomial expansion, which leads to Eq. (2.7).
Eq. (2.7) expresses the fact that, while full moments of parton distributions evolve
independently of each other, truncated moments obey a system of coupled evolution
equations. In particular, the evolution of the nth moment is determined by all the
moments qj , with j > n. In practice, the expansion in Eq. (2.6), because of its convergence,
can be truncated to a finite order p = M. The error associated with this procedure will be
discussed in Section 3. In this case, Eq. (2.7) can be rewritten as
S (Q2 ) X (M)

d
qn x0 , Q2 =
cnk (x0 ) qn+k x0 , Q2 ,
dt
2
M
(2.8)
k=0
where
(M)
cnk (x0 ) =
M
X
(1)p+k gpn (x0 )
p=k
k!(p k)!
(2.9)
To solve the system of equations (2.8), it is necessary to include a decreasing number

of terms (M, M 1, and so on) in the evolution equations for higher moments
(n + 1, n + 2, . . .), obtaining M + 1 equations for the M + 1 truncated moments
{qn , . . . , qn+M }. We will see in the next section that this approximation is fully justified. In
this case, the coupled system of evolution equations takes the form
n+M
X
d
qk =
Ckl ql ,
d
n 6 k 6 n + M,
(2.10)
S (Q2 )
,
2
(2.11)
l=n
with
Zt
=
t0
dt 0 a(t 0 ),
a(t) =
where C is now a triangular matrix:

(Mk+n)
(l > k),
Ckl = ck,lk
(l < k).
Ckl = 0
51
(2.12)
In the nonsinglet case, discussed in Ref. [12], the matrix elements Ckl are just numbers,
and the matrix C in Eq. (2.12) is triangular, which makes it easy to solve Eq. (2.10)
perturbatively. In the singlet case, all the steps leading to Eq. (2.12) are formally the
same, but now each entry Ckl is a 2 2 matrix. As a consequence, the matrix C, which
is given in terms of partial moments of the evolution kernels, is no longer triangular, but
has nonvanishing 2 2 blocks along the diagonal. This problem can be circumvented, by
writing the perturbative expansion of C as
C = C0 + aC1 + = (A0 + B0 ) + a(A1 + B1 ) + ,
(2.13)
where A = A0 + aA1 is block-diagonal, with 2 2 blocks on its diagonal,

Akl = Ckk kl ,
(2.14)
while B = B0 + aB1 , considered as a matrix of 2 2 blocks, is upper-triangular with

vanishing diagonal entries. Now one can define a matrix S that diagonalizes A0 ,
SA0 S 1 = diag(1 , . . . , 2M ).
(2.15)
Clearly, S is -independent, block-diagonal, and easily computed. Eq. (2.10) can then be
rewritten as
d
q = T q,
(2.16)
where q = S q and T = SCS 1 .

The new evolution matrix T is triangular at leading order (with the same eigenvalues
as A0 ). This is enough to solve the evolution equation to next-to-leading order, as in the
nonsinglet case of Ref. [12]. The general solution is worked out in detail in Appendix A;
the result is
q(
) = U (T , ) q(0),
(2.17)
where
Uij (T , )

bkl b1 l kl
a(0) l /b0
a(0) l /b0 T
a(0) k /b0
1
1
+
a( )
a(0)
Rlj .
= Rik kl
a( )
k l + b 0
a( )
a( )
(2.18)
In Eqs. (2.17), (2.18), T is expanded as T = T0 + aT1 ; R is the matrix which
b1 = RT1 R 1 .
diagonalizes T0 , RT0 R 1 = diag(1 , . . . , 2M ); finally, T
The matrix R can be computed recursively, using the technique applied in Ref. [12] and
proven in Appendix B. One finds
52
Rij =
j 1
1 X
pj
Rip T0 ,
i j
(2.19)
p=i
Rij1 =
j
X
1
ip 1
T0 Rpj
,
j i
(2.20)
p=i+1
which, together with the conditions Rii = 1 and Rij = 0 when i > j , determine the matrix
R completely.
The general solution for the parton distributions is then
q( ) = U (C, ) q(0),
(2.21)
where
U (C, ) = S 1 U (T , )S.
(2.22)
The splitting functions and partial moment integrals which should be used in Eq. (2.18) in
order compute this solution explicitly are listed in Appendices C and D.
For the sake of completeness, we describe a different method to solve Eq. (2.8). It is
immediate to check that the matrix
Zn1
Z
X
d1
dn C(1 ) C(n )
(2.23)
U (C, ) = I +
n=1 0
obeys the differential equation

d
U (C, ) = CU (C, ),
(2.24)
d
with the initial condition U (C, 0) = I . In general, Eq. (2.23) is not very useful, since it
involves an infinite sum. In the present case, however, the infinite sum collapses to a finite
sum. To see this, consider again the decomposition C = A + B, where A is block-diagonal
and B is upper-triangular. It is easy to prove that
e ),
U (C, ) = U (A, )U (B,
(2.25)
where
e = U 1 (A, )BU (A, ).
B
(2.26)
Since A is block diagonal, U (A, ) is also block-diagonal, and it can be computed

perturbatively using the procedure described in Appendix A. Furthermore, once U (A) is
e can be computed through Eq. (2.26). Now one can
known, the upper-triangular matrix B
use the fact that upper-triangular matrices have the property that their Mth power vanishes.
Hence, the solution can be expressed as the finite sum
e ) = I +
U (B,
M1
XZ
n=1 0
Zn1
e n ),
e 1 ) B(
dn B(
d1
(2.27)
e and U (A) one can determine the solution to the evolution

and from the knowledge of U (B)
equations explicitly.
53
3. Numerical methods and their accuracy

In this section we will assess the accuracy of our method when the series of contributions
to the right-hand side of the evolution equation (2.7) is approximated by retaining a
finite number M of terms. The loss of accuracy due to this truncation is the price to
pay for eliminating the dependence on parton parametrizations and extrapolations in the
unmeasured region. However, unlike the latter uncertainties, which are difficult to estimate,
the truncation uncertainty can be simply assessed by studying the convergence of the
series. A reasonable goal, suitable for state-of-the-art phenomenology, is to reproduce the
evolution equations to about 5% accuracy: indeed, we expect the uncertainties related to
the parametrization of parton distributions in the conventional approach to be somewhat
larger ( 10%). 2 Notice that there is no obstacle to achieve a higher level of precision
when necessary, by simply including more terms in the relevant expansions. To this level
of accuracy it is enough to study the behavior of the leading order contribution to the
evolution equation: indeed, next-to-leading corrections to the anomalous dimension are
themselves of order 10%. We have verified explicitly that the inclusion of the next-toleading corrections does not affect our conclusions.
We can compare the exact evolution equation (2.7) with its approximate form, Eq. (2.8),
by defining the percentage error on the right-hand side of the evolution equations for the
quark nonsinglet, singlet and gluon:
RNS
n,M
R
n,M
1
=
NNS
1
=
N
Z1

X

M

x0
k NS
NS
dy y n1 GNS
y
c
y, Q2 ,
n
nk q
y
x0
("
Z1
dy y
qq
Gn
n1
x0
1
=
Ng
x0
y
"
+
g
Rn,M
(3.1)
k=0

qg
Gn
("
Z1
dy y
n1

gq
Gn
x0
y
"
x0
gg
Gn
M
X
#
qq
y k cnk
y, Q2
k=0
x0
y
M
X
#
qg
y k cnk
k=0
M
X
gq
y k cnk
x0
y
M
X
g y, Q
#
y, Q2
k=0
(3.2)
#
gg
y k cnk
)
g y, Q
)
,
(3.3)
k=0
where NNS,,g are the exact right-hand sides of the evolution equation (2.7). We study
the dependence of the percentage error on the value of M for typical values of the cutoff
x0 and for representative choices of test parton distributions. In particular, we parametrize
parton distributions as
2 Notice that this is not the uncertainty associated with evolution of a given parametrization with, say, an
x-space code; rather, it is the uncertainty associated with the choice of the parametrization, and with the bias it
introduces in the shape of the distribution.
54

q x, Q2 = Nx (1 x) .
(3.4)
We begin by choosing, as a representative case, = 4 and = 1 for the singlet

distributions and = 0 for the nonsinglet. The nonsinglet is assumed to behave
qualitatively as qNS xg x, in accordance with the behavior of the respective splitting
functions. Furthermore, the normalization factors N for the singlet and gluon are fixed by
requiring that the second moments of (x, Q2 ) and g(x, Q2 ) are in the ratio 0.6/0.4,
which is the approximate relative size of the quark and gluon momentum fractions at a
scale of a few GeV2 . We will then show that changing the values of and within a
physically reasonable range does not affect the qualitative features of our results.
The accuracy of the truncation of the evolution equation is determined by the
convergence of the expansion in Eq. (2.6). Because this expansion is centered at y = 1,
and diverges at y = x0 , the small y region of the integration range in Eq. (2.4) is poorly
reproduced by the expansion. Hence, even though the series in Eq. (2.7) converges, as
discussed in Section 2, the convergence will be slower for low moments, which receive
a larger contribution from the region of integration y x0 . In fact, for low enough values
of n, the convolution integral on the right-hand side of the evolution equation (2.4) does not
exist: this happens for the same value for which the full moment of the structure function
does not exist, i.e., n 6 1 in the unpolarized singlet and n 6 0 in the unpolarized nonsinglet
and in the polarized case. Therefore, we concentrate on the lowest existing integer moments
of unpolarized distributions, i.e., the cases n = 2, 3 for the singlet distributions, and
correspondingly n = 1, 2 for the nonsinglet, which are the cases in which the accuracy
of the truncation will be worse.
NS,,g
The values of Rn,M , computed at leading order with x0 = 0.1, are shown in Table 1.
The table shows that nonsinglet moments of order n behave as singlet moments of order
n 1. This is a consequence of the fact that, as discussed above, the convergence of the
expansion is determined by the singularity of the integrand Gn (x0 /y)q(y) of Eq. (2.4)
as y x0 ; near y = x0 , the function Gn (x0 /y) is well approximated by the singular
contribution log(1 x0 /y), while parton distributions carry an extra power of y 1 in the
singlet case in comparison to the nonsinglet. We also observe in Table 1 that, as expected,
Table 1
NS,,g
Values of Rn,M
for x0 = 0.1 and different values of n and M
M
RNS
1,M
R
2,M
R2,M
RNS
2,M
R
3,M
R3,M
0.12
0.10
0.08
0.05
0.03
0.02
0.01
0.16
0.12
0.08
0.05
0.03
0.02
0.01
x0 = 0.1
5
10
20
40
70
100
150
0.63
0.49
0.34
0.20
0.12
0.09
0.06
0.43
0.36
0.27
0.17
0.10
0.07
0.05
0.55
0.38
0.26
0.17
0.10
0.07
0.05
0.16
0.13
0.10
0.06
0.04
0.03
0.02
55
the convergence is slower for the lowest moments, and rapidly improves as the order
of the moment increases. This rapid improvement is a consequence of the fact that the
convergence of the expansion of G(x0 /y) is only slow in the immediate vicinity of the
point y = x0 , and the contribution of this region to the nth moment is suppressed by a
factor of x0n1 . Due to this fast improvement, the approximation introduced by including
one less term in the expansion as the order of the moment is increased by one, which is
necessary to obtain the closed system of evolution equations (2.10), is certainly justified.
The 5% accuracy goal which we set to ourselves requires the inclusion of more than
100 terms for the lowest moment, but only about 40 terms for the next-to-lowest. The
computation of series with such a large number of contributions does not present any
problem, since the splitting functions are known and their truncated moments are easily
determined numerically. The implications of this requirement for phenomenology will be
discussed in the next section.
We can now study the dependence of these results on the value of the truncation point
x0 by plotting the exact and approximate right-hand side of the evolution equations as a
function of x0 , as shown in Fig. 1. The figures show that the case x0 = 0.1 studied in Table 1
is a generic one between the two limiting (and physically uninteresting) cases x0 = 0 and
x0 = 1, where the approximation is exact. In fact, with this particular choice of parton
distributions, x0 = 0.1 is essentially a worst case and the error estimates of Table 1 are
therefore conservative.
An interesting feature of these plots is the presence of zeroes of the lowest moment
evolution at x0 = 0 in the nonsinglet and around x0 102 in the gluon case. The physical
origin of these zeroes is clear. At leading order, the first nonsinglet full moment does not
evolve. On the other hand, the second gluon full moment grows with Q2 , while higher
gluon full moments decrease, i.e., the gluon distribution decreases at large x; this implies
that the second truncated moment of the gluon must decrease for a high enough value of
the cutoff x0 , while it must increase for very small x0 ; its derivative is thus bound to vanish
at some intermediate point. Of course, the phenomenology of scaling violations (such as
a determination of s ) cannot be performed at or close to these zeroes, where there is no
evolution. From the point of view of a truncated moment analysis, this means that the value
of x0 should be chosen with care in order to avoid these regions.
Finally, in Table 2 we study the dependence of our results on the form of the
parton distributions, by varying the parameters and within a reasonable range. Of
course, parton distributions which are more concentrated at small y give rise to slower
convergence. However, we can safely conclude that the effect of varying the shape of
parton distributions is generally rather small. We have also verified that varying the
relative normalization of the quark and gluon distributions has a negligible effect on the
convergence of the series, even though it may change by a moderate amount the position
of the zeroes in gluon evolution discussed above.
56
Fig. 1. Right-hand sides of the evolution equations for the first and second truncated moments of the
nonsinglet distribution, and for the second and third moments of singlet distributions. The overall
scale is set by s (2 GeV2 ).
4. Techniques for phenomenological applications

So far we have discussed scaling violations of parton distributions. In a generic factorization scheme, the measured structure functions are convolutions of parton distributions
and coefficient functions. When taking moments, convolutions turn into ordinary products and moments of coefficient functions are identified with Wilson coefficients. In the
57
Table 2
g
Values of R
n,M and Rn,M for x0 = 0.1 and different choices of parton distribution parameters
R
n,20
R
n,70
Rn,20
Rn,70
0.27
0.23
0.18
0.30
0.26
0.22
0.32
0.29
0.26
0.11
0.09
0.07
0.12
0.10
0.09
0.14
0.12
0.10
0.07
0.04
0.02
0.11
0.08
0.05
0.15
0.12
0.08
0.03
0.02
0.01
0.05
0.03
0.02
0.06
0.05
0.03
n = 2, x0 = 0.1
1.5
1.0
0.5
1.5
1.0
0.5
1.5
1.0
0.5
2.0
2.0
2.0
4.0
4.0
4.0
6.0
6.0
6.0
0.26
0.20
0.14
0.32
0.27
0.22
0.36
0.32
0.27
0.10
0.07
0.05
0.12
0.10
0.08
0.14
0.12
0.10
1.5
1.0
0.5
1.5
1.0
0.5
1.5
1.0
0.5
2.0
2.0
2.0
4.0
4.0
4.0
6.0
6.0
6.0
0.07
0.04
0.02
0.12
0.08
0.05
0.16
0.12
0.09
0.03
0.02
0.01
0.05
0.03
0.02
0.07
0.05
0.03
n = 3, x0 = 0.1
present case, however, as shown in Eqs. (2.4)(2.7), truncated moments turn convolutions
into products of triangular matrices. Hence, in a generic factorization scheme, truncated
moments of parton distributions are related to truncated moments of structure functions by
a further triangular matrix of truncated moments of coefficient functions.
This complication can be avoided by working in a parton scheme [10], where the
quark distribution is identified with the structure function F2 . This still does not fix
the factorization scheme completely in the gluon sector. One way of fixing it is to
use a physical scheme, where all parton distributions are identified with physical
observables [8]. This may eventually prove the most convenient choice for the sake of
precision phenomenology, once accurate data on all the relevant physical observables are
available. At present, however, the gluon distribution is mostly determined from scaling
violations of F2 , so within the parton family of schemes the choice of gluon factorization is
immaterial. Here we will fix the scheme by assuming that all moments satisfy the relations
between the parton-scheme gluon and the MS quark and gluons imposed by momentum
conservation on second moments [14]. This is the prescription used in common parton sets,
and usually referred to as DIS scheme. Explicit expressions of AltarelliParisi kernels in
the DIS scheme are given in Appendix C.
With this prescription, the phenomenology of scaling violations can be studied by
58
computing a sufficiently large number of truncated moments of structure functions, so as

to guarantee the required accuracy. If the aim is, for instance, a determination of s from
nonsinglet scaling violations, all we need is a large enough number of truncated moments
of the nonsinglet structure function. Once an interpolation of the data in the measured
region is available, the determination of such truncated moments is straightforward. This
interpolation can be performed in an unbiased way using neural networks, as already
mentioned in the introduction. One may wonder, however, whether the need to use the
values of very high moments would not be a problem. Indeed, very high moments depend
strongly on the behavior of the structure function at large y, which is experimentally known
very poorly. Furthermore, it seems contradictory that scaling violations of the lowest
moments should be most dependent on the structure function at large y.
This dependence is only apparent, however. Indeed, Eq. (2.7) for, say, the nonsinglet first
moment can be rewritten as
gp1 (x0 ) p
s (Q2 ) X

d
q1 x0 , Q2 =
q1 x0 , Q2 ,
dt
2
p!
(4.1)
p=0
where
p
qk
x0 , Q
Z1
=

dy y k1 (y 1)p q y, Q2 .
(4.2)
x0
The need to include high orders in the expansion in Eq. (2.8) is due to the slow
convergence of the series in Eq. (4.1), in turn determined by the fact that Gn (x0 /y)
diverges logarithmically at y = x0 . Correspondingly, the right-hand side of the evolution
p
equation depends significantly on q1 for large values of p, which signals a sensitivity to
the value of q(y, Q2) in the neighborhood of the point y = x0 . The dependence on high
p
truncated moments qn is introduced when q1 is re-expressed in terms of qn , by expanding
p
the binomial series for (y 1) . Since this re-expansion is exact, it cannot introduce a
dependence on the large y region which is not there in the original expression. The high
orders of the expansion do instead introduce a significant dependence on the value of the
structure function in the neighborhood of x0 , which can be kept under control provided x0
is not too small, i.e., well into the measured region. There is therefore no obstacle even in
practice in performing an accurate determination of s from scaling violations of truncated
moments.
Let us now consider a second typical application of our method, namely the determination of truncated moments of the gluon distribution. In particular, the physically interesting
case is the lowest integer moment, i.e., the momentum fraction in the unpolarized case or
the spin fraction in the polarized case. The need to include a large number of terms in the
expansion of the evolution equations seems to imply the need to introduce an equally large
number of parameters, one for each gluon truncated moment. This would be problematic
since it is appears unrealistic to fit a very large number of parameters of the gluon from
currently available data on scaling violations. We may, however, take advantage of the fact
that the dependence on high order truncated moments is fictitious, as we have just seen,
59
and it rather indicates an enhanced sensitivity to the value of q(y) as y x0 . This suggests
that a natural set of parameters to describe the gluon distribution should include the first
several truncated moments, as well as further information on the behavior of the distribution around the truncation point x0 , such as the value of the distribution (and possibly of
some of its derivatives) at the point x0 .
To understand how such a parametrization might work, notice that if q(y) is regular
around y = x0 , then it is easy to prove that
R1
p
x0 dy(y 1) q(y)
= 1,
(4.3)
lim
R
p q(x ) 1 dy(y 1)p
0 x0
by Taylor expanding q(y) about y = x0 . We may therefore approximate the series which
appears on the r.h.s. of Eq. (4.1) by
S(x0 , n0 ) =
nX
0 1
p=0
gp1 (x0 )
Z1
dy(y 1)p q(y)
p!
x0
X
gp1 (x0 )
p=n0
p!
Z1
dy(y 1)p .
q(x0 )
(4.4)
x0
Eq. (4.4) describes the evolution of the first truncated moment of q(y) in terms of the
first n0 truncated moments and of the value of q(y) at the truncation point x0 . Of course,
the approximation gets better if n0 increases. It is easy to check that when x0 = 0.1 the
accuracy is already better than 10% when n0 7. This means that a parametrization
of the distribution in terms of less than ten parameters is fully adequate. It is easy to
convince oneself that this estimate is reliable, and essentially independent of the shape
of the distribution q(y). In fact, because slow convergence arises due to the logarithmic
singularity in Gn (x0 /y), we can estimate the error of the approximation in Eq. (4.4)
by replacing the functions gp1 (x0 )/p! with the coefficients of the Taylor expansion of
log(1 x0 /y) in powers of y 1, which we may denote by gp1 (x0 )/p!. The error is then

X g 1 (x0 ) Z1

p

p
dy (y 1) q(y) q(x0)

p!
p=n
0
x0
Z1
6
x0

nX
0 1 1

g p (x0 )

p
(y 1) q(y) q(x0).
dy log(1 x0 /y)

p!
(4.5)
p=0
The expression inside the first absolute value on the r.h.s. of Eq. (4.5) is just the error
made in approximating the logarithm with its Taylor expansion around y = 1; thus, it is
a slowly decreasing function of n0 , it is integrable, and the integral receives the largest
contribution from the region y x0 ; the second absolute value, on the other hand, is a
bounded function of y in the range x0 6 y 6 1, which vanishes as y x0 for any choice
of q(y). These two facts combine to limit the size of the error. One can check directly that,
60
choosing for example q(y) = (1 y)4 as in the previous section, the accuracy is better
than 10% with n0 10 and x0 = 0.1, in agreement with the previous estimate. One may
also verify that, as expected, changing the shape of q(y) does not significantly affect the
result.
We conclude that there is no difficulty in using the evolution of truncated moments
for a direct extraction of the lowest truncated moment of the gluon distribution, provided
only the higher moments of the distributions, which are auxiliary quantities needed in the
extraction, are parametrized in an effective way. We have seen that this is possible with a
reasonably small number of parameters.
5. Outlook
In this paper, we have discussed the solution of the AltarelliParisi evolution equations
of perturbative QCD by projecting parton distributions on a basis of truncated moments.
We have seen that truncated moments give us a compromise between standard Mellin
moments, which satisfy simple linear evolution equations, but are not measurable, and
parton distributions themselves, which are measurable, but satisfy integro-differential
evolution equations. In this respect, projecting on a basis of truncated moments is akin
to projecting on a basis of orthogonal polynomials [11], but with the added advantage
that truncated moments are physical observables. We have further shown that evolution
equations for truncated moments can be solved to arbitrarily high accuracy by using a
sufficiently large basis of truncated moments, and we have discussed how this formalism
can be exploited to perform a model-independent, unbiased analysis of scaling violations,
which does not rely on parton parametrizations. We have also collected all technical tools
which are needed for this analysis, and discussed the reliability of the approximations
which are necessary in order to implement it in practice. Such an analysis could be used for
the determination of s and for the direct measurement of the contribution to the moments
of the gluon distribution from the experimentally accessible region. In both cases, it would
offer the advantage of allowing reliable estimates of the uncertainty on the result. In order
to be effective, these phenomenological applications will need an unbiased interpolation of
the data, such as could be achieved by means of neural networks. Phenomenological studies
along these lines are currently under investigation and will be presented in forthcoming
publications.
Acknowledgements
We thank L. Garrido and J.I. Latorre for discussions on neural networks and their
applications to structure functions. We thank R. Ball for a careful reading of the manuscript
and several useful comments.
61
Appendix A. Solution of matrix evolution equations at NLO

In this appendix, we find the general solution of the equation
d
q = Cq,
(A.1)
d
where q is a vector with M components, and C is a generic M M matrix. The usual QCD
evolution equations are special cases of this equation in which M 6 2. We will assume that
C has a perturbative expansion in powers of a parameter a( ):
C = C0 + a( )C1 + ,
(A.2)
with
da( )
= b0 a(1 + b1 a + ).
d
For QCD applications
S
,
a=
2
1
=
2
Zt
(A.3)
dt 0 S (t 0 ),
(A.4)
t0
with t = log(Q2 /2QCD ), and

33 2nf
b0
=
,
2
12
153 19nf
b1
=
.
2
2(33 2nf )
(A.5)
The solution of Eq. (A.1) can be obtained perturbatively. Expanding q to order a,

q = q0 + a q1 ,
(A.6)
we find
d
q0 = C0 q0 ,
d
d
q1 = (C0 + b0 ) q1 + C1 q0 .
d
The solutions of Eqs. (A.7), (A.8) are
q0 ( ) = R 1 e R q0 (0),
q1 ( ) = R 1 e( +b0 ) R q1 (0) + R 1 e( +b0 )
(A.7)
(A.8)
(A.9)
Z
b1 e R q0 (0), (A.10)
d e( +b0 ) C
where the matrix R diagonalizes C0 ,

RC0 R 1 = diag(1 , . . . , M ) ,
(A.11)
b1 = RC1 R 1 .
C
(A.12)
and
Collecting these results, and noting that a exp(b0 ) = a(0), up to terms of order a 2 , we
can write the solution as
62
q( ) U (C, )q(0)
"
=R
+ ae
( +b0 )
Z
d
b1 e
e( +b0 ) C
R q(0),
(A.13)
with the initial condition

q(0) = q0 (0) + a(0)q1(0).
The explicit expression of U (C, ) is

n e ( m +b0 )
1
n
mn e
b
+ a( ) C1
Rnj ,
Uij (C, ) = Rim mn e
n m b0
(A.14)
(A.15)
which, expanded to next-to-leading order reduces to

(
"

bmn b1 n mn
a(0) m /b0
a(0) n /b0 C
1
1
+ m
a(0)
Uij (C, ) = Rim mn
a( )
n + b0
a( )
#)
n /b0

a(0)
a( )
(A.16)
Rnj .
a( )
In the case of standard QCD evolution equations the matrix C0 is at most 2 2 and is
easily diagonalized. In the cases treated in this paper, the matrix C0 is triangular and can
be diagonalized using the methods discussed in the next appendix.
Appendix B. Diagonalization of triangular matrices

In this appendix, we show how to construct the matrix R which diagonalizes a generic
n n triangular matrix T by means of the recursion relations Eqs. (2.19), (2.20). The
matrix R is defined by the requirement that
RT R 1 = diag(1 , . . . , n ),
(B.1)
where the matrix T is upper triangular, i.e., Tij = 0 if i > j . It is easy to see, by solving
the secular equation, that the eigenvalues i of T coincide with its diagonal elements,
i = Tii .
(B.2)
Now, define eigenvectors v j associated to the j th eigenvalue Tjj , with components vi j :

n
X
Tik vk j = j vi j .
(B.3)
k=1
Clearly, the matrix R 1 coincides with the matrix of right eigenvectors, (R 1 )ij = vi j ,
P
while the matrix R coincides with the matrix of left eigenvectors nk=1 v j k Tki = j v j i ,
Rij = v i j . The eigenvector condition Eq. (B.3) immediately implies that the j th
component of the j th eigenvector is equal to one: vj j = 1. Furthermore, it is clear that
63
Eq. (B.3) can only be satisfied if all components vk j of the j th eigenvector with k > j
vanish,
vj j = 1,
vk j = 0 if k > j.
(B.4)
Using Eq. (B.4) and the fact that the matrix T is triangular, Eq. (B.3) can be written as
j
X
Tik vk j = j vi j .
(B.5)
k=i
Substituting the explicit form of the eigenvalues, Eq. (B.2), and identifying vi j = (R 1 )ij ,
this is immediately seen to coincide with Eq. (2.20). Furthermore, using the condition
vj j = 1, this equation can be viewed as a recursion relation which allows the determination
of the (k 1)th element of v j once the kth element is known, which is what we set out
to prove. The same argument, applied to the left eigenvectors, leads to the expression in
Eq. (2.19) for R.
Appendix C. Splitting functions in the DIS scheme

In this appendix, we give the explicit expressions of the AltarelliParisi kernels in the
DIS scheme [10,14]. We define the non-singlet splitting functions as
P (x) Pqq (x) Pq q (x) = P(0) (x) + aP(1) (x) +
(C.1)
and the singlet 2 2 matrix of splitting functions as

P (x) = P (0) (x) + aP (1) (x) + .
(C.2)
The MS LO and NLO kernels are given for example in Eqs. (4.94) and (4.102)(4.112)
of Ref. [3], whose notation and conventions we have followed throughout this paper.
The splitting functions in the DIS scheme can be constructed from these by a change of
factorization scheme.
To next-to-leading order, a generic change of factorization scheme for the splitting
functions is
P(1) P(1) b0 E NS ,

P (1) P (1) + E, P (0) b0 E,
(C.3)
(C.4)
where E and E NS are functions of x, and the commutator in Eq. (C.4) is defined in terms
of convolutions, as

(C.5)
E, P (0) = E P (0) P (0) E,
and is thus in general nontrivial to compute.
The transformation that takes from the MS to the DIS scheme, defined as in Ref. [14],
is given by
64
E NS (x) = C q (x),
q
C (x)
E(x) =
C q (x)

2nf C g (x)
,
2nf C g (x)
(C.6)
(C.7)
where C q (x) and C g (x) are the next-to-leading terms of the quark and gluon coefficient
functions for the unpolarized deep-inelastic structure function F2 (x, Q2 ) in the MS
scheme:

F2 x, Q2
)
(

Z1

x
dy
2
q x
2
g x
2
1
+ aC
y, Q + 2nf aC
g y, Q
= x he i
y
y
y
y
x
Z1
+x

dy
x
q x
1
+ aC
qNS y, Q2 + ,
y
y
y
(C.8)
where he2 i =
2
i=q,q ei /(2nf ).
Explicitly,

3
1
log(1 x)
q
(1 + x) log(1 x)
C (x) = CF 2
1x
2 1x +
+
2

9
1 + x2
log x + 3 + 2x
+
(1 x) ,
1x
3
2

1x
8x 2 + 8x 1 .
C g (x) = TR 1 x 2 + x 2 log
x
(C.9)
(C.10)
Notice that the expression for the scheme change given in Eq. (3.9) of Ref. [14] lacks a
factor of 2nf in the qg entry because it refers to a single flavor.
The explicit expression of the commutator is

E, P (0) qq

1
2
= CF nf 3 Li2 (x) 1 + 4 x 4 x 2 log2 x
2
3

2
5
22 x
3
22 x 2
4
+ 7x
log(1 x)
+ 9x
log x
+
3x 2
3
2
3

55 20 x 8 x 2
2
1x
2
,
+ 1 2 x + 2 x log(1 x) log 2 +
3x
6
3
3
x
(C.11)
E, P (0)

qg
= 2 nf b0 C g (x) + nf 2

1
2
Li2 (x) + log2 x
2
6

2
2 + 4 x 6 x log(1 x) + 2 + 16 x + 10 x 2 log x

59 x 2
11
+ 24 x
+
2
2
1 + 4 x + 4 x2

2
+ CF nf 1 2 x + 2 x 2 2 Li2 (x) + log2 (1 x)
3

2
1 + 10 x 10 x log(1 x) 1 4 x + 10 x 2 log x

6 + 11 x 8x 2

1
+ 2 CA nf (1 + 4x) Li2 (x) + log2 x
2

1
79x 2
2
+ 12x
log(1 x)
+
3x 2
6

1x
+ 1 2x + 2x 2 log(1 x) log
x

31x 2
1
43
1
+ 16x
log x +
2
6
3x 12
2
2x 2
281x
121x
+ 2x +
6
12
3

E, P (0) gq
= b0 C q (x)
65
(C.12)

1
2
2
2
2
20 2x x 16x
+ CF nf (1 + x) log x + 2 Li2 (x) +
3
x

2
4x 2
4
4x
x
log x 1 +
log(1 x)
+ 1 + 5x
3
3x
3

log(1 x)
3
27
4 (3) (1 x) + 3
+ log2 x
+ CF2
4
1x
2
+

2
6 5x
log(1 x) + 2 x log2 (1 x)
12 +
x
2
x

2
3x
2
9
+
2 2 x log x log(1 x)

2
2
2
x

7
4
+ 4 x log x
1 2x Li2 (x) +
x
2

2

log x log(1 x)
log2 x
log (1 x)
2
4
+
1+x 3
1x
1x
1x
+

2
2
2
2

27 2
9x
2 x
1
log x
3
+
+
+
+
1 x2
2
1x
4
3
2
3
1x +
2

log x
log(1 x)
+
+ 4 (3) (1 x)
+3
+ CA CF 6
1x
1x
2
+

2
log x log(1 x)
+ 2 1 2 x Li2 (x) 8x
x
1x

2
2
4 2 x
log x
1
log (1 x)
9x +
+ 2x
+ 6x
1x
1x
3
1x +
+

66

2
2 x 2 log2 (1 x)
1 + 2x 2 (3 log x + log2 x) +
x

4
+ 2 2 x + 4 x 2 log x log(1 x)
x

6
+ 17 4x + 6x 2 log(1 x) 2 6x
x

2
2 2 x 2
x
+ 8x 2 +
.
+
3
3
Finally

E, P (0) gg = E, P (0) qq .
(C.13)
(C.14)
Note that [E, P (0) ]qg denotes the qg matrix element of the commutator; i.e., using the
conventions of Ref. [3], upon scheme change

(1)
(1)
2nf Pqg
+ E, P (0) qg b0 Eqg .
(C.15)
2nf Pqg
Appendix D. Truncated moment integrals
The integrals which are needed in order to compute the Mellin moments of the NLO
AltarelliParisi splitting functions are well known [15]. Here we list all the truncated
moment integrals which are necessary in order to determine the evolution kernels Gn (x),
Eq. (2.5), from the expressions of the splitting functions given in Appendix C. The
triangular anomalous dimension matrices Ckl , Eq. (2.12), can be easily determined from
these formulas by Taylor expansion.
Z1
dz zn1 =
1 xn
,
n
(D.1)
Z1
dz zn1 log z =
1 x n (1 n log x)
,
n2
(D.2)
Z1
dz zn1 log2 z =
2 x n (2 2 n log x + n2 log2 x)
,
n3
(D.3)
Z1
dz zn1 log(1 z)
x
n+1
1
x
2 F1 (n + 1, 1; n + 2; x)
n (n + 1)

+ (n + 1) E + x n log(1 x) + (0) (n + 1) ,
(D.4)
67
Z1
dz zn1 log z log(1 z)
x
1h
E n x n+1 (x, 2, n + 1) x n log(1 x)
n2
+ n x n log(1 x) log x + x n+1 (x, 1, n + 1) (n log x 1)
i

2

2
+ (0) (n) + n (0) (n) n (0) (n + 1) + n (1) (n + 1) ,
Z1
dz zn1
x
xn
xn
log z
= 2 + x n+1 (x, 2, n + 1)
log x
1z
n
n
x n+1 (x, 1, n + 1) log x (1) (n),
Z1
dz zn1
x
(D.5)
(D.6)
2 xn
2 xn
log2 z
= 3 2 x n+1 (x, 3, n + 1) + 2 log x
1z
n
n
xn
log2 x
n
x n+1 (x, 1, n + 1) log2 x (2) (n),
+ 2 x n+1 (x, 2, n + 1) log x
Z1
dz zn1
x
Z1
dz zn1
x
1
1z

+
= E
n1 Z
X
dz zk1 log z log(1 z),
dz zn1
1
1+z

1 (0) n
xn
(0) n + 1
,
= 2 F1 (n, 1; n + 1; x)
n
2
2
2
(D.10)
log z log(1 + z)
1+z
"

(3) log x log2 (1 + x) log3 (1 + x)
x
n1
+
+ log(1 + x)Li2
= (1)
8
2
3
1+x
#
X

Z1
n1
x
k1
k1
(1)
dz z
log z log(1 + z) , (D.11)
+ Li3 (x) + Li3
1+x
dz zn1
(D.9)
k=1 x
Z1
Z1
(D.8)
log z log(1 z)
= Li3 (1 x) log(1 x)Li2 (1 x)
1z
xn
(0)
2 F1 (1, n; n + 1; x) (n),
n
(D.7)
k=1
68
Z1
dz zn1
x
2 xn
2
log2 z
= 3 + 2 x n+1 (x, 3, n + 1) + 2 x n log x
1+z
n
n
2 x n+1 (x, 2, n + 1) log x
xn
log2 x
n
+ x n+1 (x, 1, n + 1) log2 x

1 (2) n
(2) n + 1
,

8
2
2
Z1
(D.12)
Li2 (z)
1+z
"
7 (3) 2
log 2 + log(1 + x)Li2 (x) + log x log2 (1 + x)

= (1)n1
4
12

log3 (1 + x)
1
2
log(1 + x) + 2Li3
+
3
1+x
3
#
Z1
n1
X
(1)k1 dz zk1 Li2 (z) ,
(D.13)
dz zn1
k=1
Z1
dz zn1 log z log(1 + z)

x
"
1
4 n x n+1 (x, 2, n + 1) 4 log 2
=
4 n2
n
+ 4 x n+1 (x, 1, n + 1) (1 + nlog x) +
4 x log(1
+ x)
n
n
(0)
(0) n + 1
2
1+
4 n x log x log(1 + x) + 2
2
2
#

n
n+1
+ n (1)
,
n (1) 1 +
2
2

Z1
log(1 z)
dz zn1
1z
+
x

= (n 1) (1 x) 4 F3 (1, 1, 1, 2 n; 2, 2, 2; 1 x)

3 F2 (1, 1, 2 n; 2, 2; 1 x) log(1 x)
log(1 x)2
,
+
2
Z1
dz zn1 Li2 (z)
x
"
1
x n+1 2 F1 (n + 1, 1; n + 2; x)
= 2
n (n + 1)
(D.14)
(D.15)

(n + 1)
Z1
69
n 2
log 2 + x n log(1 + x)
12

n
1 (0)
(0) n + 1
1+
+
2
2#
2

+ n x n Li2 (x) ,
(D.16)

dz zn1 log2 (1 z) = 2 (1 x) 4 F3 (1, 1, 1, 1 n; 2, 2, 2; 1 x)
3 F2 (1, 1, 1 n; 2, 2; 1 x) log(1 x)
1 xn
log2 (1 x),
n

Z1

1
dz zn1 Li2 (z) = 2
(2 + n) E + (0) (n + 1)
n (n + 2)
x

+ (n + 1) x n+1 2 F1 (n + 1, 1; n + 2; x)
+

2
x n log(1 x) n x n Li2 (x)

,
+ (n + 1) n
6

2
Z1
n1 log (1 z)
dz z
1z
+
x
(D.17)
(D.18)
n2
X
log(1 x)3
(1 x)
2 4 F3 (1, 1, 1, j ; 2, 2, 2; 1 x)
3
j =0
2 3 F2 (1, 1, j ; 2, 2; 1 x) log(1 x)

1 x j +1
log2 (1 x) ,
+
(j + 1) (1 x)
(D.19)
where
X
zk
,
Lim (z) =
km
Z0
Li2 (z) =
k=1
d k+1 log (n)

,
dnk+1
X zk
,
(z, s, a) =
(a + k)s
(k) (n) =
dt
log(1 t)
,
t
(D.20)
E = (0) (1) 0.577216,
(D.21)
(D.22)
k=0
p Fq (a1 , . . . , ap ; b1 , . . . , bq ; z) =
X
(k + a1 ) (k + ap ) zk
k=0
(k + b1 ) (k + bq ) k!
(D.23)
Note that the integrals in Eqs. (D.9), (D.11), (D.13) and (D.19) are valid only when n is a
positive integer, which is what is usually needed. If an expression for real or complex n is
70
necessary, one should compute these integrals numerically. All other integrals are given in
a form that immediately generalizes to complex n. The special functions in Eqs. (D.20)
(D.23) are available in algebraic manipulation programs of common use.
References
[1] See, e.g., S. Catani et al., hep-ph/0005025, to be published in the proceedings of the workshop
Standard Model Physics (and more) at the LHC, CERN, 1999.
[2] See, e.g., S. Forte, Nucl. Phys. A 666 (2000) 113.
[3] See, e.g., R.K. Ellis, W.J. Stirling, B.R. Webber, QCD and Collider Physics, CUP, Cambridge,
1996.
[4] S.A. Larin et al., Nucl. Phys. B 492 (1997) 338, hep-ph/9605317.
[5] E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 (1992) 525.
[6] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[7] G. Altarelli et al., Nucl. Phys. B 496 (1997) 337, hep-ph/9701289;
G. Altarelli et al., Acta Phys. Pol. B 29 (1998) 1145, hep-ph/9803237.
[8] S. Catani, Z. Phys. C 75 (1997) 665, hep-ph/9609263.
[9] W.T. Giele, S. Keller, Phys. Rev. D 58 (1998) 094023, hep-ph/9803393;
D. Kosower, W.T. Giele, S. Keller, in: Proc. 1999 Rencontres de Physique de la Valle
dAoste, INFN, Frascati, 1999, p. 255;
R.D. Ball, J. Huston, in: S. Catani et al., hep-ph/0005114.
[10] G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 143 (1978) 521;
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 157 (1979) 461.
[11] F.J. Yndurain, Phys. Lett. B 74 (1978) 68;
G. Parisi, N. Sourlas, Nucl. Phys. B 151 (1979) 421;
W. Furmanski, R. Petronzio, Nucl. Phys. B 195 (1982) 237.
[12] S. Forte, L. Magnea, Phys. Lett. B 448 (1999) 295, hep-ph/9812479;
S. Forte, L. Magnea, in: Proc. Int. Europhysics Conf. on High-Energy Physics, EPS-HEP 99,
Tampere, 1999, hep-ph/9910421.
[13] See, e.g., S. Iyanaga, Y. Kawada (Eds.), Encyclopedic Dictionary of Mathematics, MIT Press,
Cambridge, Mass., 1980.
[14] M. Diemoz, S. Ferroni, M. Longo, G. Martinelli, Z. Phys. C 39 (1988) 20.
[15] For comprehensive tables see, e.g., A. Devoto, D.W. Duke, Riv. Nuovo Cimento 7 (1984) 1;
J. Blmlein, S. Kurth, Phys. Rev. D 60 (1999) 014018, hep-ph/9810241.

Intercepts of the non-singlet structure functions

B.I. Ermolaev a , M. Greco b,c, , S.I. Troyan d
a A.F. Ioffe Physico-Technical Institute, 194021 St. Petersburg, Russia
b CERN, 1211 Geneva 23, Switzerland
c Dipartimento di Fisica and INFN, University of Rome III, Rome, Italy
d Petersburg Nuclear Physics Institute, 188300 Gatchina, St. Petersburg, Russia
Received 6 September 2000; accepted 2 November 2000
Abstract
Infrared evolution equations for small-x behaviour of the non-singlet structure functions f1NS and
NS
g1 are obtained and solved in the next-to-leading approximation, to all orders in s , and including
running s effects. The intercepts of these structure functions, i.e., the exponents of the power-like
small-x behaviour, are calculated. A detailed comparison with the leading logarithmic approximation
(LLA) and DGLAP is made. We explain why the LLA predictions for the small-x dependence of the
structure functions may be more reliable than the prediction for the Q2 dependence in the range
of Q2 explored at HERA. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
In recent years the deep inelastic scattering (DIS) of leptons off hadrons has been the
object of intensive theoretical studies. In QCD the hadronic tensor for electron proton DIS
is usually regarded as a convolution of two objects: the partonic tensor W describing
p
DIS off a (nearly) on-shell parton (quark or gluon) and the probability Ph to find the parton
in the nucleon. As is well known, they have a different description: whereas the tensor W
can be calculated within perturbation theory, there are no model-independent methods of
p
calculating Ph , but only the general behaviour with respect to the kinematical variables
can be predicted. Usually, the hadronic tensor W is presented in terms of the structure
functions (see, e.g., [1]) f1,2 and g1,2 :

q q
(pq)
(pq)
1
f2
W = g + 2 f1 + p 2 q p 2 q
q
q
q
(pq)

(Sq)
mq
p g2 ,
+
S g1 + S
(1)
(pq)
pq
E-mail address: mario.greco@cern.ch (M. Greco).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 7 - 7
72
B.I. Ermolaev et al. / Nuclear Physics B 594 (2001) 7188
where p, m, S are the momentum, mass and polarization of the incoming parton and q
is the momentum of the virtual photon. The structure functions depend on the Bjorken
variable x = q 2 /(2pq), with (1 > x > 0) and Q2 = q 2 > 0. Each of the structure
functions consists of flavour singlet and flavour non-singlet (NS) contributions. Although
the singlet contributions accounting for gluon splitting are dominant at small x, the NS
contributions are also interesting quantities to investigate. They can be expressed in terms
of densities of quarks (1q) and antiquarks (1q)
inside a hadron. The non-singlet spinNS
whereas the
dependent structure function g1 refers to the combination of (1q + 1q),
combining them
non-singlet spin-averaged structure function f1NS is related to (1q 1q);
one could study the evolution of 1q and 1q separately.
The standard theoretical investigation of DIS structure functions, originally developed
for the kinematic region of x 1 and large Q2 , is made through the DGLAP [2] evolution
equations. This approach accounts for the evolution of the structure functions with respect
to the photon virtuality Q2 , whereas the evolution with respect to x is neglected. In other
words, the DGLAP accounts for the resummation of all powers of ln Q2 and systematically
neglects the powers of ln x, which is obviously correct when x is not small. Despite the
fact that the DGLAP equations provide a good agreement with experimental data (see,
e.g., [3]), they are not expected to work well in the small-x region, where logarithms of x
are not less important than logarithms of Q2 and therefore must be taken into account to
leads to a power-like, or Regge-like, behaviour in x and
all orders in s . This resummation
p
NS
2
a
2
in Q of the type f1 ( Q /x) . Such a behaviour cannot be obtained within DGLAP,
p
which predicts a dependence of the type f1NS exp c ln(1/x) ln ln Q2 .
In perturbative QCD the power-like behaviour was obtained long ago [4] for the
singlet structure functions of spin-averaged DIS and later in Refs. [5,6] for the structure
functions g1 and f1NS . However, contrary to the DGLAP, the results of Refs. [46] are
obtained in the leading logarithmic approximation (LLA) with a fixed QCD coupling s .
Hence the exponents calculated in [46] (also called intercepts) for the predicted powerlike small-x behaviour of all structure functions contain s fixed at a scale that is not
well defined. The very existence of such a scale, sometimes called a reasonable scale,
and estimates for its numerical value have been discussed in many works (see, e.g., [7]).
However, according to the Regge theory, the expressions for the intercepts must not contain
s running with Q2 , but, on the contrary, they just have to be numbers.
Thus, in order to obtain realistic values for the intercepts, one has to go beyond the LLA
and include the QCD running coupling effects in the evolution equations at small x, so that
the running coupling s would be integrated out in the solutions. Recently in Ref. [8] we
have obtained expressions for the non-singlet structure function 1 f1NS , with running QCD
coupling effects taken into account. The resulting intercept does not explicitly depend on
s but depends on the infrared cut-off. That calculation prompted us to study the running
QCD coupling effects for the non-singlet structure functions in more detail.
The structure function g1NS is similar to f1NS , because the same Feynman graphs
contribute to both of them. As a result, they obey the same leading order (LO) DGLAP
1 f NS was denoted as f NS in work [8].
1
73
evolution equation. A difference between them arises only in next-to-leading order (NLO)
corrections to DGLAP and can be extracted from the expressions of the second loop
anomalous dimensions (see, e.g., [9,10]).
Concerning the small-x behaviour, the difference between f1NS and g1NS can be
related to the difference in their signatures: the signature of f1NS is positive whereas
g1NS has a negative one. A difference between them means that some high-order nonladder graphs contribute to f1NS and g1NS in a different way. Historically, signatures for
scattering amplitudes of high energy processes were first introduced in the context of
the Regge theory (see, e.g., [11]). A forward scattering amplitude M(s, u), where s and
u are the Mandelstam variables, has positive (negative) signature if it is symmetrical
(antisymmetrical) with respect to the replacement s u:
M (+) (u, s) = M (+) (s, u),
M () (u, s) = M () (s, u).
(2)
Basically, any forward scattering amplitude M(s) consists of the two parts:
M(s, u) = M (+) + M () ,
M () =

1
M(s, u) M(u, s) .
2
(3)
In the small-x region, neglecting O(m2 /s) and O(Q2 /s) terms, one can use u s. As
the imaginary part in s, =s M, for s > 0, corresponds to the cross-section, this cross-section
may acquire contributions from both M (+) and M () amplitudes. The partonic tensor W
of Eq. (1) can be considered as the imaginary part of the amplitude M of the forward
Compton scattering of a virtual photon on a constituent quark:

q q
(pq)
(pq)
1
2M2
M = g + 2 2M1 + p 2 q p 2 q
q
q
q
(pq)

(Sq)
mq
S 2M3 + S
p 2M4 ,
+
(4)
(pq)
pq
so that
W =
1
=s M ,
2
(5)
and in particular,
f1 =
1
=s M1 ,
g1 =
1
=s M3 .
(6)
The amplitude M is invariant with respect to the replacement q q, , which

corresponds to the crossing transition from the s-channel to the u-channel. However, the
projection operators multiplying f1 and g1 behave differently under such a transition: the
operator (g q q /q 2 ) does not change, so the invariant amplitude M1 multiplying it
does not either. On the contrary, the operator ( mq /(pq)) acquires a negative sign,
so the invariant amplitude M3 must acquire a negative sign too. Therefore, the invariant
amplitudes
M (+) =

1
M1 (s, u) + M1 (u, s)
2
(7)
74
and

1
M3 (s, u) M3 (u, s)
(8)
2
have positive and negative signatures, respectively, and the DIS structure functions f1
and g1 can be considered as imaginary parts of scattering amplitudes with positive and
negative signatures:
M ()
f1 =
1
=s M (+) ,
g1 =
1
=s M () .
(9)
The first observation of the importance of the contribution of the negative-signature

amplitudes to the cross-sections was made in the context of QED in Ref. [12]. Later
on, in Ref. [13], quarkquark scattering amplitudes with positive and negative signatures
were calculated in the double-logarithmic approximation (DLA). Using the results of those
works, it was shown in Ref. [6] that in DLA, with s fixed, the intercept of g1NS is larger
than the intercept of f1NS .
In the present paper we obtain explicit expressions for f1NS and g1NS for small values
of x, which account for both leading double-logarithmic (DL) and sub-leading singlelogarithmic (SL) contributions to all orders in QCD coupling, including running s effects.
As logarithms of both x and Q2 are important at small x, we account for both of them,
constructing and solving a two-dimensional infrared evolution equation (IREE). In order
to take running s effects into account, we use the approach of Ref. [13], with the
improvement made in our previous work [8]. Also, we discuss similarities and differences
between running s effects in the DGLAP and in our approach. Finally we discuss the
region of applicability of the LLA and estimate the range of Q2 where it can be safely
applied. The paper is organized as follows: in Section 2 we derive the small-x evolution
equations for the non-singlet structure functions and solve them. In Section 3 we obtain
the asymptotic behaviour of f1NS and g1NS . In Section 4 we compare our results with
the intercepts of the non-singlet structure functions obtained in LLA. In Section 5 our
findings are compared with DGLAP predictions for the NS anomalous dimensions and we
discuss the difference in accounting for running s effects. Finally, Section 6 contains our
conclusions.
2. Small-x evolution equation for the non-singlet structure functions

In the Born approximation, which is the pure NS case, M is given by the sum of the
graphs in Fig. 1. They yield

p p
mq
(+)
()
Born
S 2MBorn ,
(10)
2MBorn +
M = g + 2x
(pq)
(pq)
where
()
MBorn
= eq2

1 s m2 q 2 u m2 q 2
2
,
2 m2 s
m u
(11)
75
Fig. 1. The Born graphs for the DIS amplitude M .
()
Fig. 2. Two-loop graph obtained from Fig. 1(b), contributing to =s M (a), and the corresponding
physical process (b).
with eq being the electric charge of the incoming quark and s = (p + q)2 , u = (p q)2
the Mandelstam variables. We have omitted unimportant contributions proportional to q ,
q in Eq. (10).
()
MBorn are amplitudes of defined signature (cf. Eq. (2)). Obviously, only the first term in
Eq. (11), corresponding to the graph (a) in Fig. 1, has an imaginary part in s at s > 0, which
gives the same contribution to the amplitudes of both signatures. According to Eq. (9) we
obtain the well-known results:
f1Born = g1Born =
eq2
2
(1 x).
(12)
The property f1Born = g1Born remains true in the next one-loop approximation, with one
gluon added to graphs in Fig. 1. Only the dressing of graph (a) in Fig. 1 contributes to
the imaginary part in s at s > 0. This explains why the LO DGLAP splitting functions
and the LO anomalous dimensions are the same for the structure functions corresponding
to amplitudes with different signatures. The difference between them arises only at the
second-loop order. Indeed, let us consider graph (a) of Fig. 2. It is actually a non-ladder
graph, obtained from the Born graph (b) in Fig. 1 by adding two gluons to it. On the other
hand, graph (a) in Fig. 2 can be redepicted as graph (b) in Fig. 2. It is easy to check that its
imaginary part in s (the quark propagators to be cut are marked with crosses in Fig. 2) has,
in particular, DL terms ( ln3 (1/x)); this graph therefore cannot be neglected. Then, one
can see that the DL contribution of this graph, which is symmetrical in , (contribution
76
to f1NS ) and the antisymmetrical one (contribution to g1NS ) have different signs. This
example shows that some higher-loop contributions to f1NS and g1NS are different.
Thus, a possible regular way of calculating f1NS and g1NS consists of the following steps:
(i) Calculate the forward scattering Compton amplitude M obtained by adding
gluon propagators to both Born graphs in Fig. 1.
(ii) Extract from it the amplitudes M1 and M3 proportional to (g ) and q p /
(pq) (cf. Eqs. (4) and (6)), which are the positive- and negative-signature amplitudes M () , respectively.
(iii) Calculate (1/)=s M () .
So, first we calculate M , and we do it by constructing and solving an IREE for M .
In order to account for the evolution in both x and Q2 one must consider DL and SL
contributions to all orders in s . High order Feynman graphs contributing to M may
have both ultraviolet (UV) and infrared (IR) singularities. Whereas UV singularities are
absorbed by renormalization, IR ones must be regulated explicitly. Providing gluons with
a mass violates the gauge invariance. In order to save it and to avoid IR singularities,
we use the Lipatov prescription of compactifying the impact parameter space (see, e.g.,
works [1315]). In other words, we introduce the infrared cut-off in the transverse space
(with respect to the plane formed by q and p) for integrating over the momenta of virtual
particles:
ki > .
(13)
With such a cut-off acting as a mass scale, one can neglect quark masses and still be
free from IR singularities. As a result the amplitude M depends on the cut-off and we
can study its evolution with respect to it, thus obtaining the IREE for M . Feynman
graphs contributing to the non-singlet structure functions are obtained from graphs in
Fig. 1 by adding gluon propagators to them, without breaking the quark lines, however.
With logarithmic accuracy, the region of integration over transverse momenta of virtual
quarks and gluons can be regarded for every Feynman graph contributing to M as a sum
of sub-regions so that in every sub-region only one virtual particle (quark or gluon) has
the minimal transverse momentum k . We call such particles the softest ones although
their longitudinal momenta can be large (see, e.g., [15]). It is essential that integrating
over k one includes as the lowest limit. The softest particle can be either a gluon or
a quark. When the softest particle is a quark, DL contributions come from the sub-region
where another quark has the same transverse momentum, so that there appears a t-channel
intermediate state with a soft quark pair. Therefore in order to keep the DL contributions,
one must consider M as convolution of two amplitudes connected by the softest quark
pair, each quark with the minimal transverse momentum k , as depicted by graph (b) in
Fig. 3. The first amplitude appearing in graph (b) (the upper blob) is the Compton forward
scattering M and the second one (the lower blob) is the forward scattering amplitude of
quarks M0 . On the other hand, when the particle with the minimal k is a virtual gluon, it
can be factorized out of the amplitude with the DL accuracy [16]. This means, in particular,
that the propagator of such softest gluon is attached to external lines only, yielding zero
result. Having added the Born contribution to the convolution graph (b) in Fig. 3, we arrive
77
Fig. 3. The evolution equation for the DIS amplitude M .
at the equation of the BetheSalpeter type for M . As usual the BetheSalpeter equation
relates M with on-shell quarks (lhs of equation in Fig. 3) to M with off-shell quarks
(rhs of Fig. 3). However, all amplitudes in Fig. 3 are actually on-shell because the virtual
quark pair in Fig. 3 has the minimal transverse momentum k . Integrations in blobs imply
k to be a new IR cut-off and, therefore, a new mass scale for these blobs (see [13,15,16]).
Let us note that in the DGLAP approach such softest quark pair is always in the
lowest ladder rung because of the DGLAP ordering: in this approach all ladder transverse
momenta ki are indeed ordered as
6 k1 6 k2 6 6 Q2 ,
(14)
the enumeration running from the bottom of the ladder to the top. Eq. (14) shows that
only the integration over k1 has as the lower limit. The lower limits for other ki
with i 6= 1 are expressed through the other transverse momenta. Therefore, the small-x
DGLAP equations correspond to the case where the quark scattering amplitude in graph (b)
in Fig. 3 is considered in the Born approximation. As is well known, the ordering of
Eq. (14) corresponds to accounting for lnn (Q2 /2 ) without considering terms lnm (1/x)
not accompanied by logarithms of Q2 . On the other hand, as we investigate the small-x
region, we have to lift this ordering, allowing for the transverse momentum of quarks in
any ladder rung to reach the lowest limit k = . Obviously, this increases the region of
integration over ki . Then, a quark pair with minimal k appears in any rung of the ladder
Feynman graph.
The convolution of the amplitudes depicted in graph (b) prompts us to apply the Mellin
transform for solving this equation. However, to respect the signatures of M () it is
more convenient to use the asymptotic form of the SommerfeldWatson (SW) transform
(see [11]):

M
()
s Q2
,
2 2
Z
=

d s ()
Q2
()
()F
,
,
2 2
2
(15)
where
() =
e 1
2
(16)
78
is the signature factor, for which this transform differs from that of Mellin. The inverse
transform to Eq. (15) is
F
()
2
() =
sin()
Z
d exp()=s M () (),
(17)
where = ln(s/2 ).
The small-x region corresponds effectively to the dominance of the small- region in
Eq. (15). Expanding into series in and retaining the first terms in Eq. (15, (+)
1 + /2 and () /2, we see that in the small-x (or small-) region
1
1
=s M () =

d s
Q2
()
F
,
;
2 2
2
(18)
we need to calculate F () only to obtain the structure functions f1NS and g1NS (cf. Eq.(6)).
To do this one can apply the SW transform of Eq. (15) to the equation depicted in Fig. 3 and
then differentiate it with respect to ln 2 . On the other hand the Born amplitude does not
depend on and therefore vanishes when differentiated with respect to . Then, observing
that
2
M ()
M ()
M ()
+
=
2
ln(s/2 ) ln(Q2 /2 )
(19)
corresponds to
F () +
F ()
y
(20)
for the amplitudes F () , where we have defined y = ln(Q2 /2 ), and differentiating the rhs
of the equation depicted in Fig. 3 with respect to ln 2 , we are led to the following IREE:

1
()
+ F (, y) =
(1 + ) F () (, y)F0() (),
(21)
y
8 2
where = 1/2. For more details, see Ref. [8], where the latter equation was obtained for
F = F (+) . The only difference now is that IREE (21) involves also the negative signature
()
(+)
amplitude F0 of forward quarkquark scattering instead of F0 only. The solutions to
Eq. (21) are

(22)
F () = C exp + (1 + )F0() ()/8 2 y ,
which contain two unknown quantities. The first one is an arbitrary factor C, essentially of
a non-perturbative nature, which can be fixed by comparison with the data.
()
The other unknown quantity F0 can be specified by constructing and solving the
IREE for the amplitude M0 of forward scattering of quarks. The equation for M0 is
shown in Fig. 4. The first two terms in the rhs look similar to the equation in Fig. 3.
Indeed, the rhs consists of the Born contribution depicted by graph (a), the convolution
of two quark scattering amplitudes (graph (b)), the intermediate quarks having minimal
transverse momenta k , and a new contribution depicted by graphs (c)(f). For symmetry,
79
Fig. 4. The evolution equation for the quark scattering amplitude M0 .
the contributions of graphs (c) and (d) are equal. The same is true for graphs (e) and (f).
All these graphs correspond to the case where the particle with the minimal transverse
momentum k is a gluon. The propagators of such soft gluons are attached to external lines
because by integrating over the soft gluon momenta the most singular DL contributions
come from the region where each soft virtual gluon is factorized out (see, e.g., Ref. [16]).
Self-energy graphs are absent in Fig. 4 because the Feynman gauge is used. The blobs in
these graphs imply integrations over momenta of internal virtual particles with k acting
as a new IR cut-off. Let us consider in detail the contribution of the graphs in the rhs of the
equation in Fig. 4. The Born contribution coming from graph (a) in Fig. 4 is given by

s
s
,
(23)
(s)
B () = 2CF s (s)
s
s 2 +
s 2 +
where the quark mass is substituted by in the denominator and is suppressed in the
numerator as well as q 2 . We use for s the following formula:
1
1
=

2
2
b ln(s/QCD ) b ln(s/QCD )

ln(s/2QCD )
1
+
=
,
b ln2 (s/2QCD ) + 2 ln2 (s/2QCD ) + 2
s (s) =
(24)
where b = (33 2nf )/12 (nf is the number of flavours).

Eq. (24) is written in such a way that s (s) has non-zero imaginary part when s is
positive. Applying the transform Eq. (17) to Eq. (23) we define
80
()
2
=
Z
d exp()=s B () () =
A
,
(25)
where = ln(s/2 ),
"
#
Z
d exp()
4CF
,
A() =
b
2 + 2
( + )2 + 2
(26)
introduced = ln(2 /2QCD )
assuming > QCD .

and we have
The first term in the square brackets in Eq. (26) corresponds to the imaginary part of
Eq. (24) and the second one comes from the imaginary part of 1/(s 2 + ).
The expression corresponding to graph (b) in Fig. 4 is

2
1 1
+ F0 .
(27)
8 2
Here, the softest partons (i.e., partons with minimal k ) here are given by the
intermediate quark pair. The remaining graphs, (c)(f) in rhs of Fig. 4, correspond to
the case when the softest parton is a gluon and therefore can be factorized out, i.e., its
propagator can be attached to the external lines in all possible ways.
The amplitude M0 in the lhs of Fig. 4 is colourless, i.e., it belongs to the singlet
representation of the group SU c (3). As one of the virtual gluons is removed from the blobs
in graphs (c)(f), these blobs are no longer colourless. They correspond to the colour octet
scattering amplitude M8 , where k acts as a new IR cut-off. As M8 does not depend on the
longitudinal components of k, one can integrate over them, easily arriving at

Zs
d(k 2 )
s
(k
)<
M
.
(28)
=M(cf ) = CF
s
s
8
2
(k 2 )
k
2
Eq. (28) reads that there is total compensation between the cuts of M8 , so that only cuts
that do not involve M8 (i.e., two-quark ones) contribute to =M(cf

).
()
As we are going to obtain M0 to all orders in s with SL accuracy, we must first know
(+)
M8 with the same accuracy. However, for an evaluation of the asymptotic behaviour
()
of M0 , this knowledge is not necessary. Indeed, it was shown in Ref. [17] that in DLA
the blobs in Fig. 4(c)(f) can be approximated, with a few percent accuracy, by their Born
(+)
contribution. The reason is that the octet amplitude M8 decreases with energy so quickly
that higher-loop contributions can be neglected. Motivated by this result and noticing from
the results of Ref. [8] that, apart from running s effects, SL contributions do not change
DL results drastically, we use this approximation in the present work, replacing the blobs
in graphs (c)(f) of the equation in Fig. 4 by their Born value, although with the running s .
Substituting the Born approximation M8Born for the colour octet amplitude M8 ,

s
s
= 2< s (s)
(s)
<M8 <M8Born
s
s 2 +
s 2 +

(29)
2 <s (s) <s (s) ,
81
we then obtain
()
=M(cf )

2CF
<s (s) <s (s)
=
2N
Zs
d(k 2 )
s (k 2 ).
(k 2 )
(30)
Finally, applying the Mellin transform Eq. (17) to Eq. (30), and using Eq. (24) for s , we
obtain, for contribution of graphs (c)(f) in Fig. 4:
D
()
2CF
() =
b2 N
Z
d e
+
ln

+
1
.
( + )2 + 2 +
(31)
In DLA, when s is fixed, instead of Eq. (31) one gets

(+)
() = 0,
DDL
()
DDL
() =
4s2 CF
.
3 N
(32)
The positive-signature contribution D (+) in the DLA is zero, so that M8 does not
(+)
contribute to M0 at all. In the Feynman gauge this means that the positive-signature
contribution of non-ladder graphs in the DLA is zero and only ladder graphs must be
considered. This remarkable observation was first made in Ref. [12] for QED processes,
and later in [13] for quark scattering. Eq. (31) shows that running s effects violate the
total compensation of non-ladder graph contributions. In our previous work [8] we had
neglected this non-compensation. We discuss it in more detail in Section 4.
()
Now we are able to write the full equation for F0 shown in Fig. 4, which generalizes
(+)
the equation for F0 = F0 obtained in our earlier work [8]:

2
1 1
A()
()
+
+ F0() () + D () (),
(33)
F0 () =
8 2
where D (+) and D () are given by Eq. (31); = 1/2 corresponds to SL contributions not
related to running coupling effects.
Eq. (33) has the following solution
p
2
() ())/2 2
()
2 (1 + )(A() + D
.
(34)
F0 = 4
1 +
Combining this result with Eqs. (9), (18), (22) we finally obtain the following
expressions for the structure functions f1NS and g1NS :
Z
f1NS
g1NS =
with C arbitrary.

1
d
(+)
C
exp (1 + )F0 /8 2 y ,
2
x

1
d
()
C
exp (1 + )F0 /8 2 y ,
2
x
(35)
82
3. Asymptotics of the non-singlet structure functions

The small-x asymptotic behaviour of g1NS and f1NS can be obtained by evaluating the
expressions found at the end of last section with the saddle-point method. However, it is
()
(+)
much easier to get it directly, by noticing that at small x g1NS x 0 and f1(NS) x 0 ,
()
where 0 is the rightmost singularity of F () . Eq. (34) reads that this singularity is a
square root branch point given by a solution to 2

(1 + )
2

+
2CF
b
2
2CF
b
1
4NCF

Z
0
2 + 2
Z
0
d e
( + )2 + 2

+
+
1
= 0. (36)
d e
ln
( + )2 + 2 +
We recall that = ln(s/2 ) and = ln(2 /2QCD ). This equation can be solved
numerically. Before doing so, let us notice that if we keep s fixed in Eq. (36) and, for
self-consistency, put = 0, it transforms into the well-known algebraic equation in the
DLA:

2s CF 2 1
2s CF
1
2
1 (1) = 0,
(37)

2
4NCF
with the obvious solutions
s

1/2
4
DL
DL 1
DL
1+ 1+
(+)
,
1.055e
(+)
e
() = e
2
(N 2 1)
where the positive-signature leading singularity reads:
q
DL
= 2sDL CF /.
e
(+)
(38)
(39)
Now let us come back to Eq. (35). Substituting the value of F0() of Eq. (34) at the point
()
= 0 , where the square root turns to zero, we arrive at the power-like asymptotics:
f1NS
(+) 2 (+) /2
0
1 0
Q
,
x
2
(+)
g1NS
() 2 () /2
0
1 0
Q
.
x
2
(40)
()
Obviously, the solutions 0 and 0 to Eq. (36) (also called the intercepts of f1NS
and g1NS , respectively) depend on the choice of the parameters nf , , QCD . We recall
that we keep > QCD . Also, must be greater than the mass of the heaviest involved
quark, which fixes nf . In numerical estimates below, we have used QCD = 0.1 GeV and
nf = 3. Eq. (36) has been solved numerically for different values of . The dependence
()
()
of the intercepts 0 on (in terms of ) is plotted in Fig. 5. The plot shows that 0
2 There is a misprint in the coefficient before the square brackets of Eq. (52) in Ref. [8]. The true coefficient is
(2CF / b). However, the numerical results of Ref. [8] are correct.
83
is greater than 0(+) for all . Both 0(+) and 0() first grow with increasing , getting
maximal at 1 GeV , where
(+)
(+) = 0.37
(41)
0() () = 0.4,
(42)
0
and
and smoothly fall off after that. Eq. (36) also tells us that contrary to the double-logarithmic
approximation of Eq. (38), there is no total compensation between the DL contributions
(+)
of the non-ladder graphs to 0 (see expression in the second squared brackets) when
s is running. That fact was neglected in [8]. Once taken into account, it leads to a
steep increase of both 0(+) and 0() at small . However, we regard this increase as
an artefact of our approach: using expression (24) for is becoming unreliable at such
small . Another reason for not considering the small- region is that when 6 2.86 the
values of 0(+) become complex, which leads to negative f1NS . Thus, we find that our
approach is valid for > 2.8. Recalling that = ln[(/QCD )2 ], with acting as an
input for our evolution equations, we see that, for self-consistency, we must keep >
min = 4QCD = 0.4 GeV as QCD = 0.1 GeV. A further look at Fig. 5 shows that
(+)
()
0 and 0 are maximal at 5. Then they smoothly ( 1/) fall with growing
. The reason for this fall-off is quite clear: at large , the running s is approximated
by 1/(b). The -dependence of 0() , plotted in Fig. 5, has such a complicated form
because of the competition between the relative weights of the 2 - and -contributions in
()
the denominators of Eq. (36). In the absence of the 2 -terms 0 will depend on in a
smoother way, as shown by the curves (3) and (4) in Fig. 5. We will further discuss this
plot in the next section.
Fig. 5. Dependence of the intercept 0 on infrared cut-off = ln(2 /QCD ). 1: for f1NS , 2: for g1NS ,
3 and 4: for f1NS and g1NS , respectively, without accounting for 2 -terms.
84
4. Comparing with the leading logarithmic approximation

The power-like (Regge-like) small-x behaviour of f1NS and g1NS , Eq. (40), with DL inDL given by Eqs. (38) and (39), was obtained in Refs. [5] and [6]. These
()
tercepts 0() = e
expressions explicitly include the dependence of s upon an unknown scale, because the
DL approximation treats s as fixed. On the other hand, the LLA approach postulates that
asymptotically, at very high energies, LLA results should dominate over all sub-leading
contributions. Having accounted for the SL contributions, we are now able to check the validity of the LLA predictions for the NS structure functions. First, let us notice that accounting for running s does not lead to intercepts 0() explicitly dependent on s , although
()
they depend on QCD . Then, thanks to our choice of the infrared regulation, Eq. (13), 0
DL ,
depend on . But in spite of such a difference between 0() and the DL intercepts e
()
DL e
DL ]/e
DL = 0.055 basi(+)
(+)
one can see from Eq. (38) and Fig. 5 that the DL ratio [e
()
()
cally holds for 0 also, changing slightly from 0.09 at = 3 to 0.06 at = 8. Therefore,
the DL result for the ratio in Eq. (38) is a good approximation. The 2 -terms that we have
accounted for in Eq. (36) are quite important because of their large value. In principle,
they are formally beyond the SL accuracy we have kept through the paper. Indeed, contributions 2 may appear also from integrations over phase space. Such contributions
in higher orders of the perturbative series are beyond the control of any kind of logarithmic approximation. On the other hand the 2 we have accounted for appear as the result
of respecting the analytical properties of the scattering amplitudes. So, we cannot simply
neglect them. Moreover, it has been shown recently in Ref. [18] that a value () = 0.4 is
in good agreement with the experimental data for the structure function F3 . When the 2
in Eq. (36) are dropped, () = 0.4 corresponds to 5.5 GeV (see curves 3 and 4 in
Fig. 5). At this value of the SL contributions not related to the running s are small, i.e.,
one can neglect them, putting = 0 in Eq. (36). Therefore we conclude that 0 5.5 GeV
is a good estimate of the value of the mass scale for LLA evolution equations. In other
words, 20 acts as a momentum scale for the evolution equations obtained in LLA. Therefore these equations are not expected to reproduce the correct Q2 -dependence until at least
Q2 > 20 30 GeV2 .
20
(43)
Q2
is pretty close to the typical values of

at HERA, we can see that using the LLA
As
evolution equations is not very reliable for such Q2 .
Let us discuss the x-dependence. In Refs. [1921], it is assumed that s depends on Q2
in the expression for the intercepts. On the other hand the results of Ref. [8] show that there
is no such dependence. Now we have seen that in LLA s should depend rather on 20 than
on Q2 , and as the Q2 range covered at HERA is not far from 20 the estimate s = s (Q2 )
might still be in reasonable agreement with the data.
5. Comparison with DGLAP
Now let us compare our results with the DGLAP results for the non-singlet structure
functions. First, let us check that the corresponding DGLAP evolution equations can be
obtained from Eq. (21). Writing Eq. (21) as

1
()
()
F (, y) = +
(1 + ) F0 ()F () (, y)
y
8 2
85
(44)
and noting that the term in the square brackets is absorbed by changing the Mellin
factor (s/2 ) to x (see Eq. (35)), then
1
(1 + )F0() ()
(45)
8 2
are the new small-x non-singlet anomalous dimensions. However, when x 1 the main
contribution of the Mellin factor exp[ ln(1/x)] comes from the region of rather large
values of , where ln(1/x) 6 1. That makes it possible to expand the exponent F0() in
Eq. (35) into a series in 1/. Doing so, we obtain
e() =
F ()
e () (1 + ) 0 2
8

2

A 21 1
A 1
+ +
+ + D () + O A3 .
=
2
2

8
8
Retaining only the first term in the rhs of Eq. (46), one arrives at the expression

A 1
()
+
(1) =
e
8 2
(46)
(47)
for the leading order anomalous dimension, which is similar to the expression for the
leading order DGLAP non-singlet anomalous dimension

s (Q2 )CF 1
+ .
(48)
(1) =
2
Let us first compare the two results by assuming that s is fixed. Eqs. (47) and (48)
coincide completely because in this case A = 4s CF . Let us then compare the next two
()
, again for
terms in Eq. (46) with the second order DGLAP anomalous dimension (2)
3
fixed s . Retaining the leading terms, proportional to 1/ , we can see that
2

A 21 1
()
+
=
+ D ()
(49)
(2)
e
8 2
transform into (see Eq. (32))
2 2
s CF 1
(+)
(2) =
2 3
and

()
(2)
2

CF2
CF 1
+
,
2
N 3
(50)
(51)
which coincide with the leading terms of Eq. (B18) of Ref. [10], also used in Ref. [9]
for the second order non-singlet anomalous dimensions. Unfortunately, we do not obtain
exact coincidence between non-leading terms, which are proportional to 1/2 , but this
86
can be explained by a different choice of the factorization and regularization procedures.

Therefore, the non-singlet anomalous dimensions given by Eq. (45) correspond to a
resummation of (1) to all orders in the QCD coupling and to accounting for the signaturedependent DL contributions to order s2 . Such a resummation leads to the power-like
small-x behaviour, which cannot be achieved by incorporating any finite number of NLO
contributions into expressions for DGLAP anomalous dimensions.
Another difference between IREE and DGLAP is the different treatment of the QCD
coupling. Although both approaches use the same formula (Eq. (24)) for s , the DGLAP
2 ) in ith ladder rung, k
uses s = s (ki
i being the ladder quark transverse momentum
whereas IREE suggests that in the ith rung should depend on the gluon virtuality 3
(ki ki1 )2 . Let us explain this difference. When x 6= 1, the Born contribution to f NS
is zero and f NS obeys the BetheSalpeter equation
Z
1 s ((p k)2 )
.
(52)
f NS = d 4 k (q + k)2 2 =
k
[(p k)2 ]
We have suppressed all unimportant factors in Eq. (52) and have used the notation k
instead of k1 for the ladder quark momentum in the lowest rung. We have noted as
for off-shell f NS . Usually it is convenient to integrate over k in (52), using the Sudakov
parametrization:
k = (q + xp) + p + k ,
(53)
so that
2(pk) = 2(pq),
2(qk) = 2(pq)( x),
2
.
k 2 = 2(pq) k
(54)
But here it is more convenient to use the variable

m2 (p k)2
(55)
instead of the Sudakov variable , writing Eq. (52) as

Z
NS
=
d 2 k d dm2
f

2
s( x)(1 ) m2 ( x) k
(1 x) (1 )1
=
s (m2 )
.
2 ][m2 ]
[m2 + k
The argument of in Eqs. (52) and (56) should be positive, i.e.,

2
(1 x) > 0.
s( x)(1 ) m2 ( x) k
(56)
(57)
When x 1, one can neglect the last term in Eq. (57). It leads to ( x)(s(1 ) m2 )
> 0, which allows m2 to be neglected. Therefore, instead of Eq. (56), we obtain
Z

s (m2 )
. (58)
f NS = d 2 k d dm2 s( x)(1 ) (1 )1 =
2 ][m2 ]
2
[m + k
3 We use here the standard enumeration for the momenta of the virtual ladder quarks k , so that i = 0
i
corresponds to k0 = p and, increasing the number of runs, it leads to the top of the ladder.
87
After that, one can apply the Cauchy theorem relating an analytic function to its imaginary
part. This permits to perform an integration over m2 by taking residues. As m2 is positive,
the only residue is m2 = k 2 /. Thus, we obtain
Z
2 /)

s (k
.
(59)
f NS = d 2 k d s( x)(1 ) (1 )1
2
k
As > x and x 1, one can drop in the argument of s and arrive at
Z
2)

s (k
NS
,
f = d 2 k d s( x)(1 ) (1 )1
2
k
(60)
which corresponds to the DGLAP. Usually the negative sign of the argument of s is
dropped together with the negative sign of the argument in Eq. (24). As the argument of s
in Eq. (59) is negative, the expression for s in the DGLAP approximation of Eq. (60) does
not contain 2 -terms at all, while the argument of s in Eq. (52) is positive and therefore
there are 2 -terms in Eq. (52).
Obviously, the transition from Eq. (52) to Eq. (59) holds only when x 1, which is
2 in Eq. (57)
the DGLAP kinematics, and fails for small x. In particular, neglecting k
has an important consequence for the DGLAP evolution, where one picks up logarithmic
contributions through integrations over k . At small x one should use the s -dependence
given by Eq. (26) rather than that given by Eq. (60).
6. Conclusions
In conclusion, we have obtained a generalization of the second-order DGLAP evolution
equations for the non-singlet structure functions at small x. Besides resumming the LO
DGLAP anomalous dimension to all orders in s and accounting for running s effects,
we account for the difference between f1NS and g1NS , which is due to the difference in
the second-order anomalous dimensions. We have shown that, with single-logarithmic
contributions and running s taken into account, f1NS and g1NS have the scaling-like
asymptotic behaviour
x
()
Q2
2
(() /2)
0
(61)
as also obtained earlier in the DLA [5,6] at asymptotically small x. We have appropriately
calculated the intercepts 0() . We have made detailed comparisons of our results with
the predictions obtained in the leading-logarithmic approximation and with the DGLAP.
Results obtained with the evolution equations usually depend on a mass scale 20 acting as
a starting point of the evolution with respect to Q2 , and we estimate the value for this mass
scale for a LLA evolution equation as 20 30 GeV2 . This implies that these equations can
reproduce the correct Q2 -dependence for the non-singlet structure functions (and likely,
for the singlet ones) only for approximately Q2 > 30 GeV2 . On the other hand, as far as
the x-dependence is concerned, we have also shown that the estimate s = s (Q2 ) used
88
in Refs. [1921] in LLA can be in a reasonable agreement with the x-behaviour of the
structure functions for the HERA data.
Acknowledgements
The work is supported in part by grant INTAS-97-30494, and by EU QCDNET contract
FMRX-CT98-0194.
References
[1] R.P. Feynman, PhotonHadron Interactions, Benjamin, 1972;
B.L. Ioffe, V.A. Khoze, L.N. Lipatov, Hard Processes, North Holland, 1984.
[2] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 297;
V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 438;
L.N. Lipatov, Sov. J. Nucl. Phys. 20 (1972) 95;
Yu.L. Dokshitzer, Sov. Phys. JETP 46 (1977) 641.
[3] G. Altarelli, R.D. Ball, S. Forte, G. Ridolfi, Nucl. Phys. B 496 (1997) 337.
[4] V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Sov. Phys. JETP 44 (1976) 443;
V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Sov. Phys. JETP 45 (1977) 199;
Y.Y. Balitskij, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
[5] B.I. Ermolaev, S.I. Manayenkov, M.G. Ryskin, Z. Phys. C 69 (1996) 259.
[6] J. Bartels, B.I. Ermolaev, M.G. Ryskin, Z. Phys. C 70 (1996) 273;
J. Bartels, B.I. Ermolaev, M.G. Ryskin, Z. Phys. C 72 (1996) 627.
[7] S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov, G.B. Pivovarov, JETP Lett. 70 (1999) 155
160, hep-ph/9901229.
[8] B.I. Ermolaev, M. Greco, S.I. Troyan, Nucl. Phys. B 571 (2000) 137; hep-ph/9906276.
[9] M. Gluck, E. Reya, M. Stratmann, W. Vogelsang, Phys. Rev. D 53 (1996) 4775.
[10] E.G. Floratos, C. Kounnas, R. Lacaze, Nucl. Phys. B 192 (1981) 417.
[11] P.D.B. Collins, An Introduction to Regge Theory and High Energy Physics, Cambridge, Univ.
Press, 1977.
[12] V.G. Gorshkov, L.N. Lipatov, M.M. Nesterov, Yad. Fiz. 9 (1969) 1321.
[13] R. Kirschner, L.N. Lipatov, Nucl. Phys. B 213 (1983) 122.
[14] L.N. Lipatov, Phys. Lett. B 116 (1982) 411.
[15] B.I. Ermolaev, L.N. Lipatov, Int. J. Mod. Phys. A 4 (1989) 3147.
[16] B.I. Ermolaev, V.S. Fadin, L.N. Lipatov, Sov. J. Phys. 45 (1987) 508.
[17] B.I. Ermolaev, S.I. Troyan, in: Proceedings of 5th Intern. Workshop on DIS, New York, 1998,
p. 861.
[18] A. Kataev, G. Parente, A.V. Sidorov, Nucl. Phys. A 666/667 (2000) 184;
A. Kataev, G. Parente, A.V. Sidorov, Nucl. Phys. B 573 (2000) 405.
[19] B. Badalek, J. Kwiecinski, Phys. Lett. B 418 (1998) 229.
[20] J. Kwiecinski, B. Ziaja, hep-ph/9902440.
[21] Y. Kiyo, J. Kodaira, hep-ph/9803448;
Y. Kiyo, J. Kodaira, hep-ph/9711260;
Y. Kiyo, J. Kodaira, Z. Phys. C 74 (1997) 631.

Quarkonium production through hard comover

rescattering in polarized and unpolarized pp
scattering
Martin Maul
Department of Theoretical Physics, Lund University, Slvegatan 14A, S-223 62 Lund, Sweden
Abstract
In this paper hadroproduction of charmonium states in polarized pp collisions is discussed.
A thermal picture for the gluonic cloud of comovers is given making contact between the formalism
and the measured unpolarized cross sections. The experimentally observed non-polarization of the
final J / states leads to the consequence that no correlations between the initial proton spin and the
final charmonium spin should be existent. Hence the single spin asymmetries vanish to leading order
in that model. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.39.Hg; 29.27.Hj; 29.25.Pj
Keywords: Charmonium production; Hard comover rescattering; Thermal gluon plasma hep-ph/0009279
1. Introduction
Charmonium production in polarized pp scattering at RHIC is one of the key
experiments to pin down the polarized gluon distribution amplitude [1]. The standard
formalism in which heavy quarkonium production in hadronhadron collisions is described
is still the Color Octet Mechanism (COM) [2]. However, it is known that this mechanism
fails to describe the experimentally observed non-polarization of the final J / and 0
meson [3] and, furthermore, it predicts a c1 /c2 production ratio which is far too low
[4]. Recently, it has been shown that these problems can be cured by assuming that
the charmonium formation happens through rescattering with a gluon cloud of hard
comovers [5,6]: The two colliding hadrons form through the self interacting gluon field a
gluonic medium in which the heavy quarkonium formation is directed by hard rescattering
processes. This means a crucial qualitative difference as to electroproduction where one
E-mail address: maul@thep.lu.se (M. Maul).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 9 - 3
90
M. Maul / Nuclear Physics B 594 (2001) 89112
Fig. 1. Typical diagram for the S-wave charmonium production amplitude. The formation of the J /
mesons happens through hard interaction with a gluonic cloud of comovers ( ).
of the collision partners is a lepton and cannot participate in strong interactions. As a

deeper understanding of the heavy quarkonium formation mechanism in polarized pp
scattering will be very important for the extraction of the polarized gluon density, we want
to investigate what consequences this theory implies for the charmonium production in
polarized pp scattering. In addition to the rescattering picture developed in [5,6] we will
present a thermal description of the comover cloud.
Double spin asymmetries in charmonium hadroproduction have been calculated in the
framework of the COM formalism, which is based on a systematic non-relativistic velocity
expansion [7], for the prospective HERA-NE experiment [812]. In the framework of the
Color Singlet Model (CSM) [13,14] double spin asymmetries in J / production have
been studied in [15]. The CSM has been shown to be incompatible with the absolute
size of the unpolarized J / cross section [4]. Attempts to cure this failure by the k factorization approach within the CSM have not led to satisfying results [16,17]. Therefore, one is looking for a suitable combination of the CSM and the COM formalism in the
k -factorization approach which may release the polarization problem of the final J /
and 0 [18]. A discussion for the double spin asymmetry for RHIC energies in terms of the
CSM formalism can be found in [19]. Higher order velocity corrections in the polarized
case in terms of the COM formalism have been discussed in [2022]. It is the aim of this
paper to add to this discussion the possible insights the comover rescattering picture can
offer as to polarized hadroproduction of charmonium states.
In Sections 2 and 3 we will write down the total cross section for polarized S-wave
and P-wave charmonium production. For the gluonic comovers we will develop a thermal
description as a boson gas and explain the consequences for the polarized partonic cross
sections. In Sections 4 and 5 we will fit the parameters of the theory (volume of the comover
cloud, the energy the charm quarks carry on the average, and the expansion parameter of
the charmonium system) to the numerical values of the measured unpolarized charmonium
cross sections. We will discuss the implications on the energy transfer from the colliding
particles to the cloud and from the comover cloud to the charmonium system. We give also
a picture of the geometry of the cloud. Finally, in Section 6 we will calculate the double
spin asymmetries for inclusive J /, 0 and cJ , J = 0, 1, 2, production in the framework
of the thermal description.
91
2. Cross section for S-wave quarkonium production

The gg-fusion amplitude for 3 S1 quarkonium production in the presence of a background gluon field (`) (see Fig. 1 for a typical diagram) has been derived in [5,6]:

g 3 R0
1
2
c (`) ` e(Sz )
i1
M 3 S1 , Sz = d abc
1
3
2
6m

2 z 0
+ 0c (`)
` Sz + e(1 ) `Sz2 + e(2 ) `Sz1 .
1
(1)
Here R0 is the value of the S-wave quarkonium wave function at the origin. 2m is
the energy of the two incoming gluons. To the accuracy of the first order velocity
expansion used here it is identical to the energy of the charm quarks before and after
the interaction with the comover cloud and also identical to the mass of the J / or 0
produced. Effectively, we will see later that we have to adjust the numerical value for m
to experimental data of the unpolarized cross section. 1 , 2 are the helicities of the two
incoming partonic gluons with polarization e(i ), i = 1, 2. It has been pointed out in [5,6]
that the fact that the polarization of J / and 0 is small and consistent with zero, as can
be seen from fixed target data [2328] and also from the new CDF data [3], leads to the
condition that in the c.m. system of the two quarks the relation:

c 2 c
(`) (`) `2 /`2 ,
(2)
0
must be fulfilled. a =
Nc2 1
is the four-vector potential of the interacting gluonic
field. The philosophy of our concept is that the quarkonium production happens in a heat
bath of essentially co-moving (real) gluons. The polarization tensor of the real gluons can
then be described by:
(`)0 (`) = g0 +
` n0 + `0 n ` `0 n2
.
`n
(` n)2
(3)
The introduction of a real gluon field means on the other hand that the incoming charm
quark from the gluongluon fusion amplitude must be slightly off-shell. In the calculation
this off-shellness is not included. In fact, it can be regarded as a velocity correction of
order v = 2l0 /m (see Appendix B) which we will neglect in the calculations that follow.
The other alternative, i.e., to have a virtual field leads together with the requirement
0 0 to unphysical consequences as we show in Appendix B, whereas all requirements
turn out to be natural for a real gluon field . For the 0 and c1 asymmetries discussed in
Section 6 all model dependent parameters cancel and we coincide exactly with the model
for proposed in [5,6].
In order to fulfill the condition that ensures the non-polarization of quarkonium states
Eq. (2), we parametrize n = (1, be), with e being a unit vector and 0 6 b 1. In fact,
for b = 0 we find 0 0. One can interpret b approximately as the relative velocity of
the gluon comovers to the c.m. system of the two incoming gluons and e as the direction
of this relative movement as to the axis defined by the two incoming gluons which is the
z-axis in our case. In the following we will refer to b then as displacement parameter.
92
The heat bath, in which the quarkonium production takes place has the temperature T =
1/ and is described by the BoseEinstein statistics. This takes into account the multiple
gluon interaction inside the cloud. Guided by this philosophy, we define the following three
quantities:
Z
(4)
d 3 ` |( (`) `) e (P )|2
(V )
V 2 4,
P = V
2|`|(2)3 exp((n `)) 1

Z
d 3 ` |(0 (`)(` e (P ))|2 )
(E)
P = V
2|`|(2)3 exp((n `)) 1
2

b (4)
4 |e e(P )|2 + |e e (P )|2 ,
V
2
4
10
Z
d 3 ` Im[( (`) `) e(Sz )0 (`)(e(P ) `)]
(int)
Sz ,P = 2V
2|`|(2)3
exp((n `)) 1

b(4)
(4)
V 2 4 Im (e e(P )) e(Sz ) .

One should notice that with this form we reproduce the StefanBoltzmann law, that
the energy density represented by the field squared grows T 4 with temperature. The
polarization P can take the values P = 1, 0, +1, and the polarization vector e(P ), (and
also e(Sz )) is defined by:
(1, i, 0)/ 2 if P = +1,

(5)
e(P ) = (0, 0, 1)
if P = 0,
(+1, i, 0)/ 2 if P = 1.
We then can write down the following partonic cross section:
1 2 Sz =
=

2 X
1
M 3 S1 , Sz 2
3
2
2
2(2m)
(Nc 1)
5
9
abc
3
3
s R02 h 2
(2m)6 1
(V )
2
1
1
(E)
Sz + S0z 0
(E)
(E)
(E)
+ Sz1 2 + Sz2 1 + 2Sz2 Sz1 Sz

i
1 (int)
2 (int)
S0z S(int)
Sz Sz ,2
Sz Sz ,1
z ,0
5 3 s3 R02 h 2 (V )
(E)
Sz + S
+ 412 Sz1 S(E)
z
z
9 (2m)6 1
i

2
(int)
+ 1 1 1 2S0z Sz ,Sz .
(6)
The corresponding hadronic cross section is given by:

1 2 Sz =

dh1 2 Sz
1X
=
G(x1 , (2m)2 ) + 1 1 1G(x1 , (2m)2 )
dx1 dx2
4
1 2

(2m)2
2
2
.
G(x2 , (2m) ) + 2 2 1G(x2 , (2m) ) 1 2 Sz 1
Sx1 x2
Now one can isolate the various components:
(7)
93

1
(++Sz + Sz ) + (+Sz + +Sz )
4

1X
G x1 , (2m)2 G x2 , (2m)2 1 2 Sz
=
4
1 2

(E) 1 (V )
(E)
= FS G x1 , (2m)2 G x2 , (2m)2 1 S0z Sz + Sz + Sz ,
2

1
(++Sz + Sz ) (+Sz + +Sz )
LL (Sz ) =
4

1X
1 2 1G x1 , (2m)2 1G x2 , (2m)2 1 2 Sz
=
4
1 2

= FS 1G x1 , (2m)2 1G x2 , (2m)2

(E) 1 (V )
(E)
0
1 Sz Sz Sz + Sz ,
2

1
(++Sz Sz ) + (+Sz +Sz )
L0 (Sz ) =
4

1X
1 1G x1 , (2m)2 G x2 , (2m)2 1 2 Sz
=
4
1 2

(int)
1
0
= FS 1G x1 , (2m)2 G x2 , (2m)2 Sz S(E)
+
1
Sz Sz ,Sz ,
z
2

1
(++Sz Sz ) (+Sz +Sz )
0L (Sz ) =
4

1X
2 1G x2 , (2m)2 G x1 , (2m)2 1 2 Sz
=
4
1 2

(int)
1
0
1
2
= FS 1G x1 , (2m)2 G x2 , (2m)2 Sz S(E)
Sz Sz ,Sz . (8)
z
2
00 (Sz ) =
The common pre-factor is:

FS (2m)

(2m)2
5 3 s3 R02
1
=
,
9 (2m)6
Sx1 x2
(9)
where (2m) is the energy of the two incoming gluons. Eq. (8) is valid in general, even
without any assumptions about the thermal nature of the gluonic cloud, if we modify the
=
definition of (V ,E,int) accordingly. In case we apply our model we find always S(int)
z ,Sz
(E)
0. Then, the single spin asymmetries in our model are proportional to Sz Sz , which means
proportional b 2 . As we know from the non-polarization of the final J / and 0 meson that
b must be small and consistent with zero, we consequently predict the absence of single
spin asymmetries in charmonium hadroproduction. Furthermore, if we can set b to zero as
the non-polarization of the final J / and 0 suggests, we see that no initial spinfinal spin
correlations are present and the double spin asymmetry reduces to:
LL (Sz ) 1G(x1 , (2m)2 )1G(x2 , (2m)2 )
.
00 (Sz )
G(x1 , (2m)2 )G(x2 , (2m)2 )
(10)
94
Fig. 2. Typical diagrams for the P-wave charmonium production: (a) CSM contribution without
comovers at O(s2 ); contributions with comover interaction at O(s3 ): gg fusion (b) and q q
annihilation (c).
In other words, the experimental fact that the J / and 0 are produced in a non-polarized
mode assures that the double spin asymmetry equation (10) is essentially only dependent
on the ratio of the unpolarized and polarized gluon parton distributions, which is a very
important statement as far as the possibility is concerned to isolate the polarized gluon
density from direct J / or 0 production data.
We can describe the gluon cloud of comovers with a Bose distribution of momenta in
the sense of a gluon plasma, in which the formation of the charmonium states takes place.
To get a quantitative model the following quantities have to be set:
The temperature T of the cloud which should lie above the phase transition of
hadronic matter, i.e., larger than typically 200 MeV. It should also be much smaller
than the typical charm mass of 1.5 GeV, otherwise the interaction with the gluon cloud
would rather inhibit the charmonium production than catalyze it.
The volume V of the cloud which should be large enough to comprise the cc system.
The energy (2m) of the two initial gluons.
The expansion parameter from the quarkonium wave function at the origin to the
interaction point with the comovers.
From the non-polarization of the final J / and 0 we can already set the parameter b = 0
for the following. We will do this analysis in Section 4 and Section 5 after having collected
the cross sections for the P-wave charmonium production in the next section.
95
3. Cross section for P-wave charmonium production

P-wave charmonium production for the mesons c0 and c2 can occur via two
gluons without any contribution from comovers, see Fig. 2(a). This corresponds then
to the contribution being calculated from the color singlet model (CSM). The comover
contribution for the production of cJ , J = 0, 1, 2, mesons becomes important only at the
O(s3 ) level, see Fig. 2(b), (c), where the comover contribution is supposed to be dominant
over all other CSM contribution of the same order in s . The O(s2 ) CSM partonic cross
sections are given by:
(a)
1 2
(a)
1 2

24 2 s2 |R10 |2 2
P0 , 0 =
1 ,
(a)
1 2
(2m)7
0
2
2
2

32 s |R1 | 2 1
3
P2 , Jz =
1 Jz /2 .
(2m)7
3

P1 , Jz = 0,
(11)
We show the derivation of these cross sections in Appendix A to make also sure for the
constants needed to make contact between the thermal amplitudes and the physical cross
sections. Averaging over the gluon helicities 1 and 2 and summing over all possible final
states Jz we reproduce the unpolarized partonic cross sections as given in [4,29,30]. At
the level of three gluons, i.e., O(s3 ), we can again use the formulas derived from hard
comover scattering. This time, however, it is hot possible to include a finite displacement b
as it would require a further expansion of the gluon fusion amplitude, taking into account
the Lorentz transformation from the gluongluon c.m. to the final quarkquark c.m. [5].
Such an expansion would destroy the simple structure of the theory developed so far and,
therefore, we leave it out here. Furthermore, as data indicate from the non-polarization of
the final J / and 0 state should b be small and consistent with zero. With this assumption
the P-wave amplitude reads [5]:
M(b)

g 3 R 0 /m
2
PJ , Jz = f abc 1 i1
1
3
2
6m

z
c (`) `
q
z
i 32 e (Jz ) ( c (`) `)
3e (J ) c (`) `i
3i
(J = 0),
(J = 1),
(12)
(J = 2).
We define now additionally:

(V )
(V )
J =0,0 = 0 ,
Z

3 (4)
d 3 ` 3 |[e (Jz ) ( (`) `)]z |2
)
=V
=
V
1 J0z ,
J(V=1,J
3
2
4
z
exp((n `)) 1
2
2|`|(2) 2

Z
i
2
3
|e3i (Jz )[ (`) `] |
3 (4) 4 0
d `
(V )
1
=V
+ Jz . (13)
3
J =2,Jz = V
2|`|(2)3
exp((n `)) 1
2 2 4 3 Jz
(J ) necessary for the calculation performed here can
The explicit form of the tensor e3i
z
be found in [31]. We find then for the partonic cross section of the contribution from the
diagrams of type Fig. 2(b):
96

(b)
1 2

3 s3 |R10 |2 2 (V )
PJ , Jz = 4 2
1 J,Jz .
(2m)8
(14)
This formula states that there is no correlation between the helicities 1 , 2 and the total
angular momentum J, Jz . P-wave quarkonium production can also occur through q q
annihilation, see Fig. 2(c). Using the formulas given in [6], the amplitude for this case
is given by:

1 + |1 + 2 | g 3 3 R10 a
(c) 3
T
Mij a PJ , Jz = (22 )
2 2m3 ij
(2m)2
+ 2 ) 2m0a ` + `2 a
(J = 0),
e(
q1
i 2 e (Jz ) e(1 + 2 ) 2m a ` + `2 a (J = 1),
(15)
0
3
0
(J = 2).
Here 1 , 2 are the helicities of the incoming quark and antiquark, respectively. i, j are
their color indices and a is the color index of the gluon from the heat bath. Then the
partonic cross section reads:
(c)
1 2 Jz

PJ , Jz =
2 X 1 (c)
Mij a
2(2m)3
Nc2
32 3 s3 |R10 |2 (T )
J Jz (1
3(2m)10
2
PJ , Jz
ij a
+ 2 ).
(16)
If we refer to our model, discussed in the previous section and neglect 0 we obtain:
20 (6)
(T )
J Jz (1 + 2 ) = (1 + |1 + 2 |)V 2 6

1

23 1 |e(1 + 2 ) e(Jz )|2
(J = 0),
(J = 1),
(J = 2).
(17)
The essential statement is that the contribution (c) contains a correlation between the
spin of the charmonium and the spin of the two incoming gluons. On the other hand
this contribution (c) is suppressed by a factor T 2 /(2m)2 versus the contribution (b). As
T 2 /(2m)2 |`|2 /(2m)2 we can again neglect this contribution to the accuracy of the
calculation as it is of the same order as the velocity corrections not taken into account
here.
The formalism used to derive the partonic cross sections is NRQCD to leading order.
In principle to the order of accuracy of this approximation the masses of all charmonium
mesons are the same, i.e., MJ / = M 0 = McJ ,J =0,1,2 = 2mc [4]. In the same order of
accuracy we can also identify m = mc in all the formulas above. One has to keep in mind
that in such a crude approximation the contribution of velocity corrections may be quite
substantial. Unfortunately, the inclusion of velocity corrections will destroy the simple
structure of the relations derived in [5,6] and is therefore out of the scope of the discussion
here. To face this problem we will fit as a pragmatic ansatz m for the different charmonium
states involved in a way that most experimental facts are reproduced. m plays then the
97
role of the average gluon energy involved in the production of the charmonium state under
consideration.
4. The choice of the quark energy m

The starting point of the consideration is the direct J / cross section. Here we can as
a basis identify 2m = MJ / , because the difference between the MJ / = 3.097 GeV and
2mc = 3 GeV is small:
Z

V
dx1 dx2 FS MJ2/ G x1 , MJ2/ G x2 , MJ2/
dir (J /) =
2

b2
(4)
11 + 2ez2
2 4 3+

5
Z
X (J /)
00 (Sz ).
(18)
=
dx1 dx2
Sz
To the order of accuracy of expansion we could apply the same arguments also to the
direct 0 production. However, such an approximation, which would be in accordance of
the standard COM velocity expansion, where all charmonium masses are identical MJ / =
M0 = McJ = 2mc , is in practice very unsuitable as the masses enter in high powers in the
partonic cross section. We will therefore use an effective mass for the direct 0 production:
Z

V
dx1 dx2 FS M2 0 eff G x1 , M2 0 eff G x2 , M2 0 eff
dir ( 0 ) =
2

b2
(4)
11 + 2ez2 .
(19)
2 4 3+

5
We can then fit M 0 eff [32] to the measured ratio ( 0 )dir / (J /)dir given in [32]. It has
to be noticed that this ratio is completely independent of the cloud parameters. The result
of the fit is:
M 0 eff = 3.4228 GeV,
(20)
which is a bit smaller than the real 0 mass of M 0 = 3.6860 GeV. It indicates in our
language that there is a net transfer of energy from the gluon cloud into the quarkonium
system through hard comover rescattering or to stay in a thermal picture that there is a
transfer of energy from the hotter gluon cloud to the colder charmonium system. The result
of the fit can be seen in Table 1.
We can now pay attention to the cJ masses. As no comovers enter in the CSM
contribution we should use in this case the original cJ masses, i.e.:
Z

1 X
3
dx1 dx2 G x1 , M2cJ G x2 , M2cJ (a)
(CSM) (cJ ) =
2 PJ , Jz
1
4
1 2 Jz

2
McJ
.
(21)
1
x1 x2 S
98
Table 1
Fit results using the best value M 0 eff = 3.4228 GeV in comparison with experimental data from
E705 [32]. The pA experiment was done at E = 300 GeV and the A experiment at E = 185 GeV.
For the fit we get 2 = 0.08 per degree of freedom
( 0 )dir / (J /)dir
GRV
CTEQ5L
experiment E705
pA
A
0.23
0.23
0.20
0.21
0.21 0.05
0.23 0.05
For the contribution resulting from comover rescattering we have to fit two parameters,
first the effective cJ masses and then also the expansion parameter . We will proceed as
follows. We take for all three cJ the same effective cJ mass in the spirit that the effective
mass should be lowered proportional to what was the case for the 0 particle:
!
2
M 0 eff
1 X
McJ
= 3.2451 GeV.
(22)
McJ eff =
3
M 0
J =0
For the contribution resulting from comover rescattering we can make a fit of the expansion
parameter to the E705 data by considering the reduced fraction, which is defined in a
way that it is independent of the gluon cloud parameters in our approach:
(J /)s-wave = (J /)dir + ( 0 ) Br( 0 J /),
X
(cJ )(CSM) Br(cJ J /),
(J /)(CSM)
p-wave =
J
(J /)(comovers)
p-wave
(cJ )(comovers) Br(cJ J /),
J
(CSM)
( frac)red =
(J /)incl (J /)p-wave (J /)s-wave

(CSM)
(J /)incl (J /)p-wave
(comovers)
(J /)p-wave
(comovers)
(J /)p-wave
+ (J /)s-wave
(23)
In the framework of our thermal description this ratio is independent of the gluon cloud
parameters. Unfortunately, the reduced fraction is not directly measured which brings
in an extra dependence on the parton distributions used. The fit to the reduced -fraction
yields:
= 4.40.
(24)
This value is a bit larger than the value used in Ref. [5], where the effective mass
McJ eff was set to be 2mc . This points to the problem that the relative big expansion
parameter means that it may be inconsistent to consider only the wave function of the
quarkonium system at the origin. Besides the substantial velocity corrections this is the
second indication that the NRQCD approach in general is not a suitable description of the
problem. In fact, future analysis will have to find ways to go beyond the NRQCD approach
99
Table 2
Fit results using the best value = 4.40 with M 0 eff = 3.2451 GeV in comparison with the values
cJ
extracted from E705 experiment [32]. The pA experiment was done at E = 300 GeV and the A
experiment at E = 185 GeV. For the fit we get 2 = 0.75 per degree of freedom
pA
(-frac)red
GRV
CTEQ5L
experiment
fit
experiment
fit
0.29 0.04
0.30 0.04
0.34
0.33
0.36 0.03
0.36 0.03
0.34
0.34
which we have followed here to be compatible with the standard literature of the field. For
the details of the fit using = 4.40 and McJ eff = 3.2451 GeV see Table 2.
5. Determination of the other parameters of the theory

The two parameters that are left undetermined so far are the active Volume V of the
gluon cloud and its temperature T . From the design of the theory the temperature is limited
within tight bounds. It must be well above QCD in order to make the interaction hard and
to justify the use of perturbation theory. On the other hand it must be smaller than the
mass of the charm quark mc , otherwise the comover interaction would rather destroy the
charmonium system than to catalyze it. Now we can make an ansatz taking a constant value
T = 500 MeV and let us check now what consequences this has for the active volume V .
For this purpose we will fit V to the data available for inclusive J / and 0 production.
The situation is simple in the case of the 0 production because it is a purely direct process:
Z
FS (M2 0 eff )

G x1 , M2 0 eff G x2 , M2 0 eff
dir ( 0 ) = V dx1 dx2
2

b2
(4)
2
(25)
2 4 3 + (11 + 2ez ) .
5

The case of J / production is more complicated because to a considerable amount J /
mesons can be produced indirectly via the decay of cJ , J = 0, 1, 2, mesons predominantly
through photon emission (cJ J / + ). Therefore, we have to write for the total
inclusive unpolarized cross section for J / hadroproduction:
X
(cJ ) Br(cJ J / + X)
incl (J /) = dir (J /) +
J
+ dir ( 0 ) Br( 0 J / + X).
(26)
The total cJ cross section is then given by the contribution from the diagrams in Fig. 2
(a) and (b):
(cJ ) = (cJ )(CSM) + (cJ )(comovers),
100
(cJ )
Z

(a)
1 X
=
dx1 dx2 G x1 , M2cJ G x2 , M2cJ 1 2
4
(CSM)
1 2 Jz
Z
(cJ )
(comovers)
=V
FP M2cJ eff
= 4
PJ , Jz

M2cJ
,
1
x1 x2 S

dx1 dx2 G x1 , M2cJ eff G x2 , M2cJ eff
with the pre-factor:
FP (M2cJ eff )
2
3 3 |R 0 |2
s
1
1
M8cJ eff
(2J + 1)
M2cJ eff
x1 x2 S
(4)
,
2 4

.
(27)
(28)
Numerically, the following branching ratios are used [33]:

Br(c0 J / + ) = (6.6 1.8) 103 ,
Br(c1 J / + ) = (27.3 1.6)%,
Br(c2 J / + ) = (13.5 1.1)%,
Br( 0 J / + X) = (54.2 3.0)%.
(29)
So putting all components together we get for the total inclusive J / cross section:

Z

V (4)
dx1 dx2
3 G x1 , MJ2/ G x2 , MJ2/ FS MJ2/
incl (J /) =
2
4
2

2
+ G x1 , M 0 eff G x2 , M2 0 eff FS M2 0 eff Br( 0 J /)
X

(2J + 1)G x1 , M2cJ eff G x2 , M2cJ eff FP M2cJ eff
+
J
Br(cJ J /)
+
(CSM)
(cJ ) Br(cJ J /) .
(30)
+ A J /, 0 we refer
For the data of the total cross section + A J /, 0 and p(p)
basically to [30] and add the more recent values from [25,27,3437]. The cross sections
have been rescaled to give the value over the whole range of xF (The details as to this
rescaling are explained later in this chapter). We reproduce essentially the figures in [4],
except for the data point from [25] for the N 0 cross section, which is displayed a
factor 2 too small. For the gluon parton distribution of the proton we use the two leading
order sets from CTEQ5 [38] and GRV98 [39]. For the gluon parton distribution in the
pion we use the leading order parameterization given in [40] (GRS99). For s we use the
one-loop formula:

4
(4)
; using QCD = 200 MeV,
(31)
s 2 =

(nf ) 2
2
2
11 3 nf ln / QCD
101
Fig. 3. The active volume of the gluon plasma fitted from J / and 0 data in pA collisions (right)
and A collisions (left).
which comes close to the value used in the GRV and CTEQ5L (leading order) gluon
distribution. The scale 2 is given by the only scale relevant for the partonic subprocess
of quarkonium production, i.e., the quarkonium mass, so 2 = MJ2/ , M2 0 , etc. The
quarkonium wave function at the origin R0 is determined to leading order by the decay
to e+ e :
4e2 2 R 2
J /, 0 e+ e = c 2 em 0 .
MJ /, 0
(32)
We take the values (J / e+ e ) = 5.2374 keV, ( 0 e+ e ) = 2.3545 keV,

MJ / = 3097 MeV, M 0 = 3686 MeV, em = 1/137 [33]. ec = 2/3 is the charm quark
charge quantum number. In order to fix |R10 |2 we could try to extract it from the decay
cJ , however the data basis here is not very conclusive [33], and, therefore, we take
here in accordance with [5] and [4] the value resulting from the BuchmllerTye potential
given in [41], i.e.:
|R10 |2 = 0.075 GeV5 .
(33)
It should be noticed that other potentials, also tabulated in [41] yield considerable larger
values for |R10 |2 , up to nearly
a factor of 2. This means that the expansion parameter then
will be reduced by a factor 2 which will not help as to the principle problem mentioned
above.
Using Eqs. (25) and (30) the active volume V can be fitted to the data available from
pA and A collisions. The result is shown in Fig. 3. We assume a linear dependence in the
double logarithmic scale, i.e., a dependency of the form V = c(s/GeV2 )p . The numerical
results of the fit are shown in Table 3.
It is now the place to make a few statements as to the physical meaning of V and
its geometry. V is the size of the gluon cloud at the moment the interaction with the
charmonium pair takes place. It should decrease with s as the faster the collision happens
102
Table 3
Numerical results for the fit of the active volume of the gluon cloud displayed in Fig. 3. For the fit a
functional form V = c(s/GeV2 )p is assumed
GRV
pA
A
CTEQ5L
c [fm3 ]
c [fm3 ]
5.9212 100
2.6175 101
3.7885 101
5.9388 101
6.5601
3.8535
2.1032 102
1.6390 102
8.2425 101
8.5073 101
5.7948
4.2729
Fig. 4. Geometry of the cloud of hard rescattering comovers.
Fig. 5. The fraction of energy xcloud transferred from the projectiles to the gluon cloud in pA
collisions (right) and A collisions (left).
103
Fig. 6. Total cross section for S-wave charmonium production in nb versus s: On the left the cross
0
sections for J / production are displayed and on the right the ones for production. The upper
two figures show the results for pA and the lower two figures for A collisions. All nuclear effects
have been rescaled so that in principle all the cross sections should display the result for pp and p
collisions, respectively. The solid and the dashed line show the curves obtained from the combined
(J / and 0 data) fit for the temperature depending on which parton distribution has been used.
the less time has the cloud to form and to expand. Fig. 4 shows the geometry of the cloud.
The majority of all J / are produced at small p , so we can think us a situation where the
cc pair moves essentially in the beam axis. In transversal direction the cloud should have a
radius roughly comparable to 1/ l0 = 1/|`| 1/T , its longitudinal length however depends
on a momentum pk = /(V T 2 ). For simplicity we did not take into account this geometry
at the thermal integration d 3 ` in Eq. (4) etc., pk will grow with s. The ratio xcloud =
pk / s which is displayed in Fig. 5, shows what fraction of the energy of the system is
transferred to the cloud. Whereas the CTEQ5L gluon distribution leads to a rising fraction
(which is rather unphysical), the GRV gluon distribution predicts more or less a constant
fraction xcloud 0.5% for pA collisions and xcloud 1% for A collisions.
We are now in a position to predict what cross section we will get with our fit at higher
energies and especially with RHIC energies with s = 200 GeV. Fig. 6 shows the fit
104
Table 4
Comparison of the measured 1 /2 -ratio in A reaction with the result of the theory described here.
For the gluon distribution function the GRV98 GRS99 set is used
Reference
E beam [GeV]
(1 /2 )exp
(1 /2 )theor
E705 [32]
E506 [35]
185
515
1.4 0.4
1.2 0.4
1.0456
0.9194
we made in terms of the total inclusive J / and 0 cross section. The data have been
rescaled to use the full range in xF [1, 1]. For the A collisions we have assumed
an xF distribution of the form d/dxF (1 |0.18 xF |)c , with c = 2.5 for J / and
c = 3.9 for 0 [30]. In case of the pA collisions a symmetric xF distribution is assumed. It
turns out that for the latter one the CTEQ5L distribution predicts an unphysical decreasing
cross section at large s therefore we will not consider this distribution in the following
any longer, whereas GRV shows in all cases a reasonable relaxing rising behavior. We can
now have a more closer look at the details of the various processes contributing using the
GRV set only. The first important quantity of interest in the 1 /2 -ratio. It is defined by:
(c1 ) Br(c1 J /)
.
(34)
1 /2 -ratio =
(c2 ) Br(c2 J /)
A comparison of the data and our results is shown in Table 4. It turns out that the values
for the 1 /2 -ratio are a bit smaller, but well inside the error bars of the measured A
reactions. They would be reproduced even better, if we neglected the CSM contributions
altogether. In this case we get (1 /2 )theor = 1.2 independent of the beam energy, if we
set all masses equal to 2mc . So, in principle the theory is capable to describe the large
1 /2 -ratio found experimentally in contrast to the standard COM and CSM theory.
Fig. 7 shows some details as to the various subprocesses contributing to the inclusive
J / production in pA and A scattering. It is shown that the fraction of directly produced
J / versus the whole inclusive cross section decreases with increasing energy until it
falls down to about 40% at RHIC energies. The s dependence of the ( 0 )/ (J /)dir is
governed by the scale dependence of the gluon parton distribution (here GRV) involved. It
goes towards a constant for large s which is about 1/4. The c1 /c2 -ratio is rapidly falling
with s. In all cases the pA and A curves behave similar.
6. Double spin asymmetries

The discussion on the partonic cross section has shown two important consequences:
As long as we can set the displacement parameter b to zero, which in accordance to
the non-polarization of the final J /, there is no correlation between the proton spin
and the final J / spin orientation. All single spin asymmetries are then zero as to the
order of accuracy of the approximations applied here.
The double spin asymmetries for the directly produced J / and 0 depend up to a
factor only on the ratio of the polarized and unpolarized gluon distributions. In case
(a)
105
(b)
(c)
Fig. 7. Contribution of various subprocesses to inclusive J / production in pA and A collisions.
(a) The ratio of the directly produced J / to all J / measured is displayed. (b) The ratio between
the 0 cross section and the cross section of the directly produced J / is shown. The shape of this
ratio changes with s only due to the evolution of the gluon parton density. (c) The ratio c1 /c2 as
discussed in the text is displayed. The underlying gluon distribution for all figures has been taken
from GRV and GRS respectively.
of the inclusive J / cross section the different mass-scales make the situation more
complicated.
These findings mean a big simplification for the polarized physics because they state that
the extraction of the polarized gluon density from the double spin asymmetry will not
be complicated by initial and final state spin correlations. With the model set up in the
previous chapter we are able to compute the error bars for the double spin asymmetries for
J /, 0 and cJ production. In the following we collect the expressions for the double
longitudinal spin cross section 1LL :
Z
X
LL (Sz ) (general formula),
1LL = dx1 dx2
Sz
Z
Z

V (4)
1LL (J /)dir = 3 2 4 dx1 dx2 FS MJ2/ 1G x1 , MJ2/ 1G x2 , MJ2/ ,
2
Z
Z

V
(4)
1LL ( 0 ) = 3 2 4 dx1 dx2 FS M2 0 eff 1G x1 , M2 0 eff 1G x2 , M2 0 eff ,
2
106

V (4)
0
1LL cJ
= (2J + 1) 2 4
2
Z
dx1

dx2 FP M2cJ eff 1G x1 , M2cJ eff

(CSM)
(cJ ),
1G x2 , M2cJ eff + 1LL
1LL (J /)incl = 1LL (J /)dir + Br( 0 J /)1LL ( 0 )

X
Br(cJ J /)1LL (cJ ).
+
(35)
The sign in the general formula takes into account that the standard convention for
the numerator of the asymmetry is always anti-parallel spin alignment minus parallel spin
alignment. For the color singlet (CSM) contributions we find:
Z

12 2s2 |R10 |2
(CSM)
(c0 ) =
dx1 dx2 1G x1 , M2c0 1G x2 , M2c0
1LL
7
Mc0

M2c 0
,
1
Sx1 x2
(CSM)
(c1 ) = 0,
1LL
(CSM)
1LL
(c2 ) =
16 2s2 |R10 |2
M7c2

dx1 dx2 1G x1 , M2c2 1G x2 , M2c2

M2c 2
.
1
Sx1 x2
(36)
The corresponding unpolarized cross sections can be straightforwardly obtained by

replacing the polarized by the unpolarized gluon distributions and to remove all minus
signs. Then the double spin asymmetry ALL and its statistical error ALL are simply given
by:
r
+
ALL = 2
, = 1LL .
(37)
ALL = LL / ;
L(+ + )3
Fig. 8 shows the double spin asymmetry for S-wave charmonium production, i.e., inclusive
J / and 0 production. For the polarized gluon distribution amplitude we use the leading
order set gluon A [42], which we will abbreviate in the following by GSA. For the error
band we have assumed a luminosity of L = 0.25 pb1 . It is seen that for inclusive J / and
0 production the asymmetry at RHIC energies is sizable. It is larger for 0 production,
but here also the error bars are larger. For P-wave charmonium production (see Fig. 9) the
asymmetry for the c2 production is partially negative due to the CSM contribution. In
case the CSM contribution is smaller, i.e., that a smaller value for |R10 | is more realistic,
the asymmetry will go to more positive values, so here we see a very fine test of the
interplay between CSM and comovers. In all cases we get a substantial asymmetry for
RHIC energies with small error bars. In general it is noticed that the asymmetry decreases
with increasing beam energy s. This means that in addition to the RHIC spin program
an polarized experiment like HERA-NE , which is supposed to run at s = 40 GeV will

also contribute very valuable information for the mechanism how charmonium production
happens in hadroproduction. One should notice that the asymmetries for inclusive J /
107
Fig. 8. Double spin asymmetry ALL for inclusive J / (left) and 0 (right) hadroproduction. For the
plot we use the unpolarized parton distribution set GRV and the polarized GSA. For the grey error
band the assumed luminosity is 0.25 pb1 using a beam polarization of 100%.
(a)
(b)
(c)
Fig. 9. Double spin asymmetry ALL for inclusive cJ , J = 0, 1, 2 hadroproduction. For the plot we
use the unpolarized parton distribution set GRV and the polarized GSA. For the grey error band the
assumed luminosity is 0.25 pb1 using a beam polarization of 100%.
108
and the c1 production do not depend on the temperature while the other asymmetries do,
so to measure the asymmetries will give essential new information upon the validity of the
theory.
Finally, it has to be stated that the whole calculation still is based on the velocity
expansion p/m within the NRQCD formalism, which is truncated already at the first
order. The corrections to this may be quite considerable. In this direction we find also the
big expansion parameter already determined in [5] and which has been confirmed here.
Unfortunately, the inclusion of those velocity corrections will destroy the simple relations
derived in [5,6]. It would be quite advisable for future studies to develop a formalism, that
could test the results of the hard comover rescattering from a different stand point which is
not based on the approximations of the NRQCD.
7. Summary
In this work we have tried to describe the measured unpolarized cross section for
charmonium production through the framework of hard comover rescattering and made
predictions for the asymmetries in polarized pp scattering. The generic advantage of the
hard comover rescattering mechanism is that it can explain the non-polarization and the
comparatively large value for the 1 /2 -fraction observed in experiment in a simple and
natural way.
In order to get quantitative results we have expressed the comovers as a thermal cloud of
gluons. The measured data suggest that about 0.51% of the total energy in the collision is
invested in the formation of the gluon cloud. The fact, that the final state J / is unpolarized
leads us to the conclusion that the displacement parameter b should be small and consistent
to zero. If this is true then it means that there is no correlation in polarized experiments
between the initial polarization of the protons and the final polarization of the J /
and furthermore the single-spin asymmetry 0L should be zero, a notion which should
be tested by experiment. The asymmetries ALL ( 0 ) and ALL(c1 ) depend only on the
ratio of the polarized and unpolarized gluon distribution amplitudes, while ALL (J /)incl ,
ALL (c0 ) and ALL (c2 ) are in principle sensitive to the parameters describing the gluon
cloud. The hard comover rescattering picture provides an understanding of the formation
of onium states in hadroproduction which may give the answers to some problems left
unsolved in the standard COM and CSM mechanism. The RHIC-spin experiment and a
possible HERA-NE will provide very crucial new information upon the validity and the
consequences of the theory presented here.
Acknowledgements
It is a pleasure to thank Paul Hoyer, Nils Marchal and Stphane Peign for fruitful
discussion and many explanations concerning their work. I also want to thank Marc Bertini
and Torbjorn Sjstrand for helpful suggestions and ideas.
109
Appendix A. Derivation of the 0 and 2 O(s2 ) CSM cross sections

In this appendix we reproduce the Born cross section for 0 and 2 production (O(s2 )).
This gives a cross check for the formulas derived in [5] and also a cross check for the
phase-space and flux factors we used in the text to obtain the cross section formulas from
the amplitudes.
The momenta for the gluon fusion amplitude are:
g1 = (m, 0, 0, m),
p1 = (m, p),
g2 = (m, 0, 0, m),
p2 = (m, p),
(A.1)

p
/ g/ 1 + mQ
/1 T a
= ig v p2 , /2 T b 1
(p1 g1 )2 m2Q
2
+ /1 T a

p
/ 1 g/ 2 + mQ
b
/
T
u(p1 , )
2
(p1 g2 )2 m2Q

g 2
fabc 1 2 v p2 , (/
g 1 g/ 2 )u(p1 , )
2
(2m)

pz
ig 2
b
a 1
v(p
2 , ) /2 T (/
+
p1 g/ 1 + mQ )/1 T
2m
m m2

pz
1
2 u(p1 , )
p 1 g/ 2 + mQ )/2 T b
+ /1 T a (/
m m
2

g
fabc 1 2 v p2 , 3 u(p1 , ).
+
2m
+
(A.2)
Using now the equation of motion one can write:

p1 g/ 1 + mQ )/1 T a
/2 T b (/

= T b T a /1 2 p + /2 1 p + m 1 2 3 + i[1 2 ]z 0 5 ,
p1 g/ 2 + mQ )/2 T b
/1 T a (/

= T a T b /1 2 p + /2 1 p m 1 2 3 i[1 2 ]z 0 5 .
(A.3)
The spinor combinations can be expressed as follows:

v p2 , 3 u(p1 , ) = 2m 2,
v p2 , 0 5 u(p1 , ) = 2m ,

v p2 , /u(p1 , ) = 2m (2) e (2) 2.
(A.4)
Then, using [T a , T b ] = ifabc T c and {T a , T b } = dabc T c , we arrive at the following

expression in the first order p/m:
110

1
ipz
fabc Tijc
= ig 2 i [1 2 ]z
ab ij + dabc Tijc
Nc
m

2 2 e (2) 1 (2 p) + e (2) 2 (1 p)

1 1
c
ab ij + dabc Tij
m Nc

pz 1
c
ab ij + dabc Tij .
(A.5)
+ 2 1 2
m Nc
Hereby we reproduce up to convention dependent phase factors equation (4) in [5]. For the
wave function one uses the following expression:
1
L S
(A.6)
z z (q) = LLz e(Sz ) .
2
Now the amplitude for cJ , J = 0, 1, 2, production is given by:

Z
X 1 1 J
dq [1]
L S
(a) 3
PJ , Jz =
(q)z z (q).
(A.7)
M
Lz Sz Jz
(2)3
L z Sz
With this we find for the c0 meson:

1
3
2
R10
,
M(a) 3 P0 , 0 = g 2 ij ab
1
3
Nc
2m
(A.8)
with R10 being the first derivative of the quarkonium wave function at the origin, as defined
by:
r
Z
3
q d 3q
R 0 e(Lz ).
LLz (q) = i
(A.9)
3
4m 1
(2)
Then the partonic cross section is given by:
X 1 X

2
1
M(a) 3 P0 , 0 2
1 2 3 P0 , 0 CSM =
4
2
2
2
2(2m) (Nc 1)
Nc
ab
ij
24 2 s2 |R1 |2 2
1 .
(2m)7
(A.10)
The amplitude for the c1 meson vanishes identically. For the c2 meson we find then:
r

3
(a) 3
2 2
P2 , Jz = g
ij ab
R 0 2 1 ,
(A.11)
M
Nc
2m3 1 1 Jz /2
and we obtain for the partonic cross section henceforward:
(a)
1 2
P2 , Jz
=
CSM
32 2s2 |R1 |2 2 1
1 Jz /2 .
(2m)7
(A.12)
Appendix B. On the alternative of a virtual gluon field

A virtual gluon field can be parameterized by the transversality condition:
l l0
0 = g0 2 .
l
(B.1)
111
Taking now the incoming and outgoing charm quark to be on-shell one obtains up to
velocity corrections:
m2c = (l + p)2 = m2c + l 2 + 2p l l 2 + 2l0 m + m2c ,
(B.2)
which results in:

l0
.
(B.3)
2m
If |0 | is going to be zero it requires l0 = 2m which means that the charm quark gets
a negative energy, which is unphysical. Now we can investigate how much the incoming
charm quark needs to be off-shell so that we can work with a real gluon field instead:
|0 |2 = 1 +
2l0
.
(B.4)
m
Then, for small enough l0 , we can again treat the off-shellness as a velocity correction
along the many others we have neglected in the calculation.
m2c = (l + p) = (1 )m2c + 2p l
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
G.P. Ramsey, hep-ph/0001105.

R.L. Jaffe, D. Kharzeev, Phys. Lett. B 455 (1999) 306, hep-ph/9903280.
CDF Collaboration, T. Affolder et al., hep-ex/0004027.
M. Beneke, I.Z. Rothstein, Phys. Rev. D 54 (1996) 2005, hep-ph/9603400;
M. Beneke, I.Z. Rothstein, Phys. Rev. D 54 (1996) 7082, Erratum.
P. Hoyer, S. Peigne, Phys. Rev. D 59 (1999) 034011, hep-ph/9806424.
N. Marchal, S. Peigne, P. Hoyer, hep-ph/0004234.
G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 51 (1995) 1125, hep-ph/9407339.
O. Teryaev, A. Tkabladze, Phys. Rev. D 56 (1997) 7331, hep-ph/9612301.
W.D. Nowak, O. Teryaev, A. Tkabladze, hep-ph/9711290.
V.A. Korotkov, W.D. Nowak, Nucl. Phys. A 622 (1997) 78C, hep-ph/9701371.
V.A. Korotkov, W.D. Nowak, hep-ph/9908490.
W.D. Nowak, A. Tkabladze, Phys. Lett. B 443 (1998) 379, hep-ph/9809413.
E.L. Berger, D. Jones, Phys. Rev. D 23 (1981) 1521.
R. Baier, R. Ruckl, Phys. Lett. B 102 (1981) 364.
T. Morii, S. Tanaka, T. Yamanishi, Phys. Lett. B 322 (1994) 253, hep-ph/9309336.
F. Yuan, K. Chao, hep-ph/0008302.
P. Hagler, R. Kirschner, A. Schafer, L. Szymanowski, O.V. Teryaev, hep-ph/0008316.
F. Yuan, K. Chao, hep-ph/0009224.
T. Morii, S. Tanaka, T. Yamanishi, Phys. Lett. B 372 (1996) 165.
S. Gupta, P. Mathews, Nucl. Phys. Proc. Suppl. 64 (1998) 446, hep-ph/9708479.
S. Gupta, Phys. Rev. D 57 (1998) 1858, hep-ph/9707315.
S. Gupta, P. Mathews, Phys. Rev. D 56 (1997) 7341, hep-ph/9706541.
NA3 Collaboration, J. Badier et al., Z. Phys. C 20 (1983) 101.
C. Biino et al., Phys. Rev. Lett. 58 (1987) 2523.
C. Akerlof et al., Phys. Rev. D 48 (1993) 5067.
E672 and E706 Collaborations, A. Gribushin et al., Phys. Rev. D 53 (1996) 4723.
E-771 Collaboration, T. Alexopoulos et al., Phys. Rev. D 55 (1997) 3927.
J.G. Heinrich et al., Phys. Rev. D 44 (1991) 1909.
112
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
R. Baier, R. Ruckl, Z. Phys. C 19 (1983) 251.

G.A. Schuler, Quarkonium production and decays, hep-ph/9403387.
P. Cho, A.K. Leibovich, Phys. Rev. D 53 (1996) 6203, hep-ph/9511315.
E705 Collaboration, L. Antoniazzi et al., Phys. Rev. Lett. 70 (1993) 383.
C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
E789 Collaboration, M.H. Schub et al., Phys. Rev. D 52 (1995) 1307;
M.H. Schub et al., Phys. Rev. D 53 (1996) 570, Erratum.
E672-E706 Collaborations, V. Koreshev et al., Phys. Rev. Lett. 77 (1996) 4294.
E771 Collaboration, T. Alexopoulos et al., Phys. Lett. B 374 (1996) 271.
BEATRICE Collaboration, Y. Alexandrov et al., Nucl. Phys. B 557 (1999) 3.
CTEQ Collaboration, H.L. Lai et al., Eur. Phys. J. C 12 (2000) 375, hep-ph/9903282.
M. Gluck, E. Reya, A. Vogt, Eur. Phys. J. C 5 (1998) 461, hep-ph/9806404.
M. Gluck, E. Reya, I. Schienbein, Eur. Phys. J. C 10 (1999) 313, hep-ph/9903288.
E.J. Eichten, C. Quigg, Phys. Rev. D 52 (1995) 1726, hep-ph/9503356.
T. Gehrmann, W.J. Stirling, Phys. Rev. D 53 (1996) 6100, hep-ph/9512406.

Predictive grand unified textures for quark and

neutrino masses and mixings
Zurab Berezhiani a,b , Anna Rossi c,
a Dipartimento di Fisica, Universit di LAquila, I-67010 Coppito, AQ, and INFN, Laboratori Nazionali del
Gran Sasso, I-67010 Assergi, AQ, Italy

b The Andronikashvili Institute of Physics, Georgian Academy of Sciences, 380077 Tbilisi, Georgia
c Dipartimento di Fisica, Universit di Padova and INFN Sezione di Padova, I-35131 Padova, Italy
Abstract
We propose new textures for the fermion Yukawa matrices which are generalizations of the socalled Stech ansatz. We discuss how these textures can be realized in supersymmetric grand unified
models with horizontal symmetry SU(3)H among the fermion generations. In this framework the
mass and mixing hierarchy of fermions (including neutrinos) can emerge in a natural way. We
emphasize the central role played by the SU(3)H adjoint Higgs field which reduces SU(3)H to
U (2)H at the GUT scale. A complete SO(10) SU(3)H model is presented in which the desired
Yukawa textures can be obtained by symmetry reasons. The phenomenological implications of these
textures are thoroughly investigated. Among various realistic possibilities for the Clebsch factors
between the quark and lepton entries, we find three different solutions which provide excellent fits
of the quark masses and CKM mixing angles. Interestingly, all these solutions predict the correct
amount of CP violation via the CKM mechanism, and, in addition, lead to an appealing pattern of
the neutrino masses and mixing angles. In particular, they all predict nearly maximal 23 mixing and
small 12 mixing in the lepton sector, respectively, in the range needed for the explanation of the
atmospheric and solar neutrino anomaly. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
In the last years the interest for the issue of neutrino mass generation has been renewed,
especially by the SuperKamiokande data on atmospheric neutrinos [1]. These data can be
explained by oscillation within the following parameter range (at 99% CL):
m2atm = (19) 103 eV2 ,
sin2 2atm > 0.8
E-mail addresses: berezhiani@fe.infn.it (Z. Berezhiani), arossi@pd.infn.it (A. Rossi).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 3 - 2
(1)
114
Z. Berezhiani, A. Rossi / Nuclear Physics B 594 (2001) 113168
(the best-fit values are m2atm = 3 103 eV2 and sin2 2atm = 0.99), while the explanation
through the e oscillation is strongly disfavoured [2]. On the other hand, the solar
neutrino data [3] can be interpreted in terms of e oscillation into , , or their mixture
within the following parameter range [4,5]:
m2sol = (310) 106 eV2 ,
sin2 2sol = (0.11.4) 102
(2)
(the best-fit values are m2sol = 5 106 eV2 and sin2 2sol = 7 103 ), the so-called
small-mixing angle MSW solution [6]. 1 Therefore, the experimental data provide a strong
evidence that like charged fermions, also neutrinos have masses and weak mixing. Namely,
barring the less natural possibility that the neutrino mass eigenstates 1,2,3 are strongly
degenerate and assuming that m3 > m2 > m1 , the m2 ranges in Eqs. (1) and (2) directly
translate into
1/2
= (310) 102 eV,
m3 ' m2atm

1/2
= (1.73.3) 103 eV,
(3)
m2 ' m2sol
which indicates that the neutrino mass ratio m3 /m2 is similar to that of charged fermions,
m /m or mb /ms . However, the neutrino mixing pattern inferred from (1) and (2)
is drastically different from the well-established structure of the CabibboKobayashi
Maskawa (CKM) mixing of quarks.
The explanation of the fermion flavour structure is one of the most challenging problems
of particle physics, and the neutrino case constitutes a part of this issue. In the standard
model the charged fermion masses emerge from the Yukawa terms: 2
ij
ij
ij
uci Yu qj Hu + dic Yd qj Hd + eic Ye lj Hd ,
(4)
where qi = (u, d)i , uci , dic and li = (, e)i , eic are the quark and lepton fields of three
families (i = 1, 2, 3), and Hu,d are the Higgs doublets, with the vacuum expectation values
(VEVs) vu,d determining the electroweak scale, (vu2 + vd2 )1/2 = vW = 174 GeV. There is
no renormalizable term that gives rise to the neutrino masses. However, the latter can get
Majorana masses from the lepton-number violating higher order operator
1
2
li Yij
lj Hu ,
ML
Y = YT ,
(5)
which is cutoff by some large scale ML , e.g., the grand unification or Planck scale [7]. 3
2 /M which
Hence, the charged fermion masses are vW and the neutrino masses are vW
L
makes it easy to understand why the latter are so light.
1 We concentrate mainly on this solution, whose parameter range naturally emerges in the models presented
below. One has to remark, however, that the solar neutrino data are still ambiguous and cannot discriminate other
possible solutions. In particular, the recent analysis show that the large-mixing angle MSW and long wavelength
just-so oscillation solutions have comparable statistical significance [2,5].
2 In the following we mainly deal with the supersymmetric model, though the general features of our discussion
are equally valid for the non-supersymmetric case.
3 Any known mechanism for the neutrino masses reduces to such an effective operator. E.g., in the see-saw
scheme [8] it is obtained after integrating out heavy-neutral fermions with Majorana masses ML .
115
The quark mixing is originated from the misalignment between the Yukawa matrices Yu and Yd . Analogously, the neutrino mixing is due to misalignment between Ye
and Y . In particular, one can choose a basis where Yu and Y are diagonal, Yu =
Diag(Yu , Yc , Yt ) and Y = Diag(Y1 , Y2 , Y3 ). In this basis, the quark and lepton mixing angles are determined, respectively, by the form of Yd and Ye which in general
remain non-diagonal. The latter are to be diagonalized by the bi-unitary transformations:
Ud0T Yd Ud = YD
d = Diag(Yd , Ys , Yb ),
Ue0T Ye Ue = YD
e = Diag(Ye , Y , Y ),
(6)
0 rotating
with the unitary matrices Ud,e rotating the left handed (LH) states and with Ud,e
the right handed (RH) ones. Thus, in this basis the CKM matrix of weak mixing between
the physical quark states (u, c, t) and (d, s, b) is Vq = Ud , while the matrix Vl = Ue
determines the properties of the weak current between the charged leptons (e, , ) and
the neutrinos. Namely, Vl relates the neutrino flavour eigenstates (e , , ) to the mass
eigenstates (1 , 2 , 3 ), and describes the neutrino oscillation phenomena.
The unitary matrices
Vud Vus Vub

Ve1 Ve2 Ve3
Vl = V1 V2 V3
(7)
Vq = Vcd Vcs Vcb ,
Vt d Vt s Vt b
V 1 V 2 V 3
are usually parameterised as [9]:
c12 c13
Vq,l = s12 c23 c12 s23 s13 ei
s12 s23 c12 c23 s13 ei
s12 c13
c12 c23 s12 s23 s13 ei
c12 s23 s12 c23 s13 ei
s13 ei
s23 c13 ,
c23 c13 q,l
(8)
where sij (cij ) stand for the sines (cosines) of the three mixing angles 12 , 23 and 13
while is the CP violating phase. (In the case of Majorana neutrinos, the lepton mixing
matrix contains two additional phases 1,2 , factorised out as Vl Diag(1, ei1 , ei2 ), which,
however, are not relevant for neutrino oscillations.) In the following, we distinguish the
angles and phases in Vq and Vl by the superscripts q and l, respectively.
The explanation of the fermion mass and mixing pattern is beyond the capacities of the
standard model. The Yukawa matrices Yu,d,e , are arbitrary complex 3 3 matrices, not
constrained by any symmetry property. Concerning the neutrinos, apart from the Yukawa
factors Y in (5), also the lepton-number violation scale ML is an arbitrary parameter.
The magnitude of the latter can be inferred from the mass value m3 in (3). Namely, from
2 /M we obtain
m3 Y3 vW
L
Y31 ML 1015 GeV,
(9)
which is rather close to the GUT scale MG 1016 GeV. In particular, the normalization
of the operator (5) as ML = MG implies Y3 10. Let us remark that there is no
contradiction if the effective coupling constant Y3 is large. E.g., in the context of the see-
116
Table 1
Quark and lepton mass ratios between second/first and third/second families. All mass ratios are
independent of the renormalization scale except Yt /Yc and Yb /Ys (in the case of reasonably small
tan vu /vd ). For definiteness, the latter ratios are evaluated at the electroweak scale, = MW ,
while at the GUT scale, = MG ' 1016 GeV, they are scaled respectively as Bt3 and Bt1 , with
the factor Bt = B(Yt ) varying from 0.9 to 0.7 for the top Yukawa constant at the GUT scale in the
interval YtG = 0.53 (see details in Section 4)
2nd/1st
3rd/2nd
Yc /Yu = 2501000
Ys /Yd = 1725
Y /Ye = 207
Y2 /Y1 > 1
Yt /Yc = 230330(=MW ) , 3201000(=MG )

Yb /Ys = 3085(=MW ) , 33120(=MG )
Y /Y = 17
Y3 /Y2 = 1060
2
saw mechanism [8] Y3 is determined by the ratio YDirac
/YMajorana which can easily happen
to be 10 even if the Yukawa constants YDirac and YMajorana are order 1.
One aspect of the flavour problem is related to the inter-family mass/Yukawa hierarchy
among fermions [10]. For example, the top Yukawa constant is Yt 1, while the firstgeneration constants Ye,u,d are much smaller, 105 . Concerning the neutrinos, they also
seem to indicate a mass hierarchy similar to that of the charged fermions. In particular,
Eqs. (3) show that the ratio Y3 /Y2 lies in the range 1060. There are no data restricting the
ratio Y2 /Y1 , but it may also be large. For the sake of comparison, the mass ratios between
different families are summarized in the Table 1.
On the other hand, the mixing structure is very different between quarks and leptons
(see Table 2). The most striking feature is that the atmospheric neutrino data favour the
nearly maximal 23 mixing of neutrinos, while the 23 mixing of quarks is very small. On
the contrary, the MSW solution implies a very small 12 mixing angle for neutrinos as
compared to the sizeable 12 angle of the Cabibbo mixing for quarks.
It is widely believed that the fermion flavour structure can be understood within
the context of the grand unified theories with horizontal symmetries. The latter acts
between the fermion families and thus can help to constrain the form of the Yukawa
matrices Yu,d,e, . In particular, it can dictate the hierarchical and closely aligned structures
of Yu and Yd and thus simultaneously explain the large mass splitting among different
families as well as the smallness of the quark mixing angles. However, the origin of the
large neutrino mixing is not a priori clear. Thus, the crucial point that must be explained by
any realistic flavour theory is the dramatic difference between the quark and lepton mixing
angles. 4
4 In last years many efforts have been applied to derive the neutrino mixing pattern from particular textures of
the neutrino mass matrix [11]. Differently from some point of view adopted in [11], we believe that a satisfactory
scheme can be obtained in the context of a well-defined and complete flavour theory including the charged
fermions. Such attempts have been taken, e.g., in Refs. [12] where the neutrino mixing pattern was discussed
in the context of grand unified and horizontal symmetries.
117
Table 2
l and l correspond to the 99% CL
Quark and leptonic mixing angles (in ). The values of 23
12
l comes from
regions of the AN and SN data fitting, Eqs. (1) and (2), while the upper bound for 13
q
q
the combination of CHOOZ [14] and atmospheric neutrino data [2]. The values of 12 , 23 and
q
13 reflect the present data on |Vus |, |Vcb | and |Vub /Vcb | [9]. The last two values depend on the
renormalization scale and for = MG they should be scaled by a factor Bt , i.e., become smaller by
(1030)%
12
23
13
q(uark)
l(epton)
12.7 0.1
2.3 0.1
0.21 0.03
2.4 1.0
45 13
< 15
In a previous paper [13] we have outlined that such a complementary pattern of quark
and lepton mixings can naturally emerge in the context of grand unified theories. In the
SU(5) model the quark and lepton states of each family are combined in the multiplets
5 i = (d c , l)i , 10i = (uc , ec , q)i , and the Higgs doublets Hu and Hd fit, respectively, into
The terms responsible for the fermion masses read
S 5.
the representations H 5 and H
as:
ij
2
S + 1 5 i Gij
(10)
10i Gu 10j H + 5 i Gij 10j H
5j H ,
ML
where the Yukawa constant matrices Gu and G are symmetric due to SU(5) symmetry
reasons while the form of G is not constrained. After the SU(5) symmetry breaking these
terms reduce to the standard model terms (4) and (5), with
Yu = G u ,
Y = G ,
Yd = G,
Ye = G T .
(11)
Hence, in the basis where the matrices Gu and G are diagonal, the rotation angles of
the LH leptons coincide with that of the RH quarks: Ue = Ud0 , and vice versa, Ud = Ue0 .
(One has to remark, that in grand unification the rotation angles of RH states also have a
physical sense as they define the fermion mixing in B- and L-violating currents.) Hence,
if G23 G33 while G32 G33 , the large 23 neutrino mixing will occur together with the
small 23 mixing of quarks [12,13]. In other words, the phenomenology requires that the
matrix G should have a strongly asymmetric form.
In particular, a very interesting situation emerges in the case of Fritzsch-like texture [15]
with vanishing G22 and G11 entries. In addition, the entry G33 can be taken as SU(5)
singlet thus maintaining the b Yukawa unification, while the off-diagonal entries G23
etc. should contain some SU(5) Clebsches to avoid the wrong relations Yd,s = Ye, of
the minimal SU(5). This means that the masses of the third generation emerge from the
Yukawa couplings to the Higgs 5-plet, while the lighter generation masses are contributed
also by 45-plet. In fact, it is not necessary to introduce the elementary Higgs 45-plet.
Instead, the off-diagonal entries Gij can be regarded as operators dependent on the adjoint
118
Higgs 24 of SU(5), Gij = Gij (). 5 In this case the relation Yd = YTe is not exact
anymore, but it fulfills with the precision of these Clebsch factors. Nevertheless, this
property ensures that between the 23 mixing angles of quarks and leptons the following
relation is fulfilled with a good accuracy [13]:

m ms 1/2
q
l
.
(12)
tan 23 tan 23 '
m mb
This product rule indeed works remarkably well. It demonstrates a see-saw correspondence between the lepton and quark mixing angles and tells us that whenever the neutrino
l
1, the quark mixing angle comes out small and in the correct
mixing is large, tan 23
q
range, tan 23 0.04.
Namely, in Ref. [13] we have considered the following Yukawa pattern:

0
Ad
0
Yu 0 0
0 Ad 0
Yu = 0 Yc 0 , Yd = A0d 0 Bd = a1 Ad
0
Bd ,
1
0 Bd0 D
0
B
D
0 0 Yt
b d

0
Y1 0 0
0 Ae 0
0
Ae
0
Be , (13)
Y = 0 Y2 0 , YTe = A0e 0 Be = a1 Ae
1
0
0 Be D
0 0 Y3
0
D
b Be
where the parameters b = Be /Be0 = Bd /Bd0 and a = Ae /A0e = Ad /A0d reflect the possible
asymmetry between the off-diagonal entries in Yd,e . In addition, the Clebsches Ae /Ad =
kA and Be /Bd = kB are allowed to be different from 1 and their values can be extracted
from the quark and lepton masses.
Indeed, the texture predicts the following relations between the Yukawa eigenvalues [13]:
Yd Ys
Ys Yd
1
1
Yb
(14)
= Z,
= 2 ,
= 2 ,
Y
Y Ye
Ye Y kA Z
kB Z
and
k2
md
ms
+
2 = A4
md
ms
ZkB
where

me
m
+
2 ,
me
m
(15)

m me
1 b + b1 1 kB2
.
(16)
m
In particular, the phenomenologically correct picture for the fermion masses emerges if the
asymmetry parameter b is large enough, b 10, and when kA ' 1 and kB ' 2 [13]. In this
case the relation (15) leads to the correct prediction for the strange/down quark mass ratio.
Z=
5 This should be understood as expansion series Gij () = G(0)ij + G(1)ij + , where M is some cutoff
M
mass larger than the GUT scale (e.g., the string scale). This means that the off-diagonal terms emerge from
H
S5 i G(1)ij 10j etc., just on the same footing as the last term in (10).
the effective higher-order operators M
S is represented by the tensor product 24 5 = 5 + 45, it can distinguish the
Since in general the operator H
corresponding entries in the matrices Ye and Yd and hence avoid the leptonquark degeneracy of the minimal
SU(5) theory.
On the other hand, for the 23 mixing angles in (7) we obtain the relations:
r
r
m
ms
q
l
' b1/2
,
tan 23 ' b1/2
,
tan 23
m
mb
119
(17)
from which the product rule (12) is immediate. These relations also point to a large
asymmetry factor b. 6 In particular, for b 10 the neutrino mixing becomes nearly
l 45 , whereas the quark mixing angle gets small, q (23) .
maximal, 23
23
In the general case, for arbitrary a, the product rule similar to that in (17) can be obtained
also for 12 mixing angles. However, from phenomenological grounds one can deduce that
no significant asymmetry should occur in the 12 sector of the matrices (13). Namely, the
successful relation for the Cabibbo angle:
r
md
q
,
(18)
tan 12 '
ms
indicates that Ad ' A0d , so that a = 1 is appropriate. Interestingly, in this case we obtain
for the leptonic 12 mixing [13]:
r
l
l me
' cos 23
.
(19)
tan 12
m
l
l
45 , this implies 12
3 , which is in the range relevant to the MSW
Given that 23
solution of the SN problem (cf. Table 2).
Therefore, for a = 1 the ansatz (13) depends on six parameters, so it is highly predictive.
Among these, three Yukawa entries Ae , Be , D can be expressed in terms of lepton
couplings Ye,, and b. Thus 9 physical quantities, the Yukawa eigenvalues Yd,s,b and
the mixing angles in Vq and Vl , are left as functions of three Clebsch factors, kB , kA
and b. Hence, at the GUT scale these are connected by six relations like (14), (17), etc.
As discussed above, in the first approximation these relations well reproduce the observed
pattern of the quark and lepton mixing angles for b 10 and kB ' 2, kA ' 1. However,
a closer inspection shows that for precise Clebsch values kB = 2 and kA = 1, which
could be motivated in the GUT context, and a = 1 which could follow from the U (2)H
horizontal symmetry, the predictions for |Vus | and |Vcb | substantially deviate from the
corresponding experimental values which presently are known with a very good accuracy.
The case summarized above implies a real CKM matrix since Yu is taken diagonal and
so all phases in Yd can be absorbed by the field re-definitions. Nevertheless, it would be
interesting to address the CP issue in a scenario in which Yu is still kept diagonal.
In the present paper we suggest a new grand unified ansatz for the fermion masses,
which is an extension of the pattern (13) of Ref. [13]. Namely, we consider the case when
the matrices Gu and G are both diagonal and related as G b01 Gu , with b0 being
a diagonal matrix b0 = Diag(1, 1, b0). Regarding G, we assume that it has the diagonal
part Gu and the off-diagonal one b1 A(), where A is an antisymmetric matrix
6 The symmetric case b = 1, i.e., Y
d,e having the familiar Fritzsch texture, is obviously excluded, since the 23
q
l = 13.7 (cf. the experimental
mixing of quarks is too big, 23 = (610) , while it is too small for leptons, 23
values in Table 2).
120
with -field dependent entries inducing different Clebsch factors for the down quarks and
charged leptons, and b = Diag(1, 1, b). In other words, we consider the Yukawa textures:
Yu = YD
u,
Yd = Yu + b1 Ad ,
Y = b01 Yu ,
YTe = Yu + b1 Ae ,
(20)
where and are proportionality coefficients. For b = 1 this pattern resembles the
so-called Stech texture proposed long time ago by Stech [16] and independently by
Chkareuli and one of the authors [17], where the off-diagonal entries of Yd were assumed
to be antisymmetric, Yd = Yu + Ad . This case, however, is completely excluded by
experimental data. As for our texture, the matrices b1 Ad,e remain antisymmetric only
in the 12 block, while the other entries are skew due to the factor b 6= 1. In the explicit
form the Yukawa matrices read as:
Yu
Ad
Cd
Yu 0 0
Yt D
Yc
Yd = Ad
Yu = 0 Yc 0 ,
Bd ,
Yt D
0 0 Yt
b1 Cd b1 Bd D
Yu
Yu 0
0
Ae
Ce
Yt D
Yc
(21)
YTe = Ae
0 ,
Y = 0 Yc
Be .
Y D
t
1
b0 Yt
b1 Ce
b1 Be
Clearly this texture represents an extension of the Fritzsch-like zero-texture (13)

considered in [13]. In particular, for vanishing 13 and 31 entries, Cd,e = 0, this texture
essentially reduces to the latter as the diagonal 11 and 22 entries in Yd,e (20) are quite
small, since the ratios Yu /Yt and Yc /Yt are much smaller than 1/3 and 2/3 Yukawa ratios
in the down quark and charged leptons. In practice, as we show below, one can safely
take the limit Yu /Yt 0 and ignore the 11 entries in Yd,e . As for the 22 entry, though it
is small, it can provide significant corrections (in particular, to Vcb ). In principle, it also
contains the relative (un-removable) CP violating phase which however cannot provide a
sufficient amount of CP violation. This can be remedied by non-vanishing Ce,d .
Moreover, we show that for large values of b ( 10), the texture (21) provides a very
appealing description of the quark and lepton mass and mixing structures. Interestingly,
the case b 0 b 10 also provides nice relations between neutrino and up-quark masses,
Y2 /Y3 = b0 Yc /Yt (cf. Table 1).
Hence, the complete pattern (21) is first motivated on the phenomenological point of
view as it can fit the physical observables with excellent precision. From a theoretical
point of view, we show that these textures can be justified in the framework of grand
unification by invoking also the concept of horizontal symmetry among three fermion
families [17,18].
The paper is organized as follows. In Section 2 we discuss some theoretical tools
for building-up the textures (20) in a realistic and predictive manner in the framework
of supersymmetric SU(5) model with SU(3)H horizontal symmetry. The discussion in
Section 3 is meant to provide an existence proof of those realizations in the framework
of SO(10) SU(3)H model. Three particular anstze are singled out, characterized by
121
different assignments of the Clebsch factors between the quark and lepton entries, and in
Section 4 their phenomenological analysis is presented and discussed in all details. Finally,
we summarise our findings and conclude.
2. Theoretical framework: GUT and horizontal symmetry

In this section we would like to sketch the general ideas needed to obtain the Yukawa
textures (20) in the SU(5) model with the SU(3)H horizontal symmetry [17,18]. In the
course of the presentation, it will become apparent that the discussion has to take a less
general character as for example some additional symmetries should be invoked. This
aspect will be faced in Section 2.2. In the discussion SU(5) is taken as a prototype theory
featuring the basic properties of grand unification. In a more general context one could
think of GUTs based on larger groups (e.g., SU(N) with N > 6) which contain SU(5) and
perhaps also unify both SU(5) and SU(3)H . A particular model based on SO(10)SU(3)H
will be presented in the next section.
2.1. General aspects
We consider the supersymmetric SU(5) model with the horizontal symmetry SU(3)H
where three fermion families are unified in chiral superfields:

3),
10i = uc , q, ec i (10, 3),
(22)
5 i = d c , l i (5,
(i = 1, 2, 3 is SU(3)H index), while the Higgs superfields 5, 5 are singlets of SU(3)H :
1).
S (5,
H
H (5, 1),
(23)
We also assume that the theory is invariant under R-parity. In other words, we impose the
matter parity Z2 under which the matter superfields change the sign while the Higgs ones
are invariant. 7
Since the fermion bilinears transform as 3 3 = 3 + 6, their standard Yukawa
couplings to the Higgses are forbidden by the horizontal symmetry. 8 Hence, the fermion
masses can be induced only by higher order operators involving a set of horizontal Higgs
ij
superfields Xij in two-index representations of SU(3)H : symmetric Xs S {ij } (1, 6)
and antisymmetric Xa A[ij ] = ij k Ak (1, 3): 9

ij
7 The SU(3) symmetry itself does not not prevent the R-parity violating couplings ij k 10 5 5 from where
H
i j k
the BL violating terms uc d c d c , qd c l, ec ll arise. For a general discussion on the relation between R-parity and
horizontal symmetries, see [19]. In particular, in case of the horizontal symmetry SU(4)H the R-parity could
emerge as an accidental symmetry. One has to remark, however, that in many cases considered below the matter
parity automatically follows from the discrete symmetries imposed on the model.
8 The following discussion can equally apply in both the cases of global and local horizontal symmetry, though
it would be more appealing to regard SU(3)H as gauge symmetry, like SU(5).
9 The theory may also contain conjugated Higgses X
Sij in representations S
These
S (1, 6) and A (1, 3).
usually are needed for writing non-trivial superpotential terms in order to generate the horizontal VEVs (see
Appendix A). The conjugated fields, however, do not couple the fermion superfields (22) and thus do not
contribute to the quark and lepton masses.
122
ij
S ij
S ij + Aij
S, O = S 5 i 5 j H 2 ,
10i 10j H, O =
5i 10j H
(24)
M
M
M2
where M is some large scale (the flavour scale). Needless to say, by SU(5) symmetry
reasons, the antisymmetric Higgses A can participate only in O. Therefore, in the lowenergy limit the operators (24) reduce to the Yukawa couplings:
Ou =
Yu
hSi
,
M
Yd , YTe
hSi + hAi
,
M
Y
hSi
.
ML M 2
(25)
In this way, the fermion mass hierarchy can be naturally linked to the hierarchy of the
horizontal-symmetry breaking scales [17,20]. For more details on the horizontal VEV
structures, see, e.g., [21].
In particular, let us assume that the horizontal Higgses include one sextet S with a VEV
taken diagonal:
0
S1 0
ij
(26)
S = 0 S2 0 , S3 S2 S1 ,
0
0 S3
ij
and a set of triplets An ij k Ank , n = 1, 2, 3, which in general have the VEVs towards
all three components:
0
A3 A2
X
ij
(27)
An = A3
0
A1 , A1 > A2 > A3 ,
n
A2 A1 0
(see Appendix A). Then from (25) we see that Yt 1 implies S3 M which indeed
can naturally arise from the Higgs sector as we show in Appendix A. Similarly one can
expect that also Yb, 1 which would require large tan regime. However, as we shall see
below, in realistic schemes also small tan can be naturally accommodated. In addition,
the operator O translates into Y3 /ML S3 /M 2 Yt /M. Therefore, the flavour scale M
is of the order of the BL violating scale ML 1015 GeV (cf. (9)). The closeness of the
flavour scale with the GUT scale MG 1016 GeV suggests that they may have a common
origin. (The mismatch between the estimate of M and MG can be due to some spread of
the coupling constants in the theory.)
It is well known that the effective operators (24) may arise entirely from a renormalizable
theory, after integrating out some heavy degrees of freedom [22], and thereby the flavour
scale can be related to the mass scale of the latter. It is also plausible to think that this mass
scale is set by large (order GUT scale) VEV of some Higgs , hi M.
In particular, one can introduce vector-like states in the 10- and 10-representations of
SU(5):

i
T i = U, Qc , E i (10, 3),
(28)
T i = U c , Q, E c (10, 3),
and the SU(5)-singlet fields
N i (1, 3),
Si (1, 3),
N
and consider the following Yukawa couplings in the superpotential:
(29)
123
Fig. 1. The heavy fermion exchanges giving rise to the effective higher order operators for the fermion
masses: (a) for up quarks, (b) for down quarks and charged leptons, (c) for neutrinos.
ST + TSX10 + T TS,
WT = 10H T + 5 H
N + Xs N
S
S2 + N N
WN = 5H
(30)
(family indices are suppressed, and order 1 coupling constants are understood in each
term). Clearly, in WT the horizontal Higgs X can be both of the type Xs = S or Xa = A,
whereas in WN only the symmetric Higgses Xs can participate. Since T , TS contain heavy
states with electric charges 2/3, 1/3 and 1, their exchange can induce the operators Ou
and O relevant for the masses of all charged fermions: up quarks, down quarks and leptons.
Analogously, operator O for the neutrino masses emerges via the exchange of the singlet
S. The relevant diagrams are shown in Fig. 1.
states N, N
Therefore, after integrating out the heavy states, we obtain the operators (24) in the form
1
T
Ou = 10 H M1
T X + X MT H 10
1
1
= 2 10H M1
T S)10 + 10H (MT A AMT 10,

S M1 S + M1 A 10,
SM1 X 10 = 5 H
O = 5 H
T
M1 SM1 H 5,
O = 5H
N
N
(31)
where MT , MN are 3 3 mass matrices of the heavy states, induced from their couplings
to the Higgs . Therefore, if is a gauge singlet, then MT , MN will be degenerate in
family space, MT ,N M 1, where 1 = Diag(1, 1, 1) is the unit matrix of SU(3)H . In
this case, if X are SU(5) singlets, only the symmetric fields Xs = XsT can contribute Ou
in (31), while the contributions of the antisymmetric ones Xa = XaT cancel out. On the
other hand, both Xs and Xa can contribute to O. Therefore, the operators (31) reduce to
the Yukawa couplings of the form:
Yu M1
T hSi,
1
Yd , YTe M1
T hSi + MT hAi,
ML1 Y M2
N hSi.
(32)
124
More explicitly, these can be written as:

Yu S,
Y S,
Yd = YTe S + A,
(33)
where S = M 1 hSi and A = M 1 hAi are, respectively, the symmetric and antisymmetric
matrices reflecting the form of the VEVs (26) and (27). (We omit the dimensionless
factors which keep track of the coupling constants in (30) and are not specified at the
moment).
It should be immediately noted, however, that the texture (33) is far from being realistic
for many reasons:
q
(A) Yd Yu +A with antisymmetric A leads to very big 23 mixing, s23 ' (Ys /Yb )1/2
0.10.2, much above the experimental value of Vcb .
(B) The relation Y Yu implies Y1 : Y2 : Y3 = Yu : Yc : Yt and thus too big a hierarchy
between the neutrino mass eigenvalues m3 and m2 (cf. Table 1).
(C) Finally, Yd = YTe features the minimal SU(5) degeneracy of the Yukawa couplings,
Yd,s,b = Ye,, , whose drawbacks have been already recalled.
The latter problem can be easily cured by making use of the adjoint superfield which
breaks SU(5) down to SU(3) SU(2) U (1) by the VEV proportional to the hypercharge
generator: hi y Diag(1/3, 1/3, 1/3, 1/2, 1/2). Thus if the matrix A in (25) effectively contains , A = A(), effective Clebsch factors emerge discriminating quarks
from leptons. Therefore, at least some of the antisymmetric Higgses A can be taken in the
mixed representation A (24, 3). 10 The tensor product is 5 24 = 5 + 45 + 70, where
only the first two terms are relevant for the fermion bilinear 5 10 = 5 + 45. Then the
unwanted relations Yd,s = Ye, can be avoided while maintaining the successful relation
Yb ' Y .
The first two problems can be solved in a similar way, by making use of the adjoint
Higgs of SU(3)H . The masses of the heavy fermions transform as 3 3 = 1 + 8
representations. Hence one can assume that the set of Higgses contains, besides the
singlet I (1, 1), also the superfield (1, 8) with the VEV hi pointing towards the 8
generator: 11
1 0 0
hi M 0 1 0 .
(34)
0 0 2
This VEV breaks SU(3)H down to SU(2)H U (1)H and thus discriminates the third
fermion family from the others [23]. Hence, the heavy-fermion mass matrices MT ,N
10 Alternatively, one can take all the antisymmetric Higgses as SU(5) singlets, A (1, 3), but assume that some
S (cf. (24)) so that
of them act in combination with (24, 1), to end up with the effective operator A2 5 10H
M
S A effectively contains both (5,
3) and (45, 3) representations.
Such an attitude will be
the tensor product H
taken in constructing the SO(10) model in Section 3.
11 In the following of the paper, the symbol stands for a set of Higgs fields in real representations of the
gauge groups. When needed the content of will be specified as singlet I (1, 1), octet (1, 8) or 24-plet
(24, 1).
125
maintain only the SU(2)H invariant form M = M Diag(1, 1, b), with b in general different
from 1. In this way, from (32) we can obtain the desired Yukawa pattern (20). 12
Notice however, that for achieving our goal, we have implemented the adjoint structures
( into M and into A) in Eq. (32) which is not the right place. Indeed, (32) follows
from (31) if and only if neither M nor A contain adjoint representations of SU(5) or
SU(3)H . For example, if A contains the SU(5) adjoint, then the tensor product A H
(24, 3) (5, 1) effectively contains the (45,3) representation which would induce offdiagonal antisymmetric contribution also to Yu . On the other hand, if As are SU(5) singlet,
but MT contains the SU(3)H octet, then (1, 3) (1, 8) (5, 1) effectively contains (5, 6)
which would still give rise to symmetric off-diagonal entries in Yu . Therefore, in either
case the antisymmetric VEVs hAi contribute the first term in (24) and so induce offdiagonal entries in Yu ruining in such a way the desired pattern. Tackling this issue requires
more theoretical ingredients and tools which will be discussed below.
2.2. Some specific realizations
Thus, we have seen that in order to arrive at the desired textures (20), both the SU(3)H
singlet I (1, 1) and octet (1, 8) are needed for generating the heavy fermion
masses. In addition, the diagrams contributing via the S and A type Higgses are to be
differentiated by some additional symmetry so that S will contribute to all the matrices
Yu , Y and Yd,e , while As only to Yd,e . Hence, we have to assume that besides the
local SU(5) SU(3)H symmetry, the theory is invariant under some set G of abelian
symmetries which may contain, e.g., non-anomalous or pseudo-anomalous gauge U (1)
factors, continuous or discrete R-symmetry, discrete symmetries like Z2 , Z3 , etc.
First, on the basis of the considerations outlined in the previous section, we put forward
the needed couplings in the superpotential. Second, those will be motivated by some
additional symmetry of G.
Now about the first step. We assume that the flavour scale is determined by the VEVs
of the singlet I (1, 1) and the octet (1, 8), respectively, hI i = M and hi =
xM Diag(1, 1, 2), x 1. Then the combination I + provides order M masses to the
heavy fermions, splitting the third generation from the first two and maintaining the latter
degenerate. As in the previous subsection, we take the horizontal VEVs in the form (26)
and (27) and define the dimensionless VEV matrices as S M 1 hSi and An M 1 hAn i.
We further assume that the superpotential WT (30) related to heavy 10-plets (28)
involves some set of horizontal Higgses X, which in fact consists of the sextet S (1, 6).
Therefore, WT becomes:
ST + T TSX10 + T (T I + T )TS,
WT = f 10H T + g 5 H
(35)
12 In group-theoretical language this can be rephrased as follows. The tensor product A reads as 3 8 =
3 + 6 + 15 where both 3 and 6 can match the fermion bilinears 3 3. In this way, the off-diagonal entries in Ye,d
are not anymore antisymmetric only the 12 sector keeps on the antisymmetry owing to the residual SU(2)H
symmetry.
126
where we explicitly indicate the order 1 coupling constants. So the diagonal entries in Yu
and Yd,e are induced, respectively, via the diagrams (a,b) of Fig. 1. In order to respect the
b Yukawa unification, the 24-plet (24, 1) should not couple to T TS.
The couplings in WN are left as in (30) and they induce Y via the diagram (c) in
Fig. 1:
N + N S N
S 2 + N(N I + N )N
S.
WN = h5H
(36)
Thus, Y is diagonal in the VEV basis (26).

Let us consider another set of horizontal Higgses X0 , which contains antisymmetric
fields An in the representations (1, 3) and (24, 3), and assume that they contribute only
to Yd,e via the exchange of other heavy fermions. The simplest possibility is to introduce
additional states in 5- and 5-representations

of SU(5):

i
i
c
3),
S = D , L (5,
Fi = D, Lc i (5, 3),
(37)
F
having the following couplings in the superpotential:
0 F + g0 F
S,
SH
S 0 10 + F (F I + F )F
WF = F 5X
(38)
S 0 is another Higgs 5-plet

(or 45-plet) of SU(5). The corresponding diagram is
where H
shown in Fig. 2(a). In the following, such a scenario will be called F -scheme.
S one can involve an additional pair of 10-plets
Alternatively, instead of the F + F

i
TSi 0 = U 0 , Q0c , E 0 i (10, 3),

(39)
T 0i = U 0c , Q0 , E 0c (10, 3),
with the superpotential terms:

S 0 T 0 + 0T TS 0 X0 10 + T 0 T0 I + T0 TS0 .
WT 0 = g 0 5 H
(40)
The corresponding exchange is shown in Fig. 2(b). This scenario will be referred to as the
T 0 -scheme.
In either scheme, the standard Higgs doublet Hd should be regarded a superposition of
S and H
S 0 : H d = c H
S 0 s H
S2 , which is rendered light after
the doublet components in H
2
S 0 + c H
S2 is
arranging the doublet-triplet splitting, while the orthogonal combination s H
2
left heavy, with mass MG (here is a mixing angle and c (s ) = cos (sin )). On the
other hand, we assume that Hu comes entirely from H , while the field H 0 , though present,
S and H
S0
is just a spectator in the Yukawa sector. On the contrary, Hd is contained in both H
with weights s and c , respectively.
S 0 is
Once the fields An = (Xa )n contain the mixed representation (24, 3) and/or H
T
45-plet, the diagrams in Fig. 2 induce off-diagonal entries in Yd and Ye with different
Clebsch factors.
Let us consider in some more details, e.g., the F -scheme. After substituting the VEVs
of the horizontal Higgses, the couplings (35), (36) and (38) give rise to the following fielddependent mass matrices for the up quark and neutrino states:
q
0
Qc mu
U c fHu
uc
Q
fHu
MT
0
mTu
0 ,
MT
N hHu
0
Nc
N
hHu
0
MN
Nc
0
MN ,
m
(41)
127
Fig. 2. The diagrams inducing the off-diagonal contributions to Yd,e : (a) exchange of 5-plets
(F -scheme), (b) exchange of additional 10-plets (T 0 -scheme).
and for the down quark and charged lepton states:

q
0
dc
Qc mu
S0
D c g0 d H
2
Q
S2
gH
MT
0
md
0 ,
MF
ec
0
S2
E c gH
Lc
mTl
E
mTu
MT
0
S0
g0 e H
2
0 ,
MF
(42)
S 0 . Namely, e = d
where d,e are the Clebsch factors dependent on the SU(5) content of H
0
0
S is 5-plet
S is 45-plet. In the above, each entry is a 3 3 matrix in

if H
and e = 3d if H
the SU(3)H space. In particular f = f 1, h = h1, etc. are flavour-blind (SU(3)H degenerate)
matrices.
As long as the heavy fermion mass matrices are induced by I and , they will be
diagonal and have SU(2)H invariant shape M = a Mb ( = T , F, N ), where
1 0 0
2x
.
(43)
b = 0 1 0 , a = + x , b =
+ x
0 0 b
These will give rise to the deformation factors b and b0 in (20) whose relations with the
factors b are easy to catch at any rate they will be shown later.
Thus, the main information on the fermion flavour pattern is contained in the matrices
mu,d,l, . The matrices mu, are proportional to the VEV hSi = Diag(S1 , S2 , S3 ), mu =
T MS and m = N MS. As for the matrices md,e , they are antisymmetric and emerge
from the VEVs of the triplets An sketched in Eq. (27). As far as the latter set contain
(1, 3) and (24, 3) representations, all those entries in general are non-trivial 5 5 matrices,
A k = Ak (1 + k y ), where 1 and y are the unit and hypercharge matrices in SU(5) space,
respectively, and k measures the relative weight of the singlet and 24-plet VEVs at each
component (k = 1, 2, 3). Thus, the corresponding entries in md and mTl differ by Clebsch
factors Ckd,e = 1 + k yd,l , where yd = 1/3 and yl = 1/2 are the hypercharges of the d c
and l fragments of 5-plets.

Therefore, after integrating out the heavy states, we obtain an effective low energy theory
with the following Yukawa matrices:

T 1
Yu = f M1
T mu + mu MT ,
1
0
Yd = gs M1
T mu + g c d md MF ,
1
0
YTe = gs M1
T mu + g c e ml MF
(44)
128
and
Y
= h2 M2
N m .
ML
(45)
Therefore, Yu and Y get contribution only from S:

Yu =
Y
h2 N 2
b S,
= 2
ML
aN M N
2f T 1
bT S,
aT
(46)
and thus both have diagonal forms related as:

Yu
Y
,
= b01
ML
M
(47)
where b0 = Diag(1, 1, b0 ), and

b0 =
2
bN
,
bT
h2 N aT
.
2
2f T aN
(48)
As for Ye,d , they have a diagonal contribution gs /(2f )Yu from the first term in (44),
and off-diagonal entries An b1
F from the the second term. Therefore, the full matrices Yd,e
can be rewritten as
Yd = Yu + b1 Ad ,
YTe = Yu + b1 Ae ,
(49)
g
s and b = Diag(1, 1, b), with b = bF1 , while the other dimensionless
where = 2f
S 0 and A, are
coefficients, such as the Clebsch factors related to the representations of H
absorbed into the definition of the entries of the antisymmetric matrices
0
Ad,e
Ad,e
3
2
d,e
(50)
Ad,e =
0
Ad,e
1 ,
A3
d,e
d,e
A2
A1
0
so that the relative Clebsch factors are defined as ki = Aei /Adi = e Cie /d Cid . Putting
everything together, we can write the Yukawa matrices as:
Yu
Ad3
Ad2
Yu 0 0
Yd = Ad
Yu = 0 Yc 0 ,
Yc
Ad ,
3
Yu
Y = 0
0
0
0
Yc
0
Yt
0
0 ,
1
b0 Yt
b1 Ad2
Yu
T
Ye = Ae3
b1 Ad1
Yt
Ae3
Ae2
b1 Ae2
b1 Ae1
Yc
Ae1 .
(51)
Yt
The following comments are in order.

First of all, Yt 1 implies that S3 bT M, while the relations Yu : Yc : Yt 2 : :
1
bT , well fits the upper quark mass hierarchy for 102 and bT 0.11. As already
noted in Section 2.1 this situation can naturally emerge from the pattern of the horizontal
Higgs superpotential (see Appendix A).
129
Since the (3,3) element in Yd,e is also related to largest scale S3 , we have approximate
b Yukawa unification at the GUT scale: Y ' Yb ' Yt . On the other hand, as in
S and H
S 0 can be small, i.e.,
principle the mixing angle between the doublets in H
sin < 1, the case Yb /Yt 1 is plausible even if the constants f and g are both of
order 1. Therefore, the regime of moderate tan is quite a natural possibility, in which
case the renormalization group equations can be substantially simplified. In the following
we assume that regime, though the case with large tan should not be excluded.
The large 23 lepton mixing implies Ae1 Y . This also seems natural in the context of
the model as the largest triplet VEV A1 should be of the order of the flavour scale scale M
(see Appendix A). On the other hand, the small 23 mixing of quarks needs Ad1 /b < 0.1Yb ,
i.e., a large asymmetry parameter, b 10, which in turn implies bF 0.1. In this case we
observe an inverse hierarchy between the first two and third family of the heavy F states,
F > MF .
i.e., M1,2
3
Concerning the neutrino Yukawa eigenvalues, they obey the relations
Y1 Yu
=
1,
Y2
Yc
Y2
Yc
= b0
Y3
Yt
(52)
which also points to large b 0 (cf. Table 1), b0 5100. In particular, the relation b0 = b
could work, which can occur if bT = bF and bN = 1 that is when the same singlet-octet
combination I + acts in WT and WF , while in WN only the singlet I is present (N = 0
in (36)).
On the other hand, we obtain (cf. (9))
M
ML
= 1 b0
1015 GeV.
Y3
Yt
(53)
Since Yt 1 and b0 10, this translates into 102 if M 1016 GeV, i.e., order GUT
scale. It should not be, however, very surprising to find 102 due to some spread of the
coupling constants in the theory (cf. (48)).
Let us remark that the set X appearing in Eq. (35) could contain also a triplet Higgs A3
ij
(1, 3) having a VEV towards the third component, hA3 i = ij 3 A3 . This would induce
antisymmetric 12 entries in the matrix mu . This, however, would not alter the diagonal
T 1
form of Yu , since in the combination M1
T mu + mu MT the antisymmetric entries cancel
out (thanks to the residual SU(2)H symmetry in MT ). However, in Yd,e this will induce
antisymmetric 12 entries with Clebsch factors ke = kd .
The T 0 -scheme can be developed along the same lines. The only difference will be that
in this case we have b = bT 0 , so that there should be a direct hierarchy between the first
T 0 < M T 0 . In this case b 0 = b can
two and third family of the heavy T 0 states, i.e., M1,2
3
be obtained if bT 0 = bN = bT , i.e., when WT ,T 0 ,N share the same I + combination.
Alternatively, when bT 0 = bN but bT = 1 (in WT contributes only the singlet I ), we
would obtain another interesting possibility b0 = b2 which can be also compatible with
the experimental data. Namely, it would point towards smaller 3/2 hierarchy between the
neutrino mass eigenvalues: Y3 /Y2 10 (cf. Table 1).
We now come to the second step to prove that the superpotentials (35), (36) and (38) can
be justified by some extra symmetry G. First of all, we assume that G contains a discrete
130
symmetry R under which all superfields in the theory as well as the superpotential change
sign. This will allow only trilinear terms in the superpotential (provided that they are
consistent also with SU(5)SU(3)H and other symmetries in G), while some singlet Y can
have also linear term 2 Y . In this situation all VEVs (including the GUT and horizontal
ones) can emerge from a single mass scale .
Second, we introduce a Z3 symmetry acting on the superfields as exp( 2
3 iQ ),
with the following charges Q :
S 0; T ,
Q = 0 : Y, , H
F, F
S, N,
Q = 1 : I, , S
S, An , H, H 0 ; 5,
S; 10, T
S, N ,
Q = 1 : I, S, A n , H
(54)
where as agreed (24, 1), (1, 8) and I, I and Y are some singlets, (1, 1).
The most general superpotential invariant under R Z3 is restricted to have the
following form:
WHiggs = 2 Y + Y 3 + Y 2 + 3 + Y I I + I 3 + I3 + I 2 + 3
+ Y (S S
S + An A n ) + S 3 + S
S 3 + A1 A2 A3 + A 1 A 2 A 3
S + H 0 (Y + )H
S + H 0I H
S0
+ H (Y + )H
(55)
and
ST + T (I + )TS + TSS10
WYuk = 10(H + H 0 )T + 5 H
N + NY N
S + SN
S 2,
S+ F
SH
S 0 10 + 5H
+ 5AF
+ F (I + )F
(56)
respectively, in the Higgs and Yukawa sectors.

It is worth observing the following features of the Higgs potential. It has only one
mass scale (in the linear term) which determines the physical scales present in the theory
(GUT, flavour, etc.). Notice that among other degenerate minima, this superpotential allows
a solution when all singlets, adjoints and sextet/triplets get VEVs linked to the scale ,
modulo unspecified coupling constants in (55). For example, the superpotential terms for
S and S
S considered in Appendix A, emerge from here with hY i.
There are no couplings between S, A and , so the superpotential does not fix the
relative orientation of their VEVs. However, that will be fixed by supersymmetry breaking
terms,
the lines discussed in the Appendix A. For example, these can be D-terms like
R along
1
4 z
z
[
S S + n An An ] etc., where z = m 2 stands for supersymmetry
d
2
M
breaking spurion. Depending on the form and sign of the coupling constants, these terms
can fix the relative VEV orientation of , S and An in the SU(3)H space. In particular,
the choice 1,2 > 0 favours orthogonal VEV directions of A1,2 and , while < 0
prefers the parallel VEV directions of S and . Thus, in this case the largest entry in S
is positioned at the same place as in . The same for other fields. A typical consequence
of the fact that the horizontal VEV configuration is dynamically fixed by small terms, is
the presence of light horizontal pseudo-Goldstone Higgses in the theory, reminiscent of the
familons, even if the horizontal symmetry is local [23]. As for the SU(5) part, the adjoint
131
gets a VEV and, as far as the superpotential contains also 3 term, no light fragments
of are left behind.
S 0 is not present in (55) although it seems to be allowed by the
Notice that the term H I H
symmetry. The point is that if H 0 is a 5-plet, then as it has the same charges as H , that
S 0 are 45-plets,
term can be rotated away by redefinition of H and H 0 . Instead, if H 0 and H
then that term is forbidden by the SU(5) symmetry. Thus, after substituting the VEVs, the
S, H
S 0 have the form:
bilinears of the doublet and triplet fragments in H, H 0 and H

1 Y + 2 0
,
(57)
3 Y + 4 I
where 1 , 2 , . . . are O(1) coupling constants previously understood. For rendering the
Higgses Hu,d light, the 11 element of (57) in the corresponding doublet sector has to be fine
tuned by canceling big contributions between 1 Y and 2 . (Then the triplet components
will remain heavy.) Therefore, Hu is contained entirely in H whereas Hd emerges as
S2 + c H
S 0 , with tan
S and H
S 0 , Hd = s H
a superposition of the doublet fragments in H
2
hI i/h3 Y + 4 i, while the other combination of these doublets is superheavy.
The Yukawa superpotential (56) contains all terms of Eqs. (35), (36) and (38) which lead
to the Yukawa pattern (49). Furthermore, some specific features have arisen. The heavyS mass matrix emerges only by the VEV of singlet Y , so bN = 1 and thus b0 = b1 .
states N
T
Then, if I + acts in the same reducible combination in WT and WF , i.e., bT = bF
then b0 = b, which is phenomenologically welcome (52). For example, this could be the
case if the reducible combination I + emerges from an adjoint Higgs of a larger group
containing SU(3)H , e.g., SU(N + 3)H . Following similar lines of reasoning, also the T 0
scheme can be justified.
Finally, in either schemes, the Clebsch coefficients kn = Aen /Adn , n = 1, 2, 3, depend
S 0 . Needless to say that whatever
on the SU(5) content of the tensor products An H
values of the Clebsch factors can occur when the Higgses An are in (1 + 24, 3) with
unspecified weights n of 1 and 24 component. However, it would be more attractive
(though it as a matter of taste) to find a more familiar origin of the Clebsch factors
phenomenologically needed. In other words, we assume (on similar grounds as, e.g.,
in Refs. [24]) that different entries in the Yukawa matrices are differentiated by some
clean Clebsch factor kn in which the VEV structure of the original GUT is encoded
through a specific representation of the intermediate heavy fermions (e.g., F or T 0 in
our case). Therefore, let us think that at each entry, the Clebsches kn can have some
specific rational values such as 0, 13 , 23 , 1, 32 , 2, 3, etc., possibly related to
various charges (electric, hypercharge, BL etc.).
S 0 is in 5 or 45-channel
In particular, in the context of SU(5), whenever the product An H
0
S
(e.g., simply when A is singlet, A (1, 3), and the Higgs H is in 5 or 45 representations),
we have kn = 1 and kn = 3, respectively. In Table 3 we report the values for kn attainable
for A (1, 3) and A (24, 3) (or M 24), too.
As in the next we shall consider in detail specific Clebsch patterns for the Yukawa
matrices, characterized by different values of kn , now we would like to keep the discussion
on the Clebsch factors more general to show how we arrive at those specific choices.
132
Table 3
Possible values for the Clebsch factors kn = Aen /Adn , as they can emerge in the F - and T 0 -versions
S 0 and A and to the heavy-state
for various assignments of the SU(5) representations to the Higgses H
masses M. The case when both A, M 24 gives the same pattern as A, M 1. Clearly, in each case
S 0 5 and H
S 0 45 differ just by a factor 3
the Clebsches obtained for H
SU(5) rep.
A 1,
M 1
A 24,
M 1
A 1,
M 24
F -scheme
H 0 5
H 0 45
1
3
3/2
9/2
2/3
2
T 0 -scheme
H 0 5
H 0 45
1
3
6
18
1/6
1/2
Indeed, we could think of all these different anstze in terms of larger vertical and/or
horizontal symmetries. Namely, in these cases SU(5) may emerge just a subgroup of some
larger GUT symmetry group. (The SO(10) example will be considered in the next section). Then the adjoint representation would generally emerge accompanied by a singlet
partner. Hence, the corresponding combination 1 + 24 can have a VEV with two different
eigenvalues
h1+24 i = Diag(x, x, x, y, y).
(58)
In the context of F -scheme (exchange of heavy 5-plets) this would produce a Clebsch k = y/x. In the case of 10-plet exchange (T 0 -scheme) the Clebsch value would be
k = 2y/(x + y).
For example, in the context of the SU(6) model, which, as a matter of fact, provides
a natural doublet-triplet splitting, the adjoint Higgs (35-plet) of SU(6) plays a central
in terms of SU(5) subgroup) has a VEV
role [25]. This 35-plet (35 = 1 + 24 + 5 + 5,
Diag(1, 1, 1, 1, 2, 2). Therefore, in the context of F -like scheme it can provide
a Clebsch factor k = 2, or, in T 0 -like scheme, k = 4 [25].
In fact, one can consider more general cases of SU(N) group. Such a theory
could contain adjoint Higgses with different structures which break SU(N) in various
possible channels. By imposing the traceless condition on the SU(N) adjoint VEV, it is
straightforward to obtain the relation between the two eigenvalues x and y in (58) and thus
the relative Clebsch factors projected out in the context of the F -like and T 0 -like schemes.
(I) 1 : SU(N) SU(5) SU(N 5) U (1), this Higgs contains only the SU(5)
singlet VEV and thus can produce only k = 1.
(II) 2 : SU(N) SU(N 2) SU(2)w U (1), displaying the weak isospin SU(2)w .
N2
Hence h2 i = Diag(1, 1, . . . , N2
2 , 2 ) which implemented in the F -like scheme,
generates the Clebsches k = N2
2 , i.e., k = 2, 5/2, 3, 7/2, . . ., respectively, for
N = 6, 7, 8, 9, . . . . In the context of T 0 -like schemes the same VEV provides k = 2(N2)
N4 ,
i.e., k = 4, 10/3, 3, 14/5, . . ..
(III) 3 : SU(N) SU(3)c SU(N 3) U (1), exhibiting the colour SU(3)c .
3
3
, . . . , N3
). Hence, this Higgs in the context of
Therefore, h3 i = Diag(1, 1, 1, N3
133
3
F -scheme gives k = N3
, i.e., k = 1, 3/4, 3/5, 1/2, . . ., while in T 0 -scheme k =
6
d
6N , i.e., k = , 6, 3, 2, . . . (k = should be understood as 1/k = 0, i.e., A = 0
e
while A is non-zero).
(IV) There can be also adjoints with special VEV directions, having vanishing
3
,...,
eigenvalues towards the SU(2)w components, say h0 i = Diag(1, 1, 1, 0, 0, N5
3
N5 ). It is natural to use such adjoints for the doublet-triplet splitting [26]. Obviously,
this would give k = 0 in either F - or T 0 -like schemes. Analogously, there can be adjoints
with vanishing SU(3)c components, which would lead to k = in F -like scheme and
k = 2 in T 0 -like scheme.
Therefore, in the next, while searching the realistic anstze, we shall scan all possible
Clebsch values for kn or 1/kn among 0, 1, 2, 3 plus the above N -dependent
expressions (for all possible N ). 13
It should be remarked that within this wide variety, we find only three acceptable
solutions which characterise three different anstze. We anticipate them here:
ansatz A:
k1 = k2 = 2,
ansatz B:
k1 = k3 = 3,
ansatz C:
k1 = 3,
k3 = 1,
k2 = 0,
k2 = 0,
k3 = 2.
(59)
(60)
(61)
We shall come back to this point in Section 4.3.

3. The SO(10) SU(3)H model
In the context of SO(10) model all fermion representations (22) are unified into 16-plets
i , where i = 1, 2, 3 is a family (SU(3)H ) index. In terms of the SU(5) subgroup, the
fermion content is represented as i = (5 + 10 + 1)i . So, in addition to the superfields
5 i and 10i (22), i contain also the SU(5) singlets 1i , usually referred to as RH
neutrinos.
The SO(10) symmetry can be spontaneously broken to SU(5) by the Higgs supermulS 16, having the VEVs towards the singlet components. The SU(5)
tiplets C 16, C
symmetry can be further reduced down to SU(3) SU(2) U (1) by Higgses in tensor
representations 54 and 45. As for the MSSM Higgs doublets Hu,d needed for breaking the
electroweak symmetry and generating the fermion masses, they are contained in a single
superfield 10. 14 It is well known that the SO(10) framework also offers the possibility
to work out the basic problem of the doublet-triplet splitting which has to be unavoidably
addressed in any realistic grand unified theory. In particular, one can implement the so
called missing VEV mechanism [27] which can be justified by additional symmetry reasons (see [28,29] and references therein). It allows to achieve the mass splitting between
13 In the context of SU(N ) models containing SU(5) SU(3) , (N > 8), the large value of the horizontal
H
Clebsches b, b0 could be similarly obtained if also the I + (1 + 8)-like combinations emerge from the VEV
of the big adjoint breaking SU(N ) down to SU(N 1) U (1), i.e., b = N 1.
14 We remind the content of SO(10) multiplets with respect to the SU(5) subgroup: 16 = 5 + 10 + 1, 16 =
5 + 10 + 1, 54 = 24 + 15 + 15, 45 = 1 + 24 + 10 + 10 and 10 = 5 + 5.
134
the components of the Higgs 10-plet so that the doublets Hu,d remain massless while their
colour triplet partners get masses of the order of the GUT scale, without unnatural fine
tuning of the Higgs superpotential parameters.
Since the 10-plet can couple only to the symmetric combination of the fermion
16-plets, the higher-order operators inducing the fermion masses can only involve the
(anti)sextet horizontal Higgs S (cf. the SU(5) case (24) where also triplet Higgses A can
contribute). Hence, in the context of the SO(10) SU(3)H model the operators (24) are
represented as:
S ij
i j ,
(62)
M
where, as in Section 2, the flavour scale M can be of the order of the GUT scale MG
1016 GeV, and Yt 1 implies that S3 M. One can still define S = M 1 hSi. Thus, in the
low-energy limit this operator provides equal Yukawa matrices for all fermions, including
that for the neutrino Dirac mass terms: Yu , Yd,e , YD
= S. In addition, the operator
O
S ij
S2,
i j C
(63)
M2
involving the Higgs 16-plet with a VEV VC M induces the Majorana mass terms for
the RH neutrinos: MR S. Thus, the effective operator for the neutrino masses emerges
from the familiar see-saw mechanism, and it has a form (5) with Y = Yu , where
(M/VC )2 .
Therefore, in such a straightforward case we obtain a highly unrealistic pattern for
the fermion masses and vanishing mixing angles in both the quark and lepton sectors.
Recalling the discussion in Section 2.1, the way out from this situation is to invoke SO(10)
and SU(5) breaking Higgses (45- and 54-plets) in the effective operators for fermion
masses, as well as the SU(3)H octet Higgs for breaking the horizontal symmetry.
In this section we present a consistent SO(10) SU(3)H model to motivate the desired
Yukawa textures (20). By explicitly specifying the extra symmetries, generically denoted
by G in the previous section, we will show, for example, how the Clebsch structure of the
ansatz B may be obtained. Interestingly, the same symmetry reasons can be utilized for
solving the problem of the doublet-triplet splitting via the missing VEV mechanism.
Let us describe the ingredients of the model. For the sake of simplicity, we assume that
all vertical Higgses are singlets of SU(3)H and vice versa, all horizontal Higgses are
singlets of SO(10). In particular, the vertical sector of SO(10) Higgses includes chiral
superfields in the representations 54, 45, 16 and 16, needed for the SO(10) symmetry
breaking down to SU(3) SU(2) U (1), and 10-plets for the electroweak symmetry
breaking and the fermion mass generation. The horizontal sector of SU(3)H Higgses
includes the triplet, sextet and octet representations. Namely, as in the previous section, we
three triplets A1,2,3 3 (plus the conjugated spectator
introduce one anti-sextet S 6,
and the SU(3)H adjoint 8. In addition, there can
superfields S
S 6 and A 1,2,3 3),
be some fields which are singlets under both SO(10) and SU(3)H .
Besides the matter superfields i (16, 3) we introduce a number of vector-like
representations containing the fragments like (28), (29) and (37) which are needed for
OR
135
the fermion mass generation. Therefore, these should be matter superfields in SO(10)
representations 16 + 16, 10 or singlets, to be specified below. These states mediate the
see-saw like diagrams and hence the quark and lepton Yukawa structures emerge after
integrating them out.
We invoke the additional symmetry in the form G = U (1)A Z6 R, where U (1)A is
an anomalous gauge symmetry [30] implemented in the same spirit as in Ref. [29], R is a
discrete symmetry (as in Section 2.2) under which all superfields in the theory as well as
the superpotential change the sign, and Z6 is a discrete symmetry acting on the superfields
as n n exp( 3 iQn ).
The superfield content of the model with respect to SO(10) SU(3)H is given below,
and the superfield charge assignment [Q, Q] with respect to U (1)A and Z6 symmetries,
respectively, is shown in subscript. Namely, we assume that the Higgs sector contains the
following superfields:
(i) the SO(10) representations
(54, 1)[0,0],
C (16, 1)[1/2,1],
(45, 1)[0,3],
S (16, 1)[1/2,3],
C
(45, 1)[0,2],
0 (45, 1)[0,1],
(10, 1)[1,0],
0 (10, 1)[1,3],
(64)
(ii) the SU(3)H representations

[0,0] ,
S (1, 6)
S
S (1, 6)[0,0],
A1 (1, 3)[0,1],
[0,3] ,
A 1 (1, 3)
A2 (1, 3)[0,2],
[0,0] ,
A 2 (1, 3)
A3 (1, 3)[0,1],
[0,3] ,
A 3 (1, 3)
(65)
and (1, 8)[2,0] ,

(iii) and the gauge singlets
Y (1, 1)[0,0],
Z (1, 1)[0,2],
S (1, 1)[0,2] ,
Z
I (1, 1)[2,0] .
(66)
It is convenient to split the complete Higgs superpotential WHiggs consistent with the
above symmetries, into the following parts:
S + Y SS
S
W1 = 2 Y + Y 3 + Y 2 + 3 + (Y + ) 2 + Y Z Z
3
3
2
0
2
S + Z + ZC C
S + C C
S + Z
S
+ 0
+Z +Z
+ S3 + S
S 3 + ZAn A n + A1 A2 A3 + A 1 A 2 A 3 ,
S2,
W2 = 0 + I 02 + C
(67)
(all SO(10) and SU(3)H index contractions are left understood). Here order 1 coupling
constants are implied in the trilinear terms, and is a mass parameter of the order of the
GUT scale, MG 1016 GeV. 15
An important role is played by the D-term of the anomalous U (1)A symmetry
X
TrQ
Qn |n |2 , M2 =
M2
(68)
DA = M2 +
192 2 str
15 As far as R-parity is concerned, generally it is not automatic in SO(10) models involving the Higgs 16-plets.
However, in the context of our model one can easily verify that all dangerous R-violating terms are suppressed
by the G symmetry.
136
where the sum runs over all scalar components k in the theory and Q are their U (1)A
charges. Therefore, the spontaneous breaking scale of the U (1)A symmetry is naturally
related to the string scale Mstr 101718 GeV with a coefficient determined by the charge
content of the superfields in the game.
Notice that is the unique mass parameter contained in the superpotential (67), while
another mass scale in the theory comes from the FayetIliopulos term M2 . Therefore, the
VEV magnitudes of all fields into the game should be determined by these two mass scales,
which in addition can be both of the order of the GUT scale. In particular, the Higgs fields
participating in the superpotential W1 can get VEVs violating the SO(10) and SU(3)H
S which
symmetries. In fact, the linear term 2 Y induces the VEVs of the singlets Y, Z, Z
in turn play the role of mass terms for the superfields in non-trivial representations like
, 0 , , , etc. Thus, with an accuracy of O(1) coupling constants in the superpotential
terms (67), the magnitude of all non-zero VEVs is MG . As for the fields I and , they
can get VEV of the order M from the anomalous D-term (68). We assume that the trace
TrQ of the U (1)A charges over all superfields is positive: namely, the superfields presented
in (64) and (66) plus the octet contribute as TrQ 102 . (The theory may include also
other superfields with charges arranged so that the GreenSchwarz cancellation mechanism
is at work, but this does not affect the orders of magnitude in our consideration). Therefore,
M can naturally be about the GUT scale MG or a bit larger. One has also to remark that the
anomalous term fixes only the overall VEV combination hI i2 + hi2 , but not each term
separately. The VEV values of I and will be fixed by the soft supersymmetry breaking
terms, as well as orientation of hi with respect to the VEVs of other horizontal scalars,
S and A1,2,3 , in the spirit discussed in Section 2.2. This means that the horizontal fields
should contain some light fragments, with masses order TeV, which, on the other hand,
do not lead to phenomenological problems (see, e.g., discussion in Ref. [23]). The only
important point is that, due to the simple structure of the SU(3)H group, the VEV of in
either case would acquire the simplest possible configuration hi Diag(1, 1, 2).
Remarkably, the terms contained in the superpotential W1 provide the pattern needed
to break SO(10) to the standard model group. In particular, there exists a solution
S VEVs which breaks SO(10) down to SU(5): hCi = hCi
S = VC
for the 16-plet C, C
|+, +, +, +, +i (in terms of the corresponding Cartan sub-algebra generators). The
S follows from the vanishing of the SO(10) D-terms. The 54 and 45
condition hCi = hCi
Higgses are needed to subsequently break SU(5) down to SU(3) SU(2) U (1). The
necessary VEVs are given as:
hi = V Diag(1, 1, 1, 3/2, 3/2) 0 ,
hi = V Diag(1, 1, 1, 0, 0) ,
hi = V Diag(1, 1, 1, 1, 1) ,
h 0 i = 0,
where
0 =
1
0
(69)
0
1

,
0 1
1 0

.
(70)
137
One can easily verify that these VEVs are consistent with a supersymmetric vacuum and
their configurations are not affected by some unknown factors. In particular, in the 54-plet
only its 24 fragment (in terms of SU(5) subgroup) acquires VEV and, noticeably,
into the only phenomenologically allowed direction which induces the SO(10) symmetry
breaking down to the PatiSalam subgroup: SO(10) SU(4) SU(2) SU(2)0 . The
coupling of the 45-plet to in the superpotential forces the VEV of the former to
lie in the BL direction, as a result of a combination of its SU(5) 1 and 24 fragments,
with exact zeros on the last two components, as it is demanded by the missing VEV
S but
mechanism [27,28]. 16 As for the other 45-plet , it is coupled to the 16-plets C, C,
does not couple to the 54-plet . Because of this, its VEV is directed entirely towards the
SU(5)-singlet direction, with a vanishing VEV towards the 24 fragment. 17 And finally,
S and thus it should have a
the third 45-plet 0 does not couple neither to nor to C, C,
vanishing VEV.
It is also important that all non-singlet (under the standard model group) fragments
in the SO(10)-breaking superfields get masses of the order of MG . In such a case, no
light fragment (with mass MG ) is left behind which could affect the gauge coupling
unification at the scale MG . The presence of the term 0 is crucial for achieving this
result [28], since otherwise, in its absence, the theory would have flat directions towards
non-singlet fragments which are related to extra global symmetry in the superpotential.
Remarkably, this term does not affect the VEV pattern of 45-plets while lifting the
dangerous flat directions.
Finally, the term W1 in (67) is also responsible for the horizontal SU(3)H breaking
via the VEVs of S and A1,2,3 fields (see (26) and (27)) and its features are elaborated in
Appendix A. Notice, that even if all coupling constants are taken real in the superpotential,
non-trivial CP-violating phase in these VEVs could appear in case of solutions with
different phases for the singlet VEVs hY i and hZi, in the spirit discussed in Appendix B.
Let us address now the doublet-triplet splitting problem. As well-known, the VEV of the
54-plet guarantees the VEV of the 45-plet to be towards the BL direction, hi
Diag(1, 1, 1, 0, 0) . On the other hand, the superpotential W2 contains the necessary
terms to incorporate the missing VEV mechanism for the doublet-triplet splitting. In
addition, they also induce the mixing between the 5 fragments in and C through the
large SU(5) conserving VEV VC = hCi. All this can be more easily represented in terms
of the following mass matrix:
5()
0
5()
0 ) hi
5(
5(C)
0
5( 0 )
hi
M
0
S
5(C)
VC
0 ,
(71)
MC
16 As we will see below in the analysis of the Yukawa sector, such a VEV structure of is also important for
obtaining the Clebsch factors needed for the ansatz B, namely k1,3 = 3 and k2 = 0.
17 This is also relevant for the fermion Clebsch structures, because a SU(5)-singlet VEV namely, the
combination hZ + i participating in the fermion mass generation (see below, diagrams in Fig. 4) does
not induce uncontrollable corrections to the Clebsch factors.
138
S
where M hI i and MC hZ + i is the mass of the 5-plet fragments contained in C, C.
Owing to the form of the VEV hi in (69), non-zero entries are induced only between
the triplet fragments in and 0 while the corresponding entries for the doublet fragments
vanish. Therefore, all the triplets get masses of the order of MG , while two combinations
of the doublet components, to be identified with the MSSM Higgs doublets Hu,d , remain
light. The doublet Hu comes entirely from H 5(), and so no portion of it is contained
S or 5( 0 ). As regards Hd , it emerges from the combination of 5 fragments in
in 5(C)
S = c 5(C)
s 5(),
with tan MC /VC (see also [31]).
and C. Namely, H
Let us discuss now the fermion mass generation mechanism. Apart from the already
familiar (16, 3)[1/2,0], let us introduce the following vector-like fermions:
(I) 16-plets
[3/2,0],
S (16, 3)[1/2,0];
(16, 3)
0
[1/2,3] ,
S (16, 3)
0 (16, 3)[3/2,3],
(72)
(II) 10-plets
t (10, 3)[1,3],
[1,1] ,
t (10, 3)
t1 (10, 3)[1,2],
t2 (10, 3)[1,0],
[1,2] ,
t1 (10, 3)
[1,0] , (73)
t2 (10, 3)
(III) and singlets

[0,3] ,
N (1, 3)
S (1, 3)[0,3].
N
(74)
Let us start by discussing the role played by the additional fermion 10-plets ti = (5 + 5)i ,
t i = (5 + 5)i . All the symmetries of the model allow in particular the following terms:
S t.
C t + Zt
(75)
Correspondingly, the mass matrix connecting the matter states reads as:
5(ti )
5(i )
0
i) 0
5(t
ti )
S
5(
hZi
5(ti )
VC
S ,
hZi
0
(76)
which shows that the light states 5 i are composed by a combination of 5 fragments in i
i , where the angle is defined as tan = hZi/VC ,

and ti , i.e., 5 i = s 5()
i + c 5(t)
i and 5(
t)i are heavy states.
while the orthogonal combinations 5 0i = c 5()
i + s 5(t)
Now we are in position to fix the fermion mass pattern which emerges in our theory.
The diagonal entries in the Yukawa matrices are generated by the following superpotential
terms:
S+
S S,
W = + (I + )
S + NY N
S + SN
S 2,
WN = CN
(77)
which through the see-saw like diagrams shown in Fig. 3, give rise to the effective operators
like (62) and (63). For an easy identification of the contributions to the Yukawa matrices,
the terms in W and WN can be rewritten according to their SU(5) decomposition:
139
Fig. 3. The see-saw diagrams giving rise to the effective higher order operators for the symmetric
contributions to the charged fermion Yukawa matrices and to the neutrino Dirac mass term YD

diagram (a), and to the Majorana mass matrix MR of the RH neutrinos 1 contained in
and the
diagram (b). In SU(5) language the relevant fragments are = 5 + 10 + 1, = 5 + 5,
S
singlet component of C with VEV VC .
S )S10()
W = 10()5()10( ) + 5()
5()10(
) + 10(
5(
) + 5(
S ) + 10()5()
S)S 5()
+ 10( )(I + )10(
WN
)(I + )5(
S ) + 5()5()1(
S )S1() + 1( )(I + )1(
S ),
+ 5(
) + 1(
2
S + NY N
S + SN
S .
= 1()1(C)N
(78)
Therefore, as already discussed in Section 2.1, the VEV combination hI + i provides

the SU(3)H breaking and leads to the SU(2)H invariant form of the heavy fermion mass
matrix M = Mb0 , where b0 = Diag(1, 1, b0 ) and M M. The relevant Yukawa terms
for the light fermions emerge after integrating out the heavy states. In particular, for the
Yukawa coupling matrix of up quarks (which are entirely contained in 10() fragment) to
01 S, where S M 1 hSi and
the Higgs doublet Hu 5() we obtain Yu = M1
hSi = b
it is diagonal once the VEV hSi is chosen as (26). On the other hand, recalling that 5()
contains the light 5-plet with a weight s and the Higgs 5() contains Hd with a weight
s , the terms (78) provide also the diagonal contributions to the down quark and charged
diag
lepton Yukawa matrices Yd,e = s s b01 S. From the decomposition (78) it is apparent
that the (heavy fermion) exchange contributions to Yu are of the T -scheme type (cf. (35))
diag
and those to Yd,e are of both T - and F -scheme type (cf. also (38)).
01 S. On the
Finally, for the Dirac Yukawa matrix of neutrinos we obtain YD
= s b
other hand, by considering the terms in WN , we see that the VEV hY i induces the
S, MN = MN 1, (1 = Diag(1, 1, 1))
SU(3)H invariant masses between the states N and N
and MN MG , and so the RH neutrinos 1() acquire the Majorana masses MR =
2
2
M2
N VC hSi = MS, where (VC /MN ) 1. Thus, as long as both Dirac and Majorana
140
Fig. 4. Diagrams giving rise to the effective higher-order operators for the non-diagonal contributions
i ) states.
to Ye,d . The relevant contributions of these diagrams involve Hd 5(C)
and 5 i 5(t
terms are present for neutrinos, the operator (5) emerges via the standard see-saw
mechanism:
s2 02
Y
1 D
b S.
= YD
MR Y =
ML
M
(79)
Therefore, both Yu and Y are diagonal and we have:

Y = b01 Yu ,
(80)
= s2 /.
where
Let us discuss now the origin of the off-diagonal entries in Yd,e , recalling that they
have already got the diagonal contributions Yu . We also remember that the off-diagonal
entries should contain non-trivial Clebsches between quarks and leptons.
To this purpose, we recall that the light fermion 5-plet

is also contained in 5(t)
with a
weight c , while the doublet Hd is contained in the fragment 5(C) with a weight c . Then,
the symmetry of the theory prescribes other superpotential couplings for the superfields t
S ) heavy fermions, so that the exchange of the latter induce
and C, involving other (than ,
the off-diagonal contributions to Yd,e via the VEVs of the triplet horizontal Higgses A1,2,3 .
The relevant diagrams 18 are shown in Fig. 4. The diagram (a) provides the entries (2, 3)
and (1, 2) as only the SU(3)H triplets A1 and A3 are allowed to have a direct coupling to
the matter 10-plets ti . In this case, the coupling of the 45-plet (with its VEV into the
S 0 ) gives rise to the Clebsch coefficients k1,3 = 3.
BL direction) to the 16-plets ( and
We should also observe that these off-diagonal contributions are effectively induced by
18 For the sake of brevity we do not write down the relevant superpotential terms responsible for the generation
of the off-diagonal Yukawa couplings: they can be read out directly from the diagrams in Fig. 4.
141
S 0,
the exchange of the SU(5) 10+ 10 fragments contained in the vector-like states 0 +
0
according to the T -scheme (cf. (40)).
The triplet A2 is instead involved in the diagram (b) which therefore contributes to the
(1, 3) entries. Because of the requirement of G invariance, this triplet cannot couple to the
states ti but it couples to the heavy states t2 which are mixed to the former through the VEV
of the 45-plet . This occurs through the couplings on the left of the A2 incoming-line in
the diagram (b), while the part on the right side contains exactly the same couplings as that
on the right of the A1,3 line in diagram (a). However, the coupling of the 45-plet with
the fermion 10-plets, t t2 , gives vanishing contribution to the lepton doublets contained
in 5(t)
(similarly to what happens in the missing VEV mechanism for the doublet-triplet
splitting) and therefore we obtain k2 = 0. In either case of diagram (a) or (b), the SU(3)H
breaking Clebsch b emerges from the same combination I + (i.e., from the matrix M0
Mb 0 = M diag(1, 1, b 0 )) so that the 13 and 23 entries have the same asymmetry.
Thus, the final form of the down quark and charged lepton Yukawa matrices reads as:
1
Yd = YD
u + b Ad ,
1
YTe = YD
u + b Ae ,
(81)
where the anti-symmetric matrices Ad,e have the form (50) with k1,3 = 3 and k2 = 0,
b = diag(1, 1, b) where b = b10 . In this way we have reproduced the Clebsch factors
demanded by the ansatz B (60). The coefficient is identified with s s . It may be smaller
than 1 and thus moderate or even small values of tan can naturally be accommodated.
In conclusion, we have presented a complete SO(10) SU(3)H model in which the
symmetry breaking VEV pattern as well as the form of the superpotential couplings needed
to obtain the desired predictive mass texture for the fermions (specifically, the ansatz
B presented in Section 2) are fully motivated by the additional G = U (1)A Z6 R
symmetry. Interestingly, the missing VEV mechanism for the doublet-triplet splitting can
be motivated on the basis of the same symmetry G. The other Clebsch structures demanded
by the anstze A or C can be reproduced along similar lines by rearranging the charge
assignments of the superfields in the game with respect to the Z6 symmetry, however we
will not show the corresponding models here.
4. Phenomenological analysis
This part is devoted to the phenomenological study of the anstze suggested. First, we
provide the corresponding expressions of the physical quantities quark masses as well
as mixing angles. Second, the criteria of the fit are presented. Finally the results are shown
and discussed.
4.1. Physical observables from the theoretical model
We now deal with the explicit form for the Yukawa matrices Ye,d in (51). In general
the coupling constants and the VEVs present in the theory are complex. However, all the
142
phases except three can be pulled out by field redefinitions and the Yukawa matrices can
be presented in this form: 19
Yu 0 0
Yu 0 0
Yu = 0 Yc 0 ,
Y = 0 Yc 0 ,
(82)
Yt
0 0 Yt
0 0 b
i
i
Ae
C
Yut D e
Yut D ei k3 A ei b1 k2 C
Yd = A ei Yct D ei B , Ye = k3 A ei
Yct D ei b1 k1 B .
1
bC
1
bB
k2 C
k1 B
D
(83)
Here D = Yt , Yut = Yu /Yt and Yct = Yc /Yt . Notice that in Y b0 = b has been taken
for concreteness. As compared to the expressions in (51) the off-diagonal terms have been
redefined as Ad1 = B, Ad2 = C and Ad3 = A and understood to be real. 20
The following observations can be made. The 11 entry in the matrices Ye,d is negligible
and in the following we set it to zero. Indeed, the ratio Yu /Yt is two orders of magnitude
less than Ye /Y and Yd /Yb (cf. Table 1) and thus its contribution in Yd,e is irrelevant.
As regards the 22 entry, this also leads to small corrections the ratio Yc /Yt is one order
of magnitude less than Ys /Yb (cf. Table 1). However, we find that these corrections are
relevant on comparing with the present precision in the CKM angles. So this contribution
in Yd,e cannot be neglected.
In general case the phases in (83) are arbitrary. However, they can have specific values
in the context of the spontaneous CP violation. Namely, one can assume that the original
theory has an exact CP invariance, i.e., all couplings in the Yukawa and Higgs sectors
can be simultaneously made real by phase redefinitions of the superfields. However, CP
invariance can be spontaneously broken by non-vanishing phases of some VEVs in the
theory. In particular, as we show in Appendix B, the triplets An can have complex phases.
For example, one can choose a basis where the VEVs of the sextet S, S1,2,3 are real.
Thus, in this basis all three diagonal entries of Yd in (51) are real. For definiteness, let
us take them also to be positive. On the other hand, the off-diagonal entries keep on the
phases originated from the VEVs of An . As we show in Appendix B, these VEVs can
exhibit a spontaneous CP violation when the product A1 A2 A3 is imaginary. In particular,
if all entries A1,2,3 are imaginary, the matrix Yd in (51), once recasted in the form (83),
implies the phase assignment = , = /2. In another possible case, when A1,2 are
real but A3 is imaginary, one has instead = 0 and = /2. In principle, the spontaneous
CP breaking could provide some other interesting phase assignments such as and have
some inverse -fold values like 2/3, /4, etc. In the following, to be more general, we
keep the phases and arbitrary, however we shall pay special attention to those specific
phase values which might be obtained in the context of the spontaneous CP violation.
19 As far as Y and Y are diagonal, these phase transformations are not relevant for the form of the mixing
u
matrices Vq,l and we do not show them.

20 In general the relative lepton versus quark Clebsches k = Ae /Ad can be complex, however in the
n
n n
following they are taken real.
143
Let us now proceed our analysis. The (1, 3) and (3, 1) entries in Yd,e can be
simultaneously rotated away by orthogonal and phase transformations in the 12-flavour
space:
e dT Yd O
e d P+ = e
e eT Ye O
e e P = e
Yd , P+ O
Ye ,
P O
12
12
12
12
d(e)
d(e)
i
c12
s12
e
0 0
e d(e) = s d(e) cd(e)

P = 0
1 0 , O
12
12
12
0
0 1
0
0
0
,
(84)
d(e)
d(e)
d
e
d
= C/B and tan 12
= kk21 tan 12
. Notice
where c12 = cos 12 etc. and it is taken tan 12
that these rotations do not affect the (1, 2) and (2, 1) elements as the Yukawa matrices are
antisymmetric in the 12 block. On the contrary the 23 block gets modified and by further
e B/cd , the matrices become:
re-defining everywhere the parameter B as B
12
0
k3 A
0
0
A
0
ee =
e
e
e
(85)
Y
Yd = A Yct Dei B
,
k3 A Yct Dei 1 k1 B
,
b
1e
e
D
0
D
0
k1 B
bB
q
d )2 + (
d k /k )2 . Finally, these matrices e
s12
Ye and e
Yd can be diagonalwhere k1 = k1 (c12
2 1
0 understood
ized by bi-unitary transformations as in Eq. (6) with the matrices Ue,d and Ue,d
as:
U = U32 U13 U12
1
0
0
c23
s23 ei23
0 s23 ei23
c23
s12 ei12
c12
s12 ei12
c12
0
c13
s13 ei13
c13
s13 ei13
0 .
(86)
e d P+ Ud and is parameterised as
Therefore, the CKM matrix is Vq = O
12
d i
d
d i
d
d i
d
c12 e vud + s12 vcd c12
e vus + s12
vcs c12
e vub + s12
vcb
d
d i
d
d i
d
d i
Vq = c12
vcd s12
e vud c12
vcs s12
e vus c12
vcb s12
e vub , (87)
vt d
vt s
vt b
where by vud , vus etc. with lower v we mean the matrix elements of the quark mixing
Ud as defined in Eq. (7) and parameterised as in (8). Analogously, the effective leptonic
e eT :
mixing matrix is Vl = Ue P O
12
e i
e
e
e i
c12 e ve1 + s12 ve2 c12
ve2 + s12
e ve1 ve3
e i
e
e
e i
(88)
Vl = c12
e v1 + s12
v2 c12
v2 + s12
e v1 v3 ,
e e i v + s e v
c12
1
12 2
e v + s e e i v
c12
2
1
12
v 3
144
where with lower ve1 , ve2 etc. are denoted the matrix elements of the leptonic mixing
matrix Ue . These forms of quark and leptonic mixing matrices are convenient since
the contributions vus , vud and ve1 , ve2 etc. from the Yukawa matrices e
Ye,d of (85), are
d,e
e
Ye,d the
separated from that arising from the initial 12-rotation O12 . Had the matrices e
form in (13) (i.e., vanishing 22 entry) the above transformations Ue,d would be pure
rotations. Then the only source of CP violation would reside in the residual phase in P
driven by the (1, 3), (3, 1) entries. It is worth noticing that the CP violating phase appearing
in the quark and leptonic mixing matrices differs by (cf. (84)). This feature comes from
the antisymmetry of the 12 block, while it does not depend on the initial antisymmetry in
the 23 and 13 blocks.
The GUT scale Yukawa eigenvalues Yu,c,t , Yd,s,b , Ye,, , Y1,2,3 , are linked to the
physical fermion masses through the renormalization group equations (RGE). For
moderate values of tan = vu /vd , one obtains at one-loop 21 (see, e.g., [32]):
mu = Yu Ru RuSM u Bt3 vu ,
mc = Yc Ru RuSM c Bt3 vu ,
mt = Yt Ru RuSM Bt6 vu ,
md = Yd Rd RdSM d vd ,
ms = Ys Rd RdSM s vd ,
mb = Yb Rd RdSM b Bt vd ,
me = Ye Re ReSM vd ,
m = Y Re ReSM vd ,
m = Y Re ReSM vd ,
(89)
and for the neutrino masses

m1,2,3 =
Y1,2,3
R RSM Bt6 v22 ,
M
(90)
SM
account for the gauge-coupling induced running
where the factors Ru,d,e, and Ru,d,e,
16
from the GUT scale MG ' 10 GeV to the SUSY breaking scale MS ' Mt and from MS
to the electroweak scale MZ , respectively. The factors f encapsulate the QCD + QED
running from MS down to mf for f = b, c (or to = 1 GeV for the light quarks f =
u, d, s). Namely, for s (MZ ) = 0.119 0.004 we have
Ru RuSM = 3.53+0.06
0.07 ,
Rd RdSM = 3.43+0.07
0.06 ,
Re ReSM = 1.50,
R RSM = 1.15,
b = 1.53+0.03
0.04,
c = 2.05+0.13
0.11,
u,d,s = 2.38+0.24
0.19 .
(91)
The factor Bt includes the running induced by the large top quark Yukawa constant
(Yt 1). 22 For Yt varying from the lower limit Yt = 0.5, imposed by the top pole-mass,
to the perturbativity limit Yt 3, the function Bt decreases from 0.9 to 0.7. Regarding the
CKM elements, their physical values are related to the corresponding GUT-scale quantity
(labelled by the superscript G) as follows: 23
21 The notation is the standard one: for the heavy quarks t, b, c, m
t,b,c are their running masses, respectively, at
= mt,b,c , while for the light quarks mu,d,s are given at = 1 GeV.
22 In the RG running of neutrino masses (90), the factor B is function of Y / sin .
t
t
23 We shall not take into account the analogous RGEs [33] for the neutrino mixings, since the experimental data
still contain big error bars and such an improvement is not justified. Moreover, renormalization effects are mostly
expected in case of strong mass degeneracy [34] which is not our case.

G
Vus(d) = Vus(d)
,
G
Vcs(d) = Vcs(d)
,
G
Bt1 ,
Vc(u)b = Vc(u)b
145
Vt b = VtGb ,
Vt d(s) = VtGd(s)Bt1 .
(92)
Using the procedure outlined in detail in [13], the following relations (valid at the GUT
scale) are derived from the matrices (85):

D = Y
1 (b + b1 )
Y Ye
Y
1/2

= Yb
Fe
d
1 (b + b1 ) Ys YY
b

e(d)2
Yct b + b1 cos
Fe(d) = 1 + c12
Fd
1/2
,
(93)
(94)
and
A2 D = Ye Y Y = k32 Yd Ys Yb ,
e2 = b (Y Ye )Y Ie = b(Ys Yd )Yb Id ,
B
Fe
Fd
k12
where we have defined
Y
e2
Yct cos ,
Ie = 1 + c12
Y Ye
d2
Id = 1 + c12
(95)
(96)
Yb
Yct cos .
Ys Yd
(97)
In the following we directly substitute the Yukawa constant ratios with the corresponding
mass ratios whenever the latter are RGE invariant, e.g., Y /Y = m /m , Yd /Ys =
md /ms , etc. Now, by dividing the squared of the l.h.s. and r.h.s. in Eq. (93) by the
corresponding sides of Eq. (96) and inverting, we obtain the following implicit relation:
Ie Fd k12 m me
Ie Fd k12
Ys Yd
=
0.059
,
Yb
m
Id Fe Z 2
Id Fe Z 2
where
Z=
Fd
Fe
1/2

1/2

Ie
1 m me
.
1 1
b
+
b
m
Id k 2
(98)
(99)
Substituting the expression (98) back into Eq. (93) we get:

Yb
= Z,
Y
(100)
which shows the modification induced on the expression of the b Yukawa unification
at the GUT scale by the asymmetry in the 23 block. Analogously, by utilizing the
relation (100) in Eq. (96), the latter can be rewritten as
k 2 Ie Fd
Ys Yd
= 1
,
Y Ye
Z Id Fe
(101)
while by dividing the squared of the r.h.s. and l.h.s. of (96) by the corresponding sides
of (95) we have:

k 2 Ie Fd 2 m
k 2 Ie Fd 2
md
me
ms
+
2 = 43
+
2 204.7 43
.
(102)
md
ms
me
m
k1 Z Id Fe
k1 Z Id Fe
146
Compare the present expressions (98), (101), (102) and (99) with the corresponding ones
in Eqs. (14), (15) and (16) obtained in the Fritzsch-like case (C = 0, Yct , Yut 0, i.e.,
Fe,d , Ie,d 1). From Eqs. (100) and (101) we can extract the following physical masses:
mb =

Rd RdSM b
Bt Zm = Bt Z 6.22+0.25
GeV,
0.27
SM
Re Re
m s md =

Rd RdSM d Ie Fd
4Ie Fd
(m me ) = 2
143+18
MeV.
14
SM
2
Re Re
k1 ZId Fe
k1 ZId Fe
(103)
e,d
and the corresponding
We now turn to the quark and lepton mixing. For the angles 23
phases appearing in the unitary transformation (86) we find:
s

m me
2
e
1 + Yct cos
tan 223 = 2 b
m
b

m m 1/2
1 Ie (b + b1 ) m e
,
(104)
m m
1 2b m e + (b b1 )Yct cos
s

2
Ys Yd
d
tan 223 =
(1 + 2bYct cos )
Yb
b
1/2

d
1 Id (b + b1 ) YsYY
b
,
(105)
d
1 )Y cos
1 2b1 Ys YY
+
(b
b
ct
b
e
=
tan 23
Yct
b sin
1 + Ybct cos
d
tan 23
=
bYct sin
bYct sin ,
1 + bYct cos
Yct
sin ,
b
(106)
(107)
0e,d
0e,d
and 23
are obtained from these expressions
while the right rotation angles 23
substituting, respectively, b b 1 . The phases appear strongly suppressed by the tiny
d
. 0.01. The 13 mixing angles are
Yct parameter: for Yct 3 103 and b 10, 23

0e
me m2 1/2
c23
s 0e me m
e
3
= ae 23
=
6
10
,
a
=
(108)
sin 13
e
0e
e
e ,
c23
m
c23
bc23
m3
d
0d r
d
s23
c23
s13
1
md Ys
=
,
a
=
.
(109)
d
d
0d d
0d
ad c23
s23
s23 ms Yb
c23
e,d
0 e,d
= 23
. Finally the U12
One can easily verify that for the corresponding phases 13
transformations are expressed by the following angles:

me 1/2
r
r
1 (ae + ae1 ) m
2
0.140
m
me
e
e
p e
p e , (110)
tan 212 = 2 ae
me
m
1 2ae m
c23 m me
c23

1/2
r
r
1 md
q
2
md 1 (ad + ad ) ms
md
d
0d
=
2
c23
,
(111)
tan 212
m
1
d
a d ms
ms md
1 2ad
ms
147
while the corresponding phases are

e
tan 12
d
tan 12

1/2
2
1 m
. Yct
1 m
1
4.6 102 ,
m
Yct m cos

1/2

2
sin
sin
1 4
1 m
. Yct
1
1 Y
1
.
Yct m
m
k1 Z
Yct Ybs cos
cos
2
(k Z) m
sin
(112)
d can be at most 11 and 25 with b & 1, Y = 3 103 and
We can easily estimate that 12
ct
for k1 = 2 and k1 = 3, respectively. Moreover, the larger is the asymmetry parameter b,
d
appears. In
the smaller the function Z becomes and therefore the smaller the phase 12
conclusion, the phase cannot provide by itself a sizeable source of CP violation since it
is driven by the tiny ratio Yct . That is the reason motivating us to introduce non-vanishing
(1, 3), (3, 1) entries driving the extra phase . Hence, as a matter of fact we may think of
the matrices Ue,d as pure rotations, parameterised in the standard form (7) with the phase
= [13].
In the analysis, we test the amount of CP violation through the parameter K describing
the CP violation in K . More precisely, we use the more uncertain parameter BK
S0 transition which
parameterising the deviation from the vacuum saturation limit in K 0 K
we extract from the expression of K as given in the standard model [35]:
2

f MK G2F m2W Im(F ? tu2 ) 1
,
(113)
BK = |K | K
1MK 12 2 2 |tu |2
and compare with its theoretical value which ranges from 0.6 to 1. Here fK = 161 MeV is
the kaon decay constant, MK = 497.7 MeV is the K 0 mass, and 1MK = 0.53 1010 s1
is the experimental value of KL KS mass difference. The mixing angles enter through
? V : t appears explicitly, 24 whereas t , t are contained in the
the combinations t = Vs
d u
c t
function
F = 1 tc2 S0 (xc ) + 2 tt2 S0 (xt ) + 23 tc tt S0 (xc , xt ),
(114)
where 1 1.38, 2 0.57, 3 0.47 are QCD correction factors and S0 is one of the
InamiLim functions (xc,t = m2c,t /m2W ). In this way, we are testing whether the experimental value of K can be reproduced entirely by the standard model contribution associated with the box diagrams involving the charm, top quark and the W boson. It is
known, that in the MSSM or more generally, in models with flavour-aligned soft terms
like in [36], the supersymmetric contributions are negligible. However, in general SUSY
GUT context K can receive substantial contributions from the super-partners at the weak
scale [37,38]. In view of this, we shall not use the information from K to constrain the
parameter space.
24 Note that in most of V
0
CKM parameterizations, the factor tu , coming from the amplitude of the decay K
2 , is made real and thereby the factor tu2 /|tu2 | becomes 1. In our final parametrization of the CKM matrix this is
not the case see (87).
148
4.2. Model parameters versus physical observables

We can now proceed by confronting the parameters with the physical observables
described by the model. The number of the former amount to 14, namely Yu,c,t , A,
d
, b, , , k1 , k2 , k3 , ML while the number of observables is 20: the up-quark
B, D, t12
masses mu,c,t , the down-quark masses md,s,b , the charged-lepton masses me,, , the
neutrino masses m1,2,3 , four independent CKM parameters and other four parameters
of the leptonic mixing matrix. As it will be discussed more extensively below, the three
Clebsch coefficients k1 , k2 , k3 are treated as external parameters, fixed at some specific
GUT-inspired value.
The various physical observables deserve a different treatment in the analysis for at least
two reasons. First, they are known with different degree of accuracy. Second, some of
them have a minor impact on the determination of the flavour parameters themselves.
Therefore, the observables are classified and used as follows:
(1) me , m , m : These masses are determined with high accuracy so they are used as
inputs to extract the parameters A, B, D via Eqs. (93)(96). Notice that in this way, also
the explicit dependence on tan of the fermion masses is automatically absorbed in the
parameters A, B, D.
u mc
(2) m
mt , mt : These mass ratios enter only into the determination of Yut , Yct and are used
as inputs. 25 Notice, however, that Yut , Yct quite depend on Yt through the renormalization
factor Bt : e.g., Yct = mc /mt Bt3 c1 . As this ratio is tiny, the dependence on mc /mt is weak
and hence our results are not sensitive to the corresponding errors on mc and mt . This
justifies the fact that we use the central values of the ratio mc /mt shown in Table 4. On the
other hand, the Yukawa constant Yt is left as free parameters. Indeed, it is not sensible to
fix Yt by the top mass due to its infrared behaviour. It will turn out that (in almost every
case) the preferred value for Yt is 0.5 which enhances the effect of Yct . This value is indeed
the smallest one compatible with the top mass for which mt = (160.9 2.2) sin GeV for
3
s (Mz ) = 0.119 0.004. Correspondingly, we have for the ratio Yct = (2.70.1
+0.2 ) 10 .
2
(3) md , ms , mb , |Vus |, |Vcb |, |Vub /Vcb |: These observables are fitted by a -like
procedure (see Section 4.4) to constrain the remaining parameters they depend on, namely
d , , once the Clebsch factors k , k are prescribed (see Section 4.3).
Yt , b, t12
1 3
4
s (Mt ) + 11.4
2 (Mt )) is
(4) Mt , mu /md : The top pole mass Mt = mt (1 + 3
2 s
determined as output once Yt is fixed by the fit in point 3, modulo sin . Though in
our analysis we cannot determine univocally (and in a sense is not necessary) tan , the
model we presented in Section 3 naturally yields tan in the small or moderate regime.
Using the determination of ms /md from the fit, we extract the ratio mu /md from the
combination [39]
Q= p
ms /md
1 (mu /md )2
(115)
25 Now for the sake of discussion we consider also Y but in practice, as already mentioned, we neglect it at
ut
all.
149
Table 4
Physical quantities used in the phenomenological analysis. Those marked by are used as input
S 0 and
for the fits. The bounds on |Vt d | and |Vt d /Vt s | are inferred from the evaluation of Bd0 B
d
0
0
S
Bs Bs transitions in the standard model [9] and thus they are not literally valid in the context
of supersymmetric model; even in the MSSM framework, with flavour-aligned soft terms, one can
expect substantial supersymmetric contributions to these transitions
Observables
Values
me [MeV]
m [MeV]
m [GeV]
mu /mt
mc /mt
Q
|K |
ms [MeV]
mb [GeV]
ms /md
mu /md
Mt [GeV]
|Vcb |
|Vus |
|Vub /Vcb |
|Vt d |
|Vt d /Vt s |
BK
0.511
105.7
1.777
(0.93) 105
(8.0 1.3) 103
22.7 0.8
(2.280 0.013) 103
155 75
4.25 0.15
21 4
0.20.7
173.8 5.2
0.0395 0.0017
0.2196 0.0023
0.093 0.014
0.0088 0.0018
< 0.22
0.8 0.2
(5) BK : The CP violation parameter is given as a prediction from the result of the fit
performed in point 3.
(6) |V3 |, |Ve2 |, |Ve3 | and lepton CP-phase: These are also given as predictions. More
precisely, for the sake of comparison with the experimental data, we trade these three
l
= 4|V3 |2 (1 |V3 |2 ),
elements for the corresponding oscillation parameters: sin2 223
2
l
2
l
2
2
2
2
sin 212 = 4|Ve2 | (1 |Ve2 | ) and sin 213 = 4|Ve3 | (1 |Ve3 | ). Notice, however, that
e =
the predictions for both CP violation phase and the element |Ve2 | depend also on t12
k2 d
k1 t12 , i.e., on the Clebsch k2 .
(7) m1 , m2 , m3 : The neutrino masses follow immediately as predictions m2 /m3 =
bYct , m1 /m2 = Yuc . However, the determination of any single mass does require the last
parameter, i.e., the mass scale ML = M
.
In summary, we have that from the sets (1) and (2) five parameters are fixed
A, B, D, Yut , Yct . From the set (3) other five parameters are constrained by the fit b,
d
, , , Yt once the two Clebsches k1 , k3 have been assigned. The determination of
t12
these twelve parameters allows us to make five clean predictions: m1 /m2 , m2 /m3 , BK ,
|V3 |, |Ve3 |. Also Mt is given as a prediction within the uncertainty of tan expected to
150
be in the range 110. From the result of the fit, we also gain the ratio mu /md by using the
information on the parameter Q (fixed at its central value).
Finally, the neutrino masses depend on ML and both |Ve2 | and the amount of
CP violation in the lepton mixing matrix depend on the Clebsch coefficient k2 . For
definiteness, the flavour mass scale M and the lepton-violation mass scale ML are set at the
GUT scale MG , i.e., M = ML = MG which as already discussed means that b/Yt 101
or 102 for Yt 1, b 10.
In the next section we shall discuss more extensively how we handle the Clebsch factors.
Finally, the phenomenological analysis requires also the parameters associated with the
standard model gauge sector: s (MZ ), em , sin2 w . For em , sin2 w we use the central
values quoted in [9]. As regards s (MZ ), the present world average is s (MZ ) = 0.119
0.002 [9]. On the other hand the SU(5) unification of gauge couplings implies a larger
value, s (MZ ) 0.125. Therefore, as a compromise, we have taken s (MZ ) = 0.123. We
remind that only mb and ms are sensitive to variations of s (MZ ) within the quoted error.
The corresponding variations can be inferred from the gauge-renormalization factors given
in (91). On the other hand, the s (MZ ) dependence of the running factor Bt is less than
1% and therefore all other observables are not sensitive to s (MZ ).
4.3. Clebsch prescription: three different anstze
First let us recall how the Clebsch coefficients enter into the determination of the
physical quantities. The Yukawa eigenvalues Ye,, and Yd,s,b depend only on k1 and k3
(see the analytical expressions in Section 4.1). 26 Similarly, the mixing angles vus , vud , . . . ,
also depend only on k1 , k3 . The coefficient k2 affects only the lepton mixing matrix (88)
e
d
= kk21 t12
. Thereby k2 can only be fixed for example by the 12
via the initial 12 rotations t12
l
= 4|Ve2 |2 (1 |Ve2|2 ). Notice,
lepton mixing angle Ve2 through the SN parameter sin2 212
2
2
l and sin 2 l do not depend on k .
indeed, that both sin 213
2
23
We can guess very easily the needed range for such k1,2,3 . Indeed k1,3 are mainly fixed
by the strange mass in (103) and the ratio ms /md in (102). From ms we infer that k1 is to
lie within 53 3 for b & 8. Correspondingly, from ms /md , k3 is forced to be in the range
1 38 . Hence, we have performed a preliminary scanning of the parameter space, by fitting
the quark observables in point 3 (Section 4.2) with k1 , k3 free to float in the range 03.
In such a way we have specified three Clebsch prescriptions for k1 and k3 and they are
given in Eqs. (59), (60) and (61), featuring three different anstze A, B, C, recalled here
for convenience:
ansatz A: k1 = 2, k3 = 1,
ansatz B: k1 = k3 = 3,
ansatz C: k1 = 3, k3 = 2.
On the other hand a priori we cannot stick to any specific value of k2 : only the result of
l
.
the fit will allow us to envisage the value that can give the better prediction for sin2 212
26 The dependence on k contained in k in Eq. (85) is mild for td < 0.1.
2
1
12
151
For that it is useful to write down explicitly 27 Ve2 :

e2 2
e2 2
e e
ve2 + s12
ve1 + 2c12
s12 ve2 ve1 cos .
|Ve2 |2 = c12
(116)
The present MSW range in Eq. (2) translates into |Ve2 | = (1.66) 102 . The lower
elements of Ue read as
e e
c23 ,
ve2 s12
e e
ve1 c12
c13 1,
(117)
p e
p e
then from Eqs. (110) we have |ve2 | c23 me /m 0.07 c23 which slowly decreases
with the asymmetry parameter b for b & 8 it becomes around (56) 102 saturating
by itself the present experimental upper limit. Therefore, the r.h.s. of Eq. (116) tells us
e
l
. 0.1 and < 90 . In this way sin2 212
may
that the third term can reduce |Ve2 | if t12
2
become smaller than 10 . Interestingly, it will turn out that the quark-sector fit demands
d
d
to lie in a similar range, t12
. 0.1, in all the three
the corresponding initial 12 rotation t12
anstze. We also write the CKM elements in (87) to explicitly show the interplay between
d
and the CP-phase :
the 12 rotation t12
d2 2
d2 2
d d
vus + s12
vcs + 2c12
s12 vus vcs cos ,
|Vus |2 = c12
d2 2
d2 2
d d
vcb + s12
vub 2c12
s12 vcb vub cos ,
|Vcb |2 = c12
d2 2
d2 2
d d
vub + s12
vcb + 2c12
s12 vcb vub cos ,
|Vub |2 = c12
(118)
where
d d
d
s12 s12
,
vus = c13
d d
vcs = c12
c23 1,
d
vcb s23
,
d
vub = s13
.
(119)
d . 0.1, then the choice

Therefore, if the fit of the quark quantities selects < 90 and t12
e
d
k2 = k1 is favoured as it implies t12 = t12 . On the other hand, if it comes out that >
d
e
) then k2 = 0 (i.e., t12
= 0) is preferable in order not to
90 (and irrespectively of t12
2
l
2
raise sin 212 above 1.4 10 . As we shall see in the next section the three anstze
will prefer different values of the phase such that in the ansatz A k2 = k1 = 2 can
l < 0.01 can be obtained. On the other hand, in both
be chosen and as a result sin2 212
l
> 0.01.
the other scenarios we have to select k2 = 0 and content ourselves with sin2 212
Needless to say that this picture in which all the Clebsch k1,2,3 are taken real is minimal.
For example, by allowing for complex kn , the lepton and quark mixing matrices would
have different CP violating phases and therefore, for example, also in the anstze B, C
l < 0.01 with non-vanishing te . Finally, one may wonder about
we could achieve sin2 212
12
the alternative possibility to account for the large-mixing angle MSW solution of the SN
l 0.50.98. That parameter range translates into |V | > 0.4
anomaly requiring sin2 212
e2
e
e
t12
& 0.70.8, or k2 /k1 & 78.
which could be achieved within our anstze by large t12
Such O(10) Clebsch coefficients seem to come less naturally and probably at the price of
a less economical Higgs content.
27 For the sake of brevity we are assuming that the elements of the mixing matrices U
e,d are real as we have
e,d
seen that the phases ij
are typically tiny.
152
4.4. Strategy for the fit

To study the constraints on the parameters of the model from the down-quark masses
and the CKM mixing angles we have performed an alternative 2 -analysis. 28
The CKM elements |Vus |, |Vcb |, |Vub /Vcb | are assumed to be mainly affected by
Gaussian errors and therefore the standard 2 -function can be assumed for them:
X a 2
exp
, a = x a x a ,
(120)
2 =
a
a
where the theoretical outcome for a certain observable is denoted by xa and the
exp
corresponding experimental value and statistical error by xa and a , respectively, (see
Table 4). On the contrary, the uncertainties affecting the determination of the down-quark
masses are due to the dependence on the theoretical-model used to extract the masses from
the measurements. For this reason we think that it is not correct to assign them a Gaussian
distribution and hence to define the canonical 2 -function. Nevertheless, we interpret the
1- range reported in the Table 4 as a reasonable interval accounting for all possible
(different) determinations. Therefore, we prescribe the masses a flat distribution in that
range. More precisely, we define the following 2 -like function for the masses:
0,
if |b | < 0.15,

2
2
b =
|b | 0.15
, if |b | > 0.15,
0.075
0,
if |s | < 75,
4

ms 80 , if 6 75,
s
2
10
s =

ms 230 4
, if s > 75,
30
(
0,
if |s/d | < 3,
2
4
(121)
s/d =
|s/d | 3 , if |s/d | > 3,
2 refer to m , m , m /m , respectively, and , ,
where b2 , s2 , s/d
b
s
s
d
b
s
s/d are the
corresponding deviations defined as in Eq. (120). We can note from (121) that mass
values outside the 1- range are strongly penalized by the higher-power dependence of
the (arbitrary) assigned distribution. The sum of the above functions will be denoted as
2 .
2 = b2 + s2 + s/d
By minimizing the sum 2 + 2 , we rather test the consistency between the model
and the Gaussian-distributed data within a reasonable range for the down-quark masses. In
2 and 2 separately, in Tables 5, 6 and 7
the next we shall report both the value of min
min
28 Those observables are computed by numerically diagonalizing the Yukawa matrices (85) once the parameters
A, B, D are determined from the charged-lepton masses (for given k1,2,3 ) through Eqs. (93)(96). The agreement
between the numerical and the analitycal out-comings presented in Section 4.1 is at the level of one per
mil.
153
Table 5
The analysis of the ansatz A. Taking as reference Eq. (83) for Ye,d , the columns refer to: (I) (1, 3)
and (3, 1) entries are set to zero (C = 0), i.e., three-parameter fit Yt , b, ; (II) (2, 2) entry is set to
d ; (III) complete pattern five-parameter fit Y ,
zero (Yct 0), i.e., four-parameter fit Yt , b, , t12
t
d . In all cases, from Y = 0.5 it comes out Y = 2.6 103 . The quantities marked by the
b, , , t12
t
ct
symbol ?, are not included in the fits but are predictions of the fit itself in correspondence of best-fit
2 (see
parameters (also shown). The values at the minima of the (non-Gaussian) 2 -like function min
2
Eq. (121)) and of the Gaussian function, min are also given
Ansatz A:
k1,2 = 2, k3 = 1
mb [GeV]
ms [MeV]
ms /md
|Vus |
|Vcb |
|Vub /Vcb |
?|Vt d |
?|Vt d /Vt s |
? Bk
l
? sin2 223
l
2
? sin 212
l
2
? sin 213
?mu /md
?Mt / sin
Yt
b
d
t12
2 and 2
min
min
(I) C = 0
(Yt , b, )
(II) Yct = 0
d )
(Yt , b, , t12
(III) Complete
d )
(Yt , b, , , t12
(IV) Spont. CP
d )
(Yt , b, t12
4.35
219
17.5
0.2138
0.0507
0.080
0.0148
0.30
0.96
1.2 102
1.8 102
0.6
172.7
4.00
249
20.8
0.2197
0.0447
0.097
0.0140
0.33
1.9
0.83
2.0 103
2.6 102
0.4
172.7
4.08
253
21.9
0.2193
0.0437
0.094
0.0133
0.32
0.84
0.87
5.2 103
2.4 102
0.3
172.7
4.04
254
22.0
0.2198
0.0433
0.098
0.0133
0.32
0.77
0.85
5.7 103
2.4 102
0.3
172.7
0.5
10.0
0
0.5
12.1
19.7
3.4 102
0.5
11.8
180
43.4
4.4 102
0.5
12.0
/4
4.7 102
0.06 and 50.7
2.1 and 9.5
0.4 and 6
1 and 5.1
containing all the results of the fits. We have to remark that descriptions of the quark masses
outside the reasonable range may be accounted, for example, by uncertainties in s (MZ ).
4.5. Testing theoretical models: fit and predictions
We have found it to be instructive to consider first the case with (1, 3), (3, 1) entries
set to zero (C = 0), to investigate the effect of a non-zero 22 entry on the pattern (13).
This pattern (denoted by (I)) has three parameters, b, and Yt . As a second step, we have
restored the (1, 3) (3, 1) entries and set to zero the 22-entry (formally, Yct = 0) leaving as
d , , Y (case denoted by (II)).
free parameters b, t12
t
d , , Y (case
Finally, the complete pattern is analysed with five free parameters, b, , t12
t
denoted by (III)). We can first make some general considerations. We can note that the
dependence on Yct of the physical quantities is mainly encoded in the factors Ie , Id . The
154
former can be easily estimated, Ie 1 + 0.044 cos and thus Yct can induce quite a small
correction on the leptonic mixing angles. Therefore, from the expressions (104), we can
e
(see (88)) depends mainly on the asymmetry parameter b,
infer that |V3 | |v3| s23
l
= 4|V3 |2 (1 |V3|2 ) > 0.8
increasing roughly as b. As a result, the AN bound sin2 223
requires 6 < b < 12 as it was found in the case (13) [13]. Analogously, in the case (I) it
e c e which weakly decreases with b. Then the MSW range, in the
is |Ve2 | |ve2 | s12
23
l = 4|v |2 (1 |v |2 ) gets smaller than
case (I) can be recovered for b & 7 where sin2 212
e2
e2
e
1.5 102 . As we already know, this mixing can be affected by the initial 12 rotation t12
present in the full model (case (III)).
The discussion about Id affecting the quark observables is more involved as it depends,
through the ratio Yb /(Ys Yd ) k12 , on the specific ansatz considered. However, in the
case A a rough estimate gives Id 1 + 0.12 cos , while larger correction can be achieved
in the anstze B, C, Id 1 + 0.25 cos . In the following we discuss separately our results
for the three anstze, reported in Tables 5, 6 and 7.
Ansatz A
In the case (I) (Table 5 second column), the result of the fit is acceptable though, due
to the high accuracy achieved in the determination of the CKM angles, we can observe
d
d
Id /Ie and |Vcb | s23
. The
some discrepancy in the Cabibbo angle |Vus | |vus | s12
former appears to be quite small while the latter is somehow too large. These two quantities
require in turn = 0 (to maximize Id ) and quite a large 23 asymmetry, b = 10 (to minimize
d
) as best fit points. Notice that b > 10, though preferable to further reduce |Vcb |, is not
s23
tolerated by |Vus | which would further decrease.
On the other hand, the quark masses fall in their reasonable range. Noticeably, Yt =
0.5 is preferred in all the fits and as a result the prediction Mt / sin 6 172.7 GeV turns
out to be consistent. Interestingly, the predictions (marked by ?) for the leptonic mixing
angle are quite good. We have to recall that whenever the quark mixing are reasonably
accommodated, the leptonic mixing angles fall automatically into the presently most
favoured range thanks to the remarkable product rule (12) inherent in the theoretic structure
l is almost maximal, as the
of the model. For example the AN oscillatory mixing sin2 223
large b = 10 implies. The 13 neutrino mixing remain well below the upper bound shown
l
lies in the range (13) 102 in all the anstze, as we will see
in Table 1 and sin2 213
(cf. Tables 6, 7). This range in the case of m223 close to the upper bound in (1), can be of
interest for the experimental search of e oscillation in the future CERN Neutrino
Factory [40].
We do not show in the tables the prediction for the neutrino mass ratios. However, as
regards m1 /m2 the prediction is just m1 /m2 = Yu /Yc (in any of the anstze) and so it is
typically about 3 103 . On the other hand, the prediction for the ratio m2 /m3 depends
on the value of the asymmetry parameter b 0 . In the simplest case b0 = b, it can be easily
computed as m2 /m3 = bYct and so it is in the range (23) 102 in agreement with
the present experimental hint. This out-coming will be similar in all the anstze and it
is a byproduct of the link between the neutrino and the up-quark masses and the large
23-asymmetry b demanded by the quark phenomenology.
155
Table 6
The analysis of the ansatz B. In the fit (II) from Yt = 0.97 it follows Yct = 1.9 103 . See also the
caption of Table 5
Ansatz B
k1,3 = 3, k2 = 0
mb [GeV]
ms [MeV]
ms /md
|Vus |
|Vcb |
|Vub /Vcb |
?|Vt d |
?|Vt d /Vt s |
? Bk
l
? sin2 223
l
2
? sin 212
l
2
? sin 213
?mu /md
?Mt / sin
Yt
b
d
t12
2 and 2
min
min
(I) C = 0
(Yt , b, )
(II) Yct = 0
d )
(Yt , b, , t12
(III) Complete
d )
(Yt , b, , , t12
(IV) Spont. CP
d )
(Yt , b, t12
4.65
76
20.3
0.2127
0.0430
0.014
0.0097
0.23
0.97
1.5 102
1.0 102
0.4
172.7
4.82
83
27.3
0.2194
0.0575
0.102
0.011
0.20
0.22
0.64
1.7 102
4.5 103
195.6
4.24
87
24.4
0.2196
0.0403
0.094
0.0086
0.22
0.69
1
1.3 102
1.5 102
172.7
4.37
83
23.0
0.2194
0.0409
0.099
0.0089
0.22
0.62
1
1.4 102
1.4 102
0
172.7
0.5
6.8
0
0.97
3.5
95.7
0.1
0.5
8.8
9.7
93.5
9.2 102
0.5
8.2
0
/2
9.7 102
11.4 and 45.2
150.5 and 112
0.02 and 0.24
0 and 0.82
The other predictions, Vt d , Vt d /Vt s are marginally compatible with the experimental
values. We have also reported the expected mass ratio mu /md extracted from the ellipse
parameter Q (115) which appears to be consistent with the determination of ms /md
obtained from the fit.
Clearly there is no CP violation in the CKM matrix as the infinite value of BK does
reflect. The situation gets improved in the case (II). The presence of the (1, 3), (3, 1)
d and the phase which strongly modifies the
entries introduces the initial 12 rotation t12
CKM elements (87). From the expressions of the Cabibbo angle in (118) and from the
ed is to be a small
fact that the element vus itself is around 0.2, we can deduce that O
12
d
rotation, t12 . 0.1, to prevent too large a correction from the element vcs 1. So |Vus | is
d 0.03 and 20 . At the same time |V | is
increased up to its experimental value by t12
cb
reduced, thanks to the larger value of the asymmetry parameter, b 12 which is now not
prevented by Vus . We can notice that also |Vub /Vcb | is successfully reproduced. The effect
of b larger, i.e., of the decreasing of the ratio Yb /Y has in turn induced a bigger ms and
a smaller mb , both slightly outside their reasonable range. The amount of CP violation
from the CKM matrix, though increased thanks to the non-vanishing phase , is still not
enough, BK 2. The predicted values of the leptonic mixings are good. Accordingly to
156
Table 7
The analysis of the ansatz C. In the fit (III) (Yt = 0.7) and (IV) (Yt = 0.6) it is Yct = 2.2 103 and
2.4 103 , respectively. See also the caption of Table 5
Ansatz C:
k1 = 3, k3 = 2, k2 = 0
mb [GeV]
ms [MeV]
ms /md
|Vus |
|Vcb |
|Vub /Vcb |
?|Vt d |
?|Vt d /Vt s |
? Bk
l
? sin2 223
l
2
? sin 212
l
2
? sin 213
?mu /md
?Mt / sin
Yt
b
d
t12
2 and 2
min
min
(I) C = 0
(Yt , b, )
(II) Yct = 0
d )
(Yt , b, , t12
(III) Complete
d )
(Yt , b, , , t12
(IV) Spont. CP
d )
(Yt , b, t12
4.04
115
18.2
0.2197
0.0378
0.041
0.0098
0.27
9.3
0.97
1.2 102
1.8 102
0.6
172.7
4.02
111
16.7
0.2196
0.0381
0.093
0.010
0.28
0.74
0.97
1.2 102
1.8 102
0.7
172.7
4.17
114
18.9
0.2195
0.0395
0.093
0.0098
0.26
0.65
1
1.3 102
1.4 102
0.5
187.0
4.09
120
20.4
0.2205
0.0381
0.082
0.0093
0.25
0.83
0.99
1.3 102
1.6 102
0.4
181.2
0.5
10.0
107
0.5
10.1
106
7.3 102
0.7
8.7
135
93.3
8.6 102
0.6
9.4
/2
7.3 102
0.5 and 15
4.0 and 0.7
0 and < 0.01
0.02 and 1.5
the approach elucidated in the Section 4.4, we have fixed the initial 12 lepton rotation as
e = td , i.e., k = k . Then, as expected, sin2 2 l is strongly reduced below 102 with
t12
2
1
12
12
respect to the case (I). Notice, that the amount of CP violation in both the quark and lepton
mixing matrix is the same as it is controlled by the same phase .
Finally, by comparing the results obtained in the case (III) where the 22-entry is
restored we conclude that the presence of the 22-entry does not play a significative
role the quality of the fit is quite stable in the two cases. However, the predictions are
better. Notice that the right amount of CP violation is achieved BK 0.7 as the phase
l
lies exactly in the range required by
is now bigger. For the same reason, also sin2 212
the MSW solution.
The fact that the best-fit points of the phases, = , /4, are just integers (or
half-integers) of may suggest that the CP violating phase is originated by a spontaneousbreaking mechanism. For this reason in the column (IV) we have performed again the
fit fixing the phases and at the nearest -fold values of the kind q (q =
1, 1/2, 1/4, . . .) as respect to those given by the fit (III). In this case we do not observe
any variation. Therefore we can conclude that the ansatz A provides quite a good fit of
d 0.05 and for = 180 ,
the quark observables for large 23 asymmetry b 12 and t12
157
= 45 . The corresponding amount of CP violation appears to be in the right range as

indicated by BK 0.8. We shall come back to the CP issue below.
Ansatz B
We have first to remind that this Clebsch-pattern in its Fritzsch version (vanishing 22
entry, C = 0 and b = 1) would give ms /md 25 and hence too a small |Vus | 0.20. Let
us consider the effect of the non-vanishing 22 entry. From Eq. (102) we see that a proper
trend can be achieved for 0 and b > 6 so that to maximise Id . On the other hand,
the moderate asymmetry b 67 required by the fit implies quite a large mb , small ms
and consequently also a very tiny |Vub /Vcb |. The fit is indeed rather poor (see Table 6,
column (I)). However, setting to zero the 22-entry and taking C 6= 0 spoils completely the
fit (Table 6, column (II)). Among the CKM elements only |Vus | and the ratio |Vub /Vcb | get
d 0.1 rotation. The same rotation is instead less important for
improved thanks to the t12
|Vcb | which remains too large due to the small asymmetry, b 34. For the same reason
also ms /md is quite large. In particular as ms /md is larger than the parameter Q, the lightquark mass ratio mu /md cannot consistently be evaluated. As a whole this case is definitely
disfavoured. Remarkably, the leptonic mixing angles as well as BK are not well predicted,
too.
Nevertheless, the interplay of the 22-entry with the (1, 3), (3, 1) entries can offer
a satisfactory description as the results in the case (III) show. All the quark quantities
2 and 2 are < 1). The fit requires b 9, td
are perfectly fitted (notice that both min
min
12
0.1, small and large CP phase, 94 . All the predicted quantities are within their
experimental ranges. The 23 leptonic mixing is predicted to be maximal as required by the
e = 0 or k = 0 and the SN oscillation mixing
AN anomaly. The large phase enforces t12
2
l
e
d
= 1.3 102 . Indeed, were t12
= t12
comes out to be close to the upper limit, sin2 212
(k2 = k1 ) and 90 as preferred by the fit the 12 leptonic mixing |Ve2 | would get
further increased (see Eq. (116)), contrary to what happens in the ansatz A. As a further
e = 0 the CP phase is vanishing in the lepton mixing matrix except for
consequence of t12
the small contribution driven by the phase . Finally in the last column, we have considered
the case with = 0 and = /2 fixed. The quality of the fit remains very good.
Ansatz C
The Fritzsch case for this pattern shows an opposite behaviour as compared to the
previous ansatz. Indeed, it is ms /md 12 implying too a big |Vus | 0.29. Therefore,
in the variant I , this can be cured with > 90 and a large b (see Table 7, column (I)).
In this case only |Vub /Vcb | is strongly incompatible with its experimental range. Notice
that, for the same asymmetry b = 10, mb is smaller as respect to the value obtained in the
ansatz A (I), since as k1 is larger as Z becomes smaller. All the predictions are reasonable
(except for BK ). In the second case (II) all the CKM mixings are well reproduced thanks to
the initial 12 rotation, but Yct = 0 makes ms /md below the reasonable range. Finally, the
complete model provides quite an excellent fit (III). Interestingly, also this ansatz points
to maximal CP violation 90 . In correspondence of the best-fit parameters all
the predicted quantities are in agreement with the present experimental status. We should
158
remark that like in the previous ansatz, the presence of the 22-entry is crucial. Moreover,
also in this case the choice k2 = 0 does not induce CP violation in the lepton mixing matrix.
In the last column, we have performed a three-parameter fit, setting = 90 and = 180 .
One could conclude that, by endowing the same pattern with spontaneous breaking of CPsymmetry, the description remains very good.
Finally, we can try to give an overlook at all the three anstze analysed. The case
A requires the CP-phase < 90 , whereas both the anstze B and C prefer maximal
CP violation, & 90 . This should imply in general a different relations between the
corresponding CKM elements. For example the three anstze will show up a different
shape of the unitarity triangle characterized by the following angles

Vt?b Vt d
,
arg ?
Vub Vud

?V
Vcb
cd
arg ?
,
Vt b Vt d

? V
Vub
ud
arg ?
. (122)
Vcb Vcd
Therefore in correspondence of the results of the fits obtained in each ansatz for the
complete case (III), we have plotted the unitarity triangle in the 0 , 0 plane (Figs. 68).
We recall that , , A, are the Wolfenstein parameters and 0 = (1 2 /2), 0 =
? has been chosen real
(1 2 /2). More precisely, the side CB corresponding to Vcd Vcb
and rescaled to unit length. Hence the length CA and CB in the rescaled triangle, usually
denoted by Rb and Rt , respectively, are:

q
?
Vud Vub
2 1 Vub
02
02

,
Rb
? = + = 1 2
Vcd Vcb
Vcb

Vt d Vt?b q

= (1 0 )2 + 02 = 1 Vt d .
Rt

?
V V
V
cd cb
(123)
(124)
cb
Once the parameters , A have been fixed according to the results of the (III) fit, = |Vus |,
A = |Vus |/2 , we have depicted the 1- circles of Rb from the measured value of |Vub /Vcb |
(solid lines), and the circles of Rt from the the mass difference 1Md,s describing the
0 B
S 0 mixings (dot-dashed and dashed lines, respectively). The values
strength of the Bd,s
d,s
used are 1Md = 0.471 0.016ps1 and 1Ms > 12.4 ps1 [41]. In the same plane the
hyperbola represent the iso-contours for BK (dotted lines). This overall picture of the CKM
elements shows that the ansatz A (see Fig. 5) is characterized by a large angle , > 100 ,
and by a small angle , 8 . They both appear to be in disagreement with the result
reported by the CDF collaboration sin 2 = 0.79+0.41
0.44 and with the indirect bounds on
which disfavour & 100 [41]. Moreover, the vertex A lies above the region delimited
by the circles Rt due to the large value of |Vt d | 0.013 predicted. On the contrary, the
ansatz B provides a consistent scenario where the observed amount of CP violation can
be accounted by the CKM picture (Fig. 6). The same holds for the ansatz C (Fig. 7).
Ultimately, a better discrimination of the specific ansatz will be possible when the program
of the measurements of the present and future B factories is completed.
159
Fig. 5. The unitarity triangle as emerging from the fit of the ansatz A (case (III)) in the ( 0 , 0 )
ABC,
b B CA.
b The
plane. The angles defined in Eqs. (122) are to be identified as C AB,
parameters , A are accordingly fixed as = 0.2193 and A = 0.909. We have drawn the 1- contours
of Rb = 0.414 0.062 (solid lines) Rt = 0.913 0.191 as inferred from 1Md (dot-dashed
lines) and Rt < 1.02 from the lower bound on 1Ms (dashed-line). The band enclosed by the
hyperbola (dotted) refers to BK = 0.8 0.2.
Fig. 6. As in Fig. 5 for the ansatz B (case (III)). Here = 0.2196 and A = 0.836, Rb = 0.413 0.062,
Rt = 0.989 0.207 (from 1Md ) and Rt < 1.02 (from 1Ms ).
5. Conclusions
We have considered a particular set of textures for the Yukawa matrices of quarks
and leptons which share some features with both the Stech ansatz proposed long time
ago [16,17] and the Fritzsch-like textures suggested in [13]. Our main motivation was to
show that they are not only successful from the phenomenological point of view but also
160
Fig. 7. As in Fig. (5) for the ansatz C (case (III)). Here = 0.2195 and A = 0.820,
Rb = 0.413 0.062, Rt = 1.01 0.21 (from 1Md ) and Rt < 1.02 (from 1Ms ).
grounded and compelling on the theoretical side. We have discussed how these textures
could emerge in the context of grand unified theories, in terms of the prototype SU(5)
model complemented by the horizontal symmetry SU(3)H (Section 2).
The SU(3)H group may seem too large as compared to U (2)H . Indeed the latter,
providing ab initio the 2 + 1 representation structure for the fermion fields, directly singles
out the heavy top quark from the lighter fermions of the first and second generation.
Namely, the models based on U (2)H [42] invoke the familiar paradigm according to which
the third generation has a priori order 1 (tree-level) Yukawa couplings while the small
Yukawa constants of the lighter generations emerge from higher-order terms containing
the U (2)H symmetry breaking Higgses, with VEVs smaller than the cutoff scale.
However, there are many good points in favour of SU(3)H . First, it accounts by itself
for three fermion families, and thus can be more predictive than U (2)H . Second, the
spontaneous breaking features of SU(3)H may turn the Yukawa constants of the low energy
theory (MSSM) into dynamical degrees of freedom and fix the inter-family hierarchy
in a pretty natural manner. Namely, the third generation becomes heavy (Yt 1), while
the second and first ones become lighter by successively increasing powers of small
parameters.
In the general discussion of theoretical issues, we have put forward the following
argument: just like the SU(5) theory contains the adjoint Higgs, 24-plet, which, by breaking
down SU(5) to SU(3)SU(2)U (1), does hide the existence of the large GUT symmetry,
also SU(3)H may be first broken down to U (2)H by its adjoint Higgs, an octet, which
makes the existence of a larger symmetry among all three families less visible.
In particular, the horizontal octet which breaks SU(3)H down to SU(2)H , not only
diversifies the third family from the lighter ones, but also induces a non-trivial Clebsch
structure between the 12 and 3 generations in the Yukawa matrices of quarks and leptons.
The effect of the latter is twofold. First, it induces the complementary largesmall (see-
161
saw like) pattern between the neutrino and quark mixing angles, which is indeed exhibited
by the observed small value of Vcb and the nearly maximal mixing. Second, it may
link the neutrino mass hierarchy to that of the up-quarks in realistic way.
On the other hand, the SU(5) adjoint has proved necessary to break the standard
down-quark and lepton degeneracy, providing different Clebsch coefficients in the Yukawa
matrices. Though its role is not univocal, we have featured three Yukawa patterns
characterized by what we consider as the most natural Clebsch factors. In Section 3
we have elaborated a consistent SO(10) SU(3)H model which, supplemented by the
symmetry G = U (1)A Z6 R, represents an existence proof of the Yukawa texture
suggested. In particular, the specific model outlined provides an understanding of the
Clebsch pattern of the ansatz B. Remarkably, the same symmetry content of G also
motivates the missing VEV mechanism necessary to achieve the doublet-triplet splitting.
A careful fit of the quark observables demonstrates that all the anstze A, B and C
reproduce quite well the down-quark masses and CKM mixing angles (Section 4). What
appears very interesting is that for the best-fit ranges of the parameters the predictions for
the neutrino mixing angles are in good agreement with the present experimental hints.
About the lepton mixing angles, we have to stress that while the outcome for the 23
neutrino mixing is a genuine prediction, in strict connection only with the down-quark
sector through the product rule (12), the 12 mixing angle becomes effectively a free
e
. Should the small angle MSW solution be confirmed
parameter fixed by the rotation t12
e
. On the
in the future, then all the anstze can account for that with different values of t12
contrary, should the 12 neutrino mixing angle be much larger, as indicated by the largemixing angle regime of the MSW solution, then only the lepton sector would need some
refinement.
Interestingly, also the prediction for CP violation in the quark sector may lie in the
observed range. In particular, all the anstze generate a CKM matrix with the correct
amount of CP violation. Though the experimental test of CP violation in the SM is not
yet accomplished, this result may imply that superpartners contributions to K must be
adequately suppressed [38] and it would be interesting to study in the theoretical models
presented here the implications of the horizontal symmetry on the SUSY spectrum.
Acknowledgements
We thank Andrea Brignole and Denis Comelli for useful discussions. This work is
partially supported by the MURST research grant of national interest Astroparticle
Physics.
Appendix A
Consider the following superpotential terms including antisextet superfield S = S ij and
its sextet partner S
S =S
Sij :
S + S3 + S
S 3,
WS = S S
(125)
162
where, to be explicit, S 3 = 13 ij k lmn S il S j m S kn = det S (similarly for S

S 3 ), and order 1
coupling constants are absorbed. Observe that this superpotential is manifestly invariant
2 S
S
under Z3 symmetry: S exp(i 2
3 )S and S exp(i 3 )S.
Without loss of generality, the VEV of S can be chosen in the diagonal form, hSi =
S
Diag(S1 , S2 , S3 ). Then the condition of vanishing F -terms FS , FS
S = 0 implies that hSi is
also diagonal, hS
Si = Diag(SS1 , SS2 , SS3 ), and
S1 S2 = SS3 ,
S1 S3 = SS2 ,
SS1 SS2 = S3 ,
SS1 SS3 = S2 ,
S2 S3 = SS1 ,
SS2 SS3 = S1 .
(126)
In the exact supersymmetric limit the VEV pattern of S and S

S is not fixed unambiguously
and there remain flat directions which represent a two-parameter vacuum valley. In other
words, the six equations (126) reduce to four conditions:
S1 SS1 = S2 SS2 = S3 SS3 = 2 ,
S1 S2 S3 = 3 ,
SS1 SS2 SS3 = 3 ,
(127)
while the others are trivially fulfilled (the last in Eq. (127) is not independent, it just follows
from the others). Thus, in principle the eigenvalues S1,2,3 can be different from each other,
say S3 > S2 > S1 . Then Eqs. (126) imply that SS1,2,3 should have an inverse hierarchy,
SS3 < SS2 < SS1 . More precisely, we have SS1 : SS2 : SS3 = S11 : S21 : S31 .
The flat directions of the VEVs can be fixed from the soft supersymmetry breaking
terms. The relevant ones are D-terms: 29

Z
2
L = d 4 zz Tr S S + 2 Tr S S + 2 Tr S SS S +
M
M
S
+ [S S]
(128)
having a similar form for S and S
S. Here z = m 2 and z = m 2 are supersymmetry breaking
spurions, with m 1 TeV. The cutoff scale M is taken as the flavour scale, i.e., the same
as the one in superpotential terms (24), and we assume that M . Therefore, for S1,2,3
these terms translate explicitly into the following scalar potential:

2
m2
m2 |S1 |2 + |S2 |2 + |S3 |2 + 2 |S1 |2 + |S2 |2 + |S3 |2
M

m2
+ 2 |S1 |4 + |S2 |4 + |S3 |4 + |Sk | |SSk | .
(129)
M
The stability of the potential implies that > 0 and > , whereas can be
positive or negative. In the former case the minimization of the potential (129), under
the conditions (127), would imply that S1 = S2 = S3 = , i.e., no hierarchy between the
fermion families. In the latter case, however, the largest eigenvalue of S and S
S, respectively,
S3 and SS1 , grow up above the typical VEV size and reach values of the order of the cutoff
scale M:
R 2
d zW are not relevant for the VEV orientation as far as they just repeat the holomorphic
S
invariants like S S and det S whose values are already fixed by the conditions (127).
29 The F -terms
S3 , SS1 =
2( + )
163
1/2
M
(130)
that is S3 , SS1 M. Then it follows from (127) that

S2 , SS2 = = S3 ,
2
S1 , SS3 =
= 2 S3 ,
S3
(131)
where = /S3 /M.

Let us assume now that the theory contains three triplets superfields An = Ani , and their
partners A n = A in , n = 1, 2, 3. One can incorporate them by the following terms in the
superpotential:
X
n An A n + A1 A2 A3 + A 1 A 2 A 3 ,
(132)
WA =
n
where order 1 coupling constants are understood and A1 A2 A3 ij k A1i A2j A3k . For
simplicity, we shall take all masses n equal, 1,2,3 = 0 < M. In addition, by assuming
that the triplets have Z3 charges different from that of S, we do not include terms like
SA2 S ij Ai Aj . In this way, in the exact supersymmetric limit the ground state has a
continuous degeneracy (flat direction) related to unitary transformations An U An with
U SU(3)A . In other terms, the superpotential W = WS + WA has an accidental global
symmetry SU(3)S SU(3)A , with the two SU(3) factors independently transforming the
two sets of horizontal superfields, S and A.
Similarly to the case of the Higgs fields S, S
S, now the conditions FA , FA = 0 can
only fix the values of holomorphic invariants An A n and A1 A2 A3 . Namely, by unitary
transformation An U An (U SU(3)A ), one can choose a basis where the VEV of
A1 points towards the first component. Then we see that in this basis the fields A2 and A3
should have the VEVs towards the second and third components. In other words, in this
basis the triplets have the VEVs hAni i = ni Ai , satisfying the following equations:
A1 A1 = A2 A2 = A3 A 3 = 02 ,
A1 A2 A3 = 03 ,
A 1 A 2 A 3 = 03 .
(133)
The three VEVs can be different and one can take, say A1 > A2 > A3 . The hierarchy
between the latter can be fixed by soft D-terms
X
Z

0 X
4
An An + 2
Am Am An An
L = d z z 0
M mn
n

0
X

Am An An Am + An A n
(134)
+ 2
M
m6=n
< 0. Then, similarly to what happened to the sextets, the largest VEV rises up
with
to the cutoff scale M: A1 , A3 M, while A2 , A 2 0 M and A3 , A 1 02 M, with 0
0 /M (0 /).
As for the relative orientation of the sextet and triplet VEVs, these will also be fixed
from the soft D-terms:
Z

X
An SS An + S S
S, An A n .
(135)
L = d 4 zz 2
M n
164
Indeed, for positive the favoured orientation between the S and A bases, which minimizes
the energy of the ground state, corresponds to U = 1.
The above can be interpreted in the following manner. In the view of operators like (24),
the MSSM Yukawa constants become dynamical degrees of freedom. In particular, the
operator Ou in (24) implies
Yu 0 0
S1 0
0
ij i
1
hS
0 S2 0 .
=
Yu = 0 Yc 0
(136)
M
M
0 0 Yt
0
0 S3
In the exact supersymmetric limit the values of the constants Yu,c,t are not fixed they
have flat directions and Eq. (127) translates into the constraint Yu Yc Yt = 3 . However,
we have shown that soft supersymmetry breaking terms could naturally split the Yukawa
constants so that Yt 1, while Yc and Yu 2 , which perfectly reflects the observed
pattern if 5 103 . (For discussions on dynamical Yt in supersymmetry see also
Ref. [43].)
As for the triplet VEVs, in view of the operator O (24) they provide off-diagonal Yukawa
entries
0
A3 A2
0
A3 A2
X hAij
1
i
n
A3
=
(137)
A = A3
0
A1
0
A1 ,
M
M
n
A2 A1 0
A2 A1 0
with the constraint A1 A2 A3 = 03 in the supersymmetric limit, while the soft breaking
terms can dynamically fix their values as A1 1, A2 0 and A3 02 . In the context of
our work it is clear that such a hierarchy of the off-diagonal entries in Yd,e is also well
suited for the observed fermion mass and mixing pattern if 0 0.1.
One has to remark that the above considerations are literally valid if the horizontal
symmetry is global. For the case of local SU(3)H , apart from F -flatness conditions, there
P
are eight additional conditions: the gauge D terms Da = n X T a X, should vanish on
the vacuum configuration (T a are SU(3)H generators, a = 1, . . . , 8). Although this does
not occur for the obtained VEV pattern of S and An , one can easily imagine the theory to
contain additional spectator superfields in different representations of SU(3)H and their
VEVs are oriented so that to cancel the contributions of hSi and hAi in gauge D-terms of
SU(3)H . In order not to affect the obtained solutions for S and An , these extra superfields
should not couple to the latter in the superpotential.
Appendix B
Let us assume now that the considered theory has an exact CP invariance. In other
words, we can choose a superfield basis where all coupling constants in the theory are real,
including the coupling constants with the matter superfields, i.e., those in the terms (24).
Applied to the superpotential (125) this means that the dimensional parameter as well as
S 3 are real. Without loose of generality, by means of a
the understood constants at S 3 and S
SU(3)H transformation, one can chose a basis where S3 and S2 are real and positive. Then,
165
since Eq. (127) tells us that the VEV products S1 S2 S3 = 3 is real, also the component S1
should be real. The same consideration is true for the VEVs of triplets An . Therefore, no
CP-violation can occur in this case.
Let us now consider an alternative superpotential for triplets (132) not containing the
mass terms but invoking instead an auxiliary singlet I :
WA = n I An A n A1 A2 A3 A 1 A 2 A 3 + 32 I + I 3 ,
(138)
where is a dimensional parameter, n are coupling constants and order one constants are
also understood in other terms. Applied to this superpotential, CP invariance means that n
as well as 2 are real. For simplicity, let us assume again that all n are equal, n = , and
1.
It is clear, that the F -term conditions FA , FA = 0 are the same as in (133) apart from the
fact that the mass scale 0 should be substituted by I, where I = hI i. The latter is then
fixed by the condition FI = 0 given as

1 X
Ak A k = 2 + 1 + 3 I 2 = 0.
2 + I 2 +
3
3
(139)
k=1
Hence, I 2 ' 2 , and if 2 is positive, then I should be imaginary and so we obtain

spontaneous CP violation. In other words, imaginary I means that the CP-odd component
of the superfield I acquires a VEV.
As a result, the induced mass 0 = I and so the VEV product A1 A2 A3 = 03 are
also imaginary. Therefore, at least one of the VEVs A1,2,3 should be complex and thus
would induce the corresponding phase in the fermion Yukawa matrices.
To fix the point, we have to find the relative phase orientation between the VEVs of S
and A. These should be determined by soft terms like
Z
(140)
L = d 4 zz 2 S S An An + h.c. ,
M
etc., with being a real constant. In particular, the above term induces the following
couplings in the Higgs potential:

m2
S3 S2 A21 + S3 S1 A22 + S2 S1 A23 + h.c.
(141)
2
M
We see that for > 0 the minimization of the ground state energy requires that in the
basis where Sk are real and positive, A21,2,3 < 0, i.e., all triplets have imaginary VEVs. For
< 0 we see instead that A1,2 prefer to be real. From the couplings (141) also the smallest
VEV A3 would prefer to be real; however it is forced to have a complex phase /2 by the
constraint that the product A1 A2 A3 is imaginary. Both these cases can be of interest for
the CP-breaking phases of the fermion Yukawa textures discussed in this work.
References
[1] SuperKamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1562, 2016;
166
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
M. Nakahata, in: TAUP 99, V Int. Workshop on Topics in Astroparticle and Underground
Physics, Paris, September 1999, to appear; Transparencies available at http://taup99.in2p3.fr/
TAUP99/;
M. Takita, in: Cosmo 99, Int. Workshop on Particle Physics and the Early Universe, Trieste,
September 1999; Transparencies available at http://www.ictp.it/cosmo99/.
For a review on the present status see, e.g., G.L. Fogli, in: TAUP 99, V Int. Workshop on Topics
in Astroparticle and Underground Physics, Paris, September 1999, to appear; Transparencies
available at http://taup99.in2p3.fr/TAUP99/;
N. Fornengo, M.C. Gonzalez-Garcia, J.W. Valle, hep-ph/0002147.
See, e.g., Y. Suzuki, in: LeptonPhoton 99 Proc. XIX Int. Symposium on Photon and Lepton
Interactions at High Energies, Stanford, California, USA, 1999, to appear; Transparencies
available at http://lp99.slac.stanford.edu.
J. Bahcall, P. Krastev, A. Smirnov, Phys. Rev. D 58 (1998) 096016;
M. Gonzalez-Garcia, P. de Holanda, C. Pena-Garay, J.W.F. Valle, hep-ph/9906469.
J. Bahcall, P. Krastev, A. Smirnov, Phys. Rev. D 60 (1999) 093001, hep-ph/9911248.
S.P. Mikheyev, A.Yu. Smirnov, Yad. Fiz. 42 (1985) 1441;
L. Wolfenstein, Phys. Rev. D 17 (1978) 2369.
S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566;
R. Barbieri, J. Ellis, M.K. Gaillard, Phys. Lett. B 90 (1980) 249;
E. Akhmedov, Z. Berezhiani, G. Senjanovic, Phys. Rev. Lett. 69 (1992) 3013.
M. Gell-Mann, P. Ramond, R. Slansky, in: D. Freedman et al. (Eds.), Supergravity, NorthHolland, Amsterdam, 1979;
T. Yanagida, Prog. Th. Phys. B 135 (1979) 66;
R. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
Particle Data Group, Eur. Phys. J. C 3 (1998) 1.
For a review see, e.g., Z. Berezhiani, in: Proc. 1995 Summer School, High Energy Physics
and Cosmology, E. Gava et al. (Eds.), The ICTP Series in Theoretical Physics, Vol. 12, World
Scientific, Singapore, 1995, p. 618, hep-ph/9602325;
R.N. Mohapatra, Lectures at ICTP Summer School in Particle Physics, Trieste, Italy, June 1999,
hep-ph/9911272.
Z. Berezhiani, A. Rossi, Phys. Lett. B 367 (1996) 219;
M. Drees, S. Pakvasa, X. Tata, T. ter Veldhuis, Phys. Rev. D 57 (1998) 5335;
R. Barbieri, L. Hall, D. Smith, A. Strumia, N. Weiner, JHEP 12 (1998) 017;
R. Barbieri, L. Hall, A. Strumia, Phys. Lett. B 445 (1999) 407;
G. Altarelli, F. Feruglio, Phys. Lett. B 439 (1998) 112;
G. Altarelli, F. Feruglio, JHEP 9811 (1998) 021;
H. Fritzsch, Z. Xing, Phys. Lett. B 440 (1998) 313;
B. Mukhopadhya, S. Roy, F. Vissani, Phys. Lett. B 443 (1998) 191;
Y. Grossman, Y. Nir, Y. Shadmi, JHEP 10 (1998) 007;
M. Gomez, G. Leontaris, S. Lola, J. Vergados, Phys. Rev. D 59 (1999) 116009;
S. Lola, G.G. Ross, Nucl. Phys. B 553 (1999) 81;
M. Tanimoto, Phys. Rev. D 59 (1999) 017304;
M. Tanimoto, Phys. Lett. B 456 (1999) 220;
G. Altarelli, F. Feruglio, I. Masina, hep-ph/9907532;
M.A. Diaz, J. Ferrandis, J.W. Valle, hep-ph/9909212;
D.A. Ryzhik, K.A. Ter-Martirosian, hep-ph/9909424;
E.K. Akhmedov, G.C. Branco, M.N. Rebelo, hep-ph/9912205;
For an alternative approach to the generation of maximal neutrino mixing see J.C. Romao,
M.A. Diaz, M. Hirsch, W. Porod, J.W. Valle, Phys. Rev. D 61 (2000) 071703.
Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 409 (1997) 220;
Z. Berezhiani, Z. Tavartkiladze, Phys. Lett. B 396 (1997) 150;
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
167
C. Albright, K.S. Babu, S. Barr, Phys. Rev. Lett. 81 (1998) 1167;

G. Altarelli, F. Feruglio, Phys. Lett. B 451 (1999) 388;
T. Blazek, S. Raby, K. Tobe, Phys. Rev. D 60 (1999) 113001, hep-ph/9912482.
Z. Berezhiani, A. Rossi, JHEP 9903 (1999) 002, hep-ph/9811447;
Z. Berezhiani, A. Rossi, Nucl. Phys. Proc. Suppl. B 81 (2000) 346, hep-ph/9907397.
CHOOZ Collaboration, M. Apolonio et al., Phys. Lett. B 420 (1998) 397;
CHOOZ Collaboration, M. Apolonio et al., Phys. Lett. B 466 (1999) 415;
See also the initial results by Palo Verde experiment, F. Boehm et al., hep-ex/9912050.
H. Fritzsch, Nucl. Phys. B 155 (1979) 189.
B. Stech, Phys. Lett. B 130 (1983) 189.
Z. Berezhiani, J. Chkareuli, JETP Lett. 35 (1982) 612;
Z. Berezhiani, J. Chkareuli, Yad. Fiz. 37 (1983) 1043;
Z. Berezhiani, J. Chkareuli, Sov. J. Nucl. Phys. 37 (1983) 618.
J. Chkareuli, JETP Lett. 32 (1980) 671.
Z. Berezhiani, E. Nardi, Phys. Lett. B 355 (1995) 199;
Z. Berezhiani, E. Nardi, Phys. Rev. D 52 (1995) 3087.
Z. Berezhiani, M. Khlopov, Yad. Fiz. 51 (1990) 1157;
Z. Berezhiani, M. Khlopov, Yad. Fiz. 51 (1990) 1479.
Z. Berezhiani, G. Dvali, M. Jibuti, J. Chkareuli, in: A. Tavkhelidze et al. (Eds.), Proc. Int.
Seminar Quarks 86, Tbilisi, 1986, INR Press, Moscow, 1986, p. 209.
C.D. Frogatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277;
Z. Berezhiani, Phys. Lett. B 129 (1983) 99;
S. Dimopoulos, Phys. Lett. B 129 (1983) 417.
A. Anselm, Z. Berezhiani, Nucl. Phys. B 484 (1997) 97.
H. Georgi, C. Jarlskog, Phys. Lett. B 86 (1979) 297;
J. Harvey, P. Ramond, D. Reiss, Phys. Lett. B 92 (1980) 309;
S. Dimopoulos, in Ref. [22];
G. Anderson, S. Dimopoulos, L.J. Hall, S. Raby, G. Starkman, Phys. Rev. D 49 (1994) 3660.
Z. Berezhiani, G. Dvali, Sov. Phys. Lebedev Inst. Rep. 5 (1989) 44;
R. Barbieri, G. Dvali, M. Moretti, Phys. Lett. B 312 (1993) 137;
R. Barbieri et al., Nucl. Phys. B 432 (1994) 49;
Z. Berezhiani, C. Csaki, L. Randall, Nucl. Phys. B 444 (1995) 61;
G. Dvali, S. Pokorski, Phys. Rev. Lett. 78 (1997) 807.
J. Chkareuli, A. Kobakhidze, Phys. Lett. B 407 (1997) 234;
J. Chkareuli, I. Gogoladze, A. Kobakhidze, Phys. Rev. Lett. 80 (1998) 912;
J. Chkareuli, C.D. Froggatt, I. Gogoladze, A. Kobakhidze, hep-ph/0003007.
S. Dimopoulos, F. Wilczek, Report No. NSF-ITP-82-07 (unpublished).
K.S. Babu, S.M. Barr, Phys. Rev. D 48 (1993) 5354.
Z. Berezhiani, Z. Tavartkiladze, Ref. [12].
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 585;
J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109;
M. Dine, A. Ichinose, N. Seiberg, Nucl. Phys. B 293 (1987) 253.
S. Barr, Phys. Rev. D 24 (1981) 1895;
Z. Berezhiani, Yad. Fiz. 42 (1985) 1309;
Z. Berezhiani, Sov. J. Nucl. Phys. 42 (1985) 825;
G. Dvali, S. Pokorski, Phys. Lett. B 379 (1996) 134.
V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 47 (1993) 2038.
P. Chankowski, Z. Pluciennik, Phys. Lett. B 316 (1993) 312;
C.N. Leung, K.S. Babu, J. Pantaleone, Phys. Lett. B 319 (1993) 191;
168
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
A. Brignole, H. Murayama, R. Rattazzi, Phys. Lett. B 335 (1994) 345;

A. Smirnov, F. Vissani, Phys. Lett. B 341 (1994) 173.
J. Ellis, S. Lola, Phys. Lett. B 458 (1999) 310;
N. Haba, N. Okamura, hep-ph/9810471;
J.A. Casas, J.R. Espinosa, A. Ibarra, I. Navarro, hep-ph/9910420.
For a review and references, see for instance, A.J. Buras, R. Fleischer, in: A.J. Buras, M. Lindner
(Eds.), Heavy Flavours II, World Scientific, 1997, hep-ph/9704376.
Z. Berezhiani, Nucl. Phys. B Proc. Suppl. A 52 (1997) 153;
Z. Berezhiani, in: R.N. Mohapatra, A. Rasin (Eds.), Proc. Fourth Int. Conf. SUSY 96, NorthHolland, hep-ph/9607363.
A. Kostelecky, L.J. Hall, S. Raby, Nucl. Phys. B 267 (1986) 415;
R. Barbieri, L.J. Hall, Phys. Lett. B 338 (1994) 212.
See, e.g., Y. Grossman, Y. Nir, R. Rattazzi, in: A.J. Buras, M. Lindner (Eds.), Heavy Flavours
II, World Scientific, 1997, p. 755, hep-ph/9701231;
Y. Nir, hep-ph/9911321.
H. Leutwyler, Talk given at the Conference on Fundamental Interactions of Elementary
Particles, ITEP, Moscow, Russia, 1995, hep-ph/9602255.
The Neutrino Group of the prospective study on + colliders, B. Autin et al., CERNSPSC/98-30, October 1998;
See also A. de Rujula, M. Gavela, P. Hernandez, Nucl. Phys. B 547 (1999) 21;
V. Barger, S. Geer, R. Raja, K. Whisnant, hep-ph/9911524.
F. Parodi, P. Roudeau, A. Stocchi, hep-ex/9903063;
See also the theoretical discussion by M. Neubert, hep-ph/9904321.
A. Pomarol, D. Tommasini, Nucl. Phys. B 466 (1996) 3;
R. Barbieri, G. Dvali, L.J. Hall, Phys. Lett. B 337 (1996) 76;
R. Barbieri, L.J. Hall, Nuovo Cimento A 110 (1996) 1;
R. Barbieri, L.J. Hall, S. Raby, A. Romanino, Nucl. Phys. B 493 (1997) 3.
P. Binetruy, E. Dudas, Phys. Lett. B 338 (1994) 23;
C. Kounnas, I. Pavel, G. Ridolfi, F. Zwirner, Phys. Lett. B 354 (1995) 322.

Noncommutative field theory from string theory:

two-loop analysis
Youngjai Kiem a , Sangmin Lee a, , Jaemo Park b
a School of Physics, Korea Institute for Advanced Study, Seoul 130-012, South Korea
b School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540, USA
Received 16 August 2000; accepted 10 November 2000
Abstract
Noncommutative 3 field theory in six dimensions exhibits the logarithmic UV/IR mixing at
the two-loop order. We show that open string theory in the presence of constant background NS
NS two-form field yields the same amplitude upon taking a decoupling limit. The stretched string
picture proposed on the basis of one-loop analysis naturally generalizes to the two-loop amplitudes
in consideration. Our string theory formulation can incorporate the closed string insertions as well as
open string insertions. Furthermore, the analysis of the world-sheet partition function and propagators
can be straightforwardly generalized to Riemann surfaces with genus zero but with an arbitrary
number of boundaries. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Since the realization that certain noncommutative field theories are natural decoupling
limits of string theory [18], there have been a number of startling discoveries on
the physics in noncommutative spacetime. One striking example is the UV/IR mixing
in noncommutative field theory [9]; in nonplanar amplitudes of noncommutative field
theories, novel IR divergences at zero momentum come from the UV regime of the loop
momentum integration [10,11]. To understand this phenomenon in the Wilsonian effective
description, it was suggested that some extra (closed string) degrees of freedom might
survive the decoupling limit [10].
One useful vantage point for understanding this issue is to go back to the string theory
itself, and carefully examine what mechanisms are responsible for the UV/IR mixing. In
this spirit, there have been attempts to recover the (nonplanar) noncommutative field theory
amplitudes from the direct string theory loop calculations; indeed, at the one-loop level,
E-mail addresses: ykiem@kias.re.kr (Y. Kiem), sangmin@kias.re.kr (S. Lee), jaemo@ias.edu (J. Park).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 7 3 - 8
170
Y. Kiem et al. / Nuclear Physics B 594 (2001) 169189
one now has a fairly complete understanding of the string theory calculations [1218]. The
upshot is that, at the one-loop level, while one can add some extra (closed string-like or
closed string-inspired) degrees of freedom to the effective action, which, upon integrating
out, yield the correct IR divergence at least in the field theory analysis, it appears equally
possible that the UV/IR mixing may be a purely open string phenomenon. The stretched
string interpretation of [17] gives us a concrete example of the latter. Analysis of the multiloop amplitudes should be in order.
In this paper, we develop a world-sheet approach to the noncommutative multi-loop
amplitude calculation in string theory. Based on this approach, we analyze the twoloop logarithmic UV/IR mixing phenomenon in 3 field theory in six dimensions. Our
calculation indicates that the stretched string interpretation advocated in [17] can be
extended to the multi-loop amplitudes corresponding to nonplanar vertex insertions on
planar vacuum world sheets. It remains to be seen whether the similar purely open string
interpretation is possible for the nonplanar vacuum world sheet.
In Section 2, we review the one- and two-loop UV/IR mixing in noncommutative 3
theory in six dimensions. We recast the field theory amplitudes in a form which is
straightforward to compare with the string theory calculations, following the line of
investigations originating from the work of Bern and Kosower [1921]. In Section 3, based
on the multi-loop string amplitude analysis in the absence of background NSNS two-form
field (B-field) [2227], we study the modifications due to a constant background B-field.
Both closed string and open string world-sheet propagators are constructed for Riemann
surfaces with boundaries (with genus zero), along with the world-sheet partition function.
Using these inputs, we explicitly compute two-loop nonplanar amplitudes, which yield the
two-loop amplitudes obtained in Section 2 upon taking the SeibergWitten decoupling
limit [8]. Some relevant background material and details are presented in appendices.
In Section 4, we investigate the decoupling limit and the UV/IR mixing in the two-loop
context.
Recently noncommutative multi-loop analysis was reported in Ref. [28] based on
Reggeon vertex formalism. Our approach produces the same amplitudes as the Reggeon
vertex formalism. Furthermore, it supplements that formalism in the sense that it is
straightforward in our approach to consider the closed string vertex insertions, while the
Reggeon vertex formalism applies only to the purely open string vertex insertions.
2. 3 theory in D = 6: one- and two-loop amplitudes
The noncommutative 3 theory in D dimensions is described by the action

Z
1
1 2 2 1
D
2
() + m + g ,
I= d x
2
2
3!
where the -product is defined as

i
(y)(z)|y=z=x .
(x) = exp
2
y z
We will be primarily interested in the D = 6 case in this paper.
(2.1)
(2.2)
171
Fig. 1. 3 theory Feynman diagrams.
At the one-loop level, the 1PI Feynman diagrams contributing to the two-point and
three-point vertices shown in Fig. 1 exhibit the UV/IR mixing, the occurrence of the IR
divergence from the UV corner of the momentum integral. 1 For the two-point vertex, using
the noncommutative Feynman rules, we have
V2(1) (p1 , p2 ) = g 2 (p1 + p2 )W2(1) (p1 , p2 ),
where
1
exp(ik p2 )
+ p1 )2 + m2 )

Z Z t
1
2
D/2 m2 t
+ p1 p2 .
dt d t
e
exp p1 p2
=
t
t
W2(1) =
(2.3)
dDk
(k 2
+ m2 )((k
(2.4)
Here the products are defined as p1 p2 = p1 p2 , p1 p2 = p1 p2 , and p1 p2 =

14 p1 ( 2 ) p2 and we use the Schwinger parameterization of the internal propagators
going to the second line of (2.4). To study the UV behavior (t 0) of the amplitude (2.4),
we rescale the coordinate into = /t to make the integration range t-independent. We
then find that as t 0, we may retain only the -product term in the exponential function,
(1)
resulting W2 1/(p1 p1 ) in D = 6. This shows that (2.4) is quadratically IR divergent
as the external momentum goes to zero, namely the one-loop quadratic UV/IR mixing.
Similarly, the one-loop three-point vertex
i
V3(1) (p1 , p2 , p3 ) = g 3 (p1 + p2 + p3 )e 2 p2 p3 W3(1) (p1 , p2 , p3 )

1 Hereafter, we will neglect the overall normalization of each Feynman diagram.
(2.5)
172
in Fig. 1(b) shows the logarithmic UV/IR mixing:

Z
1
exp(ik p3 )
dDk 2
W3(1) =
2
2
(k + m )((k + p1 ) + m2 )((k + p1 + p2 )2 + m2 )
Z Z t
Z1
2
dt d1 d2 t D/2 em t
=

22
(1 2 )2
+ p2 p3 2
exp p1 p2 1 2
t
t

2
2
1
+ ip2 p3 + p2 p3 + p1 p3 1 1
t
t
t

1
1
+ ip1 p3 + p1 p3 .
t
t
0
(2.6)
(1)
Rescaling the i coordinates by t as in (2.4) reveals that W3 log(p3 p3 ) as t 0

when D = 6. We again note that as t 0, we may retain only the -product part in the
exponential function.
In this paper, we will present a full analysis of two-loop (Tr )3 terms in the effective
action. In Ref. [28], one finds an analysis of two-loop two-point terms in the effective
action. Modulo the renaming of the internal propagators and external insertions, all
possible two-loop (Tr )3 Feynman diagrams are Fig. 1(c) and (d). We evaluate the diagram
in Fig. 1(c) using the noncommutative Feynman rules and find the correction to the threepoint vertex:
(2)
(2)
(p1 , p2 , p3 ) = g 5 (p1 + p2 + p3 )e 2 p2 p3 W3(c)
(p1 , p2 , p3 ),
V3(c)
i
where
(2)
W3(c)
Z
=

d D ki (k1 + k2 + k3 ) exp(ik1 p3 ik2 p2 )
3
Y
(k 2
i=1 i
+ m2 )((k
+ pi )2 + m2 )
For each internal propagator, we introduce Schwinger parameters via (i = 1, 2, 3)

1
=
2
k i + m2
(2.7)

di exp ki2 + m2 i ,
1
=
(ki + pi )2 + m2
and use
(k1 + k2 + k3 ) =

di exp (ki + pi )2 + m2 i ,

d Dw exp i(k1 + k2 + k3 ) w ,
to rewrite (2.8) in the following form:
(2.8)

3 Z
Y
Zti
(2)
W3(c)
dti
i=1 0
di (t1 t2 + t2 t3 + t3 t1 )D/2 em
2 (t
173
1 +t2 +t3 )

exp p1 p2 F12 + ip1 p2 G12 + p1 p2 H12 + (cyclic) .
(2.9)
Here the functions F12 , G12 , H12 are defined as

F12 (1 , 2 ) = 1 + 2
12 t2 + 22 t1 + (1 + 2 )2 t3
,
t1 t2 + t2 t3 + t3 t1
1 t2
,
t1 t2 + t2 t3 + t3 t1
t1 + t3
,
H12 =
t1 t2 + t2 t3 + t3 t1
G12 =
(2.10)
(2.11)
(2.12)
and similarly for their cyclic permutations. Going from (2.8) to (2.9), we explicitly perform
the ki - and w-integrals, introduce ti = i + i , and use the momentum conservation
p1 + p2 + p3 = 0. When the noncommutativity parameter = 0, the amplitude (2.9)
reduces to that of Ref. [20].
To examine the UV and IR limits of the amplitude, it is convenient to use a spherical
polar coordinate on the t space (t 2 t12 + t22 + t32 and two angles) and make a variable
change i = i /ti . This makes the integration range of i s be independent of ti s. The
scaling behavior of each type of term in (2.9) is readily seen to be
F t,
G 1,
H t 1 .
(2.13)
The UV limit of the amplitude (2.9) is where t goes to zero with the angles kept fixed. In
that limit, (2.9) becomes
"
!#
Z
Z
3
1 X
(2)
5D
dt t
exp
pi pi Ki ,
(2.14)
W3(c) d
t
S2
i=1
t 0
where Ki is defined as Ki = ti+2 /(t1 t2 + t2 t3 + t3 t1 ) and similarly for cyclic permutations.

Note that we use an identity
3
X
pi pi+1 (ti + ti+2 ) =
i=1
3
X
pi pi ti+2 .
i=1
The other terms on the exponent of (2.9), including the mass term, can be neglected near
t = 0. From (2.14), we see that there are logarithmic UV singularities when D = 6 if
p1 p1 + p2 p2 + p3 p3 = 0 (we recall that p p > 0 for an arbitrary p)
!
3
X
(2)
pi pi .
(2.15)
W3(c) log
i=1
Therefore, the contributions from the UV corner of the Schwinger parameters produce the
IR divergence when all the external momenta satisfy pa pa = 0, the two-loop logarithmic
UV/IR mixing. The potential IR divergence when t gets regulated by the mass term
that scales like t in (2.9).
174
The 1PI amplitude of the diagram Fig. 1(d)

i
V3(d)(p1 , p2 , p3 ) = g 5 (p1 + p2 + p3 )e 2 p2 p3 W3(d)(p1 , p2 , p3 ),

(2)
where
(2)
W3(d)
Z
=
(2)
(2.16)

d D ki (k1 + k2 + k3 ) exp(ik1 p3 ik2 p2 )
" 3
Y

ki2 + m2 (k1 + p3 )2 + m2 (k1 + p3 + p1 )2 + m2
i=1
(k2 + p2 ) + m
2
#1
(2.17)
can also be written in a similar fashion:

3 Z
Y
Zt2
(2)
W3(d)
i=1 0
dti
Zt1
d2
Z3
d3
d1 (t1 t2 + t2 t3 + t3 t1 )D/2 em
2 (t
1 +t2 +t3 )

e31 (3 , 1 )
exp p1 p2 F12 (1 , 2 ) + p2 p3 F12 (3 , 2 ) + p3 p1 F

1 t2 + 2 t3 + 3 t3
+ p1 p2 H12 + (cyclic) , (2.18)
exp ip1 p2
t1 t2 + t2 t3 + t3 t1
e31 is defined as
where the function F
2
e31 (3 , 1 ) = |3 1 | (t2 + t3 )(3 1 ) ,
(2.19)
F
t1 t2 + t2 t3 + t3 t1
and the momentum conservation implies p1 p2 = p2 p3 = p3 p1 . A straightforward
analysis shows that (2.18) is logarithmically divergent when p2 p2 = 0 and it is finite
otherwise.
3. Two-loop string amplitudes

One can relate a field theory Feynman diagram to a string diagram by thickening the
lines. By thickening a l-loop planar vacuum diagram, we find that the relevant world-sheet
has the topology of genus g = 0 and the boundary b = l + 1 surface, or the (0, l + 1)
surface if we introduce the notation (gb) to denote a surface with g handles and b
boundaries. This surface can be equivalently viewed as an upper half of the (l0) surface.
The boundaries are the fixed points, w = I (w), under the involution I that identifies the
upper and lower hemi-surfaces. Along these boundaries, we should impose an appropriate
boundary condition
g n X + iB t X = 0|w=I (w) ,
(3.1)
where n and t are normal and tangential derivatives to the boundaries.

The main ingredients for the computation of the string theory amplitudes are the world
sheet propagators and the partition function. The main goal of this section is to compute
175
Fig. 2. The (03) surface in Schottky representation.
both objects for the (03) surface that corresponds to the two-loop diagrams considered in
the previous section. However, most of the discussion on the world sheet partition function
and the entire subsection on the world sheet propagators will be valid for a surface with
arbitrary number of boundaries but no handles.
We will see that it is often convenient to use the Schottky representation for a Riemann
surface. For the (03) surface, the representation is depicted in Fig. 2, where we follow the
conventions of Ref. [21]. Here we will only give an intuitive picture of the (03) surface;
a more systematic and self-contained introduction to the Schottky representation is given in
Appendix B. In Fig. 2, the (03) surface is the region in the upper half plane surrounded by
the solid lines and semicircles. The involution in this representation is simply the complex
conjugation, that is, I (z) = z . The mirror hemi-surface under the involution is enclosed
by dotted curves in the lower half plane. The two circles C1 and C10 are identified and
similarly for the C2 and C20 circles. After the identification, it becomes clear that the three
boundaries are A0 A, BC C 0 B 0 and DD 0 .
3.1. World-sheet propagator
Our strategy is to compute the closed string world sheet propagators and to extract
the open string propagators from them. For this purpose, it is helpful to start from the
consideration of the one-loop annulus propagators.
3.1.1. One-loop propagator revisited
Following [13] we write down the one-loop bulk propagator,
176
X (z)X (z0 )
=
0

( G )

0
0 2
g G z, z0 +
2G g G z, z 0
x
+
x
2
2
2 0
T

0 |iT )

2i
(z
+
z
1
1
log
+
x + x0 y y0 ,
+
2
1 (z + z0 |iT ) T
(3.2)
where the open string metric G and the noncommutativity parameter are given in
= 2 0 (g
terms of the closed string variables by G = (g + B )1
+
S and
1
0
B )A . The function G(z, z ) is defined as

1 (z z0 |iT ) 2 2

2
+
y y0 .
(3.3)
G z, z0 = log
0

1 (0|iT )
T
The variables x and y are the real and imaginary parts of z, respectively, and T denotes
the annulus modulus. The two boundaries of the world-sheet are at x = 0, 1/2, and the
propagators are periodic in y y + T .
To compute open string amplitudes, one needs a boundary propagator. Naively, one
might expect to obtain the boundary propagator by taking the insertion points in (3.2) to the
boundary. When B 6= 0, however, this procedure does not give the correct answer. To see
this, we note that the quadratic terms in (3.2) as they stand do not treat the two boundaries
on an equal footing. A more rigorous way to derive the boundary propagator from the
bulk propagator is to use the factorization of the string amplitudes. When computing
the amplitudes in that derivation, care should be taken to incorporate the effect of the
self-contractions. In the presence of world-sheet boundaries, it is well-known that the
contraction between a closed string vertex and its own mirror image should be included
in such calculations. 2 For example, the tachyon amplitude contains
#
"
X
X
i j
i i
G (zi , zj )k k
Gs (zi )k k ,
(3.4)
exp
j <i
where in the case at hand, the self-contraction is defined by

0
1
2
G g
( G ) x 2 .
G(z, z )
Gs (z) =
2
2
2 0 T
(3.5)
As a closed string vertex operator approaches a boundary, the term involving the theta
function in G(z, z ) becomes singular and generates the propagator for a virtual particle
emitted from the boundary. On the other hand, the zero mode part remains and gets
absorbed into the mutual-contractions via momentum conservation. Specifically, we make
P
use of the following identity that holds when i ki = 0
X
X
X
ki kj (xi + xj )2 + 2
ki2 xi2 =
ki kj (xi xj )2 .
(3.6)
j <i
j <i
2 The normal ordering of a closed string vertex operator is V =: exp(ikX(z)) :: exp(ikX(z)): in contrast to
that of an open string vertex operator V =: exp(ik(X(z) + X(z))) : |z=z . When the world sheet has a boundary,
X(z) and X(z) do not commute and produce self-contractions. At one-loop level, the operator method illustrates
this aspect well. This method is useful since one does not need the knowledge of the world sheet propagator
beforehand. See Appendix C for an outline of the operator method for B 6= 0.
It is useful to use another identity

X
X
ki kj (xi + xj )(yi yj ) =
ki kj (xi xj )(yi + yj ),
j <i
177
(3.7)
j <i
though it cannot be accounted for by the self-contraction. Taking these zero-mode effects
into account, one obtains the planar and nonplanar boundary propagators,

i
(3.8)
GP z, z0 = 0 G G z, z0 + z z0 ,
2

( G )
2 2i

x x 0 x x 0 y + y 0 , (3.9)
GNP z, z0 = 0 G G z, z0 +
0
2 T
T
in complete agreement with Ref. [28], where the Reggeon vertex formalism is used.
3.1.2. Multi-loop propagator
With detailed understanding of the one-loop propagators, it is now straightforward
to obtain their multi-loop generalizations. The B-field background does not cause any
complications except what we already encountered at one-loop.
The first step toward the generalization is to replace the theta function in (3.2) by the
prime form reviewed in Appendix A. In this process, one should note that the definition of
Rz
Rz
the prime form (A.9) involves the integrals z0 and z 0 . These integrals depend on the
path of integration. To be precise, two paths give the same value for the integrals if and
only if the two paths are homotopic to each other. Fig. 2 gives an example of two paths P
and P 0 that are not homotopically equivalent. Therefore, in order for the propagator to be
well-defined, we should make a specific choice of the paths.
A similar ambiguity arises for the multi-loop analogue of the quadratic part in (3.2). To
see this, we define the variables x(z) and y(z) to be the real and imaginary part of the
integral of the Abelian differentials along a path from a reference point p to z (see Fig. 2);
Zz
(z) x(z) + iy(z),
(3.10)
where the reference point p is chosen to be an arbitrary point on the boundaries. When
computing x(z) + iy(z), we assume that the path from p to z lies entirely in the worldsheet and the path from p to z , the mirror point, lies entirely in the mirror world-sheet.
We also demand that the paths never warp through the pairs of circles. Using the explicit
form of the Abelian differentials given in the appendix, one can show that
x(z) = x(z),
y(z) = y(z).
(3.11)
As z approach a boundary (z = z ), one might be tempted to say that x(z) = x(z) and
conclude in view of (3.11) that x = 0. This is true when z lies on the same boundary as the
reference point p, but otherwise x(z) differs from x(z) by 1 since the difference between
the two paths form a cycle homologous to one of the four circles (Ci , Cj0 ). In this sense,
the function x(z) has branch cuts along the boundaries that do not contain the reference
point. We recall that the x + x 0 term in (3.2) measures a distance between a point and the
178
mirror image of another point reflected along the x = 0 boundary (not the x = 1/2 one).
This choice is equivalent to the choice of the reference point in our present discussion.
Using the variables x and y, we rewrite the arguments of the prime form as
Zz
Zz
= x x +i y y ,
z0

= x + x0 + i y y0 .
(3.12)
z 0
This expression together with the choices made in the definition of x and y completely
fixes the ambiguity. At one loop, we observed that despite the apparent breaking of the
symmetry between the two boundaries due to (x + x 0 ), the correct incorporation of selfcontractions restored the symmetry in the physical amplitude. In the same way, although we
must choose a reference point to define the propagators, the final answer for the amplitude
will not depend on the choice.
We are now ready to write down the propagators. The bulk propagator is given by
X (z)X z0
0

0
g G z, z0 +
2G g G z, z 0
2
2

1
0
( G ) x + x 0 (T )1
x + x
0
2

E(z, z 0 )
1
1
0
0
log
+
+ 2i x + x (T ) y y
,
2
(E(z, z 0 ))
(3.13)
and the planar and nonplanar boundary propagators are

i
GP z, z0 = 0 G G z, z0 + z z0 ,
2

1
0
( G ) x x 0 (T )1
GNP z, z0 = 0 G G z, z0 +
x x
2 0

0
2i x x 0 (T )1
,
y + y
where the function G(z, z0 ) is given by

2

0
.
G z, z0 = logE z, z0 + 2 y y 0 (T )1
y y
(3.14)
(3.15)
(3.16)
Here the matrix T is the imaginary part of the period matrix. In fact, for the Schottky
representation of the (03) surface, the period matrix is purely imaginary as explained in
the appendix. Note that the boundary propagators (3.14), (3.15) depends only on x x 0
that is defined unambiguously and independently of the reference point p. The value of
(y +y 0 ) still depends on the reference point, but the dependence drops out from the physical
amplitude due to momentum conservation as we will see shortly.
3.2. World-sheet partition function
In the absence of the B-field background, the partition function for the (03) surface has
been known for some time [23,24]. For N Dp-branes, the answer is given by (up to an
overall normalization factor):
179
Fig. 3. Getting two-loop Riemann surfaces from an annulus.
Z
Z(03) = N 3
dT11 dT22 dT12
|W (T )|
,
(det T )(p+1)/2
(3.17)
where
10

Y

a (0|iT )2 .
W (T ) =
a=1
Here a s are the ten even Riemann theta functions for the g = 2 surface.
For the one-loop (or two boundaries) partition function, an explicit computation [22]
shows that the nonzero B-field background only changes the overall normalization factor.
We argue here that the same should be true for arbitrary number of boundaries as long as
g = 0. 3
One may compute the partition function recursively by a gluing process. Specifically,
one starts from a disk (the (01) surface) and insert two vertex operators along the boundary.
By connecting these two vertex insertions and summing over all possible intermediate
vertex operators, one gets the annulus (02) partition function. Since all vertex insertions
are planar, the gluing process cannot generate any nontrivial B dependence. This explains
why the B-dependent factor of the annulus partition function factors out.
At the two-loop level, there are two possible values of g and b, giving the Euler
characteristic = 1: (11) and (03) (Fig. 3(a) and (b)). For the former partition function,
we insert two vertex operators in a nonplanar fashion, connect them and sum them over all
possible vertex operators. In Fig. 3(b), it is explained how this procedure produces g = 1,
b = 1 world sheet. In this case, the propagator expression (3.15) shows that there are zeromode contributions from the G part. As a result, the partition function now contains a
nontrivial dependence. On the other hand, for (03), one insert two vertex operators in
the planar fashion and repeat the same procedure as before. Since there are no nontrivial
B-dependent (zero-mode) contributions from the propagators (as seen from (3.14)), the
resulting partition function is the same as the one computed for the B = 0 case, up to
3 After completion of this paper, we were informed that Ref. [29] gave another argument and actually computed
the normalization factor for arbitrary number of boundaries.
180
a trivial overall multiplicative factor. By recursively adding two planar vertex insertions
along the same boundary, one can show that all g = 0, b = l + 1 partition function is the
same as the one for B = 0 case, modulo a trivial multiplicative factor.
3.3. The two-loop amplitudes
We are interested in the (Tr )3 three-point amplitudes, which are related to the field
theory amplitudes in Fig. 1(c) and (d), and the string amplitudes are given by
Z
|W (T )|
dy1 dy2 dy3 dt1 dt2 dt3
(det T )(p+1)/2

exp p1 p2 GNP (z1 , z2 ) + (cyclic) .

(3.18)
Written explicitly, the world-sheet nonplanar boundary propagator GNP (z1 , z2 ) in (3.15)
becomes

2
p1 p2 GNP (z1 , z2 ) = 0 p1 p2 logE(z1 , z2 )

+ 2 0 p1 p2 (y1 y2 ) (T )1
(y1 y2 )
2ip1 p2 (x1 x2 ) (T )1
(y1 + y2 )
p1 p2 (x1 x2 ) (T )1
(x1 x2 ) .
(2 0 )
(3.19)
In accordance with Fig. 1(c) and (d), the insertion points are z3 A0 A, z1 BC, and
z2 DD 0 when seen in Fig. 2. We parameterize the imaginary components T of the period
matrix as

t11 t12
t1 + t3
t3
=
,
(3.20)
2 0 T =
t12 t22
t3
t2 + t3
which implies that

1
t2 + t3
=
2 T
0
t3
det(2 T )

det 2 0 T = t1 t2 + t2 t3 + t3 t1 .
0
1
t3
t1 + t3

,
A direct computation in the Schottky representation gives

0
1/2
1/2
,
x3 x2 =
,
x1 x3 =
.
x2 x1 =
1/2
1/2
0
(3.21)
(3.22)
The quantities in (3.22) become topological for open string insertions; they do not change
as we locally move the position of the vertex insertions, in marked contrast to the closed
string insertions (see also [30]). Using (3.21) and (3.22), we find that
2 0 p1 p2 (y1 y2 ) (T )1
(y1 y2 ) + (cyclic)
=
2
2
(2 0 )2 p1 p2
(y1 y2 )2 t1 + (y1 y2 )1 t2
t1 t2 + t2 t3 + t3 t1
2
+ (y1 y2 )1 + (y1 y2 )2 t3 + (cyclic),
(3.23)
181
2i(p1 p2 )(x1 x2 ) (T )1
(y1 + y2 ) + (cyclic)
=i
(2 0 )(p1 p2 )
t1 (y3 y1 )2 + t2 (y1 y2 )1
t1 t2 + t2 t3 + t3 t1

+ t3 (y3 y2 )1 + t3 (y3 y2 )2 ,
p1 p2 (x1 x2 ) (T )1
(x1 x2 ) + (cyclic)
(2 0 )
p1 p2 (t1 + t3 ) + p3 p1 (t2 + t3 ) + p2 p3 (t1 + t2 )
,
=
t1 t2 + t2 t3 + t3 t1
(3.24)
(3.25)
using p1 p2 = p2 p3 = p3 p1 via the momentum conservation. The expression (3.15)

depends upon the combination y + y 0 , which in turn depends on a particular choice of the
reference point p in Fig. 2. However, at the level of the physical amplitudes, we observe
that the dependence on p drops out upon the imposition of the momentum conservation,
as can be seen from the y y 0 dependence of (3.24).
4. Reduction from string theory to field theory: decoupling limit and UV/IR mixing
Upon taking a decoupling limit, a given string theory amplitude reproduces various field
theory amplitudes represented by differing field theory Feynman diagrams, by considering
appropriate corners of the moduli space. For the two-loop field theory amplitude depicted
in Fig. 1(c), the reduction from the string theory amplitude has been worked out in detail
in Ref. [20] in the commutative field theory setup. Following the analysis of Ref. [20], in
the decoupling limit 0 0, we compute

1
0
,
2 (y2 y1 ) =
2

3
0
,
2 (y3 y2 ) =
2 + 3

1 + 3
,
(4.1)
2 0 (y1 y3 ) =
3
which relates (yi yj ) to the field theory Schwinger parameters i . Following Seiberg and
Witten [8], all the open string quantities, such as the open string metric G and , are kept
fixed as we take the 0 0 limit:
2 0 y = fixed,
2 0 T = t fixed.
(4.2)
The conventions of Ref. [20] ensure that we recover the field theory propagator from the
two-loop string Green function in the commutative case = 0. In the noncommutative
case, one can easily check that the terms (-product terms) and the 2 terms (-product
terms) of the field theory amplitude (2.9) is correctly reproduced from the string theory
amplitude ((3.24) and (3.25), respectively); the string partition function reproduces the
first line of (2.9) upon deleting the contributions from the massive string modes, and the
zero mode parts of the string propagators give the second line of (2.9), where the linear
182
terms in s in (2.10) come from the zero mode parts of the generalized theta functions. In
a similar fashion, to recover (2.18) depicted in Fig. 1(d) from the string theory amplitude,
we compute

1
,
2 0 (y2 y1 ) =
2

3
,
2 0 (y3 y2 ) =
2

1 3
,
(4.3)
2 0 (y1 y3 ) =
0
corresponding to a different corner of the moduli space, in the decoupling limit.
The outstanding issue, then, is to understand the two-loop logarithmic UV/IR mixing
term (2.15) (see also (2.12), (3.25)), which originates from the ( G ) part of the string
propagators.
4.1. Stretched string interpretation
The stretched string interpretation of the UV/IR mixing, which does not involve the
consideration of extra light (closed string) degrees of freedom was suggested in Ref. [17]
at the level of one-loop analysis. When it comes to the one-loop two-point nonplanar
amplitudes, we insert vertex operators along each of the two boundaries of an annulus.
The analog UV/IR mixing term in this context is
1
p p,
(4.4)
2 0 T
where p is the external momentum.
Even in the decoupling limit 0 0, (4.4) remains finite (recall (4.2)); it essentially corresponds to the length squared of a rigid, nondynamical stretched string, whose length
X = p corresponds to the short-distance cutoff introduced by the noncommutativity parameter . The size of the Feynman diagram cannot shrink below the length scale
X set by the stretched string, and this fact reflects the inherent nonlocality of a noncommutative theory.
As seen from (4.4), the way a stretched string contribution enters into the amplitudes
is formally similar to the (s 1/t modular transformed) contribution from the winding
modes of closed strings (see also Refs. [3133]). Comparing (4.4) with its two-loop
counterparts (3.25), we observe that the stretched string interpretation of Ref. [17] naturally
carries over to the multi-loop amplitudes, which result from the nonplanar vertex insertions
on a planar vacuum world sheet. In fact, more subtle types of amplitudes are those resulting
from a nonplanar vacuum world sheet, which necessarily has a positive genus. For these
kinds of amplitudes, there are integrations over the momenta appearing in (p p). As
a
result, the stretching length p can be larger or smaller than the string length 0
depending on the value of the loop momentum p . In contrast, the winding (closed) strings
have a fixed spacetime size. Taking the decoupling limit in this case necessarily involves
more careful analysis of the competition between the two length scales.
183
Acknowledgements
Y.K. would like to thank the high energy theory group of Princeton University for the
hospitality during his visit. We are grateful to C.-S. Chu for valuable discussions.
Appendix A. A brief review of Riemann surfaces

For a genus g Riemann surface g , we choose 2g linearly independent cycles a , b
( = 1, . . . , g) such that the intersection pairings satisfy
(a , a ) = (b , b ) = 0,
(a , b ) = (b , a ) = .
(A.1)
Any such basis is called canonical. We can also find g linearly independent holomorphic
called Abelian differentials. We
closed one-forms and their complex conjugates S
normalize the s along the a-cycles, then the periods over the b-cycles give the period
matrix;
Z
Z
= ,
= .
(A.2)
a
The period matrix is symmetric and its imaginary part is positive definite.
Consider a mapping between two canonical bases of the same Riemann surface,

0
a
D C
a
a
=M
=
,
(A.3)
0
b
b
B A
b
where M is a 2g 2g matrix composed of the g g blocks A, B, C, D. To preserve the
intersection pairing (A.1), the matrix M must satisfy

0 1
T
J=
,
(A.4)
MJ M = J,
1 0
that is, M Sp(2g, Z). Normalizing the Abelian form in the new basis, we find a relation
between the period matrices in the two bases:
0 = (A + B)(C + D)1 .
(A.5)
R
The integral of the Abelian differentials along a path on the Riemann surface =
naturally introduces a lattice in Cg . The Riemann theta functions are defined on Cg to be
the sum of Gaussian functions over the lattice,

X

1
(A.6)
exp 2i (n + ) (n + ) + (n + )(z + ) ,
(z| ) =
2
g
nZ
where 2, 2 Zg and z Cg . It follows from the definition that

(z + m1 + m2 | )

1
= exp 2i m2 m2 m2 z + m1 m2 (z| ),
2

(z| ) = exp[4i] (z| ).
(A.7)
(A.8)
184
Using the second relation, one can show that among the 22g theta functions, 2g1 (2g + 1)
are and 2g1 (2g 1) are odd.
The prime form is a (1/2, 1/2) form defined on g g by
R z

w |
p
.
(A.9)
E(z, w| ) =
(0| ) (z) (0| ) (w)
Here z and w are coordinates on g and the theta function can be any
of the
one
odd theta functions. Note that when z approaches w, E(z, w) (z w)/ dz dw. By
slight abuse of notation, we sometimes write E(z, w) in place of E(z, w) dz dw. The
transformation rule for the prime form follows from that of the theta function; for a fixed
value of w, the prime form remains unchanged when z moves around an a-cycle, while it
changes by

Zz

1
(A.10)
E bk (z), w = exp 2i kk + k E(z, w),
2
w
when z moves around a b-cycle.

The world sheet of an open string theory is a Riemann surface with boundary. An
efficient way to describe such a surface is to begin with a surface without boundary and
folding it by an involution. The involution I is an orientation-reversing diffeomorphism
from onto itself. The set of fixed points of I becomes the boundary of the resulting
surface.
Clearly, the involution preserves the intersection pairing, but changes its sign. Therefore,

0
a
H G
a
a
(A.11)
=
I
=
I J I T = J.
b0
b
F E
b
If the complex structure on the covering space is compatible with the involution,
holomorphic differential forms are mapped to antiholomorphic ones and vice versa. The
compatibility condition gives a constraint on the period matrix. Note that any integral of a
closed form over a homology cycle is invariant the involution. In particular,
Z
Z
I ( ) = ,
I ( ) = .
(A.12)
I (a )
I (b )
It follows that
= (E + F )(G + H )1 .
(A.13)
Appendix B. A brief review of Schottky representation

B.1. Generality
This subsection is based on Appendix A of [34]. Via stereographic projection, a two
sphere can be represented as the complex plane with a point at infinity added (S 2 =
C {}). One may attach a handle to the sphere by removing a pair of discs with equal
185
radii from C{} and identifying the two boundaries with opposite orientation. Repeating
this procedure g times, one obtains a genus g Riemann surface.
The Schottky representation realizes this idea quantitatively. One begins with g
independent projective transformations P SL(2, C), which act on C {} in the usual
way. The Schottky group Gg is the group generated by the P s. For our purposes, it is
convenient to specify the generators by their fixed points , and multipliers k defined
implicitly by
z
P (z)
= k
.
(B.1)
P (z)
z
The two discs D , D0 associated to each P are defined by

1 1/2
dP 1/2
dP
0

6 1,
D :
6 1.
(B.2)
D :

dz
dz
Let the circles C and C0 be the boundary of the discs. It is straightforward to show that
the radii R , R0 and the centers J , J0 of the circles are given by
p
| |
k
k
,
J =
,
J0 =
.
(B.3)
R = R0 = |k |
|1 k |
1 k
1 k
It can be shown that P maps D onto (C {} D0 ) and similarly for P1 . A bit of
thought shows that the fundamental region of the Schottky group is precisely the region
exterior to all the discs, or the Riemann surface we had in mind
g
[

D D0 .
(B.4)
g = C {}
=1
A Schottky representation has a preferred choice of canonical basis; the a cycle

corresponds to the circle C or C0 , while the b cycle corresponds to a path from a point
on C to its image by P on C0 .
The Abelian differentials and the prime form in a Schottky representation are given by

X
1
dz
1
,
(B.5)
=
z Ta ( ) z Ta ( ) 2i
a
z w Y z Ta (w) w Ta (z)
,
(B.6)
E(z, w) =
dz dw a z Ta (z) w Ta (w)
where the summation index runs over all the elements {Ta } of the Schottky group except
for the elements having P as the right-most factor, and the product index runs over all
elements except for the identity (furthermore, Ta and Ta1 are counted only once).
The Abelian differentials apparently have poles at Ta ( ) and Ta ( ). All the poles
in fact lie inside the discs, and therefore are holomorphic in the entire Riemann
surface g . Next, when one integrates along an a cycle, or the circle C , each pole in
the sum (B.5) contribute 1. It is easy to show that they cancel pair-wise except when Ta
is the identity element, so that the normalization condition (A.2) is satisfied.
Although the expression for the prime form given in (B.6) look quite different from
its definition (A.9), they can be shown to have the same analytic and periodic properties,
hence they should be equal.
186
B.2. The (03) surface

Clearly, the g = 0, b = 3 surface is obtained by folding the g = 2, b = 0 surface. The
Schottky representation of the (03) surface is obtained from that of the (20) surface in the
following way. First, place the centers of the circles along the real axis such that C1 is
adjacent to C10 and C2 to C20 . Then take the involution to be the complex conjugation on
C {}.
We may use the SL(2, R) invariance of the upper half plane to fix three of the six
parameters that define the generators of the Schottky group. Following [21], we choose
2 = 0, 2 and 1 = 1. Without loss of generality, we can also let 1 move between 0
and 1. Using (B.3), we find that the radii and the centers of the circles in Fig. 2 are
p 1 1
1 k1 1
1 k1
,
J1 =
,
J10 =
,
(B.7)
R1 = R10 = k1
1 k1
1 k1
1 k1
p
1
J2 = J20 = 0.
(B.8)
R2 = R20 = k2 ,
Note that not only the fixed points but also their images under the elements of the Schottky
group lie on the real axis. This fact has three consequences. First, it follows from (B.5) that
the period matrix is purely imaginary. It is consistent with (A.13) since in the case at hand
E = H = 1, F = G = 0. Next, we see again from (B.5) that
+ c.c. = 0|z=z .
(B.9)
dz
Finally, Eq. (B.6) shows that
E(z, w) = E(z, w
S ).
(B.10)
Appendix C. World-sheet propagators

C.1. Operator method at one loop for B 6= 0
For simplicity, we consider the case B12 = B and B = 0 otherwise. The mode
expansion of the X = (X1 , X2 ) is given by
X(, ) = RX( + ) + R T X( ),
(C.1)
where the mode expansion for the right-mover and the left-mover are given by
r
0 X n in( + )
1
0
e
,
X( + ) = x + p( + /2) + i
2
2
n
n6=0
r
0 X n in( )
1
0
e
,
X( ) = x + p( + /2) + i
2
2
n
(C.2)
(C.3)
n6=0
and as in Ref. [35], we introduced the rotation matrix,

cos
sin
R=
,
tan = B.
sin cos
(C.4)
187
The effect of the B field is completely summarized by the rotation matrix, and the
commutation relations of the modes in (C.2), (C.3) are exactly the standard ones. The
variables (, ) are related to the (x, y) in Section 3.2 by 2(x, y) = (, ). The /2
shifts in the linear terms of (C.2), (C.3) are to ensure that the same magnitude of
noncommutativity is measured at the two boundaries.
Given the mode expansion, the computation of scattering amplitudes in the operator
method is straightforward. When all the external particles are open string states, the details
are given in Ref. [36]. For closed string insertions, the only subtlety is that when one
writes down vertex operators, one should take the normal ordering for the left-mover and
the right-mover separately,
V (, ) = :eikRX( + ): : eikR
T X( )
:.
(C.5)
C.2. Analysis of the multi-loop bulk propagator

In this subsection, we show the validity of the world-sheet propagator given in
Section 3.2. It suffices to consider a simple case when the only nonzero component of
the B-field is B12 = B and g = , the propagators for X = X1 and Y = X2 reduce to

1 B2

G z, z 0
X(z)X z0 = G z, z0 +
2
1+B

4B 2
0
(T )1
(C.6)
x + x0 ,
+
x + x
2
1+B

2B
E(z, z 0 )
+ 4i T 1 x + x 0 y y 0
Y (z)X z0 =
log
. (C.7)
2
0
1+B
(E(z, z ))
Let us first check the periodicity of the hXXi propagator in (C.6). Note that under a periodic
shift along the b -cycles, the quadratic term in G changes by

= 4(y1 y2 ) + 2T
2(T )1
(y1 y2 ) (y1 y2 )
Zz1
= 4 Im
+ 2T .
(C.8)
z2
These two additional pieces precisely cancel the pieces coming from the transformation of
the prime form in (A.10), making G invariant. Note that the ak -cycles are no longer cycles
along which the periodicity should be required, since they are odd under the involution.
Since the real part of the period matrix is zero, the quadratic piece in hXXi remains
invariant under the bk -cycle shift.
For the hXY i propagators (C.7), the periodic shift along the b -cycle changes its
quadratic pieces as

4i T 1 x + x 0 y y 0 4i T 1 x + x 0 y y 0

+ 4i x + x 0 ,
(C.9)
188
for the period matrix is purely imaginary. The extra piece from (C.9) is precisely what
cancels the extra piece from the transformation of the prime form (A.10), making hXY i
periodic.
To check the boundary condition, it is easiest to use the Schottky representation of the
previous section. The boundary condition reads

(C.10)
X(z)X(w) B + Y (z)X(w) = 0|z=z ,
where the derivatives act on z only. Using (B.10), one can easily show that the terms
involving prime form satisfy (C.10) among themselves. To verifyR that the quadratic pieces
z
themselves satisfy the boundary condition, note that for (z) = p = x(z) + iy(z),

1
1
+ c.c. = 0,
(C.11)
t x = n y = ( + c.c.) =
2
2 dz
(recall (B.9)) and
t y =
1
( c.c.) = n x.
2i
(C.12)
References
[1] A. Connes, Noncommutative Geometry, Academic Press, 1994;
A. Connes, M. Rieffel, in: Operator Algebras and Mathematical Physics, Iowa City, Iowa, 1985,
Contemp. Math. Oper. Alg. Math. Phys., Vol. 62, AMS, 1987, p. 237;
A. Connes, M.R. Douglas, A. Schwarz, J. High Energy Phys. 9802 (1998) 003, hep-th/9711162.
[2] M.R. Douglas, C. Hull, J. High Energy Phys. 9802 (1998) 008, hep-th/9711165.
[3] M.M. Sheikh-Jabbari, Phys. Lett. B 425 (1998) 48, hep-th/9712199;
M.M. Sheikh-Jabbari, Phys. Lett. B 450 (1999) 119, hep-th/9810179.
[4] Y.-K.E. Cheung, M. Krogh, Nucl. Phys. B 528 (1998) 185, hep-th/9803031.
[5] C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219;
C.-S. Chu, P.-M. Ho, hep-th/9906192.
[6] V. Schomerus, J. High Energy Phys. 9906 (1999) 030, hep-th/9903205.
[7] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, hep-th/9803067;
J. High Energy Phys. 9901 (1999) 016, hep-th/9810072;
F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, hep-th/9906161.
[8] N. Seiberg, E. Witten, J. High Energy Phys. 9909 (1999) 032, hep-th/9908142.
[9] T. Filk, Phys. Lett. B 376 (1996) 53.
[10] S. Minwalla, M. Van Raamsdonk, N. Seiberg, hep-th/9912072.
[11] M. Van Raamsdonk, N. Seiberg, hep-th/0002186.
[12] O. Andreev, H. Dorn, hep-th/0003113.
[13] Y. Kiem, S. Lee, hep-th/0003145, Nucl. Phys. B, in press.
[14] A. Bilal, C.-S. Chu, R. Russo, hep-th/0003180.
[15] J. Gomis, M. Kleban, T. Mehen, M. Rangamani, S. Shenker, hep-th/0003215.
[16] A. Rajaraman, M. Rozali, J. High Energy Phys. 0004 (2000) 033, hep-th/0003227.
[17] H. Liu, J. Michelson, hep-th/0004013.
[18] S. Chaudhuri, E.G. Novak, hep-th/0006014.
[19] Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
[20] K. Roland, H.-T. Sato, Nucl. Phys. B 515 (1998) 488;
K. Roland, H.-T. Sato, Nucl. Phys. B 480 (1996) 99.
189
[21] A. Frizzo, L. Magnea, R. Russo, Nucl. Phys. B 579 (2000) 379, hep-th/9912183.
[22] E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 163 (1985) 123;
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 288 (1987) 525;
A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
[23] S.K. Blau, M. Clements, S. Della Pietra, S. Carlip, V. Della Pietra, Nucl. Phys. 301 (1988) 285.
[24] M. Bianchi, A. Sagnotti, Phys. Lett. B 211 (1988) 407.
[25] S. Samuel, Nucl. Phys. B 341 (1990) 513.
[26] G. Moore, Phys. Lett. B 176 (1986) 369.
[27] E. Martinec, Nucl. Phys. B 281 (1987) 157.
[28] C.-S. Chu, R. Russo, S. Sciuto, hep-th/0004183.
[29] O. Andreev, Phys. Lett. B 481 (2000) 125, hep-th/0001118.
[30] I. Chepelev, R. Roiban, J. High Energy Phys. 0005 (2000) 037, hep-th/9911098.
[31] W. Fischler, E. Gorbatov, A. Kashani-Poor, S. Paban, P. Pouliot, J. Gomis, J. High Energy
Phys. 0005 (2000) 024, hep-th/0002067.
[32] W. Fischler, E. Gorbatov, A. Kashani-Poor, R. McNees, S. Paban, P. Pouliot, J. High Energy
Phys. 0006 (2000) 032, hep-th/0003216.
[33] G. Arcioni, J.L.F. Barbon, J. Gomis, M.A. Vazquez-Mozo, J. High Energy Phys. 0006 (2000)
038, hep-th/0004080.
[34] P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, S. Sciuto, Nucl. Phys. B 322 (1989)
317.
[35] S. Hyun, Y. Kiem, S. Lee, C.-Y. Lee, Nucl. Phys. B 569 (2000) 262, hep-th/9909059.
[36] M. Green, J. Schwarz, E. Witten, Superstring Theory, Vol. II, Cambridge Univ. Press,
Cambridge, 1987.

Gepner-like description of a string theory on

a noncompact singular CalabiYau manifold
Satoshi Yamaguchi
Graduate School of Human and Environmental Studies, Kyoto University, Yoshida-Nihonmatsu-cho, Sakyo-ku,
Kyoto 606-8501, Japan
Received 13 July 1999; accepted 25 October 2000
Abstract
We investigate a Gepner-like superstring model described by a combination of multiple minimal
models and an N = 2 Liouville theory. This model is thought to be equivalent to the superstring
theory on a singular noncompact CalabiYau manifold. We construct the modular invariant partition
function of this model, and confirm the validity of an appropriate GSO projection. We also calculate
the elliptic genus and Witten index of the model. We find that the elliptic genus is factorised into a
rather trivial factor and a non-trivial one, and the non-trivial one has the information on the positively
curved base manifold of the cone. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Hf; 11.25.Pm
Keywords: Gepner model; Modular invariance; CalabiYau manifold
1. Introduction
The correspondence between a string theory on a Khler manifold and an N = 2
LandauGinzburg theory is interesting and is very largely investigated [14]. But, the
most results are limited to the cases of compact CalabiYau manifolds. Recently, it is
conjectured that in the case of a noncompact CalabiYau manifold, the associated CFT
consists of an N = 2 Liouville theory and a LandauGinzburg theory [5]. They claimed
that when the CalabiYau n-fold X is written as a hypersurface F (z1 , . . . , zn+1 ) = 0 in
Cn+1 by a quasi-homogeneous polynomial F , then the string theory on X is equivalent to
the CFT on
R S 1 LG(W = F ).
Here R is a real line parametrized by with linear dilaton background and LG(W = F )
is the IR theory of the LandauGinzburg model with the superpotential F . In this case,
E-mail address: yamaguch@phys.h.kyoto-u.ac.jp (S. Yamaguchi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 8 - 6
S. Yamaguchi / Nuclear Physics B 594 (2001) 190208
191
the background charge Q of R is determined by a condition of the total central charge.

From the condition that Q is a non-zero real number, we find that the base manifold X/C
should be curved positively.
The boson with linear dilaton background is strongly coupled in the region ,
so we should introduce the Liouville potential or consider an SL(2)/U (1) KazamaSuzuki
model to avoid the strong coupling singularity [68]. But we do not care about this point
in this paper.
In [5,79], they also claim that the string theory on this singular noncompact CalabiYau
manifold X is holographic dual to the little string theory.
In the case of a compact CalabiYau manifold, the string theory is solved in the
description of Gepner model in a special point of the moduli space. We want to describe
also the string theory on the noncompact CalabiYau manifold X by the Gepner-like
solvable model. If we can do it successfully, it will be possible to analyze more deeply
a noncompact CalabiYau manifold and the little string theory.
In [10], they treat the string theories with ADE simple singularities. They construct the
modular invariant partition functions, and show the consistency of these string theories.
In this paper, we consider more general cases, in which the LandauGinzburg part is
described by a direct product of a number of minimal models. A typical example of ours
is the CalabiYau n-fold X described in the form
N
n+1
=0
z1N1 + z2N2 + + zn+1
in Cn+1 .
We construct the modular invariant partition functions and show the string theory actually
exists consistently in these cases. We also calculate the elliptic genus, and find that it is
factorised into two parts a rather trivial one and a rather non-trivial one. We analyze
the non-trivial one in detail, and find that it has the information on the cohomology of the
positively curved base manifold X/C except the elements generated by cup products of
a Khler form.
The organization of this paper is as follows. In the next section, we explain the setup and
review the correspondence between a noncompact CalabiYau manifold and an N = 2
Liouville theory LandauGinzburg theory. In Section 3, we construct the modular
invariant partition function. In Section 4, we calculate the elliptic genus and compare it with
the geometric property of the associated CalabiYau manifold X. In the last section, we
summarize the results and discuss the problems and prospects. In Appendix A we collect
some useful equations of theta functions and characters that we use in this paper.
2. The string theory on a noncompact singular CalabiYau manifold

We consider the string compactification to a noncompact, singular CalabiYau n-fold X.
The total target space is expressed by a direct product of a d-dimensional flat spacetime
and the manifold X
Rd1,1 X.
Here, n is related to d by the constraint on the total dimension 2n + d = 10.
(2.1)
192
For simplicity, we concentrate the case that the noncompact singular CalabiYau
manifold X is realized as the hypersurface in Cn+1 determined by the algebraic equation
with a quasi-homogeneous polynomial F
F (z1 , . . . , zn+1 ) = 0.
By the term quasi-homogeneous, we mean that the polynomial F satisfies

F r1 z1 , . . . , rn+1 zn+1 = F (z1 , . . . , zn+1 ),
(2.2)
for some exponents {rj } and for an arbitrary C .

This manifold X is singular at (z1 , . . . , zn+1 ) = (0, . . . , 0). If we consider the manifold
(X (0, . . . , 0))/C , where the action of C is (z1 , . . . , zn+1 ) (r1 z1 , . . . , rn+1 zn+1 )
with the exponents {rj } of (2.2), then we get a compact manifold. We denote this compact
manifold simply as X/C and call it the base manifold of X.
It is conjectured in [5] that the string theory on the space (2.1) is equivalent to the theory
including flat spacetime Rd1,1 , a line with the linear dilation background R , S 1 , and the
LandauGinzburg theory with a superpotential W = F
Rd1,1 R S 1 LG(W = F ).
The part (R S 1 ) has a world sheet N = 2 superconformal symmetry. Let be the
parameter of R , Y be the parameter of S 1 , and + , be the fermionic part of (R S 1 ).
The N = 2 superconformal currents are written in terms of the above fields
1
Q
1
1
T = (Y )2 ()2 2 ( + + ),
2
2
2
2
1
Q
G = (iY ) ,
2
2
J = + QiY.
(2.3)
The associated central charge of this algebra is c(=

c/3) = 1 + Q2 .
In this paper, we consider the case in which the LandauGinzburg theory with
superpotential W = F can be described by a direct product of N = 2 minimal models.
Let MG,N be the minimal model corresponding to simply laced Lie algebra G = A, D, E
with dual Coxeter number N . We consider the theory in the following:
Rd1,1 R S 1 MG1 ,N1 MGR ,NR ,
where R is the number of the minimal models. The cases with R = 1 are treated in [10]
and R = 0 in [11,12].
A typical example is the case that all Gj are A type. In this example, the quasihomogeneous polynomial is written as
NR
2
2
+ zR+1
+ + zn+1
.
F (z) = z1N1 + + zR
The background charge of R is determined by the criticality condition. To cancel the

conformal anomaly, the total central charge is to be 0. The central charge of the ghost
sector is 15, so the total central charge of the matter sector is to be 15. The central charge
193
of the flat spacetime is 3/2 for each pair of a boson and a fermion, and that of the R S 1
is 3 + 3Q2 as mentioned above, and that of a minimal model MG,N is 3(N2)
. Therefore,
N
the criticality condition leads to the equation
X 3(Nj 2)
3d
+ 3 + 3Q2 +
= 15.
2
Nj
R
j =1
From this criticality condition, we obtain the value of Q2 as

Q2 = 4
d X Nj 2
.
2
Nj
(2.4)
By the condition Q2 > 0 for a real number Q, the right-hand side should be positive:
4
d X Nj 2
> 0.
2
Nj
(2.5)
It is equivalent to a condition that the singularity is in finite distance in the moduli space
of deformation of singular CalabiYau manifold X [5,13]. In view of the base manifold
X/C , the finite distance condition is equivalent to that X/C is positively curved.
3. Modular invariant partition function

Now, let us construct the modular invariant partition function. We take the light-cone
gauge, then the associated CFT to consider is
Rd2 R S 1 MG1 ,N1 MGR ,NR .
The toroidal partition function can be separated into 2 parts: the one ZGSO concerning to
the GSO projection and the other Z0 not concerning to it. We construct the total partition
function Z as
Z 2
d
Z0 (, )ZGSO (, ),
Z=
22
where = 1 + i2 is the moduli parameter of the torus, and d 2 /22 is the modular
invariant measure.
First, we study the rather easy part Z0 , then we investigate the rather complicated
part ZGSO .
3.1. GSO independent part of the partition function
In this subsection, we discuss the Z0 : the GSO independent part. It is completely the
same as that in [10].
The Z0 includes the contribution from the flat spacetime bosonic coordinates XI (I =
2, . . . , d 1) and the linear dilation .
194
The partition function of each flat spacetime boson is represented by the Dedekind eta
function ( ) as
1
.
2 |( )|2
The partition function of is defined as ZL = Tr q L0 cL /24 q L0 cL /24 (q = exp(2i ),

cL = 1 + 3Q2 ) in the canonical formalism. Here the trace is taken over delta function
normalizable primary fields
iQ
+ `, ` R,
2
and their excitations by oscillators. Then we obtain ZL as

Z
1 + 3Q2
1
1 2 i
p
pQ
+
dp
exp
4
ZL = Q
2
| n=1 (1 q n )|2
2
2
24
1
,
=
2 |( )|2
exp(ip),
p=
(3.1)
where the region of the integral of p is as (3.1). As a result, the partition function of is the
same as that of an ordinary boson. So we obtain Z0 as the partition function of effectively
(d 1) free bosons:

d1
1
.
Z0 =
2 |( )|2
The primary fields of (3.1) correspond to principal continuous series in terms of
the representation of SL(2). To include the other sectors is an interesting problem and
is postponed as a future work.
3.2. GSO dependent part of the partition function: d = 2, 6 cases
Now, let us proceed to the GSO dependent part ZGSO . In this subsection, we treat
d = 2, 6 cases.
This part includes (d 2) flat spacetime fermions I (I = 2, . . . , d 1), two free
fermions , Y associated to R S 1 , minimal models MG1 ,N1 , . . . , MGR ,NR , and an S 1
boson Y . We combine the d free fermions I (I = 2, . . . , d 1), and Y and construct
\ 1 is characterized by an integer
\ 1 . The Verma module of SO(d)
the affine Lie algebra SO(d)
s0 = 0, 1, 2, 3, which labels the representations of SO(d), that is scalar, spinor, vector, and
cospinor, respectively.
Let us turn to Verma modules of N = 2 minimal models. The Verma module of an
N = 2 minimal model is specified with three indices (`, m, s), which satisfy the following
conditions [2]
` = 0, . . . , N 2,
s = 0, 1, 2, 3,
m = 0, . . . , 2N 1,
` + m + s 0 mod 2.
(3.2)
We denote m`,s (, z) as the character of the Verma module labeled by the set (`, m, s).
Some properties of this character is collected in Appendix A.
195
The Verma module of the whole GSO dependent parts is specified by the index s0 of
\ 1 representation, the indices `j , mj , sj (j = 1, . . . , R) of the minimal models, and
SO(d)
the S 1 momentum p. We combine these indices except p into two vectors ,
:= (`1 , . . . , `R ),
:= (s0 ; s1 , . . . , sR ; m1, . . . , mR ).
We shall introduce the inner product between and 0 as in [2]
0 :=
R
R
0
0
d s0 s00 X sj sj X mj mj
+
.
2 4
4
2Nj
j =1
j =1
Also it is convenient to introduce special vectors 0 , j (j = 1, . . . , R)

0 := (1; 1, . . ., 1; 1, . . . , 1),
j := (2; 0, . . ., 0, 2 , 0, . . . 0; 0, . . ., 0).
Sj
Here the 0 is the vector with all components 1, and j is the vector with s0 and sj
components 2 and the others zero.
With these notations, the criticality condition (2.4) can be written in a rather simple form
as
Q2 = 4(1 + 0 0 ).
(3.3)
When we define an integer K := lcm(2, Nj ), KQ2 is shown to be an even integer because

of the Eqs. (2.4) and (3.3). Therefore, it is convenient to define an integer J by the equation
J := 2K(1 + 0 0 )
(= KQ2 /2).
(3.4)
In terms of J , the finite distance condition (2.5) can be expressed as J > 0.

Now, let us consider the character of the Verma module (, , p),
q
( )
1 2
2p
( )
\ 1 character
where ( ) is the product of characters of the minimal models and the SO(d)
s0 ( ) of the s0 representation
( ) := s0 ( )m`11,s1 ( ) m`RR,sR ( ).
1
In this character, ( ) has good modular properties, but q 2 p /( ) has bad ones. So,
we will sum up the characters with respect to certain values of p and make the modular
properties good [10,11].
Let us consider the GSO projection. By the GSO projection, we pick up the states with
odd integral U (1) charges of the N = 2 superconformal symmetry. The U (1) charge of
the states in the above Verma module is expressed as
196
20 + pQ =
d s0 X sj X mj
+
+ pQ.
2 2
2
Nj
j
From the condition that this U (1) charge must be an odd integer (2u + 1) with u Z, the
S 1 momentum p is written as
p(u) =
1
(2u + 1 20 ).
Q
If we sum up the characters for all u Z, we obtain the theta function with a fractional
level [10], which does not have good modular properties. So we perform the following
trick.
Let us write u = J v + w with integers v, w and sum up the characters for v Z. Then
the sum leads to the following theta function
X
q 2 p(u=J v+w) = 2K0 +K(2w+1),KJ ( ).
(3.5)
vZ
Note that 2K0 + K(2w + 1) is an integer, and the above theta function has good
modular properties.
Now, including oscillator modes and other sectors, we can define the building blocks
f ( ) by
f ( ) := ( )M,KJ ( )/( ),
:= (, M), M Z2KJ .
We should use only the building blocks f with the conditions
M = 2K0 + K(2w + 1)
for w Z,
s0 s1 sR mod 2.
(3.6)
(3.7)
The condition (3.6) comes from the formula of (3.5), and the condition (3.7) implies that
the boundary condition of the fermionic currents are the same in all the sub-theories, i.e.,
they must be all in the NS sector, or all in the R sector.
The modular invariant partition function can be systematically obtained by the beta
method [2].
The inner product between of two vectors , 0 is defined as

0 := 0

MM 0
.
2KJ
We also extend the vectors 0 , j to 0 , j as
0 := (0 , J ),
j := (j , 0),
and evaluate the inner products of these 0 , j vectors
J2
= 1,
0 0 = 0 0
2KJ

d
1
1 .
j 0 =
2
2
j j = j j =
197
d
1,
2
(3.8)
Note that 0 0 is an odd integer, j j are even integers (recall that we consider the
cases d = 2, 6), and j 0 are integers. Using these special vectors, the conditions (3.6)
and (3.7) are written in a simple form

2 0 2Z + 1,
j Z.
(3.9)
We call this condition the beta condition.

Using these notations, and the modular transformation laws of theta functions and N = 2
characters written in the Appendix A, we can calculate the modular transformation laws of
f as

X

`j (`j + 1) 1
1 X Nj 2 d
+ 1) = e

+ + 1 f ( ),
4Nj
2 24
Nj
2
j
j
!
even
Y 1
X
1
0
p
A0
e[ 0 ]f0 ( ),
f (1/ ) =
8Nj
8KJ
j
0 ,0
f (
where the sums of the 0 , 0 are taken only for the range (3.2) and for M = 0, . . . , 2KJ 1.
Especially we must impose the condition `j + mj + sj 0 mod 2 for each minimal model.
\ N 2 S matrices A(Nj0) :
A0 is the products of the SU(2)
`j `j
A0 =
Y
j
(Nj )
0
j `j
A`
(`j + 1)(`0j + 1)
2
sin
,
Nj
Nj
and we use here and the rest of this paper the notation e[x] = exp(2ix).
Let us note that if a vector satisfies the beta condition (3.9), the vector +b0 0 +
P
j bj j for b0 , bj Z (j = 1, . . . , R) also satisfies the beta condition by virtue
of (3.8). Using this fact, we define the function F for (, ) which satisfies the beta
conditions (3.9) as a sum of
f +b +P b s
0 0
j j j
F ( ) =
X
b0 ,bj
as
(1)s0 +b0 f +b0 0 +P
j
j bj
( ),
where the sum is taken for b0 Z2K and bj Z2 . The sign (1)s0 +b0 is (1) for the
Ramond sector.
These functions have very good modular properties. Especially by S transformation, the
functions are mixed among those which satisfy the beta condition:
198
F ( + 1) = e

X

`j (`j + 1) 1
1 X Nj 2 d

+ + 1 F ( ),
4Nj
2 24
Nj
2
F (1/ ) =
even,beta
X
A0
0 ,0
Y
1
p
8Nj

1
0
0
e 0 (1)s0+s0 F0 ( ),
8KJ
where the sums of 0 , 0 is taken for restricted subclass that satisfies the conditions (3.2)
and the beta condition (3.9).

With this function F , we obtain the modular invariant ZGSO as
ZGSO (, ) =
1
R
4 2K
even,beta
X
L F ( )F ( ),
,,
where L =
Q
j
(Gj ,Nj 2)
j `j
is the product of Gj = A, D, E type modular invariants of
L`
\ N 2 [1416].
SU(2)
j
We can check the modular invariance of the above partition function.
We expect from spacetime supersymmetry that the partition function vanishes, or
equivalently F ( ) = 0. It is a future work to check that it actually vanishes.
Here, we find a solution, but it may not be the only solution, and there can be some
variety of modular invariant partition function. Actually, for R = 1 case, there are many
other solutions associated with the other modular invariants of the theta system [10].
3.3. GSO dependent part of the partition function: d = 4 case
In this subsection, we comment on the d = 4 case. To construct the modular invariant
partition function in the d = 4 case, we combine the four fermions to construct the affine
\ 1 and label the the Verma module by indices s1 and s0 . Then,
\ 1 SO(2)
currents SO(2)
the modular invariant partition function can be constructed in almost the same way as the
d = 2, 6 cases.
First we define the vectors s and the inner product between them as
:= (s1 , s0 ; s1 , . . . , sR ; m1 , . . . , mR ; M),
0 :=

0
s1 s1
0
0
s0 s00 X sj sj X mj mj
MM 0
+
.
4
4
2Nj
2KJ
j
It is convenient to introduce special vectors 0 , j and 1
0 = (1, 1; 1, . . ., 1; 1, . . ., 1; M),
j = (0, 2; 0, . . ., 0, 2 , 0, . . . , 0; 0, . . . , 0; 0) (j = 1, . . . , R),
Sj
1 = (2, 2; 0, . . ., 0; 0, . . ., 0; 0).
199
Using these vectors, we can construct the building blocks f ( ) as
f ( ) := s1 ( )s0 ( )m`1 ,s1 ( ) m`R ,sR ( )M,KJ ( )/( ),
\ 1 characters. Then the GSO conditions and the

where s1 ( ) and s0 ( ) are SO(2)
condition of fermionic sectors are
2 0 2Z + 1,
j Z,
1 Z,
(3.10)
and we can construct the modular invariant partition function by the beta method in this
case. Next we introduce the function F ( ) as
F ( ) =
X
b0 Z2K ,bj Z2 ,b1 Z2
(1)b0 +s0 f +b0 0 +P
bj j +b1 1 ( ),
then we obtain the GSO dependent part of the modular invariant partition function ZGSO
ZGSO (, ) =
1
4R 4K
even,beta
X
L F ( )F ( ).
,,
We can check the modular invariance of the above partition function.
4. Elliptic genus
In this section, we calculate the elliptic genus of the theory [17]. The definition of the
elliptic genus is
Z(, , z) := TrRR (1)F q L0 c/24 q L0 c/24 y J0 ,

where the trace is taken for the RR states, and y = exp(2iz). This elliptic genus has the
following modular properties:
Z( + 1, + 1, z) = Z(, , z) = Z(, , z),
2
c z
Z(, , z).
Z(1/, 1/ , z/ ) = e
2
(4.1)
Here, we omit the contribution from the flat spacetime and consider only the internal
part describing the CalabiYau n-fold X. We calculate its elliptic genus and the Witten
index, and discuss its geometrical interpretation.
Let us consider again the criticality condition, in other words the CalabiYau
condition of X. Here the total c should be n because we want the theory that describes a
CalabiYau n-fold. Therefore, the total c of the N = 2 Liouville and the minimal models
should satisfy the relations
X Nj 2
.
(4.2)
n = c = 1 + Q2 +
Nj
j
200
We introduce the following vectors with R components {mj }, and the inner product
between them as
X mj m0j
0 :=
.
:= (m1 , . . . , mR ),
2Nj
j
We also introduce the special vector 0 with all components 2

0 := (2, . . . , 2).
With these notations, the condition (4.2) becomes
Q2 = n 1 R + 0 0 .
Next we let N := lcm(Nj ), and define J as
2J
:= Q2 .
(4.3)
N
In this paper, we concentrate only the case that (n 1 R) is even, then in this case, J is
an integer. In terms of J , the finite distance condition Q2 > 0 can be written as J > 0.
Because we want a CalabiYau CFT, we have to pick up only the states with
integral U (1) charges. This condition is realized as the condition
0 + pQ Z.
(4.4)
From this, p can be written with an arbitrary integer u

1
(u 0 ).
Q
Following the same manner as in the previous section, we let u = 2J v + w and sum up for
v Z. It leads to the theta function
X 1 p2
q 2 y pQ = N(w0 ),NJ (, 2z/N).
p=
vZ
Note that N(w 0 ) is an integer and N(w0 ),NJ (, 2z/N) has good modular
properties.
Collecting these, we define the building blocks g as
g (, z) :=
(, z)
s0 ,sj =1,3
s0 P
M,NJ (, 2z/N)
(1) 2 j
( )
s0
where
:= (, M). In the sign (1)J0 = (1) 2
sj
j 2
+0 + M
N
sj
2
+0 + M
N
, the part ( s20
sj
j 2
represents contributions of ordinary U (1) charges from the indices s0 , sj , and the rest
1
0 + M
N = w = u 2J v reflects contributions from the indices mj and S momentum.
0
as
Let us define the inner product between
and
0 := 0

and the special vector
MM 0
,
2NJ
201
0 = (0 , 2J ).
We also introduce the functions Im` and I

Im` (, z) := m`,1 (, z) m`,3 (, z),
I (, z) := Im`11 (, z) Im`RR (, z).
With these notations, the building blocks g can be written as
g (, z) =
1 (, z)
M,NJ (, 2z/N)
I (, z)
(1) ,
( )
( )
where we omit the overall irrelevant phase. The condition (4.4) can be rewritten as
Z,
0
(4.5)
and again we call this condition the beta condition.

Now, we construct elliptic genus using the above building blocks g which satisfy the
condition (4.5).
+b0 0 for b0 Z also satisfies the beta
Note that if
satisfies the beta condition, then
condition, because
0 0 = 0 0
2J
= (n 1 R),
N
is an integer. 1 Here we used the definition of J (4.3). So let us define the new functions
G` as follows:
G` (, z) =
X
b0 ZN
g +b0 0 (, z),
where
satisfies the beta condition (4.5). Then, from the modular properties of g
g ( + 1, z)

X
`j (`j + 1) 1
1 X Nj 2
R+1

+
+ 2 g (, z),
=e
4Nj
2
8
24
Nj
j
g (1/, z/ )
even
0 (
0 ) 0

n z2 X
1
1

A0 Q p
0 (1)
g 0 (, z),
e

= (i) e
2
2N
2NJ
j
0
0
j
,

G` have very good modular properties:
1 Actually, it is an even integer. Remember that we concentrate the case in which (n 1 R) is an even integer.
202
G ( + 1, z)

X
`j (`j + 1) 1
R+1
1 X Nj 2

+ 2 G (, z),
+
=e
4Nj
2
8
24
Nj
j
G (1/, z/ )
n z2
= (i) e
2
R
even,beta
X
1
A0 Q p
0
0 ,
0 (
0 )

1
0

e

0 (1)
G 0 (, z).
2Nj 2NJ
Using these functions, we obtain the elliptic genus in the following form:
1
1
Z(, , z) = R
2 N 2 |( )|2
even,beta
X
( , 0).
L G (, z)G
,,
\ modular invariants, and the factor 1/2 |( )|2 is

Here L is the product of SU(2)
contribution of . We can check that the above elliptic genus has the right modular
properties (4.1) with c = n.
Actually, this elliptic genus is 0 because it has an overall factor 1 ( , 0) = 0.
4.1. Hodge number and Witten index
To get some nontrivial information from the above elliptic genus, we factor out the trivial
b by the equations
parts and define Z
Z(, , z) =
1 (, z)1 (, 0) b
Z(, , z),
2 |( )|6
b , z) =
Z(,
1
2R N
b (, z) =
G
X
b0 ZN
even,beta
X
b (, z)G
b ( , 0),
L G
,,
g +b0 0 (, z),
g (, z) = I (, z)M,NJ (, 2z/N)(1)
Now, we take the limit i and consider the ground states. In this limit, M,NJ
becomes
mod NJ
,
M,NJ (i, z) = M
b can be evaluated as
so, the Gs
(
I M (i, z) (M 0 mod 2J ),
b
G = + 2J 0
0
(others).
b is expressed in the formula
Then, Z
b=
lim Z
1
R
2 N
even,beta
X
mod 2J
M
L I+
(i, z)I+
(i, 0).
M
M
2J 0
,,
203
2J 0
M
0 by , then we can perform the sum of M Z2NJ . Moreover,
When we replace the + 2J
from the fact
`
mod 2N
j
y
Imjj (i, z) = mj `j 1
`+1 1
N 2
j
mj +`j +1
y
mod 2N
`+1 1
N +2
it can be seen that the even condition `j + mj 1 mod 2 is included in this factor. We
b in this limit
obtain a formula of the Z
#
"
beta Y X
X
1
`j
(Nj ) `j
b=
L` ` Imj (i, z)Imj (i, 0) .
lim Z
j j
i
2R
`j ,`j
So far, we treat rather general cases, but from now, we take an example and restrict
ourselves to calculations in the example. We consider the example which satisfies all the
following conditions:
All minimal models are A type. So, L = ;
R = n + 1.
N1 = N2 = = NR = N .
In other words, this example is the case where the associated CalabiYau manifold X is
the hypersurface of the form
N
= 0 in Cn+1 .
z1N + z2N + + zn+1
(4.6)
We can write the finite distance condition as N < n+1, which is equivalent to the condition
that first Chern number of X/C is positive. In this case, nontrivial factor of the elliptic
genus can be calculated as
"
beta Y X
X
`+1 1
`+1 1
1
mod 2Nj
mod 2Nj
N +2
b=
N 2
y
y
lim Z
m
`
1
m
+`
+1
j
j
j
j
i
2R
j
`j
#

mod 2N
mod 2Nj
j
mj `j 1
mj +`j +1
#
"
n+1 beta
`+1
y 2 X Y X mod 2Nj `+1
mod 2Nj
+1
.
mj `j 1 y N + mj +`j +1 y N
=
2R
`j
When we put mj = aj + Nbj (bj = 0, 1; aj = 0, 1, . . . , N 1), then beta condition

becomes
X
aj 0 mod N.
j
b in this limit
We obtain the Z
b=
lim Z
n
X
p=1
hp y p
n+1
2
204
Table 1
The values of the coefficients hp for n = 3, 4, 5, N = 3, . . . , n + 1. We include the N = n + 1 case
in the table, despite it is suppressed by the finite distance condition
n=3
n=4
n=5
N \p
N \p
N \p
3
4
0
1
6
19
0
1
3
4
5
0
0
1
5
30
101
5
30
101
0
0
1
3
4
5
6
0
0
0
1
1
21
120
426
20
141
580
1751
1
21
120
426
0
0
0
1
hp :=
1=
aj =1,...,N1,
P
j aj =pN
p
X
i=0

n + 1 (p i)(N 1) + p 1
(1)
.
i
n
i
We show several examples of hp for lower n, N in Table 1. These coefficients hp seem to

coincide with the middle dimensional Hodge numbers of X/C except for the cohomology
elements generated by cup products of a Khler form of X/C , on which we mention
below.
In our model, the Witten index can also be calculated as

n
X
(1 N)n+1 1
b=
hp = (1)n+1 1 +
Z
lim
i,z0
N
p=1

(1 N)n+1 1
n .
= (1)n+1 n + 1 +
N
On the other hand, the Euler number of the (n 1)-dimensional manifold X/C is
expressed in the formula
X/C = n + 1 +
(1 N)n+1 1
.
N
The Witten index of the CFT almost coincide with the Euler number X/C of X/C .
One of the differences of the two is the sign (1)n+1 , but this is not relevant. Except this
difference of the sign, the Witten index is smaller by n than the Euler number of X/C
in our case. This difference may correspond to the cohomology elements generated by
cup products of the Khler form of X/C . On X, these cohomology elements are absent
because they appear when we take the quotient of X by C . This seems the reason why
the Witten index is smaller by n than the Euler number of X/C .
5. Conclusion and discussion

We construct the toroidal partition function of the string theory described by the
combination of an N = 2 Liouville theory and multiple N = 2 minimal models. This
205
partition function is actually modular invariant, and we can conclude that the theory exists
consistently.
This string theory is thought to describe the string on a noncompact singular CalabiYau
manifold. To check this proposition, we also calculate the elliptic genus of this theory and
the Witten index.
The Euler number defined from nontrivial factor of the Witten index in the CFT seems to
be that of the non-vanishing elements of the cohomology. In the case of a singular manifold,
there are vanishing elements of the cohomology, which are supported on the singular point
and reflect the structure of singularities.
The fact that the vanishing elements of the cohomology cannot be seen, is probably
related to our method of construction in which we include only the states in principal
continuous series of the SL(2) theory. If we can include some discrete series (but it is
difficult [18]), the structure of the singularities might be seen in the CFT.
Another reason is that we treat the N = 2 Liouville theory as free field theory in this
paper. It is mentioned in [19] that if we treat appropriately the effect of Liouville potential,
the Witten index does not vanish and gives the Euler number including the vanishing
elements of the cohomology. In this paper, since we treat the case of = 0 and not
deformed singularity, the vanishing elements of the cohomology actually vanish and it
is consistent with the vanishing Witten index.
We may not be able to use our result to analyze the structure of the singularities, but we
can use it to analyze the string on the positively curved manifold X/C . Especially it is
interesting to analyze the D-branes wrapped on infinite cycle in this noncompact Calabi
Yau manifold through the recipes of boundary states in the CFT [1922] as the case of the
ordinary Gepner models [2326].
Acknowledgements
I would like to thank Tsuneo Uematsu and Katsuyuki Sugiyama for useful discussions
and encouragements. I would also like to thank to the organizers of Summer Institute 2000
at Yamanashi, Japan, August 721, 2000, and the participants of it, especially Michihiro
Naka, Masatoshi Nozaki, Yuji Sato and Yuji Sugawara for useful discussions.
This work is supported in part by JSPS Research Fellowships for Young Scientists.
Appendix A. Theta functions and characters

We use the following notations in this paper.
e[x] := exp(2ix),

1 (m 0 mod N),
mod N
:=
m
0 (others),
where m and N are integers. The useful formula is
206
X jm
mod N
e
,
= Nm
N
j ZN
where m and N are integers.

The SU(2) classical theta functions are defined as
X
m 2
m
q k (n+ 2k ) y k (n+ 2k ) ,
m,k (, z) =
nZ
where q := e[ ], y := e[z]. The Jacobis theta functions are also defined as

X
X
1 2
1
1 2
1
(1)n q (n 2 ) y (n 2 ) ,
2 (, z) :=
q (n 2 ) y (n 2 ) ,
1 (, z) := i
nZ
3 (, z) :=
nZ
X
2
4 (, z) :=
(1)n q n y n .
n2 n
q y ,
nZ
nZ
Two kinds of theta functions are related by equations

20,2 = 3 + 4 ,
21,2 = 2 + i1,
22,2 = 3 4 ,
23,2 = 2 i1.
The Dedekind function is represented as an infinite product

1
( ) := q 24
(1 q n ).
n=1
\ 1 for d/2 2Z + 1 can be expressed as

The character s (, z), s = 0, 1, 2, 3, of SO(d)
3 (, z)d/2 + 3 (, z)d/2
,
2( )d/2
2 (, z)d/2 + (i1 (, z))d/2
,
1 (, z) =
2( )d/2
3 (, z)d/2 3 (, z)d/2
,
2 (, z) =
2( )d/2
2 (, z)d/2 (i1 (, z))d/2
.
3 (, z) =
2( )d/2
0 (, z) =
Let us denote the character of a Verma module (`, m, s) in the level (N 2) minimal
model as m`,s (, z). This character satisfies equivalence relations
`,s
N2`,s+2
= m`,s+4 = m+N
.
m`,s = m+2N
The explicit form of this character is written in [2].

We collect the modular properties of these functions. Under the T transformations they
behave as
2
m
m,k (, z),
m,k ( + 1, z) = e
4k
207

1
1
2 ( + 1, z) = e
1 (, z),
2 (, z),
8
8
4 ( + 1, z) = 3 (, z),
3 ( + 1, z) = 4 (, z),

1
( ),
( + 1) = e
24
2

d
s
s (, z),
s ( + 1, z) = e
8
48

s 2 N 2 `,s
`(` + 2) m2
`,s
m (, z)
m ( + 1, z) = e
4N
4N
8
8N
1 ( + 1, z) = e
and for S transformations they have modular properties

2 X

mm0
k z
1
e
m0 ,k (, z),
m,k (1/, z/ ) = i e
4
2k
2k
m0 Z2k
2
1z
1 (, z),
1 (1/, z/ ) = i i e
2
2
1z
4 (, z),
2 (1/, z/ ) = i e
2
2
1z
3 (, z),
3 (1/, z/ ) = i e
2
2
1z
2 (, z),
4 (1/, z/ ) = i e
2
(1/ ) = i ( ),

3

d ss 0
d z2 X 1
e
s 0 (, z),
s (1/, z/ ) = e
4 0 2
2 4
s =0

even
1 X
N 2 z2
ss 0 mm0 `0 ,s 0
`,s
0
+
m0 (, z),
A`` e
m (1/, z/ ) = e
2N
4
2N
8N `,m,s
r

(` + 1)(`0 + 1)
2
sin
,
A``0 =
N
N
P
where the sum even
`,m,s means that ` + m + s 0 mod 2 for (`, m, s).
We use the notation f ( ) for a function f (, z) of , z with substituting z = 0
f ( ) := f (, z = 0).
References
[1] D. Gepner, Exactly solvable string compactifications on manifolds of SU(N) holonomy, Phys.
Lett. B 199 (1987) 380.
[2] D. Gepner, Spacetime supersymmetry in compactified string theory and superconformal
models, Nucl. Phys. B 296 (1988) 757.
208
[3] C. Vafa, String vacua and orbifoldized LG models, Mod. Phys. Lett. A 4 (1989) 1169.
[4] K. Intriligator, C. Vafa, LandauGinzburg orbifolds, Nucl. Phys. B 339 (1990) 95.
[5] A. Giveon, D. Kutasov, O. Pelc, Holography for non-critical superstrings, JHEP 9910 (1999)
035, hep-th/9907178.
[6] H. Ooguri, C. Vafa, Two-dimensional black hole and singularities of CY manifolds, Nucl. Phys.
B 463 (1996) 55, hep-th/9511164.
[7] A. Giveon, D. Kutasov, Little string theory in a double scaling limit, JHEP 9910 (1999) 034,
hep-th/9909110.
[8] A. Giveon, D. Kutasov, Comments on double scaled little string theory, JHEP 0001 (2000) 023,
hep-th/9911039.
[9] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography,
JHEP 9810 (1998) 004, hep-th/9808149.
[10] T. Eguchi, Y. Sugawara, Modular invariance in superstring on CalabiYau n-fold with ADE
singularity, Nucl. Phys. B 577 (2000) 3, hep-th/0002100.
[11] S. Mizoguchi, Modular invariant critical superstrings on four-dimensional Minkowski space
two-dimensional black hole, JHEP 0004 (2000) 014, hep-th/0003053.
[12] D. Kutasov, Some properties of (non)critical strings, hep-th/9110041.
[13] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69,
hep-th/9906070.
[14] A. Cappelli, C. Itzykson, J.-B. Zuber, Modular invariant partition function in two dimensions,
Nucl. Phys. B 280 (1987) 445.
(1)
[15] A. Cappelli, C. Itzykson, J.-B. Zuber, The ADE classification of minimal and A1 conformal
invariant theories, Commun. Math. Phys. 113 (1987) 1.
[16] A. Kato, Classification of modular invariant partition functions in two dimensions, Mod. Phys.
Lett. A 2 (1987) 585.
[17] T. Kawai, Y. Yamada, S.-K. Yang, Elliptic genera and N = 2 superconformal field theory, Nucl.
Phys. B 414 (1994) 191, hep-th/9306096.
[18] A. Kato, Y. Satoh, Modular invariance of string theory on AdS3 , Phys. Lett. B 486 (2000) 306,
hep-th/0001063.
[19] T. Eguchi, Y. Sugawara, D-branes in singular CalabiYau n-fold and N = 2 Liouville theory,
hep-th/0011148.
[20] S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici, G. Sarkissian, D-branes in the background of
NS fivebranes, JHEP 0008 (2000) 046, hep-th/0005052.
[21] W. Lerche, On a boundary CFT description of nonperturbative N = 2 YangMills theory, hepth/0006100.
[22] W. Lerche, A. Lutken, C. Schweigert, D-branes on ALE spaces and the ADE classification of
conformal field theories, hep-th/0006247.
[23] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hepth/9712186.
[24] I. Brunner, M.R. Douglas, A. Lawrence, C. Romelsberger, D-branes on the quintic, JHEP 0008
(2000) 015, hep-th/9906200.
[25] M. Naka, M. Nozaki, Boundary states in Gepner models, JHEP 0005 (2000) 027, hepth/0001037.
[26] K. Sugiyama, Comments on central charge of topological sigma model with CalabiYau target
space, hep-th/0003166.

N = 2 supersymmetric RG flows and

the IIB dilaton
Krzysztof Pilch, Nicholas P. Warner
Department of Physics and Astronomy and CIT-USC Center for Theoretical Physics, University of Southern
California, Los Angeles, CA 90089-0484, USA
Abstract
We show that there is a non-trivial relationship between the dilaton of IIB supergravity, and
the coset of scalar fields in five-dimensional, gauged N = 8 supergravity. This has important
consequences for the running of the gauge coupling in massive perturbations of the AdS/CFT
correspondence. We conjecture an exact analytic expression for the ten-dimensional dilaton in
terms of five-dimensional quantities, and we test this conjecture. Specifically, we construct a
family of solutions to IIB supergravity that preserve half of the supersymmetries, and are lifts
of supersymmetric flows in five-dimensional, gauged N = 8 supergravity. Via the AdS/CFT
correspondence these flows correspond to softly broken N = 4, large N YangMills theory on part
of the Coulomb branch of N = 2 supersymmetric YangMills. Our solutions involve non-trivial
backgrounds for all the tensor gauge fields as well as for the dilaton and axion. 2001 Elsevier
Science B.V. All rights reserved.
1. Introduction
It has been evident over the last year that five-dimensional supergravity theories are
very powerful tools in the study of the AdS/CFT correspondence [13]. In particular,
gauged N = 8 supergravity in five dimensions [4,5] describes N = 4 YangMills
theory in the large N limit under perturbations that involve fermion or scalar bilinear
operators [2,3,68]. What has been less evident is exactly how the five-dimensional
solutions are lifted to ten-dimensional solutions. An example of such a lift was given in [9]
for the non-trivial, supersymmetric critical point of [8], however, as yet, no non-trivial lifts
of massive five-dimensional flows have been obtained. One of the purposes of this paper is
to give the exact ten-dimensional solution for the five-dimensional N = 2 supersymmetric
flows.
E-mail addresses: pilch@usc.edu (K. Pilch), warner@usc.edu (N.P. Warner).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 6 - 8
210
K. Pilch., N.P. Warner / Nuclear Physics B 594 (2001) 209228
The other, and more far-reaching purpose of this paper is to solve a beautiful subtlety
in consistent truncation: a subtlety that has significant consequences for the field theory
side of the AdS/CFT correspondence. Specifically, there is an SL(2, R) invariance of the
five-dimensional gauged supergravity theory, and perturbatively, the coset, SL(2, R)/SO(2)
corresponds to the ten-dimensional dilaton and axion. Combining these facts, it is a natural
conflation to assume that this is always true: i.e., that the SL(2, R) invariance in the fivedimensional theory sweeps out the ten-dimensional dilaton/axion coset. This turns out
to be false, and indeed false in a very interesting way.
The scalars of N = 8 supergravity are decribed by the coset E6(6)/USp(8). In terms of
SO(6) representations, the non-compact generators constitute the 200 10 10 1 1.
The two singlets are dual to the gauge coupling and theta-angle, while the 200 10 10
are respectively dual to the YangMills scalar and fermion bilinears:
1

Tr i j ,
Tr i j .
(1.1)
Tr XA XB AB Tr XC XC ,
6
The subgroup SL(6, R) SL(2, R) E6(6) describes the YangMills theory on the
Coulomb branch [10,11], or under purely scalar mass perturbations. In this sector the
SL(2, R) factor can indeed be identified with the ten-dimensional dilaton/axion coset.
However, for more general E6(6) matrices, i.e., when fermion masses or vevs are non-zero
it turns out that the relationship is far from trivial. Indeed, in this paper we conjecture that
the SL(2, R) matrix, S, that parametrizes the ten-dimensional IIB dilaton/axion is related
to the scalar E6(6) matrix, V, of five-dimensional supergravity via:

= const VI ab VJ cd x I x J ac bd .
(1.2)
4/3 SS T
In this equation x I are the cartesian coordinates on the compactification 5-sphere:
P I 2
I (x ) = 1, and is related to the determinant of the internal metric. One may also
define by taking the determinant of both sides of (1.2) and using the unimodularity of S.
The five-dimensional scalar potential is invariant under SL(2, R), and this is broken
to SL(2, Z) in the string theory. The non-compact generators of this SL(2, R) are thus
naturally identified with the N = 4 gauge coupling. Eq. (1.2) thus shows that the gauge
coupling and theta-angle on the branes is a non-trivial, but exactly known function of the
fermion and scalar masses and vevs, and of the N = 4 gauge coupling. Indeed, most of the
five-dimensional flows considered to date have constant values for the five-dimensional
dilaton and axion. Eq. (1.2) gives precisely the non-trivial running of the coupling
for such flows, and presumably for N = 1 supersymmetric flows it should subsume an
integrated version of the NSVZ exact beta-function [12].
We will examine some of these ideas in this paper, and test the conjecture (1.2) by
considering the N = 2 supersymmetric flow. In this flow, N = 4 supersymmetric Yang
Mills is softly broken to N = 2 by introducing a mass for the adjoint hypermultiplet.
Gauged N = 8 supergravity also allows us to probe a lowest mode of the Coulomb branch
of the N = 2 theory: there is a supergravity scalar that corresponds to turning YangMills
scalar vevs that, in the absence of the fermion mass, corresponds to spreading out the
branes into a uniform disk distribution. For this flow, (1.2), in principle, gives a supergravity
prediction for (, m, u), where is the running N = 2 coupling, is the N = 4 coupling,
211
m is the fermion mass and u is the non-trivial scalar vev. As we will see there are some
subtleties yet to be understood.
We begin in Section 2 by reviewing the five-dimensional description of the N = 2
supersymmetric flow, and we compute the running of the dilaton predicted by (1.2). In
Section 3 we use the results of [8] to give the exact ten-dimensional metric for the flow,
and then we examine the linearized Ansatz for the ten-dimensional 2-form fields. We then
show that the N = 2 flow must necessarily involve a running ten-dimensional dilaton.
In Section 4 we obtain the complete ten-dimensional solution, confirming our prediction
of the dilaton/axion behaviour. Finally, in Appendix A we give the consistent truncation
argument that led us to the formula (1.2).
2. The N = 2 RG flow in five dimensions
The flow that preserves N = 2 supersymmetry can be obtained from the superpotentials
considered in [13]. We need to turn on the supergravity scalar fields dual to the operators:
Ob =
4
X
j =1
Tr X X
6
X

Tr Xj Xj ,

Of = Tr 3 3 + 4 4 ,
(2.1)
j =5
Sf . In the conventions of [13], the corresponding

and to the complex conjugate operator, O
supergravity scalars are and = 1 = 2 . On this subector the tensor Wab has two
distinct eigenvalues, each with degeneracy 4. One of these two eigenvalues provides a
superpotential for the flow: 1
1
1
4 cosh(2),
2
2
where, as usual, = e , and where the potential is given by:
W =
g2
g 2 4 g 2 2
cosh(2) + 8 sinh2 (2)
4
2
16

g 2 W 2 g 2 W 2 g 2 2
+
W .
=
48
16
3
(2.2)
P =
(2.3)
The kinetic term on this sector is: 3()2 ()2 .

Taking the flow metric to have the form
2
= e2A(r) dx dx dr 2 ,
ds1,4
one then finds that supersymmetric flow equations are:

g W
g 1
d
4
=
=
cosh(2)
,
dr
12
6 2
g W
g
d
=
= 4 sinh(2),
dr
4
4
(2.4)
(2.5)
1 Note that the eigenvalue that provides the superpotential for the flow considered here is not the same as the
eigenvalue that provided the N = 1 flow in [13].
212
along with the auxilliary equation

g
dA
= W.
(2.6)
dr
3
The first step to solving the system (2.5), (2.6) is to write everything as a function of .
Thus:

2
1
cosh(2)
d
=
,
(2.7)
d
3 6 sinh(2) sinh(2)
and

2
2
cosh(2)
dA
=
.
+
d
3 6 sinh(2) sinh(2)
First note that:

cosh(2)
d
(A 2) = 2
.
d
sinh(2)
(2.8)
(2.9)
This is trivially integrated with respect to , and it yields

eA = k
2
,
sinh(2)
(2.10)
where k is a constant.
To integrate the equation for , simply define = 13 log(sinh(2)) and observe that
2
1
2 e6
d
= 6
.
=
d
3 sinh(2)
3 sinh3 (2)
It is also elementary to integrate this, and one then finds:

sinh()
6 = cosh(2) + sinh2 (2) + log
cosh()

c
1
1
,
= c + c2 1 + log
2
c+1
(2.11)
where c = cosh(2) and is a constant of integration.

This formula has different asymptotics for positive, negative or zero. Since the
superpotential has a manifest symmetry under , we focus on > 0: If is
positive then 23 + 16 log( 4 ) for large (positive) . If is negative then limits
to a finite value as goes to . If = 0 then we get the interesting ridge-line flow with
13 + 16 log( 43 ) for large (positive) . Some of these flows are shown in Fig. 1.
We claim that the choice, = 0 corresponds to the pure N = 2 flow with vanishing
scalar vev. The simplest argument for this claim is obtained from Fig. 1. There are
several obvious ridges in this figure. The ridges with = 0 and varying are two of
the N = 4 supersymmetric Coulomb branch flows identified in [11]. The two other ridges
are equivalent under and are obtained by setting = 0. They correspond to
massive supersymmetric flows, and there is only one such preferred flow, namely, the
pure N = 2 flow with no scalar vev. One should note that the flow along the -axis to
the left corresponds to the Coulomb branch in which the branes spread out in a disk in
213
Fig. 1. Contours of the superpotential showing some of the steepest descent flows. The horizontal
and vertical axis are and , respectively. The middle, ridge-line flow has = 0 and the three flows
to the left and right have < 0 and > 0, respectively.
the (X5 , X6 ) directions, whereas the other direction corresponds to a brane distribution
in the (X1 , X2 , X3 , X4 ) directions. The moduli space that we seek makes the scalars
(X1 , X2 , X3 , X4 ) massive, but leaves us with the vevs in the (X5 , X6 ) directions. Hence it
is natural to identify the softly broken N = 4 theory with the = 0 ridge and everything
to the left of it. This is essentially the argument given in [14], where is was further argued
that the flows to the right of the = 0 should be viewed as unphysical.
3. The ten-dimensionsional solution

In this section we lift the N = 2 flows to solutions of the chiral IIB supergravity in
ten dimensions [15,16]. As we have already explained above, one should expect to find
that all bosonic fields in ten dimensions are non-vanishing and as a result all bosonic field
equations become nontrivial. Those equations consist of [15]:
The Einstein equations:
(1)
(3)
(5)
+ TMN
+ TMN
,
RMN = TMN
(3.1)
where the energymomentum tensors of the dilaton/axion field, B, the three index
antisymmetric tensor field, F(3) , and the self-dual five-index tensor field, F(5) , are given
by
(1)
= PM PN + PN PM ,
TMN
(3.2)
214

(3)
TMN =

1
1
GP Q M GP QN + GP Q M GP QN gMN GP QR GP QR ,
8
6
1
(5)
= F P QRS M FP QRSN .
TMN
6
We work here in the unitary gauge in which B is a complex scalar field and
PM = f 2 M B,
QM = f 2 Im(BM B ),
(3.3)
(3.4)
(3.5)
with
f=
1
,
(1 BB )1/2
while the antisymmetric tensor field G(3) is given by

.
G(3) = f F(3) BF(3)
The Maxwell equations:

2
P iQM GMNP = P M GMNP iFMNP QR GP QR .
3
The dilaton equation: 2

1
M 2iQM PM = GP QR GP QR .
24
The self-dual equation:
F(5) = F(5) .
(3.6)
(3.7)
(3.8)
(3.9)
(3.10)
In addition, F(3) and F(5) satisfy Bianchi identities which follow from the definition of
those field strengths in terms of their potentials:
F(3) = dA(2),

1
Im A(2) F(3)
.
(3.11)
8
Our strategy for constructing the ten-dimensional solution is to start with the consistent
truncation Ansatz for the metric [8]. By examining the resulting Ricci tensor, we arrive
at identities that, together with the SU(2) U (1)2 symmetry, essentially determine the
general structure of the antisymmetric tensor fields. The next crucial step is to solve the
linearized Maxwell equation for the three index tensor field, G(3) , which turns out to yield
a non-vanishing source for the dilaton/axion field in (3.9). Using the consistent truncation
Ansatz for the dilaton/axion we are then able to completely solve the Einstein equations
to all orders, and fine tune all the phases and constants using the remaining equations of
motion and the Bianchi identities.
F(5) = dA(4)
2 A perceptive reader might have noticed that the sign on the right hand side of our dilaton equation is opposite
to that in (4.11) and (5.1) of [15]. It appears that there is an error in [15] in passing from (4.10) to (4.11). An
independent verification of the sign is to check whether the Bianchi identity M RMN 12 N R = 0 is consistent
with the field equations. Indeed, this is the case for the sign in (3.9), but not for the one in [15]. The consistency
between the Bianchi identities and the field equations was also verified in [17], with the resulting correct sign in
the dilaton equation.
215
3.1. The metric

Along the flow the ten-dimensional spacetime is topologically a product of AdS5 and a
sphere, S5 , with the warped product metric of the form:
2
2
= 2 ds1,4
ds52 ,
ds10
(3.12)
2 has been given in (2.4). The internal metric, ds 2 , and the warp factor, 2 ,
where ds1,4
5
are determined by the consistent truncation in terms of the scalar fields as discussed in
Appendix A.
The calculation of the explicit form of the internal metric using (A.8) is essentially
the same as in [9] (see also [18]). We represent S5 as a unit sphere in R6 with the
cartesian coordinates x I , I = 1, . . . , 6, and pass to suitable spherical coordinates to make
the SU(2) U (1)2 symmetry manifest. This is accomplished by setting
u1 = x 1 + ix 4 ,
u2 = x 2 + ix 3 ,
u3 = x 5 ix 6 ,
so that (u1 , u2 ) transform as a doublet of SU(2) with zero charge, and u3 is a singlet with
charge 1. The remaining U (1) rotates between the doublet and its conjugate. Then we use
the group action to reparamerize these coordinates as follows:

1
1
u
,
u3 = ei sin ,
(3.13)
= cos g(1 , 2 , 3 )
0
u2
where g(1 , 2 , 3 ) is an SU(2) matrix expressed in terms of Euler angles.
eI J ab , for the flow, and parametrizing the Killing vectors
Using explicit scalar 27-beins, V
in terms of the coordinates above we arrive at the final result for the internal metric:
2

2
22 + 32
1
a 2 (cX1 X2 )1/4 1 2
2
6
2
2 d
+
c d + cos
+ sin
,
ds5 =
2
3
cX2
X1
X2
(3.14)
where, as in Section 2, c cosh(2) and e . The warp factor is given by:
2 = 2/3 =
(cX1 X2 )1/4
.
The two functions, X1 and X2 , are defined by

X1 (r, ) = cos2 + (r)6 cosh 2(r) sin2 ,

X2 (r, ) = cosh 2(r) cos2 + (r)6 sin2
(3.15)
(3.16)
and we have introduced the constant, a, to account for the arbitrary normalization of the
Killing vectors. As usual, i , i = 1, 2, 3, are the SU(2) left-invariant forms, satisfying
di = 2j k .
Clearly, the metric (3.14) is invariant under SU(2) and the two U (1)s, where the first
one, U (1), acts by a translation in , while the second second one, U23 (1), rotates 2
into 3 .
Note that in our coordinates the metric (3.12) is almost diagonal. We choose the
corresponding orthonormal frames eM , M = 1, . . . , 10,
216
e1 dx 0 ,
e2 dx 1 ,
e3 dx 2 ,
e4 dx 3 ,
e5 dr,
e6 d,
e7 1 ,
e8 2 ,
e9 3 ,
e10 d.
(3.17)
The computation of the Ricci tensor becomes rather involved and is most conveniently
carried out on a computer. We find that the only non-vanishing off-diagonal components are
R56 = R65 , while the diagonal components satisfy the obvious identities, R11 = R22 =
R33 = R44 and R88 = R99 , which follow from the symmetries of the metric. We also
find the rather non-trivial identity:
R77 + R88 = 2R11 .
(3.18)
We will see below that this equation implies the vanishing of some components of the
3-form field strengths.
Throughout the calculation we use the flow equations (2.5) to eliminate derivatives with
respect to r, so that the final result depends on rational functions of the scalar fields c(r)
and (r) and trigonometric functions of . While most of the components of the Ricci
tensor are sufficiently complicated to prevent us from reproducing them here, we note that
the combination R10 10 R77 is rather simple. This will turn out important for solving the
Einstein equations below.
3.2. The dilaton
We now use (1.2) to obtain the SL(2, R) SU(1, 1) scalar matrix of the ten-dimensional
type IIB theory. We find that on the (, ) parameter space considered in Section 2, the
right-hand side of (1.2) yields:

1 cX1 cos2 + X2 sin2
6 s sin2 sin cos
,
(3.19)
cX1 sin2 + X2 cos2
6 s sin2 sin cos
2
where c = cosh(2) and s = sinh(2). According to (1.2) the determinant should yield
8/3 , and from this we obtain:
=
3/2
3/8 ,
c X1 X2
(3.20)
which is consistent with (3.15). Note that 2 det(gmp g pq ) where gmp is the internal
metric given by (3.14) and g pq is the inverse of the round internal metric at = = 0.
One can determine S in the symmetric gauge by taking the square root of this matrix. For
direct comparison with the IIB field equations in [15] we also pass to the SU(1, 1) basis in
which the dilaton/axion matrix takes the form:

1 B
.
(3.21)
V =f
B 1
We then find the following result:

cX1 1/4
cX1 1/4
1
+
,
f=
2
X2
X2
1
fB =
2

cX1
X2
1/4
cX1
X2
217
1/4
e2i .
(3.22)
3.3. The antisymmetric tensor fields

A Poincar and SU(2) U (1)2 invariant Ansatz for the self-dual antisymmetric tensor
field, F(5) , reads
F(5) = F + F ,
F = dx 0 dx 1 dx 2 dx 3 dw,
(3.23)
where w(r, ) is an arbitrary function. The self-duality equation (3.10) is then satisfied by
construction. The non-vanishing components of the energymomentum tensor satisfy:
T11 = T22 = = T33 = T77 = = T10 10 = A2 + B 2 ,
(5)
(5)
(5)
(5)
(5)
(3.24)
(5)
T55
(5)
= T66
(3.25)
= A2 B 2 ,
and
(5)
(5)
= T65
= 2AB,
T56
(3.26)
where
s4
w
23/2
,
4
1/8
9/2
5/8
ak c (X1 X2 )
s4
w
2
,
B = 4 11/2
5/8
k
(cX1 X2 )
r
A =
(3.27)
and k is the constant introduced in (2.10).

The most general Ansatz for the potential, A(2) , that gives an SU(2) U23 (1) invariant
field strength, G(3) , with the U (1) charge +1 is

A(2) = ei a1 (r, )d 1 + a2 (r, )2 3 + a3 (r, )1 d + a4 (r, )d d ,
(3.28)
where ai (r, ) are some arbitrary complex functions.
In solving the equations of motion we want to test the conjecture (3.22), and assume as
little as possible of its form. However, the group theory implies that the dilaton must have
the form:
B = b(r, )e2i .
(3.29)
In particular, the energymomentum tensor of the dilaton vanishes in the directions 7, 8, 9

as well as in the directions parallel to brane. It follows that the only energymomentum
tensor that can contribute to (3.18) is T (3) . One finds that the only way to satisfy (3.1) and
(3.18) is to impose:
a4 (r, ) = 0.
(3.30)
(1)
TMN
(3)
and TMN
,
If one computes the components of the energymomentum tensors,
in the orthonormal frame (3.17) we find that they have some non-zero off-diagonal
218
components where the corresponding components of the Ricci tensor vanish. These
algebraic constraints are most simply (but not necessarily uniquely) solved by requiring
that the phase of b (if any) is independent of r and , and the functions a1 and a2 are pure
imaginary, while a3 is real.
At this point it is instructive to solve the linearized equations of motion in the UV
limit, r . In this limit the
metric (3.12) approaches the product metric on AdS5 S5
with the radii L = 2/g and a/ 2, respectively, as seen from the expansions A(r) r/L,
er/L , re2r/L that follow from (2.5) and (2.6). The five-index tensor should
reduce to the usual FreundRubin Ansatz. Thus we take

1
k4
(3.31)
w=m 4 +O 2 ,
4s
s
and find that the UV limit reproduces the correct solution of the Einstein equations at the
N = 8 critical point provided
a2 =
8
g2
and m2 =
1
.
16
(3.32)
Futhermore we consider a linearized Ansatz for the two-index gauge potentials of the
form:

(3.33)
ai (r, ) = er/La i ( ) + O e(+1)r/L ,
for some constant . From the linearized analysis of [2,3], and because these linearized
fields are dual to fermion bilinears on the brane, we must have either = 1 or = 3. The
former mode is non-normalizable and corresponds to a massive flow, while the latter is
normalizable and corresponds to a gaugino condensate, which is a vacuum modulus.
The Maxwell equations do indeed imply that = 1 or = 3. There are also four
independent Maxwell equations for the functions a i , which can be reduced to a single
third order equation (which has regular singular points) for a 1 ( ). The regular solution is
a 1 ( ) = a0 i cos( ),
(3.34)
where a0 is a real constant. The remaining two functions are

a 2 ( ) = a0 i cos2 sin
and a 3 ( ) = a0 cos2 sin .
(3.35)
Substituting this solution into the right hand side of the dilaton equation we find

1
GMNP GMNP 2 9 e2r/Le2i sin2 .
24
(3.36)
This reveals two interesting features: If = 3 then the dilaton does not flow at lowest order,
whereas if = 1 then the dilaton must flow. So deforming the ground state of the Yang
Mills theory does not (at lowest order) require the dilaton to run, but if the flow involves
giving a mass to the fermions then the dilaton must run. A similar linearized analysis has
recently been done for some N = 1 supersymmetric flows, and once again it was shown
that the dilaton had to run [19].
219
4. The complete solution in ten dimensions

With the general structure of the solution determined by group theory and the linearized
form, we now turn to the full solution. We take the metric in (3.12), (2.4) and (3.14),
the dilaton in (3.21) and (3.22), the five-index tensor field, F(5) , of the form (3.23) and
the three-index tensor, G(3) , defined by the potential A(2) in (3.28) with pure imaginary
functions a1 and a2 , a real function a3 and a4 = 0.
4.1. Solving the equations of motion
We start with the Einstein equations (3.1) and cosider linear combinations for which
there are some cancellations or obvious simplifications of the energymomentum tensors
on the right hand side.
The first such example is the difference between the (10, 10) and (1, 1) equations, where
we find
R10 10 R11 6= 0.
(4.1)
Now, it is easy to see that the reality conditions on the functions ai imply the identity
e2i
GMNP GMNP .
(4.2)
24
Combined with the dilaton equation (3.9), this provides us with a nontrivial test of the
consistent truncation Ansatz for the metric and the dilaton, namely,

(4.3)
R10 10 R11 = 2|P10 |2 + e2i M 2iQM PM .
(3)
(3)
T10 10 T11 =
We find that this identity is indeed satisfied by the dilaton/axion given in (3.22).
In fact, the above calculation can be viewed as an independent confirmation that the tendimensional dilaton must run. The only imput that goes into (4.2) is a symmetry constraint
on which components, GMNP , can be non-zero and that those components are purly real
or imaginary up to the ei phase. Thus, if we tried to set the dilaton to zero, we would end
up with the vanishing right hand side in (4.3), which would contradict (4.1).
The next combination of the Einstein equations is the difference
(3)
(3)
R10 10 R77 2|P10 |2 = T10 10 T77 ,

where we have used
g2
4
(1)
T77
= 0 and
3 sinh2 (2)(2X1
(5)
T10 10
+ 6 X 2 )2
(cosh(2)X1 X2 )5/4
(5)
= T77 .
(4.4)
Evaluating (4.4) explicitly we get
9 tanh2 (2)
= |G8910|2 |G567|2 ,
4 (cosh(2)X1 X2 )5/4
(4.5)
g2
and verify that in the linearized limit the first and the second term on the left hand side
reduce to the first and the second term, respectively, on the right hand side. Thus we set
9/2 tanh(2)
g
ei ,
2 (cosh(2)X1 X2 )1/8
g sinh(2)(2X1 + 6 X2 ) i
e ,
G8910 =
2 3/2 (cosh(2)X1 X2 )5/8
G567 = i
(4.6)
220
where the signs have been chosen to agree with the linearized solution to the Maxwell
equations.
By expanding (3.7) in terms of the potential and using (3.22) one obtains
g 2 i 1/2 c3/4 sec a1
e
,
4
(cX1 X2 )1/8 r
g3
X1 X2 csc sec2
(a2 2a3 ).
G8910 = ei 3/2
8
(cX1 X2 )5/8
G567 = i
(4.7)
The first order equation for a1 (r, ) is can be easily integrated, and by invoking once more
the linearized limit we can identify the remaing two functions a2 (r, ) and a3 (r, ). The
result is a simple modification of the linearized solution (3.34) and (3.35):
4
tanh(2) cos ,
g2
4 6 sinh(2)
sin cos2 ,
a2 (r, ) = i 2
g
X1
4 sinh(2)
sin cos2 .
a3 (r, ) = 2
X2
g
a1 (r, ) = i
(4.8)
At this stage a nontrivial check for our solution is the sum of the (5,5) and (6,6) Einstein
equations in which according to (3.25) the contribution from the yet undetermined energy
momentum tensor of the five-index tensor cancels.
Now, the five-index tensor is calculated using, e.g., the (1,1) and (5,5) Einstein equations
and fixing the sign to agree with the linearized Ansatz. The result is
w(r, ) =
k 4 6 X1
.
4 sinh4 (2)
(4.9)
Finally, it is a matter of a straightforward algebra to verify that all the remaining

equations of motion and the Bianchi identities are satisfied! The conjecture (1.2) has thus
passed a collection of non-trivial tests perfectly.
4.2. Asymptotic behaviour of the solution
As was noted in Section 2, there are three distinct asymptotic flows:
(i) > 0: 23 + 16 log( 4 ) for large (positive) .
(ii) < 0: , const.
(iii) = 0: 13 + 16 log( 43 ) for large (positive) .
Here we will focus primarily on the last option as we believe it should exhibit the
most interesting new behaviour. As discussed earlier, the > 0 flow is expected to be
unphysical, while the < 0 flow will be akin to the Coulomb branch of the N = 4 theory.
First, the = 0 is most interesting in that the dilaton depends upon in the asymptotic
limit. For < 0 we find B 0 as , and the asymptotic dilaton/axion
configuration is trivial. Indeed, it approaches the constant dilaton background of the N = 4
221
UV fixed point. For > 0 we have B e2i as and the matrix S diverges in no
matter what direction we approach the core of the solution. For = 0 we have:

B=
(1 + 23 tan2 ( ))1/2 1
(1 + 23 tan2 ( ))1/2 + 1

e2i .
(4.10)
Note that generically |B| < 1, except for = /2 for which one has B = e2i . So the
transition from < 0 to > 0 can be thought of as moving from an asymptotically trivial
dilaton, to a non-trivial dilaton matrix that is finite except on the ring = /2, and thence
to a dilaton matrix that is asymptotically singular in all directions.
One should note that = /2 corresponds to setting the cartesian coordinates
1
x = x 2 = x 3 = x 4 = 0 on S5 . The remaining non-trivial coordinates x 5 and x 6 thus
define a ring, which is presumably the enhanon ring of [20]. The fact that the dilaton
is asymtotically trivial for < 0, is singular for > 0, and exhibits the milder ring
singularity for = 0 further supports the identification of the supergravity flows with the
various field theory limits.
The Einstein metric behaves similarly. Setting = 0 we find that as the vielbein
behaves according to:
ea 2e2 dx a ,
a = 1, . . . , 4;
r
3
5
6
a d,
e
e 3a d,
2

1/2
2
1
2 ,
e8 a 1 + tan2 ( )
3
2 2
r
3
10
a tan( ) d,
e
2
1
e7 ae2 1 ,
2 2

1/2
2
1
e9 a 1 + tan2 ( )
3 ,
3
2 2
(4.11)
where
1/8

1/4
2 2
2
1/2
(cos )
.
1 + tan ( )
3
3
(4.12)
As , and for 6= /2, the metric remains regular, and the D3-branes appear be at
the bottom of an infinitely long throat, much as they are at a conformal fixed point. The
metric does not quite have asymptotic conformal invariance owing to the -dependence
of e7 . It is however tempting to speculate that the near-conformality of this asymptotic
metric may be related to a flow to the large N versions of ArgyresDouglas points [21,22].
At = /2 the metric (as well as the dilaton) become singular. One can re-analyse the
asymptotics, and the precise details depend upon whether one looks at the Einstein metric
or string metric. The latter still sees an infinitely long throat, whereas the Einstein metric
is singular at finite distance. We also find that in either metric, the coefficient of d, and
hence the diameter of the ring singularity goes to infinity as .
222
5. Conclusions
In obtaining and checking (1.2), we have exposed a very interesting aspect of consistent
truncation, whose physical consequence is that the five-dimensional dilaton coset should
be identified with the SL(2, Z)-invariant N = 4 coupling, and not with coupling in the
gauge theory on the brane. Indeed, our expression gives the coupling on the brane as a
function of the N = 4 coupling and of the masses and vevs captured by gauged N = 8
supergravity.
Many of the the flows obtained in the literature to date [6,7,11,13,2325] keep the fivedimensional dilaton fixed. For these flows (1.2) yields the flow of the dilaton as a function
the masses and vevs that drive that flow, and if the flow is supersymmetric (1.2) must
capture the NSVZ exact beta function. Indeed, it is, in hindsight rather easy to identify the
IIB supergravity version of the NSVZ exact beta function: The supersymmetry variation
of the ten-dimensional fermion is [15]:
=
i
i
P G .
24
(5.1)
This must vanish along supersymmetric flows. To linear order G are the fermion
masses, while Pr is the running coupling.
The foregoing identification is, however, a little superficial in that it glosses over the fact
the dilaton doesnt just depend upon the radius, r. It also depends upon other coordinates.
For the N = 2 flow considered here it depends upon and (the latter dependence being
very simple). This makes physical sense in that one starts with an N = 4 theory, and the
flow must know which directions are getting massive. To be more specific, Seiberg
and Witten [26] showed that there was no infra-red gauge enhancement on the Coulomb
branch of N = 2 theories. Thus, even in the far infra-red, the brane description of the
N = 2 supersymmetric limit must involve a disk-like distribution of branes, or at least
something with a corresponding multipole moment. Thus it is entirely to be expected that
the metric and dilaton have non-trivial dependence on and in the IR limit. This does,
however, beg the question as to how the direction of approach is seen purely from the
perspective of the brane and, in particular, in terms of the SeibergWitten effective action.
More generally, for N = 1 flows, in which several fields are given independent masses, one
would like to relate the direction on S5 with the physics on the brane and thereby isolate
the running coupling of [12].
As yet we do not have definitive answers to these issues, but we have computed some of
the dilaton flows for other known supersymmetric flows. First, and rather surprisingly, the
dilaton and axion are constant everywhere on the two-parameter (, ) space underlying
the flow of [13]. On the other hand, the dilaton and axion do flow in a very non-trivial way
for the flow of [23]. This is presently under investigation.
Finally, and rather ironically, prior to this work we expended much effort in trying to
find supersymmetric flows in which the five-dimensional dilaton flows along with other
fields. Our failed efforts, and a heuristic argument based upon energy suggest that there
should be a no-go theorem for such supersymmetric flows. It would be interesting to try
223
to prove such a result, and then (1.2) would truly be the unique expression for the running
coupling on the brane.
Acknowledgements
We would like to thank L. Castellani, S. Gubser, J. Polchinski and E. Witten for helpful
conversations. This work was supported in part by funds provided by the DOE under grant
number DE-FG03-84ER-40168.
Appendix A. Consistent truncation revisited

The central issue in consistent truncation is to determine how the fields and field
equations of the lower-dimensional theory are embedded in those of a higher-dimensional
theory. One of the keys to this is to use the gauge invariances of both theories and
supersymmetry transformations to make this mapping precise [27]. In particular, this is
how one can find the exact metric of the higher-dimensional theory from the metric and
scalar fields of the low-dimensional theory.
The consistent truncation has been carried out in full detail for the reduction of the eleven
dimensional supergravity to four dimensions (see, e.g., [18,27,28] and the references
therein) and to seven dimensions [29,30]. The consistent truncation of IIB supergravity
has been analyzed only for various subsectors of the theory (see, e.g., [8,18,28,3034]).
Here we briefly review this technique, and then extend it to the consideration of the dilaton
and axion in IIB supergravity.
A.1. The exact form of the internal metric
The starting point is the encoding of the gauge fields of the dimensionally reduced theory
into the metric and Killing vectors of the parent theory. While this technique had always
been a staple of dimensional reduction at the linearized order, it was argued in [35] that
when properly stated, such encoding of gauge fields must be exact to all orders in fields.
The argument was based upon how the gauge symmetries in the reduced theory must be
related to the diffeomorphism invariance of the parent theory, and that this relationship
would be spoilt if the linearized Ansatz were not, in fact, exact.
To be more specific, consider a theory in D-dimensions that is reduced to d-dimensions
on a manifold, M, with isometries represented by Killing fields: K p , where indexes
the Killing fields and p is the vector index on M. Decompose the D-dimensional vielbein
according to:

e
e a
,
(A.1)
eM A =
em
em a
where the upper left corner represents the d-dimensional spacetime and the bottom right
represents the (D d)-dimensional manifold M. The claim of [35] is that
p a
ep ,
e a = A
K
(A.2)
224
is the exact consistent truncation Ansatz for the gauge fields to all orders.
It was observed in [27] that when the foregoing is combined with the supersymmetry
transformations of the gauge fields, one could obtain the exact Ansatz for the internal
metric gpq . To illustrate this we consider the S5 reduction of the IIB supergravity to five
dimensions, but we stress that the argument is very general. Consider the gravitino terms
in the supersymmetry variations of the five-dimensional gauge vector fields [4,5]:
a
VI J ab b + .
AIJ = 2i e
(A.3)
In this equation, I, J = 1, . . . , 6, and AIJ = AJI represent the SO(6) gauge fields.
The hats have been introduced to distinguish five-dimensional fields from their tendimensional antecedents.
Using (A.2) the corresponding ten-dimensional variation of the vielbein gives:
AIJ K I Jp = (e a )ea p + e a (ea p ) = 2 Im( a ) + .
(A.4)
One now recalls that the dimensional reduction to the standard Einstein action and
RaritaSchwinger actions in d dimensions requires a proliferation of warp factors. In
particular, one has:
e = d2 e ,
1
= 2(d2) ,
1
= 2(d2) ,
where the hats refer to d-dimensional quantities, and the warp-factor is given by:
q

det ep a eb p = det gmp g pq .
(A.5)
(A.6)
The inverse frame eb p and the inverse metric g pq are those of the round, maximally
supersymmetric background on M. The metric warp-factor is introduced so that the
D-dimensional Einstein action, along with its factors of g reduce to the d-dimensional
Einstein action. The RaritaSchwinger field is rescaled for the same reason, and the
supersymmetry parameter is rescaled so that one gets the canonical form for the
supersymmetry variations of and e in d-dimensions. The effect of all this warping is
that (A.4) becomes:
AIJ K I Jp = 2ea p d2 Im( a ) + .
1
(A.7)
One now compares (A.3) with (A.7). The internal indices of the former, arise
through the labelling of Killing spinors M. For the five-dimensional supergravity this
labelling is ambiguous up to a USp(8) transformation, 3 but such internal local symmetry
transformations can be eliminated by squaring and contracting with the USp(8) symplectic
form: ab . The result is the squaring and contraction of the inverse-vielbein ea p in (A.7)
yields the inverse metric, and one obtains (putting d = 5):
1 I Jp KLq e
eKLcd ac bd ,
VI J ab V
K K
a2
where the constant, a, depends upon the normalization of the Killing vectors.
2/3 g pq =
3 Recently, equations that determine those USp(8) transformations were obtained in [30].
(A.8)
225
This argument was performed in great detail in [27] for the S7 reduction of elevendimensional supergravity. The foregoing argument was also made to arrive at the
expression for the general IIB compactification metric given in [8]. The details of the local
USp(8) structure in ten dimensions were not explicitly checked for the result in [8], and
so for that reason we referred to it as a conjecture, however, a more precise statement of
that result would have been: If the truncation is consistent, then the internal metric must be
given by (A.8).
A.2. The exact form for the dilaton
In the same spirit as in [8], we will now conjecture an exact form for the ten-dimensional
dilaton.
The starting point is now the encoding of the five-dimensional tensor gauge fields in the
ten-dimensional, two-form gauge potentials AMN . At the linearized order one has:
I I
b
x ,
A = B
(A.9)
where = 1, 2 now denotes an SL(2, R) index, I is the SO(6) vector index, and x I are the
P
cartesian coordinates of the 5-sphere: I (x I )2 = 1.
It seems plausible, based on the gauge symmetries, the minimal couplings, and mixings
with the gauge fields, that this linearized Ansatz is exact to all orders. Rather than prove
this in detail, we shall assume that it is true and derive the dilaton Ansatz. The body of this
paper then represents a highly non-trivial test of this assumption.
One proceeds exactly as in the previous subsection, except that one compares specific
gravitino terms in the ten-dimensional and five-dimensional supersymmetry variations of
the two-form field strengths:
[ ]
+ 4iV [ ] +
A = 4iV+
+ 4iV [ ] + ,
= 2/3 4iV+ [ ]
(A.10)
and
I
a
b
= 2ig VIab [
] b + .
B
(A.11)
Again one can remove the local USp(8) tansformations by squaring and contracting to
obtain:

= const VI ab VJ cd x I x J ac bd ,
(A.12)
4/3 SS T
where S is the IIB dilaton/axion matrix written in the SL(2, R) basis, with the local
U (1) = O(2) acting from the right. (Note that V of [15] is generally written in the
SU(1, 1) basis.)
One obvious consistency check is that the value of obtained from taking the
determinants of both sides of (A.12) and of (A.8) agree. This was indeed confirmed in
Section 3.2.
One can choose a gauge in which S SL(2, R) is symmetric, and so one can take the
square root of (A.12), and extract S. One should also note that if the E6(6) matrix V of
226
five-dimensional scalars is, in fact, in the subgroup SL(6, R) SL(2, R) used in [4], then
the IIB dilaton and axion are indeed precisely described by the SL(2, R) factor of this
SL(6, R) SL(2, R). The formula (A.12) is consistent with this special case. However, for
general the E6(6) matrix V, the five-dimensional and ten-dimensional SL(2, R) factors have
a highly non-trivial relationship.
In terms of the AdS/CFT correspondence, this last statement means that on the Coulomb
branch of the N = 4 YangMills theory, the gauge coupling and theta angle are constant,
and are represented by the SL(2, R) of the five-dimensional theory. However, if fermion
masses are turned on in the gauge theory, then (A.12) tells us exactly how the running
of the gauge coupling and axion are determined entirely in terms of the running of all
the masses of the YangMills scalar and fermion fields. The SL(2, R) matrix of the fivedimensional theory is a global symmetry of the potential and is presumably broken to
SL(2, Z) in the quantum theory: it represents the symmetry of even the perturbed theory
under the SL(2, Z) action on the N = 4 coupling. The actual physical coupling, represented
by the ten-dimensional dilaton and axion, are non-trivial functions of this N = 4 coupling,
the masses of the fields and the scale. This relationship is given in large N theories by
(A.12).
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de-Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[4] M. Gnaydin, L.J. Romans, N.P. Warner, Gauged N = 8 supergravity in five dimensions, Phys.
Lett. B 154 (1985) 268;
M. Gnaydin, L.J. Romans, N.P. Warner, Compact and non-compact gauged supergravity
theories in five dimensions, Nucl. Phys. B 272 (1986) 598.
[5] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Gauged N = 8, D = 5 supergravity, Nucl. Phys.
B 259 (1985) 460.
[6] J. Distler, F. Zamora, Non-supersymmetric conformal field theories from stable anti-de-Sitter
spaces, Adv. Theor. Math. Phys. 2 (1999) 1405, hep-th/9810206;
J. Distler, F. Zamora, Chiral symmetry breaking in the AdS/CFT correspondence, JHEP 5
(2000) 005, hep-th/9911040.
[7] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, Novel local CFT and exact results on
perturbations of N = 4 super YangMills from AdS dynamics, JHEP 12 (1998) 022, hepth/9810126.
[8] A. Khavaev, K. Pilch, N.P. Warner, New vacua of gauged N = 8 supergravity in five
dimensions, Phys. Lett. B 487 (2000) 14, hep-th/9812035.
[9] K. Pilch, N.P. Warner, A new supersymmetric compactification of chiral IIB supergravity, Phys.
Lett. B 487 (2000) 22, hep-th/0002192.
[10] V. Balasubramanian, P. Kraus, A. Lawrence, Bulk vs. boundary dynamics in anti-de-Sitter
spacetime, Phys. Rev. D 59 (1999) 046003, hep-th/9805171.
[11] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Continuous distribution of D3-branes and
gauged supergravity, JHEP 07 (2000) 038, hep-th/9906194.
227
[12] V. Novikov, M.A. Shifman, A.I. Vainshtein, V. Zakharov, Exact Gell-Mann-low function of
supersymmetric YangMills theories from instanton calculus, Nucl. Phys. B 229 (1983) 381.
[13] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography supersymmetry and a c-theorem, CERN-TH-99-86, hep-th/9904017.
[14] S. Gubser, Curvature singularities: the good, the bad, and the naked, PUPT-1916, hep-th/
0002160.
[15] J.H. Schwarz, Covariant field equations of chiral N = 2, D = 10 supergravity, Nucl. Phys.
B 226 (1983) 269.
[16] P. Howe, P. West, The complete N = 2 D = 10 supergravity, Nucl. Phys. B 238 (1984) 181.
[17] L. Castellani, I. Pesando, The complete superspace action of chiral D = 10, N = 2 supergravity,
Int. J. Mod. Phys. A 8 (1993) 1125.
[18] M. Cvetic, H. L, C.N. Pope, Geometry of the embedding of scalar manifolds in D = 11 and
D = 10, Nucl. Phys. B 584 (2000) 149, hep-th/0002099.
[19] M. Petrini, A. Zaffaroni, The holographic RG flow to conformal and non-conformal theory,
hep-th/0002172.
[20] C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities,
Phys. Rev. D 61 (2000) 086001, hep-th/9911161.
[21] P.C. Argyres, M.R. Douglas, New phenomena in SU(3) supersymmetric gauge theory, Nucl.
Phys. B 448 (1995) 93, hep-th/9505062.
[22] P.C. Argyres, M. Ronen Plesser, N. Seiberg, E. Witten, New N = 2 superconformal field
theories in four dimensions, Nucl. Phys. B 461 (1996) 71, hep-th/9511154.
[23] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, The supergravity dual of N = 1 super Yang
Mills theory, Nucl. Phys. B 569 (2000) 451, hep-th/9909047.
[24] K. Behrndt, Domain walls of D = 5 supergravity and fixpoints of N = 1 super YangMills,
Nucl. Phys. B 573 (2000) 127, hep-th/9907070.
[25] K. Behrndt, M. Cvetic, Supersymmetric domain wall world from D = 5 simple gauged
supergravity, Phys. Lett. B 475 (2000) 253, hep-th/9909058.
[26] N. Seiberg, E. Witten, Monopole condensation, and confinement in N = 2 supersymmetric
YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2 supersymmetric QCD, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[27] B. de Wit, H. Nicolai, On the relation between d = 4 and d = 11 supergravity, Nucl.
Phys. B 243 (1984) 91;
B. de Wit, H. Nicolai, N.P. Warner, The embedding of gauged N = 8 supergravity into d = 11
supergravity, Nucl. Phys. B 255 (1985) 29.
[28] M. Cvetic, H. Lu, C.N. Pope, A. Sadrzadeh, T.A. Tran, Consistent SO(6) reduction of type IIB
supergravity on S5 , Nucl. Phys. B 586 (2000) 275, hep-th/0003103.
[29] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11d
supergravity on AdS7 S4 and self-duality in odd dimensions, Phys. Lett. B 469 (1999) 96,
hep-th/9905075;
H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS7 S4 reduction and the
origin of self-duality in odd dimensions, Nucl. Phys. B 581 (2000) 179, hep-th/9911238.
[30] H. Nastase, D. Vaman, On the nonlinear KK reductions on spheres of supergravity theories,
Nucl. Phys. B 583 (2000) 211, hep-th/0002028.
[31] M. Cvetic et al., Embedding AdS black holes in ten and eleven dimensions, Nucl. Phys. B 558
(1999) 96, hep-th/9903214.
[32] M. Cvetic, S.S. Gubser, H. Lu, C.N. Pope, Symmetric potentials of gauged supergravities in
diverse dimensions and Coulomb branch of gauge theories, Phys. Rev. D 62 (2000) 086003,
hep-th/9909121.
[33] H. Lu, C.N. Pope, T.A. Tran, Five-dimensional N = 4, SU(2) U (1) gauged supergravity from
type IIB, Phys. Lett. B 475 (2000) 261, hep-th/9909203.
228
[34] M. Cvetic, H. Lu, C.N. Pope, A. Sadrzadeh, Consistency of KaluzaKlein sphere reductions of
symmetric potentials, Phys. Rev. D 62 (2000) 046005, hep-th/0002056.
[35] U. Chattopadhyay, A. Karlhede, Consistent truncation of KaluzaKlein theories, Phys.
Lett. B 139 (1984) 279.

ChernSimons supersymmetric branes

Pablo Mora a,b
a Department of Physics, University of Maryland at College Park, College Park, MD 20742-4111, USA
b Instituto de Fisica, Facultad de Ciencias, Igu 4225, Montevideo, Uruguay
Abstract
The purpose of this paper is to continue the study of the class of models proposed in a previous
letter. The model corresponds to a system of branes of diverse dimensionalities with ChernSimons
actions for a supergroup, embedded in a background described also by a ChernSimons action.
The model treats the background and the branes on an equal footing, providing a brane-target
space democracy. Here we suggest some possible extensions of the original model, and discuss
its equations of motion, as well as the issue of currents and charges carried by the branes. We also
discuss the relationship with M-theory and Superstring theory. 2001 Elsevier Science B.V. All
rights reserved.
1. Introduction
ChernSimons Supergravity (CSS) theories in diverse dimensions have been studied in
many papers during the last years [19]. Those theories are very interesting on their own
right, having a wealth of nice properties, ranging from being topological (in the sense of
being independent of any background metric), being true gauge theories for extensions
of the standard spacetime symmetry groups (like the Poincar or anti-de Sitter group),
having coupling constants that are not renormalized (so that the classical action is also the
quantum effective action) and being exactly solvable in 2 + 1 dimensions. Some time ago
Chamseddine [4] suggested that this kind of models might be regarded as the basis for an
approach to the unification of the fundamental interactions alternative to the Superstring
Theory [10] program. More recently Horava [9] proposed that a CSS may correspond
to the M-theory [1114], the fundamental theory underlying the five known consistent
superstring theories, which would then be an ordinary field theory (against the belief of
most experts in Superstring/M-theory). 1
E-mail address: pmora@wam.umd.edu (P. Mora).
1 Troncoso and Zanelli [7] suggested earlier that the low energy limit of the M-theory might be given by a CSS.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 9 - 0
230
P. Mora / Nuclear Physics B 594 (2001) 229242
On other line of development it was attempted to recast the (1 + 1)-dimensional

superstring actions as (2 + 1)-CS theories [15] (see also [3,16,17]) by a sort of thickening
of the world sheet, as a way to benefit from the good properties of the later.
Recently we suggested [18] a way to introduce fundamental supersymmetric extended
objects on CSS, inspired on the model of Ref. [19] for the coupling of branes to Yang
Mills fields, which brought together these approaches. There is a model of that class
which has at least two of the superstring theories (IIA and IIB) as sectors of its phase
space and it describe branes with actions of the same form that the one describing the
background, which is a CSS with the branes acting as sources for the (super)gauge
fields, and interacting with that background. Also it is plausible that standard supergravity
approximately describes some regime of the theory [79]. It was then argued in [18] that
a model of the kind considered there might correspond to the quantum effective action of
the M-theory.
In this paper I continue the study of that kind of models by reviewing its invariances and
transformation properties, discussing its equations of motion and the issue of currents and
charges carried by the branes. Finally I discuss several of the questions and open problems
where I believe further developments are to be expected. The model of this paper differs
from the one on Ref. [18] in the fact that we point out that it may not be necessary to
introduce a kinetic term by hand in order to make contact with superstring theory.
2. The action
The mathematical tools used through this paper are discussed in (and were to a great
extent developed in) references [2023]. We will follow the notation and conventions of
Ref. [23]. 2 We consider a (super)group G with generators T I , the gauge potential 1-form
A = AIm T I dx m and the curvature 2-form F = dA + A2, defined on a manifold M 2N+2 of
even dimension 2N . The covariant derivative is D = d + [A, ] and it follows that F satisfy
the Bianchi identity DF = 0. An invariant polynomial P (F ) is defined as the formal sum
P (F ) =
N
X

n STr F n+1 ,
(1)
n=0
where STr(T I1 T In+1 ) = g I1 In+1 stands for an invariant symmetric supertrace on the
algebra of G. An important property is
d STr( ) = STr(D).
(2)
We define the action for our system by

2 An excellent recent work covering the mathematical background of these articles is the book on anomalies
by R. Bertlmann [24]. The reader may find amusing to know that Dr. Bertlmann socks have inspired in J.S. Bell
some deep reflections on the foundations of quantum mechanics [25]. Bertlmann reminiscences of J.S. Bell on
the topic of anomalies (where he was one of the pioneers) can be found in the Preface to [24].
S=
N
X

k01 STr F n+1 ,
n=0
231
(3)
S 2n+1
where S 2N+1 is the boundary of M 2N+2 , S 2N+1 M 2N+2 , and the manifolds S 2n+1 are
embedded into S 2N+1 . In case S 2N+1 itself has a boundary 2N , then S 2N+1 is included
into the boundary of M 2N+2 . The manifolds S 2n+1 may have boundaries in manifolds
2n .
The Cartan homotopy operator acting on polynomials P(Ft , At ), with At interpolating
between two gauge potentials A0 and A1 as
At = tA1 + (1 t)A0 ,
Ft = dAt + A2t ,
(4)
is defined to be
Z1
k01 P(Ft , At ) =
dt lt P(Ft , At ),
(5)
with the operator lt defined to act on arbitrary polynomials by

lt At = 0,
lt Ft = A1 A0 J,
(6)
and the convention that lt is defined to act as an antiderivation.

The dynamical variables are the gauge potentials AIm , the embedding coordinates of
m
i
i
((2n+1)
), m = 0, . . . , 2N + 1, where the (2n+1)
with
the submanifolds S 2n+1 , X(2n+1)
i = 0, . . . , 2n + 1, are local coordinates in S 2n+1 and the embedding coordinates of the
m ( i
i
submanifolds 2n , X(2n)
(2n+1) ), m = 0, . . . , 2N + 1, where the (2n) with i = 0, . . . , 2n,
2n
2n
2n+1
the Xm s of the same
are local coordinates in (of course, in the boundary of S
point must coincide as functions of the s or the corresponding s). Notice that all these
manifolds are supposed to be noncompact at least in what would be the time direction.
It was noticed in Ref. [6] that the dimensionless coefficients n can consistently take only
a discrete set of values if we require that the quantum theory must be independent of the
way in which the manifolds S 2n+1 could be extended into manifolds M 2n+2 included into
M 2N+2 such that S 2n+1 M 2n+2 . For instance P (F ) = h STr[eiF /(2)] would work.
A useful relationship is the Cartan homotopy formula
(k01 d + dk01)P(Ft , At ) = P(F1 , A1 ) P(F0 , A0 ),
(7)
which follows by integrating

(lt d + dlt )P(Ft , At ) =
P(Ft , At )
t
(8)
over t from 0 to 1.
Then we get
L2n+1 k01 STr
Ftn+1
Z1
= (n + 1)
dt STr J F n
0
(9)
232
and
S=
N
X
Z
L2n+1
n=0
N
X
n S2n+1
(10)

dt STr J F n .
(11)
n=0
S 2n+1
or
Z
N
X
(n + 1)n
S=
n=0
S 2n+1
Z1
0
In Ref. [18] a kinetic term was added by hand in the boundary 2n of the manifolds
given by
Z

ij
p
1
(2n)
d 2n (2n) (2n) (2n)
STr(Ji Jj ) (2n 2)
(12)
SK =
2
S 2n+1
2n
or alternatively of the BornInfeld-like form

q

Z

d 2n (2n) STr sdet Ji Jj + (F0 )ij + (F1 )ij
(13)
2n
or
q
Z
2n
d (2n) STr

sdet STr(Ji Jj ) + (F0 )ij + (F1 )ij ,
(14)
2n
where the superdeterminant is taken in the curved indices i, j of the pull-backs on S d

while the supertraces are taken on the group indices. We will ignore those kinetic terms
here, even though it may be that they appear as quantum corrections to the action of Eq. (3)
in the quantum effective action.
0
(Ft , At ), where the Chern
From the Cartan homotopy formula for P(Ft , At ) = I2n+1
0
Simons (CS) form I2n+1 (F, A) is defined as
Z1

ds STr AFsn ,
(15)
Fs = dAs + A2s = sF + s(s 1)A2,
(16)
0
(F, A) = (n + 1)
I2n+1
0
where
As = sA,
we get

0
0
0
(F1 , A1 ) I2n+1
(F0 , A0 ) d k01I2n+1
(Ft , At ) ,
k01 STr Ftn+1 = I2n+1
(17)
0
(F, A) = STr(F n+1 ). The last term is a boundary term which is
where we used that dI2n+1
given explicitly by
Z1
C2n (F1 , A1 ; A0, F0 ) n(n + 1)
Z1
ds

dt s STr At J Fstn1 ,
(18)
233
with Fst = sFt + s(s 1)A2t and At = tA1 + (1 t)A0 . Then the action of Eq. (11) can
be written as
( Z
)
Z
N
X

0
0
n
(F0 , A0 )
C2n .
(19)
I2n+1 (F1 , A1 ) I2n+1
S=
n=0
S 2n+1
2n
3. Invariances of the action

By construction the action of Eq. (3) and the kinetic terms given above are generally
covariant.
The content of this section regarding the transformation properties of CS forms and
descent equations was developed in the context of anomalies in quantum field theories in
Refs. [2023]. I find it worthwhile to review it here because I apply it in a different context
and conceptual framework.
Under (super)gauge transformations we have
Ar = g 1 (Ar + d)g,
g
r = 0, 1,
(20)
where g is an element of the group. It follows that J = A1 A0 transforms covariantly if

both A1 and A0 are transformed with the same g
J g = g 1 J g.
(21)
F g = g 1 F g.
(22)
Also
Under infinitesimal gauge transformations

v Ar = dv + [Ar , v],
r = 0, 1.
(23)
Then
v J = [J, v]
(24)
v F = [F, v].
(25)
and
From the facts that F and J are gauge covariant, the cyclicity of STr( ) and Eq. (11) it
follows that the action and also the kinetic terms of Eqs. (12)(14) are gauge invariant.
In order to compute the change of the action under gauge transformations involving
only one of the gauge fields A1 or A0 it is useful to consider elements of the gauge group
g(x, ) function of the point x on the base manifold and a set of parameters on some
parameter space, such that g(x, = 0) = 1 (the identity). In addition to the standard
exterior derivative d = dx x , we define the exterior derivative in parameter space =
d . If
A = g 1 (A + d)g = Ag
(26)
234
and
A = g 1 (A + d + )g = g 1 (A + )g = A + v,
g 1 g,
(27)
we have
=
= d + d =
= 0. It is easy to
with = d + and v =
verify that A = DA v so that generates gauge transformations with parameter v =

g 1 g. The derivative corresponds to the BRS operator and v to the FaddeevPopov
ghost [2024]. Defining F = d A + A 2 and F = A + A2 it is possible to check the
Russian formula
d2
b(A)
= g 1 F (A)g.
F (A) = F
(28)
Considering
A t = t A 1 + (1 t)A0 ,
At = tA1 + (1 t)A0 ,
and A0 = 0, then from the Russian formula and the Cartan homotopy formula with
Pt = STr(Ftn+1 ) for At and A t we get
0
0
(A 1 + v, A0 ) = dI2n+1
(A 1 , A0 ),
(d + )I2n+1
(29)
0
(A1 , A0 ) = k01 STr(Ftn+1 ) = L2n+1 correspond to the pieces of diverse
where I2n+1
dimension of our Lagrangian. If we expand by the order in v
0
(A 1 + v, A0 ) =
I2n+1
2n+1
X
k
I2n+1k
(v, A 1 , A0 )
(30)
k=0
then we obtain the descent equations

k+1
k
(v, A 1 , A0 ) + dI2nk
(v, A 1 , A0 ) = 0,
I2n+1k
k = 0, . . . , 2n + 1.
(31)
In particular,
0
1
(A 1 , A0 ) + dI2nk
(v, A 1 , A0 ) = 0
I2n+1k
(32)
gives the variation of our action under a gauge transformation involving only A1 (notice
that A 1 |=0 = A1 ) as a boundary term. A similar identity holds for gauge transformations
involving only A0 .
4. Equations of motion
In the case of CS gauge theory or supergravity without branes or boundaries the action
is
Z
S=
0
I2n+1
(F, A) =
S 2n+1
STr(F n+1 ),
(33)
M 2n+2
then under variations of the gauge potential

Z
Z

n
STr D(A)F = (n + 1)
S = (n + 1)
M 2n+2
M 2n+2

d STr(AF n ) ,
(34)
235
where we used F = D(A) and Eq. (2). From Stokes theorem

Z
STr(AF n ).
S = (n + 1)
(35)
S 2n+1
Then the equations of motion S/A = 0 are [4,5,7]

STr(T I F n ) = 0.
(36)
Whether or not these equations are related to General Relativity and/or standard
Supergravity in diverse dimensions has been discussed in Ref. [35,79].
In the case there are boundaries and branes we need to use that
1 J = A1 ,
0 J = A0 ,

r Ft = Dt (r At ) = d(r At ) + At , (r At ) ,
1 At = tA1 ,
r = 0, 1,
0 At = (1 t)A0 .
Therefore, we have for variations of A1

Z1
1 L2n+1 = (n + 1)
dt STr
A1 Ftn
Z1
+ n(n + 1)
dt t STr J Dt (A1 )Ftn1
(37)
but

d STr J A1 Ftn = STr Dt J A1 Ftn1 STr J Dt (A1 )Ftn1 ,
(38)
where we used d STr( ) = STr(Dt ) and the Bianchi identity Dt Ft = 0, then

Z1
1 L2n+1 = (n + 1)
dt STr
A1 Ftn
Z1
+ n(n + 1)
"
dt t STr A1 Dt (J )Ftn1
Z1
+ d n(n + 1)
dt t STr
A1 J Ftn1
#
.
(39)
The last term of the second member is a boundary term. Under variations of A0 we have
Z1
0 L2n+1 = (n + 1)
"
dt STr
A0 Ftn
Z1
dt (1 t) STr A0 Dt (J )Ftn1
+ n(n + 1)
Z1
dt (1 t) STr
+ d n(n + 1)
A0 J Ftn1
#
.
(40)
If we write

(r)
r L2n+1 = STr Ar Q(r)
2n + d STr Ar R2n1 ,
(41)
236
where
Q(1)
2n
Z1
= (n + 1)
Z1
dt Ftn
+ n(n + 1)
Q(0)
2n = (n + 1)
dt tDt (J )Ftn1 ,
0
Z1
Z1
dt Ftn + n(n + 1)
0
(1)
R2n1
dt (1 t)Dt (J )Ftn1 ,
0
Z1
= n(n + 1)
(0)
R2n1
dt tJ Ftn1 ,
Z1
= n(n + 1)
dt (1 t)J Ftn1 .
0
(42)
Then we can write
" Z
N
X
n
r S =
n=0
or
STr
(r)
Ar Q2n +
S 2n+1
Z
STr
(r)
Ar R2n1
(43)
2n
Z
r S =
d 2N+1 x (Ar )Im J (r)mI ,
(44)
S 2N+1
where
J (r)mI (x m ) =
N
X
n=0
" Z
n
(r)mI
d 2n+1 2n+1 J(2n+1)
(r)mI
d 2n 2n J(2n)
S 2n+1
(45)
2n
with

(r)mI
(r)
J(2n+1) = 2N+1 X(2n+1) (2n+1 ) x m STr T I Q2n m
2 m2n+1
m2
[m
2n+1 i1 i2n+1
i2 X(2n+1)
i2n+1 X(2n+1)

i1 X(2n+1)
m
(46)
and

(r)mI
(r)
m
= 2N+1 X(2n)
(2n ) x m STr T I R2n1
J(2n)
m
2 m2n
m ]
[m
m2
2n i1 i2n
i2 X(2n)
i2n X(2n)

.
i1 X(2n)
(47)
The equations of motion S/Ar = 0 are then

J (r)mI = 0.
(48)
These equations can be interpreted as the equations found before in the case there are no
boundaries or branes but now with source terms given by currents carried by the branes.
Concerning the equations of motion corresponding to extremizing the action under
variations of the embedding functions X it is convenient to write
S=
N
X
" Z
d 2n+1 2n+1 2n+1
n=0
m2n+1 ] i1 i2n+1
X[m1 i2n+1 X(2n+1)

m1 m2n+1 i1 (2n+1)
S 2n+1
2n
237
[m1
m2n ] i1 i2n
2n 2n m m i1 X(2n)
i2n X(2n)

1
2n
#
,
(49)
2n
where we separated the bulk and boundary contributions to L2n+1 as in Eq. (19). In the
previous expression the dependence of S on the functions X is through the s while the
dependence of S on X is through the pull-back factors. The EulerLagrange equations
s then give
for X(p)

mp ] i1 ip
[r
m2
= 0.
p r (p) sm mp
s ((p) )rm2 mp i1 X(p) i2 X(p) ip X(p)
2
X(p)
X(p)
(50)
In applying the EulerLagrange for the bulk S 2n+1 we left out a boundary term
h
i

m
[m2
s] i1 ip
ip X(p)p X(p)

.
i1 (p) sm mp i2 X(p)
(51)
We can require the boundary term term to vanish, in analogy with open strings,

m
[m2
s] i1 ip
ip X(p)p X(p)

= 0,
(p) sm mp i2 X(p)
(52)
which is not to be taken as a condition on what points can be swept by the boundaries of the
branes, its velocities or the allowed variations X but only on the spatial derivatives of the
functions X at the boundary. Alternatively we can add that term as an extra contribution to
the EulerLagrange equations at the boundary. If we add the kinetic terms of Eqs. (12)(14)
there would be an extra term to the current located on the boundaries 2n of the branes,
and extra terms in the EulerLagrange equations. The equations of motion of the auxiliary
metrics in the kinetic terms of Eq. (12) are algebraic.
5. Discussion
5.1. Connection with superstring/M-theory
In Ref. [18] we considered a the model of Eq. (3) plus the kinetic term of Eq. (12)
added by hand for the M-theory group Osp(32|1) [9,14,26] and related it to IIA
and IIB superstrings. That group has generators Pa (translations), Q (generators of
supersymmetries), Mab (Lorentz) and Za1 a5 , a = 0, . . . , 10. We took the symmetric trace
to be the standard symmetricized supertrace in the adjoint representation of G. 3
3 For the following discussion the relevant traces of products of generators in the adjoint representation
of Osp(32|1) are [18] STr(P P ), STr(P Q) and STr(QQ), while for the WZNW-term the relevant traces are
STr(P P P ), STr(P P Q), STr(QQP ) and STr(QQQ). Among these, the non-vanishing ones (after the limiting
process of [18]) are normalized as
STr(Pa Pb ) = ab ,
STr(Pa Q Q ) = (a C 1 ) .
238
The point we would like to make here is that that connection can be made for the action
of Eq. (3) alone. We consider as our candidate to the M-theory action

Z
5
X
F
(53)
k01 STr exp i
S =h
2
n=1
S 2n+1
for Osp(32|1). 4 The integrals in the previous expression are supposed to pick up the
differential forms of the right order. We will use the same notation as [18] and take the
InonuWigner limit as it was done in that paper. We need to consider a pure gauge A1 =
ba
g 1 dg with g restricted to the form g = ei X Pa + Q . Then essentially A0 = [dXa +
i a d ] + d Q . We will also make A0 = A1 . Then proceeding as in [18] we consider
a 11D slab with two 10D noncompact boundaries (each with the topology of R 10 ) for
ba with the 10D coordinates. We will also take
which we identify the gauge parameters X
the pull-back of the gauge superfield A1 to be self-dual or anti-self-dual A1 = A1 with
respect to some arbitrary auxiliary metric in the 2D boundaries of the 2-brane, which are
contained in the 10D boundary of the 11D slab. To fix ideas we may think that metric is
a 2D Minkowski metric = (1, 1), then the (anti)self-duality condition means that we
are keeping only the (left) right movers with respect to that metric. The bulk parts Chern
Simons of Eq. (53) for the 2-brane give boundary WZW terms because the potentials are
pure gauge, and both terms actually add because we chose A1 = A0 . On the other hand
from Eq. (18) C2 = STr(A1 A0 ), but the wedge product of a differential form with its dual
with respect to a metric is the square with that metric. It follows that the C2 in each side of
the slab would look like half the kinetic term computed with that metric (in the sense we
would only have right or left movers). The 11D Majorana spinor has 32 components. We
can split those 32 components in 10D either as (32) = (16L , 16R ) or as (32) = (16R , 16R )
(equivalently (32) = (16L , 16L ), where the subscripts L and R denote the chiralities
in 10D. Choosing properly the anti-self-duality or self-duality conditions so that we have
for instance left movers in one face and right movers in the other we can assemble left or
right movers for X and for the spinors from both faces. Then each choice of the splitting
of 11D spinors into two 10D spinors yield IIA or IIB superstrings. Of course the previous
considerations only mean that IIA and IIB strings corresponds to sectors of the phase
space of the theory, contributing to the quantum path integral. It would take more work to
check if those configurations are actually solutions of the equations of motion.
5.2. Duality
In our model we can distinguish between the base manifold coordinates X and the gauge
b and only identify them as a coordinate choice after choosing a topology for
parameters X,
4 This must be considered as a best guess in the sense that even if a model of the class considered here would
correspond to the M-theory action it may be that the group is a different one (for instance, Osp(64|1)), or that
there are extra boundary terms as the ones of Eqs. (12)(14), or that the invariant trace required is a different one
(for instance including some suitable extension of the Euler characteristic). Note incidentally that when we plug
the action of Eq. (53) into the path integral the result would be independent of any free parameter.
239
the base manifold, as done in a particular case in [18]. That means that we can get the
various dualities of [27] at the level of the Osp(32|1) algebra, corresponding to different
choices of the set of operators associated to translations in 10D, by picking different gs on
b
the MaurerCartan form as above with the proper Pa , and doing the identification X X
b
for the corresponding X on the appropriate 10D submanifold.
Concerning T-duality, the fact that the dimensional reduction of Chern characters are
Chern characters in the lower dimension would single them out of all the possible
invariant polynomials. Namely we should consider just a linear combination of terms of
the form STr(F n+1 ) instead of some arbitrary combination of products of traces or some
supersymmetric extension of the Euler characteristic.
5.3. D-branes and K-theory
Several recent works deal with the issue of D-branes and K-theory [28,29]. 5 The
situation is often stated as K-theory is to be preferred to cohomology or equivalently
gauge fields are to be preferred to p-form RR-fields. In our model, as pointed out in [18],
we can mimic the RR-fields (which would then be regarded as composite) with the CS
forms of one of the gauge potentials (say A1 ) which would couple (with an anomalous
coupling) to the other one (say A0 ). The anomalous gauge transformation rule of the RR
field is then built in. Also K-theoretic constructions involve a doubling of the gauge fields,
as our model does. It seems therefore reasonable to investigate the relationship between
both approaches.
5.4. Quantum theory and quantum effective action
The quantum theory is formally defined by the path integral
X XZ
DADX(p) eiS/h ,
Z=
topologies p
where it is understood that we must sum over all the gauge field configurations and brane
and base space geometries and topologies. Suitable gauge fixing procedures should be used
to eliminate redundancies in summing configurations corresponding to the same physical
state. The topological character of the action and the quantization of the coupling constants
may be taken to imply that the action of Eq. (3) is already the quantum effective action.
However, a careful analysis of this question is clearly required.
5.5. Anomalies and full gauge invariance
It is worthwhile to notice that the variation of our action under a gauge transformation
involving only one of the gauge fields has the right form to be canceled by an anomaly
on the boundary 2n [22], as it satisfies the WessZumino consistency condition and
comes from a chain of descent equations. It is therefore tempting to look for such
5 For a review see [30].
240
a mechanism to ensure the full gauge invariance of the quantum theory under arbitrary
gauge transformations of A0 or A1 . 6 However a priori it does not seem that we need to
have anomalies at all on the even-dimensional brane boundaries, as we could in principle
arrange the fermion fields (the fermionic gauge fields of the supergroup) in couples of
opposite chirality so that the world-volume theory is nonchiral. Yet if we chose to make
it chiral, computing the precise form of the anomaly would be highly nontrivial, as
we have fermionic gauge fields transforming as such under gauge transformations and
reparametrizations (general coordinate transformations). It follows that the usual formulas
for standard (whether gauge or gravitational) or sigma model anomalies would not
apply. A discussion of anomalies for extended objects (which unfortunately I could not
translate in any straightforward way to the problem at hand), and when and how to apply
which formulas can be found in Refs. [32] and [33]. 7 This approach seems to be the best
hope to single out the gauge supergroup and the dimension of the spacetime manifold. 8
If our model in eleven dimensions and with gauge group Osp(32|1) has as limiting
cases the five consistent superstring theories, then consistency and anomaly cancellation
considerations from the later should translate into the fact that the former is the only one
of our class of models that works. We can make a hand waving argument in the sense
that a fully consistent theory of Nature should be perturbatively smooth when expanded
around any point of its phase space, in the vague sense that each order must be finite,
even if the point is not the true vacuum and the whole series does not converge.
5.6. Vacuum and phenomenology
If the action is already the effective quantum action, as claimed, the problem of finding
the vacuum reduces to finding a solution of the classical equations of motion. Doing
phenomenology would require a realistic solution in the sense of having four large nearly
flat 3+1 spacetime dimensions (at least at some stage of cosmic evolution) and the masses
and coupling constants of particle physics could be read from the coefficients of the lower
order terms in a background field expansion quantization.
5.7. Group manifold/Superspace formulation
An interesting possibility would be to treat the base manifold and the fiber on the same
footing by a group manifold approach. That may furthermore allow to treat the BRS
operator and the exterior derivative d on a symmetric fashion, giving rise to a sort
of double group manifold approach. It may also be possible and useful to extend the
6 That seems also desirable as making the same variation on both sides of a t-parameter space slab is
reminiscent of distant parallelism, even though actually both A0 and A1 are evaluated at the same space
time point. Incidentally I wonder if this additional one dimensional t-parameter space has anything to do with
F-theory [31].
7 See also [34] and references therein. There are more recent references on brane anomalies but do not apply
to the peculiarities of our model.
8 It would be ironic if a model of the kind studied in [18] and the present paper is relevant to M-theory as I
believe it is, as the action of the model itself is essentially an anomaly.
241
definition of lt and the one-dimensional t-parameter to a manifold as in [22] and treat also
lt in a more symmetrical fashion (a triple group manifold approach?).
5.8. Pregeometric theory
I find very attractive the purely algebraic way in which the differential structure is treated
in Refs. [2023]. It is also remarkable the contrast between the simplicity and terseness of
the formalism set forth on those papers and the power and scope of those methods. I believe
that is a broad hint of the kind of conceptual and mathematical framework required to
describe a fundamental theory of Nature for which the differential structure is a dynamical
entity. 9 It seems that the more fundamental formulation of such a theory must be a discrete
one (Covariant Matrix Theory) [3640]. One can think that it is not possible to give
physical content to the de Rham complex without giving up the smooth manifold picture of
the spacetime. As pointed in [39] the pregeometric approach and the geometric approach
mentioned in the previous item might not be compatible, and I believe the pregeometric
one is more likely to give a conceptually tight picture of physical reality. Possibly asking
for a pregeometric and a geometric group manifold/superspace formulation at once may
be like asking for Newtons Laws and circular planetary orbits.
Acknowledgement
I am grateful to S.J. Gates Jr. and H. Nishino for stimulating discussions as well as
support and encouragement.
References
[1] P. Van Nieuwenhuizen, Phys. Rev. D 32 (1985) 872.
[2] A. Achucarro, P.K. Townsend, Phys. Lett. B 180 (1986) 89.
[3] E. Witten, Nucl. Phys. B 311 (1988) 46;
E. Witten, Nucl. Phys. B 323 (1989) 113.
[4] A.H. Chamseddine, Phys. Lett. B 233 (1989) 291;
A.H. Chamseddine, Nucl. Phys. B 346 (1990) 213.
[5] M. Baados, R. Troncoso, J. Zanelli, Phys. Rev. D 54 (1996) 2605.
[6] J. Zanelli, Phys. Rev. D 51 (1995) 490.
[7] R. Troncoso, J. Zanelli, Phys. Rev. D 58 (1998) 101703;
R. Troncoso, J. Zanelli, Higher-dimensional gravity and local anti-de Sitter symmetry, hepth/9907109.
[8] M. Baados, Gravitons and gauge fields in 5D ChernSimons supergravity, hep-th/9911150.
[9] P. Horava, Phys. Rev. D 59 (1999) 046004.
[10] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, 1987.
[11] P.K. Townsend, Phys. Lett. B 350 (1995) 184, hep-th/9501068.
[12] C. Hull, P.K. Townsend, Nucl. Phys. B 348 (1995) 109.
[13] E. Witten, Nucl. Phys. B 443 (1995) 85.
9 Interesting insights can be found in Ref. [35]
242
[14] P.K. Townsend, Four lectures on M-theory, in: Proceedings of ICTP Summer School on High
Energy Physics and Cosmology, Trieste, June 1996, hep-th/9612121;
P.K. Townsend, M-theory from its superalgebra, hep-th/9712004.
[15] M.B. Green, Phys. Lett. B 223 (1989) 157.
[16] I. Kogan, Phys. Lett. B 231 (1989) 377.
[17] G. Moore, N. Seiberg, Phys. Lett. B 220 (1989) 422.
[18] P. Mora, H. Nishino, Phys. Lett. B 482 (2000) 222.
[19] J.A. Dixon, M.J. Duff, E. Sezgin, Phys. Lett. B 279 (1992) 265.
[20] R. Stora, Algebraic structure of chiral anomalies, in: H. Lehmann (Ed.), Recent Progress in
Gauge Theories, NATO ASI Series, Plenum, New York, 1984.
[21] B. Zumino, Chiral anomalies and differential geometry, in: B.S. De Witt, R. Stora (Eds.),
Relativity, Groups and Topology II, North-Holland, Amsterdam, 1984.
[22] J. Maes, R. Stora, B. Zumino, Commun. Math. Phys. 102 (1985) 157.
[23] L. Alvarez-Gaum, P. Ginsparg, Ann. Phys. 161 (1985) 423.
[24] R. Bertlmann, Anomalies in Quantum Field Theory, Oxford Univ. Press, Oxford, 1996.
[25] J.S. Bell, Bertlmanns socks and the nature of reality, in: Speakable and Unspeakable in
Quantum Mechanics, Cambridge Univ. Press, Cambridge, 1987.
[26] J.W. Van Holten, A. Van Proeyen, J. Math. Phys. 15 (1982) 3763.
[27] E. Bergshoeff, A. Van Proeyen, The many faces of OSp(1, 32), hep-th/0003261.
[28] R. Minasian, G. Moore, JHEP 9711 (1997) 002, hep-th/9710230.
[29] E. Witten, JHEP 9812 (1998) 019, hep-th/9810188.
[30] K. Olesen, R. Szabo, Constructing D-branes from K-theory, hep-th/9907140.
[31] C. Vafa, Nucl. Phys. B 469 (1996) 403.
[32] J.M. Izquierdo, P.K. Townsend, Nucl. Phys. B 414 (1994) 93, hep-th/9307050.
[33] M. Duff, J. Liu, R. Minasian, Nucl. Phys. B 452 (1995) 261, hep-th/9506126.
[34] Y. Cheung, Z. Yin, Nucl. Phys. B 517 (1998) 69, hep-th/9710206.
[35] E. Witten, Mod. Phys. Lett. A 5 (1990) 487.
[36] T. Banks, W. Fischler, S. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
[37] G. Moore, Finite in all directions, hep-th/9305139.
[38] P.C. West, Physical states and string symmetries, hep-th/9411029.
[39] C.M. Hull, Duality and strings, space and time, hep-th/9911080.
[40] R. Gebert, H. Nicolai, E10 for beginners, hep-th/9411188.

D-branes and their absorptivity in BornInfeld

theory
D.K. Park a,b , S.N. Tamaryan a,c , H.J.W. Mller-Kirsten a ,
Jian-zu Zhang a,d
a Department of Physics, University of Kaiserslautern, D-67653 Kaiserslautern, Germany
b Department of Physics, Kyungnam University, Masan, 631-701, South Korea
c Theory Department, Yerevan Physics Institute, Yerevan 36,375036, Armenia
d School of Science, East China University of Science and Technology, Shanghai 200237, PR China
Received 18 May 2000; accepted 8 November 2000
Abstract
Standard methods of nonlinear dynamics are used to investigate the stability of particles, branes
and D-branes of abelian BornInfeld theory. In particular the equation of small fluctuations about the
D-brane is derived and converted into a modified Mathieu equation and complementing earlier
low-energy investigations in the case of the dilaton-axion system studied in the high-energy
domain. Explicit expressions are derived for the S-matrix and absorption and reflection amplitudes
of the scalar fluctuation in the presence of the D-brane. The results confirm physical expectations and
numerical studies of others. With the derivation and use of the (hitherto practically unknown) high
energy expansion of the Floquet exponent our considerations also close a gap in earlier treatments of
the Mathieu equation. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
Recently BornInfeld gauge theory has attracted considerable interest as the bosonic
light-brane approximation or limit of superstring theory [1], and has turned out to be a
simple and transparent model in this context [2]. Branes, defined as extended objects in
spacetime, can be fundamental or solitonic. The connection of these branes with a U (1)
gauge field was motivated by the presence of this field in the massless part of the spectrum
of open strings, and by realising that branes with open strings attached to them which
satisfy Dirichlet boundary conditions, or more generally one brane attached to another,
can become classically stable, solitonic objects. It is for this reason that the dynamics
E-mail addresses: dkpark@genphys.kyungnam.ac.kr (D.K. Park), sayat@moon.yerphi.am (S.N. Tamaryan),
mueller1@physik.uni-kl.de (H.J.W. Mller-Kirsten), jzzhang@online.sh.cn (J.-Z. Zhang).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 6 3 - 5
244
D.K. Park et al. / Nuclear Physics B 594 (2001) 243271
of D-branes [3] in BornInfeld theory is being studied in detail and generalised [48].
Since, in general, a brane may or may not be a solitonic configuration or BPS state, the
exploration of this question deserves particular attention. It is often stated that a brane is
BPS in view of the vanishing of a fraction of the supersymmetry variation of the associated
gaugino field. However, since BPS states (as classically and topologically stable states) and
Bogomolnyi bounds have been studied in great detail in a host of other theories, and the
approach in these is practically standard, one would like to understand aspects of Born
Infeld particles in a similar way, also because it is not absolutely clear that D-branes are
solitons of string theory in precisely the same way as more familiar topological solitons in
field theory. Therefore our first intention in the following is to study BornInfeld particles
with standard methods of nonlinear dynamics in the simplest case of a flat spacetime.
We begin with the free BornInfeld particles, i.e., BIon and catenoid [4]. Using a scale
transformation argument [9] we show that these static configurations which differ from
ordinary solitons of nonlinear theories in requiring a special consideration of source terms
or boundary conditions (cf. also [10,11]) require the number of space dimensions p
to be larger than 2. We assume spherical symmetry and study the local stability of these
configurations by considering the second variational derivatives of their respective actions.
Our conditions for stability are (a) that the eigenfunctions of the corresponding operator be
square integrable, and (b) that the charge e be fixed, with angular fluctuations ignored. We
then consider the case of the scalar field corresponding to a single transverse coordinate
coupled to the gauge field (here only the electric component), i.e., the catenoid or brane
with associated open fundamental string. We distinguish between two types of arguments
in deriving the linearised fluctuation equation, and infer the stability of this stringy D-brane.
In Ref. [12] an explicit and detailed consideration of the Bogomolnyi bound in a special
model of BornInfeld theory has been given where the central charge of the supersymmetry
algebra plays the role of the topological or winding number of ordinary solitons.
Our second intention in the following is the explicit study of the small fluctuation
equation about the D3-brane in the high energy domain. This equation with singular
potential has the remarkable property of being convertible into a modified Mathieu
equation which depends only on one coupling parameter which is a product of energy
and electric charge. The S-matrix for scattering of the fluctuation off the brane can be
obtained in explicit form. The D3-brane is therefore one of the very rare examples allowing
a detailed study of its properties with explicit expressions for all relevant physical quantities
in both low and high energy domains. We therefore expect that also S-duality can be
uncovered and studied in this case (although we do not attempt this here). Various other
Dp-brane models have been discovered recently whose small fluctuation equations can be
reduced to modified Mathieu equations [1315] which have then been investigated mainly
by computational methods. For the AdS/CFT correspondence the logarithmic corrections
to the low energy absorption probability are of particular interest, since these permit a
direct relation to the discontinuity of the cut in the correlation function of the dual twodimensional quantum field theory. The first such logarithmic correlation to the absorption
probability was originally obtained in Refs. [1618] without resorting to the use of Mathieu
functions. Subsequently the authors of Ref. [13] considered the modified Mathieu equation
245
and used computational methods to generate explicit series expansions up to several orders
for the low energy absorption probability. In [19] a different choice of expansions was
considered to obtain leading expressions more easily. It is natural to supplement such
investigations by exploring also the high energy case, the first such consideration being
that of Ref. [15]. The analytical high energy results obtained in the following and the
complementary low energy results of Ref. [19] (we also demonstrate how the S-matrices
are related) are therefore directly applicable to these. Singular potentials have been studied
from time to time, and have mostly been discarded as pathological. It seems, however,
that their real significance lies in the context of curved spaces with black-hole type of
absorption [20].
Sections 2 and 3 deal with the BIon and the catenoid, Sections 4 and 5 with the
Bogomolnyi limit of the D3-brane and the derivation of the linearised fluctuation equation
about it. In Section 6 we consider this equation in detail in the high energy domain and
calculate the rate of absorption of partial waves of the fluctuation field by the brane. That
this absorption occurs is attributed to the singularity of the potential. The absorptivity part
of the paper may be looked at as the high energy complement to the low energy case
of Ref. [19] with the same expression of the S-matrix. All these calculations require a
matching of wave functions. In the low energy S-wave case simple considerations of
Bessel and Hankel functions suffice as was shown in Refs. [21]. The low energy limit
is, in fact, independent of the choice of matching point, as was shown recently [22]. Our
considerations here, however, are general.
2. The BIon
We consider first purely static cases and write the Lagrangian of the static BIon in p + 1
spacetime dimensions (cf. [4])
Z
q
p p/2
(1)
L = 1 1 (i )2 p e(r),
p =
L = d p xL,
(p/2)!
(i = 1, . . . , p) with the charge e held fixed by the constraint
Z
i
1
= 0.
di q
e+
p
1 (j )2
(2)
Eq. (1) is the Lagrangian one obtains from the world brane action of the pure Born
Infeld U (1) electromagnetic action reduced to the purely electric case with field
Ei = 0 Ai i A0 and no transverse coordinate. The field A is assumed to depend on
the world brane coordinates x , = 0, . . . , p. The static BIon equation of motion is

i
(3)
i p
= p e(r).
1 (i )2
In the special case p = 3 the classical SO(3) symmetric solution, called a BIon, is given by
246
Z
c (r) =
q
r
Zr
dx
1+
x4
e2
= c (0)
q
0
dx
1+

r5
' c (0) r +
10e2
r0
x4
e2
(4)
and c (0) = 14 B( 14 , 14 )e 2 = 1.854074677 e 2 , B being the Bernoulli function. It is easily

verified that this solution satisfies the constraint (2) for any value of r. Defining E = c
(sothat c = A0 with A0 (xi , t)/t = 0 in the static case), and defining D = L/E =
E/ 1 E2 we have (with F0i = Ei )
T00 = F0i
1
L
L=EDL=
1 + 4e(r).
F0i
1 E2
(5)
The energy Hc of the BIon (obtained by integration over R3 ) is then found to be finite, i.e.,
Z
3
(6)
Hc = dx T00 = 4 3.09112 e 2
and in p dimensions the total energy of the BIon scales correspondingly as ep/p1 .
The finiteness of the energy depends on the minus sign in (1) and so with (3) on the relation
q
r2
r2
1 (c0 )2 = c0 = q
(7)
4
e
e 1 + re2
for 0 6 r 6 . It may be noted that by defining D such that the left-hand side of Eq. (3)
is i Di , the singularity of the right-hand side is associated with D rather than with E which
is the decisive difference between Maxwell and BornInfeld electrodynamics. A similar
observation applies to the catenoid equation below. The energy of the BIon is seen to
be independent of its position which hints at the existence of some kind of collective
coordinate. However, exploring this point further is expected to be difficult since a moving
charge generates a magnetic field, and hence the electric field alone would not suffice.
We can use a scaling argument [9] to show that here finite energy configurations
require p to be larger than or equal to 3. Under a scale transformation x x 0 = x,
(x) (x) = (x), i (x) [i (x)] = i (x). The charge e defined by the
constraint (2) also changes under the scale transformation, i.e.,
Z
1
di i
q
.
(8)
e e = p2
p
1 2 (j )2
In particular for p = 3 and radial symmetry
(p=3)
1
r2
p
2
1 + r 4 /e2
(9)
and for arbitrary values of the r-dependence drops out only if the limit r is taken
in the evaluation of the integral. Then
(p=3)
r 1
(10)
.
e
But also e=1 = e for any r. If c is stable and 6= 0, the energy must be stationary for
= 1, i.e., (Hc /)=1 = 0. From this one finds that p > 3. Also ( 2 Hc /2 )=1 > 0
247
for p > 3. Eqs. (6) and (10) show that changing the scale changes both the charge and
the energy, i.e., if the charge were variable, one could lower the energy and hence the
configuration could be unstable. But fixing the charge (e.g., by a quantisation condition)
no instability is implied by the scaling condition.
We investigate the stability of the BIon further in the special and exemplary case of
p = 3 by considering the second functional variation of the static Lagrangian evaluated at
c (r). This can be written and simplified in the following form (ignoring total divergences
on the way)
Z
1
(11)
d 3 x A ,
2 L =
2
where
A = i
=
i c j c
1
i i
j
2
1/2
[1 (j c ) ]
[1 (k c )2 ]3/2
r2
1 d
d
.
2
r dr (1 c0 2 )3/2 dr
(12)
The operator A can also be written

6 02 d
1
1 d 2 d
r
.
A =
r c dr
(1 c0 2 )3/2 r 2 dr dr
(13)
The classical stability of c is therefore decided by the spectrum {n } of the small

fluctuation equation
r2
d
1 d
n = n n .
2
r dr (1 c0 2 )3/2 dr
(14)
We explore first the existence of a zero mode 0 , i.e., the case = 0. In this case
r2
(1 c0 2 )3/2
d
0 = C
dr
(15)
and so with 0 () = 0
C
0 = 3
e
Z
r
x4
dx = C
4
2
3/2
e
(1 + x /e )
Z
r
c
dx
.
= C
4
2
1/2
e
(1 + x /e )
(16)
The derivative of the classical configuration c with respect to the charge e indicates that
a perturbation along c /e around c leaves the static action invariant, i.e., c (e, r) and
c (e + e, r) have the same action since

c (e, r)
c (e + e, r)
.
=

(e)
e
e=0
We now show that the operator A does not possess negative eigenvalues, and that
therefore the BIon is a classically stable configuration. We let n be an eigenfunction
Then
of the operator A.
248
n = 4
d xn A
dr n
0
Z
= 4
0
r2
d
dn
dr (1 c0 2 )3/2 dr

r2
d
r2
dn
dn 2
n
dr
dr (1 c0 2 )3/2 dr
(1 c0 2 )3/2 dr
Z
= F + 4
dr
0
where F
:= F (r)|
0
r2
(1 c0 2 )3/2
dn
dr
2
,
(17)
and
F (r) = 4n

dn
r 4 3/2 dn
3 n
=
4e
.
1
+
r4
e2
dr
(1 c0 2 )3/2 dr
r2
(18)
The second term on the right-hand side of Eq. (17) is strictly positive. Hence nonpositive eigenvalues imply a nonvanishing negative value of F . From the condition
R 2 2
r
1+ ), > 0) it follows that r 2 d /dr 0 with
n
n
0 n r dr < (i.e., n ' 1/(r
r , so that
r
dn
0
F (r) ' 4r 2 n
dr
and F () = 0. Hence

n dn
.
(19)
F = F (0) ' 4e3 4
r dr
r0
As r 0 Eq. (14) becomes

n
1 d 1 d
n = 3 n .
r 2 dr r 4 dr
e
In the case of the zero mode
0 ' C1 + C2 r 5 ,
r 0.
(20)
(21)
In this case F = 20e3C1 C2 . For C1 C2 < 0 this is in full compliance with (16) and (4)
from which we obtain
1 1

B( 4 , 4 )
r5
3 .
0 ' C
8e1/2
5e
For n 6= 0 the small-r behaviour of n is

1 n 8
16
r +O r
n ' Cn 1
24 e3
(22)
so that

4
2 1 7
= 0.
F = Cn 4 r
3
r
r=0
Thus the conclusion is that for all eigenfunctions n

n i > 0.
hn |A|
(23)
249
This inequality excludes the possibility of the existence of negative eigenvalues. Hence the
BIon is in this sense classically stable.
3. The catenoid
The Lagrangian of the static catenoid in p + 1 spacetime dimensions and with a source
term is given by (cf. [4])
Z
q
p1
L = 1 1 + (i y)2 p r0 y(r),
(24)
L = d p x L,
where the signs have been chosen such that the energy is positive. Here the scalar field
y(xi , t) originates from gauge field components Aa for a = p + 1, . . . , (d 1), d =
dimension, which represent transverse displacements of the brane; here we consider
the case of only one such transverse coordinate, i.e., y, all d p 1 of which are
essentially KaluzaKlein remnants of the d = 10 dimensional N = 1 electrodynamics after
dimensional reduction to p + 1 dimensions. The EulerLagrange equation of the static
catenoid yc (static meaning y(xi , t)/t = 0) is given by

i yc
p1
(25)
= p r0 (r)
i p
1 + (yc0 )2
so that after integration
p
yc
p1
1 + (yc0 )2
= r0
r
rp
(26)
or for r > r0
r0 p1
yc0 = (+) p
,
r 2p2 r0 2p2
1 + yc0 2 = (+) q
r p1
2p2
(27)
r 2p2 r0
In the case of the catenoid without source term the right-hand side of Eq. (26) can be taken
to originate from a boundary condition such as (r/r p ) = 0. The domain r 6 r0 is the
nonsingular throat region (i.e., yc (r0 )) is finite). One may observe that the singularity on
the right-hand side of Eq. (25) is associated with the entire expression on the left whereas,
like i c in the BIon case, so now here yc is finite, i.e., the p-brane or single throat
solution is given by
yc (r) = (+)
r
p1
r0
dr q
.
2p2
r 2p2 r0
(28)
Thus y is double valued. The two possible signs can be taken to define a brane and its
antibrane. We show at the end of this section that the solution with the minus sign is the
minimum of the action and the solution with the plus sign the maximum of the action. This
function is finite at r = r0 and can be expressed in terms of elliptic integrals. For r0 = 1 it
is even simpler and has the value yc (1) = (+)
1 K( 1 )
2
2
where K is the complete elliptic
250
integral of the first kind. Plotted as a function of r, yc (r) is a monotonically decreasing

function starting from r0 ; pictured on a 2-dimensional space it looks like an inverted funnel
(i.e., the surface swept out by a catenary with boundaries at the openings), thus suggesting
the name catenoid. As pointed out in Ref. [2], the two possible signs of the square root
allow a smooth joining of one such funnel-shaped branch to an inverted one connected by
a throat of finite thickness, the resulting structure then representing a braneantibrane pair.
This braneantibrane pair is joined by the throat of finite thickness r0 and finite length. In
fact, we can rewrite Eq. (28) in terms of yc (x) = yc (r0 x), x = r/r0 , and for the special case
of p = 3 as
dx
Zx
dx
= (+)
()
4
4
x 1
x 1
x4 1
x
1
1

1
2
2
1 1
,
cn
,
= (+) K
2
x 2
2
yc (x) = (+)
dx
(29)
where x > 1 and we used formulae of Ref. [23]. Inverting this expression we obtain the
periodic function
1

2 +
2
() 2y,
.
(30)
x(y) = cn K
2
2
Plotting this expression with x as ordinate, one obtains the picture of a cross section
through a chain of periodically recurring funnel-shaped structures to the one side of the
throat, i.e., the series representing a series of braneantibrane pairs along the
abscissa.
Proceeding as in the above case of the static BIon and calculating the second variational
derivative we obtain
Z
1
2
b y,
(31)
d p x y B
L=
2
where for r > r0
i yc j yc
i
i i
j
2
1/2
[1 + (i yc ) ]
[1 + (i yc )2 ]3/2
1 d (r 4 r 4 )3/2 d
1 d
r2
d
0
=(+) 2
.
= 2
2
r dr (1 + yc0 )3/2 dr
r dr
r4
dr
b = i
B
b can also be written

The operator B

6 02 d
1 d 2 d
1
b
r
+ yc
.
B=
dr
(1 + yc0 2 )3/2 r 2 dr dr r
(32)
(33)
Since the gauge field components Aa , a = p + 1, . . . , d 1 (of which we retain only one),
are dynamical, the Lagrangian in the nonstatic case is
p
p1
(34)
L = 1 1 ( y)( y) p r0 (r)
251
and we can obtain the same condition of stability by considering the dynamical
fluctuation , i.e.,
y(t, x) = yc (r) + (t, x),
= (r)ei
and linearising the time-dependent EulerLagrange equation. The square integrable

perturbations (r) are the so-called L2 deformations of Ref. [4]. The classical stability
of yc is therefore decided by the spectrum {} of the small fluctuation equation

(r 4 r 4 )3/2 d
1 d
r2
d
1 d
0
=
= .
(35)
(+)
r 2 dr (1 + yc0 2 )3/2 dr
r 2 dr
r4
dr
We explore first the existence of a zero mode 0 , i.e., the case = 0. In this case
r2
(1 + yc0 2 )3/2
d
0 = C
dr
(36)
and so in the case p = 3 and r > r0

Z
0 = C
dx
r
x4
C yc
=
4
4
3/2
2r0 r0
(x r0 )
(37)
so that
0 = C
yc
.
r0 2
Here again the derivative of the classical configuration yc with respect to the parameter r0 2
is indicative of stationarity of the action in a shift of r02 .
b
We now demonstrate that the operator Bwith
the minus sign has no negative eigenvalues,
and that therefore the free catenoid is a classically stable configuration like the BIon for
fixed throat radius r0 . Then
Z
b = 4
d x B
dr
r0
r2
d
d
2
0
3/2
dr (1 + yc )
dr

Z
d
r2
d 2
4
dr
= 4
(1 + yc0 2 )3/2 dr r0
(1 + yc0 2 )3/2 dr
r2
r0

Z
(r 4 r04 )3/2 d +
(r 4 r0 2 )3/2 d 2
= (+) 4
()
4
dr
,
dr r0
dr
r4
r4
r0
(38)
where we used Eq. (27). The second term is always positive if the upper sign is chosen.
R
The first term vanishes at infinity with dr r 2 2 < , since
4
(r 4 r04 )3/2 d
r4
dr
' 4r 2
d
0.
dr
252
On the other hand, in the case r r0 , we have

(r 4 r04 )3/2 d
d
' 32 r0 (r r0 )3/2
.
4
r
dr
dr
As r r0 Eq. (35) becomes
d
8 d
= .
3/2 (r r0 )3/2
dr
r dr
4
(39)
In the case of the zero mode 0 with = 0 the considerations are analogous to those of
the BIon case and the sum of the two terms in Eq. (38) vanishes. In the case of 6= 0 we
therefore have

3/2
r r0
' C 1 r0
4
and
d
= C 2 lim r0 3/2 (r r0 ) = 0.
lim (r r0 )3/2
rr0
rr0
dr
8
This proves that for all eigenfunctions
b
h|B|i
> 0.
b has no negative eigenvalues, and the free throat is classically stable with fixed r0
Thus B
b with the plus sign has no
for the sign chosen as in Eq. (38). Obviously the operator B
positive eigenvalues, which means that we have the maximum of the action. Of course, if
b also changes.
we change r0 (and so consider a different theory), the expectation value of B
One should note that the free throat we discuss here is that with vanishing gauge field.
The double valuedness of the solution of Eq. (25) implies that if one solution is classically
stable, the other one is not. Thus a multi-throat solution constructed from these by matching
both solutions, if it exists, like the braneantibrane solution of Ref. [2], is expected to be
unstable in view of negative as well as positive eigenvalues, and is therefore neither a
maximum nor a minimum of the action. In fact, as argued in Ref. [4] (after Eq. (132))
equilibrium between these should not be possible. The reason for this is that a symmetrical
configuration, symmetrical about the plane x3 = 0 for instance, implies 3 y = 0 there.
Evaluating the stress tensor element T33 (even for vanishing gauge field), one obtains a
negative quantity which is interpreted as implying an attractive force between the brane and
its antibrane in this symmetrically constructed configuration. This is, in fact, the general
instability of this configuration discussed in Ref. [2].
4. Coupled fields: the D-brane in the Bogomolnyi limit

In the case of coupled fields and y (the former with source, the latter without), the
Lagrangian of the static case is (cf. [4])
Z
L =
d p x L, L = 1 Q p e(r),
1/2

.
Q = 1 (i )2 + (i y)2 + (i .i y)2 (i )2 (j y)2
(40)
253
From the first variation of L we obtain the coupled equations of the fields and y, i.e.,
from
Z

1
i ( y)i y + (y)2 i d p x
L =
i
Q
Z

1
i y + ( y)i ()2 i y d p x
+ yi
Q
Z
(41)
p e (r)d p x
(ignoring total divergences).
The source term of the electric field again suggests spherical symmetry. In deriving the
two coupled EulerLagrange equations one new constant c (apart from e) arises in the
integration of the catenoid equation, i.e.,

L
p1 L
= c.
= 0, r p1
r r
(r y)
(r y)
We have no source term of the y field because, as before, the appropriate effect is provided
by the boundary condition defining the width of the throat. The two equations with
spherical symmetry are found to be
e
0
= p1 ,
0
2
0
2
1/2
[1 ( ) + (y ) ]
r
y 0
c
= p1
0
2
0
2
1/2
[1 ( ) + (y ) ]
r
(42)
so that
0 e 1
= .
y0
c a
(43)
Then
( 0 )2 =
1
r 2(p1)
e2
+ 1 a2
(y 0 )2 =
a2
r 2(p1)
e2
+ 1 a2
(44)
Thus the family of solutions can be parametrised in terms of the single parameter a as
already pointed out in Ref. [5]. This parameter is seen to interpolate between the two types
of static solutions. The solution y of (35) for various values of a 2 is now the p-brane, i.e.,
+
Z
dr q
y(r) =() ae
r
2(p1)
1
2(p1)
r 2(p1) r0
(45)
= e2 (a 2 1) and for the solution to make sense we must have a 2 > 1. If ae

where r0
in Eq. (36) is replaced by ae, the expression represents the corresponding antibrane.
Taking e2 0, a 2 e2 const the electric field is eliminated and we regain the free
catenoid solution. In approaching the limit a 2 1 the width of the throat becomes
infinitesimal with nonvanishing electric field and the configuration can then be considered
to be a fundamental string, as argued in Ref. [2]. We distinguish between three cases:
254
Z
|a| < 1:
p
r
|a| > 1:
1 a2
Z
p
=e
r
dx
+ x 4 /e2
dx
x4
r0
y=a
p
r
dx
1 a 2 + x 4 /e2
Z
p
y = ae
e
y= .
r
e
= ,
r
a = 1:
dx
x4
r0 4
,
(46)
We see that for = 1 Eq. (34) becomes the first order Bogomolnyi equation or linearised
field equation for y (as in Ref. [2])
y
= 0,
(47)
F0r
r
where F0r = Ec is the static electric field. This is the same equation as that obtained from
the vanishing of the supersymmetry variation of the gaugino field for half the number
of 16 supersymmetries (for d = 10 and p = 3) + , of for which = 0, i.e.,
a2
+ = 0,
6= 0,
where as discussed in the literature [24] is the constant spinor of the supersymmetry
variation and are its chiral components. Thus a 2 = 1 implies BPS configurations,
whereas those with a 2 6= 1 are non-BPS. Taking a 2 = 0 in Eq. (36) we regain the BIon
configuration as a local minimum of the energy whereas for vanishing electric field one
expects a local maximum, i.e., a sphaleron configuration (as pointed out in [2]).
Next we investigate the second variation of the static L with spherical symmetry. We set
Z

1 b
b + Ly
+ y L d p x.
(48)
M + y Ny
2 L =
2
Again ignoring total divergences one finds
02
b = 1 d r2 1 + y d ,
M
r 2 dr
Q3 dr
02
b = 1 d r2 1 d ,
N
r 2 dr
Q3 dr
1 d 0y 0 d
L = 2 r 2 3
r dr Q dr
with L = L . We can now rewrite 2 L as

Z
1
2
3
b
d x (, y)H
,
L=
y
2
where
b=
H
M
L
L
N

=
1 y 0
y 00
(50)
1 d 2 d
r h
r 2 dr
dr
and
1
h= 3
Q
(49)
y 00
2
1 0
(51)
!
,
b,
b = H
H
(52)
with
1 + 0
y 00
h1 = Q
y 00
2
1 + y0
!
,
det h =
1
.
Q4
255
(53)
The small fluctuation equation therefore becomes

b = 1 d r 2 h d = .
H
dr
r 2 dr
(54)
Again we first explore the existence of a zero mode 0 with

2 d
,
r h 0 =
dr
(55)
where and are constants. Setting

0
0 =
y0
and evaluating 0 for the solutions of Eq. (46) we obtain with
Z
=
0 (x) dx
3
the relation
0 =
e

( + a)
e

1
.
a
b of Eq. (42) becomes

In the BPS limit with y 0 = 0 = Ec , Q = 1, the operator H

1 d 2 1 Ec2
d
Ec2
b
.
H= 2 r
2
2
E
1
E
r dr
dr
c
c
(56)
(57)
Setting

s =

= (x)

1
1
for an arbitrary function (x) we have

1 d 2 0 1 Ec2
1 d 2 0 1
Ec2
1
b
= 2 r
.
H s = 2 r
1
1
Ec2
1 Ec2
r dr
r dr
(58)
bs = 0 implying 2 L = 0 or L constant in a specific

Thus for arbitrary (x), we have s H
direction about the BPS configuration. This behaviour may be interpreted as indicative of a
local symmetry, in this case of supersymmetry, and so of the cancellation of fermionic and
bosonic contributions in the one loop approximation. Here, of course, we have no fermionic
contributions and consequently those of the two bosonic fields have opposite signs.
256
5. Fluctuations about the D-brane

In the following we distinguish clearly between two different types of fluctuations. We
consider the above BPS solution for the string attached to the 3-brane as background and
consider first a scalar field propagating in a direction along the string and perpendicular
to the brane and its antibrane. The linearised equation of small fluctuations about this
background is obtained from the second variational derivative of the action which is the
standard procedure and we therefore consider this first (cf. also [25]). Our treatment here
is somewhat different (see below) from that in Refs. [25]. The resulting fluctuation equation
has also been given in Ref. [2]. It is necessary to return to the fully time-dependent
version, i.e.,
Z
p

1
(59)
d p+1 x 1 det( + F ) p e(r) ,
S=
p
(2) gs
where in 3 + 1 dimensions F = F (x0 , x1 , x2 , x3 ). In the electrostatic case with only
one scalar field y we have A = (A0 , A1 , A2 , A3 , y, 0, 0, 0, 0, 0), F0i = Ei and F4 = y
for i = 1, 2, 3 and = 0, 1, 2, 3. Then

1
E1
E2
E3
0 y

E1
1
0
0
1 y

(60)
det( + F ) = E2
0
1
0
2 y
E
0
0
1
3 y
3

y y y y
1
0
1
2
3
and so

det( + F ) = 1 E2 1 + y 2 (E y)2 + (0 y)2 .
(61)
We consider first the Lagrangian density (remembering that the relevant fields are A0 , Ai
and y)

1/2

.
(62)
L = 1 Q,
Q = 1 E2 1 + y 2 + (E y)2 y 2
The equations of the static BIon and the static catenoid discussed above follow again from
the first variations

1
L
Ei 1 + y 2 i y(E y) ,
=
Ei
Q

1
L
= i y 1 E2 + Ei (E y) ,
i y
Q
1
L
= .
(63)
0 y
Q
In the BPS background given by
i y = Ei ,
one finds
E2 = (y)2 = E y =
e2
Ec2 yc2 ,
r4
Q=1
(64)

2L
= 1 + Ec2 ij ,
Ei Ej
257

2L
= 1 Ec2 ij ,
i yj y
2L
= Ec2 ij ,
Ei j y
2L
= 1.
y 2
(65)
This enables us to write (ignoring again total divergences in shifting derivatives)

2 L = (A0 , Ai , y)
i (1 + Ec2 )i
0 (1 + Ec2 )i
i Ec2 i
i (1 + Ec2 )0
0 (1 + Ec2 )0
+i Ec2 0
i Ec2 i
A0
Ai .
0 Ec2 i
2
y
i (1 Ec )i 0 0
(66)
In the linear approximation the EulerLagrange equations of the fluctuations y ,

Ei = 0 Ai i A0 are therefore given by the following set of three equations

d2
+ i 1 Ec2 i + i Ec2 (0 Ai i A0 ) = 0,
2
dt

d
d
1 + Ec2 (0 Ai i A0 ) Ec2 i = 0,
dt
dt

i 1 + Ec2 (0 Ai i A0 ) i Ec2 i = 0.
(67)
(68)
(69)
The last of these three equations can be seen to be a constraint by applying /t and using
the second equation. Substituting from the last
i Ec2 (0 Ai i A0 ) = i Ec2 i i (0 Ai i A0 )
into the first equation we obtain
d2
+ i (0 Ai i A0 ) = 0.
dt 2
The second of the three equations can be written in the form

1 + Ec2 (0 Ai i A0 ) Ec2 i = 1 + Ec2 Ci (r),
(70)
(71)
where C(r) is an arbitrary function. Dividing Eq. (60) by (1 + Ec2 ) and taking the
derivative i , we obtain
i (0 Ai i A0 ) = i
=
Ec2
i + i Ci
1 + Ec2
E2
2Ec Ec0 xi
i + i Ci .
+
2
1 + Ec
(1 + Ec2 )2 r
(72)
Replacing on the right hand side Ec2 i by the expression in Eq. (60) this becomes
i (0 Ai i A0 ) =

xi
Ec2
2Ec0
(0 Ai i A0 ) Ci + i Ci .
+
2
2
1 + Ec
Ec (1 + Ec ) r
(73)
258
Choosing as gauge fixing condition the relation

xi
2Ec0
(0 Ai i A0 ) Ci + i Ci = 0
2
Ec (1 + Ec ) r
one obtains the following fluctuation equation for
(1 + Ec2 )
d 2
+ = 0.
dt 2
(74)
All the relations from (60) to (74) describe perturbations along the string and perpendicular
to the brane. Eq. (74) cannot be considered independently of the others as is apparent
from the linkage of the fields in the above equations. Thus if one wants to determine
the radiation of the string between the brane and the antibrane, one must connect the
asymptotic behaviour of the field with that of the vector field A .
However, an equation like (74) is also obtained if one evaluates the determinant in the
BornInfeld Lagrangian at the BPS background and with an additional time-dependent
scalar , representing the fluctuation field along a new spatial direction (cf. also Ref. [2]).
In this case this new scalar field in the D-brane background has no relevance to the string
radiation, and we have

1
E1
E2
E3
0 0

E
1
0
0
E1 1
1

0
1
0
E2 2
E2
(75)
det( + F )|BP S, =

E3
0
0
1
E3 3 y

0
1
0
E1 E2 E3

0 1 2 3 0
1
and so

det( + F )|BPS, = 1 + Ec 2 (0 )2 (i )2 1.
Thus the Lagrangian density becomes
q
L = 1 1 + ()2 (1 + Ec 2 )(0 )2 .
(76)
(77)
By expanding the square root and retaining only the lowest order terms, we again obtain
a fluctuation equation like (65), but this time for with no relevance to radiation of the
string. This is equivalent to studying the scattering of the scalar off a corresponding
supergravity background.
6. Absorption of scalar in background of D3-brane

We now consider the equation of small fluctuations, i.e., Eq. (74), in more detail. The
fluctuation (t, x) represents a scalar field that impinges on the brane which reflects part of
it and absorbs part of it depending on the energy of the field. The absorption results from
and takes place into the singularity of the real potential which corresponds to the black
259
hole with zero event horizon in the analogous case of the dilaton-axion system of, e.g.,
Ref. [19]. This absorption is a classical phenomenon. We therefore consider the equation

e2
2
(78)
r + 1 + 4 = 0.
r
One can argue that the absorption is a consequence of the nonhermiticity of the potential.
The radial part of this equation is with = r 1 Ylm and angular momentum l

l(l + 1)
e2
d 2
2
+
1
+
= 0.
(79)
dr 2
r2
r4
This equation is a radial Schrdinger equation for an attractive singular potential r 4 but
depends only on the single coupling parameter = e2 for constant positive Schrdinger
energy, i.e., for S-waves the equation is with x = r simply

2
2
d
+
1
+
= 0.
(80)
dx 2
x4
In the following we consider the general case, i.e., l 6= 0. The simplified case of the
singular potential replaced by an effective deltafunction potential has been considered
in Refs. [2] and [25]. The solutions and properties of such equations have been studied
in detail in the literature, in both the small- and large- domains and with inclusion of
the centrifugal term l(l + 1)/r 2 in Eq. (79) for the calculation of Regge trajectories
l n (2 ) [2629]. A recent investigation which attempts to treat arbitrary power
singular potentials is Ref. [30]. Eq.(79) describes waves above the singular potential well.
With the substitutions
1
1
(81)
(r) = r 2 (r), r = e ez , h2 = e2 , a = l + ,
2
the equation becomes the modified Mathieu equation

d 2 2
+ 2h cosh 2z a 2 = 0
2
dz
(82)
which has been studied in detail in the literature [31] (though some properties, such
as large-h asymptotic expansions of Fourier coefficients, have even now not yet been
published). Here we study the S-matrix in the domain of finite values of angular momentum
l and h2 6= 0, i.e., in the domain of h2 large. Relevant solutions and matching conditions for
this case have been developed in [32] and [33]. We follow the latter of these references here
since this makes full use of the symmetries of the solutions. Moreover we can determine
also the Floquet exponent which Ref. [32] leaves undetermined and only remarks that
the notion that this is a known function of (our) a 2 and h2 is partly a convenient fiction.
For convenience we set in Eq. (82) as in Refs. [33,34]
(q, h)
,
(83)
8
where q is a parameter to be determined as the solution of this equation and /8 is
the remainder of the large-h asymptotic expansion (83), the various terms of which are
a 2 = 2h2 + 2hq +
260
determined concurrently with corresponding iteration contributions of the solutions of

the equation and are known explicitly to many orders [34]. Then setting in Eq. (82)
(q, h; z) = A(q, h; z) exp[2hi sinh z]
(84)
we obtain an equation for A which can be written

1
d 2A
dA 1
+ (sinh z iq)A =
A 2 .
cosh z
dz
2
4hi 8
dz
(85)
We let Aq (z) be the solution of this equation when the right hand side is replaced by zero
(i.e., in the limit h ). Then one finds easily

1 + i sinh z q/4 z z/2 iq/4
1
2e
e
.
(86)
Aq (z) =
cosh z 1 i sinh z
Correspondingly the various solutions are
exp(ihez ) iq/4
e
,
cosh z
z exp(ihe |z| ) iq/4
e
.
(q, h; z) = Aq (z) exp[2hi sinh z] '
cosh z
z
(q, h; z) = Aq (z) exp[2hi sinh z] '
(87)
We make the important observation that given one solution (q, h; z) we can obtain
the linearly independent one either as (q, h; z) or as (q, h; z), the expression (83) remaining unchanged. With the solutions as they stand, of course (q, h; z) =
(q, h; z). Below we require solutions H e(i) (z), i = 1, 2, 3, 4, with some specific
asymptotic behaviour. We define these in terms of the function
Ke(q, h; z) :=
exp[iq/4]
Aq (z) exp[2hi sinh z] k(q, h)(q, h; z).
2ih
(88)
Since this function differs from a solution by a factor k(q, h), it is still a solution but not
with the symmetry property (q, h; z) = (q, h; z). Instead, after performing this
cycle of replacements the function picks up a factor, i.e.,
Ke(q, h; z) =
k(q, h)
Ke(q, h; z),
k(q, h)
k(q, h)
= ei 2 (q+1)
k(q, h)
(89)
in leading order. One can easily show that the quantity 80 of Ref. [32] is related to q
by 80 = iq/2 + O(1/ h). In Fig. 1 we show the behaviour of q as a function of h. In
order to be able to obtain the S-matrix, we have to match a solution valid at z = to a
combination of solutions valid at z = . This is achieved with the help of Floquet solutions
(1)
(z, h2 ) of
Me (z, h2 ). As such, these satisfy the same circuit relation as a solution M
Eq. (82) expanded in a series of Bessel functions, i.e., we have the proportionality

(90)
Me z, h2 = M(1) z, h2 , h2 = Me 0, h2 M(1) 0, h2 .
The functions Me (z, h2 ) are expansions of the modified (hence M instead of m)
Mathieu equation in terms of exponentials (hence e) which are uniformly convergent in
any finite domain of z. For large values of the argument 2h cosh z of the Bessel functions
of the modified Mathieu function M(1)(z, h2 ) can be reexpressed in terms of Hankel
261
Fig. 1. The function q(h) plotted versus h, which, of course, is valid only away from h = 0. The plot
should be compared with graphs in Ref. [32] where a similar but less convenient quantity is used.
functions. With the dominant terms of these we can obtain the large 2h cosh z asymptotic
behaviour of the Floquet function Me (z, h2 ), i.e., for |z|

cos(2h cosh z /2 /4)
,
Me z, h2 ' exp[i /2]
2h cosh z
(91)
where (with Me (z, h2 ) = Me (z, h2 ))

exp[i ] =

(h2 )
(1)
= M 0, h2 M(1) 0, h2 .
(h2 )
(92)
We now define the following set of solutions of Eq. (82) by setting

H e(2)(z, q, h) = Ke(q, h, z), H e(1)(z, q, h) = H e(2)(z, q, h),
H e(3)(z, q, h) = H e(1)(z, q, h), H e(4)(z, q, h) = H e(2)(z, q, h).
(93)
The solutions so defined have the following asymptotic behaviour (where (z) =
(2h cosh z)1/2 ):

H e(1)(z, q, h) = (z) exp ihez i 4 , <z 0,
r exp[ir i/4]
,
r

H e(2)(z, q, h) = (z) exp ihez + i 4 , <z 0,
r exp[ir + i/4]
r

H e(3)(z, q, h) = (z) exp ihe| z| i 4 , <z 0,

H e(4)(z, q, h) = (z) exp ihe| z| + i 4 , <z 0,
r0
r 1/2 exp[ie/r + i 4 ]
.
(e)1/2
(94)
262
For the following reasons we choose the latter, i.e., the solution H e(4)(z, q, h), as our
solution at r = 0. The time-dependent wave function with this asymptotic behaviour is
proportional to
eit +ie/r+i/4 .
Fixing the wave front by setting = t + e/r + /4 = const and considering the
propagation of this wave front, we have
e
r=
+ t /4
so that when t : r 0. This means that the origin of coordinates acts as a sink.
With Eq. (91) we therefore equate in the domain <z 0:
i
exp[i /2]
2

exp i 2 H e(1)(z, q, h) exp i 2 H e(2)(z, q, h) ,
i
Me (z, h2 ) = exp[i /2]
2

Me (z, h2 ) =
(95)
where the second relation was obtained by changing the sign on in the first. Changing
the sign of z we obtain in the domain <z 0:

Me z, h2 = Me z, h2

i
= exp i /2
2


i
Me z, h2 = exp[i /2]
2

exp i 2 H e(3)(z, q, h) exp i 2 H e(4)(z, q, h) . (96)
These relations are now valid over the entire range of z. Substituting Eqs. (96) into
Eqs. (95) and eliminating H e(3) we obtain
sin H e(4)(z, q, h) = sin ( + )H e(1)(z, q, h) sin H e(2)(z, q, h). (97)
In a similar way one obtains the relations
sin H e(2)(z, q, h) = sin ( + )H e(3)(z, q, h) sin H e(4)(z, q, h),
sin H e(1)(z, q, h) = sin H e(3)(z, q, h) + sin ( )H e(4)(z, q, h).
(98)
From Eqs. (89) and (93) we see that H e(2)(z, q, h) is proportional to H e(3)(z, q, h).
From (89) and (98) we see that the proportionality factor is given by

sin ( + )
.
exp i 2 (q + 1) =
sin
(99)
263
From Eq. (97) we can now deduce the S-matrix Sl e2il , where l is the phase shift. The
latter is defined by the following large r behaviour of the solution chosen at r = 0, which
in our case is the solution H e(4). Thus here the S-matrix is defined by (using (97))
r 1/2 eie/r+i/4
(e)1/2

i/4 sin (1)l
r
l sin ( + )e
i/2 ir
l ir
e
= (1)
e (1) e
sin ( + )
r

ei eil/2 ir
Sl e (1)l eir .
2i r
sin
(100)
From this we deduce that

sin
sin i(l 1 q)
2 .
ei(l+1/2) =
e
(101)
Sl =
sin ( + )
sin
We can see the relation of this high-energy (i.e., large |h|) expression of the S-matrix to
the low-energy expression of Ref. [31] by recalling that R of the latter is here exp(i ).
With this identification we can write Sl
Sl =
1
R
ei(l+1/2),
i
(Rei e R )
R ei ,
which agrees with the S-matrix of Ref. [19], i.e., we thus obtained the same exact
expression of the S-matrix here with our large-h considerations. In fact, comparison with
the considerations given there allows one to write down the reflection and transmission
amplitudes Ar and At as Ar = 2i sin and At = sin , respectively. We thus have one
and the same expression for the S-matrix for the two asymptotic regions, i.e., in the low
energy and high energy domains. One should therefore be able to proceed directly to the
large-h case from the exact S-matrix derived in the small-h domain. This is an interesting
calculation which we do not attempt to go into here. We only indicate in Appendix A the
first necessary step in that direction, i.e., the derivation of large-h asymptotic expansions
for the Fourier coefficients of Mathieu functions. In this connection we make the following
two observations. (1) Eq. (80) is invariant under interchanges x /x, x which
means that the inner or string region is equivalent or dual to the outer or brane region.
(2) Due to SL(2, R) invariance of the D3-brane its action is mapped into that of an
equivalent D3-brane by S-duality transformations [35] or, in other words, weakstrong
duality takes the D3-brane into itself [3]. It would be interesting to find some connection
between these properties, or equivalently the symmetry which the SL(2, R) invariance of
the D3-brane action imposes on the S-matrix.
The quantity is now to be determined from Eq. (99). One finds
p

(102)
sin = sin iei 2 q cos 1 + eiq sin2 .
It remains to determine the Floquet exponent in terms of q and h. In Appendix B we
derive the appropriate large-h behaviour of for the case of the periodic Mathieu equation.
Replacing there the eigenvalue by a = (l + 12 )2 and observing that h2 remains h2 , the
appropriate relation for our considerations is
264
2 +1)

1 + 3(q64h
1
e4h
+
O
cos + 1 =

(8h)q/2 34 q4 14 q4
h2

2 +1)

1 + 3(q64h
q+1
cos q
1
e4h
2
2
+
O
.
(103)
=
q/2
q/2
(8h)
h
22
Since the right-hand side grows exponentially with increasing h the Floquet exponent
must have a large imaginary part. Since the right-hand side is real, the real part of must
be an integer. Using Stirlings formula we can approximate the equation for q ' h (i.e.,
irrespective of what the value of l is) as
r
r

h/2
h
h
h
h 1.8h
7
cos
e /32
e
.
(104)
'
cos
cos + 1 =
2
2
2
2
From Eq. (101) and Eq. (102) we obtain

p

Sl = ieil cos cos2 1 eiq .
(105)
From this we obtain the absorptivity A(l, h) of the lth partial wave, i.e.,
A(l, h) := 1 |Sl |2
(106)
with near asymptotic behaviour

A(l, h) ' 1
2(16h)q

2 .
e8h q+1
2
(107)
In Figs. 2, 3 and 4 we plot A(l, h) as a function of h. One can clearly see the expected
asymptotic approach to unity and in Fig. 2 some sign of rapidly damped oscillations. This
behaviour agrees with that obtained on general grounds in Ref. [15]. We also observe
that in the high energy limit logarithmic contributions as in the low energy expansions,
Fig. 2. The absorptivity A(l, h) for l = 0.
265
discovered originally in [1618], and typical of the low energy expansions of [13] and [19],
do not arise. Of course, these plots do not extend down to h = 0, since our asymptotic
solutions become meaningless in that domain. The continuation to h = 0 can be obtained,
however, from the small-h expansions such as those derived in Refs. [13] and [19]. Thus
the absorptivity A(l, h) is known over the entire range of h. We observe that Sl = 0 for
q = 1, 3, 5, . . ., with [(l + 1/2)2 + 2h2 ]/2h ' 1, 3, 5, . . . . Only in the plot for l = 2 is h
sufficiently large to hint at these zeros.
266
7. Concluding remarks
Branes, whether fundamental or solitonic, play an important role in all aspects of

string theory. In particular D-branes have been looked at as string-theory analogues of
solitons of simple field theories, and some of their important properties such as charges
are well understood. Our first objective in the above was to investigate properties of
solitonic objects of BornInfeld theory in ways familiar from field theory, in particular
their classical stability. It was shown that the BIon and the catenoid as distinct, i.e.,
free objects, are stable configurations whereas the braneantibrane system is unstable;
we also recognized the zero modes associated with these and their significance. We then
considered the D3-brane of BornInfeld theory and recognised this as a BPS state that
preserves half of the number of supersymmetries as discussed in detail already in [2].
The equation of small fluctuations about this D3-brane was derived and shown to be
convertible into a modified Mathieu equation. The low energy solutions of this equation,
the S-matrix for scattering of a massless scalar off the brane and the corresponding
absorption and reflection amplitudes are similar to those for the dilaton-axion system
investigated first in Refs. [1618], where the important logarithmic contributions were
discovered, and then investigated in extensive detail in [13] and [19]. Here we performed
the high energy calculations which complement in particular those of [19], thus completing
the investigation of the modified Mathieu equation for the purpose of obtaining absorption
cross sections for all such cases. In particular the behaviour of the important Floquet
exponent involved in these calculations (in general a complex quantity) is now fully
understood, the Floquet exponent being vital in the evaluation of the S-matrix which
we derive and the calculation of the corresponding absorption amplitudes and cross
sections. According to our findings the high energy limit of the absorption cross
section does not involve logarithmic contributions, quite contrary to the low energy
limit.
The high energy case considered here is not only of interest in the immediate context
of the BornInfeld model considered here, but together with the low-energy case also of
considerable interest in connection with the concept of duality which links weak coupling
with strong coupling. The D3-brane with Schrdinger potential coupling e2 , which
links the gauge field charge e with energy of the incoming scalar field is presumably
the ideal example for the investigation of this property. Investigations elucidating this
aspect are of considerable interest. We also envisage interest in the study of nonBPS configurations, including sphalerons and bounces, as a matter of principle, i.e.,
even if the effect of these is not of dominant importance. Finally we remark that it
should be possible to proceed directly from the S-matrix derived in Ref. [19] to the
high-energy case here by using appropriate asymptotic expansions for the cylindrical
functions and expansion coefficients involved (for the latter such expansions do not
seem to have been given in the published literature so far, but we comment on these in
Appendix A).
267
Acknowledgements
D.K.P, S.T. and J.-Z.Z. are indebted to the Deutsche Forschungsgemeinschaft (Germany)
for financial support of visits to Kaiserslautern; the work of J.-Z.Z. has also been supported
in part by the National Natural Science Foundation of China under Grant No. 19674014
and the Shanghai Education Development Foundation.
Appendix A
In Ref. [19] on the absorptivity of the D3-brane of the dilaton-axion system it was shown
that the S-matrix for scattering of a massless scalar field off the brane is given by

R R1 eil
,
(A.1)
S=
i
Rei e R
where
(1)
R=
M (0, h)
M(1) (0, h)
M(1)(z, h) being the modified Mathieu function expanded in terms of Bessel functions,
i.e.,
Me (0, h)M(1)(z, h) =
c2r
h2 J+2r (2h cosh z)
r=
(an expansion with better convergence to use in practice is one in terms of products of
Bessel functions as shown in Ref. [19]) where Me (z, h) is the Fourier or Floquet solution
(h2 ) have only been

of the Mathieu equation. In the published literature the coefficients c2r
2
considered as power series in rising powers of h , and consequently were used in Ref. [19]
in the small h2 or low energy domain. It would be very interesting to make the transition
to the large-h2 or high energy case directly from this expression by developing large-h2
asymptotic expansions of the Mathieu function Fourier coefficients c2r

(h2 ) (for the Bessel
functions the corresponding expansions are known). We know of no publication where such
expansions have been given, but one of us (M.-K.) remembers from private communication
with the author of Ref. [37] that these Stokes-type asymptotic expansions can indeed be
obtained. One writes the recurrence relation of the coefficients (cf. [31], p. 106)
[ ( + 2)2 ]
c2
(A.2)
h2
(the Mathieu equation being y 00 + ( 2h2 cos 2x)y = 0). For |h2 | this implies
c2+2 + c22 =
c2+2 i (2+2)/2.
Setting
c2+2 b+1 ,
we have
b = i
268
[ ( + )2 ]
b ,
h2
i[ ( + )2 ]
.
+1 + 1 =
h2
From this we deduce that the next approximation to c2+2 is obtained from
b+1 + b1 =
r = 1 +
r

i X
( + )2 .
2
h
(A.3)
(A.4)
=0
The sums on the right-hand side can be evaluated. E.g.,

r
X
=0
1
2 = 12 + 22 + 32 + + r 2 = r(r + 1)(2r + 1)
6
so that one obtains

r(r 1)
i r(r + 1)(2r + 1)
2
+ 2
+ .
r = 1 + 2
h
6
2
(A.5)
Proceeding in this way one can indeed obtain the desired asymptotic expansion of the
coefficients c2 . (In fact the asymptotic expansion of the Bessel function similar to that
of a linear combination of Hankel functions can be obtained from its recurrence relation
in a very similar way).
Appendix B
For the determination of the large-h behaviour of the Floquet exponent we make use
of results of Ref. [34]. A fundamental pair yI , yII of, respectively, even and odd solutions
of the original periodic Mathieu equation with eigenvalue defined by
yI (z) = yI (z),
yII (z) = yII (z)
can be chosen to satisfy the following boundary conditions (cf., e.g., [31], pp. 99, 100)
yI (0) = 1,
yII (0) = 0,
yI 0 (0) = 0,
yII 0 (0) = 1.
From its original defining property the Floquet exponent can then be shown to be given
by (cf. [31], p. 101)

(B.1)
cos = yI ; , h2
so that (cf. [31], p. 100)

cos + 1 = 2yI /2; , h2 yII 0 /2; , h2 .
(B.2)
The solutions yI (z), yII (z) can be identified with the large-h solutions ce, se of Ref. [34]
(there Eqs. (64)) in terms of functions A(z), A(z)

as in Eq. (84) above with normalization
0
constants N0 , N0 , i.e., in leading order
ce(o) = 2N0 A(0),
se0 (0) = 4hN0 0 A(0)
269
from which we deduce in leading order for large |h| that

N0 = 23/2,
N0 0 = 25/2/ h.
Eqs. (65) of Ref. [34] give the large-h expansions of yI (/2; , h2 ) and yII 0 (/2; , h2 ).
Inserting these multiplied by the appropriate normalization constants into Eq. (B.2) and
retaining the dominant terms for large |h| we obtain
2 +1)

1 + 3(q64h
1
e4h
(B.3)
+
O
cos + 1 =

q
q
q/2
3
1
(8h)
h2
44 44
in agreement with a result cited in Ref. [31] (p. 210) from [36] with logarithmic corrections.
We, however, see no such logarithmic terms in the simpler formulation of Ref. [34]. The
relation (B.3) we rediscovered here has practically been unknown, largely in view of the
difficulty to extract it from the complicated considerations of Ref. [36]. Our derivation
above is simple and closes a difficult gap which the author of Ref. [32] commented upon
with the words: It is not likely at this stage that an analytic relation will ever be found
connecting (our) and to (our) a 2 and h2 . Our search of later literature did not uncover
other derivations. The main source summarizing more recent developments in the field of
the Mathieu equation in Ref. [38].
References
[1] G.W. Gibbons, in: C. Teitelboim, J. Zanelli (Eds.), Lecture at CECS, Santiago, Chile, August
1997, hep-th/9801106.
[2] C.G. Callan, J.M. Maldacena, Brane dynamics from the BornInfeld action, Nucl. Phys. B 513
(1998) 198, hep-th/9708147.
[3] J. Polchinski, Tasi lectures on D-branes, hep-th/9611050.
[4] G.W. Gibbons, BornInfeld particles and Dirichlet p-branes, Nucl. Phys. B 514 (1998) 603,
hep-th/9709027.
[5] A. Hashimoto, The shape of branes pulled by strings, Phys. Rev. D 57 (1998) 6441, hepth/9711097.
[6] S. Lee, A. Peet, L. Thorlacius, Brane waves and strings, Nucl. Phys. B 514 (1998) 161, hepth/9710097.
[7] D. Brecher, BPS States of the non-abelian BornInfeld action, Phys. Lett. B 442 (1998) 117,
hep-th/9804180.
[8] S.B. Giddings, Scattering ripples from branes, Phys. Rev. D 55 (1997) 6367, hep-th/9612022.
[9] H.J.W. Mller-Kirsten, J.-Z. Zhang, D.H. Tchrakian, Int. J. Mod. Phys. A 5 (1990) 1319;
See also P. Goddard, D.I. Olive, Rep. Progr. Physics 41 (1978) 1357.
[10] D. Chruscinski, Point charge in BornInfeld electrodynamics, Phys. Lett. A 240 (1998) 8, hepth/9712161.
[11] A.A. Chernitskii, Nonlinear electrodynamics with singularities (Modernized BornInfeld
electrodynamics), Helv. Phys. Acta 71 (1998) 274, hep-th/9705075, v8.
[12] S. Gonorazky, C. Nunez, F.A. Schaposnik, G. Silva, Bogomolnyi bounds and the supersymmetric BornInfeld theory, Nucl. Phys. B 531 (1998) 168, hep-th/9805054.
[13] S.S. Gubser, A. Hashimoto, Exact absorption probabilities for the D3-brane, Comm. Math.
Phys. 208 (1999) 325, hep-th/9805140.
[14] M. Cvetic, H. L, C.N. Pope, T.A. Tran, Exact absorption probability in the extremal sixdimensional dyonic string background, Phys. Rev. D 59 (1999) 126002, hep-th/0001002.
270
[15] M. Cvetic, H. L, J.F. Vazquez-Poritz, Absorption by extremal D3-branes, hep-th/0002128;

M. Cvetic, H. L, J.F. Vazquez-Poritz, Massive-scalar absorption by extremal p-branes, Phys.
Lett. B 462 (1999) 62, hep-th/9904135.
[16] I.R. Klebanov, World-volume approach to absorption by non-dilatonic branes, Nucl. Phys.
B 496 (1997) 231, hep-th/97020776 v2.
[17] S.S. Gubser, I.R. Klebanov, A.A. Tseytlin, Coupling constant dependence in the termodynamics
of N = 4 supersymmetric YangMills theory, Nucl. Phys. B 534 (1998) 202, hep-th/9805156.
[18] S.S. Gubser, I.R. Klebanov, M. Krasnitz, A. Hashimoto, Scalar absorption and the breaking of
world volume conformal invariance, Nucl. Phys. B 526 (1998) 393, hep-th/9803023.
[19] R. Manvelyan, H.J.W. Mller-Kirsten, J.-Q. Liang, Y. Zhang, Absorption cross section of scalar
field in supergravity background, Nucl. Phys. B 579 (2000) 177, hep-th/0001179.
[20] Since metrics of other supergravity theories lead to singular potentials with powers different
from that of the case considered here (i.e., 1/r 4 ), we cite some references with deal with the
general case: G. Tiktopoulos, S.B. Treiman, Weak-coupling limit for scattering by strongly
singular potentials, Phys. Rev. 134 (1964) 844;
L. Bertocchi, S. Fubini, G. Furlan, The theory of scattering by singular potentials, Nuovo
Cimento 35 (1965) 633;
R.A. Handelsman, Y.-P. Pao, J.S. Lew, Low-energy scattering by long-range singular potentials,
Nuovo Cimento LV A (1968) 453;
A. Paliov, S. Rosendorf, High-energy phase shifts produced by repulsive singular potentials,
J. Math. Phys. 8 (1967) 1829.
[21] I.R. Klebanov, World volume approach to absorption by non-dilatonic branes, Nucl.
Phys. B 496 (1997) 231, hep-th/9702076;
I.R. Klebanov, W. Taylor IV, M. Van Raamsdonk, Absorption of dilaton partial waves by D3branes, Nucl. Phys. B 560 (1999) 207, hep-th/9905174.
[22] D.K. Park, H.J.W. Mller-Kirsten, Universality or non-universality of absorption cross sections
for extended objects, Phys. Lett. B (2000), to be published, hep-th/0008215.
[23] P.F. Byrd, M.D. Friedman, Handbook of elliptic integrals for engineers and scientists, 2nd edn.,
Springer, 1971, formulae 110.06 and 260.75.
[24] D. Bak, J. Lee, H. Min, Dynamics of BPS states in the DiracBornInfeld theory, Phys.
Rev. D 59 (1999) 045011, hep-th/9806149.
[25] K.G. Savvidy, G.K. Savvidy, Von Neumann boundary conditions from BornInfeld dynamics,
Nucl. Phys. B 561 (1999) 117, hep-th/9902023;
G.K. Savvidy, Electromagnetic dipole radiation of oscillating D-branes, Nucl. Phys. Proc.
Suppl. 88 (2000) 152, hep-th/0001098;
C.G. Calan, A. Gijosa, Undulating strings and gauge theory waves, Nucl. Phys. B 565 (2000)
157, hep-th/9906153;
C.G. Calan, A. Gijosa, K.G. Savidy, O. Tafjord, Baryons and flux tubes in confining gauge
theories from brane actions, hep-th/9902197.
[26] H.H. Aly, H.J.W. Mller-Kirsten, N. VahediFaridi, Scattering by singular potentials with a
perturbation theoretical introduction to Mathieu functions, J. Math. Phys. 16 (1975) 961
970.
[27] R.M. Spector, Exact solution of the Schrdinger equation for inverse fourth power potential,
J. Math. Phys. 5 (1964) 1185.
[28] N. Limic, Nuovo Cimento 26 (1962) 581.
[29] H.H. Aly, W. Gttinger, H.J.W. Mller-Kirsten, in: H.H. Aly (Ed.), Singular interactions,
Lectures on Particles and Fields, Gordon and Breach, 1970, p. 247.
[30] G. Esposito, Scattering from singular potentials in quantum mechanics, J. Phys. A 31 (1998)
9493, hep-th/9807018.
[31] J. Meixner, F.W. Schfke, Mathieusche Funktionen und Sphroidfunktionen, Springer, 1953.
271
[32] G.H. Wannier, Connection formulas between the solutions of Mathieus equation, Quart. Appl.
Math. 11 (1953) 33.
[33] H.H. Aly, H.J.W. Mller, N. Vahedi-Faridi, Some remarks on scattering by singular potentials,
Lett. Nuovo Cimento 2 (1969) 485.
[34] R.B. Dingle, H.J.W. Mller, Asymptotic expansions of Mathieu functions and their characteristic numbers, J. Reine Angew. Math. 211 (1962) 11.
[35] A.A. Tseytlin, Self-duality of BornInfeld action and Dirichlet 3-brane of type IIB superstring
theory, Nucl. Phys. B 469 (1996) 51, hep-th/9602064;
M. Aganagic, J.P. Park, C. Popescu, J.H. Schwarz, Dual D-brane actions, hep-th/9702133.
[36] R.E. Langer, The solutions of the Mathieu equation with a complex variable and at least one
parameter large, Trans. Amer. Math. Soc. 36 (1934) 637695.
[37] R.B Dingle, Asymptotic expansions: Their derivation and interpretation, Academic, 1973.
[38] J. Meixner, F.W. Schfke, G. Wolf, Mathieu functions and spheroidal functions and their
mathematical foundations: Further Studies, Springer, 1980.

Loop dynamics and AdS/CFT correspondence

A.M. Polyakov a, , V.S. Rychkov b
a Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
b Department of Mathematics, Princeton University, Princeton, NJ 08544, USA
Received 21 June 2000; accepted 19 September 2000
Abstract
We consider the strong coupling limit of conformal gauge theories in 4 dimensions. The action
of the loop operator on the minimal area in the AdS space is analyzed, and the SchwingerDyson
equations of gauge theory are checked. The general approach to the loop dynamics developed here
goes beyond the special case of conformal theories. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
In spite of the great recent progress in understanding gauge fieldsstrings connection
and its special case AdS/CFT correspondence, there is still no true derivation of these
phenomena based on the first principles. In the paper [1] we made a step in this direction
by demonstrating that the loop equations of gauge theory can be verified on the string side
in WKB approximation and for the special contours (wavy lines).
In this note we will give a general treatment of the problem valid for any loops. Our new
approach turns out to be simpler. It reveals some amazing features, connecting the loop
Laplacian and the minimal area functional in the AdS space. As a result, we will be able to
check the loop equations in WKB approximation for arbitrary non-selfintersecting loops.
There is a direct path from here to the full quantum theory, but it will be left for the future,
apart from several comments.
Let us first formulate our main results. Consider the Wilson loop
*
+
I
1
Tr P exp A dx ,
(1.1)
W [C] =
N
C
E-mail addresses: polyakov@viper.princeton.edu (A.M. Polyakov), rytchkov@math.princeton.edu

(V.S. Rychkov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 2 - 8
A.M. Polyakov, V.S. Rychkov / Nuclear Physics B 594 (2001) 272286
273
where we average the ordered exponential over the YangMills fields. In this pure gauge
theory (1.1) satisfies the loop equation
b
L(s)W [C] = W W,
(1.2)
b is defined
where the RHS is zero for non-selfintersecting loops and the loop Laplacian L
by

2W
= b
L(s)W (s s 0 ) + non-local terms.
0
c (s) c (s )
(1.3)
The decomposition (1.3) is not always possible its existence is characteristic of the
zigzag-invariant functionals [1].
When there are other fields in the theory (like in N = 4 SYM) one can define many
different loop functionals, like the one introduced in [2,3]
*
+
I

1
e c(s), y(s) =
Tr P exp
A dx + a dya .
(1.4)
W
N
C
In principle we can integrate around the loop any operator in the adjoint representation
and thus define infinitely many various Wilson loops. It is an open question which
one of them should be used in the gauge fieldsstrings correspondence. This question is
important for the full treatment of the problem. However, we conjecture that in the WKB
approximation the asymptotic behavior for all reasonable definitions should be
Amin [C]
W [C] e
(1.5)
and in the same approximation

b
L(s)W [C] 0.
(1.6)
We are assuming here that we are dealing with any conformal version of gauge theory. The
coupling is strong and 1. When the couplings are running, the WKB approximation
in general is not applicable.
In the second variational derivative of W :

A
A
2 A
2W
=
W,
c (s) c (s 0 )
c (s) c (s 0 )
c (s) c (s 0 )
L(s). 1
the first term in brackets has no singularity for s s 0 and does not contribute to b
Thus, the check that string theory in AdS space satisfies the loop equations of motion of a
gauge theory reduces to the problem of calculating

b
L(s)Amin c(s) = ?
Solving this problem is the main objective of the present work.
1 This fact was considered in [6] as a check that the loop equation is satisfied. Notice, however, that in this
order an arbitrary functional A will pass this check.
274
2. String Lagrangians
Let us begin with setting up the general framework (see also [1]). The minimal area can
be described with the use of the Dirichlet functional
Z
Z

1 d 2
1
d 2 GMN x( ) a x M ( ) a x N ( ) =
(a x)2 + (a y)2 , (2.1)
S=
2
2
2
y
D
= (y, x) =
are the coordinates in the (D + 1)-dimensional AdS space
where
2
with the metric GMN = y MN (M = 0, . . . , D, m = 1, . . . , D). This functional must
be minimized with the boundary conditions

y| D = 0,
(2.2)
x| D = c (s) ,
xM
(y, x m )
where the reparametrization (s) must be chosen so that the classical action

F c (s) = min S x( ), y( )
(2.3)
is stationary. That gives the minimal area

A c(s) = min F c (s) .
(2.4)
{x,y}
{(s)}
Minimization (2.4) is equivalent to the Virasoro constraints imposed on the classical

solution
2
2
2
2
1
T = 2 x + y x y = 0,
y
(2.5)

1
Tk = 2 x x + y y = 0,
y
where = 0 , = 1 . Our strategy will be to calculate the second variation of F [c(s)] and
to derive its short-distance expansion as s s 0 . After that we will translate the result to
A[c(s)].
The equations of motion for (2.1) have the form

1
a 2 a x = 0,
y
(2.6)
2
2
1
a y a x .
y =
y
The boundary conditions are

x(0, ) = c( ),
y(0, ) = 0.
It is straightforward to check that the small expansion of the solution has the form
(see [1])
1
1
x(, ) = c( ) + f ( ) 2 + g( ) 3 + ,
2
3
(2.7)
1
y(, ) = a( ) + b( ) 3 + .
3
275
Direct substitution of (2.7) into (2.6) gives the relations
2
2
0
a ( ) = c ( ) ,
0
c
0 2 d
,
f = (c )
d (c0 )2
(2.8)
while the functions b and g remain arbitrary. They must be determined from the global
considerations. After this is done, all higher terms involved in the expansion (2.7) are
uniquely determined. For example the term y (1/4) e( ) 4 will be given by
2
2
e( ) = (f g) (c0 g 0 ),
a
3a
(2.9)
etc.
The stress tensors (2.5) at the boundary are also easily determined. Substituting (2.7)
into (2.5) we obtain

1
T 0 = 2 f 2 (c0 f 0 ) (a 0 )2 + 2ab ,
a
(2.10)

1 0
Tk
=
(c
g).
0
a2
Notice that the terms 1/ 2 in T cancel automatically due to (2.8). This is because T is
an analytic function which cannot have such a singularity.
One more relation which we need is the expression for the variational derivative
F /c(s). From (2.1) by the standard integration by parts we obtain
Z

d m m
x x + yy 0
F =
2
y
Z

d
=
(2.11)
fm + gm 2 xm + a + b 2 y 0
a 2 2
Z
Z
a
fm
d
1
xm +
gm xm
a2
a
a2
(we assume here that the contour remains at y = 0). The first divergent term is zero due to
(2.8), which means that the divergent part of this action is constant. We obtain
g( )
F
= 2 .
c( )
a
(2.12)
3. Second variation of the Dirichlet functional

In order to act with the loop operator, we must find the second variation of F . After
substituting in (2.1) the expression
xM xM ( ) + y( )M ( ),
and expanding to the second order, we obtain after more or less standard calculations
(3.1)
276

bdry
bdry
S = S0 + S1 + S2 + S2 ,
Z

1
d 2 ( M )2 + rMN ( )M N ,
S2 =
2
(3.2)
where
M = M + ( )MN N ,
( )MN =
rMN =
1
[ xN M0 xM N0 ].
y
1
[( xK )2 MN xM xN ],
y2
(3.3)
bdry
The terms S1,2 containing the first and second order boundary contributions are also easily
calculated, but we do not need them below.
The second variation of the action S with respect to the c( ) will be obtained if we find
the classical solution of the linearized problem with the action S2
2 M + rMN N = 0,
with the boundary condition
cm ( ) cm ( )

=
,
m 0
y
a( )

0 0 = O(1).
(3.4)
(3.5)
When this is done, we will expand m (, ) up to the term 2 , which will give us the
variation of the g-factor and, according to (2.12), the second variation of F .
bdry
Another, equivalent strategy (which requires knowing S2 ) would be to substitute this
solution in (3.2) and find the kernel of the resulting quadratic functional. The standard
integration by parts gives in this case

Z
1
bdry
(2)
M (, ) M (, ) d + S2
S = lim
0
2
Z
1
(3.6)
~mn (, 0 ) cm ( ) cn ( 0 ) d d 0 .
=
2
Here the kernel ~(, 0 ) can be expressed through the Green function of the equation (3.4).
This Green function is not known for a general classical solution x = x( ), y = y( ).
Fortunately, all we need for our task is its short distance expansion as 0 . And this
is relatively easy to obtain. Let us begin with the leading singularity. Substitution of (2.7)
into (3.3) gives in the highest order

c0
1
c0 c0
1
rmn 2 2mn m2 n ,
(3.7)
r00 2 .
(1 )0m m ,
a
Substituting this into (3.2), we obtain in this approximation

Z
0
1
2
cm
2
2
2
2
(m 0 0 m ) .
d ( M ) + ( M ) + 2 (M ) +
S2 =
2
a
(3.8)
277
We kept the terms 1/ 2 in the Lagrangian. As we will see in a moment, this is sufficient
0 . Then the
in the leading order. In this order, we can also neglect the -dependence of cm
action splits into two parts
S2 = S + S k ,

Z
1
2
d 2 ( i )2 + 2 (i )2 ,
S =
2

Z
1
1
2
d 2 ( 0 )2 + ( 1 )2 + 2 02 + 12 + (1 0 0 1 ) ,
Sk =
2
(3.9)
where i = 2, . . . , D, and we chose the contour to run in the x1 direction.
Let us now solve the Dirichlet problem for the action (3.9). In the case of S we get

2
2 p2 i 2 i = 0,

i (p, ) = 1 1 + |p| e|p| ci (p),
(3.10)
where we introduced the Fourier transformed quantities i (p, ) instead of i (, ).

The longitudinal case is slightly more complex, but complex variables solve it. If we
introduce = 1 + i0 , we get
2
2
(3.11)
= p.
2
The solution of this equation takes the form

(p, ) = 1 e|p| 1 + (p) 2|p| + 2p2 2 c1 (p),

1
1 (p, ) = (p, ) + (p, ) ,

2
(3.12)
0 (p, ) = 1 (p, ) (p, ).

2i
As we discussed above, all we need from these solutions are the g-factors, that is the
coefficients in front of 2 as 0.
Expanding expressions (3.10) and (3.12), we get
(
gi (p) = |p|3 ci (p),
(3.13)
g1 (p) = 2|p|3 c1 (p).
2 p2
The functional (3.6) can be written in a convenient form if we introduce the Fourier
transform

Z
+ 0
0
, p eip( ) dp.
~
(3.14)
~(, 0 ) = e
2
From (3.13) and (2.12) it follows that

0 0

c0 c0
|p|3
c c
e
~mn (p, ) = 0 2 2 m0 2n mn m0 2n ,
p c ( )
(c )
(c )
where we restored the general form of cm .
(3.15)
278
Let us now discuss the range of validity of this formula. In its derivation we neglected
many terms in (3.4). There are terms containing higher powers of coming from higher
order terms in (2.7) (we kept only the first ones). Also, we neglected the -dependence of
0 ( ) in (3.8).
cm
This is legitimate, since the values of involved in our calculations are . 1/|p|.
Hence, if we treat the -correction as a perturbation in (3.8), we will obtain a contribution
0 ( ) if we
to ~(p, ) suppressed by 1/|p|. The same is true for the -correction of cm
0 ( ) = c 0 ( ) + c 00 ( )( ), then in perturbation theory i/p in
expand cm
0
0
m 0
m
the (p, )-representation. So, our conclusion is that (3.15) is indeed the leading singularity,
and we need |p| |c00 |/|c0 | for it to be valid.
4. Dimensional analysis
~mn (p, ). However,
In the previous section, we found the leading singularity |p|3 in e
this is just the beginning of the story, since to calculate the action of the loop operator we
~mm (p, ). The general structure has the form
need to pick up terms p0 in e
|p|3
+ A1 ( )p2 + A2 ( )|p| + A3 ( ) + .
(4.1)
p
(c 0 ( ))2
According to our analysis, in order to calculate A1 we have to expand rMN up to 1/ and
up to 0 . A2 and A3 will require further expansion. We will also have to expand c0 ( )
to the needed order.
Direct use of the perturbation theory is straightforward because we know explicit
Green functions for the unperturbed equations (3.10) and (3.11), but cumbersome (see
Appendix A).
We can greatly simplify our task by noticing that the functions A1,2,3( ) depend locally
on the properties of the coefficients rMN and in (3.4) in the limit 0. That means
that they locally depend on the quantities c, f , g, a, b, . . . appearing in the expansion (2.7).
Moreover, we can uncover this dependence by the simple dimensional analysis.
The rules are as follows. We already know that
e
~mm (p, ) = (3 D)
1/p,
(4.2)
where by we mean deviation from the middle point. Then

c0 ( ) 0 ,
f 1/,
c00 ( ) 1/,
g 1/ 2 ,
c000 ( ) 1/ 2 ,
a 0,
b 1/ 2 .
(4.3)
Notice that with these assignments it follows from (2.10) that T Tk 1/ 2 , as it

should be.
Now, from (4.1) the scaling of A1,2,3 must be the following:
A1 p 1/,
A2 p2 1/ 2 ,
A3 1/ 3 .
(4.4)
Let us start now constructing these quantities. For A1 the only thinkable combination is
A1 = f (c02 )(c0 c00 ).
(4.5)
279
It follows that we must have A1 = 0, because the expression (4.5) is odd under the change
, while the equations (3.4) and (3.6) preserve this parity. Hence we confirm the
result of [1] that there is no term 00 (1 2 ) in 2 A/c(1 )c(2 ) (this term would
break the zigzag symmetry).
For A2 we have
A2 = 1 (c00 )2 + 2 (c0 c00 )2 + 3 (c0 c000 ) + 4 b + 5 (c0 g).
(4.6)
Notice that 5 = 0 because g is even under while c0 is odd. If the Virasoro

conditions are satisfied, we can express b in terms of the contour from (2.10). The
quantities k are functions of (c0 )2 . It is easy to find them using the symmetry c c.
Being a second derivative, ~ must scale as c2 . Hence we conclude that
1 (c0 )4 ,
2 (c0 )6 ,
3 (c0 )4 ,
4 |c0 |3 .
(4.7)
To calculate the numerical coefficients, one may use the wavy line limit of [1]. We will not
attempt it here.
Now we come to the most interesting term A3 , representing the action of the loop
operator. We have the following possible structures:
A 3 = 1
(c00 g)
(c0 c00 )(c0 g)
(c0 g)0
+
.
2
3
(c0 )4
(c0 )6
(c0 )4
(4.8)
Once again dimensionally possible terms (c0 c(4) ), (c00 c000 ), (c00 )2 (c0 c00 ), (c0 c000 )(c0 c00 ), b0 and
b(c0 c00 ) are forbidden by parity. The main novelty here is the appearance of g. This
coefficient depends on c( ) non-locally. However, we can express it from (2.12) as

F
0 2
.
(4.9)
g = (c )
c( )
Thus (4.8) gives a remarkable relation between the loop operator and the first derivative of
F:

F
1
b
+ .
(4.10)
L( )F = 0 2 c00
(c )
c
Since the remaining terms contain the combination
(c0 g) = (c0 )2 Tk ,
(4.11)
they drop out if the Virasoro conditions are satisfied.
5. Second variation of the minimal area

In order to calculate the minimal area, we have to use the relation (2.4). By taking
(s) s + (s) and cm (s) cm (s) + cm (s), and by expanding (2.4) to the second order,
we obtain

Z
1 00
F
(2)
0
2
cm (s)(s) + cm (s) (s) ds
A = min
{(s)}
cm (s)
2
280
1
2
2 F
cm (s) cn (s 0 )

0
(s) cn (s 0 ) + (s 0 )cn0 (s 0 ) ds ds 0 .
cm (s) + (s)cm
(5.1)
In principle this formula solves the problem of expressing the second derivative of the
minimal area A through the Dirichlet functional F . For that we have to exclude (s) from
it by the use of extremality condition. Again this can be done by the use of the short distance
expansions of the kernels involved. However, for our limited purposes in this paper we do
b .
not have to do it, since the dimensional analysis for b
LA is the same as for LF
b
We will discuss in Appendix A direct calculation of LA, while here we can just conclude
from (4.8) that

A [c(s)]
(c00 g)
b
.
(5.2)
LA[c(s)] = 0 4 = 0 2 c00 (s)
(c )
(c )
c(s)
To calculate the constant , it is sufficient to consider the wavy line limit of [1]. In this
case we already obtained
= D 4.
(5.3)
We can also understand why D 4 without any calculations. Namely, at D = 4

the b
L operator is conformally invariant, which means that not only A[c(s)] is invariant
under conformal transformations but also b
LA[c(s)]. 2 However these transformations can
00
be used to set c ( ) = 0 (a familiar example we can make straight lines out of circles).
Thus the only way in which (5.2) may be consistent with conformal symmetry is = 0 at
D = 4. As it was shown in [1], if D 6= 4 there is an inhomogeneous term in the conformal
b proportional to (D 4). That explains the origin of (5.3).
transformation of LA
So, the formulae (5.2) and (5.3) solve the main problem addressed in this paper
verification of the loop equation in the WKB limit. This is a first step in deriving the loop
equation for the full string theory functional integral. This problem requires much more
concrete expressions both for the Wilson loop and for the string Lagrangians. We hope
to address it in the future. Here we will just point out a general mechanism by which the
b )F Tk ( ).
non-linear terms in (1.2) can be generated. The Eq. (4.10) tells us that L(
In the classical theory Tk = 0. However in quantum theory expectation values hTk i are
dominated by the pinched disc [4]. This must be the source of the non-linearity in the loop
equations. However there are still some technical difficulties in implementing this idea.
Acknowledgements
The work of A.P. was partially supported by NSF grant PHY-98-02484.
2 This property of b
L is to be expected, since the YangMills equations are conformally invariant exactly at
D = 4. For the direct (and not completely trivial) proof see [1].
281
bA
Appendix A. Calculation of L
In this appendix we give a direct calculation of b
LA along the lines of Section 3. This
will provide an independent check of (5.3), and also to some extent justify and exemplify
the dimensional analysis method of Section 4.
It will be convenient to switch from the conformal coordinates (, ) to the static gauge
x0 (t, s) = t.
(A.1)
The minimal surface is then given by

1
1
1
(A.2)
x(t, s) = c(s) + f (s)t 2 + g(s)t 3 + h(s)t 4 + .
2
3
4
A derivation similar to (2.11) shows that in this gauge
p
A
= c02 g(s).
(A.3)
g(s)
To simplify further calculations, we choose the coordinate s on the unperturbed world
sheet so that

02
f = c00 ,
c (s) 1

(A.4)
(f c0 ) = (gc0 ) = (hc0 ) = = 0 .
x(t,
s)c0 (s) 0
Let us choose a tangent space basis {t, na }, a = 1, . . . , D, at every point of the curve c(s)
so that
c0
(na t) = 0,
(na nb ) = ab .
(A.5)
t= ,
02
c
Then every variation c can be decomposed into the normal and longitudinal part:
c = na (na c) + t(t c) = c + ck .
The second variation of A takes the form
Z Z
1
2 A
ci (s) ck (s 0 )
(2) A =
2
ci (s) ck (s 0 )
Z Z

2A
1
na c(s) nb c(s 0 )
=
a
b
0
2
n (s) n (s )

2 A
t c(s) na c(s 0 )
+2
t(s) na (s 0 )

2 A
t c(s) t c(s 0 ) ,
+
t(s) t(s 0 )
where
2 A
2 A
=
na (s)nbk (s 0 ),
na (s) nb (s 0 ) ci (s) ck (s 0 ) i
2 A
2 A
=
ti (s)nak (s 0 ),
t(s) na (s 0 ) ci (s) ck (s 0 )
2A
2 A
=
ti (s)tk (s 0 ).
t(s) t(s 0 ) ci (s) ck (s 0 )
(A.6)
(A.7)
(A.8)
282
Differentiating the reparametrization invariance condition ti (s) A/ci (s) = 0 gives

A ti (s)
2A
ti (s) =
.
0
ci (s) ck (s )
ci (s) ck (s 0 )
Now it follows from (A.7)(A.9) that
(A.9)

2 A
2 A
=
na (s)nb (s 0 )
ci (s) ci (s 0 )
na (s) nb (s 0 )

A ti (s) a 0
2nk (s ) t(s)na (s 0 ) + tk (s 0 ) t(s)t(s 0 )
0
ci (s) ck (s )

A 00
2 A
a
b 0
n
c
(s)n
(s
)
+
(s)
(s s 0 ), (A.10)
=
na (s) nb (s 0 )
c(s)
where in the last line we used

ti (s)
= ik ci0 (s)cj0 (s) 0 (s s 0 ).
0
ck (s )
(A.11)
To find 2 A/na (s) nb (s 0 ), we vary (A.3) for the second time with a normal variation
c = c . The result is
Z

1
(2)
(A.12)
g c c00 c g c ds.
A =
2
From this we see that

g a (s)
2 A
00
0
a
a
=
+
gc
),
g
:=
n
g
.
(A.13)
(s
s
na (s) na (s 0 )
na (s 0 )
It follows from this and (A.10) that

g a (s)
0
b
(A.14)
L(s)A = coefficient before (s s ) in a 0 .
n (s )
Thus we reduced the problem to studying how g(s) changes under normal variations.
It is well known (a clear-cut derivation can be found in [5]) that under a normal variation
xM xM (t, s) + M (t, s) the area of a minimal surface in curved space changes to the
second order in by
Z
X

h, B(Ei , Ej )i2 R(Ei , ), Ei
(A.15)
S2 = | |2
i,j
(modulo boundary terms). In this coordinate-free notation is the covariant derivative

in the normal bundle; {Ei } is an orthonormal basis of the surface tangent space; h , i is
the metric of the ambient space; B(Ei , Ej ) is the second fundamental form of the surface;
R is the Riemann curvature tensor; integration is taken with respect to the induced metric
volume form.
Let us specify (A.15) to our case of the surface (A.1), (A.2) in the AdS space. We assume
that at every point (t, s) of the surface we have a basis {na (t, s)} of (D 1) normal vectors
such that

0
a
(A.16)
,
naM nbM = ab ,
n (0, s) =
na (s)
283
a (t, s) = t na (t, s), so that a are AdS-normalized.

with na from above. We also put M
M
Now can be written as
a
(t, s).
M = a (t, s)M
(A.17)
The curvature term in (A.15) is simple:

R[MK][NL] = (GMN GKL GML GKN ),

R(Ei , ), Ei = h, ihEi , Ei i + h, Ei ih, Ei i = 2 a a .
(A.18)
For the kinetic term we have

Z
Z
| |2 = d 2 g g a a ,
a = a + (w )[ab] b .
Here g = h x, xi is the induced metric, which is found to be
1 + f 2 t 2 + 2(f g)t 3
1
+ O(t 2 ),
g = 2
2
1
t
(f f 0 )t 3
1 f 2 t 2 (f g)t 3
2
3

1
2
g = 2 1 + (f g)t 3 +O(t 2 ).
t
3
The spin connection coefficients w are antisymmetric and given by

(w )[ab] = a , b = naM nbM (a 6= b).
(A.19)
(A.20)
(A.21)
To simplify them, we choose the normals so that at the boundary

(na )0 = (na c00 )c0
(A.22)
(it is easy to see that this condition is compatible with (A.5)). For this choice of gauge we
see from (A.21) that w = 0 at the boundary, and thus for small t
w (t, s) = O(t).
Finally, the second term in (A.15) is O( 2 K 2 ), where

K 2 (t, s) = B(Ei , Ej ), B(Ei , Ej )
(A.23)
(A.24)
is the extrinsic curvature of the surface squared. This quantity is connected with the
ambient space and the surface curvature tensors by the Gauss equation:

(A.25)
K 2 = R(Ei , Ej )Ej , Ei r(Ei , Ej )Ej , Ei = 2 r,
where r is the scalar curvature of the induced metric g . One can calculate from (A.20)
that r = 2 + O(t 4 ), and thus
K 2 = O(t 2 ).
By using (A.18), (A.19), (A.23) and (A.26) in (A.15) we obtain
(A.26)
284
Z
S2 =
d 2
h
i

g g a a + 2g w[ab] a b + 2 a a + O(t 2 2 ) .
(A.27)
Varying (A.27), we obtain the equation of motion for

g g a 2 g a + 2 g g w[ab] b = O(t 2 ).
(A.28)
As it was explained in Section 3 (this will also be evident from the calculation below)
each power of t in the equations of motion suppresses the perturbative contribution to the
Green function by a factor of 1/p. Since the highest singularity will be |p|3 and we are
interested in finding terms p0 , the terms O(t 2 ) in (A.28) are negligible, being 4 orders
less singular than the leading term g a .

The third term in the LHS of (A.28) is in principle just 2 orders less singular than the
leading terms, so its contribution to g a (s)/nb (s 0 ) can contain terms 0 (s s 0 ) and
(s s 0 ). However, this contribution will be antisymmetric in a b and hence is irrelevant
for the subsequent substitution into (A.14).
Thus effectively we have to care only about the first two terms in (A.28). Taking this
into account and using (A.20), we get the following equation for = a = t a

00
4
t 2 + 2 = f 2 + (f g)t 00
t
t
3
3
4
(A.29)
+ (f g) (f f 0 ) 0 + (f f 0 )t 0 + negligible.
3
2
The boundary conditions are
(0, s) = (s).
(A.30)
The solution of the Dirichlet problem (A.29), (A.30) can be written as

Z
(t, s) = K(t, s|s 0 ) (s 0 ) ds 0 ,
(A.31)
where the Green function K(t, s|s 0 ) solves (A.29) with the boundary conditions
K(0, s|s 0 ) = (s s 0 ).
The small t expansion of can be found through K as follows

Z
1
s|s 0 ) + (s 0 ) ds 0 .
(t, s) = (s) +
t K(0,
s|s 0 ) + t 2 K(0,
2
(A.32)
(A.33)
In particular, we have
...
a (s) ...
= K (0, s|s 0 ) + antisymmetric in a b.
b (s 0 )
(A.34)
To calculate K perturbatively, we pass from the variables (s, s 0 ) to ( = s s 0 , s 0 ) and

perform the Fourier transform in . We have
s ip,
s + s 0 = s 0 + i/p.
Now Eq. (A.29) takes the form
(A.35)

p2 K
4
K
2
0
(f
g)t
K + p2 K
=
f
+
2i(f
f
)
+
p
2
2
3
t
t
3
4
+ (f g)K (f f 0 )ipK + (f f 0 )tipK + negligible,
3
2
where the coefficients in the RHS are taken at s 0 .

The unperturbed Green function is

K0 = e|p|t 1 + |p|t .
285
(A.36)
(A.37)
We are looking for K in the form of a series

K(p, t|s 0 ) = K0 + K1 + K2 + K3 + ,
(A.38)
where
Ki
1 |p|t
e
pi
(p 1).
(A.39)
The equation (A.36) can be schematically written as

A0 K = A2 K + A3 K + ,
(A.40)
where we indicated the relative order of singularity of the differential operators. It follows
e = K2 + K3 is
by simple counting that the K1 correction is absent from (A.38) and that K
found from the equation
(
e = (A2 + A3 )K0 ,
A0 K
(A.41)
e t =0 = 0.
K|
e
Thus substituting K0 into the RHS of (A.36) gives the equation for K
t
e p2 K
e
t K
= f (p, t)
t2
t2
4
8
(f g)p2 t 2 (f g)|p|t
|p|t + |p| e
3
3

9
3
.
(A.42)
+ i sign p(f f 0 ) |p|t 3p2 t 2
2
2
2 2 |p|t
:= 2f p e
This inhomogeneous problem has the solution

Z
e t) = Gp (t, )f (p, ) d,
K(p,
with the Green function

1
F0 |p|t< F |p|t> ,
Gp (t1 , t2 ) =
3
2|p|
t
(t)
=
e
(1
+ t),
F
F0 (t) = et (1 t) F (t).
Routine power expansion shows that
|p|t
(A.43)
(A.44)
286

e (p, t) = 1 1 t 2 p2
K0 + K
2

1 3
+ t |p|3 f 2 |p| (f g) (f f 0 )i sign p .
3
Thus it follows from this and (A.34) that
(A.45)
...
a (s)
= 2(f g) (s s 0 ) + .
b (s 0 )
(A.46)
...
Finally, it remains to relate a / b and g a / b . This relation is not immediate, because the normal variation does not preserve the static gauge (A.1). The perturbed surface
(
e
x0 (t, s) = t + a (t, s)na0 (t, s),
(A.47)
e
x (t, s) = x(t, s) + a (t, s)na (t, s),
has to be reparametrized by introducing the new coordinate e
t =e
x0 , so that

t =e
t a e
t, s na0 e
t, s + O( 2 ).
(A.48)
In the new coordinates the perturbed surface takes the static gauge form (we rename e
t t)

e
x0 (t, s) = t,

(A.49)
e
x (t, s) = x(t, s) + a (t, s)na (t, s) f t + gt 2 a (t, s)na0 (t, s).
To use (A.49), we must explicitly find na (t, s) satisfying (A.16). The result is non-unique
na (t, s) for each a by adding a
if D > 2. Our choice was to first transform na (0, s) e
0
a
n are orthogonal to the surface, and then to
linear combination of x(t,
s), x (t, s) so that e
e a }. The result is:
orthonormalize the family {n0
na (t, s) = fa t ga t 2 + O(t 3 ),
0

1
(A.50)
na (t, s) = na + nb fa fb t 2 + (fa gb + fb ga )t 3
f f t 2 (gf + f g )t 3 + c0 O(t 2 ) + O(t 4 )

a
(all functions in the RHS are taken at s). Substituting this into (A.49) gives

1 ... 3
g a = a + fa gb + fb ga b .
2
2
Now it follows from (A.46) and (A.51) that
g a (s)
= (4 D) (s s 0 ) + ,
a (s 0 )
and hence = D 4 by (A.14).
References
[1]
[2]
[3]
[4]
[5]
[6]
A. Polyakov, V. Rychkov, Nucl. Phys. B 581 (2000) 116, hep-th/0002106.

J. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
S.-J. Rey, J. Yee, hep-th/9803001.
A. Polyakov, hep-th/9711002, Proceedings of Strings 97.
T.H. Colding, W.P. Minicozzi, Minimal Surfaces, New York University, 1999.
N. Drukker, D.J. Gross, H. Ooguri, Phys. Rev. D 60 (1999) 125006, hep-th/9904191.
(A.51)
(A.52)

Vortices, monopoles and confinement

L. Del Debbio a, , A. Di Giacomo a , B. Lucini b
a Dipartimento di Fisica, Universit di Pisa, and INFN Sezione di Pisa, Pisa, Italy
b Theoretical Physics, University of Oxford, Oxford, UK
Abstract
We construct the creation operator of a vortex using the methods developed for monopoles. The
vacuum expectation value of this operator is interpreted as a disorder parameter describing vortex
condensation and is studied numerically on a lattice in SU(2) gauge theory. The result is that vortices
behave in the vacuum in a similar way to monopoles. The disorder parameter is different from zero
in the confined phase, and vanishes at the deconfining phase transition. We discuss this behaviour in
terms of symmetry. Correlation functions of the vortex creation operator at zero temperature are also
investigated. A comparison is made with related results by other groups. 2001 Elsevier Science
B.V. All rights reserved.
Keywords: Confinement; Vortices; Monopoles
PACS: 11.15.Ha; 12.38.Aw; 64.60.Cn
1. Introduction
Recently there has been renewed interest in understanding the role of vortices in
the mechanism of colour confinement [13]. Vortices were originally introduced in the
continuum theory as string-like topological defects [4], and have been studied both in the
continuum and on the lattice [5]. In the notation of Ref. [4], a vortex creation operator
B(C), for a gauge group SU(N), can be associated to each closed oriented curve C and
has the following commutation relation with a Wilson loop W (C 0 ), bounded by a closed
curve C 0 :
W (C 0 ) B(C) = B(C) W (C 0 ) exp(2in/N),
(1)
where n is the winding number of C 0 around C. In SU(2) such a vortex contributes a factor
(1) to each Wilson loop with an odd linking number with the vortex line. On a lattice,
E-mail addresses: ldd@df.unipi.it (L. Del Debbio), digiaco@df.unipi.it (A. Di Giacomo),

lucini@thphys.ox.ac.uk (B. Lucini).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 1 - 9
288
L. Del Debbio et al. / Nuclear Physics B 594 (2001) 287300
the worldsheet M of the vortex can be associated to a surface on the dual lattice. A vortex
creation operator is defined by twisting the plaquettes in the action, whose duals belong to
that surface.
While creating a vortex is a well defined procedure, detecting vortices in lattice gauge
configurations is more difficult. Attempts in this direction stemmed from [6], and results
were obtained by several groups [7], indicating that vortices can play a role in confinement.
However, it is not clear that this procedure detects the vortices as defined in Eq. (1) (see,
e.g., [8,9]). The current picture of the role of vortices in colour confinement is based
on Eq. (1): at low temperature, vortices disorder the Wilson loop to produce the wellknown area law; at high temperature, vortices are suppressed and the Wilson loop obeys a
perimeter law.
In three dimensions, a conserved topological quantum number can be associated to
a vortex, and a vortex creation operator is defined as a local complex scalar field. The
dynamics of this scalar field is described by an effective Lagrangian with a global dual
ZN symmetry [4]. The breaking of this symmetry is responsible for the different phases
of the theory. The scalar field (x0 ) in three dimensions has a non-trivial commutation
relations, analogous to those of Eq. (1), with any Wilson loop encircling the point x0 .
Operators that create such topological excitation are known in statistical mechanics as
disorder operators [10] and the expectation value of a disorder operator is a disorder
parameter and is related to the free energy of the associated topological excitation.
In four dimensions the situation is less clear. The vortex corresponds to a string-like
topological defect and the dual symmetry does not emerge naturally. In this respect,
there is a fundamental difference between the vortex and the monopole picture of colour
confinement in terms of symmetry: for the monopoles, a topologically conserved current
exists, whose zero component is the generator of the dual symmetry, and a picture based
on a mechanism, namely the dual Meissner effect [11], emerges as a consequence of
the breaking of the dual symmetry [12]. For the vortices the symmetry pattern is not
understood. In close analogy with previous work on abelian-projected monopoles, it is
nonetheless possible to define a vortex creation operator obeying the above commutation
relation, which can be associated to the free energy of the vortex. Starting from this point of
view, in this paper we study the role of vortices across the deconfinement phase transition,
using the techniques developed in Refs. [1214] for investigating the condensation of
monopoles defined via the abelian projection.
A vortex creation operator is introduced as a disorder operator for the SU(2) lattice
gauge theory. When applied to a gauge configuration, creates a vortex associated to a
string closing in the z direction via periodic boundary conditions (PBC). As well as for
other systems, both in statistical mechanics [10,15,16] and in quantum field theory [12
14,17], such an operator can be constructed without any reference to the dual theory.
In Section 2 we give the explicit construction of this operator. While our study was in
progress, other papers appeared which addressed similar problems by using quantities
related to the vortex creation operator examined in this work. The explicit relationship
between the operator studied here and other quantities in the literature is reviewed in
Section 3.
289
Correlation functions of the vortex creation operator at zero and finite temperature yield
useful informations on the behaviour of these topological defects in the SU(N) YangMills
vacuum. As in previous studies on monopoles, it is convenient to compute
=
d
loghi
d
(2)
and analogous quantities for n-point correlators. Unlike , can be determined with good
accuracy and provides all the needed information. The vacuum expectation value (vev) of
proves to be a disorder parameter for the finite temperature deconfining phase transition,
as reported in Section 4. Another interesting quantity to be investigated is the correlator
h(t0 )(t0 + t)i at zero temperature.
Our results establish an interesting connection between the behaviour of monopoles and
vortices wrapping around the lattice.
For completeness we remind that the behaviour of the vortex density across the
deconfining phase transition has already been studied in Ref. [18]. The percolation of
vortices at the phase transition has been investigated in Refs. [19,20]. In all these studies,
the vortices were defined by maximal center projection.
The approach presented in this paper can be generalised to SU(N) gauge group and to
vortex loops of generic geometry, also not wrapping through PBC. Work is in progress on
these aspects.
2. Vortex creation operator

The vortex creation operator is constructed by the same technique used for the
monopole creation operator in previous works [13,14]. The vacuum expectation value of
is defined as the ratio of two partition functions:
Z
e
Z
e
(3)
h(t0 , x0 , y0 )i = = Z 1 [dU ]e S[U ] ,
Z
where Z is the usual partition function with the Wilson action,
X
Tr[1 P (x)]
S[U ] =
(4)
x,
and e
S is obtained from S by twisting a line of plaquettes in the 0y plane:
P0y (t0 , x > x0 , y0 , z) 7 P0y (t0 , x > x0 , y0 , z),
z.
(5)
The net effect of this transformation is best understood by performing a series of changes
of variables in the functional integral as follows. The first change:
Uy (t0 + 1, x > x0 , y0 , z) 7 Uy (t0 + 1, x > x0 , y0 , z),
yields
h(t0 , x0 , y0 )i = Z 1
e(1) [U ]
[dU ]e S
(6)
(7)
290
Fig. 1. Set of plaquettes changing sign in the hyper-plane t0 + 1 after the first change of variables
described in Eq. (6). The plaquettes in the plane xy correspond to the location of the vortex lines at
(x0 , y0 ), and at (Ns 1, y0 ) due to PBC in the x direction.
where e
S (1) is defined by the following transformation of the Wilson action:
P0y (t = t0 + 1, x > x0 , y0 , z) 7 P0y (t = t0 + 1, x > x0 , y0 , z), z,
P0x 7 P0x ,
P0z 7 P0z ,
Pxy (t = t0 + 1, x0 , y0 , z) 7 Pxy (t = t0 + 1, x0 , y0 , z), z,
P (t = t0 + 1, Ns 1, y0 , z) 7 Pxy (t = t0 + 1, Ns 1, y0 , z), z,
xy
Pxz 7 Pxz .
Pyz 7 Pyz ,
The change of sign in the P0y plaquettes introduced at t = t0 in Eq. (5) has been shifted
at t = t0 + 1 and two vortices have been created in the (x, y) plane for t = t0 + 1 at (x0 , y0 )
and (Ns 1, y0 ). It is worthwhile to remark that the second vortex at (Ns 1, y0) is due
to the fact that we are working in a finite volume with PBC in the x direction. In an infinite
volume, or with free BC, the vortex at (x0 , y0 ) would have been created alone. In the z
direction, the vortices either extend to infinity (for an infinite lattice), or they form a closed
loop due to PBC. The set of xy plaquettes changing sign after this first change of variables
is shown in Fig. 1.
The same change of variables can be iterated at successive times, yielding, after n
iterations:
h(t0 , x0 , y0 )i = Z 1
e(n) [U ]
[dU ]e S
and e
S (n) is obtained from the Wilson action via:
(8)
291
P0y (t = t0 + n, x > x0 , y0 , z)
7 P0y (t = t0 + n, x > x0 , y0 , z), z,
(t
=
t
+
1,
x
,
y
,
z)
7 Pxy (t = t0 + 1, x0 , y0 , z), z,
P
xy
0
0
0
Pxy (t = t0 + 1, Ns 1, y0 , z) 7 Pxy (t = t0 + 1, Ns 1, y0 , z), z,

..
(t
=
t
+
n,
x
,
y
,
z)
7
Pxy (t = t0 + n, x0 , y0 , z), z,
P
xy
0
0 0
Pxy (t = t0 + n, Ns 1, y0 , z) 7 Pxy (t = t0 + n, Ns 1, y0, z), z.

Such a configuration corresponds to the propagation of the two vortices, from time t = t0
to time t = t0 + n.
Correlations of operators at different times {t1 , . . . , tn } are defined by repeating the
transformation in Eq. (5) for each ti . For a two-point correlation:
(t) = h(t0 , x0 , y0 ) (t0 + t, x0 , y0 )i
(9)
the change in the P0y plaquettes is reabsorbed when n = t and the correlator describes the
propagation of a pair of vortices from the time t0 to the time t0 + t.
We remark here that these prescriptions create vortex lines that are closed in space, due
to PBC in the z-direction, and propagate in time.
At large values of t, decreases exponentially to an asymptotic value, which, by cluster
property, is hi2 , the square of the disorder parameter:
(t) ' Aemt + hi2 .
(10)
At T = 0, hi can be extracted from the large-t value of the correlator, according to

Eq. (10).
At finite temperature, there is no propagation in time, and a direct measurement of hi
is needed by use of a single operator. In order to have a consistent implementation of
the changes of variables described above, C boundary conditions are needed in the time
direction, as explained in [14].
Closed vortex lines propagating in time can be created in a field configuration with the
same prescription, but with a different modification of the action. For instance, using for
the modified action:
P0y (t0 , x0 < x < x1 , y0 , z0 < z < z1 ) 7 P0y (t0 , x0 < x < x1 , y0 , z0 < z < z1 )
(11)
gives a vortex line encircling the rectangle in the xz plane:
R = {(x, y, z): x0 < x < x1 , y = y0 , z0 < z < z1 }.
(12)
The same definition is used in a recent publication [23].
3. Comparison with related works

Our disorder parameter is strictly related to observables introduced by other authors
[13]. The detailed comparison is as follows.
292
The paper [2] presents a computation of the free energy of a vortex pair. The latter is
defined as:
F = T log
where
Z(, 0 ) =
Z(, )
,
Z(, )
Z
(13)
[dU ]eS(, ) ,
X
X
1
Tr P + 0
Tr P
S(, ) =
2
P M
/
P M
and T = (Nt a)1 is the temperature of the system. M indicates a set of plaquettes with
modified coupling 0 = , which can be seen as the plaquettes transversed by the vortex
string. The following two cases are examined:
a vortex solution is placed at (x0 , y0 , z0 , 0) and an anti-vortex at (x0 , y0 , z0 + d, 0),
with a straight string of twisted plaquettes connecting them;
a single vortex is placed in the middle of the lattice and the string extends to the
boundary in the z-direction, where free boundary conditions are used.
In both cases the modification of the action is done in all time-slices.
Using the notation introduced in the previous section:
Z
Z(, )
0
= Z 1 [dU ]eS [U ]
(14)
Z(, )
and now S 0 is obtained from S by:
Pxy (t, x0 , y0 , z = z0 + 1) 7 Pxy (t, x0 , y0 , z = z0 + 1),

..
.
Pxy (t, x0 , y0 , z = z0 + d) 7 Pxy (t, x0 , y0 , z = z0 + d),
t,
t.
With a t
z relabelling of the axes, the correlator defined in this paper by Eq. (9) yields
the free energy of two vortex pairs as defined in [2] at distance Ns /2.
At T = 0 we use symmetric lattices (Ns = Nt ) and measure the time correlator of two
operators at x0 = y0 = Ns /2, which is in fact the correlator of two pairs, when the effect
of periodic boundary conditions are taken into account. At non-zero temperature, we use
a single vortex operator, which actually introduces two vortices, again due to PBC. The
extra vortices that are created by PBC are Ns /2 lattice spacings away from their partners.
If Ns , where is the correlation length, then our disorder parameter hi is the square
of exp(F /T ).
The authors of Ref. [3] are concerned with the behaviour of the t Hooft loop [4] in hot
QCD, and relate the dual string tension to the wall tension of ZN domain walls. For SU(2),
the lattice definition of the loop operator is given by [21]:
)
(
X
(Tr Pzt (x) + h.c.) ,
(15)
V [C] = exp
xS
293
Fig. 2. The operator introduced in Section 2 corresponds to V (C) for the curve C depicted here.
The box represents a time-slice of the whole lattice.
where S is the surface in the xy plane bounded by the curve C. The vev of the operator
defined in Eq. (15) is again the ratio of two partition functions: one with a modified action
divided by the (standard) Wilson action. The modified action is obtained by changing
the sign of the plaquettes in S. The definition given in Section 2 coincides exactly with
Eq. (15), for the curve C depicted in Fig. 2, after a relabelling of the axes y
z. It is
easy to realise in this formulation that our vortex creation operator is precisely a t Hooft
loop operator: every Wilson loop with non-trivial linking to the curve C receives a (1)
contribution from the vortex.
Finally, the authors of Ref. [1] compute the free energy of a closed SU(2) vortex, which
is defined from the logarithm of the ratio of two partition functions:
Z( )
.
(16)
Z
Z( ) is defined as usual by multiplying a given (co-closed) set V of plaquettes in the
action with an element of the center of the gauge group. For SU(2), the only non-trivial
possibility is = 1. For the set of plaquettes:
exp{F ( )} =
V = {P (t, x, y, z): (, ) = (0, 2), t = t0 , x0 < x < x1 , y = y0 , z0 < z < z1 }

the definition in [1] coincides with the one in Eq. (11).
4. Numerical results
In this paper, we present data only for open vortex lines wrapping in the z direction by
PBC. Work is in progress to study closed vortices as done in [23].
Due to the exponential in its definition, has large fluctuations, which make a direct
measurement a challenging task. Instead, as in previous studies [12,13], we define
=
d
logh(t0 , x0 , y0 )i = hSiS he
Sie
S.
d
(17)
294
Fig. 3. Comparison of defined with the vortex and the monopole creation operator.
Being the difference of the expectation values of two actions, can be easily computed,
and proves to yield all the relevant information. At finite temperature, is expected to
have a sharp negative peak in the critical region [1214], if h(t0 , x0 , y0 )i is related to the
deconfinement phase transition.
Our results for are displayed in Fig. 3, for several lattice sizes. For comparison, we
also report a plot of the corresponding quantity for the vev of a monopole creation operator
on a 124 4 lattice [12]. Already at a qualitative level, it is clear that, at weak coupling,
is negative, with its absolute value increasing with the volume. This behaviour implies
that hi vanishes for large in the thermodynamical limit. At low , is compatible
with 0 for all the volumes considered in this study, which means that hi has a non-zero
value in the infinite volume limit in the confined phase. The presence of the negative peak
connecting the confined and deconfined phases suggests that vortices play a role at the
deconfinement transition. This behaviour is in agreement with the results in [1,2]. As in the
case of monopoles [12], we can perform a finite size scaling analysis of the data presented
above. In neighborhood of the critical coupling c , if hi is a disorder parameter for the
deconfining phase transition, in the infinite volume limit we have
hi (c )
(18)
being the corresponding critical index. In a finite volume, the previous equation is
replaced by
hi = (c ) (Ns / ),
(19)
295
where is the correlation length and is a function of the ratio Ns / . Since

(c )
(20)
Eq. (19) can be written as

e L1/ (c )
hi = L/
being the critical index of the correlation length.

Eq. (21) implies
= f L1/ (c ) ,
1/
L
(21)
(22)
i.e., the ratio /L1/ is an universal function of the scaling variable

x = L1/ (c ).
By guessing that [12]
(23)
= +c
1/
L
x
it is possible to extract from our data the critical exponents and and the critical
coupling c . We perform three different fits: a fit with three parameters, a fit to and
at fixed c , using the value in [22], and again a two parameter fit to and at fixed c ,
using the value obtained in the three parameter fit. The error is estimated from the variation
in the fitted values when using the different methods described above and when the points
corresponding to smaller correlation lengths are excluded. Our results are:
c = 2.30(1),
= 0.5(1),
= 0.7(1).
The values of c and are in good agreement with the values obtained in [12], while the
value of varies by two standard deviations, when compared to previous results.
Fig. 4 shows how well the scaling relation is obeyed with the values of c and from
the above fit.
Assuming that in the weak coupling limit all link variables are close to unity, and
remembering that the twisted action is obtained by flipping the sign of 2 Nt L
plaquettes, a naive prediction would yield:
' 16L + const.
(24)
This is a justification for the linear dependence used in the ansatz above. However, the
actual value of the coefficient a does not need to be the one in Eq. (24), since the
configuration with all links set to the identity is not the true minimum of the modified
action which enters the definition of .
To check this behaviour, we have measured for different volumes and very large values
of the coupling. In this region, we expect the data to be independent of . Within the errors
the data, shown in Fig. 5, do lie on a straight line as predicted by Eq. (24). The fact that the
296
Fig. 4. Plot of rescaled data.
slope is negative ensures that the disorder operator vanishes in the weak coupling phase in
the thermodynamic limit.
The behaviour of at low s is displayed in Fig. 6: the value at low remains bounded
from below and consistent with zero for increasing volumes, which guarantees that does
not vanish in the thermodynamic limit above the phase transition.
Another quantity that can give information on the behaviour of center vortices in the
vacuum of SU(2) is the correlation function of two operators as a function of distance
at zero temperature. When relabelling the axes, as discussed in the previous section, this
quantity is identical to the free energy recently discussed in [2].
Again, our numerical computation is performed in terms of 2 , now defined as
d
logh(t0 , x0 , y0 )(t0 + t, x0 , y0 )i
(25)
2 (t) =
d
which is the difference between two actions. Since the usual Wilson contribution is
t-independent, the whole dependence on t is brought by e
S.
In Fig. 7 we display the behaviour of e
S on a 164 lattice at = 2.5, where the system is
confined and scaling is supposed to work.
The t dependence of the correlation is fitted using two different fitting functions, both
depending on three parameters (a, b, m):
emt
+ b,
t
2 (t) = aemt + b
2 (t) = a
Fig. 5. vs. L for large values of . The solid line corresponds to a linear fit.
Fig. 6. vs. for different spatial volumes.
297
298
Fig. 7. The modified action e

S per unit of volume as a function of the distance between the vortex and
the antivortex on a lattice 164 at = 2.5.
Table 1
Fit results for the correlation of vortex operators
Fit type
Fit I
Fit II
2 /d.o.f.
0.00018
0.00035
0.34923
0.34923
0.650
1.302
2.6
1.8
these two fits will be named respectively fit I and fit II in what follows. The functional
dependence in fit I describes a Yukawa potential between vortices, as studied in Ref. [2].
In both cases, b is the asymptotic value of 2 and is expected to be different from zero
if vortices are condensed. Both ansatz fit the data relatively well, despite the fact that the
masses obtained are quite different. The values of 2 give an indication that fit II could be a
better description of the data, although it is impossible to draw any robust conclusion given
that it is very difficult to disentangle the logarithmic correction in t which differentiate fit
I from fit II. The outcome of the fits is summarised in Table 1 and the two fitting curves
are superimposed on the data in Fig. 7. The value of the constant b is determined quite
accurately due to the plateau in the data points at large t. It is interesting to remark that a
turns out to be negative in both cases up to 90% CL, while the error on the fitted mass is
very large. This is mainly due to the fact that large variations in m can be reabsorbed by
variations in a in the range where data points are available.
We notice from fit II that the correlation length is around 1.3 lattice spacings. This
justifies the view of the pair of vortices (the one at the center of the lattice and the one
at the border produced by PBC) as independent.
299
5. Discussion and future outlooks

In previous works [12], we produced evidence of dual superconductivity of the QCD
vacuum, supporting the mechanism of Refs. [11]. A peculiar feature of this phenomenon is
that monopoles defined by different abelian projections all condense [12]. This possibility
was suggested in Ref. [24] and lattice data support it, showing that confinement is related
to condensation of magnetic charges defined by a few abelian projections.
If all or a large class of abelian projections are equivalent, there are infinitely many
physically equivalent disorder symmetries. We already observed that this fact is not
inconsistent, but suggests that maybe a more fundamental dual symmetry pattern exists,
which can manifest itself as condensation of all these magnetic charges [25].
Now it is found that also vortices show condensation. We were not able to associate
a dual conserved topological quantity to vortices in 3 + 1 dimensions, contrary to what
happens in 2 + 1 dimensions. However we consider what is observed here an important
information on the way to understand the features of the dual description.
We are extending the analysis to the vortices of the SU(3) gauge theory and to
some aspects of closed vortex lines. We are also trying to understand what happens
in the presence of quarks. Preliminary data on condensation of monopoles, as defined
by the abelian projections, show that the presence of quarks does not affect the dual
superconductor picture. This is in agreement with the idea that the gross features of the
QCD vacuum, including the mechanism of confinement, are determined already at Nc =
[26].
We are now trying to understand how to extend to full QCD the analysis of vortices.
Acknowledgements
We thank M. DElia, G. Paffuti and M. Teper for enlightening discussions. Financial
support by the EC TMR Program ERBFMRX-CT97-0122 and by MURST is acknowledged. BL is funded by PPARC under Grant PPA/G/0/1998/00567.
References
[1] T.G. Kovacs, E.T. Tomboulis, Computation of the vortex free energy in SU(2) gauge theory
UCLA-00-TEP-06, February 2000, hep-lat/0002004.
[2] Ch. Hoelbling, C. Rebbi, V.A. Rubakov, Free energy of an SU(2) monopole-antimonopole pair,
March 2000, hep-lat/0003010.
[3] C. Korthals-Altes, A. Kovner, M. Stephanov, Phys. Lett. B 469 (1999) 205;
C. Korthals-Altes, A. Kovner, Magnetic ZN symmetry in hot QCD and the spatial Wilson loop,
April 2000, hep-ph/0004052.
[4] G. t Hooft, Nucl. Phys. B 138 (1978) 1.
[5] G. t Hooft, Nucl. Phys. B 153 (1979) 141;
N.K. Nielsen, P. Olesen, Nucl. Phys. B 160 (1979) 380;
P. Vinciarelli, Phys. Lett. B 78 (1979) 485;
J.M. Cornwall, Nucl. Phys. B 157 (1980) 392;
300
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
J. Ambjrn, P. Olesen, Nucl. Phys. B [FS1] 170 (1980) 60;

J. Ambjrn, P. Olesen, Nucl. Phys. B [FS1] 170 (1980) 265;
G. Mack, V.B. Petkova, Ann. Phys. (NY) 123 (1979) 442;
G. Mack, in: Recent Developments in Gauge Theories, G. t Hooft et al. (Eds.), Plenum, New
York, 1980;
G. Mack, V.B. Petkova, Ann. Phys. (NY) 125 (1980) 117;
G. Mack, Phys. Rev. Lett. 45 (1980) 1378;
G. Mack, E. Pietarinen, Nucl. Phys. B [FS5] 205 (1982) 141;
L.G. Yaffe, Phys. Rev. D 21 (1980) 1574;
E.T. Tomboulis, Phys. Rev. D 23 (1981) 2371.
L. Del Debbio, M. Faber, J. Greensite, S. Olejnik, Phys. Rev. D 55 (1997) 2298.
See, e.g., Ph. de Forcrand, M. DElia, Phys. Rev. Lett. 82 (1999) 4582, and further references
by the same group;
J. Gattnar, K. Langfeld, A. Schfke, H. Reinhardt, hep-lat/0005016, and references therein by
the same group;
B. Bakker, A. Veselov, M. Zubkov, Phys. Lett. B 471 (1999) 214;
M. Chernodub, M. Polikarpov, A. Veselov, M. Zubkov, Nucl. Phys. Proc. Suppl. B 63 (1999)
575;
L. Del Debbio, M. Faber, J. Giedt, J. Greensite, S. Olejnik, Phys. Rev. D 58 (1998) 094501;
L. Del Debbio, M. Faber, J. Greensite, S. Olejnik, New developments in quantum field theory,
hep-lat/9708023.
T.G. Kovacs, E.T. Tomboulis, Vortices and confinement, hep-lat/9912051;
T.G. Kovacs, E.T. Tomboulis, Phys. Lett. B 463 (1999) 104.
M. Faber, J. Greensite, S. Olejnik, D. Yamada, JHEP 9912 (1999) 012.
L.P. Kadanoff, H. Ceva, Phys. Rev. B 3 (1971) 3918.
G. t Hooft, in: High Energy Physics, A. Zichichi (Ed.), EPS International Conference, Palermo,
1975;
S. Mandelstam, Phys. Rep. C 23 (1976) 245.
A. Di Giacomo, B. Lucini, L. Montesi, G. Paffuti, Phys. Rev. D 61 (2000) 034503;
A. Di Giacomo, B. Lucini, L. Montesi, G. Paffuti, Phys. Rev. D 61 (2000) 034504.
L. Del Debbio, A. Di Giacomo, G. Paffuti, Phys. Lett. B 355 (1995) 255.
A. Di Giacomo, G. Paffuti, Phys. Rev. D 56 (1997) 6816.
J.M. Carmona, A. Di Giacomo, B. Lucini, A disorder analysis of the Ising model, May 2000,
hep-lat/0005014.
G. Di Cecio, A. Di Giacomo, G. Paffuti, M. Trigiante, Nucl. Phys. B 489 (1997) 739;
A. Di Giacomo, D. Martelli, G. Paffuti, Phys. Rev. D 60 (1999) 094511.
E. Marino, B. Schroer, J.A. Swieca, Nucl. Phys. B 200 (1982) 473.
K. Langfeld, O. Tennert, M. Engelhardt, H. Reinhardt, Phys. Lett. B 452 (1999) 301.
M. Engelhardt, K. Langfeld, H. Reinhardt, O. Tennert, Phys. Rev. D 61 (2000) 054504.
R. Bertle, M. Faber, J. Greensite, S. Olejnik, JHEP 9903 (1999) 019.
J. Groeneveld, J. Jurkiewicz, C.P. Korthals-Altes, Phys. Scripta 23 (1981) 1022.
J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493.
S. Cheluvaraja, Vortices in SU(2) lattice gauge theory, hep-lat/0006011.
G. t Hooft, Nucl. Phys. B 190 (1981) 455.
A. Di Giacomo, Nucl. Phys. A 661 (1999) 13c.
G. t Hooft, Nucl. Phys. B 72 (1974) 461;
E. Witten, Nucl. Phys. B 160 (1979) 57;
P. Di Vecchia, G. Veneziano, Nucl. Phys. B 171 (1980) 243.

Probing extra dimensions using NambuGoldstone

bosons
Taichiro Kugo, Koichi Yoshioka
Department of Physics, Kyoto University, Kyoto 606-8502, Japan
Received 28 December 1999; accepted 25 October 2000
Abstract
We investigate a possibility that our four-dimensional world is a brane-like object embedded in a
higher-dimensional spacetime. In such a situation, the transverse coordinates of the brane become
the NambuGoldstone bosons which appear as a result of spontaneous breaking of the translation
symmetry. We determine the form of the effective action of the system, finding the explicit form of
the vierbein induced on the brane in terms of the NambuGoldstone boson variables and the bulk
vielbein. The KaluzaKlein mode couplings are suppressed by the effect of the NambuGoldstone
bosons and the suppression is stronger with a smaller value of the brane tension. However we here
show that the brane tension cannot be arbitrarily small since the inverse of the brane tension gives
the coupling constant of the NambuGoldstone bosons. A rather stringent bound is obtained for the
brane tension and the fundamental (string) scale from the consideration of the cooling process in
supernovas. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
To understand the existence of various hierarchical scales in nature is one of the most
important problem in particle physics. Recently it was proposed in [1] that the existence
of extra spacetime dimensions may solve the gauge hierarchy problem between the Planck
and weak scales in four dimensions. Stimulated by their work, there have appeared a large
number of papers concerning the physics of extra dimensions. They also deal with various
phenomenological problems, such as fermion mass hierarchy [2], neutrino physics [3], supersymmetry breaking [4], flavor problems [5], cosmology and astrophysics [6], and so on.
It is really an interesting possibility that our four-dimensional world may lie on a brane
like a D-brane, orientifold plane, domain wall, etc. embedded in larger dimensions [7].
E-mail addresses: kugo@gauge.scphys.kyoto-u.ac.jp (T. Kugo), yoshioka@gauge.scphys.kyoto-u.ac.jp

(K. Yoshioka).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 5 - 3
302
T. Kugo, K. Yoshioka / Nuclear Physics B 594 (2001) 301328
What are then typical signatures for such a brane world? Since there cannot exist a rigid
body in the relativistic theory, any type of brane must necessarily fluctuate. Therefore
there are always scalar fields which stand for the coordinates of the brane in the transverse
dimensions. These scalar fields are the NambuGoldstone (NG) bosons which appear as a
result of spontaneous breaking of the translation symmetry by the presence of the brane.
Then one important way to explore the possibility of brane world is to investigate the
effects of this NG boson.
In higher-dimensional models where the gravity and/or gauge fields live in the bulk,
there are infinite numbers of KaluzaKlein (KK) gravitons and/or KK gauge bosons on the
brane. They couple to the matter fields living on the brane (our world) and can give visible
effects in future collider experiments and astrophysics. These effects have been intensively
investigated so far in many articles [11]. However, in calculating some processes with
KK excitation modes, there was a puzzling problem; in summing up infinitely many KK
modes, a tree-level amplitude diverges when the number of extra dimensions is larger than
or equal to 2. In the previous work [8], we showed that the higher KK mode couplings
to the fields confined on the brane are exponentially suppressed if the effect of the above
mentioned NG bosons is properly taken into account. That is, the suppression is due to the
fluctuation of the brane. The suppression factor works as a regulator in the summations of
KK-mode contributions and cures the problematic divergences. We also discussed some
important phenomenological consequences of this factor. There, we concluded that if the
brane tension is very small, the KK-mode contributions become completely invisible from
our world.
In this paper, we will consider the effects of other interactions of this NG field and
show that we cannot freely have a small value for the brane tension. For this purpose,
we will examine two physical phenomena. One is a new long-range force mediated by the
NG bosons and the other is the cooling process of neutron stars in the supernova explosions.
Both results involve the brane tension parameter f (the decay constant of the NG boson) in
the form 1/f 8 and lead to sizable effects if the tension becomes very small. That is, when
the brane tension becomes small the KK mode couplings are suppressed but instead the
effects of the NG bosons can be measured. Therefore we do not miss the possibilities of
finding signatures of the extra dimensions in near future.
To discuss the effects of the NG bosons, we need to construct the effective action on the
brane. The construction of the action is more or less similar to that in the string theory,
and is straightforward in principle, as was done in Ref. [9]. Although Ref. [9] has given a
definition of the induced vielbein, it gives only a perturbative procedure for finding its form
in terms of the bulk gravity and the brane coordinates (the NG fields). In this paper, we first
give an explicit expression for the induced vielbein. The derivation of the explicit form,
together with its relevance to the non-linear realization theory, are given in Appendix A.
In Section 2, we give the effective action for the system of the d-dimensional brane
embedded in a higher-dimensional spacetime where the gravity and gauge fields live in
the bulk and the matter fields only on the brane. The NG-boson part of this action is
discussed in some detail in Section 3. Given those interactions in the flat background,
we consider several phenomenological effects of the NG field in Section 4. Comparing
303
them with the observations, we can obtain constraints for physical parameters of extra
dimensions such as the brane tension stated above. Section 5 is devoted to summary of this
paper. Appendixes B and C are added for the calculations of the fifth force potential and of
the cross sections for the processes relevant to the star cooling.
2. Setup
The brane action is basically well-known in string theory [10]. So we here explain the
setup briefly which we use throughout this paper, and construct the action for the system,
in particular, by presenting the explicit form of the induced vierbein.
We take the brane to be a (d 1)-brane whose world volume topology is R d . It is
embedded in the bulk spacetime of dimension D (> d) with topology R d T Dd , where
T k denotes a k-dimensional torus. The coordinates of the bulk spacetime are denoted XM ,
and those of the brane world volume are denoted x . The indices of upper-case Roman
letters from the middle, M, N, . . . , run over all dimensions, 0, 1, . . . , (D 1) and the
Greek letter indices , , . . . , run only over the first d dimensions, 0, 1, . . . , (d 1),
while the lower-case Roman letters m, n, . . . , run over the rest (D d)-dimensions,
d, . . . , (D 1); namely, we write like M = (, m), N = (, n), . . . . The local Lorentz
indices are denoted similarly by the corresponding letters from the beginning of the
alphabet, like A = (, a), B = (, b), . . . . Our notation is summarized in Table 1.
The action of this system consists of two parts; the bulk part Sbulk and the brane part
Sbrane . The bulk part action takes the form,

Z
1 MR NS
M D2
D
R G G tr(FMN FRS ) + , (2.1)
d X det E +
Sbulk =
2
4
where , M and R are the cosmological constant, the D-dimensional fundamental scale,
and the D-dimensional scalar curvature, respectively. We have shown only gravity and
YangMills terms explicitly, E A M (X) is the vielbein and AM (X) the YangMills fields.
The bulk metric GMN (X) is given by the vielbein as usual
GMN (X) = AB E A M (X)E B N (X),
(2.2)
Table 1
Summary of our notation
Coordinate
Curved indices
Local Lorentz indices
Bulk spacetime
Brane world volume
XM (M = 0, 1, . . . , D 1)
x ( = 0, 1, . . . , d 1)
0, 1, . . . , (D 1)
0, 1, . . . , (d 1)
d, d + 1, . . . , (D 1)
M, N, . . .
A, B, . . .
, , . . .
, , . . .
m, n, . . .
a, b, . . .
304
where AB is the D-dimensional Minkowski metric: AB = diag(+1, 1, . . ., 1). The

inverse matrices of GMN and E A M are denoted GMN and E M A , respectively.
Now, let Y M (x) denote a point in the bulk spacetime which the point x on the brane
occupies. Considering the distance between two points x and x + dx on the brane and the
parallel transport of a charged field from x to x + dx, we see that the following metric and
gauge fields are induced on the brane from the bulk ones, GMN (X) and AM (X):

g (x) = GMN Y (x) Y M Y N ,

(2.3)
a (x) = AM Y (x) Y M .
Similarly, a vielbein e (x) on the brane is also induced from the bulk one E A M (X). The
definition of this induced vielbein is, however, a bit non-trivial problem as was discussed
by Sundrum [9] and hence is explained in detail in Appendix A. We derive there an
explicit form of the induced vielbein, which we necessarily need in order to discuss the
phenomenological implications of the NG boson. The induced vielbein is given by
1 T
1/2
(2.4)
E E
e (x) = Ek 1 + EkT Ek
,

E M Y (x) , Y M
Ek (x)
,
E a M Y (x) Y M
E a (x)
(2.5)
a
b
T E )
T
1 is the inand (EkT Ek ) = Ek Ek , (E
= E ab E and (Ek Ek )
verse matrix of (EkT Ek ) . This vielbein satisfies the condition, g (x) = e (x)e (x)
and gives a matrix connecting the local Lorentz basis to the curved one on the brane. The
spin connection (x) induced on the brane from the bulk one M A B (X) is also given
in Appendix A. It becomes the usual connection (e) given by the induced vielbein e
if the bulk connection is the usual one M (E) by E A M .
The fields living on the brane generally couple to the bulk fields through these induced
vielbein (or metric) and YangMills fields. Let (x) be a fermion field on the brane which
is charged under the YangMills gauge group. Then the brane part action takes the form
Z
Sbrane =

iga (x) (x)

d x det e + e (x)(x)i
2

m(x)(x)
+ ,
d
(2.6)
( )
with = ( + 4i (x) ). The first term
where
is a cosmological constant term on the brane with standing for the brane tension,
and it also gives the NambuGoto action determining the motion (fluctuation) of the
brane. The ellipsis contains other bosonic part terms for scalars and gauge fields if
there exist such fields living only on the brane. These terms can be easily written
by using the induced metric g (x). It should be noted that the total system with
the action S = Sbulk + Sbrane still keeps the gauge invariance under the bulk general
coordinate, local Lorentz and YangMills gauge transformations. Under the bulk general
305
coordinate transformation, the induced fields e (x), a (x) as well as the genuine field
(x) on the brane transform as scalar fields. This can be checked easily from the
equations (2.3)(2.5) since Y M (x) transforms as bulk vectors. The bulk local Lorentz
transformation results in an SO(1, d 1) local Lorentz transformation on the induced
vielbein e (x) on the brane (see Appendix A), but Sbrane is manifestly invariant under
any SO(1, d 1) local Lorentz transformation. Moreover, under the bulk YangMills
transformation A0M (X) = U (X)AM (X)U 1 (X) + (i/g)U (X)M U 1 (X), the induced
gauge field a (x) is transformed as

(2.7)
a0 (x) = U Y (x) a (x)U 1 Y (x) + (i/g)U Y (x) U 1 Y (x) .
This is the usual gauge transformation in d-dimensions with U (x) = U (Y (x)) under which
Sbrane is manifestly invariant.
The brane action Sbrane also has an invariance under reparametrization of the world
volume coordinates x as is familiar in string theory [10]. We fix this reparametrization
invariance by choosing the gauge condition (the static gauge),
Y M= (x) = x .
(2.8)
Namely, we identify the world volume coordinates x with the first d components of
the brane coordinate Y M (x) in the bulk. The remaining (D d) components Y M=m (x)
represent the brane coordinates transverse to the brane and behave as dynamical scalar
fields on the brane. They are the NG fields appearing as a result of spontaneous breaking
of the translation symmetries transverse to the brane.
If we take a suitable gauge fixing also for the local Lorentz invariance in the bulk, the
induced vielbein can be written in a simpler form. Let us take the following local Lorentz
gauge as is usually adopted in the case of dimensional reduction:

E
0
,
(2.9)
EAM =
E a E a m Bm E a m

E
0
M
,
(2.10)
E A =
E m = Bm E E m a
which can be realized with the local Lorentz SO(1, D 1)/SO(1, d 1) transformations.
Note that E and E m a are the inverse matrices of E and E a m , respectively. In this
local Lorentz gauge (and in the static gauge), we have

(2.11)
Ek (x) = E x, Y m (x) ,
n
a
m
a
(2.12)
E (x) = E n x, Y (x) B (x),

n
n
m
n
(2.13)
B (x) B x, Y (x) + Y (x),
so that from (2.4), the induced vielbein becomes
1/2
.
e (x) = E + G Bm Gmn Bn
(2.14)
In the usual KaluzaKlein gravity, the zero modes of the off-diagonal components
of the higher-dimensional metric, Bm , become massless gauge bosons of U (1)Dd
306
in d dimensions. They correspond to the gauged translational symmetry in the extra

dimension. Since the translations are now spontaneously broken by the existence of the
brane, these gauge fields absorb the NG bosons Y m and become massive on the brane. In
the above expression of the induced vielbein (2.14), we see a part of this Higgs effect; the
gauge fields Bm appear only in the form of Bm = Bm + Y m , which correspond to the
massive gauge fields.
It is interesting to note that even if the bulk gravity is absent, i.e.,
E A M (X) = A M
(GMN (X) = MN ),
(2.15)
the induced vielbein e (x) is non-trivial; e (x) 6= . In this case, the Eq. (2.14) takes
a simple form

1/2
1 m
m
,
(2.16)
e (x) = (x) (x)
where m (x) are (D d) NG scalar fields rescaled with the brane tension factor so that
they carry the mass dimension (d/2 1) of scalar fields;
(2.17)
m (x) Y m (x).
Hereafter, we consider the flat background case, i.e., there is no object in the bulk except
for the brane. We take into account the fluctuations of metric around this background.
In this situation, the expression for Sbulk after the torus compactification and the relevant
phenomenology of graviton have been discussed in Refs. [11]. In the following part of this
paper, we will investigate the physics with the action S = Sbulk + Sbrane , focusing especially
on the role of the NG field m (x).
3. Effective action on the brane

The fluctuation of brane is governed by the NambuGoto term in the brane action
Eq. (2.6), which reads by inserting the form of the induced vierbein (2.16),
Z
d d x det e( )

Z
2
1
1 m
d
(x) m (x)
= d x + m (x) m (x) +
2
8

1 m
m
n
n
(3.1)
(x) (x) (x) (x) + .
4
Note that has a properly normalized kinetic term by the rescaling and hence all
interaction terms of (x) are accompanied by some powers of 1/ . Noting also that the
induced YangMills field (2.3) now reads

1
(x)
(x)
+ Am x,
m (x),
(3.2)
a (x) = A x,
the fermion part of the brane action is given by
307

d d x i m + SNG + Sgrav ,
(3.3)

Z
g
(x)
(x)
Am x,
A x,
+
m (x)
=
d d x g

1 m
m m m
+
2

(x)

+ g A x,
i
2

3
1 m
+ O 2 ,
m m
+
(3.4)
2

Z
(x)
T ,
= d d x h x,
(3.5)
Sfermion =
SNG
Sgrav
where h is the d-dimensional part of the fluctuations of vielbein, E = + h , and

denotes the d-dimensional gravity coupling constant which is related to the fundamental
scale M in (2.1) via M D2 V Dd = 2 . V is the volume of the compactification manifold
T Dd which is (2R)Dd and when d = 4, is related to the Newton constant GN by
2 = 8GN . The energymomentum tensor T is given by

1
1
1
+
, (3.6)
i m + i
T =
2
2
2
assuming h symmetric. Here we have neglected the interaction terms between the
fluctuations of bulk metric and the NG field, which are higher-order terms with respect
to the gravitational coupling constant and 1 .
From the above action, we can see that there are two types of coupling between the NG
boson and the fields on the brane. One is the derivative couplings of (the second
and third lines in (3.4)) which come from the expansion of the induced vielbein. The
derivative coupling is governed by which is the decay constant of this NG field. Another
type of interaction originates from the fact that stand for the coordinates of brane in
the transverse dimensions. To see the form of this type of coupling, we here ignore the
derivative interaction terms of through . Then the brane system is reduced to

Z
1
d d x m m + (i
m )
Sbrane =
2

(x)A x, (x)
,
(3.7)
+ g
aside from the gravity coupling terms. We assume, for simplicity, that the extra dimensions
are all compactified into a tori with a common radius R. The bulk gauge fields AM (X) is
expanded into the KK modes labeled by a (D d) vector n = (nm ):

1 X (n)
AM (x)einY/R .
AM X = x , X m = Y m =
V n
(3.8)
Inserting the KaluzaKlein mode expansion into A , the gauge interaction term reads
308
Z
dd x
X
n

in (x)
.
g (x)
(x)A(n)
(x)
exp
(3.9)
This interaction term apparently implies equal couplings g for all the KaluzaKlein excited
modes. However, though = 0 classically, it has fluctuations quantum mechanically which
is governed by the kinetic term (1/2) m m . We should rewrite the exponential factor
into a normal-ordered form

Z
X 1 n2 (M 1 )
in (x)
d
(n)
2
2
:,
(3.10)
ge R
(x) (x)A (x) :exp
d x
R
n
where is the free propagator of ,
1
1
.
(x y) (x)(y) =
4 2 (x y)2
(3.11)
In deriving this, a singularity at x = y is cut off at the fundamental scale M above which
the effective theory description on the brane becomes invalid. The interaction term (3.10)
implies that the effective coupling of the level n KK modes to d-dimensional fields is
suppressed exponentially. It should be noted that the same exponential factor also appears
in the coupling with the KK gravitons (Sgrav (3.5)). In the following section, we consider
the effects of the NG boson (x) with the derivative interaction terms in (3.4) and the nonderivative one (3.10). We will see that these two interaction terms play complementary
roles in obtaining the limits for the parameters in the models.
We here comment on the coupling involving the bulk gauge filed Am in the action (3.4).
When we perform a compactification from higher D-dimensional theories, the KK modes
(n)
A for the bulk gauge fields get masses and so should absorb the physical degrees of
freedom from some scalar fields. These are just supplied by the KK excitations A(n)
m of the
extra-dimensional components; i.e., the induced gauge field (3.2) can be rewritten in the
form

X
R
i Rn (x)
i Rn (x)
(n)
m (n)
e
i 2 n Am (x)e
A (x)e
+ (zero mode),
a (x) =
n
n6=0
(3.12)
e A + i(R/n2 ) (nm Am ) are massive gauge fields on the brane (with
where A
mass |n/R|) which are invariant (covariant in the non-abelian case) under the gauge
transformation in D dimensions. Since the induced gauge fields only appear in the minimal
interactions to the brane fields, the second term can be absorbed by a redefinition of
(n)
those fields (a gauge transformation). In the following, the KaluzaKlein gauge fields A
(n)
e . Note that in the above equation,
should be understood as this massive gauge field A
the zero-mode part contains the zero modes of extra components Am which remains
massless. Since such massless scalar fields in the adjoint representation have not been
experimentally observed, they must be removed from the low-energy effective theories. Of
course, it is not necessary that gauge fields are living in the bulk unlike the graviton. The
effects of the NG field, which we will discuss below, actually exist even when the gauge
fields are confined on the brane to start with. If one wish to consider bulk gauge fields,
(n)
(n)
(n)
309
however, one should incorporate some mechanism such as non-trivial compactifications in

order to remove the adjoint scalars.
4. Effects of the NG boson

In this section we investigate the effects of the NG field (x) on our four-dimensional
world using the brane action obtained in the previous section. We take the dimension of
our brane d = 4 and denote the number of extra dimensions (D d) . It is also simply
assumed that as in the usual case, the gravity (and sometimes the standard gauge fields,
too) live in the bulk and other matter fields are all confined in the four-dimensional brane.
For simplicity, we use a dimension-one parameter f for the brane tension in the following:
f 4 /4 2 .
(4.1)
Strictly speaking, when we take the bulk gravity turned on, the NG bosons m are
absorbed by the zero modes of the off-diagonal components of bulk metric, Bm , and the
1/2
latter become massive. Their masses are however on the order of GN [12], and very
small compared to the energy scales we will consider. Therefore we need not take into
account this Higgs effect thanks to the equivalence theorem [13], and will treat (and the
graviton zero modes) to be massless fields.
4.1. Suppressions of KK mode couplings
First we review the results of Ref. [8] about the couplings (3.10) between the brane
fields and the bulk ones. As we will see below, the couplings can exponentially reduce
the contributions of higher KK modes and potentially make their effects invisible in our
four-dimensional world.
The form of the interaction term (3.10) now implies that the effective coupling gn of the
level n KK mode to four-dimensional fields is actually suppressed exponentially:
gn g e
12 ( Rn )
2 M2
f4
(4.2)
The origin of this suppression is a recoil effect of the brane. This is easily seen if one notes
that the effective couplings gn can be written as
gn = g h0| e
2i Rn
(x)
f2
|0i,
(4.3)
using the perturbative vacuum |0i of the NG boson . Remembering that (x) stand for
n
the transverse coordinate, the operator ei R (x) is just like the vertex operator in the string
theory and is found to give transverse momentum n/R to the brane around the point x.
2i n
(x)
Hence e R f 2 |0i represents the recoiled state of by the absorption (emission) of

the level n KK mode carrying transverse momentum n/R (n/R). Thus the amplitude
h0|e
2i Rn
(x)
f2
|0i can be regarded as a probability amplitude of containing the original state
|0i in the recoiled state e
2i Rn
(x)
f2
|0i. As is clear from this view, the suppression is stronger
310
for higher KK modes since the larger deformation of brane occurs, and on the other hand,
in the case of stiff brane with larger f , the suppression becomes weaker.
Let us discuss phenomenological consequences of this suppression factor. Consider a
tree-level amplitude for the two charged particles scattering in the brane, which is caused
by the exchange of KK gauge bosons living in the bulk. This process leads to a correction
to the effective four-Fermi coupling constant GF . The summation over all the KK modes
in the amplitude becomes
X

(n)
gn2 A(n)
gn2
2
MW
1
,
+ (n/R)2
(4.4)
where we have assumed the momentum transfer is small compared with MW (and R 1 ). If
the couplings of KK modes were universal, gn = g, this amplitude diverges when the
number of extra dimensions transverse to the brane are greater than one. This means
something wrong because it is merely a tree-level amplitude and the divergence cannot
be renormalized by any means. In the recent analyses on the experimental implications
of such KK modes contributions, the sum has simply been cut off at the fundamental
scale M. However, we see above that the divergences are automatically cured as it should,
by properly taking into account the brane fluctuations (the fluctuations of ). 1
For a numerical estimation, we consider the = 1 case, but it is straightforward to
include more numbers of extra dimensions. The correction (4.4) is dominated by the first
mode contribution (with n = 1) since the higher modes are further suppressed by the
presence of exponential factors, so that the KK mode correction 1GF to the four-Fermi
coupling constant is estimated as
2
g2
MW
1GF
2 n=1
.
2 + 1/R 2
GF
g 2 MW
(4.5)
Since the standard model prediction precisely agrees with experiments, the KK mode
correction (4.5) must be small, say . O(102 ) [15]. For definiteness, we consider R 1
in the range MW < R 1 (< M), then the constraint 1GF /GF < 102 reads
2 2
R e
2MW
M2
R2 f 4
< 102 .
(4.6)
This constraint gives a weak upper bound to the brane tension; f . O(1) TeV, depending
on the value ofM. Clearly, since exp[(M 2 /R 2 f 4 )] < 1, no constraint appears if R 1 is
larger than 10 2 MW 1.1 TeV. Even for R 1 less than this, the constraint (4.6) can be
easily satisfied for any value of M provided that f is chosen suitably small. Therefore if the
brane fluctuation is taken into account, the constraints on the extra dimensions (R, M) so
far obtained can be substantially loosened and sometimes disappear. In the case of > 2,
since the exponential damping factor appears for each extra dimension, it is clear that the
suppression factor becomes (gn /g)2 , and the constraint of f becomes further weak.
1 Similar suppression factors are also discussed in different frameworks [14].
311
Finally, it should be noted that the exponential suppressions discussed above also work
in the couplings of KK gravitons to matter fields on the brane. This fact will have important
effects in our later analyses.
4.2. Fifth force
In the previous subsection, we have shown that the contributions from all KK excited
modes are suppressed if the brane tension f takes a moderately small value. We then
unfortunately could not find signatures of the presence of extra dimensions unless there
is a lower bound for f . In Section 3, we have seen two types of interaction terms for
the NG field (x) involving the coupling f . One is the exponential couplings between
the brane fields and the bulk ones whose effect was discussed before. It only gives upper
bounds of f . Expanding the exponential interaction of (x), we also have the n NG-bosons
coupling whose strength takes the form (1/Rf 2 )n exp[M 2 /R 2 f 4 ]. This factor takes
a maximum value at f M/R, which is independent of f , and so cannot be used

to give a lower bound of f either. On the other hand, another type of couplings in the
Lagrangian (3.4), coming from the expansion of the vierbein, has strength proportional
to f 2 . As the brane tension becomes smaller, its effects become larger and may easily
be detected. In the following two subsections, we will calculate various phenomena which
involve this type of coupling, and show that relatively severe lower bounds of f can be
obtained.
We first discuss a constraint from the so-called fifth force. So far, various types of
new long-range forces mediated by hypothesized very light particles have been suggested
in many theoretical frameworks [16]. Here we consider the strength of the long-range
force which arises as a consequence of exchanges of the NG scalars, which could be
potentially observable. In the macroscopic range, the dominant force working between
neutral systems is gravity. The Newtons universal law of gravitation (the inverse-square
distance dependence) has been experimentally tested up to 1 cm range, and new forces
which are comparable to gravity have been excluded to this range [17]. This fact restricts
the potential forms of new forces and imposes the bounds for new couplings which
determine the potentials. In the present model, this corresponds to a lower bound for f .
As seen from the effective lagrangian on the brane, the NG-mediated force between two
distinguishable standard fermions with masses m and m0 is calculated from the oneloop diagram in the leading order of f 1 (Fig. 1). Note that since a single does
not have tree-level couplings to the standard fermions, this two-particle exchange is the
leading contribution. This is because in the action on the brane, there is an SO() internal
symmetry under which transforms as a vector. So, unlike the case of usual Yukawa force
mediated by massless scalars, the above new force at a distance r is expected to behave
as 1/r n (n > 2). We calculate the above amplitude in the non-relativistic limit and extract
the two-body potential V (r). When we now consider a loop diagram and a higher-order
correction to gravity, a convenient way to compute the two-body potential is the dispersion
theoretical method [18]. This method gives a following relation between the potential V (r)
and the invariant scattering amplitude M obtained from Feynman diagrams,
312
Fig. 1. The lowest-order contribution to the NG boson mediated force.
V (r) =
i
8 2 r

dt Mt exp t r ,
(4.7)
where t denotes the momentum transfer (that is used beyond the physical region) and
Mt is the discontinuity of amplitude across the branch cut on the real t axis; Mt
2i Im M(t + i)|+0 . In order to obtain the expression for long-range potentials we
have only to calculate the t-channel discontinuity and perform a Laplace transformation.
The discontinuity of amplitudes can be evaluated by a simple prescription resulting in the
replacement of each propagator by its discontinuity (the delta function) multiplied by a
theta function to insure a positive energy [18];

1
2i m2 k 2 k 0 .
(4.8)
m2 k 2
For the diagram in Fig. 1, we obtain for Mt in the non-relativistic limit neglecting the
higher-order contributions of three-momenta,
3 imm0 2
t ,
80f 8
where is the number of extra dimensions. Here we have used a formula
Z

d 4 k k 2 (k q)2 k k k k

1
1
q q q q g ( q q ) q 2 + g ( g ) q 4 ,
=
10
8
48
where two totally symmetric sums are defined as
Mt =
(4.9)
(4.10)
g ( q q ) = g q q + g q q + g q q
+ g q q + g q q + g q q ,
g
( )
= g
+g
+g
g .
(4.11)
(4.12)
As a result, we find this force has a potential

3 mm0
,
(4.13)
8f 8 r 7
implying an attractive long-range force. In Appendix B, we show another derivation of this
result by using a conventional method.
V (r) =
313
Before performing numerical comparisons with the experiments, we give several

comments on other possible contributions besides the above potential. First, there is a
contribution from the massive KK gravitons. The deviation from the Newtons law by the
KK graviton mediated force becomes Yukawa-type potentials [19] and can be negligible
in the macroscopic region. On the other hand, when the number of extra dimensions
is two, this force can be as strong as the four-dimensional gravity at sub-millimeter
range for the fundamental scale M ' 1 TeV. (In that case, moreover, the KK graviton
mediated amplitude has a logarithmic dependence on the brane tension f and cannot be
suppressed enough for any small values of f .) However, several astrophysical constraints
have already excluded the region for M, at least, up to O(10) TeV [20] and in addition,
the KK graviton mediated force has a strong dependence of M. So, we can neglect this
KK graviton contributions even at the level of the proposed gravitational experiments.
In the case of > 3, the KK graviton force is much weaker than the gravitation due
to the very short Compton wavelengths of the KK gravitons. The second additional
contribution is the finite-temperature correction. In the long distance range, it could become
comparable to the zero-temperature contribution calculated above, and the potential V (r)
may receive the correction with the factor of order O(1). However, since V (r) has a large
power dependence on the brane tension f , such an O(1) correction is almost irrelevant
in numerically evaluating the constraint for f . Therefore we also neglect this finitetemperature correction.
There are some types of experiments which can test deviations from the inverse-square
law of the Newtonian gravity. The torsion balance experiments give the most stringent
limits on the deviations up to sub-centimeter range and have excluded the presence of new
forces whose strengths are comparable to or stronger than the gravity [21]. Furthermore,
the electromagnetic Casimir force have recently been measured very precisely and also can
impose the constraints on new forces in the range below 104 m [22]. By comparing our
hypothetical force (4.13) with the results of these experiments, we obtain a constraint for f
f > O(0.1) GeV.
(4.14)
Because of the steep slope of V (r) proportional to r 7 , this lower bound mainly comes
from the Casimir force experiments. The proposed classical gravity experiments will
further improve this bound by a factor of O(1).
The bound (4.14) we find here is unfortunately a very weak one. But, it certainly shows
that there exists a lower bound for the brane tension with the considerations of the processes
involving the NG boson (x).
4.3. Energy loss in stars
The fifth force constraint in the previous subsection is too weak to give an enough bound
for the brane tension f . However, there are other processes which actually can lead severer
bounds for f . Since does not have any gauge charges on the brane, its properties can
only be constrained by the missing energy arguments in the collider experiments and
in astrophysics. There, one usually assumes that the energy loss occurs via the standard
314
mechanism in form of photons and/or neutrinos. For example, if there are novel low-mass
particles having weak interactions with matter, they can freely escape from the interior of
stars and carry away their energy. Then it will change the course of the stellar evolution that
would be expected otherwise. With this observation in various astronomical observables
such as the Sun, horizontal branch stars, white dwarfs, supernovas, we can derive various
types of bounds on the coupling strengths in the models under consideration [23]. In this
section, we discuss constraints on the brane tension f required from the cooling process of
the neutron star in supernova explosions. That is, if there were novel particles which freely
stream out of the nascent neutron star, it may rather affect the observed data of neutrino
burst. This will lead to the most stringent bounds from astrophysics. The other collider
signatures of energy loss carried away by the NG field are also important and will be
discussed elsewhere.
We can obtain a rough estimation for this supernova bound by comparing naively with
the axion case. Though this estimation is very rough, we can see that the astrophysical
requirements actually impose severer bounds for the brane tension. The axion coupling
to the ordinary matter is proportional to 1/fa , where fa is the axion decay constant.
The cross section of axion emission from the interior core is proportional to 1/fa2 . The
astrophysical constraints from the supernova observation say that fa should be greater
than 1010 GeV [23]. On the other hand, a typical coupling of the NG boson to matter
3
/f 4 , where TSN is the supernova temperature (available energy
fields are given by TSN
in the supernova medium). If we naively convert the axion bound to the present case, we
obtain
3
TSN
1
f & 101.5 GeV,
< 10
f4
10 GeV
(4.15)
for the supernova temperature TSN O(10) MeV. This shows that, as anticipated, the star
energy loss argument certainly gives a severer bound for the brane tension than that from
the fifth force constraint.
We now perform a detailed analysis of the energy emission rate from the supernova core.
There are several channels which result in energy losses in the supernova medium. We
expect that the most important contribution to this is the nucleonnucleon bremsstrahlung
process like in the axion case [24] because the supernova temperature is near the pion
mass and the large suppressions with the pion mass do not occur. However, this process
is rather involved and there are many types of uncertainties and unknown factors, for
example, with reference to the effective pion description in the supernova medium. Instead,
in this paper we consider a more clean event, i.e., the electronpositron pair annihilation
process. This contribution is surely sub-dominant but can be as large as that of the nucleon
bremsstrahlung when we adopt a higher value of temperature in the supernova. We should
however bear in mind that more stringent bounds may be obtained by accurately taking
into account the dominant process stated above.
Since the energy scale considered is well below the fundamental scale M and the
tension f , there are two diagrams contributing to the energy losses from the electron
positron annihilation in the leading order of the couplings GN and gi (GN is the Newton
315
Fig. 2. The energy loss processes in the supernova medium by the electron-positron annihilation: (a)
the KK graviton production and (b) the NG boson pair production.
constant and gi are the standard gauge couplings). They are annihilations to (a) the KK
(n)
and (b) the NG bosons (Fig. 2). We estimate the initial-spin averaged
gravitons g
cross section for each case (a) and (b). The explicit calculation is given in Appendix C. For
the real KK graviton production process, the cross section is obtained by only attaching an
exponential suppression factor discussed before to the usual result: 2
a =
b =
1 2
2+3 ( 2 ) M
s2 e
+2
s M4
f
(4.16)
3 1 3
s ,
1920 f 8
(4.17)
where s is the center of mass energy in the electronpositron system. We have used a
+2 where V is the volume of the -torus (= (2R) ). We have
relation G1
N = 8V M
also neglected the electron mass for me TSN in the core. It is interesting to note that
the f -dependences of the two cross sections are very different from each other. Typical
behaviors of the cross sections are shown in Fig. 3. It can be seen from this figure that the
total cross section (the solid line in Fig. 3) becomes large for both small and large values of
the brane tension f . The cross section for a small value of f is governed by the NG boson
production process while for a large value, the KK graviton production rate dominates the
total cross section. In either way, it turns out that we cannot have arbitrary values of f and
it will be restricted by phenomenological requirements.
If the present light particles interact strongly with ordinary matter, they are scattered or
absorbed again in the medium after productions in the core. Then they cannot freely escape
to the outside and will be radiated from a sphere. The flux from this black-body surface
emission for the neutron star becomes smaller as the light particles interact more strongly
(the radius of sphere becomes larger). In that case, the region beyond some strong coupling
cannot be excluded from the cooling argument like in the axion case [25]. For the KK
2 If we sum up the inclusive processes e+ + e g + (any numbers of ) in place of the exclusive
process (a), the cross section a may be enhanced by a factor of exp(as 2 /f 4 ) with an O(1) coefficient a, which
is, however, not large enough to cancel the suppression factor exp(sM 2 /f 4 ).
(n)
316
Fig. 3. The typical behaviors of the cross sections against the brane tension f . The dotted and dashed
lines denote a and b , respectively. The solid line is the total cross section.
graviton, this energy reabsorption is, however, highly suppressed. This can be intuitively
understood from the fact that the brane occupies a very tiny (almost zero) region in the
whole bulk. Once the KK gravitons escape to the bulk spacetime from the brane, they will
never return, at least, over the age of the universe [12]. After all, the KK gravitons can be
considered to freely stream out of the core. On the other hand, since (x) is a field on our
four-dimensional brane and does not escape to the extra dimensions, its effect may become
less important than neutrinos for the strong coupling (small tension) region. For this, we
estimate the mean free path L for the field in the core, which is roughly given by
L
f8
6
TSN
ne
(4.18)
where ne is the number density of electron in the core, ne 103 GeV3 . We can see that
L exceeds the radius of the neutron star 10 km for f & O(10) GeV (see, (4.15) or
(4.30)). Then, is emitted from the entire volume of the star and we can surely discuss
the lower bounds of brane tension f . Note that if the couplings were strong and the mean
free path L were smaller than the star radius, other astrophysical constraints and/or too
many experimental signals in the neutrino detectors [26] would reject such strong coupling
regions.
From the calculated cross sections, we extract the production rates in order to compare
them with the observations and to have definite conclusions about the allowed range of
the parameters. However, in the hot dense medium as in the supernova, there are some
uncertainties such as many-body effects, which have not been fully understood. We expect
that those collective effects in the medium amount to change a factor of O(1) in the
production rate. This is actually the case for the axion emission from the supernova [23].
Moreover, as we will see below, the production rates have large power-dependence on the
parameters for which we would like to have limits. So, even if there is an ambiguity even
of factor of 10, the final results for these parameters are not so affected.
317
Now, let us estimate the bounds numerically. The theory of type II supernova explosions
says that in the explosion, almost all of the gravitational binding energy of a nascent
neutron star is released in form of neutrinos within a few second. The expected neutrino
flux and the duration of signal calculated from this picture is in good agreement with the
observations for the supernova 1987A in the Kamiokande II [27] and IMB [28] detectors.
This agreement implies a conservative requirement that the energy loss rate from other than
neutrinos should not exceed that from neutrinos in the standard picture [23]. Otherwise
the measured duration of the neutrino signal would be shorter than the observed ones.
Numerically, this constraint reads
Q . 1036 GeV/cm3 sec.
(4.19)
Here, Q is the energy loss rate per unit time and unit volume (the volume emissivity)
defined as follows;
Z
Z
d 3 k1
d 3 k2
2f1 2f2 (E1 + E2 ) 2sx ,
(4.20)
Qx
(2)3 2E1
(2)3 2E2
where x is the process label, a or b, and fi is the FermiDirac distribution function;
fi = 1/(exp(Ei /T /T ) + 1) (a factor of 2 denotes the spin degrees of freedom). For
in the distribution functions, we use the chemical potential for electron (positron) in the
supernova core. Since the electron is considered as a highly degenerate relativistic particle
in the core, is given by ' (3 2 ne )1/3 . We calculate Q for each process (a) and (b), and
the results become
Qa =
+7
TSN
F (T ),
+2 SN
4 2 +3 ( ) M
(4.21)
13
TSN
I (TSN ).
Qb =
75 f 8
(4.22)
The dimensionless functions F (T ) and I (T ) are given by

ZZ
F (T )
dx dy

(x + y)(xy)/2+2
4xyT 2 M 2 f 4 ,
y+(/T
)
+ 1)(e
+ 1)
(ex(/T )
0 0
ZZ
I (T )
dx dy
0 0
(x + y)(xy)5
,
(ex(/T ) + 1)(ey+(/T ) + 1)
(4.23)
(4.24)
where (z) is defined as

Z1
(z)
dt t /2+1 ezt .
Clearly, the relation +2n (z) = (d/dz)n (z) holds and we have
(4.25)
318

(z) =
1 ez
n+1
,
= 2n,
d
z
dz
Erf( z ), = 2n 1,
z
(4.26)
where Erf (z) is the error function,

2
Erf (z)
Zz
et dt.
2
(4.27)
From the above results, we can obtain two types of bound concerning with extra
dimensions in the present model. This is mainly because the two volume emission rates
have very different parameter dependences as mentioned before (see, Fig. 3). First, we
discuss the lower bound for the brane tension f . As can be seen from Fig. 3, for a small
value of f , the dominant contribution to the energy loss rate is that from the NG boson
production. We apply the supernova constraint (4.19) for the volume emissivity to Qb and
then obtain lower bounds (in the case of = 1);
f > 8 GeV
(TSN = 20 MeV),
(4.28)
f > 23 GeV
..
.
(TSN = 30 MeV),
(4.29)
f > 122 GeV
(TSN = 70 MeV).
(4.30)
Here we have adopted a wide range of the supernova temperature and displayed
corresponding bounds. Note that if there are extra dimensions, each value of the lower
bound is multiplied by 1/8 . It is interesting that the above bounds have no dependence on
the value of the fundamental scale M since the cross section for NG bosons production is
independent of it. The above limits are indeed 1023 times stronger than that from the fifth
force constraint. In other words, with these bounds the long-range force associated with
is outside the reach of the proposed gravity experiments, and it also does not modify the
effects of gravity in stars. In this way, the brane tension can be severely restricted from the
astrophysical argument and the KK mode couplings cannot be arbitrarily suppressed. This
is a welcome result for we could see signatures of the existence of extra dimensions in the
near future experiments.
More interestingly, we can also get a lower bound for the fundamental scale M. As
depicted in Fig. 3, there is an absolute minimum of the total cross section with respect
to the brane tension f . That is, the star energy is necessarily carried away to a certain
extent in form of the KK gravitons and NG bosons. At the minimum, the value of f and
then the minimum value of the energy loss rate are determined by a given value of M.
As a conservative bound for M, we therefore require that the total energy loss rate at the
minimum must not violate the supernova condition (4.19). It is clear that the value of f
minimizing the emission rate satisfies the above-obtained lower bounds, and the resultant
bound for M becomes weaker than that naively estimated in the TeV-scale quantum gravity
scenario [20], i.e., without the coupling suppression by the brane fluctuations. The cross
319
Fig. 4. The volume emissivity Q in the supernova core for = 2 case. The solid, dotted and dashed
line correspond to TSN = 20, 30, 70 MeV, respectively. (Q is denoted in units of 1036 GeV/cm3 sec.)
section for the KK graviton emissions (and hence Qa ) depends on the number of extra
dimensions. It can be easily seen that the = 2 case gives the maximum cross section and
hence we obtain the most restrictive bound. For > 2, the cross sections are considerably
reduced with increasing power of M 1 , and we only obtain very weak lower bounds
for M. The fundamental scale dependences of the energy loss rate at the minimum are
shown in Fig. 4 for the = 2 case. Remarkably, the fundamental scale M is strongly
restricted even when our four-dimensional brane is relatively soft; e.g., M > O(1) TeV for
TSN = 30 MeV. Note that we may have more severe bounds for parameters if we consider
the dominant process of nucleon bremsstrahlung in the supernova core as noted before, and
moreover, other experimentally observable processes may give considerable impacts on the
model parameters. Anyway we cannot have arbitrarily small values for the brane tension
and therefore do not miss the possibility of finding signatures of the extra dimensions.
5. Summary
In this paper, we have investigated the possibility where our four-dimensional world is
embedded as a brane in higher-dimensional spacetime. In such a situation, the translational
symmetry transverse to the brane is spontaneously broken by the presence of brane itself.
As a consequence, the massless NG fields appear which denote the position of brane in
the higher-dimensional bulk. For this setup, the effective action of the four-dimensional
fields is described by using the induced metric and vierbein on the brane. These induced
quantities can be given in terms of the bulk vielbein and the NG variables. We have
found the explicit expression for the induced vielbein and written down the effective
action for fermion fields confined on the brane. Given the interactions, we can discuss
320
phenomenological implications of the brane effective action, especially, several interesting

effects of the NG fields.
First, we discussed the KK mode couplings to the brane. We found in [8] that higher
KK mode contributions are exponentially suppressed with a suitably small tension of our
world, i.e., for a relatively soft brane. However, some other phenomenological observations
can impose limits on this softness. We have shown that severe lower bounds for the brane
tension can be obtained by considering, for example, the null observations of the fifth
force, the missing energy processes in the supernova explosions such as SN1987A, and so
on. The existence of these lower bounds implies that the KK mode contributions cannot
be arbitrarily suppressed and could be observable in the near future experiments as has
been discussed so far in many articles. We have also been able to calculate a lower bound
for the fundamental scale of the effective theory of our brane. More promising and severe
bounds for the physics of extra dimensions (the fundamental scale, the compactification
radius, etc.) would be obtained by studying their signatures in the collider experiments
and other astrophysical and cosmological requirements. Hence it will be an interesting and
challenging subject to find evidences for the existence of extra spacetime dimensions in
form of the KaluzaKlein modes and/or the NG bosons appearing in our four-dimensional
world.
Acknowledgements
We would like to thank M. Bando, K. Hashimoto, T. Hirayama, T. Noguchi, N. Sugiura
for valuable discussions and comments. T. K. and K. Y. are supported in part by the Grantsin-Aid for Scientific Research No. 10640261 and the Grant-in-Aid No. 9161, respectively,
from the Ministry of Education, Science, Sports and Culture, Japan.
Appendix A. Induced vielbein

In this appendix, we derive the exact form of the induced vielbein on the d-dimensional
brane. It is written in terms of the D-dimension bulk vielbein and the (D d) NG variables
denoting the position of d-dimensional brane in the bulk.
The d-dimensional brane breaks the bulk Lorentz group G = SO(1, D 1) down to
H = SO(1, d 1) SO(D d). Consider the coset-space variable ( ) G/H ,

(A.1)
(x) = exp i a (x) J (a) ,
with broken generators J (a). The fields a (x) is the NG boson associated with the
spontaneous breaking of G down to H . Act the bulk Lorentz transformation g G to
the element ( (x)) from the left, then the resultant g( (x)) is still an element of G and
can be decomposed uniquely into the form:

(A.2)
g (x) = 0 (x) h g, (x) , h g, (x) H.
Thus the bulk Lorentz transformation g G for the NG variable (x) is defined by
g

(x) 0 (x) = g (x) h1 g, (x) .
321
(A.3)
An important property of ( (x)) is that the G transformation g is converted through

( (x)) into the corresponding H transformation h(g, (x)).
Up to here everything is the standard story of the nonlinear realization theory [29]. We
wish to define the vielbein e induced on the brane from the bulk vielbein E A M . The
necessary conditions for the desired induced vielbein e to satisfy are that (i) e for
each transforms as a d-dimensional vector under SO(1, d 1); (ii) the metric given by
e should coincides with the induced metric;
g = e e = GMN Y M Y N = AB E A M E B N Y M Y N .
(A.4)
The conversion of the curved index M in the bulk into on the brane can easily be done
using a bulk vector tangent to the brane:
E A = E A M Y M .
(A.5)
This defines d tangent D-dimensional vectors which transform linearly under g

SO(1, D 1). What we now want is e which gives d tangent d-dimensional vectors
of SO(1, d 1). We already know that ( (x)) converts the g G rotation into the h H
rotation; g( (x)) = ( 0 (x))h(g, (x)). Therefore the following quantity defined by
A
(A.6)
eA 1 (x) B E B ,
for each fixed , receives only a (non-linear) H rotation under G transformation,

e0 = 1 0 (x) E 0 = h g, (x) 1 (x) g 1 gE = h g, (x) e .
(A.7)
So, eA splits into a d-dimensional vector e of SO(1, d 1) and a (D d)-dimensional

vector ea of SO(D d). The former d d-dimensional vectors e satisfy the property
(i) which is required for our desired vielbein on the brane. Note here that since 1 ( (x))
itself is an SO(1, D 1) rotation,
AB eA eB = e e + ab ea eb
= AB E A E B = AB E A M E B N Y M Y N .
(A.8)
We find that in order to satisfy the property (ii), the remaining (D d)-dimensional vector
components ea must vanish (see, Eq. (A.4)):
a
(A.9)
ea = 1 (x) B E B = 0.
This condition is the same as given by Sundrum [9]. It should be noted that this requirement
is invariant under G = SO(1, D 1) transformation since ea receives only SO(D d)
rotation under G. The condition (A.9) gives a constraint on the NG variable . Actually, the
number of equations in (A.9) is d(D d) which is just the same as the number of fields.
Thus the NG variables are completely determined in terms of E A = E A M Y M ;
= (E). Although has now become dependent variables (E), the transformation
property under g G still remains the same as above. That is, 0 = (gE) holds because
the constraint (A.9) is G-invariant. With this (E), the desired induced vielbein e is
given by
322
e = 1 (E)
A
AE .
(A.10)
It is a difficult problem to give an explicit form for the solution (E). However, what we
actually need is not (E) but the induced vielbein e . We can find the explicit form for it
as follows.
Using the explicit representation of D-dimensional Lorentz generators (J (AB) )C D =
CE
A B A B ), the coset-space variable can be written as
i (E
D
D E
1 ( ) = exp

= a ,

=
d
Dd
a ab T
C( )
S(
)
Dd
!
S( )
,
C( )
(A.11)
where we have introduced the functions C(x) and S(x) similar to usual cosine and sine:
C(x)
S(x)
X
(1)n
n=0
X
n=0
(2n)!
x n = cos x,
(1)n n sin x
x = .
(2n + 1)!
x
(A.12)
(A.13)
When we write the D-dimensional vectors E A in the following form splitting the parallel
and transverse components to the brane,

Ek
A
,
(A.14)
E =
E a
the constraint (A.9) reads
a
S(
) Ek = C( )a b E b ,
(A.15)
and the induced vielbein (A.10) becomes

e = C( ) Ek + S( ) b E b .
(A.16)
In the above, C( ) , C( )a b and Ek are all square matrices and invertible. We then
obtain from Eq. (A.15),
a
1
(A.17)
T ( ) = E a Ek ,
where T (x) is a function analogous to tangent:
tan x
1
T (x) C (x)S(x) = .
x
(A.18)
In deriving Eq. (A.17), we have used the shifting identities like

( ) = F ( ) ,
F
F ( ) = F ( ),
(A.19)
which generally hold for any function F (x). The equation (A.17) just determines in
terms of E. To obtain an expression of the induced vielbein e , let us define the following
quantity:
T
2 ( ).
M E Ek1 E Ek1 = T ( )T ( ) = T
323
(A.20)
Using xT 2 (x) + 1 = C 2 (x), we further find

C( ) = (1 + M)1/2 .
(A.21)
Thus, the induced vielbein (A.16) is finally rewritten as

e = C( ) 1 + T ( ) E Ek1 Ek
1
= (1 + M) 2 Ek .
(A.22)
This, with the definition of M in Eq. (A.20), gives the desired explicit expression for the
induced vielbein. With the shifting identity, we can also rewrite it into a slightly more
convenient expression as
1/2
(A.23)
e = Ek (1 + N
,

1 T
E E .
N EkT Ek
Finally in this appendix, we add the expression for the induced spin connection (x)
from the bulk one M A B (X). From the bulk local-Lorentz transformation law for the
covariant derivative,

0
(X) = g(X) M + iM (X) g 1 (X),
(A.24)
M + iM
and the fact that is a converter of the local-Lorentz indices from bulk to brane, it is clear
that the quantity (x),

(x) i 1 ( + i ) = 1 A A B (x) B i 1
(A.25)
with A B (x) = M A B (Y (x)) Y M (x), transforms as
(x) 0 (x) = h(x) (x)h1 (x) ih(x) h1 (x),
(A.26)
with h determined by (A.2). This is just the local-Lorentz transformation induced on the
brane, and thus Eq. (A.25) gives the desired induced spin connection on the brane. If the
bulk connection is the usual one M (E) given in terms of the vielbein E A
M as the solution
A E B (M N) = 0, then, the induced connection (A.25)
of the equation M E A N M
N
B
is also seen to equal the usual one (e) given in terms of the induced vierbein e . For
completeness, we cite the explicit expression of 1 in terms of E:

T !
(1 + M)1/2 E Ek1 b
(1 + M)1/2

,
1 (E) =

a
(1 + M0 )1/2 E Ek1
(1 + M0 )1/2a b
(A.27)
where M0 E Ek1 (E Ek1 )T . (Note that F (M0 )E Ek1 = E Ek1 F (M) by the
shifting identity.) is given simply by exchanging the signs of the off-diagonal elements.
324
Appendix B. The potential of the fifth force

Let us first evaluate the amplitude for the diagram in Fig. 1. If the external matter
) in the interaction
fields are on the mass shell, only the part 1/2 ( m m )(i
Lagrangian (3.4) contributes since the remaining parts vanish by the equation of motion
for . Then the amplitude is given by

(p1 + p3 )
(p1 + p3 )
u(p1 ) u(p
u(p2 ) I (q),
4 )
3 )
M = 2 u(p
2
2
4
Z
n k k k (k + q) (k + q) + k k (k + q) (k + q)
d
,
(B.1)
I (q) =
i(2)n
k 2 (k + q)2
where q is the momentum transfer q = p3 p1 = p2 p4 and n = 4 2. The terms in
I (q), in which any vector indices of , , . . . , are carried by q, cannot contribute to
the amplitude because of the conservation of the energymomentum tensor (1/2)u(p
i)
(pj + pi ) u(pj ). So, the only term we have to compute is the one proportional to
g ( g ) g g + g g + g g which comes solely from the k k k k term
in I (q), and we find

I (q)g ( g ) term = ( 2)
Z1
dx
x(1 x)q 2
4
2
1 ( )
g g
2
0
2 2

1
( ) (q )
C ln q 2 ,
g
g
30
64 2
where C is a divergent constant

1 1
1 3 47
+ ln 4 .
C= + +
2 30
(B.2)
(B.3)
( /2)]2[i
( /2)]
This divergent term is absorbed into a term g ( g ) 2[i
which is to appear in the higher-dimensional terms of our effective Lagrangian. (This is the
renormalization of the effective theory la Weinberg.) Thus the factor (C ln(q 2)) is
replaced by a finite one ln(q 2 /2 ) with a suitable renormalization scale .
In the low energy limit, the amplitude is dominated by the = = = = 0
components and reduces to

q2
q2
1
mm0 2 2 2
mm0 1
2 2
q
ln
ln
,
(B.4)
M=
4 2 64 2
30
2
160f 8
2
3 ) 0 ((p1 + p3 )0 /2)u(p1 ) and
where we have used f 4 /4 2 and replaced u(p
0
0
u(p
4 ) ((p2 + p4 ) /2)u(p2 ) by the matter net masses m and m0 , respectively.
To obtain the potential, let us now evaluate the Fourier transform:
Z
d 3 q igr 2 2 q 2
e
q ln 2
v(r)
(2)3
Z
Z
1
eiqr eiqr 4 q 2
1
2
q ln 2 = 2 Im dq eiqr q 5 ln(q/).
= 2 q dq
(B.5)
iqr
r
0
325
This is divergent in the ultraviolet region q . However, the amplitude M q 4 ln q 2

is reliable only in the infrared region and the true amplitude should be suppressed well
in the ultraviolet region. Supposing so, we put here a suppression factor eq , or making
a replacement r r + i, in the integrand and so understand that the obtained result is
reliable only for large r.
Then, we can rotate the integration contour 0 6 q 6 counterclockwise into that along
the imaginary axis, 0 6 6 with q = i since the contribution from the quarter circle
at infinity vanishes thanks to the factor eiqrq . Putting = 0, we then obtain
1
v(r) = 2 Im
r
d er 5 i 6 ln(i/)
1
= 2
r
Z
d e

60 1
=

.
2
r7
r 5
(B.6)
Thus the desired potential is found to be

3 mm0
60 1
mm0 2
=
,
(B.7)
V (r) =
r7
160f 8
8f 8 r 7
where the overall negative sign has been put since the amplitude M corresponds to the
effective action which has opposite sign to the potential energy.
Appendix C. Calculation of the cross sections a and b

The amplitude for the production process (a) in Fig. 2 of the level n KK graviton h(n)
is
calculated using the gravity interaction term (3.5) and is given by
(p1 p2 )
u(p1 , s1 ),
2 , s2 )
(C.1)
Mn = n (k, )v(p
2
2
) is the nth KK graviton coupling strength suppressed by

where n exp( 12 ( Rn )2 M
f4
the brane fluctuation effect, and is the polarization tensor of the massive KK graviton.
The initial-spin averaged cross section in the center-of-mass frame is given by the standard
formula:
1P
2
X
s1 ,s2 , |Mn |
2( s k0 ) 4 p
,
(C.2)
a =
4s s 4m2e
n
P
where s = (p1 + p2 )2 , and n denotes the summation over the level number n of the final
KK graviton. Henceforth we will set the small electron mass me equal to zero. The spin
sum for the final massive KK graviton is done by using

2
X
1
2

+ + ,
(k, ) (k, ) =
(C.3)
2
3
=2
where the ellipses denote the terms containing k k /m2n (, = , , , ), which all do
not contribute to the cross section because of the energymomentum tensor conservation.
326
The third term (2/3) does not contribute either since the energymomentum tensor
is traceless when me = 0. Thus we obtain
2
1 X
|Mn |2 = n s 2 .
4
8
(C.4)
s1 ,s2 ,
The mass of the level n KK graviton is given by m2n = n2 /R 2 which equals k02 in the
P
center-of-mass
frame,
and so the summation n over n can be replaced by the integral
R
R
R d k0 = R k01 dk0 with the solid angle = 2 /2 / ( 2 ) in -dimensions.
Using also the relation 2 = (2R) M +2 , we finally get
1 2
s /2 s Mf 42
e
.
2+3 ( 2 ) M +2
a =
(C.5)
The amplitude for the NG boson pair production process (b) in Fig. 2 is calculated using
the interaction term (3.4) and is given by
1
(p1 p2 )

k1 k2 k1 k2 v(p
u(p 1 , s1 ).
(C.6)
2 , s2 )
2
2
The initial-spin averaged cross section in the center-of-mass frame (p 1 = p2 p,
k 1 = k 2 k) is given by the standard formula:
Z
d |k| 1 X
1
|M |2 ,
(C.7)
b =
2
64 2 s|p| 4 s ,s
M =
where the factor 1/2 in front is put because the final two particles are of the same kind,
and the factor comes from the sum over the index m of the NG bosons m . Since we are
neglecting me presently, we have |k| = |p|, and

1
1X
|M |2 =
s 4 cos2 1 cos2 ,
(C.8)
2
4 s ,s
128
1
with cos p k/|p||k|. Putting = f 4 /4 2 , we find the cross section to be

b =
3 s 3
.
1920 f 8
(C.9)
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
J.D. Lykken, Phys. Rev. D 54 (1996) 3693;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257;
N. Arkani-Hamed, S. Dimopoulos, J. March-Russell, hep-th/9809124.
[2] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47;
S. Abel, S. King, Phys. Rev. D 59 (1999) 095010;
N. Arkani-Hamed, S. Dimopoulos, hep-ph/9811353;
N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005;
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
327
H.-C. Cheng, Phys. Rev. D 60 (1999) 075015;

K. Yoshioka, Mod. Phys. Lett. A 15 (2000) 29.
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 557 (1999) 25;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, J. March-Russell, hep-ph/9811448;
A.E. Faraggi, M. Pospelov, Phys. Lett. B 458 (1999) 237;
G. Dvali, A.Yu. Smirnov, Nucl. Phys. B 563 (1999) 63;
R.N. Mohapatra, S. Nandi, A. Prez-Lorenzana, Phys. Lett. B 466 (1999) 115.
I. Antoniadis, C. Muoz, M. Quirs, Nucl. Phys. B 397 (1993) 515;
E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503.
Z. Berezhiani, G. Dvali, Phys. Lett. B 450 (1999) 24;
Z. Kakushadze, Nucl. Phys. B 551 (1999) 549;
H.-C. Cheng, K.T. Matchev, Nucl. Phys. B 563 (1999) 21;
Y. Sakamura, hep-ph/9909454.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004;
M. Maggiore, A. Riotto, Nucl. Phys. B 548 (1999) 427;
N. Kaloper, A. Linde, Phys. Rev. D 59 (1999) 101303;
N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, J. March-Russell, Nucl. Phys. B 567 (2000)
189, hep-ph/9903239;
A. Riotto, Phys. Rev. D 61 (2000) 123506.
G. Shiu, S.H.H. Tye, Phys. Rev. D 58 (1998) 106007;
Z. Kakushadze, S.H.H. Tye, Nucl. Phys. B 548 (1999) 180;
I. Antoniadis, C. Bachas, Phys. Lett. B 450 (1999) 83;
Z. Kakushadze, Nucl. Phys. B 552 (1999) 3;
G. Shiu, R. Shrock, S.H.H. Tye, Phys. Lett. B 458 (1999) 274;
E.E. Flanagan, S.H.H. Tye, I. Wasserman, Phys. Rev. D 62 (2000) 024011.
M. Bando, T. Kugo, T. Noguchi, K. Yoshioka, Phys. Rev. Lett. 83 (1999) 3601.
R. Sundrum, Phys. Rev. D 59 (1999) 085009.
J. Polchinski, String Theory, Vol. 1, 2, Cambridge Univ. Press, Cambridge, 1998.
For example, G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3;
S. Nussinov, R. Shrock, Phys. Rev. D 59 (1999) 105002;
E.A. Mirabelli, M. Perelstein, M.E. Peskin, Phys. Rev. Lett. 82 (1999) 2236;
T. Han, J.D. Lykken, R.-J. Zhang, Phys. Rev. D 59 (1999) 105006;
J.L. Hewett, Phys. Rev. Lett. 82 (1999) 4765;
P. Mathews et al., Phys. Lett. B 450 (1999) 343;
P. Mathews et al., JHEP 0007 (2000) 008;
T.G. Rizzo, Phys. Rev. D 59 (1999) 115010;
T.G. Rizzo, Phys. Rev. D 60 (1999) 075001;
T.G. Rizzo, Phys. Rev. D 60 (1999) 115010, hep-ph/9910255;
K. Cheung, W.-Y. Keung, Phys. Rev. D 60 (1999) 112003;
K. Cheung, Phys. Rev. D 61 (2000) 015005;
D. Atwood, S. Bar-Shalom, A. Soni, hep-ph/9903538;
D. Atwood, S. Bar-Shalom, A. Soni, hep-ph/9906400;
D. Atwood, S. Bar-Shalom, A. Soni, Phys. Rev. D 61 (2000) 116011;
D. Atwood, S. Bar-Shalom, A. Soni, Phys. Rev. D 62 (2000) 056008.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
M.S. Chanowitz, M.K. Gaillard, Nucl. Phys. B 261 (1985) 379;
K.-I. Aoki, Talk presented at meeting on Physics at TeV Energy Scale, Tsukuba, May, 1987,
(QCD161:M376:1987).
328
[14] I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69;

J. Hisano, N. Okada, Phys. Rev. D 61 (2000) 106003.
[15] Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
[16] G.B. Gelmini, S. Nussinov, T. Yanagida, Nucl. Phys. B 219 (1983) 31;
J.E. Moody, F. Wilczek, Phys. Rev. D 30 (1984) 130;
T.R. Taylor, G. Veneziano, Phys. Lett. B 213 (1988) 450;
S. Dimopoulos, G.F. Giudice, Phys. Lett. B 379 (1986) 105;
I. Antoniadis, S. Dimopoulos, G. Dvali, Nucl. Phys. B 516 (1998) 70.
[17] J.C. Long, H.W. Chan, J.C. Price, Nucl. Phys. B 539 (1999) 23.
[18] G. Feinberg, J. Sucher, C.-K. Au, Phys. Rep. 180 (1989) 83.
[19] A. Kehagias, K. Sfetsos, Phys. Lett. B 472 (2000) 39;
E.G. Floratos, G.K. Leontaris, Phys. Lett. B 465 (1999) 95.
[20] S. Cullen, M. Perelstein, Phys. Rev. Lett. 83 (1999) 268;
V. Barger, T. Han, C. Kao, R.-J. Zhang, Phys. Lett. B 461 (1999) 34.
[21] J.K. Hoskins, R.D. Newman, R. Spero, J. Schultz, Phys. Rev. D 32 (1985) 3084;
V.P. Mitrofanov, O.I. Ponomareva, Sov. Phys. JETP 67 (1998) 1963.
[22] S.K. Lamoreaux, Phys. Rev. Lett. 78 (1997) 5.
[23] For a review, G.G. Raffelt, Phys. Rep. 198 (1990) 1, hep-ph/9903472.
[24] J. Ellis, K.A. Olive, Nucl. Phys. B 223 (1983) 252;
N. Iwamoto, Phys. Rev. Lett. 53 (1984) 1198;
R. Mayle et al., Phys. Lett. B 203 (1988) 188;
R. Mayle et al., Phys. Lett. B 219 (1989) 515;
G. Raffelt, D. Seckel, Phys. Rev. Lett. 60 (1988) 1793.
[25] M.S. Turner, Phys. Rev. Lett. 60 (1988) 1797;
A. Burrows, M.T. Ressell, M.S. Turner, Phys. Rev. D 42 (1990) 3297.
[26] J. Engel, D. Seckel, A.C. Hayes, Phys. Rev. Lett. 65 (1990) 960.
[27] K. Hirata et al., Phys. Rev. Lett. 58 (1987) 1490.
[28] R.M. Bionta et al., Phys. Rev. Lett. 58 (1987) 1494.
[29] M. Bando, T. Kugo, Y. Yamawaki, Phys. Rep. 164 (1988) 217, and references therein.

Holonomies, anomalies and the FeffermanGraham

ambiguity in AdS3 gravity
M. Rooman a, , Ph. Spindel b
a Service de Physique Thorique, Universit Libre de Bruxelles, Campus Plaine, C.P.225 Boulevard du
Triomphe, B-1050 Bruxelles, Belgium

b Mcanique et Gravitation, Universit de Mons-Hainaut, 20 Place du Parc, 7000 Mons, Belgium
Received 1 August 2000; accepted 25 October 2000
Abstract
Using the ChernSimons formulation of (2 + 1)-gravity, we derive, for the general asymptotic
metrics given by the FeffermanGrahamLee theorems, the emergence of the Liouville mode
associated to the boundary degrees of freedom of (2 + 1)-dimensional anti-de-Sitter geometries.
Holonomies are described through multi-valued gauge and Liouville fields and are found to
algebraically couple the fields defined on the disconnected components of spatial infinity. In the
case of flat boundary metrics, explicit expressions are obtained for the fields and holonomies. We
also show the link between the variation under diffeomorphisms of the Einstein theory of gravitation
and the Weyl anomaly of the conformal theory at infinity. 2001 Elsevier Science B.V. All rights
reserved.
PACS: 11.10.Kk; 04.20.Ha
Keywords: Anti de Sitter; ChernSimons; Liouville
1. Introduction
The interest of studying (2 + 1)-dimensional gravity has initially been emphasized in [1]
and have recently been revived with the discovery of black holes in spaces with negative
cosmological constant [2]. Since then, a large number of studies has been devoted to the
elucidation of classical as well as to quantum (2 + 1)-gravity [3]. The major problem
resides in the quest for the origin of the black hole entropy. This entropy seems due to a
macroscopically large number of physical degrees of freedom, as indicated by the value of
the central charge first computed by [4].
E-mail addresses: mrooman@ulb.ac.be (M. Rooman), spindel@umh.ac.be (Ph. Spindel).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 6 - 2
330
M. Rooman, Ph. Spindel / Nuclear Physics B 594 (2001) 329353
In (2 + 1)-dimensions, the field equations of gravity with negative cosmological constant

have been proven to be equivalent to those of a ChernSimons (CS) theory with SL(2, R)
SL(2, R) as gauge group [5]. Assuming the boundary of the space to be a flat cylinder
R S 1 , Coussaert, Henneaux and van Driel (CHD) [6] demonstrated the equivalence
between this CS theory and a non-chiral WessZuminoWitten (WZW) theory [7,8], and
showed that the AdS3 boundary conditions as defined in [4] implement the constraints that
reduce the WZW model to the Liouville theory [9].
In this paper, we show that, using the less restrictive AdS boundary conditions allowed
by the FeffermanGrahamLee theorems [10,11], the CHD analysis can be extended
and leads to the Liouville theory formulated on a 2-dimensional curved background
(a preliminary version of this analysis can be found in [12]).
In Section 2 we recall the FeffermanGraham (FG) asymptotic expansion of Einstein
and AdS3 metrics. Section 3 contains the main calculation of the paper, consisting of a
generalization of the relation between 3-d gravity and Liouville theory to curved boundary
metrics. Special attention is paid to the contributions from holonomies, when the topology
of the space is cylindrical. An originality of our approach resides in the use of multivalued gauge group elements, which allows to simply show the correspondence between
the EinsteinHilbert (EH) action and a multi-valued Liouville field action.
In Section 4, we discuss the coupling between the fields living on the different
components of spatial infinity, which is induced by the holonomies. We show that this
coupling reduces to simple algebraic conditions on-shell. The aim of Section 5 is to
illustrate the link between Liouville fields, holonomies, AdS3 metrics and SL(2, R) parallel
transport matrices in the framework of flat asymptotic boundaries.
Finally, in Section 6, by taking into account all terms resulting from integrations by
parts during the reduction process from the EH action to the Liouville action, performed in
Section 3, we make explicitly the connection between the variance of the EH action under
diffeomorphisms and Weyl transformations on the boundary of AdS3 space.
2. Asymptotically anti-de-Sitter spaces

Graham and Lee [10] proved that, under suitable topological assumptions, Euclidean
Einstein spaces with negative cosmological constant are completely defined by the
geometry on their boundary. Furthermore, Fefferman and Graham (FG) [11] showed that,
whatever the signature, there exists an asymptotic expansion of the metric, which formally
solves the Einstein equations with < 0. The first terms of this expansion may be given
by even powers of a radial coordinate r:
ds 2 ' `2
r
(2)
dr 2 r 2 (0) i
+ 2 g (x ) + g (x i ) + .
2
r
`
(1)
On (n + 1)-dimensional spacetimes, the full asymptotic expansion continues with terms

n+1
of negative even powers of r up to r 2([ 2 ]2) , with in addition a logarithmic term
of the order of r (n2) log r when n is even and larger than 2. All these terms are
331
(0)
completely defined by the boundary geometry g (x i ). It seems thus natural to take them
as the definition of asymptotic Einstein metrics. These terms are followed by terms of
1
negative powers starting from r 2([ 2 (n+1)]1) ; for even n, the trace-free part of the r (n2)
(0)
coefficient is not fully determined by g . It contains degrees of freedom in Lorentzian

spaces [13,14]. Once this ambiguity is fixed, all subsequent terms become determined.
It is instructive to look at the first iterations of this expansion. We write the metric in
terms of forms as ds 2 = 0 0 + ab a b with , [resp. a, b] running from 0
to n [resp. 1 to n] and ab a flat n-dimensional minkowskian metric diag(1, . . ., 1, 1). The
forms and the Levi-Civita connection read as: 1

r
`
a = a + a + o r 2 ,
`
r
0 = `
dr
,
r
a0 =

r a 1 a
2
+
o
r
,
`2
r
(2)

ab = ab + o r 1 ,
(3)
where the forms a , ab ab c c , a a b b and a b a b are r-independent. These

provide the dominant and sub-dominant terms of the metric expansion:
(0)
g = ab a b ,
(2)
(2)
g g ab a b = (ab + ba ) a b .
(4)
Here and in what follows, the n-dimensional indices and the covariant derivatives are
(0)
defined with respect to the metric g . Using these definitions, the (n + 1)-dimensional
Riemann curvature 2-form R becomes:

1
1
a 0 d a + ab b + o r 2 ,
r
`2
(0)

1
1
= 2 a b + R ab + 2 a b + a b + o r 1 ,
`
`
R a0 =
(5)
R ab
(6)
(0)
(0)
where R ab is the n-dimensional curvature 2-form defined by the metric g , and

(2)
a g ab b . If we impose the metric of the (n+1)-dimensional space to be asymptotically

2
, these equations, at order r 2 , fix:
Einsteinian, i.e., R = n1
= 1/`2.
(7)
r 1 ,
they yield:
Moreover, at order 1 and
(0)
(2) i
(2)
1h
c
R ab + 2 (n 2) g ab + ab g c = 0,
`
(2)
(2)
g bb;a g ba;b
= 0,
(0)
where R ab are the components of the n-dimensional Ricci tensor.

1 By definition, o(r n ) = u (r) means that lim
n
n
r un (r)/r = 0.
(8)
(9)
332
We define asymptotically AdS spaces by the stronger condition that the Riemann tensor
tends to that of AdS, i.e., R = . This implies, in addition to Eq. (7):

(0)
(2)
(2)
(2)
(2)
1
(10)
R abcd = 2 ad g bc ac g bd + bd g ac bc g ad ,
`
(2)
g ab;c
(2)
g ac;b = 0,
(11)
whose traces are the asymptotic Einstein conditions (8), (9). When n 6= 2, Eq. (8) fully
(2)
(0)
specifies the metric g in terms of g and Eq. (9) becomes the Bianchi identity satisfied by
the n-dimensional Einstein tensor. Furthermore Eq. (10) is an identity if n = 3 and implies
that the Weyl tensor of the n-dimensional geometry vanishes if n > 3, while Eq. (11)
implies that the CottonYork tensor vanishes for n = 3 and becomes a consequence of
Eq. (10) and the Bianchi identities for n > 3. Thus, AdS asymptotic spaces are Einstein
(0)
asymptotic spaces whose g metric tensor is conformally flat. The same conclusion holds
for n = 2 as in three dimensions Einstein spaces with < 0 are locally AdS and metrics
on cylindrical boundaries are conformally flat.
(2)
On the other hand, when n = 2 only the trace of g is fixed by Eq. (8):
(2)
g cc
= 2cc 2 =
`2 (0)
R,
2
(12)
(2)
and the other components of g have only to satisfy the equations:

`2 (0)
(13)
R ,b .
2
The subdominant metric components are thus not all determined by the asymptotic metric
in three dimensions, but there remains one degree of freedom, which we shall explicit in the
(2)
g ab;a
(n)
next section. Similar indeterminacy, involving g , arises for all even values of n [13,14].
To illustrate the meaning of the FG coordinates, let us consider the special case of BTZ
black holes [2] whose metric can locally be written as:
2

d 2
+ 2 d 2 + J d dt,
(14)
ds 2 = 2 M dt 2 + 2
J2
`
M
+
2
2
`
4
where M is the mass and J the angular momentum. To shorten the discussion, we restrict
ourselves to the cases M > |J /`| > 0. In this case, a single coordinate patch (, , t), on
which the metric is given by (14), covers the three regions denoted I, II and III (see Fig. 1),
corresponding to > + , + > > and < , where
s
s
!
!
2
`
J
`
J2
2
2
M+ M 2
M M 2 ,
+ =
and
=
(15)
2
`
2
`
are the values of the constant surfaces defining the inner and outer horizons. The
complete spacetime is obtained by glueing similar overlapping coordinate patches.
333
Fig. 1. Penrose diagram of BTZ black hole with M > |J /`| and domains of definition of the radial
FG coordinate.
The FG radial coordinate r is obtained from by the transformations:

s
!
2
2
2
1
M`
J
2
`
M + 2 ,
r2 =
2
2
`2
4
(16)
yielding the metric expression:

M 2 2

dr 2 r 2 2 2
+ 2 ` d dt 2 +
` d + dt 2 + J dt d
2
2
r
`

M 2 `2 J 2 2 2
` d dt 2 .
(17)
+
16r 2
This r coordinate is only defined on region I of a coordinate patch (, , t), between the
exterior horizon = + and infinity. It can be analytically continued
p so as to cover two
adjacent regions I and I0 (see Fig. 1). On the first (I), we have r 2 > 4` M 2 J 2 /`2 , which
p
is obtained by taking the plus sign in Eq. (16); on the other (I0 ) r 2 < 4` M 2 J 2 /`2 ,
corresponding to the minus sign. This r coordinate accounts for the global nature of the
black holes, where the diffeomorphic regions I and I0 are connected by an EinsteinRosen
bridge. Near the space-like infinity of region I (r ) the metric (17) coincides with
the FG asymptotic expansion (1), whereas on region I0 , where the space-like asymptotic
ds 2 = `2
334
region is given by r 0, we recover the FG expression of the asymptotic metric after the
transformation r 2 (M 2 `2 J 2 )`2 /(16r 2 ). This transformation leaves the metric (17)
invariant, ensuring the consistency of the definitions of mass and angular momentum of
the black hole, whatever the asymptotic region is. Of course we may define a single radial
coordinate that goes to on the asymptotic of regions I and I0 and leads on both ends
to the FG expression (1); it is however not related analytically to the r coordinate (16). We
shall use such a coordinate (also called r) in Section 3.
Furthermore, if we introduce the radial variable y = r 2 , corresponding to the original
FG radial coordinate [11], we may extend its domain of definition to negative values, which
correspond to regions III and III0 . Note that, as r has been taken to be intrinsically spacelike, it cannot cover regions II, between the horizons, unless we turn it into a time-like
coordinate.
3. From EinsteinHilbert to Liouville action

In this section we perform explicitly the transformations leading from the EH action
to the Liouville action, assuming the metric to be asymptotically AdS as defined in the
previous section. Our first aim is to show how the asymptotic conditions generalize in case
of curved metrics, and lead to a Liouville action on curved background. Our second aim is
to keep track of all kinds of boundary contributions that appear during the transformations
(which renders this section rather long and detailed), in view of accounting for the effect
of holonomies (see Section 4) and showing the precise link between the EH action which
is invariant under the combined action of diffeomorphisms and Weyl transformations and
the Liouville action which is not (see Section 6).
3.1. First step: ChernSimons action
The EH (2 + 1)-gravity action with cosmological constant , evaluated on a spacetime
domain V, consists of the bulk term:
Z
1
(R 2),
(18)
SEH =
16G
V
where is the volume element 3-form on V.

Let us specify the topological structure of the domain of integration V. Inspired by the
BTZ black holes [2], we assume it to consist of the product of a time interval [t0 , t1 ] with
a space-like section chosen to be homeomorphic to a disk of radius r or to a finite
2-dimensional space-like cylinder of length 2r ; afterwards r will be pushed to infinity. The
disk will provide extremal black holes if we accept a conical singularity at the origin r = 0,
and global AdS space if not. The cylinder will describe non-extremal black holes, its non
trivial topology describing the EinsteinRosen bridge. In fact, we may not exclude a priori
more complicated topologies, such as several cylinders attached to an arbitrary compact
manifold, but we shall not consider them here.
335
We use usual cylindrical coordinates (r, , t) to parametrize the manifold V, where |r|
coincides with the FG radial variable in a neighbourhood of |r| = . Accordingly, the
coordinate domain of integration will be the cube [t0 , t1 ] [rL , rR ] [0, 2]. In case
is a disk, rL = 0, rR = r , and all the considered quantities are periodic in the angular
variable . In case is a cylinder, we choose rL = r and rR = r ; to account for the
holonomies allowed by this non trivial topology, we may not assume a priori all quantities
to be periodic in . Hence, the boundary terms obtained upon application of Stokes theorem
will involve when is a disk the geometrical surfaces t = t0 , t = t1 and r = r , with in
addition when is a cylinder the extra boundary component r = r and the coordinate
surfaces = 0 and = 2 .
When < 0, one can re-express [5] the EH action (18) in terms of the two gauge fields
= A = J A , with J generators of the sl(2, R) algebra
A = A = J A and A
satisfying Tr(J J ) = 12 and Tr(J J J ) = 14 , with 012 = 1 (see Appendix A for
conventions). These fields are given in terms of the metric by:
1
1
1
1
A = + .
(19)
A = + ,
`
2
`
2
They allow to express the action SEH as the difference of two CS actions SSC [A] and
plus a boundary term that mixes the two gauge fields:
SSC [A]
SEH = SCS + BCS ,
with
`
S[A] =
16G
and
BCS =
`
16G
Z
V
SCS = S[A] S[A],
(20)

2
Tr A dA + A A A ,
3
(21)

A .
Tr d A
(22)
V
(r)
()
(t )
The integral BCS is composed of three terms, noted BCS , BCS and BCS , coming from
integrations over r, and t, respectively. As we assume that the metric and thus the fields
are globally defined on V, i.e., periodic in , B () vanishes.
A and A
CS
We now discuss the large r behaviour of the EH and CS actions (18), (20). The insertion
But
of Eqs. (2), (3) in Eq. (19) gives the asymptotic behaviour of the CS fields A and A.
before we pursue our discussion, the following clarifying remark seems necessary. The FG
expression of the metric is, in general, only locally valid. When is a cylinder there are
two asymptotic regions and we may always fix the asymptotic behaviour of the fields on
one of them according to Eqs. (2), (19). With such a choice, the frames { 0 , a } on the
other asymptotic region become defined up to a sign that depends on the continuation of the
may indeed
CS fields across the whole manifold. The asymptotic behaviours of A and A
have to be exchanged (see Section 5 for an explicit example), but this has no consequence
on the rest of our analysis.
Let us in a first stage focus on one asymptotic region. To fix the notation, we adopt
the convention defined by Eqs. (2), (19), with the radial coordinate r belonging to a
336
neighbourhood of +. We find convenient to write explicitly the large r behaviour of the

CS fields using the null frame = 1 2 and its dual vectorial frame eE = 12 (Ee1 eE2 ),
which are defined on the surfaces |r| = r . At order r 1 , we obtain:

1 1 0
(23)
' A r ,
Ar '
r 2r
0 1 r

+
0
2r
e
2
2
K ,
K+ ,
A + '
(24)
A '
+

r
r
2
2r
2
A+
r
'
r
+
+ +
r
+
`2
r
2
(2)
g ++
+
2
' (2)
r
g
r
r + +
+
`2
r
(25)
where we have introduced the null components of the connection 2-form 12 =

e+ . Note that the
+ + + and for further convenience the notation K and K
combination (+ + )/r can always be canceled by an infinitesimal (in the limit
r ) Lorentz transformation acting on 1 and 2 ; this gauge freedom explains that
this term never appears in the subsequent calculations. Using these relations, we see that in
the large r limit, the EH action (18) presents two types of divergences, one in r 2 and if ,
(0)
i.e., R (see Eq. (12)), does not vanish, another in log r . In contrast, the CS action (20) only
(r)
. Indeed,
presents a log r divergence. The quadratic divergence of SEH is found in BCS
q
Z
(0)

r 2
(r)
g d dt + O r 2 .
=
(26)
BCS
3
8G`
|r|=r
Note that in the limit r , this term does not contain dynamical degrees of freedom.
(t )
, which can be written as:
The logarithmic divergence of SEH appears in both SCS and BCS
q
Z
(0) (0)
`
(t )
log r
(27)
= '
BCS
R g d dt.
r 32G
|r|=r
This term will no longer be considered in this section but discussed in Section 6, where the
origin of the anomaly will be analyzed.
(r)
As a remark en passant, let us note that the boundary contribution BCS (26) can here
be expressed in terms of the extrinsic curvature 2-form K, defined as the product of the
337
trace of the extrinsic curvature tensor with the induced surface element 2-form. Indeed, on
a surface of constant r, this 2-form is given by:
1
,
2`
which implies that on this surface:

= K.
` Tr A A
K=
(28)
(29)
Hence, we may introduce a gravity action SG suitable for Feynman path integral, which is
finite when the boundary metric is flat (and has well-defined functional derivatives) [15]:
Z
Z
1
1
(R 2) +
K,
(30)
SG =
16G
16G
M
|r|=r
This gravity action differs from that obtained using the GibbonsHawking procedure [16],
as the term involving the extrinsic curvature is equal to half of the usual one [17]. This is
not in contradiction with [16], as we are working in the Palatini formalism.
Let us now consider the on-shell variation of the actions SEH and SCS . We therefore
re-write S[A] (21) as:
Z

`
Tr 2 At Fr + A A r Ar A dr d dt + B (r)[A],
(31)
S[A] =
16G
V
where the boundary term is:

Z
`
Tr(At A ) d dt.
B (r)[A] =
16G
(32)
|r|=r
given by Eqs. (23), (24), (25), we

Using the asymptotic behaviour of the fields A and A
find that at spatial infinity (|r| = r ):

(33)
A = O r 2 = A + ,

2
= Tr A + A ,
(34)
Tr(A A+ ) = O r

2
= Tr A+ A .
(35)
Tr A A+ = O r
Accordingly, the variations of both actions SEH and SCS vanish:
Z
Z

O r 2 d dt,
O r 2 d dt.
SEH =
SCS =
|r|=r
(36)
|r|=r
It is noteworthy that, owing to the boundary conditions (24), (25), the variation of the action
SCS vanishes by itself, without needing to add any extra boundary term [18], contrary to
what was sometimes stated.
3.2. Second step: WessZuminoWitten action
Let us now return to the action SCS given by Eqs. (20), (31). The time components At
and A t play the rle of Lagrange multipliers and can be eliminated from the bulk action
338
er = 0. On the cylinder, their general

by solving the constraint equations Fr = 0 and F
solutions [8] are given by gauge transforming non trivial flat connections hi and h i :
1
Ai = G1
1 hi G1 + G1 i G1 ,
A i = G1
2 hi G2 + G2 i G2 ,
(37)
with i labelling the coordinates (r, ). The flat connection components can be chosen as
where h(t) and h(t)

are sl(2, R) generators
hr = 0, h r = 0, h = h(t) and h = h(t),
that only depend t. The SL(2, R)/Z2 matrices G1 and G2 are assumed univoquely defined
on the cylinder: G1 (r, , t) = G1 (r, + 2, t) and similarly for G2 . Instead of using this
we choose to express them in terms of non-periodic group
representation of A and A,
elements Q1 and Q2 :
Ai = Q1
1 i Q1 ,
A i = Q1
2 i Q2 ,
(38)
where

Q1 (r, , t) = exp h(t) G1 (r, , t),

Q2 (r, , t) = exp h(t)

G2 (r, , t). (39)
These two representations of the gauge fields are obviously equivalent. We choose to use
the second one, as it will naturally lead to a single non-periodic Liouville field on the
spatial boundary.
Due to the asymptotic behaviour of Ar and A r (23), Q1 and Q2 asymptotically factorize
into:
Q1 (r , , t) ' q1 (, t)S(r ),
r
with: 2
S(r ) =
r /`
0
Q2 (r , , t) ' q2 (, t)S(r )1 ,
r
(40)
!
p0
`/r
(41)
On the other hand, the components At and A t in the boundary actions may be eliminated
e+ using the boundary conditions (24). The remaining
in terms of A , A , K and K
conditions, given by Eq. (25), restrict the matrix elements of q1 and q2 , but are difficult
to implement at this stage, and we choose not to do so for the moment. Dropping these
conditions (25) implies that we must add a boundary term to the action SSC ensuring the
vanishing of its on-shell variation, independently of (25). This modified action reads as:
0
= SCS + BCS0 GCS0 ,
SCS
(r)
(42)
(r)
The additional dynamical boundary term BCS0 is given by:

Z

`
(r)
Tr A A+ + A A + d dt,
BCS0 =
16G
(43)
|r|=r
2 Note that if the asymptotic behaviours of A and A

are interchanged (as discussed before Eq. (23)), S(r ) must
be replaced by S(r )1 in Eq. (40).
339
(0)
where = 2 g . The GCS0 term contains no degrees of freedom in the limit r , is

finite and reads as:

Z
`
1
(44)
+ + 2 d dt.
GCS0 =
16G
`
|r|=r
(r)
This term is equal to BCS0 when all boundary conditions are imposed, and ensures that the
value of the action SCS remains unmodified in the limit r . Here and in the following,
such boundary terms without degrees on freedom will be qualified as geometrical.
0 the expression of the gauge fields in terms of the Q matrices (40), we
Inserting in SCS
i
get a sum of four terms:
0
= SWZWC + BWZWC + GWZWC GCS0 .
SCS
()
The first is the chiral WZW action:

SWZWC = [Q1 ] +
`
16G
Z
|r|=r
`
+ [Q2 ]
16G
|r|=r
(45)

1
Tr t q10 q11 q1 2k dt d
e

1
Tr t q20 q21 + q2 2k+ dt d,
e+
(46)
where the derivatives + and are taken along the vectors eE+ and eE , q 0 = q 1 q,
e+ S(r), and the bulk WZW action reads as:
k = S(r)K S(r)1 , k+ = S(r)1 K
Z

`
(47)
Tr Q1 dQ Q1 dQ Q1 dQ .
[Q] =
48G
The second term:
()
BWZWC =
`
16G

1
Tr (Q1
2 r Q2 Q2 t Q2

1
Q1
1 r Q1 Q1 t Q1 ) dr d dt,
(48)
does not vanish in general due to the holonomy encoded in Q1 and Q2 . The last two terms
are geometrical; GCS0 is given by (44) and GWZWC by:

Z t
t
e+ 2
e
`
2
k
k
+
(49)
GWZWC =
t
t + d dt.
16G
e
e+
|r|=r
Defining the new variables

Q = Q1
1 Q2 ,
q = q11 q2 ,
(50)
one of the fields q1 or q2 can be eliminated from the action SWZWC using its equation of
motion. Indeed, the variables q1 or q2 only appear in the chiral WZW action in quadratic
expressions of their derivatives with respect to the angular variable . Their equations of
motion lead to:
340
t

t
t t 1
q10 = e
k + e+
e q1 n1 (t)q1 ,
+ qq 1 q k+ q 1 + e+
t 1

t
t t 1
e q2 n2 (t)q2 ,
q q + q 1 k q e
k + + e+
q20 = e+
(51)
(52)
where n1 (t) and n2 (t) are sl(2, R) generators depending only on t, appearing upon
integrations. It is easy to see that these degrees of freedom decouple from q in the action
and, using their own equation of motion, may be directly set equal to zero. The resulting
action becomes:
(,t )
SWZWC = SWZW + BWZW GWZWC ,
(53)
with the geometrical contribution canceling that occurring in Eq. (45). The (non-chiral)
WZW action is given by:

Z
`
1
Tr q 1 + qq 1 q + + qq 1 k
SWZW = [Q]
8G
2
|r|=r

q 1 q k+ k+ q 1 k q dt d.
(54)
(,t )
The boundary term BWZW reads as:

Z

`
(,t )
Tr Q1
[r Q2 Q1 t ] Q
BWZW =
2
8G
V

1
+ t Q1
2 [ Q2 Q r] Q dr d dt.
(55)
3.3. Boundary equations of motion as consistency equations

Let us for a moment focus on the equations of motion (51), (52) of q1 and q2 as functions
of q. Using the remaining boundary conditions (25) and the Gauss decomposition for
SL(2, R)/Z2 elements:

/2
e + xye/2 xe/2
,
(56)
q=
ye/2
e/2
Eqs. (51), (52) lead to 6 equations. Four of them:
`
y = ,
2
1
`( + )x + + e = 0,
2
`
x = + ,
2
(57)
1
`(+ + )y + + e = 0,
2
(58)
determine x and y as functions of and combine to give:

2 +
4
8
e + 2 = 0.
2
`
`
(59)
This is the Liouville equation on a curved background, the curvature being given by
Eq. (12). The last two equations are:
341
1
2 (2)
(+ + + )+ (+ )2 + 2 g ++ = 0,
(60)
2
`
1
2 (2)
(61)
( ) ( )2 + 2 g = 0.
2
`
Using Eq. (4) and the expression of the energymomentum tensor of the Liouville field,
these equations and Eq. (12) can be summarized as:
(0)
(2)
`2
g ab =
Tab ab R .
(62)
2
3.4. Third step: Liouville action
We now return to the non-chiral WZW action (54). Using for the matrix Q a Gauss
decomposition in terms of , X and Y similar to (56), the bulk term (Q) can be written
as a sum of three kinds of boundary contributions:
Z

`
(,t )
(r)
(63)
d e dX dY = B + B .
(Q) =
16G
The contribution on surfaces of constant |r| = r yields in the limit r :
Z

`
(r)
e [ yt ] x d dt,
(64)
B =
8G
|r|=r
where we have used Eqs. (40), (50), which give the asymptotic expressions:
`
r 2
`
Y y.
e ,
X x,
r
r
`2
The last contribution is:
Z

`
(,t )
B =
e [t Y r] X + t e [r Y ] X dr d dt.
8G
e
(65)
(66)
Accordingly, the non-chiral WZW action (54) may be written as:

(,t )
SWZW = S(x,y,) + B
where
S(x,y,) =
`
16G
+ G(x,y,),
Z
|r|=r
(67)
1
+ + + +
2

+ y + y +
d dt,
+ 2e x + x +
2`
2`
and the geometrical term G(x,y,) is:
Z
`
+ d dt.
G(x,y,) =
16G
(68)
(69)
|r|=r
This term cancels the first term in the geometric contribution appearing upon modifying
the CS action, see Eqs. (42), (44).
342
Finally, the action S(x,y,) (68) can be expressed in terms of only by eliminating [6,19]
the variables x and y in terms of the constants of motion defined by (58), using the same
trick as the one that leads to the Maupertuis action in classical mechanics with conserved
energy. From the definition of the connection and Eqs. (57), (58), we find:

e
+
+
+
,
( x + x) = t x t x =
`
2`

e
+
.
(70)
(+ y + y) = t y + t y =
`
2`
Using these equations, we add to the action S(x,y,) zero written as:

Z
e ( x + x) + y + y +
0 =
8G
2`
|r|=r

d dt
+ e (+ y + y) x + x +
2`
Z h

i
t+ x t y t + x d dt.
`
8G
(71)
|r|=r
Adding this null expression to S(x,y,) allows to eliminate x and y as functions of using
Eqs. (57), (58), while keeping the correct field equations and holonomy contributions.
Accordingly, we get:
S(x,y,) = SL + BL + GL ,
(72)
where SL is the Liouville action on curved background:

q
Z
(0)
`
8 (0)
1 (0)ab
g a b e + R g dt d.
SL =
2
32G
2
`
(73)
|r|=r
Let us emphasize that the curvature term appearing here comes directly from its definition
(0)
in terms of the asymptotic metric g , and not through as it is the case in Eq. (59). The
term BL is defined on the r, and r, t boundaries:
!
#
q
Z "
(0) (0)
`
g g ab b + (t ) t ( ) d dt,
(74)
a
BL =
16G
|r|=r
and the geometrical term is given by:

Z
`
d dt.
GL =
8G
`2
(75)
|r|=r
4. Holonomies
In the previous section, we have shown that the EH action can be expressed as a sum
of terms defined on the r, and t boundaries without any remaining bulk terms. The
343
dynamical equation resulting from imposing the stationarity of the action on the spatial
boundary |r| = r , is that of Liouville on a curved background. We now analyze the
dynamical content of the -boundary terms B () that occur when has the topology of
the cylinder.
The first two non vanishing holonomy terms are given by Eqs. (48), (55) and appear
when going from the CS action to the non-chiral WZW action. Together they give, after
substitution of Q1 by Q2 Q1 :
Z

`
()
()
1
2Q1
BWZWC + BWZW =
2 r Q2 Q t Q
16G
V

Q1 r QQ1 t Q dr d dt.
(76)
Contrary to what happens on the boundary |r| = r , Q2 cannot be eliminated in terms

of Q by using its equation of motion. When going from the non-chiral WZW action to the
()
S(x,y,) action, another holonomy term appears, B , given by Eq. (64). Adding this term
to (76), we obtain the -boundary action B () :

Z
`
()
1
=
2Q1
B
2 r Q2 Q t Q
16G
V

1
(77)
t r 2t X r Y e dr d dt,
2
which encodes all bulk terms that survive in addition to the Liouville action (73) on the
spatial |r| = r boundary.
The equations of motion on the -boundary can be obtained from this action by varying
Q2 , X, Y and . But it is easier to obtain them by varying Q1 and Q2 in the CS action (45)
expressed in terms of these fields. We get:
Z

`
1
0
Tr Q1
SCS =
1 r t Q1 Q1 Q1
8G
V

1
Q1
2 r t Q2 Q2 Q2 dr d dt.
(78)
In terms of the periodic group elements G1 and G2 defined in Eqs. (39), which factorize
in the same way as the non-periodic ones (cf Eq. (40)):
G1 (r , , t) ' g1 (, t)S(r ),
r
G2 (r , , t) ' g2 (, t)S(r )1 ,
r
the variation of the action becomes:

Z

`
0
=
Tr t g1 g11 t g2 g21 =0 dt,
SCS
8G
(79)
(80)
|r|=r
with (t) = exp(2h) exp(2h) and similarly for . The equations of motion are
thus:
r=r
r=r
t g2 g21 =0 r=r = 0,
(81)
t g1 g11 =0 r=r = 0,
whose solutions are simply
344
g1 ( = 0, t)|r=r = g1 ( = 0, t)|r=+r C1 ,
g2 ( = 0, t)|r=r = g2 ( = 0, t)|r=+r C2 ,
(82)
with C1 and C2 constant matrices. Hence, the bulk field configuration does not appear in
the equations of motion; only the fields defined on the spatial boundary do.
It is interesting at this stage to compare these results with the equivalent ones obtained
in [20] on the basis of expression (37) instead of (38) for the gauge potentials. In the latter
approach, the CS action reduces to a sum of actions on the spatial boundaries involving
globally defined (single-valued) fields which are coupled to the flat connections h and h
and hence to each other. In contrast, our analysis based on multi-valued fields appears to
be simpler, as the only remaining coupling between the fields in the equations of motion is
through Eqs. (82).
5. Flat boundary metric

To further analyze the significance of the holonomy contributions taken into account
in the previous section via -boundary terms, we compute the gravitational holonomy
encoded in the solutions of Einsteins equations. For this purpose, we restrict our analysis
(0)
to the flat g metric (infinite cylinders have no moduli). For such asymptotic geometry, the
expansion described in Eq. (2) stops at order r 1 and the complete expression of locally
AdS metrics reads as [17]:

r + `
dr 2
r `
dx + L (x ) dx
dx + L+ (x + ) dx + .
(83)
ds 2 = `2 2 +
`
r
`
r
r
This metric describes two asymptotic regions: one in the neighbourhood of r = , the
other near r = 0. In order to have a single radial coordinate that leads to the FG metric
expression on both asymptotic regions simultaneously we have to introduce a new radial
coordinate defined by:
`2
for r < r1 ,
(84)
r
where r1 < r2 are two arbitrarily chosen positive values of r; between them, is given by
any smoothly interpolating increasing function of r. So, near = we may choose as
frame:
d
`
+ = dx + + L (x ) dx ,
0 = ` ,
`
(85)
= dx + L+ (x + ) dx + .
`
=r
for r > r2 ,
It continues near = into:

d
`
,
+ = L (x ) dx dx + ,
`
= L+ (x + ) dx + dx .
`
0 = `
(86)
345
To avoid closed causal curves we must assume the functions L (x ) and L+ (x + ) to

be non-negative. In this case the geometry (83) presents a singularity on the surface
r 4 = `4 L (x )L+ (x + ). Such a singularity deserves special attention: in the case of BTZ
black holes it corresponds to the troath of the EinsteinRosen bridge, but it is not clear if
it still corresponds to a coordinate singularity in the more general case or becomes a true
singularity.
On the boundary r = , the Liouville field solution of Eq. (59) can be expressed as:
2 0 0
` f+ f
,
(87)
= log
(f+ + f )2
where f+ (resp. f ) is a function of x + (resp. x ) only and f+0 (resp. f0 ) its derivative
with respect to its argument. To yield the metric (83) these functions must be related to the
components of the subdominant term of the asymptotic metric as:

`2 3 f+00 2 f+000
`2 3 f00 2 f000
,
L
.
(88)
L+ =
2 2 f+0
f+0
2 2 f0
f0
Here we see the important role played by the local character of the gauge fields. Indeed,
though L+ (x + ) and L (x ) are periodic in their arguments, the functions f , and thus
the Liouville field , are not necessarily so.
Let us consider the right handed sector. For a given L+ function, the solution f+ of
2 , the function w being a solution of the
Eq. (88) can be obtained by posing f+0 = 1/w+
+
linear second order equation:
L+ (x + )
w+ .
(89)
`2
Floquets theory [21, Section 19.4] gives us the general functional form of the solutions of
this equation. These can be written as linear combinations of particular solutions given by
the product of exponential and periodic functions:

(90)
F x + = exp + x + F x + and K x + = exp + x + K x + ,
00
=
w+
where + is a real or purely imaginary constant. As a consequence, the function f+

appearing in Eq. (88) can, for real values of + , always be chosen as the product of
exp (2+ x + ) with a periodic function; but of course more complicated functional forms
are also possible and are even required when + is purely imaginary. The most general
expression of f+ depends on three parameters, and is obtained as the integral of the inverse
of the square of the general solution of Eq. (89).
The discontinuities of the Liouville field (of the function f ) when their argument
increases by 2 reflect an important geometrical property of asymptotic AdS3 spaces.
They encode in a simple way the global holonomy properties of the general metric (83).
We now attempt to clarify this.
In the SL(2, R) basis, the equation of parallel transport by means of the connections A
is:
+ Aa Ja = 0,
(91)
346
where has the two components 1 and 2 . Note that these equations, together with those
of the left handed sector, are exactly those defining the Killing spinors of the manifold. The
general solutions of Eq. (91) for the metric (83) are of the form:
r
r

`
r
0
+
`+ x ,
+ x + ,
2 =
(92)
1 =
r
`
with the chiral field + (x + ) satisfying:
L+
00
= 2 + .
(93)
+
`
the general solutions of Eq. (93)
This is nothing else but equation (89) defining
q f+ . Hence, q
can be written as a linear combination of 1/ f+0 and f+ / f+0 . We find:

+
1 (r, x + )
+ 1 (r0 , x0 )
+
;
r
,
x
=
P
r,
x
,
0
0
2 (r, x + )
2 (r0 , x0+ )
(94)
where the parallel transport matrix P (r, x + ; r0 , x0+ ) is given by:

P = S 1 (r)pS(r0 ),
(95)
with (we drop the + subscript to lighten the notation):

s

2
2
f0
f 00 (f0 f )
`2 2(f 0 0 f 00 f 0 f000 ) + f000 f 00 (f f0 )
1+
f00
2(f 0 )2
4
(f 0 f00 )3/2
,
s
p=

1 (f0 f )
f000 (f f0 )
f00
q
1
+
0 )2
0
f
`2
2(f
0
0
0
f0 f
(96)
whose eigenvalues are simply exp(2+ ). Thus, since + is given by the zeros of an
infinite determinant built from all the Fourier coefficients of L+ [21, Section 19.4], the
holonomies depend on all the on-shell values of the classical generators of the asymptotic
e solution can be obtained from
Virasoro algebra. The matrix p(r,
x ; r0 , x0 ) giving the
T
1
+
by the substitutions f+ f and x x .

p
The matrix p immediately yields the Wilson loop matrix S 1 (r0 )wS(r0 ) of parallel
transport around close loops, by posing x + = x0+ + 2`. They also yield the matrices q1
and q2 defining the flat connection. Indeed, inserting Eqs. (38) into Eq. (91), we obtain:

q2 x = q2 x0 p 1 .
(97)
q1 x + = q1 x0+ p1 ,
To make the decomposition (39) explicit, we use as set of solutions of Eq. (93) the
following linear combination of the quasi-periodic solutions (90):

(98)
U x + = ` K00 F x + F00 K(x + ,

+
+
+
(99)
V (x = F0 K x K0 F x ,
where the subscript 0 indicates that the functions must be taken at an arbitrary reference
point x0+ ; the functions F and K are (partially) normalized so that their wronskian is 1/`.
This implies that

U x0+ = 1,

V x0+ = 0,

U 0 x0+ = 0,
1
V 0 x0+ = ,
`
347
(100)
and the wronskian of U and V is also equal to 1/`.

In terms of these solutions, we can write explicitly the non-trivial flat connection h(t)
generating the holonomies (Eq. (39)). When + is real we have:

h(t) = 2+ q1 x0+
`K00
K0
`F00
F0

J0
F0
K0
`F00
`K00

q11 x0+ .
(101)
Thus h(t) is a just a similarity transform of the matrix 2+ J0 . As a consequence, the

matrix g1 (see Eq. (79)), which depends on both t and (and not only on x + ), is globally
defined (as it must be), its entries being given by products of exponentials of + t and
periodic functions of x + . When + is purely imaginary, we pose + = i+ , and see
that the generators of holonomies h(t) are now similarity transforms of 2+ J2 , thus of a
timelike generator, whereas the generator found for real + is space-like.
Finally note that the expressions (92) and the subsequent developments could also be
obtained by limiting ourselves to an asymptotic evaluation of Eqs. (91). This is due to the
special form (83) of the metric, for which the connection form + vanishes to all orders.
As a consequence, the factorization (40) is valid up to r 7 0, and Eqs. (81) are trivially
satisfied, i.e., C1 = 1 and C2 = 1.
To illustrate the above results, let us consider the special case of BTZ black holes.
From its metric written in FG form (17), we immediately obtain the subdominant metric
components near r = in the null frame:
(2)
g ++ L+

J
1
M+
,
4
`
(2)
g L

J
1
M
.
4
`
(102)
For such constant components, Eq. (88) is easily integrated in terms of purely exponential
solutions:
+
ae2+ x + b
,
f+ = 2 x +
ce + + d
(103)
q
where + = 12 M + J` and ad bc = 1; f is obtained by the substitutions x + x
and J J . As expected, the Liouville field constructed from these solutions is in
general not globally defined. Only for J = 0 can be extended periodically (it becomes
+
independent) on the interval
[0, 2] of the -coordinate if we choose f+ = exp(x )
and f = exp(x ), = M/2.

Furthermore, we easily obtain (compared with [22] where the same calculation was
originally performed in another context) the holonomy matrix w corresponding to a Wilson
loop surrounding the throat of the EinsteinRosen bridge:
348
r
r

M + J`
J
J
cosh M +
sinh M +
`
2
`
w=
, (104)
r

r

J
J
q 2
sinh M +
cosh M +
`
`
J
M+ `
and similarly for w with J J . The holonomy is thus always non-trivial, and there
exists no globally well defined Killing spinor fields, except in two cases. The first is J = 0
and M = 1, in which case w = 2J0 and the geometry is that of the usual AdS3 space.
e2 in order to
The second case is J = 0 and M = 0, where we have to impose 1 = 0 =
obtain global Killing spinor fields. In the former case, we have thus 4 Killing vector fields,
and in the latter 2, in agreement with [23].
6. Anomalies
As a preliminary to the discussion of the link between the diffeomorphism anomaly of
the gravitational action and the conformal anomaly of the Liouville action, let us briefly
summarize the various steps that led in Section 3 from the EH action to the Liouville
action, through the CS action. Using Eqs. (20), (22), and assuming the metric to be globally
defined on V, we find that the EH action (18) and CS action (20) are related by:
(r)
(t )
+ BCS
,
SEH = SCS + BCS
(105)
(r)
where BCS is an r 2 -divergent term defined on the spatial boundary given by Eq. (26), and
(t )
is a log r -divergent term corresponding to a total derivative with respect to t and given
BCS
by Eq. (27). The EH action presents an r 2 -divergence, whereas the CS action is at most
logarithmically divergent.
Furthermore, using Eqs. (42), (45), (53), (67), (72) we find the relation between the CS
action (20) and the Liouville action (73):
SCS = SL + B () + B (t ) + B.
(106)
B () is defined in Eq. (77) and corresponds to the sum of all -boundary terms encountered
when going from the CS action to the Liouville action. Similarly, B (t ) is the sum of all
t-boundary terms and is given by:
(t )
(t )
B (t ) = BWZW + B
Z

`
1
=
t Tr Q1
[r Y ] X dr d dt,
2 [ Q2 Q r] Q + e
8G
(107)
(t )
with BWZW
(t )
and B
defined by Eqs. (55), (66). Finally, the term B is sum of all geometrical
terms (44), (69), (75) and the boundary term BL (74). It is equal to:
q
Z
i
h
(0) (0)
`
g g b b + t
B =
16G
|r|=r
q
+ t
(0) (0)
g g tb
349

b + 2 d dt.
(108)
This term is defined on the (r, ) and (r, t) boundaries.

We now analyze the variation under Weyl transformations and under diffeomorphisms
of the EH, CS and Liouville actions. Consider a diffeomorphism generated by the vector
field , which moves the boundary |r| = r and keeps the metric in the FG form (1). Its
radial component is asymptotically of the form r = (, t)r, where we assume and
all its derivatives vanishing on the space-like boundaries of V; the t and components may
be assumed to be of order r 2 [14,24]. Under the action of this diffeomorphism, the metric
varies as:
(0)
(0)
D g ab = 2 g ab ,
(109)
(2)
D g ab
(110)
= `2 ,b;a .
The on-shell variation of the EH action (18) under these transformations can be computed
as:
Z

`
Tr r 6[r A At ] 2Ar A[ At ] + A [ A t ] d dt. (111)
D SEH =
16G
|r|=r
Inserting the boundary conditions (23), (24), (25) in this equation, we obtain by expanding
the metric:
q
Z
(0) 4
(0)
r2
`
g d dt,
(112)
D SEH =
R 4
16G
`
|r|=r
in agreement with [14,25,26]. Similarly, the on-shell variation of the CS action (20) is
given by:
q
Z
r
(0)
`
g d dt,
Tr 2 Ar A[ At ] + A[ At ]
(113)
D SCS =
16G
|r|=r
and using the boundary conditions (23), (24), (25):

q
Z
(0) (0)
`
D SCS =
R g d dt.
32G
(114)
|r|=r
Using (20), we see that the difference between the variations of the EH and CS actions is
transferred in the BCS term (22). We find indeed:
Z

`
Tr r 6[r A At ] d dt
(115)
D BCS =
16G
|r|=r
`
=
16G
Z
|r|=r
1 (0) 4r 2
R 4
2
`
q
(0)
g d dt.
(116)
350
Thus, the variations of the EH and SC actions under diffeomorphisms are related by:
Z q
(0)
r 2
g d dt.
(117)
D SEH = 2D SCS +
3
4G`
|r|=r
The variation D SEH presents a r 2 divergence whereas D SSC is finite. More strikingly, the
finite contributions of D SEH and D SSC differ by a factor of 2.
Since the Liouville action SL (73) is finite and defined on the spatial boundary |r| = r ,
it is invariant under the 3-dimensional diffeomorphisms. Hence, when going from the
CS action to the Liouville action, the variation under diffeomorphisms must be totally
transferred to the additional terms given in Eq. (106), in particular to the B boundary terms,
as the B term (108) is invariant under 3-dimensional diffeomorphisms. It is easy to check
that it is transferred to the B (t ) term (107) that presents a logarithmic divergence in r . We
find indeed that on-shell:
Z
`
1
t dr d dt.
(118)
B (t ) '
r 16G
r
q
(0) (0)
Using the fact that g R is equal to 2t up to a -derivative, we obtain the expected

variation under diffeomorphisms:
q
Z
(0) (0)
`
(t )
(119)
D B =
R g d dt.
32G
|r|=r
Let us now consider the variation of the actions under a Weyl transformation on the
spatial boundary |r| = r , which compensates the effect of the diffeomorphism (109) on the
boundary metric:
(0)
(0)
W g ab = D g ab .
(120)
The EH action is known to be invariant under the combined action of this Weyl
transformation and of the diffeomorphism (109) [13,14]:
D SEH + W SEH = 0.
(121)
The variation of the Liouville action SL (73) under this Weyl transformation is easily
computed using (62):
q
q
Z
Z
(0) (0)
(0)
(0)
`
`
ab
g d dt =
Tab W g
W SL =
R g d dt.
32G
16G
|r|=r
|r|=r
(122)
Inspection of Eqs. (105), (106) shows that the only other term which has a non-vanishing
(r)
contribution to the Weyl variation is BCS . We find:
q
Z
`
4r 2 (0)
(r)
g d dt.
(123)
W BCS =
16G
`4
|r|=r
351
We have thus:
(r)
.
W SEH = W SL + W BCS
(124)
7. Conclusion
In this paper we have explicitly shown that the 3-d EH action with negative cosmological
constant is equivalent to a Liouville action on the spatial r boundaries, plus terms
on the t- and -boundaries. The equations of motions derived from the -boundary terms,
in the case of non-trivial topologies such as those of BTZ black holes, relate the two nonconnected components constituting spatial infinity. An originality of our approach resides
in the fact that we encoded the holonomy in multi-valued functions.
Owing to the fact that we considered arbitrary curved metrics on the spatial r boundaries,
we gave an explicit demonstration that the variation under diffeomorphisms of the 3-d EH
action is equal to the Weyl anomaly of the asymptotic Liouville theory.
We moreover discussed in detail the asymptotically flat solutions, which generate
upon diffeomorphisms all other solutions. We obtained explicitly the link between the
subdominant term of the metric on the spatial boundary, which encode the FG ambiguity,
and the Liouville field. This reveals that though each regular boundary metric can be
expressed in terms of a 3-parameter family of Liouville fields, there are acceptable
Liouville fields that lead to singular metrics. The Liouville (and CS) theory thus contains
much more solutions than the EH gravity theory. Let us furthermore stress that a Liouville
field defines a solution of Einstein equations only if it can be extended to CS connections
whose difference yield driebein fields that are regular or only present singularities
A and A
interpretable as coordinate singularities. In the latter case, the topology of the space on
which these CS fields are defined will differ from that of AdS.
Note that the developments following Eq. (45), which lead from the non-chiral WZW
action to the Liouville action (73), are classically valid, but have to be re-examined in the
framework of quantum mechanics. Indeed, in quantum mechanics, the changes of variables
leading to Eq. (54) and the subsequent elimination of the X and Y variables in terms of
involve functional determinants that have been completely ignored.
Finally, our analysis does not solve the problem of the degrees of freedom at the origin of
the black hole entropy, but leads to formulate the hypothesis that these degrees of freedom
could possibly be encoded in a generalized holographic principle involving a multiply
connected surface at infinity, each component having its own independent Liouville field
only related to the others by consistency relations dictated by the holonomies.
Acknowledgements
We are grateful to F. Englert for numerous enlightening discussions and encouragements. We also acknowledge K. Bautier, M. Baados and M. Henneaux for fruitful discussions. M.R. is Senior Research Associate at the Belgian National Fund for Scientific
352
Research. This work was partially supported by the a F.R.F.C. grant and by IISN-Belgium
(convention 4.4505.86).
Appendix A.
In this appendix we recall some 2-d formula that we found useful for the calculations of
the main text.
sl(2, R) generators:

1 1 0
1 0 1
1 0 1
,
J1 =
,
J2 =
.
(A.1)
J0 =
2 0 1
2 1 0
2 1 0
Null frame
p
p = i dx i ,
eEp = p = epi x i ,
epi = p q ij j ,
+ = 1,
p,
q,
. . . {+, },
t = 1.
x i {, t},
(A.2)
Metric in the null frame

ds 2 = p q p q ,
p,
q {+, },
s2 = p q eEp eEq ,
p
= det i = t+ + t ,

0 1/2
p q =
.
1/2
0
(A.3)
Connection coefficients
d + = + ,
d = ,
= = ++ = + + + ,
p
ij p q x i j = q .
Gauss curvature

R = 4 (+ ) + ( ) + 2+ ,
i

i
+ e+
.
R = 4x i e
(A.4)
(A.5)
Weyl variance
p 7 e p ,
eEp 7 e eEp ,
7 e2 ,
7 () + + () + ,

7 e () .
Dalembertian
(A.6)

2 = 2 + + + + + + .
Liouville energymomentum tensor

1
4
1
;c
;c
Tab = ;a ;b ;ab ab ;c ;c 2 e .
2
4
`
353
(A.7)
(A.8)
References
[1] S. Deser, R. Jackiw, G. t Hooft, Ann. Phys. (N.Y.) 152 (1984) 220;
S. Deser, R. Jackiw, Ann. Phys. (N.Y.) 153 (1984) 405.
[2] M. Baados, C. Teitelboim, Z. Zanelli, Phys. Rev. Lett. 69 (1992) 1849, hep-th/9204099;
M. Baados, M. Henneaux, C. Teitelboim, Z. Zanelli, Phys. Rev. D 48 (1993) 1506, grqc/9302012.
[3] S. Carlip, Quantum Gravity in 2 + 1 Dimensions, Cambridge Univ. Press, 1998.
[4] J. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
[5] A. Achucarro, P.K. Townsend, Phys. Lett. B 180 (1986) 89;
E. Witten, Nucl. Phys. B 311 (1988) 46.
[6] O. Coussaert, M. Henneaux, P. van Driel, Class. Quant. Grav. 12 (1995) 2961, gr-qc/9506019.
[7] G. Moore, N. Seiberg, Phys. Lett. B 220 (1989) 422.
[8] S. Elitzur, G. Moore, A. Schwimmer, N. Seiberg, Nucl. Phys. B 326 (1989) 108.
[9] P. Forgacs, A. Wipf, J. Balog, L. Feher, L. ORaifeartaigh, Phys. Lett. B 227 (1989) 213.
[10] C.R. Graham, J.M. Lee, Adv. in Math. 87 (1991) 186.
[11] C. Fefferman, C.R. Graham, Soc. Math. de France, Astrisque, hors srie (1985) 95.
[12] M. Rooman, Ph. Spindel, Ann. Phys. (Leipzig) 9 (2000) 162, hep-th/9911142.
[13] N. Boulanger, Lanomalie conforme en dimensions 2 et 4, Mmoire de Licence U.L.B. (1999),
unpublished.
[14] K. Bautier, F. Englert, M. Rooman, Ph. Spindel, Phys. Lett. B 479 (2000) 291, hep-th/0002156.
[15] M. Baados, F. Mendez, Phys. Rev. D 58 (1998) 104014, hep-th/9806065.
[16] G.W. Gibbons, S.W. Hawking, Phys. Rev. D 15 (1977) 2752.
[17] M. Banados, Three-dimensional quantum geometry and black holes, Presented at 2nd La Plata
Meeting on Trends in Theoretical Physics, Buenos Aires, Brazil, November 28December 4,
1998, hep-th/9901148.
[18] F. Englert, private communication.
[19] M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton University Press, 1992.
[20] M. Henneaux, L. Maoz, A. Schwimmer, Ann. Phys. 282 (2000) 31, hep-th/9910013.
[21] E. Whittaker, G. Watson, Modern Analysis, 4th edition, Cambridge University Press, 1965.
[22] D. Cangemi, M. Leblanc, R.B. Mann, Phys. Rev. D 48 (1993) 3606, gr-qc/9211013.
[23] O. Coussaert, M. Henneaux, Phys. Rev. Lett. 72 (1994) 183, hep-th/9310194.
[24] C. Imbimbo, A. Schwimmer, S. Theisen, S. Yankielowicz, Class. Quant. Grav. 17 (2000) 1129,
hep-th/9905046.
[25] M. Henningson, K. Skenderis, JHEP 9807 (1998) 023, hep-th/9806087.
[26] K. Bautier, Diffeomorphisms and Weyl transformations in AdS3 gravity, Presented at the
Meeting on Quantum Aspects of Gauge Theories, Supersymmetry and Unification, Paris,
France, 17 September, 1999, hep-th/9910134.

Quantum inconsistency of Einstein supergravity

Jonathan A. Bagger a, , Takeo Moroi b,1 , Erich Poppitz c
a Department of Physics and Astronomy, The Johns Hopkins University, Baltimore, MD 21218, USA
b School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540, USA
c Physics Department, Yale University, New Haven, CT 06520, USA
Abstract
We show that N = 1, D = 4 Einstein-frame supergravity is inconsistent at one loop because of
an anomaly in local supersymmetry transformations. A Jacobian must be added to the Einsteinframe Lagrangian to cancel this anomaly. We show how the Jacobian arises from the super-Weyl
field redefinition that takes the superspace Lagrangian to the Einstein frame. We present an explicit
example which demonstrates that the Jacobian is necessary for one-loop scattering amplitudes to be
frame independent. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
The component Lagrangian of matter-coupled supergravity can be derived from
a superspace formulation or a tensor calculus [1,2]. Both approaches inevitably lead to
a component theory in which the gravitational action is of a generalized BransDicke form:
e1 L = 12 eK/3 R + .
(1.1)
In this expression, e is the vielbein determinant, R is the curvature scalar and K is the
Khler potential, a function of the scalar fields A , A.
The Lagrangian (1.1) leads to a kinetic mixing between the graviton and the scalar fields.
In addition, the kinetic terms of the scalar fields appear in non-Khler form. It is therefore
convenient and customary to carry out a field-dependent Weyl rescaling of the metric to
bring the Lagrangian into canonical Einstein form. In supergravity, this Weyl rescaling
must also be accompanied by a chiral rotation of the fermions. As we shall see, this rotation
gives rise to an anomalous Jacobian in the supergravity action.
The field redefinitions needed to go to this Einstein frame are usually performed
in terms of component fields [1,2]. This obscures the symmetries of the theory and
E-mail address: bagger@bohr.pha.jhu.edu (J.A. Bagger).

1 Address after April, 2000: Department of Physics, Tohoku University, Sendai 980-8578, Japan.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 5 1 - 4
J.A. Bagger et al. / Nuclear Physics B 594 (2001) 354368
355
complicates the study of anomalies and their consequences. Therefore, in this paper we
will use the superspace approach of [2] to study anomalies in supergravity theories. We
will show that:
1. The Einstein-frame field redefinitions can be carried out directly in superspace
through a super-Weyl transformation of the vielbein. The corresponding componentfield Lagrangian gives rise to ordinary Einstein gravity.
2. Local supersymmetry transformations in the Einstein frame involve chiral rotations
of the fermions. They are anomalous at one loop.
3. The local supersymmetry anomaly is cancelled by a Jacobian that arises from the
transition to the Einstein frame. This Jacobian is necessary to ensure the quantum
consistency of Einstein-frame supergravity.
4. The anomalous Jacobian can have important physical consequences. For example, it
is necessary to ensure the quantum equivalence of scattering amplitudes computed in
different frames.
This paper is organized as follows. In Section 2 we define super-Weyl transformations
and derive the corresponding Jacobians. In Section 3 we study the transition to the Einstein
frame. We show that one-loop supergravity invariance of the Einstein-frame Lagrangian
requires that a certain superspace Jacobian be added to the bare Lagrangian. In Section 4
we present an explicit example which illustrates the physical importance of this Jacobian.
We summarize our results in Section 5.
2. Super-Weyl transformations
2.1. Classical level
In this section we study super-Weyl transformations in classical and quantum supergravity. These transformations will play an important role throughout this paper.
In what follows we use the notation and conventions of [2]. We take the matter-coupled
supergravity Lagrangian to be of the form
Z

S 2 8R exp 1 K , + , , V
L(X) =
d2 2E 38 D
3

(2.1)
+ 14 Hab ()W (a) W (b) + P () + h.c.,
where X = (, V , EM A ) denotes the set of fields in the supergravity Lagrangian, and V
are the chiral and vector superfields, and EM A is the supervielbein. In this expression, K is
the Khler potential, P the superpotential, and W (a) the field strength superfield, where (a)
is the index for the adjoint representation. 2 In addition, Hab is the gauge kinetic function,
and is the gauge counterterm which renders the Lagrangian gauge invariant.
In the superspace formalism, supergravity transformations are given by translations in
superspace. Chiral and vector superfields transform as follows:
2 In this paper, we sometimes omit the spin index . Therefore W (a) W (b) should be understood as W (a) W (b) .
356
SUSY = A DA ,
SUSY V = A DA V ,
(2.2)
while the vielbein transforms as

SUSY EM A = DM A B TBM A .
(2.3)
In these expressions, the DA are covariant derivatives and the TBM A are the superspace
torsion. The superspace formalism ensures that the Lagrangian (2.1) is invariant:
L(X + SUSY X) = L(X),
(2.4)
up to a total derivative, under the supersymmetry transformations (2.2) and (2.3).

Super-Weyl transformations are defined as rescalings of the superspace vielbein that
preserve the torsion constraints [3]. 3 They are parameterized by a chiral superfield . If
b we can write
we denote the super-Weyl transformed X field as X,
b + SW X,
X=X
(2.5)
where SW X is the super-Weyl variation of X.

To linear order in , the super-Weyl transformations are given by
SW = S
,
S E ,

S 2 8R 4 2 U S D
S 2 8R U = D
S 2 8R U,
SW D

W ,
SW W = 3 W S

where U is any real vector superfield of Weyl weight zero, and S is defined by

S = 2 + 2 D .
SW E = 6E +
(2.6)
(2.7)
The bar | denotes the = = 0 component of the superfield.

Note that super-Weyl transformations also induce chiral rotations of the component
fermions. Taking the appropriate components of Eq. (2.6) and reexponentiating, we find
b,
= exp(3|),
(2.8)
= exp | 2 |
where and are the fermions in the chiral and gaugino multiplets, respectively.
In general, super-Weyl transformations are not symmetries of the classical supergravity
Lagrangian. Indeed, substituting the transformed variables (2.6) into the Lagrangian (2.1),
all derivative terms cancel. Nevertheless, a nontrivial dependence remains. At the
classical level, the Lagrangian (2.1) becomes
Z

b2

S 8R
b exp 1 K
b 6 6 + b
b
d2 2Eb 38 D
L X + SW X =
3

bab W
b (a) W
b (b) + exp(6)Pb + h.c.,
(2.9)
+ 14 H
3 It is important to distinguish between super-Weyl and super-WeylKhler transformations. The latter are
symmetries of the classical supergravity Lagrangian. A super-WeylKhler transformation is a super-Weyl
transformation, with chiral superfield parameter , combined with a redefinition of the Khler potential and
superpotential, K K + 6 + 6 , P exp(6)P .
357
where the hatted objects are evaluated using the Weyl-transformed fields. The Khler and
superpotential are different, so the Lagrangian (2.1) is not invariant.
2.2. Quantum level
We are now ready to discuss the anomalous Jacobian associated with a given superWeyl transformation. A proper framework is provided by the one-particle-irreducible (1PI)
effective Lagrangian L1PI , defined by
Z
Z

Z
4
4
[dXQ ] exp i d x Lbare (XC + XQ ) .
(2.10)
d x L1PI (XC ) = i log
Here Lbare is the bare Lagrangian, and XC and XQ are classical and quantum parts of the
X field, respectively.
In general, super-Weyl transformations are anomalous; they have a mixed super-Weylgauge anomaly. 4 5 Anomalies generate a set of non-local terms in the 1PI effective action
[5]. For the case at hand, the one-loop anomaly-induced terms are [6,7]:
Z

1 S2
1
8R
d2 2EW (a) W (a) D
1L =
2
2
256

(2.11)
4(TR 3TG )R 13 TR D2 K + h.c.,
where we have omitted a term from the sigma-model anomaly that is irrelevant for our
discussion. In Eq. (2.11), TG is the Dynkin index of the adjoint representation, normalized
to N for SU(N), and TR is the Dynkin index associated with the matter fields. A sum over
all matter representations is understood. The first term, which contains the R superfield,
arises from the superconformal anomaly. It is proportional to the beta function coefficient,
b0 = 3TG TR . The second term expresses the Khler anomaly [6,7].
The variation of 1L can be computed by considering a super-Weyl transformation with
superfield parameter . Under such a transformation, the superfield R changes as follows:

S 2 .
(2.12)
SW R = 2 2 R 14 D
This induces a shift by in the R term in Eq. (2.11). A second shift comes from replacing
b 6 6 . These two shifts induce the following change in 1L,
K by K
1L 1L + LJ , where
Z
1
(3TR 3TG ) d2 2EW (a) W (a) + h.c.
LJ =
16 2
(2.13)
(2.14)
The Lagrangian LJ can be interpreted as the superspace Jacobian that arises from the superWeyl transformation (2.5). (Note that the imaginary part of | corresponds to the Jacobian
from the anomalous U (1)R transformation.)
The nonvanishing Jacobian implies that the functional measure is not invariant. It
transforms as follows under an arbitrary super-Weyl transformation:
4 It also has a mixed super-Weyl-gravity anomaly. We ignore the gravity anomaly here.
5 For a discussion of supergravity anomalies in the compensator formalism, see [4].
358
Z

b + SW V ) = d
b exp i d4 xLJ . (2.15)
b + SW ) d(V
b dV
[d][dV ] = d(
The super-Weyl-rescaled 1PI Lagrangian is then
Z

Z
Z

b + SW X)
b + SW X) exp i d4 x Lbare (X
d(X
exp i d4 x L1PI =

Z
Z

b K
b exp i d4 x Lbare (X)|
=
dX
, P
b K66
b
b + LJ
be6 P

Z
Z

bbare (X)
b ,
b exp i d4 x L
(2.16)
=
dX
where
bbare (X)
b Lbare (X
b + SW X) + LJ
L
b K
= Lbare (X)|
,P
b K66
b
be6 Pb + LJ .
(2.17)
bbare is the bare Lagrangian for the quantum theory with superIn these expressions, L
b The bare Lagrangian does not contain the anomaly term 1L,
Weyl-rescaled variable X.
which arises from integrating out the massless quantum fields. It does, however, contain
the Jacobian LJ . As we will see, LJ is important for ensuring the quantum consistency of
supergravity in the Einstein frame.
3. Einstein supergravity
3.1. The Einstein frame
In this section, we find the field-dependent Weyl rescaling that takes the supergravity
frame Lagrangian of Eq. (2.1),
e1 L = 12 eK/3 R + ,
(3.1)
into the Einstein frame. In the literature, this rescaling has traditionally been done in terms
of component fields [1,2]. Here we perform the transformation in superspace. This allows
us to keep better track of the symmetries of the theory.
The relevant superfield rescaling is, as we will see below, a super-Weyl transformation
with transformation parameter E :
Z

S 2 8R exp 1 K 6E 6 +
d2 2E 38 D
LE (X) =
E
3

(3.2)
+ 14 Hab W (a) W (b) + exp(6E )P + h.c.,
where we have omitted all hats in the above equation. (Here and hereafter, all quantities
should be understood as being defined in the frame obtained after the super-Weyl
transformation, unless specified otherwise.)
The parameter E can be found by demanding that K 6E 6E have no lowest,
and 2 components since this combination appears in the exponent of the first term in
359
Eq. (3.2). To see, for example, why (K 6E 6E )| must vanish, note that Eq. (3.1)
involves the lowest component of eK/3 . If the lowest component of this term is scaled to 1,
the factor eK/3 is absent and gravity is canonically normalized. The other two conditions
lead to a canonical kinetic term for the gravitino, and to canonical Khler kinetic terms for
the matter multiplets.
The conditions on E are, therefore,

(3.3)
(D K)| = 6(D E )|,
D2 K | = 6 D2 E ,
K| = 6E | + 6E |,
or the vanishing of the lowest, and 2 components of K 6E 6E . The Einstein
frame conditions (3.3) almost completely determine the parameter E :
(3.4)
E = A + 2 + 2 F with
A =
1
12 K
+ i,
= 16 Ki i ,
F = 16 Ki F i
i j
1
12 Kij ,
(3.5)
where the subscript i on K denotes the derivative with respect to Ai (Ki K/Ai ), and
F i is the highest component of i . Note that the conditions (3.3) do not completely fix E ;
the imaginary part of its lowest component is left undetermined. Also note that does
not contribute to the conditions (3.3): we assume that a field redefinition is performed so
that the matter fields are in the WessZumino gauge, where has no lowest, and 2
components.
With the above E , it is not hard to find the component Lagrangian. If we substitute
Eq. (3.4) into (3.2), all terms leading to non-Einstein gravity vanish. The rest of the
component Lagrangian can be readily evaluated and gives the well-known Einstein
supergravity Lagrangian with canonical kinetic terms. The complete expression for the
Lagrangian is given in [1,2]. We have seen that the component Lagrangian can be directly
obtained from superspace, without any extra Weyl rescalings of the component fields.
Let us now show that we can safely set the field to zero. This field appears in the
kinetic terms of the matter and gauge fermions,

i
1
e1 Lkin = iKij j a a + ba Kk a Ak Kj a Aj 12ia i

6
6

i
i a a ba .
(3.6)
2
(We omit terms with spin, sigma-model, and gauge connections because they are not
relevant for our discussion.) The field also appears in the solution to the equations of
motion for the auxiliary field ba :
1
i
(3.7)
Kj a Aj Kj a Aj 12ia + Kij i a j + .
2
4
(For the complete expression for ba , see Appendix B.) In addition, it appears in the
superpotential terms in the Einstein frame,

(3.8)
e1 LYukawa = 12 exp 12 K + 6i Pij i j + h.c. + .
ba =
(The complete expression for the Yukawa terms can be found in [1,2].)
360
Upon inspection of Eqs. (3.6)(3.8), one can check that the classical component
Lagrangian can be made independent of by redefining e3i and e3i . In
fact, the dependence also cancels at the quantum level. The field redefinitions used to go
to the Einstein frame are eRe E |+3i and e3 Re E |3i . The Jacobian from
these transformations exactly cancels the Jacobian from the redefinitions used to eliminate
from the component Lagrangian. Therefore the field is unphysical, and we can safely
set it to zero.
3.2. Supersymmetry transformations in the Einstein frame
In this subsection, we discuss the invariance of the classical supergravity Lagrangian
in the Einstein frame. Then, in the next subsection, we consider quantum effects, and in
particular, anomalies.
It is simple to see that the Einstein-frame Lagrangian is not invariant under the
supersymmetry transformations (2.2). Under a supersymmetry transformation, the Khler
potential transforms as SUSY K = A DA K, in which case E transforms to E0 ,
E0 E |KK+SUSY K = E A DA E , where

1
= i Ki i Ki + O() + O 2
6 2

i + O() + O 2 .
(3.9)
(3.10)
The second term on the RHS of Eq. (3.9) is the supersymmetry transformation of a normal
chiral superfield. The term is an additional super-Weyl transformation under which the
action is not invariant.
To see this explicitly, consider the supersymmetry transformation of the Einstein-frame
Lagrangian:
LE (X + SUSY X)
Z

S 2 8R exp 1 K 6(E ) 6 +
= d2 2E 38 D
E
3

(3.11)
+ 14 Hab W (a) W (b) + exp 6(E ) P + h.c.
Since is nonvanishing, the Lagrangian is not invariant under the supersymmetry
transformation (2.2).
This lack of invariance stems from the fact that takes the Lagrangian out of the
Einstein frame. Invariance can be restored by returning to the Einstein frame through a
compensating super-Weyl transformation. This is similar to gauge invariance in globally
supersymmetric gauge theories. There, a supersymmetry transformation in the Wess
Zumino gauge must be supplemented by a superfield gauge transformation to restore
the WessZumino gauge condition. It is instructive to consider this case in some detail
because of the close analogy to supergravity [8]. To that end, we review the supersymmetry
transformations of globally supersymmetric gauge theories in Appendix A.
361
For the case at hand, the compensating super-Weyl transformation has parameter .
In the Einstein frame, therefore, we define a supersymmetry transformation to include
a frame-restoring super-Weyl transformation with parameter :
SUSY + SW .
(3.12)
Under such a transformation, chiral and vector superfields transform as follows:

= A DA + SW ,
V = A DA V + SW V ,
(3.13)
and analogously for the vielbein. In these expressions, the first terms on the RHS are
the original supersymmetry transformations; the second are the compensating super-Weyl
transformations with parameter . These transformations eliminate the in Eq. (3.11)
and restore the classical invariance of the action,
LE (X + X) = LE (X).
(3.14)
The transformation properties of the individual component fields can be derived by

expanding Eq. (3.13). We have checked that they agree with the transformations given
in [1,2] after eliminating the auxiliary fields.
3.3. Quantum consistency in the Einstein frame
We are now ready to discuss anomalies in the supersymmetry transformations (3.12). As
we have seen, these transformations include frame-restoring super-Weyl field rescalings
that induce chiral rotations on the matter fermions,
= + 3i ,
= 3i ,
(3.15)
where i = |. At the quantum level, these transformations are anomalous, so they

should give rise to an anomalous variation of the 1PI Lagrangian,

1
(a) emn(a)
(3T
3T
)
F
+
,
(3.16)
F
L1PI = e
R
G mn
16 2
If nothing were to cancel this variation, local supersymmetry in the Einstein frame would
be anomalous. In what follows, we will show that the full 1PI effective action is, in fact,
invariant. The variation (3.16) is cancelled by the variation of the Jacobian (2.14) that arises
in passing to the Einstein frame.
In the Einstein frame, the complete 1PI effective Lagrangian is of the following form:
L1PI = LE + 1L + LJ .
(3.17)
The first term is the classical part of the Einstein-frame Lagrangian, the second is the
non-local term induced by anomalies, and the third is the Jacobian (2.14). The first term is
invariant under the local supersymmetry transformation (3.12), as discussed in the previous
subsection. The second and third terms are not. Under the supersymmetry transformation
(3.12), the nonvanishing variation of 1L expresses the anomaly associated with the
362
frame-restoring super-Weyl transformation. 6 If this variation were the only change of the
Lagrangian, supersymmetry would be explicitly broken by anomalies at the quantum level.
Fortunately, however, it is not. There is also LJ , the Jacobian that arises in the Einstein
frame. This term is not invariant under (3.12). From Eq. (3.9), we have
E = A DA E .
(3.18)
The first term on the RHS is the supersymmetry transformation of a normal chiral
superfield. The second is a super-Weyl transformation of . This gives
Z
1
(3T
3T
)
d2 2E W (a) W (a) + h.c.
LJ =
R
G
16 2

1
(a) emn(a)
(3TR 3TG ) Fmn F
+ .
(3.19)
= e
16 2
Equation (3.19) exactly cancels the variation (3.16) and restores supersymmetry invariance
in the Einstein frame.
Thus we have seen that the Einstein-frame 1PI effective Lagrangian is invariant under
local supersymmetry transformations provided the Jacobian (2.14) is added to the bare
Lagrangian. Otherwise, local supersymmetry is explicitly broken at the quantum level
because of the anomaly associated with the frame-restoring super-Weyl transformations.
The Jacobian (2.14) is essential for the consistency of the quantum theory. The component
expression for the Jacobian is given in Appendix B.
4. Physical implications of the Jacobian: an example

In this section, we show that the anomalous Jacobian LJ has physical consequences. We
illustrate this with an example of a scattering amplitude which requires the Jacobian to give
a frame-independent result.
In what follows we consider a model with a no-scale Khler potential of the form

(4.1)
K = 3 log 1 13 .
This Khler potential (4.1) is chosen for simplicity; upon substitution into the superspace
Lagrangian (2.1), it gives rise to canonically normalized scalars with a conformal coupling
to gravity:

(4.2)
LSUGRA = g 12 1 13 A A R g mn m A n A + ,
where we used the metric gmn instead of a vielbein. In Eq. (4.2), the subscript SUGRA
indicates that this is the supergravity-frame Lagrangian. As discussed in the previous
sections, we can use a super-Weyl transformation to pass to the Einstein frame.
6 If we make the argument before integrating out the light fields, 1L does not exist. In this case the anomaly
arises as a change of the functional measure of the path integral.
363
Fig. 1. Schematic picture of the amplitude calculation.
In this model, we study the scattering process A A vm vm via a graviton exchange,

where vm is a gauge boson and A is a massless neutral scalar field. For simplicity, we
assume that the gauge theory is pure supersymmetric YangMills. 7
The amplitude of this process can be written as follows (see Fig. 1):
iM(A A vm vm ) =
vm (1 , p10 )vm (2 , p20 )T kl |0i i1kl,mn

2

i
h0|T mn A (p1 )A(p2 )

2
+ vm (1 , p10 )vm (2 , p20 )i LJ A (p1 )A(p2 ) ,
(4.3)
where the pi denote the momenta of the scalars, while i and pi0 denote the polarizations
and momenta of the gauge bosons. Here, Tmn is the energy momentum tensor:
2
L
,
Tmn =
g g mn
(4.4)
while 1kl,mn is the graviton propagator, given by

1mn,kl (q) =
2
(km ln + kn lm kl mn ),
q2
(4.5)
where mn is the flat-space metric. We omit the part that depends on the gauge parameter
because it does not contribute to the amplitude of interest.
For the scalars, the matrix element of the energy-momentum tensor is

h0|T mn A (p1 )A(p2 ) = p1m p2n + p1n p2m (p1 p2 )mn + q 2 mn q m q n , (4.6)
3
where q p1 + p2 , and is a coefficient proportional to the coupling of the scalars to the
scalar curvature R; = 1 in the supergravity frame and = 0 in the Einstein frame. Note
that = 1 corresponds to conformally coupled scalars; this is easy to see upon computing
the trace of Eq. (4.6) with q 2 = 2p1 p2 .
Substituting Eqs. (4.5) and (4.6) into Eq. (4.3), we obtain the amplitude
7 If the matter fields had gauge quantum numbers, the following discussion would still hold, provided we
replace 3TG by 3TG TR .
364
p1k p2l hvm vm |T kl |0i + hvm vm |Tkk |0i + hvm vm | LJ |A Ai

2
q
6
(4.7)
M0 + hvm vm |Tkk |0i + hvm vm |LJ |A Ai,
6
where M0 is frame independent. At the classical level, with LJ absent, the amplitudes in
both frames are identical because the gauge boson energy momentum tensor is traceless.
At the quantum level, the frame independence is more subtle. First, the gauge boson
energy momentum tensor is no longer traceless. Second, there is a Jacobian in the Einstein
frame, but not the supergravity frame. To see what happens, let us consider the supergravity
frame. We take = 1 and use the trace anomaly relation
M =
Tkk =
3
(a) mn(a)
TG Fmn
F
,
32 2
(4.8)
to find
MSUGRA = M0 +
3
(a) mn(a)
TG hvm vm |Fmn
F
|0i.
192 2
(4.9)
In the Einstein frame, where = 0, there is no contribution from the Tkk term. However, in
this frame there is the super-Weyl Jacobian,
Z
1
(3T
)
d2 2EE W (a) W (a) + h.c.
LJ =
G
16 2
3
(a) mn(a)
TG A AFmn
F
+ ,
(4.10)
=
192 2
where we have used E | =
element is
1
12 A A
+ . With this Jacobian, the Einstein-frame matrix
3
(a) mn(a)
TG hvm vm |Fmn
F
|0i,
(4.11)
192 2
which is in complete agreement with the amplitude in the supergravity frame. The results
in the two frames are identical because of the super-Weyl Jacobian.
ME = M0 +
5. Summary
In this paper, we studied the quantum consistency of the supergravity Lagrangian. We
used a superspace approach in which the supergravity Lagrangian does not automatically
give canonically normalized Einstein gravity. In the literature, Einstein gravity is recovered
after a redefinition of the component fields. In this paper, we showed that the field
redefinition is, in fact, a super-Weyl transformation, and we demonstrated a systematic
way to do the field redefinition in superspace. This approach provides us with a clear
understanding of supersymmetry transformations in the Einstein frame.
Supersymmetry transformations in the Einstein frame must preserve the Einstein frame
condition, so they differ from the original supersymmetry transformations defined in the
supergravity frame. We showed that the Einstein-frame supersymmetry transformations are
365
ordinary supersymmetry transformations, SUSY , combined with compensating super-Weyl

transformations, SW , which are necessary to maintain the Einstein frame condition.
The compensating super-Weyl transformations are, at the quantum level, anomalous.
Because of this fact, one must be careful when studying the supersymmetry invariance
of the quantum effective action. In this paper we emphasized that the super-Weyl
transformation used to pass to the Einstein frame is anomalous. It gives rise to an
anomalous Jacobian that must be included in the bare Einstein-frame Lagrangian. The
1PI Einstein-frame Lagrangian is supersymmetric because the variation of this Jacobian
precisely cancels the anomaly arising from the frame-restoring super-Weyl transformation.
If the Jacobian were omitted, the 1PI Lagrangian would not be invariant under Einsteinframe supersymmetry transformations. Consistency demands that the Jacobian be included
in the bare Einstein-frame Lagrangian.
Acknowledgements
The work of J.B. is supported by the US National Science Foundation, grant NSFPHY-9970781. T.M. is supported by US National Science Foundation under grant NSFPHY-9513835, and by the Marvin L. Goldberger Membership. E.P. is supported by the
Department of Energy, contract DOE DE-FG0292ER-40704.
Appendix A. Globally supersymmetric gauge theories

In this appendix, we show how the supersymmetry transformations are defined in
globally supersymmetric gauge theories. In particular, we demonstrate how the Wess
Zumino gauge condition is maintained after supersymmetry transformations. We will
see that there is a close analogy between supersymmetry transformations in globally
supersymmetric gauge theories and supergravity transformations in the Einstein frame.
Consider a globally supersymmetric gauge theory, with the Lagrangian
Z
(A.1)
L = Lgauge + d4 eV ,
where Lgauge is the kinetic term for the gauge multiplet. This Lagrangian is invariant under
the superspace supersymmetry transformations:
SUSY V = A A V ,
SUSY = A A ,
(A.2)
where a = i a + i a and is the supersymmetry transformation parameter. The

component transformations can be determined by expanding (A.2) in powers of .
The Lagrangian (A.1) contains the lower components of V , which are gauge degrees of
freedom. Usually, it is convenient to use a Lagrangian in which these lower components
are eliminated by a gauge transformation. This frame is often called the WessZumino
gauge; it is obtained by the following field redefinitions:
366
VWZ = V ,
WZ = e ,
(A.3)
where is a chiral superfield. The gauge parameter is chosen so that all the unphysical
fields are eliminated from the Lagrangian. The conditions are

(A.4)
(D V )| = (D )|,
D 2 V | = D 2 ,
V | = | + ,
so in the chiral basis is simply
= 12 C + i + i + 2i 2 (M + iN),
(A.5)
where C, , M + iN are the lowest, and 2 components of the vector superfield V ,

respectively. Note that the field is not determined by the WessZumino gauge conditions;
it is the parameter of an ordinary gauge transformation. In terms of the new variables, the
Lagrangian becomes
Z
eV WZ
LWZ = Lgauge + d4 WZ
Z
= Lgauge + d4 WZ
eVWZ WZ .
(A.6)
The WessZumino gauge Lagrangian LWZ contains only the physical fields.
The WessZumino gauge conditions, however, are not preserved by the supersymmetry
transformations (A.2). They must be supplemented by compensating superfield gauge
transformations:
VWZ = A A VWZ + ,
WZ = A A WZ + WZ ,
(A.7)
where, in the chiral basis,

= m (2vm + 2m ) + 2 .
(A.8)
Here vm and are the gauge boson and gaugino fields, respectively.
The transformations (A.7) are combinations of the original supersymmetry transformations (A.2) and the frame-restoring gauge transformations. They leave invariant the Wess
Zumino-gauge Lagrangian. Furthermore, if we expand the LHS of (A.7) in powers of ,
we obtain the WessZumino-gauge supersymmetry transformations given, for example,
in [2]. Note that in WessZumino gauge, a theory with a gauge anomaly would also have a
supersymmetry anomaly because the transformations (A.2) contain ordinary gauge transformations.
Finally, we comment on the difference between (A.8) and (3.10), that is, on the
different way we treat the imaginary part of the lowest component of the compensating
transformation parameters. In each case, the term is not determined by the WessZumino
gauge/Einstein frame conditions. In globally supersymmetric gauge theory, we choose
not to fix in |; it is the degree for freedom associated with an ordinary gauge
transformation. By contrast, in supergravity, we completely fix it and demand that the
imaginary part of E | vanish. The first term in (3.10) ensures that the imaginary part of
E | does not reappear in the Lagrangian after a supersymmetry transformation.
367
Appendix B. Component expression for the Jacobian

In this appendix, we present the complete component expression for the Jacobian that
arises from the super-Weyl transformation required to pass to the Einstein frame.
As we have seen in this paper, the bare Lagrangian in Einstein-frame supergravity is
given by
bbare = LE + LJ ,
L
(B.1)
where LE is the classical supergravity Lagrangian whose component expression is given,

for example, in [1,2]. LJ is the Jacobian. At one-loop level, LJ is given by
Z
1
(3TR 3TG ) d2 2EE W (a) W (a) + h.c.,
(B.2)
LJ =
16 2
where the chiral superfield E is given in Eqs. (3.4) and (3.5) with = 0:
E = A + 2 + 2 F , with
A =
1
12 K,
i
1
6 Ki ,
F =
i
1
6 Ki F
i j
1
12 Kij .
(B.3)
(B.4)
Expanding Eq. (B.2), we obtain

e

1
(a) mn (a)
LJ =
(3TR 3TG ) A Fmn
F
16 2

(a) m
(b) (c)
+
2iA Dm (a) f abc vm

(b) (c)

2iA (a) m Dm (a) f abc vm

i
(a)
bm
2

i
(a)
bm
2
+ 2A Daux (a) Daux (a)

b(a)
+ iA m kl m (a) + m kl m (a) Fkl(a) + F
kl
(a)
2 mn (a) + mn (a) Fmn

+ 2 i mn (a) m n (a) + 14 m m (a) (a)

+ 2 i mn (a) m n (a) + 14 m m (a) (a)

+ 2 i (a) (a) Daux (a)

F (a) (a) F (a) (a) ,
(B.5)
(a)
bmn
is the supercovariant field strength:
where F

i
(a)
(a)
bmn
= Fmn
m n (a) + m n (a) n m (a) n m (a) .
F
2
(B.6)
For a detailed explanation of the notation, see [2].

The full component expression is given by substituting the solutions to the Einsteinframe auxiliary field equations of motion:
368

eK/2Dj P + 12 Kj kl k l + 14 j h(ab) (a) (b) ,

i
R(ab) 1
(b)
i (c)
i
(c)
= h
D + i h(bc) i h(bc)
,
2 2

i
em Ai Ki D
em Ai + 1 Kij i m j 3 hR (a) m (b)
Ki D
=
4
4 (ab)
2
1

(a)
+ i 2 Ki Xi(a) Ki Xi(a) + iD (a) vm
,
F i = K 1
Daux (a)
bm
ij
is the
where Di P Pi + Ki P ,
(a) i
(a)
i
i
e
X(a) [2].
with X , and Dm A Dm A vm
X(a)
Killing vector, D (a)
(B.7)
is the Killing potential associated
References
[1] E. Cremmer, B. Julia, J. Schenk, S. Ferrara, L. Girardello, P. van Nieuwenhuizen, Nucl. Phys.
B 147 (1979) 105;
J. Bagger, Nucl. Phys. B 211 (1983) 302;
E. Cremmer, S. Ferrara, L. Girardello, A. van Proeyen, Nucl. Phys. B 212 (1983) 413.
[2] J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton University Press, 1992.
[3] P. Howe, R. Tucker, Phys. Lett. B 80 (1978) 138.
[4] B. de Wit, M.T. Grisaru, in: I.A. Batalin, C.J. Isham, G.A. Vilkovisky (Eds.), Quantum Field
Theory and Quantum Statistics, Adam Hilger, 1987.
[5] A. Dolgov, V. Zakharov, Nucl. Phys. B 27 (1971) 525;
Y. Frishman, A. Schwimmer, T. Banks, S. Yankielowicz, Nucl. Phys. B 177 (1981) 157;
S. Coleman, B. Grossman, Nucl. Phys. B 203 (1982) 205.
[6] G. Lopez-Cardoso, B. Ovrut, Nucl. Phys. B 419 (1994) 535;
See also, I.L. Buchbinder, S.M. Kuzenko, Phys. Lett. 202 (1988) 233.
[7] J.A. Bagger, T. Moroi, E. Poppitz, JHEP 0004 (2000) 009.
[8] V. Kaplunovsky, J. Louis, Nucl. Phys. B 422 (1994) 57.

Reaction operator approach to non-abelian

energy loss
M. Gyulassy a , P. Levai a,b, , I. Vitev a
a Department of Physics, Columbia University, 538 W 120-th Street, New York, NY 10027,USA
b KFKI Research Institute for Particle and Nuclear Physics, P.O. Box 49, Budapest, 1525, Hungary
Received 20 June 2000; accepted 2 November 2000
Abstract
A systematic expansion of induced gluon radiation associated with jet production in a dense QCD
plasma is derived using a reaction operator formalism. Analytic expressions for the induced inclusive
gluon transverse momentum and light-cone momentum distributions are derived to all orders in
powers of the opacity of the medium for gluons, Ng /A = L/g . The reaction operator approach
leads to a simple algebraic proof of the color triviality of single inclusive distributions and leads
to a solvable set of recursion relations. The analytic solution to all orders in opacity generalizes
previous continuum solutions (BDMPS) by allowing for arbitrary correlated nuclear geometry and
evolving screening scales as well as the inclusion of finite kinematic constraints. The latter is
particularly important because below LHC energies the kinematic constraints significantly decrease
the non-abelian energy loss. Our solution for the inclusive distribution turns out to have a much
simpler structure than the exclusive (tagged) distribution case that we studied previously (GLV1).
The analytic expressions are also obtained in a form suitable for numerical implementation in Monte
Carlo event generators to enable more accurate calculations of jet quenching in ultra-relativistic
nuclear collisions. Numerical results illustrating the first three orders in opacity are compared to the
self-quenching hard radiation intensity that always accompanies jet production in the vacuum. A
surprising result is that the induced gluon radiation intensity is dominated by the (quadratic in L)
first order opacity contribution for realistic geometries and jet energies in nuclear collisions. 2001
Elsevier Science B.V. All rights reserved.
PACS: 12.38.Mh; 24.85.+p; 25.75.-q
Keywords: Jet quenching; Non-abelian energy loss; Opacity expansion; Reaction operator
1. Introduction and summary

One of the expected signatures of the quarkgluon plasma created at RHIC and LHC
energies of s ' 200 A GeV and s ' 5400 A GeV is jet quenching [128]. At the
E-mail address: plevai@rmki.kfki.hu (P. Levai).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 2 - 0
372
M. Gyulassy et al. / Nuclear Physics B 594 (2001) 371419
SPS energies of s ' 20 A GeV, on the other hand, no quenching of moderate p <
4 GeV hadrons [24] was observed even in Pb + Pb. This may be due to the break-down of
pQCD at such low momentum scales and the difficulty of disentangling non-perturbative
multiparticle production effects such as the Cronin and intrinsic kT effects [25]. Also
finite kinematic constraints may limit the energy loss at low jet energies. In addition, at
SPS energies the duration of jet propagation in high density matter is limited to at most a
few fm/c due to longitudinal expansion. This motivated our previous work (GLV1) [26,
27] to study the problem of energy loss in thin plasmas. In that work we considered
the exclusive (tagged) case of energy loss associated with a fixed number of interactions.
Rough numerical estimates for nominal 5 GeV jets suggested that at those relative small
energies and small opacities, the radiative energy loss may indeed be much smaller than
predicted by Baier, Dokshitzer, Mueller, Peign and Schiff (BDMPS) [1012,16] and
Zakharov [13,14] for asymptotic jet energies.
Several approaches have been advanced to compute non-abelian energy loss. One
approach is aimed at treating relatively thick targets, which while small compared to the
jet coherence length are large enough that many collisions occur in the medium and allow
for a continuum formulation of the problem. In [1012,16] an effective 2D Schroedinger
equation formulation was introduced, and in [13,14,1820] a path integral formulation of
the problem was developed. Another approach [6,7,2628], which we pursue here, aims to
address the problem of radiative energy loss in thin plasmas, a few mean free paths thick,
by computing directly the radiation pattern from the finite number of Feynman diagrams
for the case of a few collisions.
The advantage of the former is that it can make direct contact with the conventional
LandauPomeranchukMigdal effect [30,31] in QED. It has the disadvantage that the
continuum solutions obtained for the case of many collisions cannot be directly applied
to the case of a few collisions. Also the effects of finite kinematical constraints are difficult
to include in such approaches. The advantage of the second approach is that the few
collision case can be computed directly from the finite number of amplitudes involved.
The disadvantage is that the LPM limit is out of reach, and the work involved in summing
diagrams increases exponentially with the number of collisions. In the present paper we
introduce a new approach to bridge the gap between the two previous approaches. Our
new approach is based on the construction of a suitable reaction operator, R n , from
which recursion relations for the inclusive gluon distribution can be derived and solved
analytically at arbitrary order n.
For thin QCD plasmas that can be formed in nuclear collisions, the gluon radiation
intensity can be studied systematically through an expansion in powers of opacity defined
by the mean number of collisions in the medium

Z
Z
z
RG
dN el
L Nel
2 del (z)
z, =
= dz d q
log
,
(1)
n = =
2
2
A
d q
c
dy 2RG
0
where N is the number of targets in the medium of transverse area A . An opacity
expansion in terms of the path integral formulation was introduced in [19,20,34]. In [26,
27] we considered the exclusive tagged case where all N target partons interacting with
373
the jet. Here (and [28]) we extend that calculation to the inclusive case where fluctuations
of the number of collisions are allowed and only the geometry is fixed.
For a homogeneous rectangular target of thickness, L, the density is = N/(LA )
and the mean free path is = 1/(el ). The opacity is then simply L/ = Nel /A .
For application to high transverse momentum jets propagating through cylindrical nuclear
reaction geometries, we can interpret L = 1.2 A1/3 Rs . For even more realistic (3 + 1)D
Bjorken and transverse expanding Gaussian cylindrical geometry such as
2

xE + 1 2
0
2|Ex|1
exp
I0
,
(2)
(Ex, ) = 0
2
2
RG
RG
L is replaced by the equivalent rms Gaussian transverse radius RG = 0.75 A1/3 fm. The
rightmost expression in (1) is obtained by averaging the number of collisions of a transverse
jet in this expanding Gaussian cylinder with the initial jet production coordinate, xE 0 ,
averaged over (Ex0 , 0 ). Here 0 is the formation time of the plasma. At RHIC energies
( s ' 200 A GeV) the expected rapidity density of the gluons is dN/dy ' 1000 for A =
200 [29]. For a typical elastic gluongluon cross section el 2 mb and a plasma formation
time 0.5 fm/c, the opacity is moderately small n < 10. This suggests that neither the
thin nor the thick plasma approximations may hold for applications to nuclear collisions.
Therefore, an alternate method is needed to handle the intermediate (mesoscopic) case.
An important simplification may be anticipated due the non-abelian analog [6,10,28] of
the LandauPomeranchukMigdal (LPM) effect [30,31]. While the number of scatterings
is moderately large, the radiation intensity angular distribution and total energy loss are
controlled by the combined effect of the number of scatterings n = L/ and the formation
probability pf L/ lf = Lk2 /(2xE) of the gluon in the medium. Here x is the lightcone momentum fraction and k xE is the transverse momentum of the radiated gluon.
Therefore, the opacity expansion may converge more rapidly than one would first expect.
One of the surprising results derived here and summarized in [28] is that the inclusive
induced gluon radiation intensity is dominated by the (already quadratic in L) first order
opacity contribution for realistic geometries and jet energies in nuclear collisions.
In GLV1 [27], we showed in contrast that in the exclusive tagged case, i.e., with all
recoiled partons measured in coincidence with the jet and gluon, a much more complex and
nonlinear non-abelian radiation pattern emerges. The tagged recoil partons act as additional
color dipole antenna that interfere in a nontrivial way with the jets color dipole antenna.
The resulting color algebra cannot be reduced to simple powers of the color Casimirs. We
found that the gluon number distribution for the case of ns = N tagged target partons has
the form

(3)
dNg(ns ) = dNg(0) 1 + (1 + R)ns 1 f (ns ) (, ) /Zns ,
where f (ns ) depends on two dimensionless ratios = k2 /2 and = / lf = 2 /2xE,
and depends only weakly on ns . The color dependence enters through a rapidly increasing
function of R = CA /CR , the ratio of the Casimirs in the adjoint (dA = (Nc2 1)dimensional) gluon representation and the Casimir, CR , of the dR -dimensional jet
representation. In GLV1 unitarity was imposed by a wave function renormalization factor,
374
Zns , that was computed perturbatively. For the inclusive case studied here, unitarity is
assured by the inclusion of certain contact double Born (virtual) terms as emphasized in
BDMPS [12].
To address energy loss of hard probes produced in nucleus-nucleus collisions we
concentrate as in [27,28] on the case of hard jet produced inside the plasma at a finite point
(t0 , z0 , x0 ) rather than on the GunionBertsch problem [32] in which the jet is replaced
by a high energy beam of quarks or gluons prepared in the remote past. We employ as in
GW [6,27] static color-screened Yukawa potentials to model interactions in a deconfined
quarkgluon plasma. The Fourier and color structure of those potentials are assumed to
have the form
Vn = V (qn )eiqn xn = 2(q 0 )v(Eqn )ei qE n Exn Tan (R) Tan (n),
where xE n is the location of the
v(Eqn )
nth
(4)
(heavy) target parton and
4s
4s
,
=
E 2n + 2 (qnz + in )(qnz in )
q
(5)
where 2n = 2n = 2 + q2n . The small transverse momentum transfer elastic cross section
between the jet and target partons in this model is
del (R, T ) CR C2 (T ) |v(q)|2
=
.
dA
d 2q
(2)2
(6)
In our notation transverse 2D vectors are denoted as p, 3D vectors as pE = (pz , p), and four
E ) = [p0 + pz , p0 pz , p].
vectors by p = (p0 , p
The color exchange bookkeeping with the target parton n is handled by an appropriate
SU(Nc ) generator, Ta (n), in the dn -dimensional representation of the target. (Tr Ta (n) = 0
and Tr (Ta (i)Tb (j )) = ij ab C2 (i)di /dA .) We will assume all target partons are in the same
dT -dimensional representation with Casimir C2 (T ).) We denote the generators in the dR dimensional representation corresponding to the jet by a ta with aa = CR 1. The elastic
cross section with target parton i with the jet is therefore proportional to the product of
Casimirs, CR C2 (T ).
The analytic results derived below with the reaction operator approach do not depend
on the actual form of v, but the Yukawa form will be used for numerical estimates. Recall
that in a thermally equilibrated medium at temperature T , the color screening mass in
pQCD is given by = 4s T 2 . Also there is a cut-off frequency for soft gluon modes
pl . We take here ' pl ' 0.5 GeV for numerical estimates. In perturbation theory,
2 / 4s3 provides a measure of the , the density of plasma partons weighed by
appropriate color factors.
This paper is organized as follows: in Section 2 we review the effect of unitarity
corrections in the simple elastic jet scattering case. We show how the jet probability
is conserved up to first order in opacity through the inclusion of contact double Born
graphs. This is the simplest example illustrating how a factor 1/2 arises from longitudinal
momentum contour integrations in the contact limit. In Section 3 we proceed to the case of
gluon bremsstrahlung associated with hard probes produced in a dense medium. We discuss
the basic diagrammatic rules and specify the assumptions and approximations used in this
375
work. We extend the algebraic classification of diagrams given in Ref. [27] to include
virtual double Born amplitudes needed in the inclusive case. Section 3.3 summarizes
the rules of diagrammatic calculus that emerge from detailed analysis of diagrams in
Appendix A through Appendix E.
In Section 4 the new reaction operator formalism is developed. First, operators D n , Vn
in Eqs. (76), (82) are constructed from the diagrammatic rules. Products of these operators
create partial sums of direct and virtual amplitudes from the initial hard vacuum amplitude.
Those partial sums, Eq. (60), form 3n classes of diagrams that can be conveniently
enumerated via a tensor notation and used to construct recursion relations. In Section 4.2,
the reaction operator, R n = D n D n + Vn + Vn , that relates the nth order in opacity inclusive
radiation probability distribution to classes of diagrams of order n 1, is derived. The
resulting simple recursion relation, Eq. (96), can be solved in closed form. The general
solution, Eq. (101), is suitable for implementation in Monte Carlo event generators to study
observable consequences of jet quenching in nuclear collisions.
Color triviality of the inclusive distribution is proven to all orders algebraically with
Eq. (101). The proof is much simpler and more transparent than in the path integral
formulations [13,20,34] and is not limited to quark jets.
In Section 4.3 a compact general expression for the momentum transfer averaged
inclusive distributions, Eq. (113) is derived. Appendix F provides an independent check
of this solution through second order starting from the amplitude iteration technique.
Numerical results comparing angular distributions of gluons up to the first three orders
in opacity are presented in Section 5.1. Analytic and numerical results for the angular
integrated intensity distributions are compared in Section 5.2. It is shown that the induced
intensity is dominated by the first order in opacity result that is already quadratic in L.
A brief summary of these results up to second order in opacity was reported in Ref. [28].
The main result of this paper is the derivation through the reaction operator approach of the
solutions, Eqs. (101), (113), that specify the inclusive non-abelian radiation distribution to
any order in opacity.
2. Elastic scattering and unitarity

To illustrate how the double Born graphs cancel direct contributions to preserve unitarity
we review here the simplest case of elastic scattering. Consider a wave packet j (p) of a
parton prepared at time t0 and localized at xE 0 = (z0 , x0 ) in color representation R. The
E in the absence of final state
(color matrix) amplitude to measure its momentum as p
interactions is
M0 ieipx0 j (p).
(7)
Multiplying |M0 |2 by the invariant one particle phase space element d 3 pE /((2)3 2|Ep|) and
taking the color trace gives the unperturbed inclusive distribution of jets in the wave packet
d 3 N0 = Tr |M0 |2
3E
E
d 3p
2 dR d p
=
|j
(p)|
.
2|Ep|(2)3
2|Ep|(2)3
(8)
376
E from an initial wave packet formed at z0 followed by single

Fig. 1. Graphs that produce a jet with p
and double Born scattering center located at z1 .
Consider next the effect of final state elastic interactions with an array of static potentials
localized at xE i = (zi , bi ) using
Z
N
X
d 3 xE
v(Ex xE 1 )Ta (i) (Ex, t)Ta (R)D(t)(E
x, t),
(9)
HI (t) =
i=1
it
and Tr Ta (i)Tb (j ) = ij ab C2 (T )dT /dA . We will compute the three

where D(t)
=
graphs in Fig. 1. The first order, direct amplitude to scatter with one of the (static) target
partons is
Z
d 4q
j (p q)(p q)v(q)D(2p q)
M1 = ieipx0
(2)4
N
X
eiq(xj x0 ) Ta (j )Ta (R),
(10)
j =1
where (p) (p2 + i)1 and D(p) = p0 . The sum of double Born amplitudes in the
same external potential is
Z
d 4 q1 d 4 q2
j (p q1 q2 )(p q1 q2 )
M2 = ieipx0 Ta (R)Tb (R)
(2)4 (2)4
D(2p 2q2 q1 )v(q1 )(p q2 )D(2p q2 )v(q2 )
N
X
ei(q1 +q2 )(xj x0 ) Ta (j )Tb (j ).
(11)
j =1
Double Born with two different centers will not contribute because Tr Ta (j ) = 0. To first
order in opacity, the probability distribution of jets is given by
d 3N =
2 d 3 p
E
1
Tr M0 + M1 + M2 +
dT
2|Ep|(2)3
= d 3 N0 + d 3 N1 +
377
d 3 pE
1
Tr 2Re(M1 M0 ) + 2Re(M2 M0 )
+ ,
dT
2|Ep|(2)3
(12)
where we separated the direct (|M1 |2 ) and unitary correction contributions with
Z
d 4q d 4q 0
d 3 N1
dR j (p q)j (p q 0 )D(2p q)D(2p q 0 )
2|Ep|(2)3 3 =
d pE
(2)4 (2)4
1
1
v(q)v (q 0 )
(p q)2 + i (p q 0 )2 i
CR C2 (T ) X i(qq 0 )(xj x0 )
e
.
dA
N
(13)
j =1
To proceed further, we consider a Yukawa potential as in Eqs. (4), (5) and assume that all
the xj are distributed with the same density
N
(z),
(14)
(Ex) =
A
R
where dz(z)
= 1. Second, we assume that the observed p = (E, E, 0) = [E + , 0, 0] is
high as compared to the potential screening scale, i.e.,
E + ' 2E .
(15)
We also assume that the distance between the source and scattering centers are large
compared to the interaction range
zi z0 1/.
(16)
Finally, we assume that the source current or packet j (p) varies slowly over the range of
momentum transfers supplied by the potential. We can then approximate the qz contour
integrals ignoring the qz dependence of the source j (p q). With (16) we can also
neglect the contributions to the qz contour integrals due to singularities of the potentials
at i , 2 = 2 + q2 . With these simplifying assumptions, only the residues with qz
i + q2 /E + and qz0 +i + q0 2 /E + contribute
Z
d 3 N1
dqz d 2 q dqz0 d 2 q0
j (p q)j (p q 0 )v(q )v (q0 )
2|Ep|(2)3 3 = NdR
d pE
(2)3 (2)3
E+
E+
CR C2 (T )
dA
E + qz qz2 q2 + i E + qz0 qz0 2 q0 2 i
i(EqEq0 )(Ex Ex )
1
0
e
Z

d 2 q d 2 q0
0
j (p q)j (p q0 ) ei(qq )(x1x0 )
NdR
2
2
(2) (2)
0 CR C2 (T )
.
(17)
v(q)v (q )
dA
A major simplification (that we will also use in the radiation case) occurs if the relative
transverse coordinate (impact parameter) b = xi x0 varies over a large transverse area,
378
A , relative to the interaction area 1/2 . In this case, the ensemble average over the
scattering center location reduces to an impact parameter average as follows
Z 2
d b
.
(18)
hi =
A
The ensemble average over the phase factor in Eq. (17) then yields
(2)2 2
0
(q q0 ).
ei(qq )b =
A
(19)
Note that we ignored a small phase shift in Eq. (17) exp[q2 (z1 z0 )/E + ] 1 on
account of our high energy (eikonal) assumption. The ensemble averaged first order in
opacity direct contribution to the jet distribution therefore reduces to the familiar form
noting Eq. (6)
Z
N
del (R, T )
d 2q
2|Ep|(2)3 d 3 N1
|j (p q)|2
3
dR
d pE
A
(2)2
d 2q
Z
N del (R, T ) L
= |j (p)|2 ,
(20)
|j (p)|2 d 2 q
A
d 2q
where the last line only holds if the initial packet is very wide in momentum space
compared to the momentum transfer scale .
Next we turn to the unitary corrections. The first order term hM1 M0 i = 0 on account of
Tr Ta (j ) = 0. A non-vanishing unitarity correction arises however from
Z

d 4 q1 d 4 q2
1
Tr M2 M0 =
j (p q1 q2 )j (p)
dT
(2)4 (2)4
D(2p q2 )D(2p q1 q2 )

V (q1 )
V (q2 )
Tr Ta (R)Tb (R)
2
2
(p q1 q2 ) + i (p q2 ) + i
+
* N
X

i(q1 +q2 )(xj x0 ) 1
e
Tr Ta (j )Tb (j ) .
dT
(21)
j =1
Note that the impact parameter average constrains q1 + q2 = 0 in this case (q10 = q20 = 0
for static potentials). The phase factor requires that we close the q1z contour in the lower
E2 = 0
half-plane. We pick up the 2i Res(q1z = i q2z ). This results in setting qE 1 + q
throughout. Again the residue from the second pole in the lower half-plane qz = i is
suppressed by the phase factor. We are left with
Z

d 2 q dq2z (4s )2 CR C2 (T )
1
Tr M2 M0 i NdR |j (p)|2
dT
(2)2 (2) A
dA
1
E+
, (22)
2
2 q2 + i
(q2z + 2 )2 E + q2z q2z
where 2 = 2 + q2 . The q2z can be performed yielding
1
1
i
E+
dq2z
.
2
2
2
2
2
+
2
2 (q2z + ) E q2z q2z q + i
2 (q + 2 )2
379
(23)
The double Born contact contribution to the differential yield at first order in opacity is
therefore
2

L
1
Tr 2 Re M2 M0 j (p) dR .
(24)
dT
The inclusive first order in opacity elastic distribution is therefore

2|Ep|(2)3 d 3 N
dR
E
d 3p

2

Z
2 del (R, T )
N
L
L
d 2 q

+
+
O
j
(p
q)
= |j (p)|2 1
2
2
(2)
d q

2

Z
Z
2

2
d q del (R, T )
L
2
2

el (q) j (p q) + O
= |j (p)| + dz (z, 0)
.
(2)2
d 2q
(25)
This is the first order term in the Glauber multiple scattering series. Note that the double
Born term cancels exactly the momentum integrated direct contribution, thereby enforcing
probability conservation. In addition, for very narrow coordinate space packets, i.e. wide
in momentum space so that |j (p q)|2 |j (p)|2 , we see that it actually cancels the first
order direct distribution differentially.
We derived this cancellation assuming p = 0 throughout, but it holds generally since
all we needed was that the phase shift (zi z0 )(q p)2 /E + 1. The point of this brief
review was to emphasize that the unitarity cancellation arose due to the factor of 12 that
appeared in Eq. (23) from the longitudinal momentum contour integration and from the
factor 1 that appeared due to moving the potential from one side of a cut to the other.
This is a generic property of contact interactions that also holds in the radiation case as
shown in detail the Appendix.
3. Soft gluon radiation

3.1. Kinematics, triple gluon vertices and color algebra
Consider a source J that produces a jet that subsequently radiates a gluon with four
momentum k, polarization (k), and emerges with momentum p. In light-cone components

k
k = xE + , k 0 , k , (k) = 0, 2 + , , p = (1 x)E + , p , p . (26)
xE
Soft radiation is defined as x 1 so that, for example, p+ k + and p = p2 /(1
x)E + k = 0 = k2 /xE + .
We adopt the shorthand notation of Refs. [26,27]
380
(k qi qj )2
k2
(k qi )2
,
i =
,
(ij ) =
,
2
2
2
Pj
(k m=i qm )2
.
(ij ) =
2
In the soft eikonal kinematics that we consider
0 =
(27)
(p + k)2
.
(28)
E+
The soft gluon and soft rescattering approximations will allow us to simplify the analytic
results using
E + k + (ij )
J (p Q + k) J (p + k) J (p),
k2
k
,
k(p Q) kp
,
(k)(p Q + k)
x
2x

1
(p Q) (p Q + k) .
(p Q)(p Q + k)
kp
(29)
Scattering of the jet orR radiated gluon with the potential centered at position xEn
introduces an integration d 4 qn /(2)4 in the amplitude over the potential V (qn ). We
use the GW model potential Eqs. (4), (5). If the scattering involves a momentum exchange
with the high energy jet, the vertex factor is simply iE + in the eikonal limit. Neglecting
the spin of the jet parton, each intermediate jet propagator brings in a factor +i(p Q),
where (p) = 1/(p2 + i). The gluon emission vertex then gives in this approximation a
factor +igs (2p + k 2Q) , where Q is the sum of the subsequent momentum transfers.
For scattering of the radiated gluon, consider the triple gluon tensor, 0 (k, q) for a
g(k q, , a) + V (q, = 0, b) g(k, , c). In our model, qn0 = 0, |qnz | = |qn+ | = |qn |
|qn |, and thus the qn correspond to small (almost) transverse momentum transfers. If the
last interaction of the gluon is with center m, then that last vertex is
0 (k, qm , k qm ) = 2g (k + qm ) g 0 + (2qm k) g 0 .
(30)
The color factor for the above clockwise kinematic convention is f cba .
The contraction of Eq. (30) with the final polarization gives
(k; qm ) (k, qm ) (k) = 2 (k + qm ) 0 + 2g 0 (qm ).
(31)
For the case of two gluon rescatterings the two vertices combine
(k; qn , qm ) (k qm , qn )g (k; qm ),

(k; qn , qm ) = 42 (3k + qn + qm ) 0 + g 0 4(qn + qm )

2 0
2(k qm + qn ) (qm ).
g 0 2qn (k qm ) + qm
(32)
Note that using +g to combine the vertices above requires us to insert i(k Q) for
each gluon propagator, where we use (p) = 1/(p2 + i).
The gluon vertex factors contracted by emission vertex factors then give

m 2p + k qm (k; qm ) 2E + (k qm ) ,

(33)
mn 2p + k qm qn (k; qn , qm ) 2E + k + (k qm qn ) .
381
We neglected above corrections of O(x). Eq. (33) also applies approximately if the
emission occurs while the jet is off shell with p Q, with Q = (0, , q) total momentum
transfer still to be acquired from subsequent jet re-interactions. Replacing p by p Q adds
for example 2(k + )2 q which is again O(x = k + /E + ) smaller than the terms retained
above.
The full triple gluon vertices including coupling and color algebra for producing a final
color c from initial color a followed by color potential interactions an and am are then
given by

m m (f cam a )(igs ta ) Tam (m)
2gs E + (k qm )[c, am ]Tam (m),

mn mn (f can e )(f eam a )(igs ta ) Tan (n) Tam (m)

2igs E + k + (k q1 q2 )[[c, an ], am ] Tan (n)Tam (m) .
(34)
To complete the Feynman rules for our problem, we note that there is a factor
iJ (p + k Q)ei(p+kQ)x0
for the jet production vertex at x0 if the net subsequent target momentum transfer to the jet
plus gluon system is Q. The hard jet radiation amplitude to emit a gluon with momentum,
polarization, and color (k, , c) without final state interactions is then
M0 = iJ (p + k)ei(p+k)x0 (igs )(2p + k) (k)i1(p + k)c
k
k
J (p + k)ei(p+k)x0 (2igs ) 2 c J (p)eipx0 (2igs ) 2 ei0 z0 c,
k
k
corresponding to Eqs. (9), (10) in Ref. [27].
(35)
3.2. Graphical shorthand

In Ref. [26] we introduced the following notation, Mns ,m,l for the direct (tagged) ns
scattering center amplitude for emitting a gluon between points zm and zm+1 with a
P
final state interaction pattern encoded by l = n1s i 2im1 , with i = 0, 1 depending
on whether the interaction was with the jet of the gluon. To include virtual double Born
corrections, it is convenient to associate a more specific graphical coding to keep track of
the many different possibilities.
A graph consists of a gluon emission vertex, Gm , for emitting a gluon between centers
zm and zm+1 as before, but also a specific set of direct interactions, Xi,i with i = 0, 1
for center i, and double Born interactions denoted, Oj,aj , for center j . The index aj =
0, 1, 2 denotes a contact interaction at center j with the jet, gluon and both jet+gluon,
respectively. In this notation
" m
#
" n
#
Y
Y
Xi,0 Gm
Xj,j ,
(36)
Mn,m,l =
i=0
j =m+1
here j = 0, 1 is the j th binary bit in the expansion of l 2m+1 . For example, the
graph corresponding to gluon emission before the first scattering center followed by a jet
382
interaction with center 1 and the gluon rescattering at center 2 is given by M2,0,2 =
G0 X1,0 X2,1 . This is simply an algebraic way of representing the corresponding graph.
Similarly, M0,0,0 = G0 and M1,0,1 = G0 X1,1 . This notation is particularly useful to specify
more complicated diagrams that arise from multiple direct and double Born interactions.
It extends the algebraic approach to the topologically distinct graphs of Ref. [27] by the
inclusion of virtual corrections. Any diagram can be written in the form
#
" n
#
" m
Y
Y
Ti,i Gm
Tj,j , Ti,i , Ti,i (Xi,i , Oi,ai ).
(37)
M=
i=0
i=m+1
While Eq. (37) can be used to enumerate all single gluon emission with rescatterings
diagrams arising from a target with n aligned centers, it will prove convenient to group
diagrams into classes of graphs that can be iteratively built from a diagrammatic kernel.
The two important kernels correspond to the two distinct possibilities for preparing an offshell parton. In the first case the jet originates asymptotically at infinity and then enters
the medium. It needs at least one inelastic interaction to get off-shell. This is the more
extensively studied GunionBertsch (GB) limit [10,11,32]. Depending on whether the first
interaction in the plasma is real or virtual

B (R) = 2igs k2 kq1 2 ei0 z1 [c, a1], real initiator,
1
k (kq1 )

(38)
Ker(GB) =
B (V ) = 2igs CA k kq1 ei0 z1 c,
virtual
initiator,
2
2
1
2
k
(kq )
1
is the effective color current [6,10] (here given in the eikonal approximation in the small
x limit). In Eq. (38) c and a1 are the color matrices of the radiated gluon and the soft
momentum transfer.
In the second and more experimentally relevant case, the parton is produced inside the
medium (e.g., through A + A q + q + X with high Q2 E +2 ) with accompanying
gluon radiation. The kernel or initiator for the hard production vertex is
Ker(H) = G0 = 2igs
k i0 z0
e
c,
k2
(39)
where c the color (generator) of the radiated gluon. Classes for both hard and Gunion
Bertsch cases can be constructed starting from the appropriate kernel. We focus here on
the hard jet case relevant to nuclear collisions with the hard production vertex localized
at z0 , as in Eq. (39). We consider the effect of final state interactions at positions zi > z0
along the direction of the jet.
Let A denote a class of graphs with NA members in which the last interaction has
occurred at position zj < zi . Each class includes one special graph, denoted A0 , where the
gluon is emitted after all real and virtual interactions in that class. We can enlarge this class
of graphs to include a direct hit at zi and label it by ADi , where Di specifies the direct
insertion iteration. This new class contains 2NA diagrams AXi,0 and AXi,1 obtained by
a direct interaction of either the jet or the previously emitted gluon. In addition, there is
new special diagram, AG1 Xi,0 Gi = (ADi )0 , where all interactions (direct or virtual)
including the one at zi are with the jet and the gluon is emitted after all interactions at zi <
383
z < . (G1 amputates the gluon emission vertex of A0 .) The new class with 2NA + 1
graphs is constructed operationally as
A H ADi = AXi,0 + AXi,1 + AG1 Xi,0 Gi .
(40)
Starting with A = G0 , AD1 Dns is then the set of 2ns +1 1 direct interaction diagrams
with centers 1 through ns as constructed in [27].
Similarly we can consider the possibilities that arise from inserting a double Born hit
at location zi . This case differs from the above only in that now a new subclass appears
because one of the legs can be attached to the jet line and the other one to the gluon line,
i.e., AOi,2 . The new class AVi with 3NA + 1 graphs including these virtual interactions
at center i are constructed operationally as
A H AVi = AOi,0 + AOi,1 + AOi,2 + AG1 Oi,0 Gi .
(41)
The classes and conjugate classes contributing to the gluon radiation distribution in the
opacity expansion at order (L/)n (el /A )n for n = 0, 1, 2 are listed in the table below.
Each class is obtained through ordered insertions from the hard kernel Ker(H) indicated
by G0 .
1. (L/)0 (el /A )0 contributions
1 (ns = 0) (ns = 0)
G0 G0
1 graph
(42)

4 (ns = 2) (ns = 0)
G0 V1 G0
4 graphs
(43)
9 (ns = 1) (ns = 1)
G0 D1 D1 G0
9 graphs
(44)

13 (ns = 4) (ns = 0)
G0 V1 V2 G0
13 graphs
(45)
57 (ns = 3) (ns = 1)
G0 D1 V2 D1 G0
30 graphs
(46)
G0 V1 D2 D2 G0
27 graphs
(47)
65 (ns = 2) (ns = 2)
G0 D1 D2 D2 D1 G0
G0 V1 V2 G0
49 graphs
16 graphs
(48)
(49)
The left column sorts contributions in terms of diagrams with fixed number of interactions
ns , n0s in the amplitude and complex conjugate amplitude. For example, there are 57
contributions involving amplitudes with 3 interactions times complex conjugate amplitudes
384
with 1 interaction that contribute to second order in opacity. Since two of the three
interactions in the amplitude must involve the same center, these 57 diagrams subdivide
naturally into the two class structures (46), (47) indicated in the right column. We note that
(42), (44) and (48) correspond to the direct diagrams computed in [27] for the exclusive
tagged spectrum.
3.3. Diagrammatic calculus
Whereas the time-ordered perturbation techniques [10] used in [27] greatly facilitated
the listing of higher order contributions, it is useful to show how those shortcuts emerged
from Feynman diagrams especially to check the shortcut rules that apply to the double
Born contributions. Detailed diagrammatic evaluation from which those rules emerge are
presented in Appendix A through Appendix E. They exhaust all possibilities that we
can encounter for single and double Born interactions in a potential model with multiple
centers. It is important to note that a double Born interaction at position zi can be thought
of as the contact limit of two direct hits at positions zi and zi+1 respectively
AOi,ai lim AXi,bi Xi+1,bi+1 .
zi+1 zi
(50)
We summarize here the kinematical and color parts of this diagrammatic calculus:
In all diagrams the jet production factors out in the form
J (p) eipx0
under the soft-gluon soft-interaction assumption Eq. (29). Therefore, the jet distribution can be factored out under the assumption that J is broad.
For graphs involving direct interactions only, i.e., of class G0 D1 D2 Di , the
Feynman diagrams reproduce the time ordered perturbation results listed in Ref. [27].
The simplest one interaction example is discussed in Appendix A. We note that
although Appendix B through Appendix E are aimed at clarifying the contact limits,
their intermediate results in the well-separated zi+1 zi = 1/ case are directly
comparable to [27].
A diagram which involves a contact contribution of the form AOi,0 or AOi,1 , i.e.,
both interactions are on a single line, is kinematically identical to a diagram in
the well-separated case, Xi,0 Xi+1,0 and Xi,1 Xi+1,1 respectively, but with zi =
zi+1 (50). This can be seen in Appendix B for the final state virtual interaction of the
radiated gluon and in Appendix C for a virtual final state scattering of the jet. There
is, however, a multiplicative factor of 12 per contact interaction that emerges from
contour integrals. Eq. (B.8) shows how the strength interpolates between 1 and 12 as
the separation varies for a Yukawa potential. Eq. (B.15) shows the generality of the
factor 12 in the contact limit.
For the case when one of the legs in the contact contribution is attached to the jet line
and the other to the gluon line, i.e., AOi,2 , the amplitude is kinematically identical to
a diagram AXi,0 Xi+1,1 in the well-separated case with zi = zi+1 (50). There are no
additional numerical factors for such mixed contact term as shown in Appendix D.
385
We also note that the contact limit of the diagram Xi,1 Xi+1,0 should not be added
to avoid over counting (i.e., they are topologically identical in the contact limit).
We show in Appendix E that the class of diagrams that have the gluon emission vertex
between the two momentum transfers of a double Born interaction are suppressed in
the contact limit by factors O(k + /E + ) and O(/k + ) and thus can be neglected
in the framework of our approximations (28). This verifies the naive argument from
time-ordered perturbation theory that such diagrams are of zero-measure and vanish
in the contact limit.
Each power of the potential comes with a weight (elastic scattering amplitude)
Z
d 2 qi
v(0, qi )eiqi bi Tai (i),
(i)
(2)2
that factors out as well under the assumptions (28), (29) and is still to be integrated
over. The amplitudes above eventually build into the elastic scattering cross section
el .
The appropriate color factor for the jet + gluon part can be constructed following
Ref. [27]. In the case of a virtual interaction certain simplifications, namely,
CA
c,
(51)
2
can occur even at the amplitude level.
The last critical rule that emerges is that direct terms come with a plus sign while each
virtual interaction contributes a minus.
These rules are illustrated to compute the classes of amplitudes needed up to second
order in opacity in Appendix F.
ai ai = CR 1,
ai [c, ai ] =
3.4. Impact parameter averages

Consider next the averaging over the transverse impact parameters bi = xi x0 . A
G0 . This involves
typical term that appears to second order in opacity is G0 X1,0 O2,2 X1,0
one direct interaction with center 1 and a double Born with center 2. We consider
separately the average over bi involving direct and virtual interaction centers.
For a scattering center, i, that appears in a direct interaction the impact parameter
average takes the form
* Z
Z
d 2 bi
d 2 qi
(i)
v(0, qi ) eiqi bi (+i)
h iA =
A
(2)2
+
Z
d 2 q0i
0
v (0, q0i ) e+iqi bi
(2)2
Z
d 2 qi d 2 q0i (2)2 2 (qi q0i )
v(0, qi )v(0, q0i )
=
(2)2 (2)2
A
Z
Z
d 2 qi
2
|
v(q
)|
(52)
d 2 q0i 2 (qi q0i ) ,
=
i
A
(2)2
386
where |v|
2 is defined as the normalized distribution of momentum transfers from the
scattering centers. In terms of Eq. (6), it is given by
2eff
1
1 d 2 el
=
,
(53)
el d 2 q
(q2 + 2 )2
where in the Yukawa example the normalization depends on the kinematic bounds through
1
1
1
= 2 2
,
(54)
2
qmax + 2
eff
Rq
2 = 1. In numerical estimates we take q2
and insures that max d 2 q |v(q)|

max = s/4 ' 3E.
For asymptotic energies eff . Note also that unlike v, the barred v is independent of
color factors and thus applies to both quark and gluon rescattering.
If the interaction with center i is a double Born rather than a direct one, then the
only change relative to (52) is that both qi and q0i come with the same (i)2 = 1 rather
than the (i)(+i) = 1 factor and the phase shifts change relative sign. If the double Born
interaction is in the amplitude, as in the example noted above, then the impact parameter
average leads to
* Z
Z
d 2 bi
d 2 qi
(i)
v(0, qi ) eiqi bi (i)
h iA =
A
(2)2
+
Z
d 2 q0i
0 iq0i bi
v(0, qi ) e
(2)2
Z
d 2 qi d 2 q0i (2)2 2 (qi + q0i )
v(0, qi )v(0, q0i )
= (1)
(2)2 (2)2
A
Z
Z
d 2 qi
2
|
v(q
)|
(55)
d 2 qi 2 (qi + q0i ) .
= (1)
i
A
(2)2
The same holds true if the double Born term is in the conjugate amplitude. The results
above is the basis for the last graphical rule of the last section. A contribution to nth opacity
involving m virtual hits produces a factor (1)m (/A )n .
A key to the analytic derivation in the next section is that the transverse impact parameter
averages in Eqs. (52), (55) diagonalize the products of amplitudes and complex conjugate
amplitudes in the qi , q0i variables. The question of which cross section enters above
depends on the details of the color algebra. At this point it is not at all clear since any
particular contribution to a given order in opacity has an extremely complicated color
structure in general [27]. However, as we prove in the next section the color algebra
simplifies drastically in the sum over all possible real and virtual contributions [11,13].
The color triviality of single inclusive distributions proved in the next section via the
reaction operator approach, states that after summing all contributions, each integration
over qi is accompanied simply by a factor (CA C2 (T )/dA ). This implies that = g is the
gluon elastic jet cross section that appears in both Eqs. (52), (55), rather than the jet one
( CR C2 (T )). Recall that
CA
el (R, T )
(56)
g =
CR
2
|v(q)|
387
and hence
CA g = CR .
(57)
Color triviality implies that the opacity expansion is in terms of the gluon mean free path,
i.e., (L/g )n .
4. Recursive method for opacity expansion

The inclusive gluon distribution to O((L/)1 (el /A )1 ) is a sum of 32 direct and
2 4 virtual cut diagrams. If we want to proceed to second order in opacity O((L/)2
(el /A )2 ) there are 72 direct plus 2 86 virtual contributions. It is therefore useful to
develop an recursive procedure for writing the sum of amplitudes in a certain class of
diagrams. We will ignore the jet and elastic scattering parts of the diagrams which leaves
us with simpler amplitudes similar to the ones discussed in Refs. [26,27]. We recall the
definitions of the Hard, GunionBertsch and Cascade terms
(k qi1 qi2 qim )
k
,
C(i1 i2 im ) =
,
2
k
(k qi1 qi2 qim )2
B(i1 i2 im )(j1 j2 in ) = C(i1 i2 jm ) C(j1 j2 jn ) .
Bi = H Ci ,
H=
(58)
4.1. Amplitude iteration

There are two basic iteration steps that one has to consider to construct the inclusive
distribution of gluon. The first one represents the addition of a direct interaction Dn that
changes the color or momentum of the target parton located at zn as illustrated by Fig. 2.
The second iteration step corresponds to a double Born virtual interaction Vn that leaves
both the color and momentum of the target parton unchanged and is illustrated in Fig. 3.
Fig. 2. Diagrammatic representation of the sum of amplitudes generated by ADi = AXi,0 +

AXi,1 + AG1 Xi,0 Gi .
388
Fig. 3. Diagrammatic representation of the

AVi = AOi,0 + AOi,1 + AOi,2 + AG1 Oi,0 Gi .
sum
of
amplitudes,
generated
by
Our goal here is to construct operators D n , Vn that can be used to construct recursively
the new amplitudes. It is important to note that each new diagram produces a difference
of two phase factors with the exception of one special diagram, (ADn )0 AG1 Xn,0 Gn ,
corresponding to the emission of the gluon after zn . The upper limits of emission is z =
in that case, and only the lower bound phase factor, ei0 zn survives on account of the
adiabatic damping factor exp(|z|). Therefore all but one diagram can be divided into
two parts which can then be recombined into more easily interpretable Hard, Gunion
Bertsch and Cascade amplitudes (58) as in [26,27].
The sum of amplitudes in class A can be denoted by
X
A (x, k)Col(c) ,
(59)
A(x, k, c)
where A (x, k) represents the kinematical part, Col(c) stands for the color matrix for the
distinct graphs in this class enumerated by . There is of course considerable freedom in
rearranging the terms of this sum. Since by definition classes are constructed by repeated
D i , Vi it is convenient to enumerate the 3n
operations of either one of three operations, 1,
different classes of diagrams via a tensor notation, Ai1 in , where the indices ij = 0, 1, 2
specify whether there was no, a direct, or a virtual interaction with the target parton at zj .
The sum of amplitudes for class, Ai1 in , can be written explicitly in the form
Ai1 in (x, k, c) =
n
Y

0,im + 1,im D m + 2,im Vm G0 (x, k, c).
(60)
m=1
Here G0 is the color matrix amplitude that corresponds to the initial source or kernel of the
jet and gluon in the limit of no final state interactions.
The inclusive induced probability distribution at order n in the opacity expansion can
then be computed from the following sum of products over the 3n classes that contribute
at that order
Pn (x, k) = A i1 in (c)Ai1 in (c) Tr
2
X
i1 =0
2
X
in =0
389
A i1 in (x, k, c)Ai1 in (x, k, c), (61)
where the unique complementary class that contracts with Ai1 in is defined by
Ai1 in (x, k, c)
n
Y

0,im Vm + 1,im D m + 2,im G0 (x, k, c),
m=1
Ai1 in (x, k, c) G0 (x, k, c)
n
Y
+ 2,im .
0,im Vm + 1,im D m
(62)
m=1
includes products
For example, A0210 = D 3 V2 G0 for A2012 = V4 D 3 V1 G0 . Note that AA
of classes and complementary classes with different powers of the external coupling, gs ,
but every product in the sum is the same order, gs4n+2 . These mixed terms are the unitarity
corrections to the direct A 11 A11 term that contribute to inclusive processes when the
target recoils are not tagged, i.e., measured exclusively. Note that Pn is not positive definite
except for n = 0, and we refer to it as a probability only figuratively.
With this tensor classification and construction, it becomes possible to construct Pn
recursively from lower rank (opacity) classes through the insertion of a reaction operator
as follows
Pn = Ai1 in1 R n Ai1 in1 ,
R n = D n D n + Vn + Vn .
(63)
We emphasize that it is possible to write the inclusive probability in this simple form
only because the transverse impact parameter averages in Eqs. (52), (55) diagonalize the
products of amplitudes and complementary conjugate amplitudes in the qi , q0i variables.
If the transverse area of the target were not large in comparison to the cross section, then
off diagonal components in those variables would appear that would have to folded with
a suitable transverse Glauber profile functions as discussed in [6]. The simple algebraic
structure of the problem in Eq. (63) therefore hinges on the validity of the transverse
averaging via Eqs. (52), (55).
To construct D n , consider first the contribution AXn,0 from Eq. (40). In this case, a
new direct interaction is attached to the jet line at zn . For the partial sum of amplitudes in
Ai1 in1 corresponding to emitting the gluon before the last real or virtual interaction at
zf 6 zn1 , the direct interaction of the jet at zn simply multiplies that partial sum by an .
No extra phases are introduced since the energy of the jet is assumed to be very high. As
noted before, for every class, there is a term in Ai1 in1 corresponding to a graph where
the gluon is emitted after the last interaction point zf . That term has the explicit form

1 Nv (Ai1 in1 )
H(ei0 zf ) c Tel (Ai1 in1 ),
(64)
2
where the elastic color matrices for that class and its complementary are denoted by
Tel (Ai1 in1 ) (an1 )in1 (a1 )i1 ,

Tel A i1 in1 (a1 )2i1 (an1 )2in1 .
(65)
390
For example, Tel (A1021 = D 4 V3 D 1 G0 ) = a4 (a3 a3 )a1 = CR a4 a1 . Eq. (65) implies that the
following identity holds for elastic color factors

(66)
Tel A i1 in1 Tel (Ai1 in1 ) = CRn1
valid for all 3n1 components of rank n 1 classes.
The factor ( 12 )Nv in Eq. (64) arises because every virtual contact interaction in the class
acquires a factor 12 from the the contact limit of the contour integration over longitudinal
momentum (see Section 2 and the Appendix). The numbers, Nv , N v , of such contact
interactions in class, Ai1 in1 , and its complementary class, Ai1 in1 , are
Nv = Nv (Ai1 in1 ) =
n1
X
2,im ,
m=1
n1
X
N v = Nv Ai1 in1 =
0,im .
(67)
m=1
For example Nv = 2, N v = 2 for A01202 = V5 V3 D 2 G0 . A useful bookkeeping identity that

we will need is
m
X 1 Nv (A i1 im ) 1 Nv (Ai1 im ) 1 1
= +1
= 0.
(68)
2
2
2 2
i1 ,...,im
Now consider how a jet interaction at zn changes the graph Eq. (64). Besides being
multiplied by an , the only other change is that the phase factor changes into a difference of
phase factors as
Xn,0
ei0 zf ei0 zn ei0 zf .
(69)
The part, ei0 zf , on the right hand side can be added back to the sum of all the other
graphs with emission before zf . The modified sum an Ai1 in1 therefore includes this part
of Eq. (69). In summary, the component of the D n operator that takes into account the jet
rescattering at zn is simply multiplication by the color matrix, an , in the jet representation.
This then identifies the first part of the D n operator as
n + .
D n = 1a
(70)
The left over part ei0 zn of Eq. (69) can be combined with the graph AG1 Xn,0 Gn (e.g.,
M1,1,0 , M2,2,0 in Appendix A, Appendix C), that accounts for gluon emission after zn . This
combination leads to the following extra term in Ai1 in

1 Nv (Ai1 in1 ) i0 zn

He
[c, an ]Tel (Ai1 in1 ).
(71)
2
The sum of graphs A(Xn,0 + G1 Xn,0 Gn ) is therefore an Ai1 in plus Eq. (71).
To finish constructing D n , we need to consider the effect of an elastic scattering of the
gluon produced at an earlier point in the medium. The gluon cascade interaction graph
AXn,1 (see, e.g., M2,0,3 in Appendix B) introduces the following changes in the amplitude:
first an extra phase shift exp[i(0 n )zn ] arises due to the difference of gluon energies
before and after the transverse momentum exchange, qn , at zn . This is in contrast to the jet
rescattering where that energy change was negligible. Second, the gluon color is rotated by
391
c [c, an ]. Finally, the transverse momentum in A(x, k, c) is shifted backward by k

k qn so that the same final state k is reached. This part of the D n operator, therefore,
simply transforms

Xn,1
(72)
Ai1 in1 (k, c) Sn Ai1 in1 = ei(0 n )zn Ai1 in1 k qn , [c, an ] .
The shift or gluon scattering operator that implements gluon rescattering at zn is thus
defined by
Sn = if can d ei(0 n )zn eiqn b .
Here, b i
(73)
is the impact parameter operator conjugate to k such that
eiqb f (k) = f (k q).

This transverse momentum shift together with a class independent phase shift and a color
rotation (via the structure constants f abc ) enlarges Ai1 in1 to include all but one of the
gluon final state interaction diagrams with an interaction at zn .
The left over graph not included in (72) is the one where the jet scatters elastically up to
previous last interaction point zf , emits a gluon with transverse momentum kqn between
zf and zn , and the gluon scatters at zn (see, e.g., M1,0,1 in Appendix A). The integral over
the emission interval leaves a characteristic differences of phases ein zn ein zf , while
the elastic scattering introduces a phase shift factor ei(0 n )zn as well. The kinematic part
of this cascade amplitude is Cn . The part proportional to ein zn contributes therefore the
following extra term

1 Nv
Cn ei(0 1 )zn ei1 zn [c, an ] Tel (Ai1 in1 ),
(74)
+
2
where again Nv is the number of contact interactions before zn . Recalling that Bn = HCn
has the physical interpretation of a GunionBertsch gluon source term at zn , the two extra
terms from Eq. (74) and Eq. (71) can be conveniently combined to form

1 Nv (Ai1 in1 ) can d
if
Bn ei0 zn dTel (Ai1 in1 )
Bn Ai1 in1 (c)
2
=
n1
Y
m=1
1
2
2,im
n1
Bn ei0 zn [c, an ]an1
a1i1 .
(75)
Combination of three steps Xn,0 + Xn,1 + G1 Xn,0 G can therefore be summarized by

the follow operation on rank n 1 class elements

D n Ai1 in1 (x, k, c) an + Sn + B n Ai1 in1 (x, k, c)

= an Ai1 in1 (x, k, c) + ei(0 n )zn Ai1 in1 x, k qi , [c, an ]

1 Nv (Ai1 in1 )
Bn ei0 zn [c, an ] Tel (Ai1 in1 ).
(76)

2
We note that the common factor to all diagrams 2igs ( ) is suppressed for clarity.
392
We now turn to the second class of diagrams, Vn Ai1 in1 , created by inserting contact
interaction at zn . This new class is constructed via Eq. (41) and illustrated in Fig. 3.
For a double Born interaction at zn , the transverse coordinate integration can be
performed before multiplying by a complementary hermitian conjugate amplitudes since
the coordinates of a virtual interaction cannot occur in the complementary class. The
transverse coordinate integration on the center of the i th interaction enforces (see Eq. (19))
that the net momentum transfered in the two legs vanishes via a factor 2 (qi + q0i ).
Consider first the sum of graphs where the two legs of the contact interaction are attached
c
in Appendix C). The situation is similar to Eq. (70)
to the jet line (AOn,0 ) (see, e.g., M2,0,0
except that two color factors an an = CR multiply the rank n 1 amplitudes instead of one.
In addition a factor of 12 appears due to the contact limit of the contour integration. Thus,
this part of the virtual interaction simply transforms
CR
Ai1 in1 .
(77)
2
c
in
The second term, AOn,1 , in Eq. (41) is the contact with the gluon at zn (see M2,0,3
ca
d
da
e
Appendix B). Since if n if n = CA c,e , the color structure of this term is also trivial.
In addition, since the total momentum transfer is now zero, no extra phase factor arises and
the gluon momentum in Ai1 in1 remains unchanged. The contact gluon interaction at zn
therefore contributes
On,1
CA
Ai1 in1 .
(78)
Ai1 in1
2
c
in Appendix D) is the most subtle.
The mixed contact interaction AOi,2 (see, e.g., M2,0,1
Even though the net momentum transfer vanishes, in this case both the jet and gluons
receive (opposite) non-vanishing transverse momentum transfers, qn . This leads to a
highly non-local modification of Ai1 in1 . The contact limit of the longitudinal momentum
transfer contour integration produces in this case a factor of 1 as shown in the Appendix.
The elastic scattering of the gluon produces multiplicative phase factor, ei(0 n )zn . The
jet elastic scattering introduces no new phase in the small x limit as before. The transverse
momentum exchange, to the gluon requires that k is again shifted back by qn as in Eq. (72).
The gluon color is also rotated as in (72). Furthermore, the scattering of the jet introduces
another color matrix an . This mixed contact interaction therefore simply produces the
following variant of (72)

On,2
(79)
Ai1 in1 (k, c) an Sn Ai1 in1 = ei(0 n )zn an Ai1 in1 k qn , [c, an ] ,
On,0
Ai1 in1
with again one extra term that can be recombined with the amplitude of gluon radiation
after the contact interaction at zn to form a virtual GunionBertsch source

1 Nv (Ai1 in1 ) CA
Bn ei0 zn c Tel (Ai1 in1 ),
(80)
an B n Ai1 in1 =
2
2
where we used an [c, an ] = 12 CA c to simplify the color algebra.
The double Born interaction at zn can therefore be implemented by the following
operator
1
1
Vn = (CA + CR ) an Sn an B n = an D n (CA CR ).
2
2
393
(81)
Specifically, the virtual iteration of a rank n 1 class gives

CR + CA
Ai1 in1 (x, k, c)
Vn Ai1 in1 (x, k, c) =
2

ei(0 n )zn an Ai1 in1 x, k qn , [c, an ]

1 Nv CA
in1
Bn ei0 zn c an1
a1i1 .

2
2
(82)
We illustrate the amplitude iteration technique Eqs. (76), (82) in Appendix F.

4.2. Reaction operator recursion to all orders in opacity
We emphasize that this recursive process is completely general and applies for both the
hard emission kernel Ker(H) Eq. (39) and the asymptotic GunionBertsch Ker(GB) Eq. (38).
The key to unraveling the interplay between direct and virtual interaction is the operator
identity Eq. (81) between D n and Vn . This makes it possible to relate the nth order in
opacity gluon emission probability distribution Pn to the probability at (n 1)th order
by expressing the reaction operator R n in Eq. (63) as

(83)
R n = D n an D n an CA = Sn + B n Sn + B n CA .
A further major simplification occurs because both S and B involve the same gluon color
rotation through if can d . We show next how this unravels the color algebra and reduces
it to simply powers of CA and CR . We therefore prove color triviality of the reaction
operator, a property that is implicit in in the work of Refs. [11,13] and which is proved in
a somewhat more involved way via path integral techniques in [34] for quark jets only.
First, we note that the shift operator has the property that
Sn Sn = CA eqn k eqn k .
(84)
Therefore, S S is diagonal in color space. This allows us to compute the effect of an extra
gluon cascade interactions through simple transverse momentum shifts
Ai1 in1 (Sn Sn CA )Ai1 in1

= CA Ai1 in1 (k qn , c)Ai1 in1 (k qn , c)

A i1 in1 (k, c)Ai1 in1 (k, c)

= CA Pn1 (k qn ) Pn1 (k)
= CA eiqn b 1 Pn1 (k).
(85)
Next, consider the modulus square of the extra GunionBertsch source for two or more
scatterings of the jet + gluon system (n > 1)
394
Ai1 in1 B n B n Ai1 in1

X 1 N v 1 Nv
= CA |Bn |2
Tel (Ai1 in1 ) cc Tel (Ai1 in1 )
2
2
i1 ,...,in1
= CA CRn |Bn |2
n1
Y
i1 ,...,in1 m=1
2,im

1 0,im 1,i
(1) m = 0,
(86)
where we used Eq. (66) and the bookkeeping identity (68). Therefore, the virtual
corrections cancel the direct extra GunionBertsch source term for n > 1. For the special
case n = 1, however, this diagonal term survives
G0 B 1 B 1 G0 = CA CR |B1 |2 .
(87)
Finally, we have to calculate the interference between the cascade and GunionBertsch
source terms in Eq. (83). The two gluon color factor f can d in Eqs. (73), (75) again contract
to form a color diagonal factor CA , leaving
2Re Ai1 in1 B n Sn Ai1 in1 = 2CA Bn Re ein zn eiqn b In1 ,
(88)
where the In1 term in the iteration step is easily seen to be

In1
i i
X 1 Nv (A 1 n1 )

Tel Ai1 in1 cAi1 in1
i1 in1
i i

2
X 1 Nv (A 1 n2 )
X
1 0,in1 2in1
i1 in2
=
Tel A
an1
2
2
i1 in2
in1 =0

c 0,in1 + 1,in1 D n1 + 2,in1 Vn1 Ai1 in2
X 1 Nv (A i1 in2 )

=
Tel Ai1 in2
2
i1 in2

12 an1 an1 c + an1 cD n1 + cVn1 Ai1 in2 .
(89)
This contribution builds up on the class of diagrams Ai1 in2 under the action of an
iteration operator which can be reduced using Eqs. (73), (81) into the form

CA
1
c = [an , c] Sn + B n CA c
an an c + an cD n + cVn = [an , c]D n
2
2

i(0 n )zn iqn b
e
1 c + [an , c]B n .
= CA e
(90)
Next, we note that the contributions from B actually cancel exactly for n > 2 in the sum
over all classes due to the bookkeeping identity
i i
X 1 Nv (A 1 n2 )

Tel A i1 in2 [an1 , c]B n1 Ai1 in2
i1 in2
= CA CR Bn1 ei0 zn1
395
X 1 Nv (A i1 in2 ) 1 Nv (Ai1 in2 )
2
2
i1 in2
Tel

A i1 in2 Tel (Ai1 in2 ) = 0.
(91)
For the special case with n = 2 we can evaluate I1 directly from the definition Eq. (89) and
the initial hard amplitude to obtain
I1 = CA ei(0 1 )z1 eiq1 b 1 c G0 + [a1 , c]B 1 G0

= CA CR ei(0 1 )z1 eiq1 b 1 Hei0 z0 + B1 ei0 z1

= CA CR ei(0 1 )z1 eiq1 b 1 H ei0 z0 ei0 z1 .
(92)
In the last line the GunionBertsch amplitude was rewritten with the help of the shift
operator Eq. (73) as being derived from the hard vertex kernel
(93)
B1 ei0 z1 = ei(0 1 )z1 eiq1 b 1 Hei0 z1 .
Eqs. (89), (90), (92) therefore imply that In obeys the recursion relation
In = CA ei(0 n )zn eiqn b 1 In1 n,1 CA CR B1 ei0 z1 .

With I0 = CR Hei0 z0 and Eq. (93) for n > 1 we solve (94) in a closed form
#
" n
Y

ei(0 m )zm eiqm b 1 H ei0 z1 ei0 z0 ,

In = CR CAn
(94)
(95)
m=1
where the product is understood as ordered for left to right in decreasing order in operators
labeled by m.
Combining Eqs. (85)(88), (95) that specify how the reaction operator acts between the
contraction of rank n 1 classes, we find that the inclusive radiation probability, Eq. (63),
obeys a simple recursion relation:

Pn (k) = CA Pn1 (k qn ) Pn1 (k) 2CA Bn Re ein zn eiqn b In1

+ n,1 CA CR |B1 |2 .
(96)
We can solve Eq. (96) with the initial condition

P0 = CR H2 .
(97)
We introduce the following notation for the separation of the scattering centers 1zn
zn zn1 . For n = 1 the solution is

P1 = CA CR C21 H2 + B21 + 2B1 C1 cos(1 1z1 )

(98)
= 2CA CR B1 C1 1 cos(1 1z1 ) .
For n = 2 we need
396

Re ei2 z2 I1 (k q2 ) = CA CR Re ei2 z2 C(12) ei(2 z1 (12) 1z1 ) ei2 z1

C2 ei2 z0 ei2 z1

= CA CR C(12) cos(2 1z2 + (12) 1z1 ) cos(2 1z2 )

+ C2 cos(2 (1z1 + 1z2 )) cos(2 1z2 ) . (99)
Therefore, the n = 2 distribution is given by

P2 = 2CA2 CR B1 C1 1 cos(1 1z1 ) B2(12) C(12) 1 cos((12)1z1 )

+ B2 C(12) cos(2 1z2 + (12) 1z1 ) cos(2 1z2 )

(100)
B2 C2 cos(2 (1z1 + 1z2 )) cos(2 1z2 ) .
For n > 1, we can use (95) to obtain the general solution for the gluon probability at nth
order in opacity
Pn (k) = CA eiqn b 1 Pn1 (k)

2CR CAn Bn
Re e
in zn iqn b
" n1
Y
i(0 m )zm iqm b
#
m=1

H ei0 z1 ei0 z0
#
" n
n
X
Y
n
iqj b
1 Bi eiqi b ei0 zi
e
= 2CR CA Re
i=1
j =i+1
" i1
Y
i(0 m )zm iqm b
#

H ei0 z1 ei0 z0 ,
(101)
m=1
where the last applies to n = 1 as well. This is therefore the complete solution to the
problem and is the central result of this paper. Eq. (101) provides not only an algebraic
proof of color triviality of the inclusive gluon distribution (Pn CR CAn ), but in fact gives
the complete angular dependence for arbitrary zi and qn . It also applies to both quark
and gluon jets. The derivation through the reaction operator and recursion relations is
furthermore much simpler and transparent than previous derivations [11,13,34]. The result
is also more general and versatile for applications to nuclear collisions.
In this form, it is for example possible to implement Pn directly as an after-burner in
a Monte Carlo event generator, e.g., ZPC [35,36] or MPC [37,38]. Such programs solve
a relativistic (3 + 1)D transport theory and provide realistic distributions of target parton
coordinates zi along the path of a high energy jet. Eq. (101) also makes it possible to
consider evolving potentials. For example, the screening scale that characterizes the
mean qn transfered at zn , could itself depend on n because of the variation of the local
density at that point. In this way, jet quenching in realistic, evolving nuclear geometries
can be calculated more accurately.
397
4.3. Ensemble average over momentum transfers

We now consider the integrated gluon distribution over the qm via the normalized
squared scattering potential Eq. (53)
Z Y
n

2
2
(qm ) Pn (k; q1 qn ),
(102)
d qm vm
hPn (k)iv =
m=1
where we allow for the possibility that the effective potential could change along the path
of the jet. We hold the zm fixed as yet. From Eq. (96) we need to consider for example
Z

d 2 qn vn2 (qn ) Pn1 (k qn ) Pn1 (k)
Z

(103)
= d 2 qn vn2 (qn ) 2 (qn ) Pn1 (k qn ).
Since 2 (qn )Bn 0, we can also replace vn2 (qn ) by vn2 (qn ) 2 (qn ) in the average over
the inhomogeneous term. However, due to the special form of the operators defining In in
Eq. (95) we can also rewrite the integral over qi for i = 1, . . . , n 1 in analogous form
Z Y
n1
n

2
Y i(0 m )zm iqm b
e
1
d qj vj2 (qj ) Bn ein zn eiqn b
e
j =1
Z Y
n
m=1
n
Y

ei(0 m )zm eiqm b ,

d 2 qj vj2 (qj ) 2 (qj ) Bn ei0 zn
j =1
(104)
m=1
where the product of operators is again path ordered. This product of operators can be
simplified as follows
ei0 zn
n
Y
ei(0 j )zj eiqj b = ein zn ei(n (n1,n) )zn1 ei((2n) (1n) )z1 eiQb
j =1
"
= exp i
n
X
(j n) (zj zj 1 ) ei(1n) z0 eiQb ,
j =1
where Q =
(105)
qm is the total momentum transfer, and
(k qm qn )2
.
2xE
Therefore, we can evaluate
(m,...,n) =
ei0 zn
n
Y
ei(0 j )zj eiqj b H ei0 z1 ei0 z0
(106)
j =1

= C(1n) ein,n ei(1n) (z1 z0 ) 1 ,
(107)
gluon elastic scattering phase shift from z0 to zn . The partial (eikonal)

where n,n is
R zthe
n
dz [kz (z) xE] ) due to gluon rescattering from zm1 to zn is given by
phase shift ( zm1
398
n,m =
m
X
(kn) (zk zk1 ) =
k=1
m
X
(kn) 1zk .
(108)
k=1
Eq. (107) makes it possible to write the recursion relation for the momentum transfer
averaged probability for n > 1 in the form
Z

d 2 qn vn2 (qn ) 2 (qn ) Pn1 (k qn ) v
hPn iv = CA
2CR CAn
Z Y
n

d 2 qi vi2 (qi ) 2 (qi ) Bn C(1n) Re ein,n
i=1

ei(1n) (z1 z0 ) 1 .
(109)
With P1 given by Eq. (98), the solution for n > 0 is

Z Y
n

d 2 qi vi2 (qi ) 2 (qi )
hPn iv = 2CR CAn
i=1
n
X

B(m+1,...,n)(mn) C(1n) Re ein,m ei(1n) (z1 z0 ) 1 ,
(110)
m=1
where for m = n, B(n+1n)(n) B(0)(n) B(n) .

Recall that we cannot take the contact limit z1 = z0 , which apparently vanishes, because
our derivation assumed that zm zm1 were larger than the range of the force 1/.
However, it is clear from Eq. (110), that it is the first step phase difference
ei(1n) (z1 z0 ) 1
(111)
that controls the bulk of the LPM destructive interference effect that suppresses radiation
when the formation length 1/(1n) is long compared to (z1 z0 ). The subsequent
interactions merely modulate this effect with an elastic scattering phase shift.
Finally, we can restore the proportionality constants by multiplying with s / for the
Q
production vertex, 1/ for the d 2 k measure, and j (g (j )/A ) along the path to convert
the vj back into vj from (53). In addition, we must multiply by the combinatorial factor,
Nn
N!
,
(112)
n!(N n)!
n!
that counts the number of ways n target partons out of N can be within the interaction range
of the jet + gluon system. Including these factors, the general formula for the induced gluon
number distribution can finally be written as

n Z Y

n

g (1) 2
CR s 1
L
dN (n)
2
2
=
v
q
(q
)
(q
)
d
x
i
i
i
i
dx d 2 k
2 n! g (1)
g (i)
i=1
2 C(1,...,n)
"
cos
n
X
m=1
m
X
k=2
B(m+1,...,n)(m,...,n)
!
(k,...,n) 1zk cos
m
X
k=1
!#!
(k,...,n) 1zk
(113)
399
P
where 12 0 is understood. We emphasize that Eq. (113) is not restricted to uncorrelated
geometries as in [11,34]. Also it allows the inclusion of finite kinematic boundaries on the
qi as well as different functional forms and elastic cross sections g (i) along the eikonal
path.
For particular models of the target geometry one can proceed further analytically. For a
sharp rectangular geometry, the average over the longitudinal target profile with
n!
(L zn ) (zn zn1 ) (z2 z1 )
(114)
Ln
leads to an oscillatory pattern [34] that is of course artifact of the assumed sharp edges.
A somewhat more realistic model [27,28] utilizes normalized exponential longitudinal
distributions of scattering center separations
(z
1 , . . . , zn ) =
(z
1 , . . . , zn ) =
n
Y
(1zj ) 1zj /Le (n)
e
.
Le (n)
(115)
j =1
This converts the oscillating formation physics factors in Eq. (113) into simple Lorentzian
factors
!
Z
m
m
X
Y
1
.
(116)
(k,...,n) 1zk = Re
d cos
1 + i(k,...,n) Le (n)
k=j
k=j
In order to fix Le (n), we require that hzk z0 i = kL/(n + 1) for both geometries (114),
(115). This constrains Le (n) = L/(n + 1) [27,28].
Note finally that the ratio of the medium induced to the medium independent
factorization gluon distributions vanish at both small and large |k|
lim
|k|=0 and
dN (n)
= 0.
dN (0)
(117)
As |k| 0, dN (0) R while all but one term in dN (n) are finite. The potentially
P
singular term (1/k2 )
j k qj , however, vanishes due to azimuthal integrations. In
the |k| limit, on the other hand, all GunionBertsch currents B(),() O(1/k4 )
and hence vanish faster than 1/k2 from dN 0 . This observation partly justifies neglecting
the kinematical |k| boundaries (120) in analytical approximations.
5. Numerical estimates
5.1. Angular distributions
Fig. 4(a)(c) illustrate Eq. (113) for an equivalent exponential geometry (115) to a box
of thickness L = 5 fm and g = 1 fm including kinematical constraints appropriate for
a 50 GeV jet. We used adaptive Monte-Carlo integration [39,40] to integrate over the
momentum transfers q1 qn in the exponential geometry (115).
The medium induced gluon differential distributions up to third order in opacity are
plotted divided by the zeroth order in opacity, hard distribution Eq. (119) for radiated
400
gluons with light-cone momentum fractions x = 0.05, 0.2, 0.5. Note that dN (ind) /dN (0)
tends to vanish at both small and large |k| in accordance with Eq. (117). As x increases the
relative magnitude of the medium induced contribution decrease and the overall magnitude
is set already by the first order in opacity result. Higher orders redistribute the moderate
k2 /2 10 toward higher values due to elastic rescattering of the gluon, but they also
tend to fill in the low |k| region with additional soft radiation. The overall angular pattern
appears somewhat complex, but one must keep in mind that the variation is rather slow
since the logarithmic scale varies over several orders of magnitude.
Fig. 4(d) shows the actual modest size of the induced radiation contribution on top of
the hard 1/k2 distribution for the case of x = 0.05.
5.2. Intensity distribution and energy loss to first order in opacity. Analytical approach
5.2.1. Zero opacity limit
The jet distribution in the absence of final state interactions is given by (see [27])
d 3 NJ = (0) (Ep)

2
E
d 3 pE
d 3p
E )
= dR J (|Ep|, p
.
3
2|Ep|(2)
2|Ep|(2)3
(118)
In the Leading Pole Approximation (LPA) approximation [33], the radiation distribution
accompanying the such hard processes for a spin 12 jet is given by

x2 1
CR s
dN (0)
1x +
,
(119)
x
2 k2
dx dk2
where in the eikonal approximation the light-cone momentum fraction x = k + /E + /E
and k is assumed to be small compared to . For other spin jets, the corresponding splitting
function replaces the x dependence above.
We consider radiation outside a cone with |k| > . and with the upper |k| bound
determined from the three body (jet + jet + gluon) kinematics. The gluon kinematic
boundaries are therefore

k2max = min 4E 2 x 2 , 4E 2 x(1 x) .
(120)
k2min = 2 ,
The radiation intensity integrated over this range of k is

x2
|k|max
dI (0) 2CR s
=
1x +
E log
,
dx
2
|k|min
(121)
where |k|max and |k|min are given by Eq. (120). Note that the differential intensity is
roughly uniform with the exception of the kinematical edges x 0 and x 1. In the
Leading Log singularity of the Leading Pole Approximation the radiative total energy loss
of a quark jet originating from a hard vertex outside a cone defined by Eq. (120) is given
by
1E (0) =
E
4CR s
E log .
3
(122)
While this overestimates the radiative energy loss in the vacuum (self-quenching), it is
important to note that 1E (0) /E 50% is typically rather large. As shown below, the
401
(a)
(b)
Fig. 4. (a)(c) The medium-induced double differential gluon distributions are plotted vs. k2 /2 up
to first (dN (1) ), second (dN (1+2) ) and third (dN (1+2+3) ) order in the opacity (L/g ) expansion,
divided by the medium independent hard radiation (dN (0) ) 1/k2 . Curves are obtained numerically
for exponential geometry with L/g = 5, Ejet = 50 GeV and = 0.5 GeV. The first three figures
(a)(c) are for typical soft, semi-soft and hard gluons respectively, x = 0.05, 0.2, 0.5. (d) The full
gluon differential distribution up to third order in opacity dN (0) + dN (1) + dN (2) + dN (3) is shown
for x = 0.05 for L = 0, 3, 6 fm.
402
(c)
(d)
Fig. 4. Continued.
medium induced energy loss is small by comparison. For a gluon jet one has to replace
the quark splitting function q qg by the gluon one g gg in Eq. (121) and use CA
Nc 2CF .
The total energy loss should reduce to the medium independent one in the limit of
vanishing opacity,
1E (tot)
= 1,
L0 1E (0)
lim
(123)
403
and for jets of asymptotically high energies due to factorization

1E (tot)
= 1.
E 1E (0)
(124)
lim
5.2.2. First order in opacity correction

The first order (n = 1) in opacity contribution, dI (1) /dx, to the induced radiation
intensity can be read of from Eq. (113). The longitudinal coordinate average over the
equivalent exponential target profile (115) is done with Le (2) = L/2. Including the q
qg splitting function as in (119) we have

CR s
x2 L
dI (1)
=
E
1x+
dx
2 g
2
k
Zmax
dk2
q21 max
d 2 q1
k2
0
k2min
2 k q1 (k q1 )2 L2
.
+ 2 )2 16x 2E 2 + (k q1 )4 L2
2eff
(q21
(125)
To obtain a simple analytic result, we ignore the kinematic boundaries and set |k|min = 0,
|k|max = that is motivated by Eq. (117). We also set |q1 |max = (i.e., 2eff = 2 ). This
allows us to change variables q0 k q1 in Eq. (125)

CR s
x 2 2 L
dI (1)
=
E
1x+
dx
2
g
Z
q0 2 L2
dq0 2
16x 2E 2 + q0 4 L2
0
dk2
k2
Z2
0
2k (k + q0 )
d
2 (q0 2 + k2 + 2|q0 | |k| cos() + 2 )2
(126)
and express the integrand in the azimuthal integral as a partial derivative with respect
to k2
Z2
2k k2
2
1
d
2 (q0 2 + k2 + 2|q0 ||k| cos() + 2 )
1
.
= 2k2 k2 p
2
0
2
((k + q + 2 )2 4k2 q0 2 )
The remaining q0 integral

Z
CR s
x2 L
22
q0 2 L2
dI (1)
=
1x+
E dq0 2 0 2
dx
2 g
q + 2 16x 2 E 2 + q0 4 L2
(127)
can be performed then analytically, resulting in

x2
L
dI (1) CR s
=
1x +
E f ( ),
dx
2
g
L2
,
4xE
(128)
404
where is a measure of the formation probability. The formation function f ( ) is given

by

( + 2 log )
if 1
.
(129)
f ( ) =
2 log if 1
(1 + 2 )
It is the 1 limit of the formation function Eq. (129) in which the the first
order in opacity medium-induced intensity distribution reduces to a simple form with a
characteristic quadratic dependence on L
dI (1) CR s 1 x +
dx
4
x
x2
2
L2 2
.
g
(130)
This formula breaks down at both x 0 and x 1 because |k|max and |q1 |max cannot be
approximated by and because the small x approximations used above break down as
x 1.
The induced radiative energy loss to first order in opacity in the framework of the above
approximations is then given by
1E (1) =
CR s L2 2
E
log .
4
g
(131)
This equation is similar to the one derived in Ref. [11], but the log-enhancement factor in
the BDMPS and Zakharov approaches is due to small impact parameters (Coulomb nature
of the scattering potential at small distances), while in Eq. (131) it comes from an entirely
different source the broad logarithmic integration over the gluon energies.
5.3. Induced radiation intensity to higher orders in opacity. Numerical results
Whereas Eq. (131) displays analytically the main qualitative features of non-abelian
energy loss, in practice at the finite energies available kinematical bounds do affect
quantitatively the results. This naturally leads to a reduction relative to the analytic
estimate.
In Fig. 5(a) the induced intensity distribution dI (ind) /dx (sets of dashed curves) is
compared to the medium independent dI (0) . We consider the example of a 50 GeV quark
jet ( = 0.5 GeV) at RHIC. The hard radiation intensity result is roughly constant with x,
whereas for most of the x region the induced radiation intensity falls like 1/x , 1. We
note that for relatively thin plasmas L/g 6 3 the medium induced energy loss remains
small compared to dI (0) .
It is surprising how much the first order result dominates the induced intensity
distribution [28]. Fig. 5(b) illustrates how rapidly the contributions from higher orders
in opacity decrease. While the differential angular distribution continue to show some
weak sensitivity to higher order contributions in opacity, the angle integrated intensity
is much less affected beyond first order. The third order contribution is almost buried
in the numerical noise. For opacities L/g ' 6 some more pronounced probability
redistribution can be seen in the double differential level Fig. 4. However, when integrated
405
(a)
(b)
Fig. 5. The radiation intensity distribution is plotted vs. the light-cone momentum fraction x of the
gluon. We consider a 50 GeV quark jet in a plasma with screening scale = 0.5 GeV and g = 1 fm.
The solid curve shows the dominant medium-independent radiation intensity. The medium-induced
gluon spectrum is plotted for up to third order in opacity (dI (1) , dI (1+2) and dI (1+2+3) ) for
opacities L/g = 1, 2, 3. (b) The absolute value of the orders in opacity dI (1) , dI (2) and dI (3)
that contribute in part (a) are plotted for the same energy and opacity L/g = 3.
406
Fig. 6. The radiated energy loss of a quark jet with energy Ejet = 5, 50, 500 GeV (at SPS, RHIC,
LHC) is plotted as a function of the opacity L/g (g = 1 fm, = 0.5 GeV). Solid curves show the
first order in opacity results. The dashed curves show results up to second order in opacity, and two
third order results are shown by solid triangles for SPS energies.
over k2 and then over xmin < x < xmax to obtain the induced non-abelian energy loss
1E (ind), the first order result dominates very strongly, as seen in Fig. 6.
The induced energy loss shown in Fig. 6 is for quark jets with energies Ejet =
5, 50, 500 GeV typical for SPS, RHIC and LHC. Higher orders in opacity (L/g )n , n > 2
have little effect except at very low SPS energies. The kinematic bounds at SPS suppress
very much the energy loss in comparison to RHIC and LHC energies. An analysis of the
slopes as a function of the opacity, L/g , shows that 1E (ind) L20.1 at all energies even
with finite kinematic boundaries included. As a measure of the deviation of the simple first
order analytic estimate, we generalize Eq. (131) as follows,
1E (ind) =
CR s L2 2
E
log .
N(E) g
(132)
If the kinematic bounds are ignored as in Eq. (130), then N(E) = 4. Including finite
kinematic constraints cause N(E) to deviate considerably from this asymptotic value. We
find numerically that N(E) = 7.3, 10.1, 24.4 for E = 500, 50, 5 GeV. Together with the
logarithmic dependence of energy, these kinematic effects suppress greatly the energy
loss at lower (SPS) energies as seen in Fig. 6. This is in contrast to the asymptotic,
energy independent result, quoted in Ref. [11], where the finite kinematic bounds were
neglected. Our numerical results, however, agree with Ref. [11] near LHC energies as
reported in [28].
407
Acknowledgements
We thank A.H. Mueller and U. Wiedemann for numerous clarifying discussions. This
work was supported by the DOE Research Grant under Contract No. De-FG-02-93ER40764, partly by the US-Hungarian Joint Fund No. 652 and OTKA No. T032796.
Appendix A. Diagrams M1,0,0 = G0 X1,0 , M1,1,0 = G0 G1 X1,0 G1 and
M1,0,1 = G0 X1,1
To make contact with the results in Refs. [26,27] we present explicit calculation of the
first nontrivial diagrams shown in Fig. 7. As a first application, consider the one rescattering
amplitude M1,0,1 , in the notation of Ref. [27].
Z
d 4 q1
iJ (p + k q1 )ei(p+kq1)x0 1 (p, k, q1 )V (q1 )eiq1 x1
M1,0,1 =
(2)4
i(p + k q1 )(i)(k q1 )
Z 2
d q1 iq1 (x1 x0 )
i(p+k)x0
[c, a1 ]Ta1 (i)
e
2gs (k q1 )
J (p + k)e
(2)2
Z
dq1z
v(q1z , q1 )(p + k q1 )(k q1 )eiq1z (z1 z0 ) .
(A.1)
E+
2
The longitudinal momentum transfer integral
Z
dq1z
v(q1z , q1 )(p + k q1 )
I1 (p, k, q1 , z1 z0 )
2
(k q1 )eiq1z (z1 z0 )
(A.2)
can be performed via closing the contour below the real axis since z1 > z0 . In addition to
the potential singularities at i1 (2i 2i = q2i + 2 ), the two propagators have four
poles in the q1z plane, which up to leading correction are located at
q1 = E + + i,
q2 = 0 i,
q3 = k + 1 + i,
q4 = 0 + 1 i.
(A.3)
Fig. 7. Three direct terms M1,0,0 = G0 X1,0 , M1,1,0 = G0 G1 X1,0 G1 and M1,0,1 = G0 X1,1
contribute to the soft gluon radiation amplitude to first order in opacity L/ el /A .
408
The residues give

ei0 (z1 z0 )
,
E + k + 1
ei(0 1 )(z1 z0 )
,
Res(q4 ) v(1 0 , q1 )
E + k + 1
Res(q2 ) v(0 , q1 )
(A.4)
while
Res(i1 )
4s e1 (z1 z0 )
,
(2i1)E + k + (i1 )2
(A.5)
where we assumed that k + 1 i . In the well-separated case where (z1 z0 ) =

1, this potential residue is exponentially suppressed and, therefore,

ei0 (z1 z0 )
v(0 , q1 ) v(1 0 , q1 )ei1 (z1 z0 )
+
+
E k 1

ei0 (z1 z0 )
1 ei1 (z1 z0 ) .
(A.6)
iv(0, q1 ) + +
E k 1
I1 (p, k, q1 , z1 z0 ) i
We thus recover in this eikonal limit the time ordered perturbation result in [26,27]
Z 2
d q1
v(0, q1 )eiq1 b1
M1,0,1 = J (p)eipx0 (i)
(2)2

(k q1 ) i(0 1 )z1 i1 z1
2igs
e
ei1 z0 [c, a1]Ta1 ,
(A.7)
e
2
(k q1 )
where b1 = x1 x0 . Similarly, we recover the other two amplitudes for one scattering
center, e.g.,
Z 2
d q1
v(0, q1 ) eiq1 b1
M1,1,0 = J (p)eipx0 (i)
(2)2
k
(2igs ) 2 ei0 z1 ca1 Ta1 .
(A.8)
k
Note that while the direct hM1,m,l M1,m

0 ,l 0 i ensemble averages lead to non-vanishing
modifications to the initial hard spectrum, |M0 |2 |J (p)|2 /k2 , the interference between
M0 and M1,l,m vanishes because the ensemble average is over a color neutral medium with
Tr Ta (i) = 0.
Appendix B. Diagram M2,0,3 = G0 X1,1 X2,1

Consider next the gluon two-scatterings amplitude M2,0,3 in [27]. Fig. 8 shows that for
inclusive processes two interesting cases arise. In the Feynman diagram approach
Z
d 4 q1 d 4 q2
iJ (p + k q1 q2 )ei(p+kq1 q2 )x0 V (q1 )eiq1 x1 V (q2 )eiq2 x2
M2,0,3 =
(2)4 (2)4
12 (p, k, q1 , q2 )i(p + k q1 q2 )(i)(k q1 q2 )(i)(k q2 )

J (p + k)ei(p+k)x0 [[c, a2], a1 ] Ta2 (2)Ta1 (1)
Z 2
Z 2
d q2
d q1
(i)
2igs (k q1 q2 )eiq1 b1 eiq2 b2
(i)
2
(2)
(2)2
Z
dq1z dq2z
(2) (2)
(E + k + ) v(Eq1 )v(Eq2 )eiq1z (z1 z0 ) eiq2z (z2 z0 )
,
((p + k q1 q2 )2 + i)((k q1 q2 )2 + i)((k q2 )2 + i)
409
(B.1)
where bi = xi x0 are transverse impact parameters, and we used the soft gluon and rescattering kinematical simplifications Eqs. (29), (34), e.g., J (p +k q1 q2 ) J (p +k)
J (p). For the q1z integral, it is convenient to rewrite the phase as
eiq1z (z1 z0 ) eiq2z (z2 z0 ) = ei(q1z +q2z )(z1 z0 ) eiq2z (z2 z1 ) .
The first longitudinal integral is closely related to Eq. (A.2)
E 2 , z1 z0 )
I2 (p, k, q1 , q
Z
v(q1z , q1 )ei(q1z +q2z )(z1 z0 )
dq1z
.
=
2 ((p + k q1 q2 )2 + i)((k q1 q2 )2 + i)
(B.2)
Since z1 z0 1/, we again close the contour in the lower half q1z plane and neglect
the pole at i1 . The remaining q1z poles are shifted by q2z and q1 q1 + q2 relative
to Eq. (A.3):
q1 = q2z + E + + i,
q2 = q2z 0 i,
q3 = q2z + k + (12) + i,
q4 = q2z 0 + (12) i,
(B.3)
where now k + (12) = (k q1 q2 )2 . The residues at q2 , q4 then give

I2 i

ei0 (z1 z0 )
i(12) (z1 z0 )
,
q
)
v(q
,
q
)e
v(q
,
2z
0
1
2z
0
1
(12)
E + k + (12)
(B.4)
where we have neglected O(exp()) contributions. This differs from Eq. (A.6) mainly
in that the potential is evaluated near q2z , which still remains to be integrated over, and
1 (12) .
Fig. 8. M2,0,3 = G0 X1,1 X2,1 direct contributes to second order in opacity (el /A )2 , whereas
c
M2,0,3
= M2,0,3 (z2 = z1 ) G0 O1,1 contact-limit may contribute to first order in opacity
(el /A )1 as well.
410
Next, we need the following critical q2z integral

Z
dq2z v(q2z + , q1 )v(q2z , q2 )eiq2z (z2 z1 )
.
I3 (k, q1 , q2 , z2 z1 )
2
((k q2 )2 + i)
(B.5)
Fortunately, we are interested in only two extreme limits:

the limit of well-separated scattering centers z2 z1 1/;
the special contact z2 = z1 limit to compute unitary contributions.
In the first case we can proceed ignoring the q2z = i1 and i2 potential singularities.
Then only the pole at q2z = 2 0 i contributes and yields

i
(B.6)
I3 k, q1 , q2 , z2 z1 1/ + v(0, q1 )v(0, q2 )ei(2 0 )(z2 z1 ) .
k
In this case we recover the result quoted in Ref. [27] for M2,0,3 ,
Z 2
Z 2
d q1
d q1
iq1 b1
v(0,
q
)
e
(i)
v(0, q2 ) eiq2 b2
M2,0,3 J (p)eipx0 (i)
1
2
(2)
(2)2

(k q1 q2 ) i(0 2 )z2 i(2 (12) )z1 i(12) z1
(2igs )
e
e
ei(12) z0
e
2
(k q1 q2 )
[[c, a2], a1 ](Ta2 Ta1 ).
(B.7)
In the general case (including the special contact case with z2 = z1 ) both q2z =
i2 , i1 singularities in the Yukawa potential contribute together with the pole at q2z =
2 0 i, resulting in
I3 (k, q1 , q2 , z2 z1 )

i
+ v(0, q1 )v(0, q2 )ei(2 0 )(z2 z1 )
k
2 (z2 z1 )

e
(4s )2
e1 (z2 z1 ) ei(z2 z1 )
.
2 (21 22 )
22
21
(B.8)
For z2 z1 = 1/ this reduces to Eq. (B.6). For the special contact contribution
z2 z1 = 0 it reduces to
I3 (k, q1 , q2 , 0)
i.e., exactly
c
M2,0,3
1
2
i
v(0, q1 )v(0, q2 ),
2 k+
(B.9)
of the strength in Eq. (B.6). The contact limit of this amplitude is therefore
Z 2
Z 2
d q1
d q2
iq1 b1
J (p)eipx0 (i)
v(0,
q
)
e
(i)
v(0, q2 ) eiq2 b2
1
2
(2)
(2)2

1
(k q1 q2 ) i0 z1
(2igs )
e
1 ei(12) (z1 z0 )
2
(k q1 q2 )2
(B.10)
[[c, a2], a1 ](Ta2 Ta1 ).
After this explicit derivation of the factor 12 in the contact limit, we can generalize
it to any functional form of the potential as follows. First we assume that a contact
E 1 = Eq2 , corresponding to no net transverse momentum
interactions only contributes for q
411
exchange inside the potential function only, i.e., v(q1z , q1 ) = v(q2z , q2 ). However,
the distinction has to be kept in the propagators and phases. In this case the potential
function v(Eq1 ), does not appear in the first integral I2 , Eq. (B.2), which is modified to
Z
ei(q1z +q2z )(z1 z0 )
dq1z
E 2 , z1 z0 ) =
I2 (p, k, q1 , q
2 ((p + k q1 q2 )2 + i)((k q1 q2 )2 + i)
i

e 0 (z1 z0 )
1 ei(12) (z1 z0 ) .
(B.11)
i + +
E k (12)
This is independent of the functional form of the potential. The second integral, I3 , is then
modified to
Z
dq2z |v(q2z , q2 )|2 eiq2z (z2 z1 )
I3 (k, q1 = q2 , z2 z1 )
2
(k q2 )2 + i
Z
1
dq2z |v(q2z , q2 )|2 eiq2z (z2 z1 )
,
(B.12)
+
k
2
q2z + i
where z2 z1 and we have used the finite range of |v(qz , q)|2 , i.e., v is limited within
(, +), to neglect the residue at q2z k + . For large z2 z1 compared to the range of
v, we close below and obtain the dominant residue
2
i
(B.13)
I3 + v(0, q2 ) ei(z2 z1 ) ,
k
independent approximately of the actual form of v as long as all singularities of v in the
lower half plane have imaginary parts ii with i = i (z2 z1 ) 1.
To extract the contact limit, z2 z1 = 0, we can integrate along the real q2z axis and use
the Dirac representation of a pole approaching the real axis
q2z
1
=
i(q2z ).
q2z + i (q2z )2 + 2
This gives
2
i
k + ei(z2 z1 ) I3 v(, q2 )
2
Z
2 qz cos(qz ) i qz sin(qz )
d qz
v(qz + , q2 )
+
, (B.14)
2
qz2 + 2
where we have introduced qz = q2z . For our high energy eikonal applications
, and therefore we can now expand |v|2 around = 0. The correction of O(/) can
be neglected. This leads to a vanishing real part since that part of the integrand is odd. On
the other hand, the imaginary part has a finite = 0 limit,
Z
i

0
d qz
sin(qz )
|v(qz , q2 )|2
=
2i |v(0, q2 )|2
2
qz
if = (z2 z1 ) 0,
if = (z2 z1 ) ,
(B.15)
where is the range of the potential v as indicated above. We see that while the detailed
interpolation between the asymptotic limits depends on the actual functional form of v, the
412
limits Eqs. (B.6), (B.9) and the factor of 12 reduction in the contact ( = 0) limit are very
general.
We thus confirm that Oi,1 diagrams where the two legs are attached to the gluon line
can be obtained from the higher order direct diagram by setting zi+1 = zi and reducing the
strength twice (i.e., times 12 ).
Appendix C. Diagrams M2,0,0 = G0 X1,0 X2,0 and M2,2,0 = G0 X1,0 G1 X2,0 G2
In those graphs it is the jet rather than the gluon that suffers two sequential scatterings
as seen from Fig. 9. It is straightforward to write the expression for the amplitude
Z
M2,0,0 =
d 4 q1 d 4 q2
(2)4 (2)4
(iE + )2 igs (2p + k) i(p + k q1 q2 )i(p q1 q2 ) i(p q2 )

a2 a1 c(Ta2 Ta1 )
Z 2
Z 2
d q1 iq1 b1
d q2 iq2 b2
ipx0
(i)
e
(i)
e
J (p)e
2
(2)
(2)2
2igs ( k) i0 z0
e
a2 a1 c(Ta2 Ta1 )(E + )2

x
Z
dq1z dq2z
(2) (2)
v(Eq1 )v(Eq2 )eiq1z (z1 z0 ) eiq2z (z2 z0 )

. (C.1)
((p + k q1 q2 )2 + i)((p q1 q2 )2 + i)((p q2 )2 + i)
Fig. 9. M2,0,0 = G0 X1,0 X2,0 and M2,2,0 = G0 X1,0 G1 X2,0 G2 graphs in the well-separated case
c
c
together with their z2 = z1 limits M2,0,0
G0 O1,0 , M2,2,0
G0 G1 O1,0 G1 .
413
In this case we define

E 2 , z1 z0 )
I2 (p, k, q1 , q
Z
v(q1z , q1 )ei(q1z +q2z )(z1 z0 )
dq1z
.
=
2 ((p + k q1 q2 )2 + i)((p q1 q2 )2 + i)
(C.2)
Since z1 z0 1/, we neglect the pole at i1 . The remaining q1z poles are
q1 = q2z + E + + k + + i,
q2 = q2z 0 i,
q3 = q2z + E + + i,
q4 = q2z i,
(C.3)
where we discarded (p + k q1 q2 )2 /E + relative to 0 . The q2 , q4 residues then give

I2 i

v(q2z , q1 ) i0 (z1 z0 )
1 .
e
+
2
(E ) 0
(C.4)
Note that 0 has been neglected in the potential relative to 1 . The second integral, I3 , is
then modified to
Z
dq2z v(q2z , q1 )v(q2z , q2 )eiq2z (z2 z1 )
I3 (p, q1 , q2 , z2 z1 )
(2)
((p q2 )2 + i)

1, if = (z2 z1 ) ,
i
+ v(0, q1 )v(0, q2 ) 1
E
2 , if = (z2 z1 ) 0.
(C.5)
Note that E + 0 = k2 /x. With the help of Eqs. (C.4), (C.5) in the case of well-separated
scattering centers we obtain
Z 2
Z 2
d q1
d q2
iq1 b1
v(0,
q
)e
(i)
v(0, q2 )eiq2 b2
M2,0,0 = J (p)eipx0 (i)
1
2
(2)
(2)2

2igs ( k) i0 z1
ei0 z0 a2 a1 c (Ta2 Ta1 ).

(C.6)
e
2
k
We thus recover the result from [27]. In the contact limit there is extra factor of 12 relative
to the naive expectation and Eq. (C.6). Analogous calculation leads to the expected result
for M2,2,0 .
We thus arrive at a conclusion similar to the one in Appendix B, i.e., in the case when
the two legs of the virtual contribution are attached to the quark line, Oi,0 , there is an
additional factor of 12 relative to just setting zi+1 = zi in the higher order well-separated
case amplitudes.
Appendix D. Diagrams M2,0,1 = G0 X1,1 X2,0 and M2,0,2 = G0 X1,0 X2,1
In the case when one of the hits is on the parent parton and the other hit is on the radiated
gluon we encounter a different situation. Explicit calculation shows this on the example of
c
= M2,0,1 (z2 = z1 ) in Fig. 10.
M2,0,1
414
c
c
Fig. 10. M2,0,1
= G0 O1,2 and M2,0,2
= G0 O1,2 topologically indistinct contact diagrams. There
1
are no additional factors of 2 arising from the integration in taking the z2 = z1 limit in
M2,0,1 = G0 X1,1 X2,0 and M2,0,2 = G0 X1,0 X2,1 .
Z
M2,0,1 =
d 4 q1 d 4 q2
(2)4 (2)4
(iE + )1 i(p + k q1 q2 )(i)(k q1 ) i(p q2 )

Z 2
Z 2
d q1 iq1 b1
d q2 iq2 b2
e
(i)
e
J (p)eipx0 (i)
(2)2
(2)2
2igs ( (k q1 ))ei0 z0 a2 [c, a1 ](Ta2 Ta1 )(E + )2
Z
v(Eq1 )v(Eq2 )eiq1z (z1 z0 ) eiq2z (z2 z0 )
dq1z dq2z
.
(2) (2) ((p + k q1 q2 )2 + i)((k q1 )2 + i)((p q2 )2 + i)

(D.1)
We perform the q1z integral first

E 2 , z1 z0 )
I2 (p, k, q1 , q
Z
v(q1z , q1 )ei(q1z +q2z )(z1 z0 )
dq1z
.
=
2 ((p + k q1 q2 )2 + i)((k q1 )2 + i)
(D.2)
The pole at i1 is again exponentially suppressed. The poles of interest in the lower half
plane are q1z = q2z 0 i and q1z = 0 + 1 i. Taking the residues leaves us
with
I2 = i

ei0 (z1 z0 )
v(q2z 0 , q1 ) v(1 0 , q1 )ei(q2z +1 )(z1 z0 ) .
+
+
E k (q2z + 1 )
(D.3)
It is important to notice that there is no pole at q2z = 1 in Eq. (D.3). The remaining
integral over q2z is
I3 (p, k, q1 , q2 , z1 z0 , z2 z1 )
iq2z (z2 z1 )
Z
1
e
dq2z
v(q2z 0 , q1 )v(q2z , q2 )
=
2 q2z + 1 (p q2 )2 + i

ei(q2z (z2 z0 )+1 (z1 z0 ))
v(1 0 , q1 )v(q2z , q2 ) .
(p q2 )2 + i
(D.4)
415
Unlike in the previous examples, we now show that no factor of 12 arises in the contact
limit z2 = z1 > z0 . The poles in the lower half plane are q2z = i, q2z = i2 , and
q2z = 0 i2 and the corresponding residues contribute
Res(i)

v(0, q1 )v(0, q2 )
1 ei1 (z1 z0 ) ,
+
E 1
Res(i2 )
(4s )2 e2 (z2 z1 )
E + 22 (2i2 )(21 22 2i0 2 )
Res(i1 0 )
(4s )2 e1 (z2 z1 ) e+i0 (z2 z1 )

.
E + 21 (2i1 )(22 21 + 2i0 1 )
(D.5)
The exponentially suppressed contributions exp[2 (z1 z0 )] in the second and third
residues have been neglected and we made use of the approximation i j . In the limit
of well separated scattering centers, the second and third residues are also exponentially
suppressed. In the contact limit, on the other hand, both the second and third residues in
Eq. (D.5) contribute. However, impact parameter averaging via Eq. (19) sets 1 = 2 , and
the those two contributions cancel exactly. The result is then
Z 2
Z 2
d q1 iq1 b1
d q2 iq2 b2
c
ipx0
(i)
e
v(0, q1 )(i)
e
v(0, q2 )
M2,0,1 = J (p)e
(2)2
(2)2
2igs

(k q1 ) i(0 1 )z1 i1 z1
e
ei1 z0 a2 [c, a1](Ta2 Ta1 ).
e
2
(k q1 )
(D.6)
We emphasize that in this case, there is no factor of 12 emerging in the z2 = z1 contact limit
Oi,2 when one of the legs is attached to the quark line and the other on to the gluon line.
Similarly, the diagram M2,0,2 leads to the result quoted in Ref. [27]. In the contact limit,
it reduces to M2,0,2 (z2 = z1 ) = M2,0,1 (z2 = z1 ). However, it is important to point out that
in the contact limit only one of the diagrams must be taken into account in order to avoid
over counting. This can be directly seen from the expansion in the powers of the interaction
Lagrangian. Alternatively, we get the correct contact answer by taking the limit of both
cross contact diagrams and multiplying each by 12 .
Appendix E. Zero measure contact limit of M2,1,0 = X1,0 G1 X2,0 and
M2,1,1 = X1,0 G1 X2,1
In calculating the different contributions coming from two interactions with the same
potential centered around xE 1 we have to take into account the two graphs given in Fig. 11,
where one of the hits occurs before the gluon emission vertex and the other one after.
In the framework of time-ordered perturbation theory of [26,27] the graphs are
Rt
identically zero in the contact limit because t11 dt 0. We here present a more detailed
c
= X1,0 G1 X1,0 .
study of the validity of this argument. In the contact limit M2,1,0 M2,1,0
+
+
We will show that this contribution vanishes in the small x = k /E limit
416
c
c
Fig. 11. Diagrams M2,1,0
= X1,0 G1 X2,0 (z2 = z1 ) and M2,1,1
X1,0 G1 X2,1 (z2 = z1 ) of O(0)
according to the time-ordered perturbation theory.
Z
c
M2,1,0
=
d 4 q1 d 4 q2
iJ (p + k q1 q2 )ei(p+kq1 q2 )x0 V (q1 )eiq1 x1
(2)4 (2)4
V (q2 )eiq2 x1 (iE + )2 igs (2(p q2 ) + k) i(p + k q1 q2 ) i

(p + k q2 ) i(p q2 )a2 ca1 (Ta2 Ta1 )
Z 2
Z 2
d q1 iq1 b1
d q2 iq2 b1
e
(i)
e
J (p)eipx0 (i)
2
(2)
(2)2
Z
k
dq1z dq2z
v(Eq1 )v(Eq2 )
2igs 2 ei0 z0 a2 ca1 (Ta2 Ta1 )(E + )2
k
(2) (2)

1
1
ei(q1z +q2z )(z1 z0 )
. (E.1)
(p + k q1 q2 )2 + i (p q2 )2 + i (p + k q2 )2 + i
In writing this we have made use of the simplifying soft gluon and weak interaction
approximations in Eq. (29).
We can now perform the q1z integral, which brings a factor of 1/E + and sets the overall
phase to 0. We have to now perform

Z
1
q2z
1
v(q2z , q1 )v(q2z , q2 )
. (E.2)
I3 =
2
(p q2 )2 + i (p + k q2 )2 + i
We close in the upper half plane where there are four poles q2z = i1 , q2z = i2 , q2z =
E + + i, q2z = E + + k + + i. The residues at the poles involving E + are suppressed by
a factor (1/E + )4 and have negligible contribution in the high energy approximation. The
remaining two residues give

(4s )2
k+
,
Res(i1 )
E+
E + (221 )(22 21 )

(4s )2
k+
+ .
(E.3)
Res(i2 )
E
E + (222 )(21 22 )
We have discarded 0 relative to i and carried the k + /E + expansion to first order. It is
important to notice the sum of the residues is not singular for the special case of interest
when the contact limit contribution is averaged over the transverse position of the scatterer,
i.e., 1 = 2 . The above contribution is suppressed by an O(k + /E + ) factor relative to
the other graphs. In the high energy E + limit, where the time-ordered perturbation
c
X1,0 G1 X1,1
theory works, M2,1,0 (z2 = z1 ) 0. Similar calculation for M2,1,1 M2,1,1
417
shows suppression of the order O(/k + ) and is approximately zero according to the k +
regime (28) in which the problem is set up.
This leads us to the conclusion that the naive zero measure argument for such graphs
is approximately valid for our kinematics.
Appendix F. Amplitude iteration technique to second order

To illustrate the general iteration procedure and check the general iteration results using
the reaction operator approach, we construct the classes for rank 1 and 2 explicitly.
For the two rank 1 classes, we apply the operators D 1 and V1 on the hard vertex
amplitude Eq. (39) once to obtain
(F.1)
G0 D1 = H ei0 z0 a1 c C1 ei(0 z1 1 z10 ) [c, a1] B1 ei0 z1 [c, a1],
CR + CA
C
C
A
A
H ei0 z0 c
C1 ei(0 z1 1 z10 ) c
B1 ei0 z1 c.
(F.2)
G0 V 1 =
2
2
2
e recall that zij zi zj , 1zi zi zi1 . These are all amplitudes needed for the first
order in opacity L/ calculation.
Some of the rank two classes are obtained from the rank 1 classes Eqs. (F.1), (F.2)
through relabeling, i.e., G0 D2 G0 D1 (1 2), G0 V2 G0 V1 (1 2). The rest are
readily derived from Eqs. (F.1), (F.2) through our iteration scheme Eqs. (76), (82)
G0 D1 D2 = H ei0 z0 a2 a1 c C1 ei(0 z1 1 z10 ) a2 [c, a1 ]
C2 ei(0 z2 2 z20 ) a1 [c, a2] B1 ei0 z1 a2 [c, a1]
B2 ei0 z2 [c, a2 ]a1 C(12) ei(0 z2 2 z21 (12) z10 ) [[c, a2], a1 ]
B2(12) ei(0 z2 2 z21 ) [[c, a2], a1 ],
CR + CA
CR + CA
H ei0 z0 a1 c +
C1 ei(0 z1 1 z10 ) [c, a1]
G0 D1 V2 =
2
2
+ C2 ei(0 z2 2 z20 ) a2 a1 [c, a2]
CA
CR + CA
B1 ei0 z1 [c, a1 ]
B2 ei0 z2 ca1
+
2
2
+ C(12) ei(0 z2 2 z21 (12) z10 ) a2 [[c, a2], a1 ]
(F.3)
+ B2(12) ei(0 z2 2 z21 ) a2 [[c, a2], a1 ],

CR + CA
H ei0 z0 a2 c
G0 V1 D2 =
2
CR + CA
CA
C1 ei(0 z1 1 z10 ) a2 c +
C2 ei(0 z2 2 z20 ) [c, a2]
2
2
CR
CA
B1 ei0 z1 a2 c +
B2 ei0 z2 [c, a2 ]
2
2
CA
C(12) ei(0 z2 2 z21 (12) z10 ) [c, a2]
2
CA
B2(12) ei(0 z2 2 z21 ) [c, a2],
(F.4)
(F.5)
418
(CR + CA )2
CA (CA + CR )
H ei0 z0 c +
C1 ei(0 z1 1 z10 ) c
4
4
CA (CR + CA )
C2 ei(0 z2 2 z20 ) c
+
4
CR CA
CA (CR + CA )
B1 ei0 z1 c +
B2 ei0 z2 c
+
4
4
C2
C2
A C(12) ei(0 z2 2 z21 (12) z10 ) c A B2(12) ei(0 z2 2 z21 ) c. (F.6)
4
4
With this explicit construction of the relevant classes, we can compute the differential
gluon probability up to second order in the opacity expansion. Similar calculations can be
performed for the GunionBertsch case as well.
To first order in opacity L/ the probability to radiate a gluon from either quark or gluon
jets is proportional to
D
E
E
1 D
= CR CA (2C1 B1 ) 1 cos(1 1z1 )
Tr G0 D1 D1 G0 + 2 Re G0 V1 G0
v
v
dR
Z

= CR CA
(F.7)
d 2 q1 v 2 (q1 ) 2 (q1 ) (2 C1 B1 ) 1 cos(1 1z1 ) ,
G0 V 1 V 2 =
where the 17 terms in Eqs. (43), (44) collapse to a single term with color trivial factor
CR CA dR interpretable as gluon final state interaction only. We note that if the normalized
potential v 2 (q1 ) is replaced by v 2 (q1 ) 2 (q1 ) the result of Eq. (F.7) remains unchanged.
To second order in opacity we find that

1 D
Tr G0 D1 D2 D2 D1 G0 + G0 V1 D2 D2 G0 + h.c. + G0 D1 V2 D1 G0 + h.c.
dR

E
+ G0 V1 V2 G0 + h.c. + G0 V1 V2 G0 + h.c.
v
D

2
= CR CA 2C1 B1 1 cos(1 1z1 )

+ 2C2 B2 cos(2 1z2 ) cos(2 (1z1 + 1z2 ))

2C(12) B2(12) 1 cos((12)1z1 )
E
2C(12) B2 cos(2 1z2 ) cos((12) 1z1 + 2 1z2 )
v
Y Z

= CR CA2
d 2 qi v 2 (qi ) 2 (qi )
i=1,2

2 C(12) B2(12) 1 cos((12) 1z1 )

2C(12) B2 cos(2 1z2 ) cos((12) 1z1 + 2 1z2 ) .
(F.8)
Eqs. (F.7), (F.8) have simple physical interpretation. We showed that for both quark and
gluon jets a simple color trace CAn survives. In general this can be seen from Eqs. (76),
(82). Thus the distribution of the radiated gluons is interpretable as the interference of the
Cascade amplitude that has the knowledge of all final state interactions with the effective
419
color currents, i.e., the GunionBertsch amplitudes, originating at the scattering centers
and also having all possible subsequent interactions with the gluon line. The formation
physics effects arise from phase differences that store the information on the cumulative
formation lengths before the effective color current emission. The averaging over the
momentum transfers is seen to be effectively over a modified potential that vanishes upon
integration.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
J.D. Bjorken, FERMILAB-PUB-82-59-THY, unpublished.

M. Gyulassy, M. Plmer, Phys. Lett. B 243 (1990) 432.
M.H. Thoma, M. Gyulassy, Nucl. Phys. B 351 (1991) 491.
M. Gyulassy, M. Plmer, M.H. Thoma, X.-N. Wang, Nucl. Phys. A 538 (1992) 37c.
X.-N. Wang, M. Gyulassy, Phys. Rev. Lett. 68 (1992) 1480.
M. Gyulassy, X.-N. Wang, Nucl. Phys. B 420 (1994) 583.
X.-N. Wang, M. Gyulassy, M. Plumer, Phys. Rev. D 51 (1995) 3436, hep-ph/9408344.
X.-N. Wang, Z. Huang, I. Sarcevic, Phys. Rev. Lett. 77 (1996) 231, hep-ph/9605213.
X.-N. Wang, Phys. Rept. 280 (1997) 287, hep-ph/9605214.
R. Baier, Yu.L. Dokshitzer, A.H. Mueller, S. Peign, D. Schiff, Nucl. Phys. B 483 (1997) 291.
R. Baier, Yu.L. Dokshitzer, A.H. Mueller, S. Peign, D. Schiff, Nucl. Phys. B 484 (1997) 265.
R. Baier, Yu.L. Dokshitzer, A.H. Mueller, D. Schiff, Nucl. Phys. B 531 (1998) 403.
B.G. Zakharov, JETP Lett. 63 (1996) 952.
B.G. Zakharov, JETP Lett. 65 (1997) 615.
I.P. Lokhtin, A.M. Snigirev, Phys. Lett. B 440 (1998) 163.
R. Baier, Y.L. Dokshitzer, A.H. Mueller, D. Schiff, Phys. Rev. C 60 (1999) 064902.
I.P. Lokhtin, A.M. Snigirev, Eur. Phys. J. C 16 (2000) 527, hep-ph/0004176.
B.Z. Kopeliovich, A.V. Tarasov, A. Schafer, Phys. Rev. C 59 (1999) 1609, hep-ph/9808378.
U.A. Wiedemann, M. Gyulassy, Nucl. Phys. B 560 (1999) 345, hep-ph/9906257.
U.A. Wiedemann, Nucl. Phys. B 582 (2000) 409, hep-ph/0003021.
X. Wang, Phys. Rev. C 61 (2000) 064910, nucl-th/9812021.
X. Guo, X.-N. Wang, Phys. Rev. Lett. 85 (2000) 3591, hep-ph/0005044.
S. Bass et al., Nucl. Phys. A 661 (1999) 205c.
X.-N. Wang, Phys. Rev. Lett. 81 (1998) 2655.
M. Gyulassy, P. Levai, Phys. Lett. B 442 (1998) 1.
M. Gyulassy, P. Lvai, I. Vitev, Nucl. Phys. A 661 (1999) 637c.
M. Gyulassy, P. Lvai, I. Vitev, Nucl. Phys. B 571 (2000) 197.
M. Gyulassy, P. Lvai, I. Vitev, CU-TP-976, nucl-th/0005032.
K.J. Eskola, K. Kajantie, P.V. Ruuskanen, K. Tuominen, Nucl. Phys. B 570 (2000) 379.
L.D. Landau, I.J. Pomeranchuk, Dokl. Akad. Nauk. SSSR 92 (1953) 92.
A.B. Migdal, Phys. Rev. 103 (1956) 1811.
J.F. Gunion, G. Bertsch, Phys. Rev. D 25 (1982) 746.
R. Field, Applications of Perturbative QCD, AddisonWesley, 1989.
U.A. Wiedemann, Nucl. Phys. B 588 (2000) 303, hep-ph/0005129.
B. Zhang, Comput. Phys. Commun. 109 (1998) 193.
M. Gyulassy, Y. Pang, B. Zhang, Nucl. Phys. A 626 (1997) 999.
D. Molnar, M. Gyulassy, Phys. Rev. C 62 (2000) 054907, nucl-th/0005051.
D.Molnr, MPC0.1.2, code available at site http://nt3.phys.columbia.edu/people/molnard.
A.C. Genz, A.A. Malik, J. Comput. Appl. Math. 6 (1980) 295.
A. van Doren, L. de Ridder, J. Comput. Appl. Math. 2 (1976) 207.

QCD in a finite box: numerical test studies

in the three LeutwylerSmilga regimes
Stephan Drr
Paul Scherrer Institut, Theory Group, 5232 Villigen PSI, Switzerland
Abstract
The LeutwylerSmilga prediction regarding the (ir)relevance of the global topological charge
for QCD in a finite box is subject to a test. To this end the lattice version of a suitably chosen
analogue (massive 2-flavour Schwinger model) is analyzed in the small (V m 1), intermediate
(V m ' 1) and large (V m 1) LeutwylerSmilga regimes. The predictions for the small and
large regimes are confirmed and illustrated. New results about the role of the functional determinant
in all three regimes and about the sensitivity of physical observables on the topological charge in the
intermediate regime are presented. 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
One of the relevant concepts in an attempt to understand the mechanism of spontaneous
breakdown of chiral symmetry 1 in QCD is provided by instantons, i.e., topologically
nontrivial solutions of the classical field equations which are both a local minimum of
the classical action and localized in space-time. They are known to influence the local
propagation properties of the light flavours (Instanton Liquid Model [1]).
What is not so clear is whether the global concept of the number of instantons minus
anti-instantons for QCD in a finite box
Z
g2
dx
GG
(1)
=
32 2
plays a role, too. This is important, because in lattice studies involving dynamical fermions
standard simulation algorithms tend to get stuck in a particular topological sector if the
E-mail address: stephan.duerr@psi.ch (S. Drr).
1 We shall only consider the theory with several light dynamical fermions (N > 2).
f
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 8 - 1
S. Drr / Nuclear Physics B 594 (2001) 420438
421
Fig. 1. LHS: Distribution of the unrenormalized field-theoretic topological charge nai in an ensemble
of 5000 SU(3)-configurations generated from SWilson at = 6.1, after 6 cooling sweeps and with a
2-smeared charge operator (figure taken from [4]). RHS: Time history, in units of MD time, of nai
for a HMC simulation with a quadruple of staggered quarks at = 5.35, m = 0.01; the analogous
distribution lacks the symmetry and shows hi 6= 0 (figure taken from [4]).
sea-quarks are taken sufficiently light 2 and obviously one would like to know whether this
affects physical observables. In other words, the question is whether non-ergodicities of
the sample w.r.t. the topological charge as visible, e.g., from the r.h.s. of Fig. 1, tend to
afflict measurements performed on such a sample.
The correct sampling of the topological sectors is known to be crucial for quantities
such as the 0 mass, which depend directly on the distribution of topological charges via
the explicit breaking of the U (1)A symmetry [5]. On the other hand, standard observables
which do not relate to the U (1)A issue (M , MK , M , M , Vq q , . . .) are known not to
depend on the global topological charge if the four-volume V of the box is taken
sufficiently large. An early investigation trying to assess the issue on a more quantitative
level is the one by Leutwyler and Smilga [6]. Their main observation is that for the case
of pionic observables a complete answer can be given from purely analytic considerations
for two extreme regimes of quark masses and box volumes, both of which involve the
LeutwylerSmilga (LS) parameter 3
x V m.
(2)
Their first statement is that in a sufficiently small box (x 1) the question whether
the algorithm managed to tunnel between the various topological sectors at a sufficient
rate shouldnt emerge, because the partition function is completely dominated by the
topologically trivial sector. In lattice language this amounts to the statement that the
algorithm is not supposed to tunnel into a nontrivial sector anyway. Their second statement
is that in the opposite regime (x 1 plus a condition to be discussed below) the total
2 The phenomenon was first observed for HMC with staggered quarks [2]. For Wilson type sea-quarks mobility
between the topological sectors was reported to be sufficient for M /M > 0.56, but the same problem emerges
if is tuned sufficiently close to crit [3,4].
3 V is the four-volume of the box, = lim
(this order) is the chiral condensate in the

m0 limV hi
chiral limit and m is the (degenerate) sea-quark mass; note that both and m are scheme- and scale-dependent,
but the combination is an RG-invariant quantity.
422
topological charge is an irrelevant concept. This means, in lattice language, that the
observable is unaffected by whichever topological path the simulation followed. From a
lattice perspective, however, the analysis by Leutwyler and Smilga is not quite sufficient,
as their paper doesnt state explicitly whether a simulation in the small x regime can be
trusted, if the algorithm got stuck in a higher topological sector even though it was not
supposed to get into it, and also because there is no statement about the situation in the
intermediate (x ' 1) regime, which might be interesting to use in dynamical simulations.
Last but not least the relevance of the second condition in the LS-analysis for large x
(a condition which, as we shall see, is never fulfilled in QCD simulations) is not discussed.
Below numerical results will be used to elucidate in which regimes of box volumes and
quark masses a standard observable like the heavy quark potential is indeed insensitive
to the global topological charge and by which mechanism a dependence sneaks in if the
insensitive regime is left. I start with a brief review of the approach taken by Leutwyler
and Smilga, followed by an argument why the massive multi-flavour Schwinger model is
a suitable testbed for numerical studies. Sections 4 through 6 present distinctive features
by which the three LS-regimes differ from each other both on a formal level (role being
played by the functional determinant) and w.r.t. observables (Polyakov loop, heavy quark
potential). The strategy used is to disentangle the complete sample into contributions from
the individual topological sectors and to observe how the generating functional weights
the fixed- expectation values to form the physical observable [7]. I conclude with a few
remarks on the potential relevance of the results for future QCD simulations.
2. LeutwylerSmilga analysis for QCD

Here I shall briefly summarize the considerations by Leutwyler and Smilga regarding
the distribution of topological charges for QCD in a finite box [6].
Leutwyler and Smilga start from the partition function for QCD with Nf > 2 degenerate
light (dynamical) flavours on an Euclidean torus
R
XZ

/ + m Nf e(1/4) GG ,
(3)
DA() det D
Z
Z
where the integration is over all gauge potentials in a given topological sector and the quark
fields have been integrated out. In (3) the VafaWitten representation [8] for the functional
determinant

Y0
Nf
/ + m)Nf = m||
(4)
2 + m2
det(D
may be used which splits the latter into a factor representing the contribution from the zero
/ on the given background times a chain of factors i + m
modes of the Dirac operator D
from the paired non-zero modes. Upon using the nontrivial input for the first few non-zero
eigenvalues
n
(5)
n '
V
423
(which is just the familiar BanksCasher relation [9]) the primed factor in the decomposition (4) is seen not to depend on m if x 1, and hence
Z m||Nf
(x 1).
(6)
This means that in the chiral symmetry restoration regime (x 1) topologically nontrivial
sectors are heavily suppressed w.r.t. the trivial one in particular for small quark masses
(in lattice units) and large Nf .
On the other hand, if the box volume is sufficiently large, the onset of chiral symmetry
breaking gets manifest as the long-range Greens functions get dominated by the Goldstone
excitations and the theory is effectively described by a chiral Lagrangian 4
Z
R
Z
DU e LXPT ,
SU(Nf )

F2
U U
U M + MU ,
(7)
4
2
where the integration is over all smooth fields taking values in SU(Nf ), and standard chiral
perturbation
theory (XPT) to order O(p2 ) has been used. 5 In the representation (7) U =
exp(i 2 /F ) is the Goldstone manifold and M = diag(m, . . . , m) > 0 the quark mass
matrix; h. . .i means a flavour trace, and F is the pion decay constant to O(p2 ). The key
observation by Leutwyler and Smilga is that if they take not only the box length sufficiently
large for XPT to apply 6 but also the quark masses sufficiently light
LXPT =
1
1
L
XPT
M
(8)
the collective (constant mode) excitations are no longer suppressed, but provide the
dominant contribution [11], i.e., (7) is approximately given by
Z
(9)
Z dU0 eV hU0 M +MU0 i/2 ,
where the path integral has collapsed into a simple group integral given by the Haar
measure. From (9) one gets the contribution of the topological sector to the partition
function in the standard way by substituting M Mei/Nf and Fourier transforming from
to , whence
Z
Z2
Z
d
0
dU0 ei eV hU0 e
i/Nf
M +h.c.i/2

d U (det U ) eV hU M +h.c.i/2 ,
(10)
U (Nf )
4 References to all the subtle points (e.g., why in the particular case of the torus no boundary terms show up in
LXPT ) are found in [6].

5 For a discussion in the framework of generalized chiral perturbation theory see [10].
6
XPT denotes a QCD intrinsic low-energy scale, e.g., QCD , , F .
424
where in the last line U0 and the factor ei/Nf have been combined into the U (Nf ) matrix
U . From (10) it follows that, if V m is large, the dominant contribution to the group
integral will come from the area where U is in the vicinity of the identity matrix, which
means that the determinant is approximatively one and hence Z is independent of .
A refined analysis shows that this is true for 2 V m/Nf , and the overall distribution
is [6]
Z e
2
2h 2 i
with h 2 i =
V m
Nf
(x 1).
(11)
I shall attach a few comments:

(i) It is worth noticing that the LS-classification of small, intermediate and large box
volumes (x 1, x ' 1, x 1) does not coincide with the usual classification on the
lattice (L M1 , L ' M1 , L M1 ). Note also that one involves the masses of the
sea-quarks, the other of the current-quarks.
(ii) The analysis by Leutwyler and Smilga covers the symmetry restoration regime
(x 1) and, on the other side, the regime where SSB is manifest, but the Goldstone
bosons overlap the box (x 1, L M1 ). The latter condition besides not being useful
because it prevents 7 extraction of pionic observables is, in fact, completely inaccessible
in present days simulations: Putting numbers into Eq. (8) one gets
XPT ' 200 MeV H L ' 3 fm H M ' 20 MeV,
(12)
which amounts to a sea-quark mass 8 of the order of 70 keV. From this we conclude that,
in order to be useful in dynamical QCD simulations, the result (11) must turn out to rely
exclusively on the condition x 1 being fulfilled, not on the r.h.s. of (8). In other words,
our hope is that the latter turns out to be a purely technical condition, immaterial to the
result (11).
(iii) Comparing the announcement of the results by Leutwyler and Smilga in the
introduction to their considerations as sketched above, one might worry because the latter
aims at the relative weight of the topological sectors in the partition function, whereas
the introduction mentioned possible -dependencies of observables. From a formal point
of view one may argue that this is not an issue, because the LS-analysis goes through if
the theory is coupled to tiny external sources, and infinitesimal sources are sufficient to
produce Greens functions and observables. Hence one expects that a strong dependence
or approximate independence of the partition function w.r.t. transfers into a sensitivity
or immunity of correlation functions and observables, but still one would desire to see this
explicitly in numerical data.
7 Unless the current-quark mass is taken considerably heavier than the sea-quark mass.
8 Since M 2 m, a sea-pion which is lighter than the physical pion by a factor 7 amounts to a sea-quark mass
49 times smaller than the phenomenological value of (mu + md )/2. For quark masses the usual conventions
(MS, = 2 GeV) are adopted.
425
3. Adaptation to 2-flavour QED(2)

The aim is to provide simulation data in the three LS-regimes x 1, x ' 1 and x 1,
while in addition fulfilling the usual condition L 1/M , i.e., the opposite of the r.h.s.
of (8).
A toy theory where the LS-issue can be studied at greatly reduced costs (in terms of CPU
power) while maintaining all essential features is the massive multi-flavour Schwinger
model, i.e., QED(2) with Nf > 2 light degenerate fermions. The physics of this theory
resembles in many aspects QCD(4) slightly above the chiral phase transition, i.e., the
chiral condensate vanishes upon taking the chiral limit and the expectation value of the
Polyakov loop is real and positive. Furthermore, the required formal analogy holds true:
like in QCD the gauge fields in the continuum version of the Schwinger model quantized on
the torus fall into topologically distinct classes [12], i.e., they can be attributed a topological
index. The latter agrees with the number of left- minus right-handed zero-modes of the
massless Dirac operator, i.e., the index theorem holds true [12].
There is, however, an (apparent) problem: as we have seen in the previous section, the
physics in the large LS-regime (x 1) is governed by the onset of SSB for the global
SU(Nf )A flavour-group (though the box volume is still finite), but it is well known that in
2 dimensions no SSB with the associate production of Goldstone bosons takes place [13].
The reason why this apparent deficiency proves immaterial relates to the fact that the
multi-flavour Schwinger model (with Nf massless flavours) shows a second order phase
transition with critical temperature Tc = 0, a result due to Smilga and Verbaarschot [14].
The theory is most appropriately seen as the limiting case of an extension where the axial
flavour symmetry is explicitly broken, e.g., by tiny quark masses. In the strong-coupling
limit 9 m g the spectrum is found to involve a heavy particle with mass [15]
p
g
(13)
M+ ' Nf + O(m)
and Nf2 1 light particles with mass [13,16]

M ' const g 1/(Nf +1) mNf /(Nf +1) .
(14)
The state (13) is often called a massive photon, but it is more appropriately seen as the
analogue of the 0 in QCD, as it is the lightest flavour singlet and stays massive in the
chiral limit. The states (14) are quasi-Goldstones and we shall call them pions, but
it is important to keep in mind that they differ from the usual pseudo-Goldstones (as
encountered in QCD) as they do not turn into true Goldstone bosons upon taking m 0;
they rather become sterile if the chiral limit 10 is performed. That conforms with Colemans
theorem [13] which forbids the existence of massless interacting particles in 2 dimensions.
9 Note that in 2 dimensions the charge has the dimension of a mass.
10 Analogous phenomena are observed, if the axial flavour symmetry is broken by boundary conditions rather
than quark masses: The chiral condensate (i.e., the would be order parameter) vanishes with a power-law
dependence as the spatial box length is sent to infinity at zero temperature, but exponentially fast if the temperature
is non-zero [17].
426
The point I shall stress is this: as long as the axial flavour symmetry is explicitly broken
which, in the following, is true, as sea-quarks are taken massive the quasi-Goldstones
are, for practical purposes, as good as pseudo-Goldstones; the difference shows only up
upon taking the chiral limit. Hence even the large LS-regime, where long-range Greens
functions are dominated by pionic excitations, finds its analogue in the (massive) multiflavour Schwinger model.
One more practical issue has to be resolved: the original definition (2) of the LSparameter x involves the quantity , i.e., (the absolute value of) the chiral condensate
in the chiral limit. The latter, however, is exactly zero, due to Colemans theorem. A practical way out of this is based on Smilgas observation that upon combining (14) with the
expression [16]
|hi|
' const g 2/(Nf +1) m(Nf 1)/(Nf +1)
(15)
for the chiral condensate one gets the relation [16]
m|hi|
'
0.388
M2 ,
(2.008)2
(16)
where I have already plugged in the nonuniversal (i.e., Nf -dependent) constants for the
2-flavour case [18]. Relation (16) being the QED(2) analogue of the Gell-MannOakes
Renner PCAC-relation
1
(17)
m ' M2 F2 (Nf )
2
in QCD means that I may define the LS-parameter in the lattice studies presented below as
x
0.388
V M2
(2.008)2
(Nf = 2)
(18)
as one may rewrite it in the case of QCD (to leading order in XPT)
1
x = V M2 F2
2
(Nf )
(19)
in a way which involves just physical degrees of freedom. 11

In order to investigate the LS-issue, I have chosen to implement the Schwinger model
with a pair of (dynamical) staggered fermions, using the Wilson gauge action Sgauge =
P
(1 cos 2 ). Since the staggered Dirac operator is represented by relatively small
matrices (see below), I have decided to compute the determinants exactly, using the
routines ZGEFA and ZGEDI from the LINPACK package.
I compare the three regimes x 1, x ' 1, x 1 to each other using three dedicated
simulations: I have chosen to vary the box volume at fixed = 1/(ag)2 = 3.4 and
fixed staggered quark mass m = 0.09 (everything in lattice units). The three regimes
are represented by the three volumes V = 8 4, V = 18 6, V = 40 10, where
always periodic boundary conditions in the first (spatial) direction and thermal boundary
conditions in the second (Euclidean timelike) direction have been used. The LS-parameter
11 Note that in order to determine x in a lattice study through (19) one has to plug in M and F of the sea-pion,
i.e., in practice of a pion constructed from current-quarks having exactly the same mass as their sea-partners.
427
(18) takes the values x ' 0.33, x ' 1.12, x ' 4.16 respectively, while the pion (pseudoscalar iso-triplet) has a (common) mass 12 M = 2.008 0.5421/3 0.092/3 = 0.329 and
hence a correlation length = 3.04 as to fit into the box (almost) in all three regimes as
is usual in lattice context. Each time 3200 decorrelated (w.r.t. the plaquette) configurations
have been generated.
For the type of investigation I am aiming at configurations must be assigned an
1 P
log U2 and the
index . I have implemented both the geometric definition geo = 2
P
field theoretic definition fth = nai , nai = sin 2 with the renormalization factor '
1/(1 hSgauge i/V ) [19]. A configuration is assigned an index only if the geometric
and the field-theoretic definition, after rounding to the nearest integer, agree. This turned
out to be the case for 3197, 3163, 2818 configurations, i.e., on a 99.9%, 98.8%, 88.1%
basis on the small/intermediate/large lattice. This fraction being so high means that in
practice an assignment can be done without cooling for the overwhelming majority of
configurations. The remaining ones are just not assigned an index; they are used to compute
the unseparated observable (e.g., the complete heavy quark potential; see below), the
latter, however, turns out to be unaffected by whether they are included or not.
Having simulation data which are supposed to reflect the situation in the small/intermediate/large LS-regimes respectively, one may want to check the overall distribution of
topological charges as encountered in the three runs. As one can see from the r.h.s. of
Fig. 2, the overall distribution is noticeably different in the three regimes. The predictions
by Leutwyler and Smilga for the two extreme cases (x 1 and x 1) seem to be well
reproduced: for small x the partition function Z (and hence the sample) is dominated
by Z0 , the contribution from the topologically trivial sector, whereas for large x the
distribution gets broad and seems compatible with a Gaussian. It is worth mentioning that
the latter holds true even though the pion did not overlap the box; rather the usual relation
L applied. This is a first indication that the r.h.s. of (8) in the LS-analysis might
indeed be a purely technical condition, immaterial to what they claim, namely that in the
large x regime the topological charge is an irrelevant concept or, in other words, that
physical observables do not depend on if x 1 [6].
4. Sectoral reweightings and renormalizations

Since the LS-issue is peculiar to the full (unquenched) theory, an attempt to understand
by which mechanism the three regimes differ from each other may lead one to investigate
how the functional determinant or specifically its contribution to the total action per
continuum-flavour

/ + m + const
(20)
Sfermion = log det D
relates to the contribution from the gauge field and, in addition, how such a relationship
might depend on the topological charge of the background. The idea is thus to study such a
12 The formula used is the prediction (14), which seems adequate since m/g ' 0.166 1.
428
Fig. 2. Distribution of fth (left) and geo (right) in the small, intermediate and large
LeutwylerSmilga regimes, respectively.
relationship sectorally, i.e., after the complete sample has been separated into subsamples
with a fixed topological charge. Since the theory (at = 0) is P/CP-symmetric, the
configurations from the sectors may be combined into a single subsample characterized
by ||.
As one can see from the rightmost column of Fig. 3, the contribution per continuum
flavour to the total action, Sfermion , shows a positive correlation with the gauge action
Sgauge . This means that, on a qualitative level, switching on the functional determinant
amounts to an increase of , which is a well-known feature in (full) QCD too. However, the
correlation is weak, i.e., the cigars in the unseparated plots are thick. A key observation
is that the correlation improves, if one separates the sample into subsamples with fixed
||, as is done in Fig. 3. It is worth noticing that in general (i.e., without specifying the
LS-regime) the best linear fit for Sfermion as a function of Sgauge depends (in offset and
slope) on ||. This means that, to a first approximation, the functional determinant results

429
Fig. 3. Scatter plot of Sfermion /V = log(det(D

/ + m))/V versus Sgauge/V on the complete sample (rightmost column) and on subsamples with
|| = 0, 1, 2, 3 in the small (top), intermediate (middle) and large (bottom) LeutwylerSmilga regime. In the sectoral plots the constant in (20) is chosen
such that hSfermion i=0 = 0.
430
in an overall suppression of higher topological sectors w.r.t. lower ones and to a sectorally
different renormalization of . As we shall see, these two features constitute the place
where the three LS-regimes differ from each other.
First, we shall consider the offset of the regression line in the = 1 sectors compared
to the line in the sector with = 0, as this is a measure how much extra suppression the first
topologically nontrivial sector receives on average compared to the trivial one due to the
functional determinant. From the first and second column in Fig. 3 one sees that this offset
(and in general the offset between neighboring sectors) decreases with increasing x. This
provides an explanation [20] for the broadening of the -distribution inherent to passing
from the small to the intermediate or from the intermediate to the large LS-regime (cf.
Fig. 2). For x 1 the offset hSfermion i1 hSfermion i0 is larger than the fluctuations in
Sfermion within the zero-charge sector, and this means means that in the small LS-regime
the functional determinant acts as an almost strict constraint to the topologically trivial
sector. For x ' 1 the offset has roughly the same size as typical fluctuations of Sfermion
in the sector with = 0; as a result of this one can say that in the intermediate regime
the functional determinant acts as a soft constraint to low-lying || values. For x 1
finally, hSfermion i1 hSfermion i0 is small compared to the fluctuations in each one of these
sectors; hence no such effective description proves appropriate in the large LS-regime.
Passing on to the slope of the regression line in the correlation hSfermion i versus hSgauge i,
we notice from Fig. 3 that this slope decreases (in general) with increasing ||. This means
that the effective renormalization

()
= 1 + Nf slope() ()
(21)
() eff
due to the functional determinant is (in general) stronger for sectors with low absolute
value of the net topological charge. This effect of sectoral dependence of the effective
renormalization factor for diminishes with increasing x; eventually, in the large x regime,
it is completely gone.
Thus, the large LS-regime is unique, as the determinant results only in an overall
suppression of higher topological sectors, but not in a sectoral dependence of eff (and
hence, in 4 dimensions, of the lattice spacing in physical units). This seems consistent with
the claim by Leutwyler and Smilga that physical observables do not depend on in the
large x regime, but it is of course not a sufficient condition for the latter to be true.
5. Sectoral heavy quark free energies

A quantity which is easily accessible in lattice studies and which might help to elucidate
the physical meaning of the LS-classification is the Polyakov loop, i.e., the trace of a chain
of link variables which winds once around the torus in the Euclidean timelike direction. Its
logarithm represents (up to a factor) the free energy of an external (heavy) quark which
is brought into the system without possibility to influence it through back-reaction. As in
the previous section the analysis shall be done sectorally, i.e., on classes of configurations
with a fixed value of ||.

431
Fig. 4. Scatter plot of the Polyakov loop on the complete sample (rightmost column) and on subsamples with || = 0, 1, 2, 3 in the small (top), intermediate
(middle) and large (bottom) LeutwylerSmilga regime.
432
The result of such an analysis for the Polyakov loop is displayed in Fig. 4. As one can
see from the rightmost column, the expectation value hLi on the complete (unseparated)
sample decreases if the box volume is increased. What is more interesting is the following:
the physical (unseparated) expectation value may be understood as a weighted mean
X
p|| hLi||
(22)
hLi =
||>0
of sectoral expectation values hLi|| , where the factor p|| reflects the combined weight
of the sectors in the histogram on the r.h.s. of Fig. 2. From the first line in
Fig. 3 one notices that the sectoral expectation values for || = 0, 1 in the small
LS-regime differ quite drastically on the scale set by the physical expectation value,
i.e., |hLi|0|,x1 hLi|1|,x1 |/hLix1 ' 0.7. In the intermediate regime the analogous
ratios for neighboring sectors (i.e., |hLi|0|, x'1 hLi|1|, x'1 |/hLix'1 and |hLi|1|, x'1
hLi|2|, x'1 |/hLix'1 ) happen to be smaller, but comparing the highest to the lowest available
||, one finds again |hLi|0|, x'1 hLi|2|, x'1 |/hLix'1 ' 0.7. Finally, in the large LS-regime,
there is an almost perfect consistency between sectoral Polyakov loop expectation values
for low-lying || (i.e., || {0, 1, 2}), but there is no numerical evidence in favour of
an overall consistency; if one builds the analogous ratio for the lowest and highest ||
available, the quantity |hLi|0|, x1 hLi|3|, x1 |/hLix1 is found to be again of order one.
The Polyakov loop is the first example of a physical observable for which we see
a difference between expectation values in neighboring sectors (for sufficiently low ||
values) disappear upon taking the large x limit. However, even in the large LS-regime no
supporting evidence has been found that hLi might be completely independent of . It
seems that hLi|| proves sensitive on topology even in the large LS-regime when jumping
from ||min to ||max .
6. Sectoral heavy quark potentials

The second observable which has been evaluated for each sector separately is the heavy
quark potential as determined from the correlator of two Polyakov loops see Fig. 5 and
Tables 13. The first thing to notice is that in each regime the physical potential (the one in
the rightmost column in Fig. 5) bends to the right. This curvature is a direct indication of the
screening brought by the dynamical fermions; the quenched potential is strictly linear. 13
In the small LS-regime the sectoral potentials for = 0 and = 1 differ quite
drastically. In this case the potential in the topologically trivial sector agrees (on a 1
basis, cf. Table 1) with the potential on the full sample as expected, since for x 1
the partition function is dominated by Z0 . What might come as a surprise is that the errorbars in the first plot in the top line of Fig. 5 are smaller than those in the rightmost one,
even though the configurations analyzed in the former case represent a subset of those
used in the latter case. The reason, as we have seen, is that the complementary subset (the
13 Unlike in QCD(4) there is no 1/r short-distance tail; the one-photon exchange contributes a linear piece in 2
dimensions.

433
Fig. 5. Physical heavy quark potential (rightmost column) and sectoral heavy quark potential on subsets with || = 0, 1, 2, 3 in the small (top), intermediate
(middle) and large (bottom) LeutwylerSmilga regime.
434
Table 1
Sectoral heavy quark potentials and unseparated (physical) potential at spatial separations 1, . . . , 3 in the small
LeutwylerSmilga regime (x 1)
|| = 0
dist = 1
dist = 2
dist = 3
0.119 +0.003
0.003
0.190 +0.009
0.009
0.224 +0.013
0.012
|| = 1
0.222 +0.028
0.025
all conf.
0.123 +0.003
0.003
+
0.793 0.224
0.201 +0.010
0.009
indet.
0.240 +0.016
0.015
Table 2
Sectoral heavy quark potentials and unseparated (physical) potential at separations 1, . . . , 4 in the intermediate LeutwylerSmilga regime (x ' 1)
dist = 1
dist = 2
dist = 3
dist = 4
0.121 +0.002
0.002
0.201 +0.004
0.004
0.250 +0.006
0.006
0.281 +0.012
0.011
|| = 2
0.206 +0.030
0.025
+
0.658 0.305
indet.
indet.
all conf.
0.128 +0.002
0.002
0.278 +0.013
0.012
0.319 +0.020
0.018
|| = 0
|| = 1
0.145 +0.005
0.005
0.266 +0.015
0.014
0.218 +0.008
0.007
0.388 +0.038
0.031
0.516 +0.125
0.071
Table 3
Sectoral heavy quark potentials and unseparated (physical) potential at spatial separations
1, . . . , 5 in the large LeutwylerSmilga regime (x 1)
|| = 0
|| = 1
|| = 2
dist = 1
dist = 2
dist = 3
dist = 4
dist = 5
0.128 +0.006
0.006
0.226 +0.021
0.017
0.291 +0.041
0.029
0.340 +0.054
0.035
0.387 +0.136
0.056
0.134 +0.007
0.006
0.259 +0.032
0.024
0.383 +0.137
0.056
+
0.444 0.119
+
0.478 0.136
0.130 +0.004
0.003
|| = 3
0.145 +0.022
0.018
all conf.
0.130 +0.003
0.003
0.229 +0.008
0.007
0.276 +0.120
0.053
0.233 +0.011
0.010
0.308 +0.023
0.019
+
0.357 0.126
0.310 +0.018
0.015
0.364 +0.028
0.022
+
0.457 0.223
0.366 +0.025
0.020
0.407 +0.057
0.036
indet.
0.417 +0.026
0.021
one comprising the configurations with = 1) provides measurement data which are
inconsistent with the data won from the topologically trivial subset, but has a total weight
which is too small to affect the complete ensemble average significantly (cf. the first plot
in Fig. 2). Hence, in the small LS-regime, the tiny contribution from the higher topological
sectors primarily adds some noise to a standard observable like the Polyakov loop. The
practical recipe to get rid of this effect is to do by hand (for sufficiently small x) what the
435
functional determinant does anyway in the limit x 0: cut away the higher topological
sectors!
In the intermediate LS-regime the difference between the two sectoral potentials for
= 0 and = 1 is smaller than in the previous regime, but the sectoral potentials are
typically several away from each other and even the potential in the topologically trivial
sector lies significantly below the unseparated physical potential (cf. Table 2). From this
we see that the sector with = 0 does not always yield the right result. The intermediate
LS-regime shares thus with the small x regime the uneasy property that different sectors
give inconsistent values for physical observables, but unlike in the small x regime, several
sectors yield sizable contributions to the complete (physical) expectation value hence
perfect ergodicity w.r.t. is mandatory.
In the large LS-regime differences between expectation values measured on subsets of
definite topological charge diminish further, if the latter differs only by one (neighboring
sectors). On the other hand, in the large x regime there is a multitude of topological
sectors yielding sizable contributions to the physical expectation value, hence an obvious
question is which effect will win in the limit x . Hence, what we would like to
know is whether the difference between the heavy quark potential on the subsets with the
lowest and the highest (reasonably populated) || will fade away, stay constant or grow if
x increases unlimitedly. From Table 3, one gets the impression that the difference between
the two extreme fixed-|| expectation values persists in the large LS-regime. It seems
that the statement by Leutwyler and Smilga that in the large x regime the topological
charge has vanishing influence on physical measurements is reproduced for sufficiently
low || values (in our case, e.g., || {0, 1, 2} but not necessarily for || = 3). It is also
worth noticing that the topologically trivial potential lies at least 0.5 below the physical
potential. Hence, a constraint to the topologically trivial sector seems permissible for
x 1, though in our simulation a constraint to the sectors || = 1 would have been a
better choice. Of course it is difficult to get definitive answers from numerical data; it may
well be that our LS-value (x ' 4.16) is still too small for the large x behaviour to be fully
pronounced. Nevertheless, the situation seems to be analogous to what we have seen in
the case of the heavy quark free energy: any differences between expectation values of
neighboring sectors (with sufficiently low ||) disappear upon taking the large x limit.
However, even in the large LS-regime no supporting evidence has been found that hVq q i
might be completely independent of .
7. Summary
The massive multi-flavour Schwinger model has been used to test the statements by
Leutwyler and Smilga about the (ir-)relevance of the topological charge for QCD in a finite
box. It is expected that an analogous study for full QCD will result in figures which look,
as far as qualitative issues go, exactly the same as the ones presented here and that, for this
reason, the statements below are valid for QCD too. The key points are the following:
436
(i) The three LS-regimes show distinctive properties already on a formal level, by the
way how the contribution per flavour to the total action (the sign-flipped logarithm of
the one-flavour functional determinant) correlates with the contribution from the gauge
field. Introducing the fermion determinant at fixed results in every regime in an
overall suppression of higher topological sectors w.r.t. low-lying ones. For x 1 this
suppression is so strong that the functional determinant effectively acts as a constraint to
the topologically trivial sector. In addition, the sea-flavours effectively result in a sectoral
multiplicative renormalization of with a factor greater than 1. For small and intermediate
x this factor is found to decrease as a function of ||, while in the large LS-regime the
renormalization factor seems to be uniform for all topological sectors.
(ii) For the overall distribution of topological charges the predictions by Leutwyler and
Smilga are well observed: in the regime x 1 the partition function (and hence the sample)
is entirely dominated by the topologically trivial sector, i.e., the histogram of topological
charges essentially consists of a delta-peak at = 0. In the intermediate regime (x ' 1)
the distribution is narrow but nontrivial, i.e., sizable contributions stem essentially from
the sectors = 0, 1, 2. In the regime x 1 the topological charge distribution gets
broad and is approximated by a Gaussian with width h 2 i ' x/Nf .
(iii) In the small LS-regime (x 1) the sectoral expectation value of an observable
in the first topologically nontrivial sector (typically won from very few configurations)
differs drastically from the analogous expectation value in the topologically trivial
sector (represented by the overwhelming majority of configurations). This means that a
simulation getting stuck at = 1 (or even higher) is absolutely disastrous to the result,
i.e., tiny non-ergodicities w.r.t. the topological charge render an overall sample average
meaningless. The good news is that in this regime the correct distribution is known and
hence there is a simple remedy: a strict constraint to the topologically trivial sector proves
beneficial.
(iv) The intermediate LS-regime (x ' 1) is characterized by the distinctive feature that
several mutually inconsistent sectors yield sizable contributions to a standard (i.e., not
relating to the U (1)A issue) physical observable. This means that the intermediate regime
is the one which is most delicate to simulate in: there is a strong sectoral dependence of
physical observables, yet there is no simple recipe how to guarantee the correct sectoral
weighting. In other words: physical measurements on a sample won in the intermediate
LS-regime do rely on the simulation algorithm having achieved perfect ergodicity w.r.t.
the topological charge.
(v) In the large LS-regime (x 1) sectoral averages seem consistent between
neighboring topological sectors (for sufficiently low ||), but no evidence has been found
that measurements might be completely independent of . It seems that jumping (within a
well-distributed sample) from the lowest to the highest accessible || may affect sectoral
averages even in the large x regime. This, if correct, means that the statement by Leutwyler
and Smilga according to which the topological charge is an irrelevant concept in the large
x regime [6] should to be interpreted in the following way: in the large LS-regime physical
observables are immune to modifications of the -histogram which are small compared
to its natural width (x/Nf )1/2 ; they may, however, be sensitive to distortions which go
437
beyond this limit. From a lattice perspective the implication is that the result of a full QCD
simulation in the large x regime may be trusted without hesitation as long as the algorithm
did not get stuck in a sector with unreasonably high ||, say at = 10 if the simulation was
in the regime x 40 and Nf = 2.
(vi) In this work, the transition from the small to the intermediate and the large LSregime has been made by adding more sites to the grid while keeping , m fixed. It would
be interesting to test the universality property, i.e., the claim that it is only the product
V m [6] which decides on the regime.
(vii) It is worth emphasizing that in the large x simulation used in this work the
predictions by Leutwyler and Smilga turned out to be fulfilled even though the condition
in the r.h.s. of (8) according to which the pion would overlap the box was not obeyed, but
rather the opposite relation 1/M L would hold true, i.e., the pion would fit into the
box, as is usual in a lattice context. Hence the r.h.s. of (8) seems to be a purely technical
condition, immaterial to the result of the LS-analysis in the large x regime.
(viii) From the fact that the massive multi-flavour Schwinger model follows the LSclassification even though it resembles QCD slightly above the chiral phase transition one
is led to suspect that this classification might be more general, even within QCD, than its
original derivation indicates. In other words: the conjecture is that the LS-classification
may prove useful not only in the broken phase but also in the high-temperature phase of
QCD as long as one stays in the immediate vicinity of the phase transition, i.e., as
long as pions keep being visible (i.e., keep dominating the long-range Greens functions
with pion-type quantum numbers; for a recent study see, e.g., [21]). This, if correct, would
imply that the version of the LS-parameter as given in Eq. (19) is more general 14 then the
original version (2).
(ix) The results of this investigation may be seen as an additional aposteriori justification
of the resources phenomenological lattice groups have been provided with in order to be
able to study (full) QCD in the large x regime. 15 In addition, they support attempts to
develop algorithms which decorrelate the topological charge faster than standard HMC
(see, e.g., [22]), as this is necessary for trustworthy simulations in the intermediate LSregime, towards which one moves when comparing different runs with increasing sea in
a fixed physical volume, or even more so, if is kept fixed. The fact that even in the
large x regime full QCD simulations may yield unreliable results upon generating extreme
distortions in the topological charge distribution should be taken as a motivation to monitor
in all phenomenological studies (i.e., not just with an observable relating to the U (1)A
issue) and, for the time being, as a warning against pushing the sea-quark mass too light 16
(even when increasing the volume together with sea so as to keep the product x = V m
fixed and large compared to 1). The latter constraint is certainly not a disaster, since there is
a well-elaborate framework [23] which allows one to extrapolate from samples generated
14 Note that in the high-temperature phase = 0, hence (2) is useless for T > T .
c
15 This is the result of a private attempt to estimate x in dynamical simulations aiming at the light hadron
spectrum from published data (e.g., msea and a in physical units).

16 Here it is assumed that each algorithm has a critical M /M (which may well depend on the actions used)
beneath which it tends to get stuck (cf. footnote 1).
438
with a somewhat larger sea-quark mass down to the physical M /M . Hence it seems that
our results do represent a little caveat but no real obstacle to future progress in Lattice
QCD.
Acknowledgements
I would like to thank Steve Sharpe for fruitful physics discussions in an early phase of
this work and Roland Rosenfelder for advice regarding presentation.
References
[1] T. Schfer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323.
[2] Y. Kuramashi et al., Phys. Lett. B 313 (1993) 425;
M. Mueller-Preussker, in: J.R. Sanford (Ed.), Proc. of XXVI Int. Conf. on High Energy Physics,
Dallas, TX, 1992, AIP Conf. Proc. 272 (1993) 1545;
B. Alls et al., Phys. Lett. B 359 (1996) 107.
[3] B. Alls et al., Phys. Rev. D 58 (1998) 071503.
[4] B. Alls et al., in: Proc. of ICHEP 96, Warsaw, 1996, p. 1599.
[5] M. Fukugita et al., Phys. Rev. D 51 (1995) 3952;
A. Ali Khan et al., CP-PACS, Nucl. Phys. Proc. Suppl. 83-84 (2000) 162.
[6] H. Leutwyler, A.V. Smilga, Phys. Rev. D 46 (1992) 5607.
[7] P.H. Damgaard, Nucl. Phys. B 556 (1999) 327.
[8] C. Vafa, E. Witten, Nucl. Phys. B 234 (1984) 173.
[9] T. Banks, A. Casher, Nucl. Phys. B 169 (1980) 103.
[10] S. Descotes, J. Stern, hep-ph/9912234.
[11] J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477.
[12] H. Joos, Helv. Phys. Acta 63 (1990) 670;
I. Sachs, A. Wipf, Helv. Phys. Acta 65 (1992) 652.
[13] S. Coleman, Commun. Math. Phys. 31 (1973) 259.
[14] A.V. Smilga, J.J.M. Verbaarschot, Phys. Rev. D 54 (1996) 1087.
[15] G. Segr, W.I. Weisberger, Phys. Rev. D 10 (1974) 1767.
[16] A.V. Smilga, Phys. Lett. B 278 (1992) 371.
[17] M.A. Shifman, A.V. Smilga, Phys. Rev. D 56 (1997) 7978;
S. Drr, Ann. Phys. 273 (1999) 1.
[18] A.V. Smilga, Phys. Rev. D 55 (1997) 443.
[19] M. Lscher, Commun. Math. Phys. 85 (1982) 39;
J. Smit, J.C. Vink, Nucl. Phys. B 303 (1988) 36;
J.C. Vink, Nucl. Phys. B 307 (1988) 549.
[20] C.R. Gattringer, I. Hip, C.B. Lang, Phys. Lett. B 409 (1997) 371;
C.R. Gattringer, I. Hip, C.B. Lang, Nucl. Phys. B 508 (1997) 329;
C.R. Gattringer, I. Hip, Nucl. Phys. B 536 (1998) 363.
[21] Ph. de Forcrand et al., hep-lat/0008005.
[22] E.M. Ilgenfritz et al., hep-lat/0007039.
[23] S. Sharpe, N. Shoresh, hep-lat/0006017.

Construction and analysis of anomaly-free

supersymmetric SO(2N)/U(N) -models
S. Groot Nibbelink , T.S. Nyawelo, J.W. van Holten
NIKHEF, P.O. Box 41882, 1009 DB, Amsterdam, The Netherlands
Abstract
This paper discusses a procedure for the consistent coupling of gauge- and matter superfields
to supersymmetric sigma-models on symmetric coset spaces of Khler type. We exhibit the
finite isometry transformations and the corresponding Khler transformations. These lead to the
construction of a generalized type of Killing potentials. In certain cases a charge quantization
condition needs to be imposed to guarantee the global existence of a line bundle on a coset space.
The results are applied to the explicit construction of sigma-models on cosets SO(2N)/U(N). Only
a finite number of these models can consistently incorporate matter in representations descending
from the spinorial representations of SO(2N). We investigate in detail some aspects of the vacuum
structure of the gauged SO(10)/U(5) theory, with surprising results: the fully gauged minimal
anomaly-free model is shown to be singular, as the kinetic terms of the quasi-Goldstone fermions
vanish in the vacuum. Gauging only the linear isometry group SU(5) U(1), or one of its subgroups,
can give a physically well-behaved theory. With gauged U(1) this requires the FayetIliopoulos term
to take values in a specific limited range. 2001 Elsevier Science B.V. All rights reserved.
PACS: 02.40.-k; 11.30.Na; 12.60.Jv
Keywords: Supersymmetry; Khler geometry; Coset space; Bundle; Anomalies
1. Introduction
N = 1 supersymmetry in 4D spacetime is likely to be a major ingredient of effective
field theories of fundamental interactions at energies below or near the Planck scale. When
restricted to theories at most quadratic in field gradients, the full spectrum of models
is characterized by the field content (e.g., the spectrum of scalar and vector multiplets),
which includes fixing the group of local gauge symmetries and their representations; and
furthermore by the choice of three functions of the scalar multiplets, the Khler potential,
E-mail address: t79@nikhef.nl (S. Groot Nibbelink).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 6 6 - 0
442
S. Groot Nibbelink et al. / Nuclear Physics B 594 (2001) 441476
the superpotential and the holomorphic kinetic functions of the gauge fields. Even with
such a restricted set of choices, a surprisingly rich variety of structures can be realized; as
yet their classification is far from complete.
In this paper we discuss supersymmetric -models on Khler coset spaces G/H . Such
models have been studied previously by various groups of authors [1]; for some reviews
see Refs. [24]. They have been considered in the context of non-standard superunification
models, of effective low-energy models for gauge theories in the strong-coupling limit,
or as models for string-inspired low-energy phenomenology. Supersymmetric -models
of various kinds are also part of supergravity theories. In N = 1 supergravity Khler type
models are of interest because they can realize many varieties of non-linear symmetries
on chiral fermions. In extended supergravity -models are a basic part of the theory, cf.
the non-compact models on SU(1, 1)/U(1) in N = 4, and on E7 (+7)/SU(8) in N = 8
supergravity in 4D spacetime.
Supersymmetry requires the target space of N = 1 scalar superfield theories in D = 4
spacetime to be a complex manifold of the Khler type [5]. For a coset model to be
Khler imposes special conditions on the groups G and H ; in particular, the stability
group H always factorizes so as to possess at least one commuting U(1) subgroup. The
more special symmetric cosets with such a structure include the Grassmannian models
on U(N, M)/U(N) U(M), the orthogonal unitary coset models on SO(2N)/U(N), as
well as models on exceptional cosets like E6 /SO(10) U(1). A non-symmetric model
of phenomenological interest is for example the supersymmetric version of E8 /SO(10)
SU(3) U(1).
By themselves, homogeneous supersymmetric coset-models are known to be inconsistent quantum field theories, because of the appearance of anomalies in the holonomy
group [6]. These can not be compensated by WessZumino type modifications [4]. A particular solution to this problem has been proposed in [7,8], involving a procedure known as
Goldstone boson doubling. This procedure takes a complexification of the broken isometry
group as the starting point for the construction of field-theory models. Alternatively, in [10]
it was proposed to cancel such anomalies by coupling additional (chiral) matter superfields
in non-trivial representations of the isometry group of the -model. More details of this
procedure have since been worked out in [11,12,14]. We wish to stress, that although the
two approaches have rather different starting points, they are not mutually exclusive.
Continuing our line of investigation, in this paper we describe several new results.
First, we perform a quite general analysis of the global aspects of the geometry and
isometries of Khler-type coset manifolds. From the results we derive a better and
more detailed understanding of the consistency conditions on the bundles which can be
constructed over such manifolds. As these bundles can be interpreted as target spaces of
fields coupled to the -model, the consistency conditions have direct implications for the
existence of interacting field theories constructed on the basis of pure coset models. We
apply the results of this analysis to the particular case of symmetric orthogonal unitary
cosets on SO(2N)/U(N). We show that only a finite number of these models can be
consistent when coupled to matter superfields with U(N) quantum numbers reflecting
spinorial representations of SO(2N). Among these are in particular the ones based on
443
SO(10)/SU(5) U(1), with matter in representations descending from the 16 of SO(10),

which are interesting for phenomenological applications.
This paper is structured as follows. In Section 2 we review some basic aspects of
Khler geometry and its role in supersymmetric scalar field theories in D = 4 space
time. We discuss the symmetries of these models in the geometric language of Killing
vectors (generating isometries), which represent infinitesimal, but generally non-linear,
transformations. A procedure for coupling chiral superfields in other representations of the
isometry group, first described in [911], is reviewed emphasizing its role in anomalycancelation. In Section 3 we present the construction of non-linear realizations of the
SL(N + M; C) starting from the approach of Ref. [15]. By imposing certain constraints
on the group elements one obtains non-linear representations of various classical groups,
like SU(N + M), SO(2N), or USp(2N) and their non-compact relatives, in finite form.
We discuss the realization of the non-linear transformations on various types of bundles
over the manifold, and examine the consistency conditions to be satisfied. In the context
of field theory, this results in the quantization of U(1) charges for matter fields coupled
to the -model. In Section 4 we turn specifically to non-linear realizations of SO(2N).
The bundles of interest for supersymmetric field theory applications are constructed. We
use the conditions for existence of these bundles to examine the possibility of cancelling
anomalies in Subsection 4.5. We identify U(N)-bundles over the SO(2N)/U(N) cosets
which together build a spinor representation of SO(2N). It is shown that only a finite
number of spinor models can be made anomaly-free. The line bundle does not appear
explicitly in these models, in particular the SO(10)/U(5) spinor model, because it is used
internally as a U(1)-compensator to construct the necessary well-defined physical matter
representations. We finish in Section 5 by discussing a number of physical aspects of the
model on SO(10)/U(5), like internal and supersymmetry breaking. A rather surprising
result, which generalizes to other coset models, is that the model with fully gauged SO(10)
is singular: the kinetic terms of the Goldstone superfields vanish in the vacuum. Gauging
the linear subgroup U(5) can give consistent models, but only in a range of non-zero values
of the FayetIliopoulos term. Finally, the appendices describe some mathematical details
of our constructions.
2. -models on Khler manifolds

The kinetic terms of complex scalar fields (z , z ) on Khler manifolds are of the form
(1)
Lkin = g g G (z, z ) z z ,
with the spacetime metric g and the target-space metric subject to the Khler condition
G , = G , .
(2)
Locally on the target manifold the condition is satisfied by deriving the metric from a
Khler potential:
G = K, .
(3)
444
In terms of the Khler two-form

(K) = i K, d z dz ,
(4)
Eq. (2) can be written as

d(K) = 0.
(5)
Obviously, the Khler potential is defined by this equation only up to holomorphic terms:
S(z).
e z ) = K(z, z ) + F (z) + F
K(z,
(6)
As a result, if two complex local coordinate charts {zi } and {zj } have non-empty overlap,
the Khler potentials in the charts are generally related by
S(ij ) (zj ).
Ki (zi , z i ) = Kj (zj , z j ) + F(ij ) (zj ) + F
(7)
In this paper we focus on the class of Khler manifolds formed by coset spaces G/H . Such
cosets are Khler manifolds if H is the centralizer of a torus in G; that is, the stability
group H contains one or more U(1) subgroups commuting with the rest of H . A general
procedure for constructing a Khler potential for these coset manifolds was developed by
the authors of Ref. [15]. It is discussed in some detail in the context of our applications
below.
Scalar lagrangeans of the type (1) can be extended to incorporate N = 1 Poincar
supersymmetry, by taking the complex fields z to be the scalar components of chiral
superfields ; we denote its chiral fermion components by L . In Minkowski space, the
component lagrangean is [5]
Z

d 4 K ,
Lchiral =
h
i 1
/ L + R L L L L , (8)
= G (z, z ) z z + L D
4
where the covariant derivative and curvature tensor are those of the Khler manifold.
In general, Khler metrics admit a set of holomorphic isometries Ri (z), with conjugates
S (z), satisfying the Killing equation

R
i
Si , = 0.
Ri , + R
(9)
These isometries define infinitesimal symmetry transformations on the manifold:

z = z0 z = i i z = i Ri (z),
(10)
with i the parameters of the infinitesimal transformations. As a result, the isometries

define a Lie algebra with structure constants fij k via the Lie derivative by:

Rj Ri,
= fij k Rk .
(11)
LRi [Rj ] = Ri Rj,
The invariance of the metric implies, that under these isometries the Khler potential
generally is invariant modulo holomorphic functions, as in Eq. (6):
Si (z).
i K = Fi (z) + F
(12)
445
From the Lie-algebra property (11) it follows that one can choose the transformations of
the functions Fi (z) to have the property
i Fj j Fi = fij k Fk .
(13)
Eq. (9) for holomorphic Killing vectors has a local solution in terms of a set of scalar
potentials Mi (z, z ), transforming in the adjoint representation of the Lie-algebra (11):
iMi = K, Ri Fi ,
i Mj = fij k Mk .
(14)
Supersymmetry generalizes the isometries to transformations of the chiral superfields :

i = Ri ().
(15)
For the chiral fermions this implies the infinitesimal transformation rule
(z) L .
i L = Ri,
(16)
In Section 3 we present the finite form of the transformations (15), of which (10) and (16)
are special cases, for a large class of symmetric coset spaces G/H .
As the chiral fermions couple to the connection and the curvature in Lchiral , the
consistency of the quantum theory is generally spoiled by anomalies [2]. Therefore, we
extend the model with additional chiral superfields generically called matter superfields
on which the isometry group is realized, with the representations chosen to cancel the
anomalies.
A general procedure for matter coupling has been worked out in [911]; the generalization to supergravity was presented in [12]. The mathematical framework used to construct
matter representations of the isometry group of the Khler manifold is the theory of complex bundles over Khler manifolds. These bundles are defined locally on the Khler manifold by sets of complex fields with specific transformation character under the isometries.
The basic pattern is that exhibited by the transformation rule (16) for the chiral fermions.
This rule shows how a vector (an element of the tangent bundle) transforms under the
isometries. Similarly, one can define a representation transforming as a 1-form (an element
of the co-tangent bundle):
i v = Ri, (z) v .
(17)
More general transformations are obtained by taking tensor products of the tangent or cotangent bundles. However, for our applications this is not sufficient. The reason is, that the
U(1) charges of such representations are completely fixed in terms of the charge of the
scalars z : a contravariant holomorphic tensor t 1 p of rank p carries a relative charge p,
whereas a covariant holomorphic tensor s1 k of rank k carries a relative U(1) charge k.
But in actual models, if one requires anomaly cancellations with a phenomenologically
interesting set of matter superfields, one usually needs a different assignment of U(1)
charges. Therefore, the spectrum of representations must be extended with bundels which
differ from tensor bundles by the assignment of U(1) charges. This is achieved for instance
by the introduction of complex line bundles [11].
A line bundle is the target space of a single-component complex scalar field over the
manifold. We consider line bundles carrying non-trivial representations of the isometry
446
group; these can be defined locally on the Khler manifold as complex scalar matter fields
S(x) coupled to the -model, with the infinitesimal transformation law given by
i S = Fi (z)S.
(18)
In the context of supersymmetric field theories such a representation of the isometry group
was introduced in [10], and subsequently considered in [13]; it is a representation because
of the property (13). From the line-bundle S one can obtain other line bundles with different
U(1) weights by taking powers:
A S
i A = Fi (z)A.
(19)
Furthermore, using the line bundle construction, one can modify the transformation rules
of fields in tensor representations of the isometry group. For example, defining
T 1 p S t 1 p ,
(20)
the new field T obeys the transformation rule

i T 1 p =
p
X
Ri,k T 1 p + Fi T 1 p .
(21)
k=1
In this way the U(1) charges can be adjusted, be it subject to the charge quantization
conditions mentioned above.
However with the introduction of the line-bundle we still have not exhausted all
possibilities for consistent non-linear realizations of symmetries over Khler manifolds.
Some coset spaces allow factorization of the Goldstone-boson transformations and the
Khler metric. Then one can define sub-bundles of the tangent-space bundles, and their
line-bundle extensions as well. A general description of matter representations that can
be associated with coset spaces can be found in Refs. [2,22]. Examples of this type of
structures are presented below.
The bundles introduced here are characterized locally on the Khler manifold by their
transformation properties. An important question is, if these definitions can be extended
globally over the manifold. This is always possible for tangent and co-tangent bundles.
However, for line bundles (18), this requires in particular that the holomorphic transition
functions introduced in (7) satisfy the cocycle condition
F(ij ) + F(j k) + F(ki) = 2iZ.
(22)
Manifolds with this property are known as KhlerHodge manifolds [23]; their Khler
forms satisfy the condition
Z
(K) = 2Z,
(23)
C2
for any closed two-cycle C2 .

The existence of the generalized line bundles (19) and (21) often requires the powers
to satisfy certain integrality conditions: there is a minimal line bundle which by Eq. (22) is
globally defined and single-valued, and all other line bundles carry integral charges w.r.t.
447
the minimal line bundle. Thus it follows that the U(1) charges of fields transforming as
line bundles are quantized [14]. These consistency requirements are discussed in detail in
Section 3 below.
3. Non-linear realization of SL(N + M, C)

In this section we discuss the method developed by Bando, Kuramoto, Maskawa and
Uehara (BKMU) [15] to obtaining non-linear transformations and Khler potentials, but
here we consider more general transformations of the complex coordinates and we discuss
matter coupling in detail. The basis of our construction is to define a transformation rule
for a complex (M N )-matrix z under the action of an arbitrary element of special linear
group SL(M + N; C). It will become clear below, why we restrict ourselves to the special
linear group. As the special linear group contains all (classical) Lie groups as subgroups,
this construction can be used to obtain Khler potentials for Khlerian coset spaces based
on these groups. It explains why the transformation rules for the different coset spaces
based on classical groups are much alike. Given a coset the group of isometries is fixed,
but we can still use the full SL(M + N; C) transformations to determine the effect of
coordinate redefinitions. In particular for a non-compact coset this allows us to interpolate
between two seemingly different representations of its Khler potential.
Let g SL(M + N; C) be an arbitrary element of the special linear group and g 1 its
inverse, we write
a

a

,
(24)
g=
and g 1 =
a a

where , , and are (M M)-, (M N )-, (N M)- and (N N )-matrices respectively.
The submatrices of the inverse g 1 are given by
1
1
,
a = 1 ,
a = 1
1 1
1
= 1 1 ,
a = 1
1
1
.
(25)
a = 1 1 = 1 1
To obtain the infinitesimal transformations, one considers infinitesimal deviations from the
unit element of SL(M + N; C)

1 u y
1+u
y
,
(26)
g=
and g 1 =
x
1+v
x
1v
where u, v, x, y are infinitesimal submatrices and the minus in front of the v is useful
later. However in the following we are primarily concerned with finite transformations.
A non-linear realization is found by defining the matrix (z) like the BKMU-parameter by

1 0
(z) =
,
(27)
z 1
and requiring that
448
(z) z = g(z)h
g
(z; g),
1
h+
with h =
0

h 0
.
h
(28)
We have written (h + )1 instead of h + in the matrix h for later convenience, at this stage
this is merely notation. We find that z transforms as
1 a

g
z = ( + z)( + z)1 = a z a
z a .
(29)
Under the action of g and the matrix h takes the form

1
h0
+ z

h(z; g) = h+
=
1 .
0
a z a
0
h
(30)
Notice that it follows from the transformation rule of z that general linear transformations
have the same effect as special linear transformations. Under the composition of two
transformations g 0 and g we find using (28) that the non-linear transformation (29) respects
0
0
this composition g(g z) = g g z and furthermore we find that

(31)
h (z; g 0 g) = h g z; g 0 h (z; g) and h + (z; g 0 g) = h + (z; g)h + g z; g 0 .
In the following we employ two projector operators defined by

1 0
0 0
and =
.
+ =
0 0
0 1
(32)
+ and h ' h = h
,
+ = + h
These definitions allow us to write (h + )1 ' h
where the symbol ' denotes equality of the left-hand side as the unique non-vanishing
submatrix of the right-hand side.
Now let J SL(M + N; C) be a fixed matrix; its properties we develop along the way.
We define the (M M)-matrix function of z, z by
J1 (z, z ) + (z )J(z)+ = A + Bz + z C + z Dz,
(33)
and obtain the transformation property

)h + (z; g).
J1 (z, z ) J1 g z,g z = h + (z; g) g1
Jg (z, z
(34)
Define the subgroup SLJ (M + N; C) consisting of elements g SL(M + N; C) that leave

J invariant

a

A B
A
Ba
1
,
J
.
(35)
g Jg = J
C D
C a Da
Hence if g SLJ (M + N; C), the function
KJ (z, z ) = ln det J1 (z, z )
transforms as a Khler potential

S(z; g),
KJ (z, z ) KJ gz,g z = KJ (z, z ) + F (z; g) + F
with
(36)
(37)
F (z; g) = ln det h + (z; g),

S(z; g) = ln det h + z ; g .
F
449
(38)
If we want to interpret KJ as a Khler potential, KJ has to be a real function KJ (z, z ) =

(KJ (z, z )) . This only happens iff J is Hermitean J = J. The composition rule for F
follows directly from Eq. (31)

(39)
F (z; g 0 g) = F (z; g) + F gz; g 0 .
We define a real finite Killing pre-potential M(z, z; g 0 ) by

0
0
S(z; g 0 ).
2i M(z, z; g 0 ) = K z,g z K gz, z + F (z; g 0 ) F
(40)
It is a function of the group element g 0 , of which the infinitesimal (linearized) form

reproduces the standard Killing potentials (14). Using the transformation property of the
Khler potential (37) together with the composition property (39) of F , it follows that
M(z, z; g 0 ) transforms in the adjoint representation

(41)
M gz,g z ; g 0 = M z, z ; g 1 g 0 g .
Again, inserting a group element close to the identity, we obtain for the Killing potentials
the infinitesimal transformation rule (14).
The metric associated with KJ can be written as

(42)
G d z dz = tr J d z J dz ,
where we define J in analogy of J in (33)
J1 (z, z ) (z )J(z)
1
= D a C a z zB a + zAa z .
(43)
This can be shown either by a direct calculation of the metric in the standard way as the
second mixed derivative or by first proving this for a block-diagonal J and showing that
the diagonalization procedure has no effect on the metric (42). This is easy as under the
action of g SLJ (M, N) the differential dz transforms as
1
dz( + z)1 ,
(44)
dz g(dz) = h (z; g) dz h + (z; g) = a z a
and J , J transform as Eq. (34) and as

J1 (z, z ) J1 gz,g z = h (z; g)J1 (z, z )h (z; g).
(45)
Hence it follows that (42) is invariant.

Until this point the matrix J used in the definitions (33) and (43) of J and J can be
any Hermitean matrix of SL(M + N; C). However if we want to use the invariants (58) as
Khler potentials for supersymmetric model building, the resulting kinetic terms have to be
positive definite. By going to the unitary gauge (z = zT = 0), we infer that both A and D a
have to be sign definite. (Of course an overall sign can be compensated by an appropriate
minus sign.) On the other hand using a unitary transformation, we can diagonalize J
with real eigenvalues i . If this is followed by an appropriate scale transformation of the
coordinates and possibly some relabeling, we bring the matrix J into the canonical form
450

J=

1 0
,
0 1
= 1.
(46)
This shows that we can restrict SLJ (M + N; C) to SU (M, N) when we want to study
the isometries of the metrics J , J or the Khler potential KJ . Here = +1 refers to
the compact special unitary group SU(M + N), while = 1 refers to the non-compact
version. We assume from now on that we have chosen this canonical form of J and consider
SU (M, N) only. Notice that by putting further restrictions on the group elements g we can
reduce the isometry group to a subgroup of SU(M, N), such as SO(2N) or USp(N). The
form of the metrics and Khler potential does not change under this; they always take the
form
(z, z ) = (1 + zz)1 ,
(z, z ) = (1 + zz)1 ,
K (z, z ) = ln det 1 = ln det 1 ,
(47)
in the canonical basis. However this leads to restrictions on the coordinates z as we see
later: that is the coordinates z parameterize a submanifold of SU (M, N)/S[U(M)
U(N)]. Even though the SL(M + N; C) group is not the isometry group, it is still
worthwhile to know its action on the fields, as it can be used to describe field redefinitions.
We give an example of this now. In the previous analysis we used that we can set B
and C in the matrix J to zero by a unitary transformation. Sometimes we can also do the
opposite: set A and D to zero. To analyze the situation we start with J in the canonical
form and perform an arbitrary transformation g of SL(M + N; C) on it

0
+
+
1
g=
.
(48)
g
0 1
+
So to remove the A and D entries of this matrix we need to have that
+ = 0
+
= 0.
and
(49)
Notice that there is no solution g SL(M + N; C) of these equations when = 1. On the

other hand in the case = 1 with M = N we can use

i
1 1
,
(50)
g=
2 1 1
to bring J into the form

0 1
J=
.
1 0
(51)
Using this matrix J we obtain a Khler potential

Kno-sc = ln det(z + z )
(52)
of the no-scale type [20]. The low energy effective actions for the moduli sectors of string
theory often take this form, see for example [2527].
Matter coupling is the next topic we discuss. As we want to interpret SU (M, N) as the
symmetry group of the models we construct, this implies that all matter representations
451
should be well defined representations of SU (M, N). To obtain a section of the tangent
bundle, we define the transformation of the tangent space vector T in analogy of (44) by
1
g
T = h (z; g)T h + (z; g) = a z a
T ( + z)1 .
(53)
A section C of the cotangent bundle transforms as
1
1

g
C = h + (z; g) C h (z; g)
= ( + z)C a z a .
(54)
When we take g SU (M, N), we obtain the following invariants for the sections of the
tangent and cotangent bundles

S 1 C .
and tr (J )1 C
(55)
tr J TSJ T
J
Next we construct subbundles of the tangent bundle. To do this we notice that the
transformation rule (44) for the differential dz factorizes [14,22]. Using this we define
the sections L and R by the transformation rules
1
g
L = h (z; g)L = a z a L and gR = R h + (z; g) = R( + z)1 . (56)
To show that these transformations do indeed define consistent bundles we proceed as
follows. All manifolds we consider here are submanifolds of Grassmannian coset-space
SU (M, N)/S[U(M) U(N)]. As this is a homogeneous space we can reach any point on
it by a transformation using a group element g SU (M, N). Therefore, we can describe
all coordinate transformations as actions of elements of SU (M, N), and the holomorphic
transition functions on overlapping complex coordinate charts for the bundle of which L is
a section are given by elements h (z; g). The global consistency conditions for this bundle,
mentioned in Section 2, then take the form

h gz; g 1 = h (z; g)1 ,
h (z; e) = 1,

(57)
h g2 g1z; g3 h g1z; g2 h (z; g1 ) = 1,
when g1 g2 g3 = e, being e the SU (M, N) identity. The composition property (31) of two
group elements show that these conditions are satisfied. Using the metric of the tangent
bundle (42), which factorizes as well, we obtain the following SU (M, N)-invariants
S
L and R R.
L
J
J
(58)
We will discuss tensor products of these types of matter representations extensively when
we consider matter coupling to SO(2N)/U(N).
Until this point our discussion was general, in the sense that we only demanded that
we construct isometries of the metrics J and J without any reference to a particular
coset space. We saw that we only obtain isometries of these metrics if we restrict the
transformations to be unitary g SU (M, N). It is now easy to describe non-linear
realizations of (classic) groups, that are subgroups of SU (M, N). For this we only have
to describe what the group and algebra of the groups look when embedded in the unitary
group SU (M, N). We have summarized our results in Table 1. We describe the ingredients
of this table which are partly taken from Ref. [21]. A complete classification of Khler
452
Table 1
This table gives an overview of the (classical) Lie-groups that can be embedded into SU (M, N).
With the parameter we distinguish between compact ( = 1) and non-compact ( = 1) groups.
For these Lie-groups the non-linear SL(M + N; C) transformation rules given in this section can be
used directly. P = gcd(M, N) is defined a the greatest-common-divisor of M and N . When the U(1)
is not diagonal, we have to perform a special unitary transformation to make it diagonal; when doing
so the transposition properties may change. The Hermitean form of an element of the algebra after
possible diagonalization is denoted by aD . The matrices u, v, x are all taken to be complex, their
additional properties are given in second last row in the table. The last row summarizes the symmetry
properties of the coset coordinate matrices
Group G
SU (M, N )
USp (N, N )
Sp(2N )
SO(2N )
Compact subgroup H
S[U(M) U(N )]
U(N )
U(N )
U(N )
SL(2N ; R)
gT K g = K =
SL(2N ; C)

1 0
0 1

0 1
1 0
SL(2N ; R)
g Jg = J =
SL(M + N ; C)

1 0
0 1
U(1) embedding
ei
0
TK g =K =
gD
D D
D
M
ei P
i
e
0

gD = eaD , aD =
N
P
u
x
x
v
0
1

ei
1
0
u x
x uT
0 1
1 0
cos
sin
sin cos

0 1
1 0

u x
x uT

1
0
0
1
cos
sin
sin cos

0 1
i
1 0

u x
x uT
Restrictions
u = u, v = v
tr u = tr v
u = u, x T = x
u = u, x T = x
u = u, x T = x
z G/H , zij C
zT = z
zT = z
zT = z
cosets can be found in Ref. [16]. A discussion on SO(2N)/U(N), Sp(2N)/U(N) cosets

can also be found in Refs. [1719].
The classic groups are either real or complex groups that satisfy certain Hermitean
conjugation and transposition properties
g Jg = J
and g T K g = K,
(59)
where J and K are fixed matrices. We discriminate between the unitary (SU), orthogonal
(SO), symplectic (Sp) and unitary symplectic (USp) groups. Furthermore, with = 1
we make a distinction between compact ( = 1) and non-compact ( = 1) groups.
We require the maximal subgroups H of these groups to have a compact U(1)-factor.
For example, we do not consider the non-compact SO(N, N) here, as the non-compact
abelian SO(1, 1) subgroup corresponds to Lorentz transformations that are not bounded.
A compact U(1)-factor is needed to ensure that the resulting coset-space is Khler; because
of its importance we give the U(1) embedding explicitly. For the real groups SO(2N) and
Sp(2N) the U(1) is not realized in a diagonal way. By making a similarity transformation
453

1 1 i
and V =
,
(60)
2 i 1
using the unitary matrix V , the U(1) is turned into a diagonal form. Here the subscript D is
used to indicate that gD , for example, is considered in the basis where the U(1) is diagonal.
As V is unitary, gD has the same unitary properties as g. However the transposition
properties may change
T
T
KD gD = KD = V K V .
(61)
gD
gD = V gV ,
g = V gD V
For this it is crucial that we have embedded the real groups Sp(2N) and SO(2N) in the
special unitary group SU(N, N) and SU(2N) respectively; else the multiplication with i
has no meaning. In the remainder we work in the basis where the U(1)-factor is diagonal.
We can now represent any element of any of these groups as a unitary matrix gD = eaD , that
is obtained by exponentiating an anti-Hermitean algebra element aD . The group definition
properties (59) can be written down for the algebra elements aD as well
= JaD J1
aD
T
and aD
= KD aD K1
D .
(62)
Using these properties it is possible to give a unique representation of the algebra

elements aD . For the different groups we give this representation in the row of gD
in Table 1. Notice that algebra elements of Sp(2N) and USp(N, N) have the same
representation in the basis where the U(1) is diagonal; therefore their corresponding
cosets are isomorphic. From this representation of the algebra, it is easy to see what the
restrictions are on the coset coordinates z for the different coordinates. For the non-compact
coset, the coordinates in addition satisfy tr(zz ) < 1 for the Khler potential and metrics in
(47) to be well defined. Notice that for the USp, Sp and SO cosets the submetrics ,
are each others transposed = T .
We now turn to the construction of the minimal complex line bundles. In this discussion
we have to make a distinction between the different coset spaces as becomes clear below.
Our discussion here is complementary to Ref. [14] where general results have been
presented, which we apply here to the particular cosets discussed in this section.
A section S of a complex line bundle can be defined to transform as
S = det h + (z; g)S = det h (z; g)S.
(63)
Here we have used that det h + = det h , which follows from (28) since g SU (M, N).
The consistency of this complex line bundle follows directly from (57) and the properties
of the determinant. To show that we have obtained the minimal line bundle in the compact
situation, we have to show that the integral over the corresponding Khler form
Z
(K) = 2 n, with n = 1,
(64)
C2
when integrated over a generating two-cycle C2 .

We first turn to a Grassmannian coset SU(M + N)/S[U(M) U(N)]. Let v be the
complex coordinate of the stereographic projection of the complex projective line CP 1 .
We define a generating two-cycle by the embedding of CP 1 in the coset by taking all the
454
coordinates zij zero except for one which is equal to v. Now since the Khler potential
= KCP 1 (v, v),
restricted to this embedding to CP 1 is given by K(z, z )|CP 1 = ln(1 + vv)

R
1
which is the Khler potential of CP that satisfies CP 1 (KCP 1 ) = 2 , it follows that we
have obtained a minimal line bundle. Next we discuss the compact USp(2N)/U(N) and
SO(2N)/U(N) coset spaces. The coordinates of these spaces satisfy zT = z, respectively,
zT = z, see Table 1. Therefore, it is not possible to set all coordinates to zero except for
one, except when one takes this symmetrization into account: z = zT = v. Hence we find
= 2KCP 1 (v, v),
so that n = 2 in Eq. (64).
in these cases that K(z, z )|CP 1 = 2 ln(1 + vv)
This implies that the section S is the square of the minimal line bundle. Since the Khler
potential of a coset is unique up to a normalization factor, it follows that a section of a
minimal line bundle over USp(2N)/U(N) or SO(2N)/U(N) is given by
1/2
1/2
g
S = det h + (z; g)
S = det h (z; g)
S.
(65)
The only possible ambiguity for a global definition resides in the square root, it can be
removed by using the BKMU-construction with the representation with highest weight
that has all its Dynkin label zero except for the N th one [14].
We now determine the relative charges of the coordinates z, the matter fields L and R,
and the sections of the minimal line bundles, using the U(1) embedding presented in
Table 1. We first discuss the Grassmannian cosets and after that the cosets USp(2N)/U(N)
and SO(2N)/U(N). The U(1)-factor in SU (M, N), that is not in SU(M) SU(N), can
be given by

iN/P
e
1
0
,
(66)
u =
0
eiM/P 1
where P = gcd(M, N) is the greatest-common-divisor of M and N . The smallest period of
this U(1) is = 2 , since the integers N/P and M/P are relatively prime by construction.
It follows that the coordinates z have charge (M + N)/P in this normalization. For the
matter couplings L and R we find the charges N/P , respectively, M/P . The section
of the minimal line bundle has a charge MN/P . For the cosets USp(2N)/U(N) and
SO(2N)/U(N) we always obtain integer charges when we choose a slightly different
normalization for u given by

i2
e
1
0
.
(67)
u =
0
ei2 1
In this case L and R have the same charge 2 and the section of the minimal line bundle has
charge N , while the charge of the coordinates is 4.
4. SO(2N )/U(N ) coset models

We discuss supersymmetric models build using the Khler geometry of the coset
space SO(2N)/U(N). For this we first discuss the decomposition of the SO(2N) algebra
into U(N) representations and the vector representation of SO(2N). We discuss the
construction of the Khler potential using the the general BKMU method and relate that
455
to our discussion on special linear transformations of Section 3. Next we discuss matter

representations that can be coupled to the supersymmetric -model of the coset in a
consistent way, heavily relying on the non-linear transformation discussed in Section 3.
For applications to a chiral spinor representation of SO(2N) later in this article we confine
ourselves to the construction of completely anti-symmetric tensor representations with an
arbitrary rescaling charge. We discuss their transformation properties and their invariant
Khler potentials that can be used in supersymmetric model building. Some relevant results
and conventions have been collected in the appendices.
4.1. SO(2N) algebra in a U(N) basis
In this section we discuss how the algebra of SO(2N) can be decomposed into SU(N)
U(1) representations. We split the SO(2N) generators Mab into SU(N) generators T i j ,
Sij which are anti-symmetric
a U(1)-factor generator Y and broken generators Xij , X
tensors of SU(N). We first discuss the embedding of U(N) in SO(2N), then we discuss
the vector representation; the spinor representation is discussed in Appendix A.
The 2N(2N 1)/2 anti-Hermitean generators Mab = Mba of SO(2N) satisfy the
commutation relations
[Mab , Mcd ] = ac Mdb bd Mac ad Mcb + bc Mad .
(68)
We denote the N 2 generators of U(N) by U i j (i, j = 1, . . . , N). The remaining

N(2N 1) N 2 = N(N 1) generators form two anti-symmetric tensor representations
Sij , each of dimension N(N 1)/2. The U(N) generators satisfy the
of U(N): Xij and X
algebra

i
(69)
U j , U k l = i l U k j k j U i l .
We decompose the SO(2N) algebra w.r.t. U(N) by writing the SO(2N) generators Mab
using indices i, j = 1, . . . , N as

1
Sij U i j + U j i ,
Xij X
2

i ij S
X Xij U i j U j i ,
Mi j +N =
2

1 ij S
X + Xij U i j + U j i .
Mi+N j +N =
2
Sij as U i j = Ai j + iS i j with
Inversely we can express U i j , Xij and X
Mij =
1
1
S i j = (Mi j +N + Mj i+N ),
Ai j = (Mij + Mi+N j +N ),
2
2
Sij = iQij P ij with
and Xij = iQij P ij and X
(70)
(71)
1
1
Qi j = (Mi j +N Mj i+N ).
(72)
P i j = (Mi j Mi+N j +N ),
2
2
The U(1)-factor generator Y in U(N) is defined as minus twice the trace of the U(N)
generators
456
Y = 2
N
X
U i i = i2S i i = 2iMi i+N
(73)
and the remaining SU(N) generators T i j are define as the traceless part of U i j
T ij = Uij +
1
Y i j .
2N
(74)
Sij the SO(2N) algebra

Using the U(N) generators U i j and the broken generators Xij and X
(68) takes the form
ij kl

i
Skl = 0,
Sij , X
X ,X = X
U j , U k l = i l U k j k j U i l ,

Sij , Xkl = k i U l j l j U k + l i U k j + k j U l ,
X
i
i
i

i
i S
i S
kl
l
S
U j , X = j Xik k j Xil .
(75)
U j , Xkl = k Xj l l Xj k ,
The closure of the algebra can be checked explicitly by computing the Jacobi identities.
The SO(2N) generators in this basis carry the following U(1)-charges:

Sij = (0, 0, 4, 4).
(76)
U(1)-charges of Y, T i j , Xij , X
Here we have chosen the U(1)-charges such that they match the convention of Slansky
[28].
4.2. The vector representation of SO(2N)
In the vector representation of SO(2N), the generators Mab take the form: (Mab )cd =
ac bd bc ad , therefore an element of the SO(2N)-algebra reads

a s
p q
ij
ij
ij
ij
+
,
(77)
= a Aij s Sij + q Qij p Pij
s a
q p
where a, p, q are N N real anti-symmetric matrices and s is a real symmetric N N
matrix; these matrices define the parameters of the SO(2N)-algebra elements. Here we
have used the definitions of the algebra elements A, S, P and Q given in Eqs. (71) and
(72). The U(1)-factor generator Y (73) in the vector representation takes the form

0 1
Y = 2i
.
(78)
1 0
Notice that the U(1) generator Y is off-diagonal. However it is more convenient to use a
basis in which Y is diagonal. Using a unitary transformation we can diagonalize Y :

1
1 0
1
i1
with V =
.
(79)
YD V Y V = 2
0 1
1
2 i1
We use the subscript notation D on any (2N 2N )-matrix A to indicate that A is evaluated
in the basis where Y is diagonal. The effect of this similarity transformation on an element
of the SO(2N) Lie algebra (77) is given by

a is q ip
u x
=
,
(80)
D = V MV =
q + ip a + is
x uT
457
where u = u = a is, uT = a + is, x = q + ip and x = q + ip. This is coincides

with the SO(2N)/U(N) entry in Table 1. Notice that in the basis where Y is diagonal, the
defining property g 1 = g T of SO(2N) becomes

0 1
1
T
.
(81)
gD = KgD K with K
1 0
Writing gD in terms of the submatrices , , and as introduced in Eq. (24) this group
property can be stated as
T

a
a
T
=
,
(82)
a a
T T
using the notation (26) for the inverse of gD . From now on we will only work in the basis
where the U(1)-charge Y is diagonal, dropping the subscripts D.
4.3. Khler and Killing potentials
We now construct the Khler potential for the coset spaces SO(2N)/U(N) using the
BKMU-method [15]. We apply their method to the 2N -dimensional vector representation of SO(2N). The BKMU-projection + projects (32) on the part of this vector representation with positive Y -charge, which is an N -dimensional vector representation of
b(N), with SOC (2N), the comSU(N). The coset spaces SO(2N)/U(N) and SOC (2N)/U
b(N) is defined as the group generplexification of SO(2N), are isomorphic because U
ated by all generators of U(N) together with the broken generators Xij over the complex
b(N)
numbers. The representative (z) SOC (2N)/U
= SO(2N)/U(N) of the equivalence
1
b
class (z)U (N) is given in terms of the 2 N(N 1) coordinates zij of SO(2N)/U(N) by

i S
1 0
(83)
(z) = exp Z =
,
Z = zij X
ij .
z 1
2
On the r.h.s. of the equation for (z) we used the vector representation in the diagonal
U(1)-charge Y basis, where Z is nilpotent Z 2 = 0. The normalization factor 2i in the
definition of Z is chosen such that we get the simple matrix expression for (z) expressed
in terms of z which coincides with (27). Notice the distinction between z and Z: Z is
Sij contracted with the
the linear combination of negatively charged broken generators X
ij
complex coordinates z of the coset space. Therefore, Z is represented by a (2N 2N )matrix, while z is an (N N )-matrix. Using the projection operator defined in Eq. (32)
and (z) the Khler potential is given by (47) and (43)

(84)
K(z, z ) = ln det (z) (z) = ln det 1 , 1 = 1 + zz.
Here the det denotes that the determinant is defined on the subspace on which the
projection acts as the identity. Notice that the submetric defined in (33) is the
transposed = T of because of the anti-symmetry of z.
We next determine the non-linear transformations of the anti-symmetric coordinates zij
under the finite g SO(2N) transformation. From Eq. (29) we know directly that
458
z = ( + z)( + z)1 .
(85)
The submetric transforms under these finite SO(2N)-transformations as

1
1
1
h (z; g) = a z a ,
gz,g z = h (z, z ) h ,
(86)
using Eq. (30). Notice that according to Eq. (82) h + (z; g) = h T (z; g) is the transposed of
h , therefore, we only use h in the following. The Khler potential (84) transforms as
follows

S(z; g),
(87)
K gz,g z = K(z, z ) + F (z; g) + F
where the holomorphic function F (z; g) is given by
F (z; g) = ln det h (z; g).
(88)
The complex Hermitean metric of the coset is obtained from the Khler potential (84) in
the standard way as the second mixed derivative

(89)
G (dz, d z ) = tr dz T d z = tr dz (1 + z z)1 d z (1 + zz )1 .
We next discuss the Killing potentials M for the Goldstone scalar fields z and z . The
Killing potential M , defined by Eq. (14), can be written for the coset SO(2N)/U(N) as

e = tr uM u + xM x + x M x ,
= Tr M
(90)
M (u, x, x)
where the trace Tr is over (2N 2N )-matrices, while the trace tr is over (N N )-matrices.
We have used a notation similar to Eq. (80)

e x
e
M
u x
e = M u
and
M
(91)
=
e x M
e uT ,
x uT
M
e x , M x = M
e x and M u = M
e u + (M
e uT )T . We now determine the
so that M x = M
Killing potentials explicitly. We will introduce some notation that might seem somewhat
cumbersome at this stage, but which will be convenient when we discuss the Killing
potentials due to additional matter coupling. Define the matrices R and RT by
R(z; ) = x uT z zu + zx z,
RT (z; ) = uT + zx .
(92)
Notice that z = R(z; ) is a compact notation for the Killing vectors of the coset
space, and tr RT = F (z), the holomorphic Khler transformation. Computing the Killing
potentials M in the standard way (14) gives
iM (z, z ; ) = tr (z, z ; ),
(93)
where we have defined the matrix in analogy to the Killing potentials associated with
the Grassmannian cosets [12] by

(94)
(z, z ; ) RT R z = zuz uT x z + zx .
The matrix can also be written in terms of the BKMU-variable (z) and and the
T = ( 0 1 ) as
projector

T
(z, z ; ) =
(z)
1
(z )
1
e is given by
Using that ((z))1 = (z), the Killing potential matrix M

z z z
T
e
.
i M = (z ) (z) =
z
459
(95)
(96)
From this we can read off the Killing potentials M x , M x and M u to find
iM x = z,
iM x = z ,
iM u = 2zz + 1.
(97)
4.4. Matter coupling

In this section we discuss different types of matter couplings to the supersymmetric
SO(2N)/U(N) -model. As we only need the decomposition of the chiral spinor
representation of SO(2N) in completely anti-symmetric SU(N) tensors in our construction
of anomaly-free models later, we focus here primarily on these representations. We first
introduce a matter representation x which transforms in the same way as a differential.
Under a finite transformation (85) the real superfield x transforms as
x = h (z; g)x h T (z; g),
using that h + = h T . An invariant Khler potential for x is given by

.
K(x, x;
z, z ) = tr x T x
(98)
(99)
Below we discuss non-linear SO(2N) realizations on the irreducible completely antisymmetric SU(N)-tensor representations with p indices and arbitrary rescaling charge q.
i1 ip
, or without indices by T(p;q) , when no confusion
We denote these tensors by T(p;q)
is possible. We interpret them as matter multiplets and construct their invariant Khler
i
potentials. To define their transformation properties we first consider a vector T i = T(1;0)
without a rescaling charge. It transforms as
T = h (z; g)T ,
(100)
under finite non-linear SO(2N) transformations (85). An invariant Khler potential for the
vector T = T(1;0) is given by
K(1;0) = TST = TSi i j T j ,
(101)
with the metric defined in Eq. (84).

It is also possible to couple a singlet chiral multiplet S to the coset, which can be
interpreted as a section of the minimal line bundle. It transforms as (65)
1
1
g
S = e 2 F (z) S = det h 2 S,
(102)
so that its Khler potential
Se 2 K ,
K(0;1) = S S
1
(103)
is invariant. With this singlet S, we can rescale any given chiral multiplet, for example
i
i
S q T(1;0)
transforms as
T(1;q)
460

q
q
T(1;q) = g S q T(1;0) = e 2 F (z) h T(1;q) = det h 2 h T(1;q).
(104)
Since S is a section of the minimal line bundle over the coset SO(2N)/U(N) the rescaling
charge q is integer. The generalization of Khler potential (101) is given by
K(1;q) = TS(1;q)(1;q)T(1;q),
(105)
with the modified metric

q
(1;q) = e 2 K = (det ) 2 .
(106)
Now we construct completely anti-symmetric tensor representations of higher rank.

By taking the completely anti-symmetric tensor products of a set of SU(N) vectors
i
{T1i1 , . . . , Tpp } we obtain an SU(N) tensor of rank p with rescaling charge q
i i
p
1
A
= T(p;q)
T(p;q)
1 q [i1
i ]
S T1 Tpp .
p!
(107)
Here we have introduced the multi-index notation A = (i1 ip ) and [. . .] denotes

the complete anti-symmetrization of the indices inside the brackets. In analogy to the
transformations of T(1;0) and S we obtain
q
i
i
j1 jp
g i1 ip
T(p;q) = det h 2 h 1 j1 h p jp T(p;q)
.
(108)
The Khler potential for this tensor T(p;q) is the direct generalization [11] of the Khler
potentials for the vector (101) and singlet (103)
q
1S
i1 ip
A
T(p;q)j1 jp e 2 K j1 i1 jp ip T(p;q)
, (109)
K(p;q) = TS(p;q)B GB
(p;q) A T(p;q) =
p!
with the generalized metric

GB
(p;q) A =
q
1
(det ) 2 j1 i1 jp ip .
p!
(110)
The SU(N) Levi-Civita tensor i1 iN is invariant under SU(N) transformations. We can
use it to defined an SU(N) dual tensor T(Np;q) ip+1 iN with N p indices and rescaling
charge q by
T(Np;q) ip+1 iN
1 ip c...i1
T
i i ,
p! (p;q) 1 N
(111)
which transforms under the finite transformation (85) as

j1
jp
1+ q
g
2.
T( p;q)i1 ip = T( p;q)j1 jp h 1
h 1
(112)
det h
i1
ip
Note, that as G is not holomorphic, we have preferred to absorb it in a redefinition of the

metric, rather than in the definition of the dual. The power 1 + q2 of the det h instead of
q/2 arises because we have changed from h to its inverse at the expense of an additional
factor of the determinant of h . In our conventions tensors have superscript indices while
dual tensors have subscript indices. Clearly, working with anti-symmetric tensors or dual
tensors is equivalent. The invariant Khler potential for a dual tensor is given by
SB
K(p;q)
= T( p;q) A GA
( p;q) B T( p;q) ,
461
(113)
where the metric is given by

GA
( p;q) B =
i
i
q
1
(det )1+ 2 1 1 j 1 p jp .
1
p!
(114)
In addition we can construct a matter representation A that transforms in the adjoint of

SU(N). The index structure of this matrix is Ai j and in addition it is traceless trA =
Ai i = 0. It transformation properties under the full non-linear SO(2N) symmetries takes
the form
A = h (z; g)Ah 1
(z; g).
(115)
This transformation rule can be obtained by defining A as the tensor product of a vector
with rescaling charge 2
T(1;0) and a dual vector T(1;2)
.
A = T(1;0) T(1;2)
(116)
It is easy to see that this gives the right generalization of the SU(5) adjoint by restricting
SO(2N) to an U(5) transformation:

0
g
T 1
A T .
(117)
g=
1 H A =
T
0
Clearly, A does not transform under the U(1) factor of U(5). Notice that the condition that
A is traceless, is respected by the transformation rule (115). The simplest invariant Khler
potential for this matter field A is

(118)
KA = tr A 1 A .
to the Killing potentials
We next turn to a discussion of the contributions M(p;q) and M(p;q)
of rank p with a rescaling charge q, respectively.

for a tensor T(p;q) and a dual-tensor TS(p;q)
are invariant, their contributions to the Killing

As the Khler potentials K(p;q) and K(p;q)
potentials are obtained from

iM(p;q) = K(p;q), R ,
iM(p;q)
= K(p;q),
R ,
(119)
where i Z = Ri denote the Killing vectors (cf. Eq. (92))

z = R,
p
X
q
i1 ip
tr(RT )T(p;q)
,
2
r=1

p
X
q
j
tr(RT )T(p;q)
=
T
(R
)
+
1
+
T(p;q)i
T jr
(p;q)i
i1 ip .
1 ip
1 j ip
2
i1 ip
T(p;q)
i j ip
1
(RT )ir j T(p;q)
(120)
r=1
They follow from expanding the finite transformations (85), (108) and (112) to first order
in the infinitesimal parameters u, x, x .
The Killing potential for a rank p tensor with rescaling charge q is given by
462

C
A
iM(p;q) = TS(p;q)B GB
(p;q)C 1(p;q)A T(p;q) ,
(121)
where, using the notation (94),

1C
(p;q)A =
p
X
k1 i1 1kr ir kp ip +
r=1
q
tr 1 k1 i1 kp ip .
2
(122)
To obtain this result we have made the following steps. We first obtained the Killing
potential for a rank 1 tensor (a vector) with rescaling charge zero. This result can easily be
generalized to a rank p tensor with rescaling charge zero. Next we construct the Killing
potential for a rank 0 tensor (a singlet) with an arbitrary rescaling charge. Finally we put
all results together to obtain Eq. (121). We can proceed similarly to obtain the Killing
for a rank p dual tensor with a rescaling charge q. As the dualization has
potential M(p;q)
introduced a determinant det h in the finite transformation (112), it is more convenient to

first consider a rank p dual tensor with rescaling charge 2, which precisely cancels the
determinant. To obtain the final result for a rank p dual-tensor with a rescaling charge q,
we have to rescale the rank again, which introduces a factor 1 + q2 . Finally, the Killing
potential reads
= T(p;q)B
1B
GC
TS(Ap;q)
iM(p;q)
(p;q)C
,
(p;q)A
(123)
defined as
with 1C
(p;q)A
=
1B
(p;q)C
p
X
r=1

q
tr 1 j1 k1 jp kp .
j1 k1 (1)jr kr jp kp + 1 +
2
(124)
The infinitesimal form of the transformation of the adjoint matter field A is given by
A = RT A ART = [RT , A]
and the resulting Killing potential can be written as

iMA = tr 1A 1 A A1 1 A = tr [1, A] 1A .
(125)
(126)
4.5. Consistent SO(2N)/U(N) spinor models

In this subsection we construct an anomaly-free model based on the spinor representation of SO(2N) that contains the coordinates of the coset SO(2N)/U(N). Only for a
limited number of choices for N such a model satisfies the line bundle constraint.
A supersymmetric model built on the SO(2N)/U(N) coset space is not free of anomalies
by itself, as all the 12 N(N 1) anti-symmetric coordinates zij and therefore, also their
chiral fermionic partners carry the same charge 4 in the standard normalization. To
construct a consistent supersymmetric model around this coset one can try to embed
the coordinates in an anomaly-free representation. All representations of SO(2N) are
anomaly-free, unless SO(2N) is isomorphic to a non-anomaly-free unitary group. This
happens for SO(2)
= U(1) and SO(6)
= SU(4), hence we disregard the cases N = 1, 3
below. In Appendix B we derive this result by calculating the possible U(1) anomalies
463
of the chiral spinor representation. An SO(2N) representation that branches to an antisymmetric 2-tensor of SU(N) is the chiral spinor representation of SO(2N). The other
U(N) representations that arise from the spinor representation transform under the full
SO(2N) symmetries via non-linear transformations. For global consistency this means
that these matter representations are sections of bundles. If one of these sections is a line
bundle we run into the cocycle condition, which greatly restricts the freedom of charge
assignments. In Section 3 we have determined the section of the minimal line bundle
over SO(2N)/U(N). As the dimension 2N is even, the irreducible representations carry
definite chirality; we show that it is sufficient to consider only the positive chiral spinor
representation for our purpose of extending the coset to the spinor representation. After
that we turn to the main result of this subsection: the cocycle condition only allows for a
very restricted class of consistent SO(2N)/U(N) spinor models: N = 2, 5, 6, 8.
As argued in Appendix A, to construct a consistent model on SO(2N)/U(N) using
irreducible spinor representations, we need to identify the anti-symmetric coordinates zij
of the coset space with an anti-symmetric 2-tensor of the branching of the spinor. We have
the following two states 2 ij or 2 ij as possible candidates. According to Eq. (A.15),
Appendix A, the charge of 2 ij is N 4; it has positive chirality. The charge of 2 ij is
opposite and its chirality is ()N . Notice that for N = 4 we can never construct a consistent
model using the spinor representations as the charges of 2 ij and 2 ij are zero, while
the charge of the coordinate zij is non-zero. For N is even both 2 ij and 2 ij have the
same chirality, hence they are in the same irreducible representation. The duality operation
(A.16), Appendix A, maps the positive chirality states into themselves. Therefore, for even
N it is sufficient to consider only the state 2 ij as the candidate for the coordinates zij of
the coset. For odd N the only odd length state that can be associated with the coordinates
zij has length N 2, but it is dual to the state with length 2. Therefore, for all N it is
sufficient to consider only the positive chirality spinor representation and only the state
2 ij as candidate for the coordinates zij of SO(2N)/U(N).
We next discuss the restriction that the consistency of the line bundle poses on the
construction of anomaly-free extensions of cosets SO(2N)/U(N) using the positive
chirality spinors of SO(2N). We remarked before that the case N = 4 does not work as
the state 2 ij does not carry Y charge. Therefore we consider the cases N = 2 and N > 5
from now on. It was shown in Section 3 that the minimal charge of the line bundle over the
coset space SO(2N)/U(N) is equal to N when the charge of the coordinates is taken to
be 4. This is the normalization employed in our detailed discussion of the SO(2N) algebra
in Subsection 4.1. As all states of a positive chirality spinor have an even number of indices,
the tensor structure of these states can be obtained from completely anti-symmetric tensor
products of the tangent vectors of the coset SO(2N)/U(N) tensored with an integral power
of the minimal line bundle. In particular the state (2p;q) with length 2p and rescaled with
the q(p; N)th power of the minimal line bundle has a charge 4p + Nq(p; N). For each
p this charge should be proportional to the charge N 4p of the anti-symmetric tensor
with 2p indices within the positive chirality spinor representation. Therefore we obtain the
relation (N 4p) = 4p + Nq(p; N) where R is a constant to be determined. Since
464
the anti-symmetric tensor with 2 indices (p = 1) is identified with the coordinates zij of
4
. Solving for
the coset, it does not have a rescaling charge, hence we find that = N4
q(p; N) gives
4
(1 p).
(127)
N 4
For consistency of the line bundle we need that q(p; N) is an integer for all 0 6 p 6
[N/2]. Notice that q(p; N) is integer whenever q(0; N) is integer. q(0; N) is only an
integer if N 4 is a divider of 4, which implies that N = 0, 2, 3, 5, 6, 8. Of course N = 0
is impossible, and though the case N = 3 satisfies the line bundle quantization condition,
it does not lead to an anomaly free model. Therefore the possible choices are:
q(p; N) = q(0; N)(1 p) =
N
4
q(0; N) = N4
The case of N = 2 is trivial in the sense that the coset is isomorphic to the simplest coset
SU(2)/U(1) (i.e., the 2-sphere) because SO(4)
= SU(2) SU(2). Notice that except for
the last case N = 8 we only use squares of the minimal line bundle.
We finish this section by giving the Khler potentials for the anomaly-free SO(2N)/
U(N) models based on the positive chiral spinor representation. The matter content is
fixed by the discussion above: we need for each 0 6 p 6 [N/2] a rank 2p completely
anti-symmetric SU(N) tensor with rescaling charge q(p; N) given in Eq. (127), except for
p = 1; this case corresponds to an anti-symmetric tensor with two indices, for which we
take the coordinates of the SO(2N)/U(N) coset itself. Using the Khler potentials for the
coset (84) and for anti-symmetric tensor representations with an arbitrary rescaling charge
(109), we can express the Khler potential for the complete system by
1
K = K +
2
[N/2]
X
K( 2p; q(p;N)).
(128)
p=0, p6=1
It is important to stress that the line bundle S, which we have used as a U(1)-compensator,
does not appear in the model as an independent field. However, it is used to ensure that the
construction of the matter superfields T(2p; q(p;N)) is well-defined. Here we have included
a factor 1/2 so as to get the standard normalization of the kinetic terms of the Goldstone
boson fields. In Section 5 we discuss the consistent SO(10)/U(5)-spinor model in detail.
There we give the explicit expression for the Khler potential, using dual tensors to reduce
the number of indices.
5. Analysis of the SO(10)/U(5)-spinor model

In this section we apply the constructions presented so far to obtain anomaly-free
SO(10)/U(5)-spinor models in the context of global supersymmetry. We study in particular
465
the role of the potentials in determining the realization of internal symmetries and
supersymmetry in the original -model, as well as in various gauged versions. We find
some rather surprising results concerning the gauged versions of the model, implying that
in spite of selecting anomaly-free combinations of representations it is not possible
to gauge just any arbitrary global symmetry. In particular, we find that gauging the full
global SO(10) is not possible, whilst the consistency of gauging all or part of the linear
SU(5) U(1) symmetry depends crucially on the vacuum expectation values and choice
of parameters in the model. We present strong arguments that the natural value of the
FayetIliopoulos parameter in the models with a linear gauge group containing U(1) is
determined by the scale f of the -model, by a relation of the type | |f 2 O(1). It is
not difficult to see that some of these results are valid beyond the particular model chosen.
A more general and extensive discussion will be given elsewhere [29].
The choice for the model on SO(10)/SU(5) U(1) is motivated by its fermionic field
content, corresponding to one complete family of quarks and leptons, including a righthanded neutrino. This can be seen by looking at the SU(5) representations of the chiral
multiplets that the model contains: the coordinate multiplets ij form the 10 of SU(5).
The completely anti-symmetric tensor with 4 indices is equivalent to the 5 with a relative
U(1) charge 3; we denote it by i . And finally we have a singlet of SU(5), with U(1)
charge +5.
We denote the full set of chiral superfields by = ( ij , i , ), their physical
components collectively by (Z , L ). The scalar components of the various SU(5)
representations are denoted by Z = (zij , ki , h). To guarantee that the physical scalars h
and ki are well-defined, the line bundle S has been used internally as a U(1)-compensator
and as such it does not represent a physical degree of freedom itself. In the absence of any
local gauge couplings, the kinetic part of the lagrangean for the model is given in terms
S ) by the supersymmetric expression (8), which is
of real composite superfields K(,
equivalent to

b H
S Z + D
b
S = G Z
/
H
LK = K ,
L
L
D

1

+ R R L L R .
2
(129)
In the present case, the Khler potential from which the metric G is derived, is given by
(128)

S Z = 1 K + K(0;4) + K
K Z,
(1;4)
2
1
=
ln det 1 + (det )2 |h|2 + (det )1 k 1 k,
2f 2
(130)
with the submetric 1 = 1 + f 2 zz and ef K = (det )1 . This is the explicit form of

Eq. (128) in the SO(10)/U(5) case, after rescaling the Goldstone fields z by the mass
b are
parameter m = 1/f which sets the scale of the -model. The auxiliary fields H

defined as H = H R L with the connection = G G, .

2
466
It is of particular importance to have an explicit expression for the kinetic terms of the
Goldstone fields, which are modified by the presence of the matter terms in the Khler
potential (130). Following the procedures of [10] and [12], the kinetic terms for the scalars
ij
zij and the quasi-Goldstone fermions L are determined by the matter-extended Khler
metric

2
= G (ij ) (kl)x ij xkl = f 2 E tr x T x

+ ef K f 2 kx T x k,
(131)
G (x, x)
where x ij stands for components of the Goldstone superfield ij or their gradients, whilst
1
2
2
+ ef K k 1 k 2e2f K |h|2 .
2f 2
For some applications it is convenient to write this as

2
= tr x T x , = f 2 E + ef K f 2 kk.
G (x, x)
E=
(132)
Clearly, the physical requirement that the model be ghost-free implies that this metric has
is
to be sign-definite. As is positive definite, and the second term proportional to kk
non-negative, positive definiteness of the metric is guaranteed if E > 0. For E < 0, there
always are negative kinetic-energy ghosts. However, for E = 0 a more detailed analysis is
required.
In particular, we note that

2
(133)
det = f 10 E 4 E + ef K k 1 k det .
As a result, for E = 0 the metric has a four-fold zero eigenvalue. This implies the
existence of four complex orthogonal eigenvectors of with zero eigenvalue; one can
then construct six independent (complex) anti-symmetric tensor zero-modes, of the form
x = vwT wv T , with v, w independent zero eigenvectors. Of course, if k = 0 at the same
time, the whole metric vanishes; then also the kinetic terms of all Goldstone fields and
their fermion partners vanish.
As concerns mass-terms, we observe that for the model (130) it is not possible to
construct an SO(10)-invariant superpotential. First, the non-linear transformations of the
coordinates z exclude their appearance in an invariant expression. Next, there is no nonvanishing holomorphic SU(5) invariant for ki . Finally, as h transforms under U(1) and
there is no field that compensates for its transformation, it also cannot appear in the
superpotential. In the absence of a superpotential, all fields in the action (129) the
Goldstone bosons and their superpartners as well as the chiral superfields defining the
matter representations describe massless spin-0 and chiral spin-1/2 particles.
This situation changes if we add a second family of quarks and leptons, with superfields
ij
(2) = ((2) , (2) i , (2) ). It is then possible to construct an invariant superpotential
X
ij
a (a) (1) i (2) j (2) .
(134)
W () =
a=1,2
The a are coupling constants of dimension (mass)1 .

As a next step towards a physical interpretation of the fermions as describing quarks
and leptons, we introduce gauge interactions. This can have important implications for the
467
spectrum of the theory, as in supersymmetric theories gauge-couplings are accompanied

by Yukawa couplings and a D-term potential. We first consider gauging the full SO(10).
A local transformation of the form (85) then always allows one to go to the unitary gauge
z = z = 0. Thus all Goldstone bosons disappear from the spectrum as a result of the
BroutEnglertHiggs effect; this is confirmed by the finite mass-terms for the gauge fields
corresponding to the broken generators of SO(10).
However in the presence of matter fields as in (130), required for the cancellation of
anomalies, the analysis of the D-terms in the potential shows that in the unitary gauge
the model becomes singular: in the minimum of the potential the expectation value of the
Khler metric vanishes: E = 0 and k = 0. Thus the kinetic energy terms of the Goldstone
and quasi-Goldstone fields all vanish. Actually, this seems to happen in other fully gauged
supersymmetric -models on Khler cosets with anomalies cancelled by matter as well.
As an alternative to gauging SO(10), one can gauge only the linear subgroup SU(5)
U(1) instead. This explicitly breaks the non-linear global SO(10). It is then allowed
in principle to construct superpotentials which are invariant only under the local gauge
symmetry, although one would expect the strength of this potential to be proportional to
the gauge coupling constant. In fact, this happens automatically with the D-term potentials.
In addition, when gauging any group containing the U(1) as a factor, the introduction of
a FayetIliopoulos term is allowed. It turns out, that the corresponding models are indeed
well-behaved for a range of non-zero values of this parameter.
We now present details of this analysis. The theory defined by the Lagrangian (129),
(130) has a global SO(10) symmetry. This global symmetry allows vector bosons to be
coupled to the model by turning the SO(10) group, or its subgroup SU(5) U(1), into a
local gauge group by introducing covariant derivatives into the Lagrangian. The covariant
derivatives are defined by:
D Z = Z Ai Ri ,
D L = L Ai Ri, L + D Z L .
(135)
Here the Ai are the gauge fields corresponding to the local symmetries. They are
components of the vector multiplets V i = (Ai , i , D i ), with i representing the gauginos
and D i the real auxiliary fields. The isometries Ri are generated by Killing vectors as in
Eq. (120); here they take the form
z = R,
h = 2 tr(RT )h,

k = k RT + tr(RT )1 ,
(136)
with R(z; ) = f1 x uT z zu + f zx z and RT (z; ) = uT + f zx . Adapting its

normalization to that of the kinetic terms (130), the full Killing potential generating
these Killing vectors is M = 12 M + M(0;4) + M(1;4)
, which takes the explicit form
(cf. Eq. (94)):

1
2
iM = tr 1 2 K(1;4) + 2K(0;4) ef K k1 1 k.
2f
(137)
468
After introduction of the gauge fields in lagrangean (129), via the covariant derivatives
(135), the -model itself is no longer invariant under supersymmetry transformations.
Supersymmetry is restored by adding terms
i D i (Mi + i ).
(138)
1LK = 2G Ri L iR + R
i R L
We have added a FayetIliopoulos term with parameter i in case there is a commuting
U(1) vector multiplet. The full lagrangean for this model after introducing gauge
interactions becomes
L = LYM + Lchiral ,
where LYM is the usual supersymmetric YangMills action, of the generic form

1
1 2
1
1
+ D
/ D2 .
LYM = 2 Tr F
4
2
2
g
(139)
(140)
We use Tr to denote a trace over 2N 2N -matrices; in contrast, traces over (N N )matrices are denoted by tr. When gauging a product of several commuting subgroups of G,
e.g., SU(5) U(1), there is a coupling constant gi for each of the subgroup factors. Lchiral
is given by (129), but with ordinary derivatives Z , L replaced by the covariant
derivatives (135), while adding 1LK :
Lchiral = LK ( D ) + 1LK .
(141)
Next we analyze the scalar potential obtained by elimination of the D-fields for various
gaugings. By substituting the expression (94) for 1 we obtain in index-free notation 1

1

2
2
,
iMu = 1 2f z z
+ K(1;4) 2K(0;4) + ef K k T k T f 2 z kkz
2f 2

1
2
iMx = f z 2 K(1;4) + 2K(0;4) + f ef K z kk,

2f

1
2
(142)
iMx = f z 2 K(1;4) + 2K(0;4) f ef K kkz.
2f
If the full SO(10) is gauged, the unitary gauge can be chosen in which all Goldstone
bosons (z, z ) vanish. This implies that the broken Killing potentials Mx and Mx vanish
automatically, leaving us with the U(5) Killing potentials only. If we only gauge U(5) then
the Killing potentials Mx and Mx are irrelevant, and again we have to consider only the
U(5) Killing potentials. However, in this case z represents a physical degree of freedom,
and its vacuum expectation value does not necessarily vanish: hzi = 0 is guaranteed only
if SU(5) is not broken.
To analyze both gauged SO(10) and gauged SU(5) U(1) at once, we consider the Dterm potential arising from the gauging of SU(5) U(1) including a FayetIliopoulos term
with parameter for the U(1):
1 The factors i here result from 1 being anti-hermitean, Eq. (93).
469
g12
g2
( iMY )2 + 5 tr(iMt )2 .
(143)
2N
2
Here the U(5) Killing potentials MY and Mt are trace and the traceless part of Mu :
V=
M t = Mu
1
MY 1,
N
MY = tr Mu .
(144)
We can derive tr M2t from MY and tr M2u by

1
(iMY )2 .
N
An explicit expression for Mt in terms of the matter-extended submetric is
tr(iMt )2 = tr(iMu )2
iMt =

2 T
2
2 1 ef K f 2 z k kz + k T k T ,
2
f
(145)
(146)
where is defined by

1 2
(147)
N = (tr )E + ef K k 1 f 2 zz k.
2
The terms in the potential (143) are proportional to the square of the coupling constants
g1 and g5 of the U(1) and SU(5) gauge groups, respectively. The case of fully gauged
SO(10) is reobtained by taking the coupling constants equal: g1 = g5 = g10 , and the Fayet
Iliopoulos term to vanish: = 0. We have left the rank N = 5 of SO(10) in, so as to keep
track of some of the dependence on this rank.
The potential (143) is non-negative. In order for supersymmetry to be preserved, the
minimum must be at Vmin = 0; in contrast, Vmin > 0 implies spontaneous supersymmetry
breaking by the potential. Being a sum of squares, a vanishing potential is possible only if
MY = 0 and Mt = 0 at the same time.
In the case of gauged SO(10) one can always work in the unitary gauge z = z = 0.
However, in the case of gauged SU(5) U(1) the potential can cause further symmetry
breaking by generating a vacuum expectation value for the would-be Goldstone bosons.
Because of its antisymmetry, an SU(5) U(1) transformation can be performed to put hzi
into the standard form
a2
,
(148)
b2
hf zi =
0
with real a, b > 0. Of course, the unitary gauge is included as the special case a = b = 0.
The vacuum expectation value (148) preserves a subgroup SU(2) SU(2) U(1). If the
5 gets a vacuum expectations value, this residual symmetry can be used to chose
hki = (k1 , 0, k3 , 0, k5 ).
(149)
We first investigate the existence of zeros of the potential, compatible with supersymmetry.
The condition hMt i = 0 then implies k1 = k3 = 0, and

(150)
E = 1 + a 2 = 1 + b2 = 1 + a 2 1 + b2 |k5 |2 .
There are three separate solutions to these conditions; the first is
470
a = b = k5 = 0,
E = .
(151)
This solution includes the unitary gauge. A second solution (which coincides with the
previous one for a = b = 0) is the case E = = k5 = 0. It can be seen immediately to yield
= 0. Therefore, in this case the kinetic terms of the Goldstone superfield components
vanish. Such a solution is unacceptable, not only because part of the quarks and leptons
disappear from the spectrum of physical states, but even more importantly as this upsets
the cancellation of anomalies, which is guaranteed only if all chiral fermions in the model
contribute. This holds in particular for the case of fully gauged SO(10) in the unitary gauge.
The third solution of the supersymmetric vacuum conditions, which exists only for
E, < 0, is
a 2 = b2 =
4
1
1 + a 2 |k5 |2 ,

E = 1 + a2 .
(152)
Inserting this solution into the expression (132) one obtains = f 2 1, which in this case
is negative definite. As it is not possible to change the overall sign of the Khler potential
without creating negative kinetic energy terms for the matter fields, this solution always
contains ghosts and is again physically unacceptable.
The upshot of this discussion is, that physically consistent models (i.e., anomaly-free,
with positive definite kinetic energy), in which the potential has zeros, require z = k = 0
and E = > 0. Such models can be realized with gauged SU(5) U(1), but the model
with fully gauged SO(10) is excluded. We observe, that positivity of E for these solutions
implies
0 6 |h|2 <
1
.
4f 2
(153)
Thus, unless h = 0, these solutions always spontaneously break U(1), whilst SU(5) is
manifestly preserved.
The conditions (151) and (153) are necessary to have physically consistent models with
hMt i = 0. This is sufficient for a zero of the potential in a model with gauged SU(5) only.
If U(1) is gauged, a zero of the potential requires the additional condition:
2
2

1 a 2 |k1 |2 + 1 b2 |k3 |2 + |k5 |2
= hiMY i = 1 + a 2 1 + b2

1 b2
1 a2
+2
E.
(154)
1+2
1 + a2
1 + b2
Combining this with z = k = 0, it follows that (with N = 5)
E= =
,
N
|h|2 =
.
+
4f 2 2N
(155)
A consistent solution of this type exists only for N/(2f 2 ) 6 < 0. The kinetic energy
for the Goldstone superfield components is now proportional to (f 2 )/N .
Clearly, for values of in this range it is necessary to perform a finite renormalization of
the Goldstone superfields to obtain the canonical value of the kinetic terms; in the Khler
potential this is equivalent to a rescaling of the -model scale such that f 2 N/ . In
471
these models the natural value of the FayetIliopoulos-parameter is therefore the -model
scale, thereby relating internal and supersymmetry breaking.
We finish this section by observing that, in addition to zeros of the potential, there can
also be ranges of the parameters (g12 , g52 , ), or models with only some proper subgroup of
SU(5) gauged, for which the minimum of the potential occurs at a positive value: hV i > 0.
In this case supersymmetry is manifestly broken by the potential. This could happen for
example in the domain < N/(2f 2 ). However, we have not performed an exhaustive
analysis of this case.
6. Conclusion
In this paper we have considered supersymmetric models based on classic Khlerian
coset spaces: U(M + N)/U(M) U(N), USp(2N)/U(N) and SO(2N)/U(N), and their
non-compact versions. Starting from a non-linear realization of the group SL(N + M, C) in
finite form, we constructed their Khler potentials. A generalization of the Killing potential
for finite transformations has been obtained. The Khler potential of such a coset can be
written as a function of a fundamental submetric. This submetric also allows us to construct
Khler potentials for superfields as sections of bundles over the original classical coset.
For most of these matter representations the naive definitions are sufficient to guarantee
the existence of these bundles globally. However, the consistency of line bundles requires
that the cocycle condition is satisfied.
We have discussed various aspects of these general constructions for classical Khlerian
coset spaces in more detail for the class of orthogonal cosets SO(2N)/U(N). All
supersymmetric matter fields which form completely anti-symmetric representations of
SU(N) with arbitrary integer charges satisfying the cocycle condition have been obtained
explicitly.
Pure supersymmetric coset models are often anomalous due to their chiral fermions. This
is also the case for orthogonal cosets SO(2N)/U(N), but as all SO(2N) representations
are anomaly free (with the exceptions of SO(2)
= U(1) and SO(6)
= SU(4)), the
supersymmetric field content can be extended such that all anomalies cancel. The
completely anti-symmetric SU(N) representations descending from the positive-chirality
spinor representation of SO(2N) provide possible candidates for anomaly free models,
which can include the Goldstone bosons. However, the U(1) charges of these antisymmetric representations can often not be realized using the bundles at our disposal. In
fact, only for N = 2, 5, 6, 8, these SO(2N)/U(N)-spinor models can fulfil the consistency
requirements of the line bundle.
Some phenomenological aspects of the SO(10)/U(5)-spinor model have been investigated. This model contains the SU(5) U(1) fermionic field content of one generation of
quarks and leptons, including a right-handed neutrino. The matter-extended metric for the
Goldstone bosons of the coset is not automatically positive definite. In order that the theory
is ghost-free when expanded around a minimum of the potential, the quantity E has to be
positive, see Eq. (132). The consequences of this physical requirement have been analysed
472
for supersymmetric minima, if part of the isometry group is gauged. If the whole SO(10)
is gauged, the analysis is straightforward as one can employ the unitary gauge to put the
Goldstone bosons to zero. We find the kinetic energy of the would-be Goldstone modes
and their fermionic partners to vanish. Therefore, the quasi-Goldstone fermions no longer
contribute to the cancellation of anomalies.
Gauging (part of) the linear subgroup U(5) calls for a more involved investigation. First
we have obtained all supersymmetric minima for the case where SU(5) is gauged. We
found three classes of such vacua, of which two are physically problematic as the kinetic
terms of the Goldstone multiplets either vanish or have negative values. The third type
of supersymmetric vacuum only exists for a finite range of vacuum expectation values of
the scalar partner of the right-handed neutrino. If the U(1) factor is gauged in addition,
the FayetIliopoulos parameter is related directly to the vacuum expectation value of this
scalar. This shows that only for a finite range of values of the FayetIliopoulos parameter
U(5) can be gauged consistently.
Appendix A. Decomposition of SO(2N ) spinors into anti-symmetric tensors

An arbitrary spinor of SO(2N) can be represented using anti-symmetric tensors
p i1 ip of SU(N) with p indices as
i i
(A.1)
= 0 , 1i1 , . . . , N1 N .
The invariant inner-product of two spinors and is given by
=
N
X
1
i i
p1 p ,
p! p ip i1
(A.2)
p=0
i i
where p ip i1 = p 1 p . We want to construct a basis for the anti-symmetric SU(N)tensors, and also a basis for the SO(2N)-spinors, using the Clifford algebra of fermion
creation and annihilation operators i and Si , as introduced by Mohapatra and Sakita
[24], see also [27] and [25]. They satisfy the usual anti-commutation relations
i j

i
, = Si , Sj = 0.
(A.3)
, Sj = i j ,
Assume that we have constructed a Hilbert space on which these Clifford operators act.
In this Hilbert space we define the vacuum state |0i by i |0i = 0 for any i. The ket- and
bra-states
ep i1 ip = Si1 Sip |0i,
i i1
ep p
= h0| ip i1 ,
(A.4)
satisfy the orthonormality relations

i i1
ep p
eq j1 jq = 0,
i
for p 6= q
i i1
and ep p
i1
ep j1 jp = [j
jpp ] ,
1
(A.5)
i1
jpp ] is the complete anti-symmetrized Kronecker-delta. Therefore the states
where [j
1
ep i1 ip form a basis of anti-symmetric rank p tensors of SU(N). Using the complete antisymmetry it is easy to show that the number of the vectors ep with length p is equal to

N
p , hence the total number of
ep for 0 6 p 6 N form a basis
473
vectors {ep } is equal to 2N . The collection of these states

for SO(2N)-spinors, hence and can be expanded in
this basis
=
N
X
1 i1 ip
p
ep i1 ip
p!
and =
p=0
N
X
1
i i
p ip i1 ep 1 p .
p!
(A.6)
p=0
It is straightforward to check that in this basis the inner-product of two spinors is

consistent with the definition (A.2) using the Clifford properties (A.3).
In terms of the Clifford algebra, we define the 2N gamma-matrices a with a =
1, . . . , 2N by
(

i i Si , a = i = 1, . . . , N,
(A.7)
a =
a = i + N = N + 1, . . . , 2N,
i + Si ,
with the property
{a , b } = 2ab .
(A.8)
This property can be used to show that the sigma-matrices

1
1
(A.9)
Mab = ab = [a , b ],
2
4
are the generators of the SO(2N)-algebra (68) in the spinor representation. With respect
to the spinor inner-product (A.2) the gamma-matrices are Hermitean a = a and hence
= ab . Furthermore it implies that w.r.t. this

the sigma-matrices are anti-Hermitean ab
inner product the fermion creation/annihilation operators are Hermitean conjugates:

Si = i .
(A.10)
i = Si ,
For products of Clifford operators A and B we have (AB) = B A .
The Hermitean chirality operator e defined by
e = () 2 N(N1) i N
1
2N
Y
a ,
(A.11)
a=1
can be written in terms of the Clifford elements as

e =
Y
N
Y
i , Si = (1 2n i ) = ()n .
(A.12)
i=1
Here we have defined the ith number operator n i = Si i and the total number operator n =
P
i . Using this chirality operator, we can define positive and negative chirality spinors
in
in 2N dimensions
e = .
(A.13)
Using the form of the chirality operator (A.12), it follows that the positive chirality
components of a spinor are given by completely anti-symmetric SU(N)-tensors of even
length p, while the negative chirality components have odd length p.
474
The generators U i j can be expressed in terms of the fermion operators as

Uij =

1 i
, Sj ,
2
(A.14)
and satisfy the U(N) algebra (69). Their anti-symmetric part Ai j , and symmetric part S i j ,
take the form

1 i
i i
, Sj j , Si ,
, Sj + j ,
Si .
Si j =
Ai j =
4
4
ij
Sij can be represented by
Furthermore the broken SO(2N) generators X and X
Xij = i j
Sij = Si Sj .
and X
The U(1)-charge operator (73) is given in terms of the total number operator n by
X

i , Si = N 2n,
(A.15)
Y=
i
hence the charge of an anti-symmetric tensor with p indices, that occurs in the
decomposition of a spinor, is N 2p: Y ep = (N 2p)ep . We define the dual vectors
e Np respectively, eNp of the basis vectors ep and ep , respectively, by
e Np iN ip+1 =
1 iN i1

ep i1 ip
p!
and eNp i
p+1 iN
1 ip i1
ep
i1 iN .
p!
(A.16)
For the components p we use analogous definitions. Notice that under dualization the
charge does not change, only the number of indices does.
Appendix B. Anomaly cancellation of the spinor representation

We now show that the positive chirality spinors of SO(2N) have no pure U(1)-anomaly,
unless SO(2N) is isomorphic to a non-anomaly-free unitary group, SO(2)
= U(1) or
SO(6)
= SU(4), by computing the possible U(1)-anomalies. However it is straightforward
to also compute the U(1)-anomalies for negative chiralities, so we calculate both here. The
Y k -anomaly A (Y k ; N) = Tr Y k for the chirality spinor representation is given by
A Y ; N =
k
N
X
N 1 ()l
l=0
(N 2l)k .
(B.1)

This follows using the multiplicities Nl and charges N 2l of the states l i1 il . The factor
1()l
is introduced to project onto the positive or negative chirality states. The necessary
2
details to obtain these results can be found in Appendix A. To calculate these anomalies it
is convenient to introduce the functions
1
q (x) = + x,
x
of a variable x, in terms of which we define
(B.2)
P (x; N)
N
X
1
N 1 ()l N2l
x
.
(q+ )N (q )N =
2
2
l
475
(B.3)
l=0
d
. The anomaly A (Y k )
Notice that the charge operator Y can be represented by Y = x dx
can be calculated using the functions P by

d k
P (x; N) .
(B.4)
A Y k ; N = x
dx
x=1
To compute this we use the properties of the functions q

x
d
q = q ,
dx
q+ |x=1 = 2,
and q |x=1 = 0.
We obtain the following results for the Y and Y 3 anomalies in D = 4 dimensions

1, N = 1,
A (Y ; N) =
0,
N 6= 1,
and
A
3!22 , N = 3,
3
Y ; N = 1,
N = 1,
0,
N 6= 1, 3.
(B.5)
(B.6)
(B.7)
Hence we see that the cases N = 1 and N = 3 have indeed an anomalous spinor
representation. We conclude from this anomaly analysis that for N = 2 and N > 4 the
spinor representation of SO(2N) is U(1) anomaly-free.
References
[1] J. Bagger, E. Witten, Phys. Lett. B 118 (1982) 103;
C.L. Ong, Phys. Rev. D 27 (1983) 911, 3044;
W. Buchmller, R.D. Peccei, T. Yanagida, Nucl. Phys. B 227 (1983) 503;
W. Buchmller, R.D. Peccei, T. Yanagida, Nucl. Phys. B 244 (1994) 186;
M.P. Mattis, Phys. Rev. D 28 (1983) 2649;
J.W. van Holten, Z. Phys. C (1985) 57;
Y. Achiman, S. Aoyama, J.W. van Holten, Nucl. Phys. B 258 (1985) 179.
[2] M. Bando, T. Kugo, K. Yamawaki, Phys. Rep. 164 (1988) 217.
[3] U. Ellwanger, Fortschr. Phys. 36 (1988) 881.
[4] W. Buchmller, W. Lerche, Ann. Phys. (NY) 175 (1987) 159.
[5] B. Zumino, Phys. Lett. B 87 (1979) 203;
D. Freedman, L. Alvarez-Gaum, Commun. Math. Phys. 80 (1981) 433.
[6] G. Moore, P. Nelson, Phys. Rev. Lett. 53 (1984) 1519;
P. di Vecchia, S. Ferrara, L. Giardello, Phys. Lett. B 151 (1985) 199;
D. Nemeschansky, R. Rohm, Nucl. Phys. B 249 (1985) 157;
E. Cohen, C. Gomez, Nucl. Phys. B 254 (1985) 235;
L. Alvarez-Gaume, E. Witten, Nucl. Phys. B 234 (1984) 269.
[7] W. Lerche, Nucl. Phys. B 238 (1984) 582.
[8] G. Shore, Nucl. Phys. B 248 (1984) 475.
[9] T.E. Clark, S.T. Love, Nucl. Phys. B 254 (1985) 569.
476
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
J.W. van Holten, Nucl. Phys. B 260 (1985) 125.

S. Groot Nibbelink, J.W. van Holten, Phys. Lett. B 442 (1998) 185.
S. Groot Nibbelink, J.W. van Holten, Nucl. Phys. B 588 (2000) 57.
U. Ellwanger, Nucl. Phys. B 281 (1986) 489.
S. Groot Nibbelink, Phys. Lett. B 473 (2000) 258.
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Phys. Lett. B 138 (1984) 94;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 313;
M. Bando, T. Kuramoto, T. Maskawa, S. Uehara, Prog. Theor. Phys. 72 (1984) 1207.
S. Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces, Academic Press,
London, 1978.
M. Gomes, Y.K. Ha, Phys. Lett. B 145 (1984) 235.
C. Ong, Phys. Rev. D 27 (1983) 3044.
F. Delduc, G. Valent, Nucl. Phys. B 253 (1985) 494.
A.B. Lahanas, D.V. Nanopoulos, Phys. Rep. 145 (1987) 1.
R. Gilmore, Lie groups, Lie Algebras and Some of Their Applications, J. Wiley & Sons, 1974.
M.A. Luty, J. March-Russell, H. Murayama, Phys. Rev. D 52 (1995) 1190, hep-ph/9501233.
J.R.O. Wells, Differential Analysis on Complex Manifolds, Springer, New York, 1980.
R.N. Mohapatra, B. Sakita, Phys. Rev. D 21 (1980) 1062.
M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Vol. 2, Cambridge Univ. Press, 1987.
F. Quevedo, hep-th/9603074.
J. Polchinski, String Theory, Vol. 2, Cambridge Univ. Press, 1998.
R. Slansky, Phys. Rep. 79 (1981) 1.
S. Groot Nibbelink, T.S. Nyawelo, J.W. van Holten, in preparation.

The geometry of W3 algebra: a twofold way

for the rebirth of a symmetry
G. Bandelloni a,b and S. Lazzarini c,,1
a Dipartimento di Fisica dellUniversit di Genova, Via Dodecaneso 33, I-16146 Genova, Italy
b Istituto Nazionale di Fisica Nucleare, INFN, Sezione di Genova via Dodecaneso 33, I-16146 Genova, Italy
c Centre de Physique Thorique, CNRS Luminy, Case 907, F-13288 Marseille Cedex, France
Abstract
The purpose of this note is to show that W3 algebras originate from an unusual interplay between
the breakings of the reparametrization invariance under the diffeomorphism action on the cotangent
bundle of a Riemann surface. It is recalled how a set of smooth changes of local complex coordinates
on the base space are collectively related to a background within a symplectic framework. The
power of the method allows to calculate explicitly some primary fields whose OPEs generate the
algebra as explicit functions in the coordinates: this is achieved only if well defined conditions
are satisfied, and new symmetries emerge from the construction. Moreover, when primary fields
are introduced outside of a coordinate description, the W3 symmetry byproducts acquire a good
geometrical definition with respect to holomorphic changes of charts. 2001 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.Hf
Keywords: Symplectic geometry; -trick, Broken symmetry; W3 -algebras; Anomalies
1. Introduction
It is hard to decide if it is more relevant, from the physical point of view, either the birth
or the decease of a symmetry.
The origin of a symmetry clarifies its deep purest form of manifestation and discloses
its harmonies, while the symmetry breaking mechanisms, even if they display the physical
world faults, shed some new light on the dynamical realization of the symmetry. Indeed the
breaking consistency conditions [1] prevent a wild discharge of the symmetry constraints.
E-mail addresses: beppe@genova.infn.it (G. Bandelloni), sel@cpt.univ-mrs.fr (S. Lazzarini).

1 And also Universit de la Mditerrane, Aix-Marseille II.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 8 - 9
478
G. Bandelloni, S. Lazzarini / Nuclear Physics B 594 (2001) 477500
Moreover, if we are in presence of several symmetry violation mechanisms, a mutual liking

of them could generate a fascinating guide-line leading to a new symmetry.
The purpose of this note is to show that one of the possible origins of W3 [29] algebras
comes from an unusual interplay between diffeomorphism symmetry breakings between
the base space (with local complex coordinates (z, z )) and a well defined chain of spaces
S(n) (z, z ))) whose construction preserves the
(with local complex coordinates (Z (n) (z, z ), Z
symplectic diffeomorphism invariance.
Indeed, the W algebras saga [10] comes from many sources: they were originally
discovered within an OPE construction by Zamolodchikov [11,12] as a natural extension
of the Virasoro algebra; later on they were derived by Drinfeld and Sokolov [13] through a
reduction procedure, which is in fact (classically) simply a Poisson reduction in infinitely
many dimensions, taking as Poisson structure the KirillovPoisson structure naturally
associated to any Lie algebra [14,15].
The same algebras were derived within the Kortevegde Vries hierarchy of equations an
approach which generalized Toda system formalism [16] giving fundamental insights in
the field of the 2D matter physics [17], integrable models [18], topological 2D field theory
[19], 2D conformal field theory [20], 2D quantum gravity [21] as well as matrix models
[15].
So, having a lot of possible sources, one may suppose that a common general trend could
link the widespread variety of these physical and mathematical fields of research.
The two dimensional space exclusive feature is that it admits an infinite dimensional
diffeomorphism group of transformations [22,23] for which higher spin extensions of the
group of representations are allowed.
This fact was already sensed within the W field of interest from the very beginning:
indeed in the instance of W chiral sector, the transformation laws of the BRS ghosts,
C (r)(z, z ), was found to be
X
s C (s)(z, z )C (rs+1)(z, z ) + (stuff).
(1.1)
SC (r) (z, z ) =
s
This gives a slight indication that this symmetry is involved with coordinate transformation
laws (the so-called W diffeomorphisms) [21,2428].
S(r) (z, z )) spaces are
This ansatz addresses both the questions to which the (Z (r) (z, z ), Z
locally diffeomorphic to the (z, z ) base space and what could be the general construction
law which originates them.
Already Witten [29] pointed out, with argument of algebraic topology, that the use of
Poisson brackets induces a kind of symplectic geometry, and gave the conjecture that W
algebras could be related to symplectomorphisms. In a recent paper [30] we have taken
inspiration from this idea in an attempt to provide an alternative approach to these algebras.
S(n) (z, z )) will be
In our scheme a sequence of local complex coordinates (Z (n) (z, z ), Z
found. The latter are built up in such a way to preserve a symplectic form, and turn
out to be the images under canonical diffeomorphisms of the (z, z ) base. Under the
diffeomorphism action on the cotangent bundle, these coordinate transformations give
rise to BRS variations similar to that of Eq. (1.1) and introduce an alternative attempt
479
to a W algebra realization. For example, some questions will find an easy answer: first of
all a Lagrangian formalism can be naturally embedded and the role of complex structure
[24,31] (along the lines of [22]) can be explained [32]. However, not all the W algebras
that can be found in the literature are naturally explained in our formalism, but the W3
algebras in Refs. [2] and [3] acquire a particular role in our approach: on the one hand, it
does not maintain the reparametrization invariance previously introduced, but on the other
hand, the breaking mechanisms are fully under control in our treatment by imposing a
reparametrization invariance in a (z, z , ) space, where is a constant Grassmann variable
which allows to manage in an algebraic way the symmetry breakings with respect to the
(z, z ) background.
This problem is the subject of the paper, and it will be shown that in the most general
case [2], there are two distinct mechanisms for the symmetry breakings, whose accordance
produces a liking mechanism which generates a symmetry criterion.
On the other side, in the particular case of the W3 of [3] a surviving reparametrization
invariance maintains its validity.
As a byproduct of our approach we shall try to use the symmetry as a firepower to
construct primary fields (whose OPE gave historically the origin to W algebras) as explicit
S(n) (z, z )) coordinates. This construction will be possible, and
functions of the (Z (n) (z, z ), Z
in that case an explicit reparametrization invariance will survive under the diffeomorphism
action as in the situation of [3].
Otherwise [2] primary fields have to be introduced as independent fields depending on
the background (z, z ) space. This point of view enhances for a good geometrical definition
of the theory, in particular a well defined anomaly under holomorphic changes of charts
will be found: this feature in not naively shown in the first approach.
Our paper is organized as follows:
In Section 2 our approach to W algebras from symplectomorphisms is reviewed and the
W3 algebras are derived from a general construction.
In Section 3 the W3 algebra is studied as a breaking of the symmetry under
reparametrizations on the base space in a well defined geometric scenario. In particular,
we discuss the different aspects which emerge if an explicit coordinate description of the
algebra is required.
2. W algebras from symplectomorphisms: the chiral Wn -gravity
In this Section we sum up our approach to W algebras in term of symplectomorphisms
we have performed in Refs. [30,32] to which we refer for more details.
The canonical 1-form, , on the cotangent bundle T writes in a local chart frame U(z,y)

|U( z,y) = yz dz + yz d z .
(2.1)
In a frame U(Z,Y ) : will take the form

SS d Z
S .
|U( Z,Y ) = YZ dZ + Y
Z
The (2,0)(0,2) form d is globally defined both in U(z,y) and in U(Z,Y )
(2.2)
480

U (z,y) dyz dz + d yz d z ,
(2.3)

SS d Z
S .
U (Z,Y ) = dYZ dZ + d Y
Z
(2.4)
Recall that a change of charts is canonical if in U(z,y) U(Z,Y )
U (z,y) = U (Z,Y ),
which implies
U (z,y) U (Z,Y ) = dF.
(2.5)
(2.6)
F is a function on U(z, y) U(Z, Y ). In the (z, Y ) plane, an extra coordinate is now

introduced which will turn out to be fundamental in the sequel. A generating function
(z, Y, ) is thus defined as:

S Y, )
SS Z(z,
d(z, Y, ) d F (z, Y, ) + YZ Z(z, Y, ) + Y
Z

= yz (z, Y, ) dz + yz (z, Y, ) d z

S Y, )
SS Z(z,
(2.7)
+ dYZ Z(z, Y, ) + d Y
Z
showing that the mappings:

yz (z, Y, ) = (z, Y, ),
Z(z, Y, ) =
(z, Y, )
YZ
(2.8)
are canonical.
So that in the (z, Y, ) chart, |U(z,Y, ) takes the elementary form:
SS dz Z(z,
S Y, )
= dYZ dz Z(z, Y, ) + d Y
U
Z
(z,Y, )
= dYZ yz (z, Y, ) dz + dYSS yz (z, Y, ) d z

Z
= dz dY (z, Y, ),
(2.9)
SS and (z, z ) respectively.

where dY and dz are the differentials operating in the YZ , Y
Z
S
At this point a crucial remark is in order: in Eq. (2.9) the terms dYZ Z dYZ + dYZ Z
S
d YZS and dz yz dz + dz yz d z will identically vanish in . So an infinitesimal variation
of Z(z, Y, ) in YZ does not modify, for fixed (z, z , ), the 2-form .
From this comment originates the important Theorem [30].
Theorem 2.1. On the smooth trivial bundle R2 , the vertical holomorphic change of
local coordinates,

SS Z (z, z , ), F (YZ ), Y
SS ,
(2.10)
Z (z, z , ), YZ , Y
Z
Z
where F is a holomorphic function in YZ , and the horizontal holomorphic change of local
coordinates,

SS yz f (z), z , YZ , Y
SS ,
(2.11)
yz z, z , YZ , Y
Z
Z
where f is a holomorphic function in z, are both canonical transformations.
481
So the local changes of complex coordinates z Z((z, z ), YZ ) and z Z((z, z ), YZ +

dYZ ), will be related to the same two form .
SS = 0 the generating function will be written as a
Expanding around, say, YZ = 0, Y
Z
formal power series in Y :
(z, Y, ) =
nX
max
max
nX

n (n)
S (z, z , ) .
S Z
YZn Z (n) (z, z , ) +
Y
S
Z
n=1
n=1
(2.12)
The extra coordinate previously introduced is now specified as a constant Grassmann
variable with FaddeevPopov charge equal to 1. One has the splitting
(z, Y, ) = 0 (z, Y ) + (z, Y ),
(2.13)
6 n are local coordinates defined by

r

1
(r)
(z, Y, )
= Z0(r) (z, z ) + Z(r)(z, z ).
Z (z, z , ) =
r! YZ
YZ =0
where the
Z (r)(z, z , ), 1 6 r

(2.14)
We introduce:

1 r
(z, Y, ),
r! YZ

r
(r) (z, z , ) = 1
(z, Y, )
(r) (z, z , )(r, (z, z , )) Z
r! YZ
(r) (z, z , ) Z (r)(z, z , ) =
(2.15)
and we shall denote (1) (z, Y, ) =: (z, Y, ), (1, (z, Y, )) =: (z, Y, )

Note that for a given level r the coordinate Z (r)(z, z , ) is related to other ones with
index less than r by [30]

r
1
(r)
((z, Y, ))
Z (z, z , ) =
r! YZ
YZ =0

mj
a
r
X
Y
(Z (pi ) (z, z , )) i
(j )
X
) M (z, z
=
j!
, )
( X
ai = j ,
ai p i = r

ai !
j =1
i=1
i
i
p 1 > p 2 > > p mj
(2.16)
Mj (z, z , )
are given by:

for all 0 6 r 6 n and where the functions

j
1
1
(j )
(z, Y, )
M (z, z , )

j ! (z, Y, ) YZ
YZ ,YZ =0
(j )
(j )
M0 (z, z ) + M (z, z ).
1
(z,Y,) YZ
(2.17)
can be formally defined using the usual GrassThe differential operator

mannian inverse procedure

(z, Y )
1
=
D0 (z, Y ) + D (z, Y ) (2.18)
(z, Y, ) YZ
0 (z, Y )
(0 (z, Y ))2 YZ
so that for 1 6 r 6 n
482

r
1
D0 (z, Y ) 0 (z, Y )Y =0 ,
Z
r!

1
r
D0 (z, Y ) (z, Y ),
)
M(r)
(z, z
r!

X

(nj )
j
D0
(z, Y )D (z, Y )D0 (z, Y )0 (z, Y )
+
(r)
M0 (z, z )
j =0,...,r1
. (2.19)
YZ =0
The previous geometrical structure allows us to derive now a W-symmetry in terms

of the algebra of diffeomorphisms combining both the diffeomorphism action and
the canonical transformations via the BRS machinery. Calling S the BRS differential
operator for the diffeomorphism action on the cotangent bundle [30,35], according to the
decomposition (2.13), one has
S(z, Y, ) = (z, z , ) 0 (z, Y ) + (z, Y ),
(2.20)
where is a constant field with FaddeevPopov charge equals to 1, and we shall fix its
BRS variation as S = 1. One identifies
0 (z, Y ) = S0 (z, Y ) (z, Y ),
(z, Y ) = S (z, Y )
(2.21)
and the nilpotency condition on the generating function, S 2 (z, Y ) = 0 means:

S0 (z, Y ) = 1 (z, Y ),
S (z, Y ) = 0.
(2.22)
So for each n we define the diffeomorphism action on the local complex coordinate Z (r)
by,

1 r
(r)
(r)
, 1 6 r 6 n, (2.23)
(z, Y, )
SZ (z, z , ) (z, z , ) =
r! YZ
YZ =0
which, once decomposed according to its content, gives:

1
r
(r)
(r)
0 (z, Y )
,
SZ0 (z, z ) Z (z, z ) =
r! YZ
YZ =0

1
r
(r)
(z, Y )
.
SZ (z, z ) =
r! Y
Z
(2.24)
YZ =0
One can easily verify, that S 2 Z (r) (z, z , ) = 0 in the (z, z , ) space (as in Eq. (2.23)), but
note that S 2 Z0 (z, z ) 6= 0 in the (z, z ) space (as in Eq. (2.24)) which leads to a diffeomorphism symmetry breaking through the local smooth changes of complex coordinates
S(r) (z, z )), while the one for (z, z ) (Z (r)(z, z , ), Z
S(r)(z, z , ))
(z, z ) (Z (r)(z, z ), Z
holds its full validity.
This shows the important role played by the field, and how the decomposition can
be managed in the present treatment: in order to fix all the algebraic relations, calculations
must be performed in the (z, z , ) space, and only after the decomposition is performed.
To sum up the construction:
483
Statement 2.1.
The introduction of the coordinate in the (z, z ) space gives rise to a superselection
sector where the BRS breaking terms of the diffeomorphism transformations (z, z )
S(r) (z, z )) lie.
(Z (r) (z, z ), Z
Moreover, the consistency conditions of these breakings and the B.R.S. behaviour of the
S(r)(z, z , )).
field preserves the diffeomorphism symmetry (z, z ) (Z (r)(z, z , )Z
The diffeomorphism, ghosts in the (z, z , ) space, are introduced as usual [30] and can
also be written with respect to the decomposition
K(r) (z, z , ) =
(r) (z, z , )
K0(r) (z, z ) + K(r) (z, z )
Z (r)(z, z , )
(2.25)
with the following transformation laws

SK(r) (z, z , ) = K(r) (z, z , )K(r) (z, z , )
(2.26)
SK0(r) (z, z ) = K0(r) (z, z )K0(r)(z, z ) + K(r) (z, z ),
(2.27)
and
(r)
(r)
(r)
(r)
(r)
SK = K0 (z, z )K (z, z ) K (z, z )K0 (z, z )
(2.28)
respectively.
S(r)(z, z )) spaces brings to a decomposition [30] of the
The nesting of the (Z (r) (z, z ), Z
ghost (r) (z, z , ) into the spaces of order lower than r:

pi
ai
r
Y (Z
X
(z,
z
))

z
(r)
) Y (z, z , ),
X
j!
mj
(2.29)
(r) (z, z , ) =
( X
ai = j ,
ai p i = r

ai !
j =1
i=1
i
i
p 1 > p 2 > > p mj
where it has been defined for 1 6 r 6 n

Y (r) (z, z , ) = C (r) (z, z ) + X (r) (z, z ) ,

r
1
(r)
D0 (z, Y ) 0 (z, Y )
,
C (z, z ) =
r!
YZ =0

r
1
D0 (z, Y ) 1 (z, Y ) + r (z, Y )0 (z, Y )
.
X (r) (z, z ) =
r!
YZ =0
(2.30)
With some tedious but straightforward algebraic manipulations the BRS variations of
the previous fields can be computed by using:

(z, z )
D(z, z )
(2.31)
S, D(z, z ) = C(z, z ) ln 0 (z, z ) + C(z, z ) +
0 (z, z )
(from now on the summation procedure is explicitly expressed: repeated indices do not
mean summation). One obtains as said in the introduction variations of the type of Eq. (1.1),
484
SC (r) (z, z ) =
r
X
sC (s)(z, z )C (rs+1)(z, z ) + X (r) (z, z ),
r <n
(2.32)
s=1
and, to insure nilpotency,

SX (r) (z, z ) =
r
X

sC (s)(z, z )X (nr+1) (z, z ) sX (s) (z, z )C (rs+1)(z, z )
s=1
r
X

(r s + 1)X (rs+1)(z, z )C (s)(z, z ) sC (s)(z, z )X (rs+1)(z, z ) .
s=1
(2.33)
The X (r) (z, z ) fields are still undetermined objects: our aim is to fix a recipe in order to
identify them as functions of the C (r)(z, z ) ghost fields; we now show that if we impose by
hand
X (n) (z, z ) = 0
(2.34)
then all the X (r) (z, z ), r < n fulfill our requirements. Indeed from the fact that the ghost
of maximum order C (n) (z, z ) has no breaking term, that is:
SC (n) (z, z ) =
s=n
X
sC (s)(z, z )C (ns+1)(z, z ),
(2.35)
s=1
so we can derive:
S C
2 (n)
(z, z ) =
" n1
X
sX (s) (z, z )C (ns+1)(z, z )
s=1
n
X
#
sC (s)(z, z )X (ns+1) (z, z ) = 0,
(2.36)
s=2
and try the more general solution as a differential polynomial in the ghost fields
X (nr+1) (z, z ) =
n
X
0
0
r r C (t )(z, z ) m m C (l)(z, z )
r=2 r 0 ,t,m,m0 ,l>0

0
T (ns+1lt +m+r),(r +m ) (z, z )

we obtain (there is no summation):
X (nr+1) (z, z ) = C (r)(z, z )C (r)(z, z )T (n3r+2) (z, z )

r
(2r+3)
(r)
2 (r)
(r) 3 (r)
C C (z, z ) (nr+1)
.
+ (nr+1) C (z, z ) C (z, z )
n+1
(2.37)
So the existence of terms is possible only if the condition n 3r + 4 = 0 is solved by n

and r integers. This is achieved for example for n = r = 2 which turns out to be the W(3)
case. Rewriting the above solution for 1 6 r < n as
485
X (r) (z, z ) = C (nr+1) (z, z )C (nr+1) (z, z )T(2n3r+1) (z, z )

+ (r) C (nr+1) (z, z ) 2 C (nr+1)(z, z )

n r + 1 (nr+1) 3 (nr+1)
2(nr+1)3
C
C
(z, z ) (r)
,
n+1
(2.38)
we then substitute into Eq. (2.32) in order to get the properties of coefficients T .
0
The terms of the type C () C ( ) (Stuff) fix the BRS variations of T(2n3r+1)(z, z ) the
0
00
other terms must cancel. Let us first consider the monomials C () C ( ) 2 C ( ) . The terms
(1)
r
X

(s)
sC (z, z )C (n(rs+1)+1)(z, z ) 2 C (n(rs+1)+1)(z, z )T(2n3(rs+1)+1)(z, z ) ,
s=1
(2)
(nr+1)
nr+1
X

sC (s) 2 C (nrs+2) T(2n3r+1)
(2.39)
s=1
cancel each other only for s = 1 while for s > 1 they give inconsistencies; so a priori we
have to put in the term (1) r = 1. In this case this term collapses into
C (1)(z, z )C (n) (z, z ) 2 C (n) (z, z )T(2n2) (z, z ),
(2.40)
which is the first term (s = 1) of the series (2). The remaining terms in the latter can be
dropped out giving prescription for the parameter n.
The a priori logical choice would be n = 1 which reproduces the previous term, and
trivializes the W content. But it has to be noted that if n = 2 the top term of (2.40) is zero
for obvious FaddeevPopov reason and the cancellation mechanism is then achieved.
For the highest value of n the cancellation mechanism does not hold unless all the ghosts
(j
C ) (z, z ); j = 2 n 1 are put to be zero and we next repeat the same FaddeevPopov
trick for the nth term of the sum in order to get
SC(z, z ) = C(z, z )C(z, z ) + C (n) (z, z )C (n) (z, z )T(2n2) (z, z )

n
(n)
C (n) 3 C (n) (z, z ) (2) ,
+ C (n)(z, z ) 2 C (n) (z, z )
n+1
SC (n) (z, z ) = C(z, z )C (n) (z, z ) + nC (n) (z, z )C(z, z )
(2.41)
for n 6= 2 this algebra was found in [37]. So we conclude:

Statement 2.2.
The most general algebra of the type Eqs. (2.32), (2.33) with constraints Eqs. (2.35),
(2.34) is given in Eq. (2.41). In particular, the W3 algebra is the only one which contain
an dependent term.
2.1. The chiral W3 -gravity algebra
We specialize here to the case n = 2 (W3 ) whose peculiarity among all the other
situations is stated just above: it is the only case which admits 6= 0. We shall see that the
486
presence (or the absence) of a term proportional to is fundamental for the geometrical
setting of the problem.
For different values of n the general discussion can be easily performed along the lines
here for n = 2.
From (2.35), let us recall that the highest order X term has been put to zero in order to
fix all the lowest order terms; here for n = 2 we have to impose:

1
(z, Y ) 2
D 0 (z, Y )
X (2) (z, z ) D02 1 (z, Y )
2
0 (z, Y ) 0

(z, Y )
D0 (z, Y )
= 0.
(2.42)
D0
0 (z, Y )
YZ =0
This constraint has a twofold face. The first one concerns the implications on the Cs
algebra, as in the previous section. The second involves the role that X (2) (z, z ) is supposed
to play in the symplectic formalism by exploiting all the geometrical aspects emerging
from the vanishing of the previous formula. Only the former will be discussed here, the
latter being postponed to the next section.
First of all, the condition X (2) (z, z ) = 0 implies:
SC (2) (z, z ) = C(z, z )C (2)(z, z ) + 2C (2)(z, z )C(z, z )
(2.43)
and setting X := X (1) Eq. (2.36) yields

X (z, z )C (2)(z, z ) = 2C (2)(z, z )X (z, z ),
(2.44)
which is solved by
16
X (z, z ) = C (2)(z, z )C (2)(z, z ) T (z, z )
3

2
(2)
2 (2)
+ C (z, z ) C (z, z ) C (2)(z, z ) 3 C (2) (z, z ) ,
3
(2.45)
where we have been forced to introduce a spin (2, 0)-conformal field T (z, z ). Summing
up, we find the algebra:
16
SC(z, z ) = C(z, z )C(z, z ) T (z, z )C (2)(z, z )C (2)(z, z )
3

2
+ C (2)(z, z ) 2 C (2) (z, z ) C (2)(z, z ) 3 C (2)(z, z ) ,
3
SC (2) (z, z ) = C(z, z )C (2)(z, z ) + 2C (2)(z, z )C(z, z ).
(2.46)
Note that the X (z, z ) as in our primary purpose in Eq. (2.30), expresses the breaking
at the level of the diffeomorphism algebra. It depends only on the C (2)(z, z ) ghost and its
derivative; this moves us to make the following remark:
Remark 2.1. In the limit of C (2) (z, z ) going to zero, we find the diffeomorphism
S z )) reparametrization. So the ghost
algebra describing the ordinary (z, z ) (Z(z, z ), Z(z,
(2)
C (z, z ) parametrizes the breaking of this symmetry at the level of BRS algebra.
487
The BRS variation of T (z, z ) is obtained from Eq. (2.41) leading to:
ST (z, z ) = C(z, z )T (z, z ) + 2T (z, z )C(z, z ) W(z, z )C (2)(z, z )
2
(2.47)
C (2)(z, z )W(z, z ) + 3 C(z, z ),
3
in terms of a spin (3, 0)-conformal field W whose BRS behaviour can be calculated from
the nilpotency condition applied to (2.47),

SW(z, z ) = C(z, z )W(z, z ) + 3C(z, z )W(z, z ) + 16T (z, z ) C (2)(z, z )T (z, z )
+ 5 C (2) (z, z ) + 2C (2)(z, z ) 3 T (z, z ) + 10T (z, z ) 3 C (2)(z, z )

(2.48)
+ 15T (z, z ) 2 C (2)(z, z ) + 9 2 T (z, z )C (2)(z, z ) .
The algebra here closes, since the nilpotency condition holds whatever . In order to realize
the role played by the parameter , the stability under holomorphic changes of charts z
w(z) of the algebra defined by the Eqs. (2.46)(2.48) is now discussed. One has
1
z
(2.49)
w0
and the glueing rule of T will be also worked out. Writing first (2.46) in the w system of
complex coordinates and then going back to the z one, we have
C w = w0 C z ,
C ww = (w0 )2 C zz ,
w =
1
z (w0 C z ) = w0 C z z C z ,
w0
16
16
Tww C ww w C ww = (w0 )3 Tww C zz z C zz ,
3
3
while the -term is more involved

1
1
0 2 zz 1
0 2 zz
z (w ) C
z 0 z (w ) C
w0
w0
w

1
1 0 2 zz
2 0 2 zz 1
z 0 z
z (w ) C
(w ) C
3
w0
w
w0

2 zz 3 zz 16
0
zz 2 zz
zz
zz
= w z C z C C z C {w, z}C z C ,
3
3
C w w C w = w0 C z
where {w, z} denotes the Schwarzian derivative. Covariance requires that

(w0 )3 Tww + w0 {w, z} C zz z C zz = w0 Tzz C zz z C zz ,
(2.50)
so that
(w0 )2 Tww + {w, z} = Tzz
(2.51)
showing that 1 T is a projective connection. On the other hand if = 0, T is a tensor. It is

also easy to recover from Eqs. (2.47), (2.48):
(w0 )3 Wwww = Wzzz ,
(2.52)
showing that W behaves as a true tensor of order three.

The solution can be found if we notice that the parameter can be reabsorbed in the
BRS operator by rescaling:
488
C (2)(z, z ) = 1/2 C (2)(z, z ),

W(z, z ) =
3/2
T (z, z ) = T (z, z ),
W(z, z ).
(2.53)
So if 6= 0 we can fix it to be equal to one without any trouble. In the case = 1 the BRS
transformations of T (z, z ) and W(z, z ) can be rewritten in terms of Bol derivatives [18].
3. W3 algebra and coordinate transformations: two different approaches for the

same symmetry
The purpose of this section is to discuss the algebra in Eqs. (2.46), (2.47), (2.48)
found previously in terms of coordinate transformations. Indeed, we have introduced
the coordinates Z (r) (z, z , ) = Z0(r) (z, z ) + Z(r) (z, z ), r = 1, 2 in Eq. (2.14) and the
reparametrizations:

S z , ) ,
(z, z ) Z(z, z , ), Z(z,

S(2)(z, z , ) ,
(3.1)
(z, z ) Z (2)(z, z , ), Z
whose algebra under symplectomorphisms reads: coordinates represents the breaking
terms of the diffeomorphisms:
SZ(z, z , ) = K(1) (z, z , )Z(z, z , ),
,SZ (2) (z, z , ) = K(2) (z, z , )Z (2)(z, z , )
(3.2)
with ghosts:
K(1) (z, z , ) = C(z, z ) + X (1) (z, z ),
(Z0 (z, z ))2 (2)
C (z, z )
K(2) (z, z , ) = C(z, z ) +
Z0(2)(z, z )

(Z0 (z, z ))2 (2)
X (z, z )
+ X (1) (z, z ) +
(2)
Z0 (z, z )
{z
}
|
(3.3)
vanishing in W3

(2)
Z (z, z )Z0(z, z ) Z (z, z )(Z0 (z, z ))2 (2)
+ 2
(z,
z
)
.
C
2
(2)
Z0(2)(z, z )
Z (z, z )
(3.4)
The previous symmetry is broken at the (z, z ) level, such that the mappings:

S0 (z, z ) ,
(z, z ) Z0 (z, z ), Z

S(2) (z, z ) ,
(z, z ) Z0(2)(z, z ), Z
0
(3.5)
do not yield any reparametrization.

However, a closed algebra can be written in terms of the BRS operation acting on the
previous coordinates:
489
SZ0 (z, z ) = C(z, z )Z0 (z, z ) + Z (z, z ),

SZ (z, z ) = C(z, z )Z (z, z ) Z0 (z, z )X
(3.6)
(1)
(z, z ),
(3.7)
(2)
(2)
(2)
SZ0 (z, z ) = C(z, z )Z0 (z, z ) + C (2)(z, z )(Z0 (z, z ))2 + Z (z, z ),
SZ(2) (z, z ) = C(z, z )Z(2) (z, z ) X (1) (z, z )Z0(2)(z, z )
2Z0(z, z )Z (z, z )C (2)(z, z ) X (2) (z, z )(Z0 (z, z ))2 ,
{z
vanishing in W3
(3.8)
(3.9)
where the 0 label identifies good coordinate frames and the ones the breakings of the
symmetry under reparametrizations.
So this algebra contains anomalies X (z, z ) and X (2) (z, z ) which in the same formalism
can be expressed in terms of coordinates as:
X (z, z ) := X (1) (z, z )
X (2) (z, z )
(z, z )
Z
,
Z0 (z, z )

Z (2)(z, z )
(z, z ) 0
(2) (z, z ) Z
Z
Z0 (z, z )
(Z0 (z, z ))2

2
(2) Z (2) (z, z ) ,
(Z (z, z )) Z
0
Z0 (z, z )
(3.10)
(3.11)
where we have introduced the operator associated to [BRS] := S

z ) [BRS] C(z, z ) .
(z,
(3.12)
(z, z ) and
These parameters are related to the breaking consistency conditions Z
Z (2) (z, z ): since the realization of the W3 algebra requires, as shown before, the constraint
X (2) (z, z ) = 0, we have to investigate in which manner this condition could affect these
transformations. In the symplectic framework and from Eqs. (2.30), (2.42) the previous
condition will imply
(z, z )
(2) (z, z ) = Z
Z
(2)
Z0 (z, z )
+ 2C (2)(z, z )Z0 (z, z )Z (z, z ),
Z0 (z, z )
(3.13)
when written in terms of coordinates. This formula encodes the geometrical content of the
W3 symmetry.
(z, z ) and
(2) (z, z ) ,Z
So this interplay between the breakings consistency condition Z
coordinate transformations induces the symmetry criterion which gives the origin to W3
algebra through the condition X (2) (z, z ) = 0. Hence we can state:
Statement 3.1. The condition X (2) (z, z ) = 0 not only fixes the breaking terms of
the differential algebra, but also generates a mutual conspiracy between the breaking
consistency conditions: the relationship between these violation mechanisms generates the
coordinate counterpart symmetry of the W3 algebra.
490
3.1. Primary fields as explicit functions of coordinates

This part of our paper wants to investigate an intriguing aspect of our problem to which
our approach gives a full meaning.
The conclusion reached in the previous statement (3.1) transforms our algebra with
the help of a differential complex for the infinitesimal transformations written in terms
(2)
of the coordinates Z0 (z, z ) and Z0 (z, z ) in presence of breaking terms Z (z, z ) and
Z(2) (z, z ). From the physical (and even the mathematical) point of view a coordinate
system represents a tool which is hard to do without.
Our approach embeds them in a (z, z ) background whose presence gives a meaning to
our model; nevertheless the explicit presence of coordinates amounts to investigating the
possibility of writing Z (z, z ) and Z(2) (z, z ) as explicit expression of both Z0 (z, z ) and
(2)
Z0 (z, z ).
We have seen before in the ansatz (2.1) that the FaddeevPopov content of the breaking
terms can only be carried by the C (2) (z, z ) so we argue as follows. Let us consider in a
matrix-like notation the following ansatz.
Ansatz 3.1.
(
)
(
)
Z
X
P(2r)
Z0 (z, z ), Z0(2) (z, z )
Z (z, z )
r (2)
,
C (z, z )
=
(2)
(2)
Z(2) (z, z )
RZ
(2r) Z0 (z, z ), Z0 (z, z )
r>0
(3.14)

(2)
(2)
Z
Z0 (z, z ), Z0(2) (z, z ) and RZ
where P(2r)
(2r) Z0 (z, z ), Z0 (z, z ) are taken to be functions of both Z0 (z, z ) and Z0(2) (z, z ) and their derivatives. One computes
)
(
Z (z, z )
(2)
Z (z, z )
(
!
)
r
Z
XX
Z0 (z, z ), Z0(2) (z, z )
P(2r)
r
j
rj +1 (2)
C (z, z )
C(z, z )
=

(2)
(2)
j
), Z0 (z, z )
RZ
(2r) Z0 (z, z
r>1 j =1
r
XX
r>0 j =0
r (2)
r
j
(z, z )
r>0
n
!
j +1
C(z, z )
"
n>0
C(z, z )Z0 (z, z ) +
rj (2)
(z, z )
n
( Z0 (z, z ))
X
( n Z0(2) (z, z ))
(2)
(2)
(2)
)
(2)
(z, z )
(2)
)!
(2)
Z
P(2s)
Z0 (z, z ), Z0 (z, z )

Z
Z0 (z, z ), Z0(2) (z, z )
P(2s)
Z
P(2r)
Z0 (z, z ), Z0 (z, z )
(2)
(2)
RZ
(2r) Z0 (z, z ), Z0 (z, z )
RZ
(2r) Z0 (z, z ), Z0 (z, z )
s>0
)
Z
P(2r)
Z0 (z, z ), Z0 (z, z )
(
s (2)
(2)
Z
Z0 (z, z ), Z0 (z, z )
P(2r)
)
), Z0(2) (z, z )
RZ
(2r) Z0 (z, z
(2)
n C(z, z )Z0 (z, z ) + C (2) (z, z ) Z(0)(z, z )
(
C
s (2)
(z, z )
s>0
(
C(z, z )
(2)
(2)
RZ
)
(2s) Z0 (z, z ), Z0 (z, z
(2)
(2)
)
RZ
(2s) Z0 (z, z ), Z0 (z, z
(2)
)!
(2)
Z
Z0 (z, z ), Z0 (z, z )
P(2r)
(2)
)
RZ
(2r) Z0 (z, z ), Z0 (z, z
491
2
)!#
(3.15)
The previous variations have FaddeevPopov charge equal to two and contain both
s C(z, z ) r C (2)(z, z ) and n C (2)(z, z ) m C (2)(z, z ) ghost monomials. In order to implement
Eqs. (3.7) and (3.9), recall that in the solution (2.45) for X (z, z ), all the mixed terms
s C(z, z ) r C (2)(z, z ) have to cancel out. This gives rise to the following set of constraints
on both the unknowns P and R. It will be useful for the sequel to proceed as follows;
we have two distinct systems of conditions. First the vanishing of the coefficient of the
monomials C (2) r+1 C, r > 0, yields
(
(
)
)
(2)
(2)
Z
Z
Z0 (z, z ), Z0 (z, z )
Z0 (z, z ), Z0 (z, z )
P(2r)
1 (r) P(2)
= V
, r > 0, (3.16)
(2)
(2)
(2)
(2)
2
), Z0 (z, z )
), Z0 (z, z )
RZ
RZ
(2r) Z0 (z, z
(2) Z0 (z, z
and the one for the monomials r C (2) s+1 C, r > 1, s > 0 gives rise to
(
)
(2)
Z
Z0 (z, z ), Z0 (z, z )
P(2(r+s))

(2)
(2)
), Z0 (z, z )
RZ
(2(r+s)) Z0 (z, z
(
)
(2)
Z
)
r!(s + 1)!
(s) P(2r) Z0 (z, z ), Z0 (z, z
V
=
,
(2)
(2)
(s + r)!(2(s + 1) r)
), Z0 (z, z )
RZ
(2r) Z0 (z, z
for r > 1, s > 0, and r 6= 2(s + 1),
(3.17)
where we have set for r > 0:

X n
(r)
nr Z0 (z, z )
V =
r +1
( n Z0 (z, z ))
n>(r+1)
nr
(2)
Z0 (z, z )
n Z0(2)(z, z )

.
(3.18)
It is worthwhile to remark that V (0) is the counting operator with respect to the little indices
z. Note however that the conditions (3.16) is contained in (3.17) by allowing for r = 0. This
decomposition of the constraints leads advantageously to the following
Z
Z0 (z, z ),
Statement 3.2. Owing to Eq. (3.17) for r, s > 0, the functions P(2(r+2))

(2)
(2)
(2)
Z
Z0 (z, z ) and R(2(r+s)) Z0 (z, z ), Z0 (z, z ) can be completely fixed in terms of

(2)
(2)
(2)
Z
Z0 (z, z ), Z0 (z, z ) and RZ
P(2r)
(2r) Z0 (z, z ), Z0 (z, z ) , respectively, in an independent way.
492
The operator V (r) defined in Eq. (3.18) decreases the order of all derivatives by r.

Z
Z Z (z, z
), Z0(2) (z, z )
Z0 (z, z ), Z0(2) (z, z ) in terms of P(2)
Thus the construction of P(2r)
0
means that, if m
e is the highest order of derivative terms of Z(z, z ) or Z (2) (z, z ) contained

(2)
Z
Z
in P(2) Z0 (z, z ), Z0(2) (z, z ) , then the functions P(2e
mr) Z0 (z, z ), Z0 (z, z ) , r > 1 will
be zero.

Z Z (z, z ), Z (2) (z, z
) involving
In particular, any V (r) for r > 1 gives no action on P(2)
0
0
(2r)
zero and first order derivative terms. The same argument holds for RZ
(2)

(2)
(2)
(2)
Z
Z0 (z, z ) relation to R(2) Z0 (z, z ), Z0 (z, z ) .
Z0 (z, z ),
Z Z (z, z
),
An important remark is in order. By acting with V (r) on the lower states P(2)
0

(2)
Z(2)
(2)
Z0 (z, z ) and R(2) Z0 (z, z ), Z0 (z, z ) all the upper states can be constructed. But due

Z(2)
Z
to (3.17) the states P(2(r+s))
Z0 (z, z ), Z0(2)(z, z ) and R(2(r+s))
Z0 (z, z ), Z0(2) (z, z )

(2)
Z(2)
(2)
Z
Z0 (z, z ), Z0 (z, z ) and R(2r) Z0 (z, z ), Z0 (z, z ) acting with
are obtained from P(2r)
V (s) only if r 6= 2(s + 1). The latter limits the cancellation mechanism of the r C (2) s+1 C
monomials in Eq. (3.15) only to selected states (since the sum (r + s) can be reached in a
non unique way without spoiling the inequality r 6= 2(s + 1)) in the decomposition (3.14).
The previous cancellation mechanisms could hide an a priori inconsistency: this is not
the case. In fact the local operators V (r) defined in (3.18) verify the algebra

(r + s + 1)!
V (r+s),
V (r), V (s) = (r s)
(r + 1)!(s + 1)!
(3.19)
so that Eqs. (3.16) and (3.17) can be related. Indeed, combining both Eqs. (3.16) and (3.19)
one obtains
(
(
)
)
Z
Z
Z0 (z, z ), Z0(2)(z, z )
Z0 (z, z ), Z0(2) (z, z )
P(2(r+s))
V (r+s) P(2)
=

(2)
(2)
(2)
(2)
2
), Z0 (z, z )
RZ
RZ
(2(r+s)) Z0 (z, z
(2) Z0 (z, z ), Z0 (z, z )
(
)
(2)
Z
(s + 1)!(r + 1)!
(s) P(2r) Z0 (z, z ), Z0 (z, z )
V
=

(2)
(s + r + 1)!(s r)
), Z0(2) (z, z )
RZ
(2r) Z0 (z, z
(
)!
(2)
Z
(r) P(2s) Z0 (z, z ), Z0 (z, z )
.
V

(2)
(2)
RZ
(2s) Z0 (z, z ), Z0 (z, z )
Using Eq. (3.17) we get an identity which holds for r 6= 2(s + 1) and s 6= 2(r + 1). The
(2)
Z
and/or RZ
latter formula shows moreover that P(2(r+s))
(2(r+s)) can be obtained without
any restriction. Hence (3.15) reduces to
)
(
(z,
z
)
Z
Z(2)(z, z )
(
)
(2)
Z
X
Z0 (z, z ), Z0 (z, z )
V (r) P(2)
s (2)
r (2)
C (z, z ) C (z, z )L(2s) (z, z )
=
, (3.20)
(2)
(2)
2 RZ
)
(2) Z0 (z, z ), Z0 (z, z
r,s>0
where the differential operator L(s) is defined by
L(2s) (z, z ) :=
493

(l)
Xn
(2)
nk V
P(2) Z0 (z, z ), Z0 (z, z )
n
k
2
( Z0 (z, z ))
k, l > 0 n>k
k+l =s
!

(l)
(2)
nk V
Z (2)
R(2) Z0 (z, z )Z0 (z, z )
+

(2)
2
n Z0 (z, z )
Xn
ns (Z0 )2 (z, z )
+
.
(2)
n
s
Z (z, z )
n>s
(3.21)
The L(2s) (z, z ) operator looks like a coordinate transformation operator which maps

(2)
Z0 (z, z ) and its derivatives into particular derivatives of P(2l) Z0 (z, z ), Z0 (z, z ) and

(2)
(2)
(2)
similarly for Z0 (z, z ) with respect RZ
(2l) Z0 (z, z ), Z0 (z, z ) plus an inhomogeneous
term depending on Z0 (z, z ).
(z, z ) has a well defined FaddeevPopov ghost content due to the condition
But Z
(2.45). Only the C (2) (z, z )C (2)(z, z ), C (2)(z, z ) 3 C (2)(z, z ), and C (2)(z, z ) 2 C (2)(z, z )
ghost monomials are present in the expansion Eq. (3.20): the first term will provide a
Z0 (z, z ), Z0(2) (z, z ) for the T function; while the other ones give the term in Eq. (2.45).
All the other terms must be zero.
These conditions provide a system of equations for P and R.
A capital role is played by the equations for terms, since they provide a coordinate
(2)
representation for the constant which has to be true for each Z0 (z, z ) Z0 (z, z ) change of
charts. This covariance requirement, combined with the algebraic structure of the two local
operators V (s) and L(2s) show that only the value = 0 is consistent for this approach.

(2)
L(2s) P(2r) Z0 (z, z ), Z0 (z, z ) ,

(2)
(3.22)
= L(2r) P(2s) Z0 (z, z ), Z0 (z, z ) , (r + s) 6= 1.
This equation can be treated as an integrability condition and contains both a linear
(2)
(2)
(2)
), Z0 (z, z )
and a bilinear part in the P(2r) Z0 (z, z ), Z0 (z, z ) and the RZ
(2r) Z0 (z, z
functions.
Filtering with the counting operators of the previous functions we select the linear
part
(2)
of Eq. (3.22). So we have an infinite set of conditions on P(2r) Z0 (z, z ), Z0 (z, z ) .
For r = 0 , s > 1 we get:

(2)
(2)
lin
) ,
(3.23)
Llin
(2s) P(2) Z0 (z, z ), Z0 (z, z ) = L(2) P(2s) Z0 (z, z ), Z0 (z, z
where:
Llin
(2s) =
Xn
n>s
ns (Z0 )2 (z, z )

(2)
n Z0 (z, z )
(3.24)
(r)
This condition must be
compatible with the V algebra stated in the Eq. (3.16) for
(2)
P(2) Z0 (z, z ), Z0 (z, z ) .
The form of the operator Llin
(2s) forbids this possibility; the only way to escape is to
prevent the dependence of these function on the derivatives of Z0(2) (z, z ) with order greater
494
or equal than two. Going on with the same filtration mechanism we forbid the dependence
on the derivatives of Z0 (z, z ) of the same order.
(2)
This shows that Z (z, z ) depends only on the derivatives of Z0 (z, z ) and Z0 (z, z ) of
(2) (z, z ) in the same
order less than two: now the same argument can be repeated for Z
FaddeevPopov and derivatives sectors as before;

so
we
obtain
an
analogous equation

(2)
(2)
Z
as Eq. (3.22) for the R(2r) Z0 (z, z ), Z0 (z, z ) functions obtaining the same results as

(2)
(2)
before for the RZ
(2) Z0 (z, z ), Z0 (z, z ) functions.
So we get:

Z Z (z, z ), Z (2) (z, z ) ,
Statement 3.3. The only non zero functions in Eq. (3.14) will be P(2)
0
0

(2)
), Z0(2) (z, z ) .
and RZ
(2) Z0 (z, z
The other functions will be zero if, as stated by Eq. (3.16), the previous functions depend
(2)
only on Z0 (z, z ) and Z0 (z, z ) and their first order derivatives.

(2)
(2)
Z Z (z, z
), Z0(2) (z, z ) , RZ
), Z0(2) (z, z ) and T (z, z ) are
This means that P(2)
0
(2) Z0 (z, z
well-defined tensors under holomorphic change of charts.
Assuming the general expressions:

Z
Z0 (z, z ), Z0(2) (z, z )
P(2)

X Y mi
ji ni (2)
ji
X
j
j
Ai
Z0 (z, z )
Z0 (z, z )
=
ji + 2ji = 1

j
i

fi
(2)
(2)
Bi
lji
(3.25)
Z0 (z, z )
ji
hij

(2)
Z0 (z, z )
ji
X
i
i

j + 2j = 2

j
(2)
(2)

,
2
Z0 (z, z )
Z0 (z, z )
(Z0 (z, z ))
(Z0 (z, z ))
,
2
Z0(2)(z, z )

gi
ji mij + ji nij = 2, i, j

,
,
(Z0 (z, z ))2 (Z0 (z, z ))2
Z0(2) (z, z )
RZ
(2) Z0 (z, z ), Z0 (z, z )
X
X
j
ji lji + ji hij = 2, i, j
(3.26)
where fi , gi are arbitrary scalar functions, with the constraints imposed in Eqs. (3.25),
(3.26) by the Statement (3.3):
mij , nij , lji , hij = 0, 1.
(3.27)
At this stage the role of Eq. (3.13) appears as the condition which
links together the
(2)
(2)
Z
(z,
z
),
Z
(z,
z
)
, since
Z0 (z, z ), Z0(2) (z, z ) and RZ
Z
functions P(2r)
(2r) 0
0
(2)
(2)
Z0 (z, z )L(1) (z, z )RZ
(2) Z0 (z, z ), Z0 (z, z )

Z
= Z0(2)(z, z )L(1) (z, z )P(2)
Z0 (z, z ), Z0(2)(z, z )

(2)
Z
Z0 (z, z ), Z0 (z, z ) ,
+ 2(Z0 (z, z ))2 P(2)
495

(3.28)
we get

(Z0(z, z ))3
(2)
Z
,
Z0 (z, z ), Z0 (z, z ) = A
P(2)
Z0(2) (z, z )

(2)
(2)
RZ
))2 ,
(2) Z0 (z, z ), Z0 (z, z ) = B(Z0 (z, z
so
(3.29)
B = A 1,

(Z0 (z, z ))2
(2)
Z0 (z, z ),
SZ0 (z, z ) = C(z, z ) + AC (z, z )
Z0(2)(z, z )

(Z0 (z, z ))2
(2)
(2)
Z0 (z, z ),
SZ0 (z, z ) = C(z, z ) + AC (2)(z, z )
Z0(2) (z, z )
(3.30)
(3.31)
so that for a rescaling C (2)(z, z ) A1 C (2)(z, z ) we find a perfect covariance of Z0 (z, z )

(2)
Z0 (z, z ) under the ghost

(Z0 (z, z ))2
(2)
.
K2 (z, z ) = C(z, z ) + C (z, z )
(2)
Z0 (z, z )
Finally we find:

2
3 (Z0 (z, z ))2
,
T (z, z ) =
8 Z (2) (z, z )
0

3
3 (Z0 (z, z ))2
.
W(z, z ) =
4 Z (2)(z, z )
0
(3.32)
(3.33)
Indeed, it is easy to verify that the Eq. (3.28) is fully equivalent to say that the part of
Eq. (3.4) becomes zero.
The explicit dependence on the coordinates reconstructs the diffeomorphism symmetry.
In the case of the algebra Eq. (2.41):
If we put:
Z = AC (n) (z, z )
(n)
(Z0 (z, z ))(n+1)

Z0(n) (z, z )
Z (z, z ) = B(Z0 (z, z ))(n) ,
B =A1
we find the covariance

(Z0 (z, z ))(n+1
(n)
Z0 (z, z ),
SZ0 (z, z ) = C(z, z ) + C (z, z )
Z0(n) (z, z )

(Z0 (z, z ))(n+1)
(n)
(n)
(n)
Z0 (z, z ),
SZ0 (z, z ) = C(z, z ) + C (z, z )
(n)
Z0 (z, z )
under the ghost
(3.34)
(3.35)
(3.36)
(3.37)
496

(Z0 (z, z ))(n+1)
.
K(n) (z, z ) = C(z, z ) + C (n) (z, z )
(n)
Z0 (z, z )
(3.38)
3.2. Advantages and drawbacks of the different approaches

In this, we will point out the worths and the faults of choosing the primary fields as
explicit functions of the coordinates or not.
First of all the coordinate description honors its scoreboard with the explicit residual
S(2) (z, z )). The price to pay for it is to
reparametrization symmetry (z, z ) (Z (2)(z, z ), Z
lose in the way the T (z, z ) fields as a projective connection. These geometrical objects are
welcome in conformal covariant theories, when the holomorphic derivative operators are
to be defined in a covariant way.
It is a well-known problem for a right definition of the diffeomorphism anomaly [18].
Its solution is quite involved and its origin is not a clearly identified.
We want to show that the anomaly of our algebra is, in the explicit coordinate approach
( = 0) not naively well defined, so that complicated surgery methods have to be applied,
while if T (z, z ) can be considered as a projective connection ( 6= 0), the anomaly becomes
well defined under change of charts.
We define the anomaly in terms of GelfandFuchs cocycles as in reference [34].
The holomorphic cocycles (\,n) (z, z ) of FaddeevPopov charge equal to n is defined
as:
S(\,n) (z, z ) = 0.
(3.39)
So if we decompose:
cn (z, z ),
(z, z ) +
(\,n) (z, z ) = C(z, z )(n1)
z
0
(3.40)
the Eq. (3.39) implies:

cn (z, z ) = 0,
z )(n1)
(z, z ) C(z, z )(n1)
(z, z )
(z,
z
z
0
cn (z, z ) + X (z, z )(n1) (z, z ) = 0.
(z, z )
z
0
(3.41)
The FeiginFuchs cocycle is the FaddeevPopov charge three well defined element of the
cohomology space.
Somewhat elaborated calculations will give:

30 (z, z ) = SC(z, z ) 2 C(z, z ) + C(z, z )C(z, z ) SC(z, z ) C(z, z )
2
W(z, z )C (2) (z, z )C (2)(z, z ) 2 C (2) (z, z )
3

1
+ 2 C (2) (z, z ) 3 C (2) (z, z ) C (2) (z, z ) 4 C (2) (z, z )
3
+ 2 2 T (z, z )C (2) (z, z )C (2)(z, z ) 10T (z, z )C (2)(z, z ) 2 C (2) (z, z )

(3.42)
2T (z, z )C (2) (z, z ) 2 C (2) (z, z ) C(z, z ).
497
Going at = 0 and substituting the explicit expression of T (z, z ) and W(z, z ) as in

Eqs. (3.33), (3.33) we recover the explicit coordinate description, so Eq. (3.42) can be
written in term of the K(2) (z, z ) as:
30 (z, z ) = K(2) (z, z )K(2) (z, z ) 2 K(2) (z, z ).
(3.43)
The anomaly can be calculated along the lines found in Refs. [33,34]:
12 (z, z ) dz d z =
3 (z, z ) dz d z
c(z, z ) c(z,
z ) 0
= (z, z ) 2 C(z, z ) C(z, z ) 2 (z, z )

1
+ 3 (2) (z, z ) 3 C (2) (z, z ) 2 C (2)(z, z ) 3 (2) (z, z )
3

(2) (z, z ) 4 C (2)(z, z ) + C (2)(z, z ) 4 (2) (z, z )

+ 2 2 T (z, z ) (2)(z, z )C (2)(z, z ) C (2) (2)(z, z )
10T (z, z ) (2)(z, z ) 2 C (2)(z, z ) C (2)(z, z ) 2 (2)(z, z )

2T (z, z ) (2) (z, z ) 2 C (2) (z, z ) C (2) (z, z ) 2 (2)(z, z ) ,

(3.44)
where the (extended) Beltrami multipliers are [6,7,24,32]:

(z, z ) =
C(z, z )
,
c(z,
z )
(2) (z, z ) =
C (2)(z, z )
.
c(z,
z )
(3.45)
for = 0 we get the usual anomaly

(Z0 (z, z ))2 (2)
(z,
z
)
,
12 (z, z ) dz d z = K(2) (z, z ) 3 (z, z ) +
(2)
Z0 (z, z )
since (z, z ) +
(Z0 (z,z))2 (2)

(z, z )
(2)
Z0 (z,z)
(3.46)
is the Beltrami multiplier of Z0(2)(z, z ) with respect to
the (z, z ) background.

This object is not well defined under holomorphic change of charts [18,36] and an
intricate game, necessary to introduce a projective connection (useful for geometrical
purposes but inessential for the dynamical ones) is needed.
On the other hand for 6= 0, (so we can fix = 1 due to the rescaling property
Eq. (2.53)) a projective connection T (z, z ) is at our disposal. So, adding total derivatives
and cocycles to Eq. (3.44) (unessential from the cohomological but essential from the
geometrical point of view) we can rewrite:

1
2 (z, z ) dz d z = C(z, z )L3 zz (z, z ) zz (z, z )L3 C(z, z )

1 (2)
5 (2)
) (2)
)
C (z, z )L5 (2)
z (z, z
z (z, z )L C (z, z
3

8 C(z, z )(2) (z, z ) zz C (2) (z, z ) W(z, z )
) (2)
)C(z, z )
24W(z, z ) C(z, z )(2)
z (z, z
z (z, z
498
+ C (2) (z, z )zz (z, z ) zz (z, z )C (2)(z, z )

dz d z ,
(3.47)

L5 = 5 + 10T (z, z ) 3 + 15 T (z, z ) 2 + 9 T (z, z ) + 16T 2 (z, z )

+ 2 3 T (z, z ) + 8T (z, z ) T (z, z ) .
(3.48)
where the Bol derivatives are recalled to be:

L3 = 3 + 2T (z, z ) + T (z, z ) ,
We see that the anomaly is well defined under holomorphic change of charts, due to the
fact that T (z, z ) is a projective connection.
4. Conclusions
We have shown in this paper the geometrical origin and the strengths and weaknesses of
W3 and the other algebras related to the same construction.
The symplectic approach to this problem fits in a coordinate scenario the O.P.E method
from which, historically speaking, the W algebras were derived. There are more questions
that the reader can ask. What deeper insights will lock the symplectic geometry? Is there
a further relation between O.P.E. and the symmetry constraints? The answers could reveal
the connections between Physics and Geometry, but unfortunately they are not at our hand.
Acknowledgements
We would like to thank Alberto Blasi for several discussions and patience.
References
[1] C. Becchi, A. Rouet, R. Stora, in: E. Tirapegui (Ed.), Field theory, quantization and statistical
physics, 1981, pp. 332.
[2] H. Ooguri, K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, The induced action of W3 gravity,
Commun. Math. Phys. 145 (1992) 515539;
K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, Nucl. Phys. B 349 (1991) 791;
K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, Phys. Lett. B 243 (1991) 248;
K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, Commun. Math. Phys. 124 (1989) 87;
K. Schoutens, A. Sevrin, P. van Nieuwenhuizen, Int. J. Mod. Phys. A 6 (1991) 2981.
[3] C.M. Hull, Phys. Lett. B 240 (1990) 110;
C.M. Hull, Nucl. Phys. B 353 (1991) 707;
C.M. Hull, Phys. Lett. B 259 (1991) 621;
C.M. Hull, Nucl. Phys. B 367 (1991) 731;
C.M. Hull, Commun. Math. Phys. 156 (1993) 245;
C.M. Hull, Int. J. Mod. Phys. 8 (1993) 507.
499
[4] K. Shoutens, A Sevrin, P. van Nieuwenhuizen, Commun. Math. Phys. 124 (1989) 87;
E. Bergshoef, C.N. Pope, L.J. Romans, E. Sezcin, X. Shen, K.S. Stelle, Phys. Lett. B 243 (1990)
245;
E. Bergshoef, C.N. Pope, L.J. Romans, E. Sezcin, X. Shen, K.S. Stelle, Nucl. Phys. B 363
(1991) 163.
[5] A.T. Ceresole, M. Frau, J. McCarty, A. Lerda, Phys. Lett. B 256 (1991) 72.
[6] M. Carvalho, L.C. Queiroz Vilar, S. Sorella, Algebraic characterization of anomalies in chiral
W(3) gravity, Int. J. Mod. Phys. A 10 (1995) 38773900.
[7] D. Garajeu, S. Lazzarini, R. Grimm, W gauge structure and their anomalies: an algebraic
approach, J. Math. Phys. 36 (1995) 70437072.
[8] A. Abud, J.-P. Ader, L. Cappiello, Consistent anomalies of the induced W gravities, Phys. Lett.
B 396 (1996) 108116.
[9] J.-P. Ader, F. Biet, Y. Noirot, A geometrical approach to super W-induced gravities in two
dimensions, Nucl. Phys. B 466 (1996) 285314.
[10] T. Tjin, Finite and infinite W algebras, Doctoral Thesis, hep-th/9308146;
J. de Boer, F. Harmsze, T. Tjin, Nonlinear finite W symmetries and applications in elementary
systems, Phys. Rep. 272 (1996) 139214, hep-th/9503161.
[11] A.B. Zamolodchikov, Theor. Math. Phys. 65 (1985) 1205.
[12] V.A. Fateev, A.B. Zamolodchikov, Nucl. Phys. B 280 [FS18] (1987) 644.
[13] V. Drinfeld, V. Sokolov, J. Sov. Math. 30 (1984) 1975.
[14] I.M. Gelfand, L.A. Dickey, Funct. Anal. Appl. 10 (1976) 4;
I.M. Gelfand, L.A. Dickey, Funct. Anal. Appl. 11 (1976) 2.
[15] C. Itzykson, W-geometry, Cargese Lectures, in: O. Alvarez et al. (Eds.), Random Surfaces and
Quantum Gravity, Plenum Press, New York, 1991.
[16] M.A. Olshanetski, A.M. Perelomov, Classical integrable finite dimensional systems related to
Lie algebras, Phys. Rep. 71 (1981) 313.
[17] A. Cappelli, C.A. Trugenberger, G.R. Zemba, W1+ minimal models and the hierarchy in the
quantum Hall effect, Int. J. Mod. Phys. A 12 (1997) 11011111, hep-th/9610019;
F. Barbarin, E. Ragoucy, P. Sorba, Finite W algebras and intermediate statistics, Nucl. Phys.
B 442 (1995) 425443, hep-th/9410114.
[18] F. Gieres, Conformally covariant operators on Riemann surfaces (with applications to
conformal and integrable models), Int. J. Mod. Phys. A 8 (1993) 158.
[19] X. Shen, W infinity and string theory, Int. J. Mod. Phys. A 7 (1992) 69536993.
[20] P. Bouwknegt, K. Shoutens, W symmetry in conformal field theory, Phys. Rep. 223 (1993) 183.
[21] C.M. Hull, Geometry and W-gravity, Talk at Pathways to Fundamental Interactions, 16th John
Hopkins Workshop on Current Problems in Particle Theory, Goteborg 1992, hep-th/9301074;
C.M. Hull, Classical and quantum W -gravity, hep-th/9201057;
C.M. Hull, W -geometry, Commun. Math. Phys. 156 (1973) 245275, hep-th/9211113.
[22] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Infinite conformal symmetry in two
dimensional quantum field theory, Nucl. Phys. B 241 (1984) 333.
[23] A.A. Belavin, V.G. Knizhnik, Phys. Lett. B 168 (1986) 201;
A.A. Belavin, V.G. Knizhnik, Sov. Phys. JETP 64 (1986) 214.
[24] A. Bilal, V.V. Fock, I.I. Kogan, Nucl. Phys. B 359 (1991) 635;
Y. Matsuo, Commun. Math. Phys. 152 (1993) 317;
Y. Matsuo, Phys. Lett. B 274 (1992) 309;
G. Sotkov, M. Stanishkov, Nucl. Phys. B 356 (1991) 439;
S. Govindarajan, Higher dimensions uniformisation of w geometry, hep-th 9412078;
Y. Matsuo, Commun. Math. Phys. 152 (1993) 317;
Y. Matsuo, Phys. Lett. B 274 (1992) 309;
A. Gerasimov, A. Levin, A. Marshakov, Nucl. Phys. B 360 (1991) 537;
A. Das, W-J. Huang, S. Roy, Int. J. Mod. Phys. A 7 (1992) 3447;
500
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
J. de Boer, J. Goeree, Nucl. Phys. B 381 (1992) 329;

J. de Boer, J. Goeree, Phys. Lett. B 274 (1992) 289;
K. Yoshida, Int. J. Mod. Phys. A 7 (1992) 4353.
J. De Boer, J. Goeree, Covariant W-gravity and its moduli space from gauge theory, Nucl. Phys.
B 401 (1993) 369, hep-th-906098.
G. Sotkov, M. Stashnikov, C.J. Zhu, Nucl. Phys. B 356 (1991) 245.
G. Sotkov, M. Stashnikov, Nucl. Phys. B 356 (1991) 439.
J.L. Gervais, Y. Matsuo, Phys. Lett. B 274 (1992) 309.
E. Witten, Surprises with topological field theories, in: R. Arnowitt et al. (Eds.), String
90, Proceedings, Superstring Workshop, College Station, USA, March 1217, 1990, World
Scientific Publishing, Singapore, 1991.
G. Bandelloni, S. Lazzarini, w-algebras from symplectomorphisms, J. Math. Phys. 41 (2000)
22332250.
R. Zucchini, Light cone Wn geometry and its symmetries and projective field theory, Class.
Quant. Grav. 10 (1993) 253278.
G. Bandelloni S. Lazzarini, The role of complex structure in w symmetry, Nucl. Phys. B, in
press.
G. Bandelloni, Diffeomorphism cohomology in quantum field theory models, Phys. Rev. D 38
(1988) 1156.
G. Bandelloni, S. Lazzarini, Diffeomorphism cohomology in Beltrami parametrization, J. Math.
Phys. 34 (1993) 5413.
C. Becchi, A. Rouet, R. Stora, Ann. Phys. (N.Y.) 98 (1975) 287.
C.M. Becchi, Nucl. Phys. B 304 (1988) 513.
R. Grimm, private communication.

Discrete symmetries of functional determinants

D.V. Vassilevich a,1 , A. Zelnikov b,,2
a University of Leipzig, Institute for Theoretical Physics, Augustusplatz 10/11, 04109 Leipzig, Germany
b Theoretical Physics Institute, University of Alberta, Edmonton, Alberta, T6G 2J1, Canada
Abstract
We study discrete (duality) symmetries of functional determinants. An exact transformation of the
effective action under the inversion of background fields (x) 1 (x) is found. We show that in
many cases this inversion does not change functional determinants. Explicitly studied models include
a matrix theory in two dimensions, the dilaton Maxwell theory in four dimensions on manifolds
without a boundary, and a two-dimensional dilaton theory on manifolds with boundaries. Our results
provide an exact relation between strong and weak coupling regimes with possible applications to
string theory, black hole physics and dimensionally reduced models. 2001 Elsevier Science B.V.
All rights reserved.
PACS: 04.62.+v (Quantum field theory in curved spacetime); 11.15.Tk (Other nonperturbative techniques);
11.30.Ly (Other internal and higher symmetries)
Keywords: Duality; Index theorem; Effective action; Dilaton
1. Introduction
There is an enormous amount of literature devoted to calculations of the one-loop
effective action on various backgrounds. It is well known that the divergent part of the
effective action is local and can be calculated exactly, at least in principle. The finite part
of the effective action is typically non-local and usually cannot be obtained in a closed
form. The famous exceptions when the effective action is known exactly are the Polyakov
action [1] and the WZNW action [2,3]. There is also an example of a spinorial action [4]
for which the finite part of the effective action can be calculated exactly. Some other exact
results on fermion determinants can also be found in [5].
E-mail addresses: Dmitri.Vassilevich@itp.uni-leipzig.de (D.V. Vassilevich), zelnikov@phys.ualberta.ca

(A. Zelnikov).
1 On leave from V.A. Fock Department of Theoretical Physics, St. Petersburg University, 198904 St.
Petersburg, Russia.
2 On leave from P.N. Lebedev Physics Institute, Leninskii prospect 53, Moscow 117 924, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 0 - 7
502
D.V. Vassilevich, A. Zelnikov / Nuclear Physics B 594 (2001) 501517
In some cases one can obtain powerful relations between functional determinants and
spectral functions of relevant operators even though the functional determinants (effective
actions) themselves are not known exactly. These cases are governed by the Index Theorem
(see [6,7] for a detailed review). The idea behind the Index Theorem is quite simple. Let the
two operators 1 and 2 be represented in a factorized form: 1 = Q1 Q2 , 2 = Q2 Q1 .
In such a case, the eigenvalues of 1 and 2 coincide up to zero modes. Consequently,
differences of all spectral functions of 1 and 2 (e.g., the zeta function or the heat
kernel) come from the zero mode contributions and typically can be expressed in terms
of topological invariants.
In this paper we study relations between the functional determinants of

(1)
+ = 1 1 and = 1 1 ,
where (x) is a matrix valued function or a positive scalar field, = e . Below we
demonstrate that in certain cases log det + log det can be calculated exactly. In
most cases this difference is just zero. The two operators are related by the transformation
1 . Hence, the relation between the determinants that has been advertised above
is a kind of duality relation between effective actions in strong and weak coupling limits.
The key difference between the problem we study here and that of the standard Index
Theorem is the summation over the index in operators (1). Because of this summation
the eigenvalues of the operators + and generically are different. Nevertheless, we find
a relation between the functional determinants of these operators even though this property
cannot be extended to other spectral functions (such as higher heat kernel coefficients).
At first sight the operators (1) look a little bit artificial, but in fact they are quite general
and appear in a number of physical applications. Consider, e.g., a (spherical) reduction of a
field theory from higher dimensions (see [8] for a recent review). Let be a minimal scalar
field living on a D-dimensional manifold that has the structure of a semi-direct product:
(ds)2 = g (x) dx dx + e D2 (x)(d)2 .
4
(2)
Here g (x) is the two-dimensional metric and (d)2 is the line element on a (D 2)dimensional manifold. Both the dilaton (x) and the metric g (x) are assumed to depend
only on the first two coordinates x . Then the action for the scalar field reduces to the
2-dimensional action,
Z
Z
q
D
D i
k
D
(3)
d x g gik ( )( ) d 2 x g g e2 ( )( ).
To quantize the theory one should also fix an inner product of the quantum fields. The
reduced inner product may also depend on the dilaton field:
Z
0
(4)
h, i = d 2 x g e2() (x) 0 (x) .
p
If we start with the natural choice of the measure g D for the inner product in D
dimensions, then the preferable choice for the scalar function () in 2D is, of course,
() = . For simplicity we have omitted in Eqs. (3), (4) an unimportant constant factor of
503
the volume of the (D 2)-dimensional manifold. By performing the functional integration

over with the measure defined by (4) we obtain the effective action

1
(5)
W [ , ] = log det e e2 e ,
2
where the conformal gauge for the two-dimensional metric g = e2 has been used.
The indices in (5) are contracted with the Kronecker symbol. The dependence of the
effective action (5) on is governed by the scale anomaly which can be easily
calculated for arbitrary . The corresponding effective action can be obtained by
integration of the anomaly [9,10]. This means, in particular, that
W [ , ] = W [, ] + W scale ,
(6)
where W scale can be easily calculated by the methods of [9,1113]. An explicit expression
will be given below (see Eq. (70)). Thus, the problem of the calculation of the effective
action (5) is reduced to the evaluation of the determinant of + with = e .
One of the most interesting applications of the reduced theories is the evaluation of
the s-wave Hawking radiation [10] (see [8] for more references). Typically,
corresponds to the asymptotically flat region and + to the region near the black
hole singularity. Thus the transformation indeed maps the strong coupling region
to the weak coupling one. Of course, reduction and quantization do not commute. The
difference is the so called dimensional-reduction anomaly [1416] (see also [17]). The
reduction of other models (as, e.g., that of the Maxwell theory in D-dimensions to the
dilaton Maxwell theory in 4 dimensions) can be considered along the same lines.
In fact, the operator + with = e is a representation of the scalar Laplacian with a
potential V+ = ( 2 ) + ()2 . Indeed, we have traded one scalar function V for another
scalar function . In a recent paper [18] it has been noted that 3rd order terms in are
surprisingly absent in the expansion of the effective action in powers of . In the present
paper we explain this result and extend it to all odd orders of .
Another important model where the operators (1) appear is the bosonic string. Consider
a model with the action
Z
(7)
S = d 2 x Gab (x) a b .
The natural inner product for the field a reads:
Z
h1 , 2 i = d 2 x Gab 1a 2b .
(8)
Performing the path integral over with the measure defined by (8) we arrive at det +
with ( 2 )ab = Gab . Note that if one expands the ordinary bosonic string action in small
fluctuations of the string coordinates one more term proportional to the target space
Riemann tensor appears in (7). Therefore, (7) describes the one-loop partition function
for the bosonic string only in the case of a flat background metric Gab which is perhaps
not of particular interest. Our analysis is rather motivated by general ideas of string duality
(see [19,20] for introductory reviews). The target space duality transformation changes the
size of the target manifold and is used to relate strong and weak coupling limits (usually
504
with respect to small and large compactification radii). We go further by considering

arbitrary matrix transformations with arbitrary dependence on the world-sheet coordinates,
and still find a duality symmetry in the quantum determinants! Although we cannot give
immediately an exact relation to known formulations of the T -duality, we note that the
one-component case is just the dilaton-shift problem in the T -dual models [21]. Also, our
transformations interchange Dirichlet and Neumann boundary conditions (see Section 3),
just as it happens in the open string duality [2227].
Throughout this paper we use zeta function regularization [28,29]. In the next section
we demonstrate that functional determinants for the operators (1) coincide on a twodimensional manifold without boundary. Our method is similar to that used in [21] for
the one-component case. We also clarify some subtle mathematical points, such as nonstandard logarithmic terms in the heat kernel expansion and non-locality of some of
the heat kernel asymptotics. In Section 3 we extend our formalism to manifolds with
boundaries. For simplicity, the one-component case is considered. This time the difference
of the effective actions is non-zero. It is given by a surface integral plus a non-local
contribution of the zero modes. In Section 4 we go to four dimensions and demonstrate
that the effective action in the dilaton Maxwell theory is independent of the sign in front
of the background dilaton field. In that section we extensively use the technique of the
multidimensional Darboux transform [30].
2. Determinants of matrix operators in 2D

Consider the first order differential operators
A = + = 1 ,
B = A = = 1 ,
(9)
where is an hermitian matrix-valued field, = . is expressed in terms of :

= 1 ( ).
(10)
From the two first-order operators (9) we construct two Laplacians

+ = A B ,
= B A .
(11)
Note that + is mapped to under 1 .

Our aim is to compare functional determinants (one-loop effective actions) for the
operators + and . For a general elliptic operator the regularized effective action reads
W
reg
1
2s
= log det(D)reg =
2
2

dt t s1 Tr exp(tD) ,
(12)
where we have introduced the (zeta-function) regularization parameter s, which should

be set to zero after calculations, and a massive parameter needed to make the action
505
dimensionless. To proceed further we need a definition of the -function of the operator

D:

D (s) = Tr D s =
1
(s)

dt t s1 Tr exp(tD) .
(13)
Eqs. (12) and (13) are valid for general elliptic (pseudo) differential operators of positive
order. From now on we use the fact that D is an operator of Laplace type, i.e., D is an
elliptic second-order differential operator with a scalar leading symbol. For such operators
the function is regular at s = 0 (see, e.g., [6]) and, therefore, the regularized effective
action is
W reg =
2s
1
1
(s) D (s) = D (0) D (0)0 log() D (0) + O(s),
2
2s
2
(14)
where prime denotes differentiation with respect to s. We have used (s) = 1/s E +
and have absorbed the Euler constant E in a redefinition of . The pole part of (14) should
be canceled by a counter-term. The renormalized effective action reads
1
W ren = D (0)0 log() D (0).
2
(15)
The log() term describes the renormalization ambiguity. By means of the Mellin
transformation one can demonstrate that D (0) = a1 (1, D) where a1 is defined through
the heat kernel asymptotics
X
t n1 a1 (f, D),
K(f, t, D) = Tr f exp(tD)
=
(16)
n=0
as t +0 valid for arbitrary smooth matrix-valued function f . The coefficient a1 is locally

computable for operators of Laplace type. D (0)0 is non-local and in general no closed
expression for this quantity is available.
Any operator of Laplace type can be represented in the form

b
(17)
D = g + E
b For the
with a suitable auxiliary metric g,
covariant derivative = + and potential E.
operator + (11) we obtain

1
,
2
1
2 1

1
b
E = + + , .
2
4
2
g = ,
The coefficient a1 reads:

Z
p

1
b + 6E
b .
tr d 2 x g f R
a1 (f, D) =
24
b is the scalar curvature of g and tr denotes ordinary trace over all matrix indices.
R
(18)
(19)
506
Consider the variation of the zeta function + of the operator + :

()B s1
A ,
+ (s) = 2s Tr ()A B s1
+
+
(20)

1
(21)
= () 1 + 1 () .
2
In the first term in (20) the operators re-combine in a power of + , giving the generalized
-function:

(22)
= 2s + (, s).
2s Tr ()A B s1
+
The contribution of this term to the variation of the effective action (15) can therefore be
treated exactly. Indeed, since + (f, s) is regular at s = 0, we obtain

1
= + (, 0) = a1 (, + ).
(23)
s 2s Tr ()A B s1
+
s=0
2
This reflects a general property of functional determinants. If the variation of an elliptic
operator D has the form D = (1 )D + D(2 ) for some local operators 1,2 , then the
variation of the zeta function reads: D (s) = s Tr((1 + 2 )D s ). Consequently, the
variation of 0 can be expressed through the heat kernel D (0) = a1 ((1 + 2 ), D).
The second term in (20) requires more care because the operators under the trace cannot
be recombined in a power of + . Nevertheless, according to general method [31], it can
be represented in terms of of -function of a different operator F :
Tr

A
()B s1
+
1
=
(s + 1)
dt t s Tr ()B exp(t+ )A
1
=
(s + 1)
dt t s Tr ()F exp(tF )

= Tr ()F s .
(24)
Here F is a matrix operator having both spacetime vector indices and internal indices
of the matrix field (x):
F = B A .
(25)
In Eq. (24) the trace is taken over all matrix indices.

It is clear from the construction that the operator F should be understood as acting on the
space V L of B-longitudinal vector fields which can be represented as v = B , where
is a scalar function. Such fields satisfy the condition
B v = 0.
(26)
VL
can be constructed easily. Let

be a normalized
The spectrum of the operator F on
eigenfunction of the operator + with a non-zero eigenvalue . Then the functions v =
B (+ )1/2 are normalized eigenfunctions of the operator F . Accordingly, we write
the off-diagonal heat kernel
K(F, x, y; t) =
v (x)v (y)et
2 0
507
d x

1/2
1/2
d 2 y 0 Bx (+ )xx 0 By (+ )yy 0 K + , x 0 , y 0 ; t . (27)
For the traced heat kernel this relation simplifies to

Tr exp(tF ) = Tr exp(t+ ) .
(28)
The operator F contains a non-local projector on the space V L . Therefore, F is a pseudo

differential operator rather than a differential one. According to the general theory [6], the
generalized zeta function for such operator can have a simple pole at s = 0:
1 F
(f ) + 0F (f ) + O(s).
F (f, s) = pole
s
By substituting (29) into (20) one obtains
(29)
+ (0)0 = 2 + (, 0) + 20F ().

Next we relate the coefficient
obvious relation
(f, s) = Tr f F
F
0F (f )
1
=
(s)
(30)
to the heat kernel asymptotics. We invert the

dt t s1 Tr f exp(tF ) ,
(31)
to show that
1
K(f, t, F ) = Tr f exp(tF ) =
2i
ds (s) F (f, s)t s ,
(32)
where the integration contour encircles all poles of the integrand. The contribution of the
pole at s = 0 reads
F
(f ).
0F (f ) (E + log t)pole
(33)
The Euler constant E can be absorbed again in re-definition of the normalization scale
which is not shown explicitly in (31) and (32). Variation (30) is again defined by the t 0
term in the heat kernel asymptotics.
We see that the heat kernel K(f, t, F ) has a quite unusual log-term in the small t
asymptotics. Even though this term does not contribute to the variation of the effective
action, it is instructive to calculate its value. Translating equation (27) from kernels to the
operator notation we obtain:
h
i
K(f, t, F ) = Tr f B (+ )1/2 exp(t+ ) B (+ )1/2

= Tr A f B 1
+ exp(t+ )

(34)
= Tr f exp(t+ ) Tr f, B 1
+ exp(t+ ) ,
where
f, = f + [ , f ].
(35)
508
The first term on the last line in (34) obviously does not contain log t terms. The second
term can be represented as an integral over the proper time:

Tr f, B 1
+ exp(t+ ) =

ds Tr f, B exp(s+ ) .
(36)
To evaluate (36) we use the method of [32]. Consider the operator + = + f, B .
Then
1

(37)
Tr f, B exp(s+ ) = Tr exp s+ =0 .
s
The 1/s term in the small s asymptotics of (37) is immediately calculated with the help of
(19):
Z

1 1
d 2 x tr f, + .
(38)
s 8
This term generates the log t term in the small-t asymptotics of (34) and (36). We conclude
that
Z

1
F
(f ) =
(39)
d 2 x tr f, + .
pole
8
Note that this term is local. This supports the quite general observation that the leading
singularity (log t in the present case) is usually local, while the subleading terms (t 0 ) are
non-local.
In order to be able to apply standard expressions for the heat kernel coefficients it is
convenient to extend the operator F to the space V of all vector fields (which also carry
the internal indices). The space V splits into a direct sum
V = VL VT ,
(40)
where the space V T consists of the vector fields u = A . It is easy to check that the
R
decomposition (40) is orthogonal, i.e., d 2 x u v = 0 for v V L and u V T . It is clear
that F is zero on V T . Define the operator
e = 0 0 A0 B 0
F
(41)
that is zero on V L .
It can be demonstrated that the zero modes of the map v V L (resp. u
e have
V T ) coincide with the zero modes of + (resp. ) and that the operators F and F
L
T
no more zero modes on V and V respectively. Suppose that the underlying manifold
is R2 and (x) goes to a non-degenerate constant matrix as |x| . In such a case
0
(x) = (x)1 c (c is a constant vector in the internal space) of the
the zero modes
operators are non-normalizable and, therefore, can be neglected. Zero-modes of the
decomposition (40) (analogs of harmonic one-forms) are also absent.
e is an elliptic operator of Laplace type,
The operator F + F

e + F = + , .
(42)
F
The coefficient a1 for this operator can be found with the help of (19):
Z
2
1
e
d 2 x tr f + .
a1 (f, F + F ) =
8
509
(43)
e there is a representation similar to (27),

For the heat kernel of the operator F
Z
Z
1/2
e, x, y; t) =
d 2 x 0 d 2 y 0 0 Ax0 ( )xx 0
K(F
1/2
0 A0 ( )yy 0 K( , x 0 , y 0 ; t).
y
(44)
From equation (44) follows an important observation: the trace over vector indices
e, x, y; t) is obtained from K(F, x, y; t) if we exchange and 1 . By repeating
K(F
the calculations above for the operator we obtain
e
(0)0 = 2 (, 0) 20F ().

Now we subtract (45) from (30) to demonstrate that

e
+ (0)0 (0)0 = 2 + (, 0) 2 (, 0) + 2 F +F (, 0).
(45)
(46)
All zeta functions on the right-hand side of (46) correspond to elliptic operators. We can
replace them by the heat kernel coefficients:

e + F = 0. (47)
+ (0)0 (0)0 = 2 a1 (, + ) a1 (, ) + a1 , F
This equation demonstrates that the difference + (0)0 (0)0 does not depend on .
Since this difference is zero for = 1 = 1,
or
+ (0)0 = (0)0 ,
(48)

W ren [] = W ren 1 .
(49)
In a particular (one-component) case this relation has been obtained in [21] where the
method of [33] was used. The equation (48) can be also formulated as a Generalized Index
Theorem:
log det(A B ) = log det(B A ).
(50)
Normally, the Index Theorem does not contain summations under the determinant but
contains contributions of the zero modes which are absent in the topologically trivial
situation considered here.
There is an interesting special case for the one-component matrix = e where is
an external scalar field (dilaton). Let the dilaton depend on one coordinate only, say x 1 . In
this case
= 22 + H ,
H = Q Q ,
(51)
where Q = 1 + (1 ) is (up to a multiplier) the supercharge used in supersymmetric

quantum mechanics (for an introduction see [34]). Obviously, the operators Q are
intertwining the Hamiltonians H , leading to pairing of their non-zero eigenvalues. This
510
pairing is the key ingredient of the Witten index construction [35] which governs dynamical
supersymmetry breaking. Since in this particular case the operators H commute with 22 ,
the eigenvalues of coincide up to zero modes with those of H . Consequently, if we
neglect the zero mode contributions, as we have agreed to do in this chapter, all traced
spectral functions of + and should coincide, including (s) and K(1, t, ), for
arbitrary value of their arguments (s and t respectively). Here comes the main difference
from our result. If the dilaton depends on both coordinates, the operators and +
have, in general, different eigenvalues. This can be confirmed for instance by calculation
of the next heat kernel coefficient a2 (1, ) which differs for + and .
3. Manifolds with boundaries
Let us extend the results of the previous section to manifolds with boundaries. For
simplicity we consider here the one-component case only, = e . Accordingly,
+ = A B ,
A = D , = e D e ,
B = A = D + , = e D e .
(52)
Since we will use a curved coordinate system near the boundary we have restored upper
and lower indices and introduced the Riemannian covariant derivative D .
As we will see in the next section, the structure of the elliptic complex plays an important
role in our construction. Thus, we could use one of the admissible sets of boundary
conditions for that complex [6]. However, we prefer a more direct (though also more
lengthy) way to derive the boundary operators. Let the operator + act on the fields
satisfying Dirichlet boundary conditions
|M = 0.
(53)
Obviously, + |M = A B |M = 0. Therefore, the vector v = B (see (26))

should satisfy
A v |M = 0.
(54)
To proceed further we need some notations. Let N be the inward-pointing unit normal
at the boundary. Let be a coordinate on the boundary. We can extend this coordinate
system to a neighborhood of the boundary keeping orthogonality (N = 0). The extrinsic
curvature k is defined as k = N g = N . We can simplify the analysis by choosing
a coordinate system in which g = const along the boundary (this relation cannot be
extended inside the manifold).
In the coordinate frame defined above

(55)
A v |M = e (N k) e vN + g e v M .
A local extension of the condition (54) on the space of all vector fields V (40) can be
obtained by requiring that vN and v satisfy these boundary conditions independently:

v |M = 0.
(56)
(N k) e vN M = 0,
In particular, the vectors
u = g A V T
511
(57)
should satisfy (56). This leads to Neumann boundary conditions for the scalar field :
N e |M = 0.
(58)
In our procedure the field forms a functional space for the operator . Therefore,
the conditions (58) become the boundary conditions for . Since (58) depends on the
dilaton, the variation of det contains two contributions, one coming from variation of
the operator, the other from the variation of the functional space on which this operator acts.
At this time we do not know how to evaluate the second contribution. To avoid difficulties
with the -dependence of (58) we impose a Neumann boundary condition on the dilaton:
N |M = 0.
(59)
Due to (59) the conditions (53), (56) and (58) are equivalent to the following -independent
set:
|M = 0,
(N k)vN |M = 0,
v |M = 0,
N |M = 0,
(60)
where one can recognize the relative boundary conditions for the de Rham complex [6].
Let us discuss now the zero mode structure of the theory. To be more precise let us
restrict ourselves to compact manifolds M with the topology of a two-dimensional disk.
The only zero mode of + is eliminated by the Dirichlet boundary condition. There is
0
= e of . There are no zero modes in the vector sector. This last
a zero mode
statement can be derived from [6] or from the simpler analysis of [36].
The boundary conditions (60) are mixed. This means that one of the components of the
vector satisfies a Dirichlet and the other one a Neumann boundary condition. The heat
kernel expansion for this type of the boundary conditions has been studied in [3639]. Let
us write the boundary conditions for a multi-component field v in the form:

(61)
D v + (N + S)N v M = 0,
where D,N are two complementary projectors and S is some function on the boundary
which can be matrix-valued. Clearly, the mixed boundary conditions (61) contain Dirichlet
and Neumann conditions as particular cases. The boundary term
Z

1
tr dy h f (2k + 12SN ) + 3f,N (N D ) ,
(62)
a1 (f, D) =
24
should be added to the volume part (19). Here y is a coordinate on the boundary and h is
the determinant of the induced metric.
0
= e of the operator to the heat
Obviously, the contribution of the zero mode
kernel reads
K0 ( , x, y; t) = R
0 (x) 0 (y)
0 (z))2
d 2 z g (
(63)
512
There is no t-dependence on the right hand side of (63). Therefore we can easily derive the
zero mode contribution to a1 (f, ):
R 2
d x g f e2
(0)
(64)
a1 (f, ) = R 2 2 .
d z ge
By acting exactly as in the previous section we obtain

1
e
(0)
+ (0)0 (0)0 = a1+ () a1 () a1 (f, ) + a1F +F ()
2
R 2
Z
d x g ()e2
1
dy h k + R 2 2 .
=
2
d z ge
(65)
This equation can be integrated to give (with the help of (15))

Z

Z
1
1
ren
ren
2
2
d z ge
dy h k log
+ W [0] . (66)
W (+ ) W ( ) =
2
2
M
Equation (66) represents the main result of this section. Here W [0] describes the difference
between the two effective actions for = const. Since the boundary conditions for + and
are different there is no reason to believe that this difference is zero. For the standard
Euclidean disk W [0] has been calculated in [40,41]. The non-local term in (66) is typical
for the effective action on a compact manifold [42].
As a consistency check let us show that if = c = const, the right hand side of (66)
does not depend on c . The first term reads c where is the Euler characteristic. The c
dependent part of the second term is c . Since we have assumed that M has the topology
of the 2-disk ( = 1) the two contributions cancel against each other.
To make our discussion self-contained we conclude this section with a calculation of
the change in the effective action under a conformal transformation of the external metric.
Since any metric in two dimensions is conformally trivial this will provide an extension of
our previous results to arbitrary curved manifolds. We use the methods of Refs. [9,1113].
[]
Consider the operator + = e2 + , which is unitarily equivalent to the operator (5)

= + . It is clear that under variation of the effective action changes as
[]

[]

+ .
(67)
W ren + = a1 ,
The coefficient a1 on the right hand side of (67) is given by equations (19), (62) where all
quantities should be calculated with the effective metric g = e2 g :
p
b = g (, , , ),
g E
p
b = 2g 2 ,
g R
(68)
k h = h (k N ).
All indices in (68) are contracted with g . For Neumann boundary conditions we choose
S = 0. We obtain
ren
[]

+
1
=
24
Z
+
"Z
513
d 2 x g 2 2 + 6( 2 ) 6()2

[]

(0)
dy h 2 (k
N )
3N ( )
+ a1 ,
+ .
(69)
The sign in the surface integral corresponds to Dirichlet () or Neumann (+) boundary
conditions. The last term in (69) describes the contribution of the zero modes. It is present
for Neumann boundary conditions only and is given by (64) with and g
e2 g. From this equation we easily obtain

[]

W ren + W ren (+ )
(Z
)
Z

1
2
2
2
2
d x g ( )
+ 6 () + dy h 2 k 3N
=
24
M
M
R
2
2(
)
d xe
1
R
.
(70)
+ (1 1) log M
2
2
4
Md xe
Here again the upper sign corresponds to Dirichlet and the lower to Neumann boundary
conditions. The action W scale in (6) is given by the right hand side of the same equation
(70) with = + . Now it is clear which modifications should be made in (66) for
arbitrary curved 2D manifolds (for a non-zero conformal factor ):
Z
Z
p
p

1
[]
[]
b + 1
d 2 x g R
dy h k
W ren + W ren =
4
2
M
M
Z

p
1
d 2 z g e2 + W [0] .
(71)
log
2
It is interesting to note that in the volume and surface integrals the constant mode of the
dilaton again couples to the Euler characteristic.
4. Dilaton Maxwell theory in 4D
Consider the dilaton Maxwell system in flat four-dimensional Euclidean space,
Z
1
(72)
d 4 x e2 F F ,
S=
4
where F is an abelian field strength, F = a a . Such an interaction appears in
extended supergravities [43], string theory, and after a reduction from higher dimensions.
Mechanisms for the generation of the action (72) and its classical properties were
considered, e.g., in [4449] (see also [5052] for some quantum calculations).
If the action (72) is obtained by a reduction from higher dimensions, the inner products
on the spaces of vector and scalar (ghost) fields should also undergo the reduction and
should thus contain the dilaton field
Z
ha, a 0 i = d 4 x e2 a a 0 ,
514
h, 0 i =
d 4 x e2 0 .
(73)
It is convenient to choose the gauge condition as

( , )e a = 0.
In this gauge the action for a and the FaddeevPopov ghosts is
Z

1
d 4 x a e ( , )( + , ) + 2, e a
S=
2

e ( , )( + , )e .
(74)
(75)
Performing the functional integration over a and with the measure (73) we arrive at the
effective action
det(+ + 2, )
1
,
(76)
W = log
2
det(+ )
where + = ( , )( + , ) and the vector determinant in (76) is restricted to the
fields e a satisfying the gauge condition (74).
The aim of this section is to study transformations of the effective action (76) under
change of the sign in front of the dilaton field. We adapt to this problem the technique of
the multidimensional Darboux transform [30].
Let us introduce the multi-index Kronecker symbol,
n
11...
...n =
1
... ... 1 ...n n+1 ...4 ,
(4 n)! 1 n n+1 4
(77)
and the supercharges

...
(n)
(n)+
B = Q[][] .
Q[][] = 11 ...n1
n
(78)
The operator Q(n) maps (n 1)-forms to n-forms and Q(n)+ maps n forms to (n 1)forms. With the help of these operators we define the matrix Hamiltonians
(1)
H| = B A ,
e(1) = 1 Q(2)+ Q(2) ,

H
|
2 |1 2 1 2 |
1
1 (3)+
e(2)
Q
= Q(2)
Q(2)+ ,
H
Q(3)
,
H(2)
1 2 |1 2 =
1 2 |1 2
2 1 2 | |1 2
12 1 2 |1 2 3 1 2 3 |1 2
1 (3)
(3)
(3)+
(79)
H1 2 3 |1 2 3 = Q1 2 3 |1 2 Q1 2 |1 2 3 .
12
e of Section 2. The operator H
e(3) is constructed in a
These operators replace F and F
e(n) act on the space Vn of the n-forms. We observe that
similar way. H (n) and H
Q(n) Q(n1) = Q(n1)+ Q(n)+ = 0.
(80)
Thus Q and Q+ have the properties of the d and operators of an elliptic complex.
There is an orthogonal decomposition Vn = Vn+ Vn such that Vn+ = Q(n1)+ Vn+1 and
Vn = Q(n) Vn1 . The spaces Vn+ (resp. Vn ) are spanned by the zero modes of H (n) (resp.
e(n) ). If const as |x| there are no more zero modes. In the construction of the
H
515
e(n) are restricted to Vn

spectral zeta functions below we always assume that H (n) and H
+
and Vn respectively.
The operators
e(n)
H(n) = H (n) + H
(81)
are elliptic on the space of n-forms.

The effective action (76) can be represented as
W=
e(1)
det H
1
log
.
2
det +
(82)
By acting as in Section 2, we write

1 + 0
(1)
(0) = 0+ () + 0H ().
2
The second term on the right hand side of (83) can be expressed as
(83)
e(1)
0H () = 0H () 0H ().
(1)
(1)
(84)
For the un-smeared -functions we have

e(1)
H (s) = H (s) H (s) = H (s) + (s).

(1)
(1)
(1)
(85)
The functions on the right hand side of (85) correspond to elliptic operators of the
e(1)
Laplace type. They are regular at s = 0. Therefore, H (0)0 is well defined, and
1 e(1)
(2)
e(1)
0H () = H (0)0 + 0H ().
2
Repeating these steps once again we obtain
(86)
e(2)
0H () = 0H () 0H (),
1 e(2)
(3)
e(2)
0H () = H (0)0 + 0H ().
2
(2)
(2)
(87)
It is convenient to rewrite H (3) in the dual representation:

1
(3)
= 1 2 3 1 2 3 H(3)
= B A + B A .
H|
1 2 3 |1 2 3
6
e(3) and H(3) in the dual representation:
We also introduce H
e(3) = A B ,
H
|
(88)
(3)
H| = 2, .
(89)
Since in the zeta functions in (87) all vector indices are contracted we can evaluate them in
the dual representation as well:
e(3)
0H () = 0H () 0H ().
(3)
(3)
H (3) ,
H(3)
(90)
e(3)
H
and
can be obtained from
Next we note that the operators
e(1) by changing the sign in front of the dilaton field . Therefore,
H
1
e(3)
0H () = (0)0 + 0 ().
2
Combining (83)(87), (90) and (91) we obtain
H (1) ,
H(1)
and
(91)
516

1
e(1)
e(2)
+ (0)0 + H (0)0 H (0)0 + (0)0
2
(1)
(2)
(3)
= 0+ () 0H + 0H 0H + 0 ().
(92)
The second line of (92) contains elliptic operators only. For them 0H () = a2 (, H ).
Consequently the second line of (92) vanishes by the GaussBonnet theorem [6]. One can
also check this by direct calculations. Therefore,
W ren [] = W ren [].
(93)
5. Conclusions
In this paper we have found several remarkable duality relations between functional
determinants. We have shown how the effective action of quantum fields interacting with
the (matrix-valued) scalar background field transforms under inversion of the background
field. For flat manifolds without boundaries the effective action is proved to be invariant
under this inversion. This property has been explicitly demonstrated for two-dimensional
models and for the dilaton Maxwell theory in four dimensions. The presence of boundaries
leads to additional terms that can also be calculated exactly (see Eq. (66)). We have studied
in detail the 2D matrix theory, the 4D dilaton Maxwell system, and the 2D dilaton theory
on a manifold with boundary. We have considered the simplest topologies of the field
configurations only. Less trivial topologies can be included in our approach by taking into
account the zero-mode contributions. The application of our results and methods include
duality properties of strings, quantum black holes, and quantum systems obtained after
reduction from higher dimensions, to mention a few. We are sure that this list can be
considerably extended.
Acknowledgements
We would like to thank M. Bordag, S. Solodukhin, and A. Wipf for very useful
discussions of related questions. Research of one of the authors (DV) has been supported by
the Deutsche Forschungsgemeinschaft, grant Bo 1112/11-1, the Alexander von Humboldt
foundation and the FWF (Fonds zur Frderung der wissenschaftlichen Forschung) project
P-12.815-TPH. AZ is grateful to DAAD (Deutscher Akademischer Austauschdienst) for
its financial support and Institute of Theoretical Physics at Friedrich Schiller University
(Jena) for hospitality. AZ is also grateful to the Killam trust for its support.
References
[1]
[2]
[3]
[4]
A.M. Polyakov, Phys. Lett. B 103 (1981) 207.

E. Witten, Commun. Math. Phys. 92 (1984) 455.
S.P. Novikov, Sov. Math. Dokl. 37 (1982) 3.
W. Kummer, D.V. Vassilevich, Phys. Rev. D 60 (1999) 084021, hep-th/9811092.
517
[5] O. Alvarez, Nucl. Phys. B 238 (1984) 61.

[6] P.B. Gilkey, Invariance Theory, the Heat Equation, and the AtiyahSinger Index Theorem, CRC
Press, Boca Raton, 1995.
[7] T. Eguchi, P.B. Gilkey, A.J. Hanson, Phys. Rep. 66 (1980) 213.
[8] W. Kummer, D.V. Vassilevich, Ann. Phys. 8 (1999) 801, gr-qc/9907041.
[9] W. Kummer, H. Liebl, D.V. Vassilevich, Mod. Phys. Lett. A 12 (1997) 2683, hep-th/9707041.
[10] V. Mukhanov, A. Wipf, A. Zelnikov, Phys. Lett. B 332 (1994) 283, hep-th/9403018.
[11] O. Alvarez, Nucl. Phys. B 216 (1983) 125.
[12] J.S. Dowker, J.P. Schofield, J. Math. Phys. 31 (1990) 808.
[13] N. Drukker, D.J. Gross, A.A. Tseytlin, JHEP 04 (2000) 021.
[14] V. Frolov, P. Sutton, A. Zelnikov, Phys. Rev. D 61 (2000) 024021, hep-th/9909086.
[15] P. Sutton, Phys. Rev. D 62 (2000) 044033, hep-th/0003290.
[16] G. Cognola, S. Zerbini, On the dimensional reduction procedure, hep-th/0008061.
[17] S. Nojiri, S.D. Odintsov, Phys. Lett. B 463 (1999) 57, hep-th/9904146.
[18] Yu.V. Gusev, A.I. Zelnikov, Phys. Rev. D 61 (2000) 084010, hep-th/9910198.
[19] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77, hep-th/9401139.
[20] E. Alvarez, L. Alvarez-Gaume, Y. Lozano, Nucl. Phys. Proc. Suppl. 41 (1995) 1, hepth/9410237.
[21] A.S. Schwarz, A.A. Tseytlin, Nucl. Phys. B 399 (1993) 691, hep-th/9210015.
[22] P. Horava, Phys. Lett. B 231 (1989) 251.
[23] J. Dai, R.G. Leigh, J. Polchinski, Mod. Phys. Lett. A 4 (1989) 2073.
[24] M.B. Green, Phys. Lett. B 266 (1991) 325.
[25] M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365.
[26] H. Dorn, H.J. Otto, Phys. Lett. B 381 (1996) 81, hep-th/9603186.
[27] S. Forste, A.A. Kehagias, S. Schwager, Nucl. Phys. B 478 (1996) 141, hep-th/9604013.
[28] J.S. Dowker, R. Critchley, Phys. Rev. D 13 (1976) 3224.
[29] S.W. Hawking, Commun. Math. Phys. 55 (1977) 133.
[30] A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Teor. Mat. Fiz. 61 (1984) 183.
[31] V.N. Romanov, A.S. Schwarz, Teor. Mat. Fiz. 41 (1979) 190.
[32] T.P. Branson, P.B. Gilkey, D.V. Vassilevich, J. Math. Phys. 39 (1998) 1040, hep-th/9702178.
[33] A.S. Schwarz, Commun. Math. Phys. 67 (1979) 1.
[34] F. Schwabl, Quantenmechanik, Springer, Berlin, 1988.
[35] E. Witten, Nucl. Phys. B 202 (1982) 253.
[36] D.V. Vassilevich, J. Math. Phys. 36 (1995) 3174, gr-qc/9404052.
[37] I.G. Moss, Class. Quantum Grav. 6 (1989) 759.
[38] T.P. Branson, P.B. Gilkey, Commun. PDE 15 (1990) 245.
[39] T.P. Branson, P.B. Gilkey, K. Kirsten, D.V. Vassilevich, Nucl. Phys. B 563 (1999) 603, hepth/9906144.
[40] M. Bordag, B. Geyer, K. Kirsten, E. Elizalde, Commun. Math. Phys. 179 (1996) 215, hepth/9505157.
[41] J.S. Dowker, Class. Quantum Grav. 13 (1996) 585, hep-th/9506042.
[42] J.S. Dowker, Class. Quantum Grav. 11 (1994) L7, hep-th/9309127.
[43] E. Cremmer, J. Scherk, S. Ferrara, Phys. Lett. B 74 (1978) 61.
[44] G.W. Gibbons, Nucl. Phys. B 207 (1982) 337.
[45] K. Maeda, Class. Quantum Grav. 3 (1986) 651.
[46] V. Frolov, A. Zelnikov, U. Bleyer, Ann. Phys. 44 (1987) 371.
[47] G.W. Gibbons, K. Maeda, Nucl. Phys. B 298 (1988) 741.
[48] D. Garfinkle, G.T. Horowitz, A. Strominger, Phys. Rev. D 43 (1991) 3140.
[49] G.T. Horowitz, A. Strominger, Nucl. Phys. B 360 (1991) 197.
[50] S. Nojiri, S.D. Odintsov, Phys. Lett. B 426 (1998) 29, hep-th/9801052.
[51] A.A. Bytsenko, S. Nojiri, S.D. Odintsov, Phys. Lett. B 443 (1998) 121, hep-th/9808109.
[52] S. Ichinose, S.D. Odintsov, Nucl. Phys. B 539 (1999) 643, hep-th/9802043.

The renormalized thermal mass and critical

temperature of a complex scalar field theory
with non-zero charge density
Hugh F. Jones , Philip Parkin
Physics Department, Imperial College, London, SW7 2BZ, UK
Abstract
The linear expansion is used to obtain the self-energy up to O( 2 ) for a complex scalar
field theory with a ( ? )2 interaction at high temperature and non-zero chemical potential .
The calculation is performed in the imaginary-time formalism via the Hamiltonian form of the
path integral and non-perturbative results are generated by a systematic order by order variational
procedure. The behaviour of the renormalized mass as a function of T and is found and the critical
temperature extracted. This is plotted as a function of and compared with the result of resummed
perturbation theory. A detailed calculation of the high-temperature limit of the two-loop settingsun diagram is given, which can easily be adapted to standard perturbation theory. 2001 Elsevier
PACS: 11.10.Wx; 11.30.Hv; 11.30.Qc
1. Introduction
It has been known for some time that conventional perturbation theory is inadequate
for describing high-temperature field theory [1], with the perturbation expansion breaking
down at some order in the coupling constant in the parameter regime where the temperature
is very much greater than the bare mass. There have been several attempts to circumvent
this problem. For example Refs. [2,3] use resummed perturbative expansions while a
temperature-dependent renormalization group approach has been used in Ref. [4]. In
this paper we tackle the problem using the linear expansion (see, for example, [5]),
which contains elements of both of the above methods. It involves a mass shift which is
determined order-by-order in the expansion by a non-perturbative criterion, but prior to the
E-mail addresses: h.f.jones@ic.ac.uk (H.F. Jones), p.parkin@ic.ac.uk (P. Parkin).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 6 2 - 3
H.F. Jones, P. Parkin / Nuclear Physics B 594 (2001) 518532
519
(crucial) optimization stage it corresponds to low-order perturbation theory with modified

propagators and vertices.
Pinto and Ramos [6] have successfully applied the method to finite-temperature
symmetry breaking in a scalar field theory with a 4 interaction up to second order in ,
which in particular involves evaluating the non-trivial setting-sun diagram. In the present
paper we extend this approach to the case of non-zero chemical potential, which necessarily
involves considering a complex, rather than a real, scalar field. The chemical potential
induces a shift in the energy component of the Matsubara propagator, which presents no
problem of principle, but in practice makes the diagrams considerably more difficult to
evaluate.
To our knowledge there have been only a few attempts at this problem. Benson et
al. [7] performed a one-loop calculation, while Funakubo and Sakamoto [8] used a
renormalization group approach in the large-N limit of an O(N) theory (in which the
setting-sun diagram does not appear). Except for a few special cases the standard lattice
Monte-Carlo approach is unable to cope with a non-zero chemical potential, because the
Euclidean action then becomes complex and cannot be used as a real, positive statistical
weight.
We outline the linear expansion technique below and explain how it can generate nonperturbative results. In Section 2 we show how to formulate the theory in the imaginary
time formalism and include the chemical potential associated with a conserved charge. We
then go on to calculate all diagrams contributing to the self-energy, up to second order
in . Section 3 includes results for the renormalized thermal mass as a function of T and
and for the -dependence of the critical temperature. Appendix A contains a detailed
calculation of the two-loop setting-sun diagram in the high temperature expansion.
1.1. The linear delta expansion
The linear expansion (LDE) is a technique that allows the use of an analytic approach
to probe the non-perturbative sector of the field theory to which it is applied. It has
been employed with success in a wide variety of areas [5], with convergence of the
expansion rigorously demonstrated in some simple zero- and one-dimensional models
(see, for example, [9]). The method involves the introduction of an artificial expansion
parameter, , and the modification of the action of the theory under consideration via

(1)
S S = (1 )S0 {i } + S,
where {i } is some set of variational parameters and S0 represents the action of some
soluble theory. This trial action is not determined by the method; however, an appropriate
choice is usually suggested by the form of the theory under consideration. The -modified
action can then be used to evaluate some desired physical quantity as a power series in
(truncated at some finite order, N ), with then set equal to 1 at the end of the calculation.
We label this quantity PN , noting that it will, in general, have a residual dependence on
the {i }. The variational element is introduced by fixing these parameters according to
some specified criterion. We shall use the principle of minimal sensitivity (PMS) [10]
whereby the {i } are chosen at a stationary point of the quantity PN , namely:
520

PN
= 0.
i i = i
(2)
It is this order by order fixing that allows for non-perturbative behaviour to emerge (the
optimized variational parameters becoming functions of the order of truncation, and so
being more correctly labelled by {i (N)}), and can also provide for convergence of the
expansion, two particularly desirable features.
2. The charged scalar field at finite temperature

In Minkowski space, the Lagrangian density of a massive, complex scalar field (x)
with a V( ? ) interaction term is given by
L = ( ? )( ) m2 ? V( ? ).
(3)
This Lagrangian density has a well-known global U (1) gauge symmetry which leads to
a conserved charge Q and an associated chemical potential , so that the grand partition
function is
Z(, ) = Tr e(H Q) .
(4)
Proceeding through the Hamiltonian form of the path integral for Z, in the imaginary-time
formalism of finite-temperature field theory [12], we arrive at
Z
(5)
Z(, ) = D( ? , )eSE (,) ,
where the Euclideanized action is (dropping the suffix for future notational convenience)
S(, ) = SF (, ) + SI (, ) + SC (, )
with
Z
SF (, ) =
(6)

2
2
2
2
+ m ,
d x 2 + 2
(7)
d 4 x V( ? ),
(8)
SI (, ) =
T

2

2
2
2
+
d 4 x ? A
2
+
+
Bm
2
T
Z
+ d 4 x CV( ? ),
Z
SC (, ) =
(9)
R
R
R
where = it, x = (, x) and T d 4 x = 0 d d 3 x. Here SC (, ) represents those
counterterms required to render the model finite.
A sensible choice for our delta-modified action, in the presence of a ( ? )2 selfinteraction, V( ? ) = (/4)( ? )2 , would then be
S (, ) = SF (, ) + SI (, ) + SC (, )
(10)
with

2
2 + 2 2 ,
d 4 x ? 2 + 2
T

Z
? 2
4
2 ?
( ) ,
d x
SI (, ) =
4
521
SF (, ) =
(11)
(12)

2

2
2
2
+
d 4 x ? A
2
+
+
B
T

Z
4
? 2
2 ?
( ) B ,
+ d x C
4
Z
SC (, ) =
(13)
introducing the variational parameter , or equivalently , given by 2 = 2 m2 . In

subsequent sections we will evaluate the thermal mass up to O( 2 ), using dimensional
regularization to renormalize the theory. It is an important point, raised in [6], that the
arbitrary variational parameter becomes a function of the bare parameters (which is
how non-perturbative behaviour arises in the LDE), so we must eliminate any divergences
before we apply the PMS optimization procedure. The renormalization procedure we use
is identical to that of [6] and is based on [11]; we forgo any detailed discussion of the
cancelling of temperature-dependent divergences and any systematic calculation of the
coefficients of the counterterms A , B and C , the details being essentially identical to
those discussed in [6].
2.1. Calculating the thermal mass
We recall that the Feynman rules in frequency space in the presence of an overall charge
are as follows:
1. To every line of the diagram assign
a factor F (in , k) [(in + )2 + k2 ]1 ,
2
where n = 2n/ and k = k + 2 , and an arrow in the direction of momentum
flow
2. Assign a factor to each vertex
3. Assign a factor 2 to each insertion
P R
4. Integrate over every internal line with the measure T n d 3 k/(2)3
5. There must be conservation of charge at each vertex, i.e., the number of arrows
entering a vertex must be the same as the number leaving
We are now in a position to estimate corrections to the thermal mass (which we do up to
O( 2 )), defined by
m2T , = 2 + (in , p),
(14)
evaluated on-shell, i.e., in = , p = 0, where is the thermal self-energy. This

choice of four-momentum is discussed in Appendix A, where the setting-sun diagram is
calculated, the only momentum-dependent contribution to up to O( 2 ).
522
Fig. 1. Diagrams contributing at O().
2.2. The thermal mass at first order

To lowest order, the relevant contributions are
T , = 1 + 2
= 2 + T
XZ
n
d 3k
F (in , k).
(2)3
(15)
The frequency sum in (15) can be treated in a particularly concise and efficient manner
through the mixed representation [13] F (, k) of F :
X
ein F (in , k)
F (, k) = T
n

1
(k )
(k +)
+ n+
(1 + n
,
=
k )e
ke
2k
(16)
where
n
k =
1
e(k )
(17)
is the BoseEinstein distribution function in the presence of a non-zero chemical potential.

The Matsubara propagator is recovered by Fourier transforming the mixed propagator with
respect to the variable:
Z
F (in , k) =
d ein F (, k).
(18)
The contribution of 2 in (15) can then be calculated trivially to give

2
= T
XZ
n
Z
= T
d 3k
(2)3
Z
d ein F (, k)
0
d 3k 1
+
1 + n
k + nk .
3
(2) 2k
(19)
Using dimensional regularization [14] one finds that the O() contribution to the thermal
mass is

2
1
2
+
ln
1
m2T , = 2 2 2 +
16 2

4M 2
+
T 2 e
h (y, r),
2 3
(20)
523
Fig. 2. Diagrams contributing at O( 2 ).
where M is a mass scale introduced by dimensional regularization and, in the notation

of [15],

1
h3 (y, r) + h3 (y, r) ,
2

Z
1
x2
1
p
,
dx p
h3 (y, r) =

0(3)
x 2 + y 2 exp x 2 + y 2 ry 1
he3 (y, r) =
(21)
(22)
where x = k, y = and r = /. Using the minimal subtraction scheme (MS) we

can eliminate the divergent term arising from the temperature-independent part of the self
energy using the O() mass counterterm

2.
(23)
ct =
16 2
We now analyze (21) in the limit y 0 (the high temperature expansion) keeping r
fixed [15]:

y
1
2 y p
y2
e
2
2
ln
+ + r + ,
1r
(24)
h3 (y, r) =
12
4
8
4
2
leading to the following expression for the renormalized thermal mass at first order:
T p
T 2
1 r2
m2T , = 2 2 +
12
4

2
4T 2
2
+
ln
2r .
16 2
M2
(25)
In this and all subsequent calculations we neglect terms of O(1/T ).

2.3. The thermal mass at second order
At O( 2 ) there are five diagrams (shown in Fig. 2) which provide corrections to the
self-energy. We begin by evaluating the first diagram, which in the high temperature limit
is

2
2 1
1
4T 2
2
2 T
2
+ ln
.
(26)
1 =
8 1 r 2

16 2
M2
The next two diagrams are constructed using mass and vertex counterterms to give
2
O( 2 ) diagrams that eliminate the temperature-dependent divergences arising from 4 .
Explicitly, we obtain in the high temperature limit
524

2

2 T
1
2
2 2
4T 2
2
+
ln
8
(16 2)2 2
16 2
M2
1 r 2 16 2

2
2
2 2
2
,
(27)
+
ln
+
2
6
2(16 2 )2
4M 2
2 = 2
52 2
2(16 2)2 2

2
52 T 2 T p
4T 2
2
2+
+ 2
1
r
ln
2r
4
32 2 12
16 2
M2

2
2 2
2
2
2 5
.
(28)
ln
+ 1 +1+
6
(32 2 )2
4M 2
3 = 2
The momentum-independent two-loop diagram is given by

2
2 2
(16 2 )2
2
2
16 2
4 = 2
1
2

1
T
1 T2 T p
1 r2
12
4
8
1 r2

2
4T 2
+
ln
r2
2
8
M2
1
2 T 3
2 T 2
+ 2
96 1 r 2
32 2
2
2
p
T
1
T
+ 2
1 r2
2
(8)
3
2
1 r2

2
4T 2
4T 2
2
+
ln
2r
ln
4 2
M2
M2
2 T
r2
64 3
1 r2

2
2
2
2
2
2
+
ln
+ (2 1) ln
+ 2.4 .
(16 2 )2
4M 2
4M 2
+ 2
(29)
Finally, we have the momentum-dependent setting sun diagram, which we evaluate

with the Euclidean self-energy on shell (the details of the calculation are provided in
Appendix A):
2
2 2
2 2
32 2 1
1
2 3 1
2 p
+
2
2
2
2
2
2
2
(32 )
(32 )
(32 ) 2
2

2
p
T
T
2
3
4T 2
2
2+
r
ln
2r
2
8
16 2 24
32 2
M2

2 2
2
2
17
2 3
2
ln
ln
+ 2
+ 1.9785
+
2(16 2)2
4M 2
6
4M 2
5 = 2
525

32
2
ln
+
2
16 2
4M 2

2
T p
2
1
T
2
2
1r
+ +r
ln
24
8
16 2
4T
2
2
2
2
2

T
3
2r
ln
+ 5.0669 + ln 1 r 2 2 (2 ln 2 1)
+ 2
2
128 2
T2

1+r
1
ln 2 .
(30)
2 ln2
1r
The divergent parts of these diagrams can all be eliminated by a suitable choice of
the counterterm in Eq. (13). As mentioned above, we use the minimal subtraction
prescription.
3. Results
Having obtained an expression for the renormalized thermal mass, we can set = 1
where is determined via the PMS
and obtain numerical results for this mass, m2T , (),
condition:

m2T , ()

= 0.
(31)

=
At O(), does not depend on the coupling and so does not generate non-perturbative
information, but non-perturbative behaviour emerges at O( 2 ). Fig. 3 is a typical plot,
showing a clear maximum in m2T , ().
It will be of particular interest to study the dependence of the critical temperature Tc ,
the signal for the phase transition being taken to be m2Tc , = 2 , which leads to an implicit
equation for Tc = Tc () (with dependence on the renormalized bare mass and coupling
of the theory suppressed in the notation). Fig. 4 shows a plot of the contours of m2T , for
m2 = M 2 , = 0.5 in a region of the (T , ) plane, with the thicker line indicating the
critical temperature. Below this line the LDE breaks down the thermal mass being a
monotonically decreasing function of and thus lacking extrema.
In Fig. 5 we present a comparison of Tc () as calculated by the LDE with the first-order
estimate
m2 +
Tc2
= 2
12
(32)
provided by perturbation theory in the high temperature limit [7].

The two curves are in fact quite similar in shape, and converge at large Tc . The reason
for this behaviour is that in the high T limit, it so happens that 2 2 m2 , which is
from 4 in (29) we see that if this is
equivalent to r 2 1. Examining those contributions
2
the case, we require the coefficients of the 1/ 1 r terms to disappear, which is exactly
the same condition as (32).
526
Fig. 3. Dependence of the thermal mass squared on the variational parameter . All masses measured
in units of M.
Fig. 4. Contours of the thermal mass squared in intervals of 12 M 2 in the (T , ) plane for = 0.5
with m2 = M 2 . Both axes are in units of M.
Fig. 6 illustrates the high-T behaviour of the second-order approximation to the thermal
mass, with /Tc approaching a constant as Tc increases; the constant being determined
from (32) to be /12.
527
Fig. 5. Dependence of the critical temperature on at O( 2 ) for m2 = M 2 , = 0.5, with both

axes in units of M. The dotted line indicates the first-order approximation given by Eq. (32).
Fig. 6.The variation of /Tc with Tc for m2 = M 2 , = 0.5. The dashed line indicates the constant
value /12. Tc is in units of M.
4. Conclusions
In this paper we have shown how the problem of finite chemical potential for a charged
scalar theory can be formulated in the context of the linear delta expansion. The graphs to
be evaluated are essentially those of ordinary perturbation theory, with modified mass and
528
Fig. 7. The two-loop momentum-dependent diagram.
coupling parameters. The chemical potential appears explicitly in the Lagrangian, and also
in the BoseEinstein factors occurring in the momentum integrals. Renormalization has
been implemented on the lines of Ref. [6], where the importance of renormalizing before
applying the variational aspect of the method was emphasized. We obtain unambiguous
PMS points, as illustrated in Fig. 3.
The results have been plotted in Fig. 4 as a contour plot of the thermal mass in the (T , )
plane, and in Fig. 5 as Tc versus for given values of m2 , . This latter curve approaches
the result of resummed perturbation theory at high temperature, but differs significantly
from it at lower temperatures. The reason for the convergence at higher temperatures can be
understood in terms of the properties of the stationary points in the variational parameter .
The calculation given in Appendix A of the high-temperature limit of the setting-sun
diagram with 6= 0 may also be useful in other contexts: it can be adapted to standard
perturbation theory simply by substituting m for .
As emphasized in the introduction, the problem of a non-zero chemical potential is not
amenable to treatment by the usual Monte-Carlo lattice method. A lattice version of the
present calculation is in progress.
An extension of the present work which we intend to pursue in the near future is to free
it from the dependence on the high temperature expansion. We anticipate that at lower
temperatures the differences between the LDE and resummed perturbation theory will
become more pronounced.
Acknowledgements
We are very grateful to Dr. T.S. Evans for useful discussions and the benefit of his
expertise in thermal field theory. One of us (P.P.) wishes to thank the Particle Physics and
Astronomy Research Council of the UK for financial support.
Appendix A
In this appendix we will derive in some detail the expression for the two-loop setting sun
diagram. We begin by mimicking the approach of Parwani [2] to split up the diagram into
three parts; containing zero, one and two BoseEinstein factors, respectively. Explicitly,

2
5 =
2 2 4 2 X
M T
2
l,m
529
d d1 k d d1 q
F (il , k)F (im , q)
(2)d1 (2)d1
F (ir , r),
(A.1)
where d = 4 2 is the spacetime dimension and r = p k q. If we now use the mixed

representation of the Matsubara propagator defined in (18), the frequency sums become
trivial and eventually we have
2
5 (in , p) =
where
G0 (in , p) =

2 2
G0 (in , p) + G1 (in , p) + G2 (in , p) ,
2
(A.2)
d[k, q]S (k , q , r ),
(A.3)
Z
Z
G1 (in , p) =

d[k, q] n+
k S (k , q , r ) + S (k , q , r )
+ S+ (k , q , r ) + S (k , q , r )
+ n
k S (k , q , r ) + S (k , q , r )

+ S+ (k , q , r ) + S (k , q , r ) ,
Z

+ +
d[k, q] n+
G2 (in , p) =
k nq S (k , q , r ) + S (k , q , r )
+ S (k , q , r ) S+ (k , q , r )
+ n+
k nq S (k , q , r ) + S (k , q , r )
+ S+ (k , q , r ) S (k , q , r )
+
+
+ n
k nq S (k , q , r ) + S (k , q , r )
+ S (k , q , r ) S (k , q , r )
(A.4)

+ n
k nq S (k , q , r ) + S (k , q , r )

+ S (k , q , r ) S (k , q , r ) ,
(A.5)
with the definitions

S (k , q , r ) =
1
,
(in + ) + k + q + r
S (k , q , r ) = S+ (k , q , r ) + S (k , q , r ),
d[k, q] = M 4
d d1 k d d1 q
1
.
(2)d1 (2)d1 8k q r
We choose to evaluate the self-energy on shell (p = 0, in = ). The reason we

evaluate at this point rather than at p = 0, in = is best explained as follows[16]. In
b Q
b rather
beff = H
setting up the theory, we are dealing with an effective Hamiltonian, H
b. We would naturally choose to evaluate the self-energy at
than the real Hamiltonian, H
in = for the real Hamiltonian, which means evaluating at in = in the effective
theory. Following this prescription, we find that
530

< G0 ( , 0)

2
2
3 2
3 2
1
1
p2
ln
+
=
2
2
2
2
2
2
2(16 )

4M
(32 )

2
2
17
3 2
2
ln
ln
+
2
+
1.9785
.
(16 2 )2
4M 2
6
4M 2
(A.6)
We now calculate the contribution from G1 . Writing
G1 (in , p) = G+
1 (in , p) + G1 (in , p),
G
1
(A.7)
n
k,
represents that part of G1 with an associated BE factor of

respectively, and
where
decomposing this factor into a UV divergent part and a UV finite part in the manner of the
= 0 case one obtains

2
(A.8)
< G
1 ( , 0) = F0 + F1 + F2 .
Here
3
32 2
3
=
32 2
F0 =
F1
1
T2
h3 (y, r) ,

2

4M 2
T2
h
(y,
r)
ln
+
2
3
2
2
(A.9)
(A.10)
and
F2
1
=
4(2)4
Z
0
with
kn
dk k
k
Z
0

X
dq
6k
q ln +

q
X

= 2 (k + q + kq )2 2 (k + q + kq )2
X

( k )2 (q + kq )2 ,
(A.11)
(A.12)
where the subscript refers to the kq term. Following a similar procedure to that of
Ref. [2] we find that in the high-T limit

T2
0.54597
.
(A.13)
ln
F2 2
T
128 2
Thus, collecting all the terms together,

< G1 ( , 0) = F0 + F1 + F2 2
(A.14)
with
3T 2 e
1
h3 (y, r) ,
4

16

3T 2 e
2
h (y, r) ln
F1 =
+2
16 4 3
4M 2
F0 =
and
F2

T2
0.54597 .
ln
64 2
T
(A.15)
(A.16)
(A.17)
531
Finally, we consider the contribution from G2 , which contains a BE factor for each loop
and so is UV finite. We write

(A.18)
< G2 ( , 0) = H ++ 2 + H + 2 + H + 2 + H 2
with
H ( 2 ) =
1
4(2)4
Z
dk
0
kn
k
k
Z
dq
qn
q
q

Y
ln + ,
Y
(A.19)
where

Y++ = 2 (k + q + kq )2 2 (k q + kq )2

2
,
( + k + q )2 kq

Y+ = 2 (k + q + kq )2 2 (k + q kq )2

2
,
( + k q )2 kq
2
2

+
2
Y = (k + q + kq ) (k + q kq )2

2
,
( k + q )2 kq

Y = 2 (k + q + kq )2 2 (k q + kq )2

2
.
( k q )2 kq
(A.20)
(A.21)
(A.22)
(A.23)
Each of the H pieces has a logarithmic IR divergence as 0. To deal with this,

we extract the leading order behaviour of ln |Y+ /Y | in this limit, separating into three
terms as follows:

+
Y+

3 k + q
Y+
kq

(A.24)
ln + ln ln 2 ln p
+ 2 ln k q .
Y
Y
k2 q 2
Collecting those terms involving the ln 2 piece alone, which we can take outside the
integral, we calculate the coefficient of this term in the high-temperature limit a 1,
which we call Cln 2 , to be

T2
2 1+r
ln
.
(A.25)
Cln 2 =
(8 2 )2
1r
The remainder of <[G2 ( , 0)] splits into two pieces which we define by
1
1
=
4(2)4

H 2 2 =
3
8(2)4

k dk +
nk n
k
k

k dk +
nk + n
k
k

q dq +
p kq
, (A.26)
nq n
ln
q

q
k2 q 2

k + q
q dq +

nq + n
ln
q
k q .
q
(A.27)
Extracting the high-temperature limit of H 1 ( 2 ) gives, after using the symmetry of the
integrand under k q to restrict the range of the q integration and making the changes of
variable = k , = q ,
532
2
=
4(2)4
Z
d
n+

n
where
= 1/(ea( r) 1).
n+
n 2ra
Z
d
1
n+

( 2 1)(2 1)
, (A.28)

ln

2 2
After making the successive approximations 1
ea
2r
2
a
2
(e 1)
a
in the integrand, we have, for a 1,

T 2r 2
1
1
2
.
ln 2
H =
(2)4
2
The calculation of H 2 ( 2 ) parallels the case for = 0, leading to
2

2
3T 2
1
ln
H 2 2 =
1.50699
.
64 2
2
T2
(A.29)
(A.30)
(A.31)
References
[1] A. Linde, Rep. Prog. Phys. 42 (1979) 389;
D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43.
[2] R.R. Parwani, Phys. Rev. D 45 (1992) 4695.
[3] N. Banerjee, S. Mallik, Phys. Rev. D 43 (1991) 3368.
[4] K. Funakubo, M. Sakamoto, Phys. Lett. B 186 (1987) 205.
[5] I.R.C. Buckley, A. Duncan, H.F. Jones, Phys. Rev. D 47 (1993) 2554;
I.L. Solovtsov, Phys. Lett. B 327 (1994) 335;
C.M. Bender, A. Duncan, H.F. Jones, Phys. Rev. D 49 (1994) 4219;
T.S. Evans, H.F. Jones, A. Ritz, Nucl. Phys. B 517 (1998) 599.
[6] M.B. Pinto, R.O. Ramos, Phys. Rev. D 60 (1999) 105005.
[7] K.M. Benson, J. Bernstein, S. Dodelson, Phys. Rev. D 44 (1991) 2480.
[8] K. Funakubo, M. Sakamoto, Prog. Th. Phys. 76 (1986) 490.
[9] C. Arvanitis, H.F. Jones, C.S. Parker, Phys. Rev. D 52 (1995) 3704.
[10] P.M. Stevenson, Phys. Rev. D 23 (1981) 2916.
[11] S. Chiku, T. Hatsuda, Phys. Rev. D 58 (1998) 076001.
[12] M. Le Bellac, Thermal Field Theory, Cambridge University Press, 1996.
[13] R.D. Pisarski, Nucl. Phys. B 309 (1988) 476.
[14] J.C. Collins, Renormalization, Cambridge University Press, 1983.
[15] H.E. Haber, H.A. Weldon, J. Math. Phys. 23 (1982) 1852.
[16] T.S. Evans, The condensed matter limit of relativistic QFT, in: Y.X. Gui, F.C. Khanna, Z.B. Su
(Eds.), Fourth Workshop on Thermal Field Theories and their Applications, 1995, Dalian,
China, World Scientific, Singapore, 1996, pp. 283295.
1 The second approximation in (A.29) is strictly valid only for |a | 1; however, its validity is assured by the
convergence of the resulting integral and has been checked numerically.
Nuclear Physics B 594 [FS] (2001) 535606

Quantum brownian motion on a triangular lattice

and c = 2 boundary conformal field theory
Ian Affleck a,b , Masaki Oshikawa c , Hubert Saleur d,1
a Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106-4030, USA
b Canadian Institute for Advanced Research and Department of Physics and Astronomy,
University of British Columbia, Vancouver, BC, V6T 1Z1, Canada

c Department of Physics, Tokyo Institute of Technology, Oh-okayama, Meguro-ku, Tokyo 152-8551, Japan
d S.Ph.T., C.E.N. Saclay, 91191 Gif-sur-Yvette, Cedex, France
Received 12 July 2000; accepted 31 July 2000
Abstract
We study a single particle diffusing on a triangular lattice and interacting with a heat bath,
using boundary conformal field theory (CFT) and exact integrability techniques. We derive a
correspondence between the phase diagram of this problem and that recently obtained for the
2-dimensional 3-state Potts model with a boundary. Exact results are obtained on phases with
intermediate mobilities. These correspond to nontrivial boundary states in a conformal field theory
with 2 free bosons which we explicitly construct for the first time. These conformally invariant
boundary conditions are not simply products of Dirichlet and Neumann ones and unlike those trivial
boundary conditions, are not invariant under a Heisenberg algebra. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 11.25.Hf; 05.40.Jc; 73.40.Gk
1. Introduction
Conformal field theory (CFT) with boundaries finds applications both to open string
theory and to various quantum impurity problems in condensed matter physics. These
generally describe gapless bulk excitations interacting with some localized degrees of
freedom. In these problems the gapless bulk excitations can be represented by a conformal
field theory in (1 + 1) spacetime dimensions, often simply free bosons. It is generally
found that the boundary dynamics renormalize, at low energies, to a conformally invariant
boundary condition (b.c.).
1 Physics Department and CIT-USC Center for Theoretical Physics, University of Southern California, Los
Angeles, CA90089-0484, USA.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 9 9 - 5
536
I. Affleck et al. / Nuclear Physics B 594 [FS] (2001) 535606
Quantum brownian motion (QBM) provides an intriguing example of such a problem.

Here one considers a heavy particle moving in a d-dimensional periodic potential and
interacting with a heat bath. A simplified model for the heat bath [13] is an infinite set
of harmonic oscillators coupled linearly to the particle co-ordinate. (This is the obvious
generalization of the CaldeiraLeggett model [4] for a particle in a double well potential.)
When the oscillator spectral weight vanishes linearly at low frequencies, corresponding
to ohmic dissipation, the set of oscillators may be represented by a (1 + 1)-dimensional
quantum conformal field theory of free massless bosons living on a fictitious half-line with
the heavy particle at the origin. The number of bosonic fields required is given by the
dimensionality of the space in which the heavy particle is diffusing (normally three). The
E
boson fields at the origin, (0),
correspond to the momentum of the particle and the dual
E
fields, (0) to the particles position. The boson CFT may be regarded as compactified on
the lattice on which the heavy particle is diffusing. The dimensionless compactification
radius (scaled by the coupling to the bath) is a crucial parameter for QBM. The fictitious
extra dimension is analogous to the coordinate along the string, , in open string theory,
while the field E plays the role of the string coordinates in D-dimensional spacetime, X .
Conformal invariant boundary conditions obtained in the dissipative quantum mechanics
framework have therefore immediate applications to open string theory, and potential
interpretations in terms of D-branes [5].
Until recently it was generally believed that, depending on the strength of the dissipation
relative to the period of the potential, the particle could only be in either localized or freely
E
E =
diffusing phases corresponding to Dirichlet (D): (0)
= constant, or Neumann (N): (0)
constant, boundary conditions in the CFT, respectively. However, it was recently shown [6]
that, for certain lattices, other phases are possible with intermediate mobility, in which the
particle is neither perfectly localized nor freely diffusing. These correspond to nontrivial
conformally invariant boundary conditions in the free boson CFT which are neither D
nor N. In fact, the existence of such phases was shown earlier in another quantum impurity
problem: tunneling though a single impurity in a quantum wire [7,8].
While these interesting phases seem to cry out for a general boundary CFT solution this
has, so far, eluded us. The problem of classifying all (or all physically relevant) conformally
invariant boundary conditions in c = 2 conformal field theory (2 free bosons) remains very
much open. In the much simpler case of 1 free boson, c = 1, it is widely believed that only
D and N phases generally occur. On the other hand, for conformal field theories with a finite
number of conformal towers, an elegant way of generating nontrivial conformal boundary
conditions is provided by fusion. This method can be used to generate an apparently
complete set of conformally invariant b.c.s in, for example, the Ising model [9] or 3-state
Potts model [10]. In the case of WessZuminoWitten (WZW) models, one can obtain
by fusion the boundary conditions that describe the low temperature phases of various
generalized Kondo problems [11]. However, the infinite number of conformal towers
present in c = 2 CFT makes the generalization of the fusion technique rather subtle. Indeed,
it is not even clear in general how to understand the simple D and N b.c.s this way. Whether
or not some generalized notion of fusion will ultimately provide a complete solution to this
537
problem for c = 2 is unclear at present. Needless to say, the generalization to larger values
of c, in particular c = 3 corresponding to 3-dimensional QBM, is also an open problem.
The original argument [7,8] for the existence of nontrivial phases came from a study
of the phase diagram in the quantum wire problem. This was studied [8] by a type of expansion. By fine-tuning the compactification radius the nontrivial fixed points can be
moved arbitrarily close to either N or D fixed points, allowing a perturbative expansion. The
other solved case for c = 2 was obtained in the context of QBM on a triangular lattice. For
a special value of dissipation strength (compactification radius) a beautiful exact mapping
of the problem onto the critical point of the 3-channel Kondo problem [11] was obtained
by Yi and Kane [6], allowing use of results from that problem [11] based on boundary CFT
in the SU(2)3 WZW model. Although it can be seen that nontrivial phases exist for a range
of dissipation strength on the triangular lattice, no general solution has been found.
Several years after the CaldeiraLeggett type model was proposed for QBM [1] it was
argued [12,13] that this model does not provide a valid description of a heavy particle
interacting with phonons or electronhole pairs. In particular, the localized phase does not
occur in physical models of quantum brownian motion [14]. Thus physical applications
of the model may only be to the quantum wire and related problems. Nevertheless,
the CaldeiraLeggett type QBM model is perfectly consistent as an ultraviolet (UV)
regularization of a boundary conformal field theory.
In this paper we re-examine the soluble triangular lattice QBM YiKane model. Our
purpose is both to understand this fascinating system better and also to shed light on
the general boundary CFT problem. Rather than relating this problem to the SU(2)3
WZW model we instead relate it to the 3-state Potts model. This is done via a conformal
embedding whereby the bulk degrees of freedom are represented by a direct sum of the
3-state Potts model, an Ising model and a tricritical Ising model, satisfying:
c = cPotts + cIsing + ctricritical = 4/5 + 1/2 + 7/10 = 2.
(1.1)
The Z3 symmetry of the Potts model corresponds to the point group symmetry of the
appropriate triangular lattice model. This seems like a more natural formulation of the
QBM problem since it does not have an SU(2) symmetry. The 4 conformally invariant
boundary conditions in the Potts model: free, fixed, mixed and new are shown to
correspond to the 4 phases that can occur in the QBM problem. In this way it is possible to
obtain the various fixed points by fusion in the Potts sector of the theory. These four fixed
points correspond to localized and freely diffusing fixed points, the nontrivial fixed point of
Yi and Kane and one new fixed point not previously known. The mobilities of these phases
and also the groundstate entropies can be calculated from the fusion approach. We also
show that it is possible to use an alternative conformal embedding to Eq. (1.1) where the
Ising and tricritical Ising sectors are replaced by a Z3(5) conformal field theory, which is a
sort of multicritical Potts model [16]. We demonstrate that some of the crossovers between
these fixed points induced by relevant boundary operators can be studied exactly using
integrability techniques. This would allow, for example, exact calculations of universal
quantities at all temperatures for a system which is crossing over between freely diffusing
behavior at higher temperatures and localized or nontrivial behavior at T = 0. The
538
integrability also helps to understand the phase diagram of the model by determining all
RG flows from a given fixed point.
In the next section we review the fusion approach to boundary CFT and the standard
D and N b.c.s for free bosons. We also show there that it is possible to obtain the D b.c.
from the N b.c. by fusion with a twist field for the c = 1 case of a single periodic boson,
for certain rational radii. In Section 3 we introduce the QBM problem and its connection
with boundary CFT, with careful attention to the boundary conditions and the groundstate
degeneracy. We show that this degeneracy scales as the size of the space on which the
particle moves. We derive, apparently for the first time, the connection between the particle
co-ordinate and the value of the dual field at the origin. In Section 4 we introduce a model
of QBM on a triangular lattice which generalizes the model of Yi and Kane, and conjecture
its phase diagram, suggesting the analogy with the Potts model. In Section 5 we discuss
our conformal embedding and show how fusion in the Potts sector gives the various fixed
points conjectured in Section 4. We also calculate the groundstate entropies and mobilities
of these fixed points and discuss the corresponding boundary states. We also discuss some
additional fixed points, that occur for more general Hamiltonians, and are obtained by
(5)
fusion in the Ising, tricritical Ising or Z3 sectors. In Section 6 we discuss the integrable
flows between the various fixed points. In Section 7 we discuss generalizations of this
model to (n 1) bosons, corresponding to QBM in (n 1) dimensions and also discuss
the relationship with the n-channel Kondo problem.
2. Boundary conformal field theory

As will be shown in the next section, QBM can be formulated in terms of (1 + 1)dimensional massless bosons on the half-line, x > 0, with interactions only at the boundary,
x = 0. As such it is related to a number of other problems in particle physics and condensed
matter physics that have been successfully studied using boundary conformal field theory
(BCFT). The basic assumption in this approach is that the boundary dynamics renormalize,
at low energies, to an effective conformally invariant b.c. A constructive approach to
classifying possible fixed points is then to attempt to enumerate all possible conformally
invariant b.c.s corresponding to a given critical bulk theory. This has met with great success
in situations where the bulk theory can be formulated in terms of a finite number of
conformal towers, with the set of conformally invariant b.c.s being generally in one to
one correspondence with the bulk conformal towers. In this case, the fusion technique is
very useful in constructing the conformally invariant b.c.s. Unfortunately, many interesting
quantum impurity problems apparently cannot be formulated in terms of a finite number of
bulk conformal towers, generally corresponding to integer values of c. Consequently the
problem of studying the fixed points in these cases remains very much open. As we will
show in Section 5, for a special choice of parameters, QBM on a triangular lattice can be
related, using a conformal embedding, to a finite number of conformal towers.
In this section we review two quite different approaches to boundary conformal field
theory. The fusion approach of Cardy which has been very successful for CFTs with a
539
finite number of conformal towers, and the straightforward D and N b.c.s for free bosons.
It is not known, in general, how to apply the fusion approach for free bosons but we
demonstrate that this is possible for a single free boson at a special (discrete, infinite)
set of compactification radii.
2.1. Fusion approach to boundary CFT
We review briefly some important results due to Cardy on the case of a finite number of
conformal towers.
Precisely what is meant by a conformally invariant boundary condition? Without
boundaries, conformal transformations are analytic mappings of the complex plane:
z + ix,
(2.1)
into itself:
z w(z).
We may Taylor expand an arbitrary conformal transformation around the origin:
X
an z n ,
w(z) =
(2.2)
(2.3)
where the an s are arbitrary complex coefficients.

They label the various generators of the conformal group. It is the fact that there is
an infinite number of generators (i.e., coefficients) which makes conformal invariance so
powerful in (1 + 1) dimensions. Now suppose that we have a boundary at x = 0, the real
axis. At best, we might hope to have invariance under all transformations which leave the
boundary fixed. This implies the condition:
w( ) = w( ),
(2.4)
where denotes imaginary time.

We see that there is still an infinite number of generators, corresponding to the an s of
Eq. (2.3) except that now we must impose the conditions:
an = an .
(2.5)
We have reduced the (still ) number of generators by a factor of 1/2. The fact that there
is still an number of generators, even in the presence of a boundary, means that this
boundary conformal symmetry remains extremely powerful.
Boundary conformal invariance implies that the momentum density operator, T TS
vanishes at the boundary. This amounts to a type of unitarity condition. Since T (t, x) =
T (t + x) and TS(t, x) = TS(t x), it follows that
TS(t, x) = T (t, x),
(2.6)
i.e., we may regard TS as the analytic continuation of T to the negative axis. Thus instead
of working with left and right movers on the half-line we may work with left-movers only
on the entire line.
540
Fig. 1. Cylinder of length l, circumference with boundary conditions A and B at the two ends.
When the bulk theory contains a conserved current, J , it is possible to impose additional
symmetry requirements on the boundary conditions of the form:
J (0) = J(0).
(2.7)
In the case of free bosons the generic symmetry consists of a product of U (1) current
algebras (one for each boson) known as a Heisenberg algebra. It is then possible, more
generally, to impose an invariance of the form:
JE(0) = RJE (0),
(2.8)
where the vector JE contains the U (1) currents and R is a rotation matrix. Since the chiral
energy density operators, T and TS can be written in Sugawara form:
1 E 2
1 E2
J ,
J ,
TS =
(2.9)
4
4
it follows that Eq. (2.6) is still obeyed. One of the lessons of this paper is that, while it
is fairly straightforward to construct boundary conditions in free boson theories that are
invariant with respect to the full Heisenberg algebra, it is much more difficult, apparently,
to find the other boundary conditions which obey only the Virasoro condition of Eq. (2.6)
and not the additional constraints of Eq. (2.8).
To exploit the conformal invariance, following Cardy, it is very convenient to consider a
conformally invariant system defined on a cylinder of circumference in the -direction
(imaginary time) and length l in the x direction, with conformally invariant boundary
conditions A and B at the two ends (see Fig. 1). From the quantum mechanical point
of view, this corresponds to a finite temperature, T = 1/. The partition function for this
system is:
T=
ZAB = tr eHAB ,
l
(2.10)
where we are careful to label the Hamiltonian by the boundary conditions as well as
the length of the spatial interval, both of which help to determine the spectrum. (We
sometimes refer to this as the open string channel.) Alternatively, we may make a modular
transformation, x. Now the spatial interval, of length, , is periodic (the closed string
channel). We write the corresponding Hamiltonian as HP . The system propagates for a

time interval l between initial and final states A and B. Thus we may equally well write:
ZAB = hA|elHP |Bi.
(2.11)
541
Equating these two expressions, Eqs. (2.10) and (2.11) gives powerful constraints which
allow us to determine the conformally invariant boundary conditions.
Due to the condition of Eq. (2.6), in the purely left-moving formulation, the energy
momentum density, T is basically unaware of the boundary condition. Hence, in
calculating the spectrum of the system with boundary conditions A and B introduced
above, we may regard the system as being defined periodically on a torus of length 2l with
left-movers only. The conformal towers of T are unaffected by the boundary conditions,
A, B. However, which conformal towers occur does depend on these boundary conditions.
We introduce the characters of the Virasoro algebra, for the various conformal towers:
X E a (2l)
i
e
,
(2.12)
a e/ l
i
where
Eia (2l)
are the energies in the ath conformal tower for length 2l, i.e.,
a c
x
,
(2.13)
l i
24l
where the xia s correspond to the (left) scaling dimensions of the operators in the theory
and c is the conformal anomaly.
l can only consist of some combination of these conformal towers,
The spectrum of HAB
i.e.,
X

naAB a e/ l ,
(2.14)
ZAB =
Eia (2l) =
naAB
are some non-negative integers giving the multiplicity with which the
where the
various conformal towers occur. For minimal conformal field theories, a runs over the finite
set of irreducible representations of the Virasoro algebra. For more complicated theories
like the ones of interest in this paper, the corresponding set is infinite, and it is not always
clear what kind or representations to expect on the right hand side of this equation. We
stress that we are interested in boundary conditions restricted by conformal invariance
only, so characters of generalized symmetry algebras are not relevant here.
Importantly, only these multiplicities depend on the boundary conditions, not the
characters, which are a property of the bulk left-moving system. Thus, a specification
of all possible multiplicities, naAB amounts to a specification of all possible boundary
conditions A. The problem of specifying conformally invariant boundary conditions has
been reduced to determining sets of integers, naAB . For minimal conformal field theories,
where the number of conformal towers is finite, only a finite number of integers needs to
be specified.
Now let us focus on the boundary states, |Ai. These must obey the operator condition:

T (x) TS(x) |Ai = 0 (x).
(2.15)
(Note that, after the modular transformation, x denotes the periodic co-ordinate.) Imposing
periodic boundary conditions along the boundary and Fourier transforming with respect to
x, this becomes:
[Ln L n ]|Ai = 0.
(2.16)
542
This implies that all boundary states, |Ai must be linear combinations of the Ishibashi
states [17]:
X
|a; mi |a; mi.
(2.17)
|ai
m
Here m labels all states in the ath conformal tower. The first and second factors in
Eq. (2.17) refer to the left and right-moving sectors of the Hilbert space. Here, a belongs to
the set of representations of the Virasoro algebra which appear simultaneously in the right
and left sector of the theory with periodic boundary conditions. It is in general a smaller
set than the one appearing in Eq. (2.11). Thus we may write:
X
|aiha0|Ai.
(2.18)
|Ai =
a
Here,
|a0i |a; 0i |a; 0i.
(2.19)
(Note that while the states, |a; mi |b; ni form a complete orthonormal set, the Ishibashi
states, |ai do not have finite norm.) Thus, specification of boundary states is reduced to
determining the matrix elements, ha0|Ai. For minimal conformal field theories, there is a
finite number of such matrix elements. Thus the partition function becomes:
X
hA|a0iha0|Biha|elHP |ai.
(2.20)
ZAB =
a
From the definition of the Ishibashi state, |ai we see that:

X
a
e2lEm () ,
ha|elHP |ai =
(2.21)
the factor of 2 in the exponent arising from the equal contribution to the energy from T
and TS. This can be written in terms of the characters:

(2.22)
ha|elHP |ai = a e4l/ .
We are now in a position to equate these two expressions for ZAB :
X
X

hA|a0iha0|Bia e4l/ ) =
naAB a e/ l .
ZAB =
a
(2.23)
This equation must be true for all values of l/. It is very convenient to use the modular
transformation of the characters [18,19]:
X b

Sa b e4l/ .
(2.24)
a e/ l =
b
Here we refer to Sba is as the matrix of modular transformations. We thus obtain a set
of equations relating the multiplicities, naAB which determine the spectrum for a pair of
boundary conditions and the matrix elements ha0|Ai determining the boundary states:
X
Sba nbAB = hA|a0iha0|Bi.
(2.25)
b
We refer to these as Cardys equations; they basically allow a determination of the

boundary states and spectrum. The problem is to find a set of boundary states A (defined
543
by the coefficients hA|a0i) satisfying Cardys equations, such that, for any pair A, B in this
set, the nbAB are non-negative integer coefficients, the identity representation appearing at
most once, n0AB 6 1. An important point in deriving (2.25) is the linear independence of
the characters. This property may actually not hold for nondiagonal theories. In such
cases, there will generally be some other quantum number that allows one to distinguish
the corresponding operators, and still prove Eq. (2.25). The study of these equations has
given rise to considerable theoretical activity recently [2022]. In the case of rational
conformal field theories, and with the additional key hypothesis of completeness, it has
been shown that finding sets of conformally invariant boundary conditions is equivalent to
finding integer valued representations of the fusion algebra. The latter problem is under
reasonable control, and as a result, conformal boundary conditions for minimal models
have for instance been classified in Ref. [21].
In the present paper, we are interested in a more difficult problem: in fact, since we look
for boundary conditions respecting only conformal invariance, the two boson problem has
the complexity of an irrational theory, even for rational values of the radius. Very little is
known about this case, and we rely heavily on a not entirely systematic but highly efficient
method called fusion, which is inspired by Cardys seminal paper [9], and was used with
success in the Kondo problem [11].
Generally, boundary states corresponding to trivial boundary conditions can be found
by inspection, i.e., given nbAA we can find ha|Ai. We can then generate new (sometimes
nontrivial) boundary states by fusion. I.e., given any conformal tower, c, we can obtain
a new boundary state |Bi and new spectrum naAB from the fusion rule coefficients,
c
. These non-negative integers are defined by the operator product expansion (OPE) for
Nab
(chiral) primary operators, a . In general the (OPE) of a with b contains the operator
c times. We denote by B the new boundary condition and boundary state, generated
c Nab
by fusion from the boundary condition A by fusion with the operator c. The partition
function, ZBD , with an arbitrary boundary condition D at the other end of the system, has
a as
multiplicities given by the fusion rule coefficients Nbc
X
a b
Nbc
nAD ,
(2.26)
naDB =
b
a
are non-negative integers, the new
hence the name fusion construction. Because Nbc
a
multiplicities nDB are also non-negative integers, and thus physical. Alternatively, in the
closed string channel, the new boundary state |Bi is defined through
ha0|Bi = ha0|Ai
Sca
,
S0a
(2.27)
where 0 labels the conformal tower of the identity operator.

Let us now show that the boundary state (2.27) gives the multiplicity (2.26), i.e., they
satisfy Cardys equation (2.25). The right-hand side of Eq. (2.25) becomes:
hD|a0iha0|Bi = hD|a0iha0|Ai
The left-hand side becomes:
Sca
.
S0a
(2.28)
544
Sba nbDB =
b d
Sba Ndc
nDA .
(2.29)
b,d
We now use a remarkable identity relating the matrix of modular transformations to the
fusion rule coefficients, known as the Verlinde formula [23]:
X
Sa Sa
b
Sba Ndc
= d ac .
(2.30)
S0
b
This gives:
X
Sa X a d
Sa
Sba nbDB = ca
Sd nDA = ca hD|a0iha0|Ai = hD|a0iha0|Bi,
S0
S0
b
(2.31)
proving that fusion does indeed give a new solution of Cardys equations. The multiplicities, naBB are given by double fusion:
X
a
b d
Nbc
Ndc
nAA .
(2.32)
naBB =
b,d
(Recall that |Bi is obtained from |Ai by fusion with the primary operator c.) It can be
checked that the Cardy equation with A = B is then obeyed.
As mentioned before, the key problem in boundary CFT is the construction of a
complete set of boundary states (and b.c.s), that is the largest possible set of boundary
states, |Ai i such that ZAi Aj is a physical partition function. By physical we mean that
the partition function with any two boundary conditions is always given by non-negative
integer multiplicities naAi Aj . Noting that any linear combination of physical boundary states
with non-negative integer coefficients,
X
ni |Ai i,
(2.33)
i
also gives a physical partition function in this sense, we see that we must impose an
additional condition to eliminate such states. The lowest energy state with the same b.c.
at both ends of the system is independent of the b.c. corresponding to the absolute finite
size groundstate, with xi = 0. (This follows from making a conformal mapping to the
half-plane and using the fact that the identity boundary operator exists with any b.c.)
We may choose to impose an additional condition that ZAi Ai contains the zero energy
state exactly once, n0Ai Ai = 1. This eliminates linear combination states. Of course, once
we have found a complete set, we may always form linear combinations with nonnegative integer coefficients. Actually, the fusion construction sometimes gives such linear
combination states. These can also arise from integrable flows [2426] and have physical
applications [27] in some cases. (The existence of extra dimension 0 boundary operators is
often associated with first order phase transitions.)
In the case of minimal diagonal theories, conformal boundary conditions are in one to
one correspondence with the set of Virasoro representations [9,21]. The corresponding
defined
boundary states can be generated by fusion starting from the the boundary state |0i
a
a0
by its diagonal partition function, n = .
00
545
In more complicated cases, fusion has proven an invaluable tool. One of the key
advantages of the method is that one can use fusion with fields that are not in the spectrum
of the bulk theory. Indeed, as already explained above, while the the boundary state (in the
closed string channel) must be compatible with the bulk spectrum, the conformal towers
appearing in the open string channel do not have to be in the bulk spectrum. In many cases,
it is then possible to use not the fusion algebra of the original bulk theory but another
one, still compatible with the modular transformation matrix, to generate boundary states.
For instance, in the case of the three state (A4 , D4 ) Potts model, the characters appearing
in the bulk partition function are a subset of all the possible characters at c = 45 . Using
the full matrix of modular transformations and fusion algebra for the associated diagonal
tetracritical Ising model (A4 , A5 ) led to the discovery of the new boundary condition in
Ref. [10]. We will discuss in the next section how the fusion method can successfully be
applied to the case of a free boson.
Before leaving this section, we should emphasize that complete sets of boundary states
form compatible sets of states which may be overlapping. For example, in the Ising model
the three states corresponding to spin up, spin down and free b.c.s are a complete mutually
compatible set. However, one could also consider imposing an up or down b.c. on the dual
spins in the Ising model. Such states should also be compatible with the free b.c. but not
with the fixed b.c.s on the original (nondual) spins. This incompatibility is presumably
related to the nonlocal nature of the dual spin variables when expressed in terms of the
original spins. This implies that one could not define a physical partition function with the
original spins pointing up at one boundary and the dual spins pointing up (or down) at the
other. In a physical application one typically begins with a particular b.c. and then wants
to find all other ones which are compatible with it. For instance, we may be interested
in adding arbitrary local interactions near the boundary and finding all b.c.s to which the
system can renormalize.
2.2. Dirichlet, Neumann, and rotated boundary conditions for free bosons
We review the simplest conformally invariant boundary conditions, namely Dirichlet
and Neumann boundary conditions and some of their generalizations, for multicomponent
compactified free bosonic field theory [15]. The Lagrangian density of the free boson
theory is given by
L=
2
1
E ,
2
(2.34)
where E is a c-component boson field. We identify the boson field as

E E + 2 uE
(2.35)
where uE is any vector in the compactification lattice , which is a c-dimensional Bravais

lattice.
Let us consider the finite size system of length l, with periodic boundary conditions. The
standard canonical quantization gives the canonical equal-time commutation relations
546
X

(x x 0 ml),
j (t, x), k (t, x 0 ) = i j k
(2.36)
E Together with the equation of motion E = 0, we obtain the mode

E = .
where
E
expansion of the field :

X
pt
E
2x
aEnL i 2 n (x+t )
E
E
l
uE +
+
+ h.c.
e
(t, x) = 0 +
l
l
4n
n=1

X
aEnR i 2 n (xt )
e l
+ h.c. ,
(2.37)
+
4n
n=1
where uE belongs to and pE is the conjugate momentum of E0 . (h.c. denotes Hermitean
E which
conjugate.) Because E0 should be identified as E0 E0 +2 , the eigenvalues of p,
we label vE, belong to the lattice which is the dual of . This is defined by the condition:
uE vE Z,
(2.38)
for all vectors uE and vE . Each component of the vector aEnL is an annihilation
j
j
operator; the anL satisfies the commutation relation [anL , a k mL ] = j k nm and all the other
j
commutators involving anL vanish. A similar definition applies to aEnR , which commutes
j
with amL and a j mL .
Now we can decompose the field E into chiral components, namely left-mover EL and
right-mover ER :

X
1
E0 E0
aEnL i 2 nx +
+
+ (2 uE + p)x
E ++
+ h.c. ,
e l
EL (x + ) =
2
2
2l
4n
(2.39)
n=1
where x + = t + x. The dual field E = EL ER has the mode expansion:

X
x
2t
aEnL i 2 n (x+t )
(t,
E x) = E 0 +
l
e
uE + vE +
+ h.c.
l
l
4n
n=1

X
aEnR i 2 n (xt )
e l
+ h.c. .
4n
n=1
(2.40)
E 2 . The
The chiral component of the energymomentum tensor is given by T (x + ) = (+ )
mode expansion of the energymomentum tensor gives the generators Lm of the Virasoro
algebra:
2 X
+ 2
Lm eimx l .
(2.41)
T (x + ) = 2
l m
Using the mode expansion (2.39), we obtain
1X
: E ml,L E lL : ,
Lm =
2
l
(2.42)
where :: denotes the normal ordering and

i n aEnL
(n > 0),
1
E (n = 0),
E nL (2 uE + p)
4

(n < 0).
+i n aEnL
547
(2.43)
The generators L m for the other chirality are given by

1X
: E ml,R E lR : ,
L m =
2
(2.44)
where

i n aEnR
(n > 0),
1
E (n = 0),
E nR (2 uE + p)
4

(n < 0).
+i n aEnR
(2.45)
An oscillator vacuum |vaci of this theory, which satisfies anL/R |vaci = 0, is thus
characterized by two sets of zero-mode quantum numbers uE and vE , where vE is
the eigenvalue of the operator p.
E We will denote the oscillator vacuum with these quantum
numbers as |(E
u, vE)i.
Now let us consider the possible conformally invariant b.c.s. As was discussed in
Section 2.1, the conformal invariance of the b.c., Eq. (2.16), implies that these are linear
combinations of the Ishibashi states, defined in Eq. (2.17). Of course, the primary states
must be contained within the Hilbert space of the system (with periodic b.c.s.). A physical
boundary state must satisfy Cardys equation (2.25), which gives a strong constraint on the
coefficients of the Ishibashi states.
The (multicomponent) free boson field theory, which we discuss in the present paper,
has an infinite number of primary fields. The classification of the boundary states for this
case is much more difficult than that for CFTs with a finite number of conformal towers
discussed in Section 2.1. In fact, at present we are far from the complete classification.
Even for the single-component c = 1 theory, for which the Dirichlet/Neumann b.c. and
their generalizations are generally believed to form the complete set, we know no complete
proof of this. Here we just present some simple boundary states for a multicomponent
free boson theory. Some examples of more nontrivial boundary states for c = 2 will be
constructed using a conformal embedding in Section 5.
An oscillator vacuum |(E
u, vE)i is naturally a primary state with respect to the Virasoro
algebra. However, there are an infinite number of Virasoro primaries which are not
oscillator vacua. The most general boundary states would include linear combinations of
Ishibashi states based on these complicated Virasoro primaries. Nevertheless, as we will
show in the following, the simplest boundary states can be written in terms of oscillator
vacua and creation operators of oscillator bosons.
A sufficient (but not necessary) condition for the free boson theory to satisfy the
conformal invariance (2.16) is given by
548
(E
nL E nR )|Bi = 0,
(2.46)
for all integer n. This corresponds to the imposition of invariance under the Heisenberg
algebra, as discussed in the previous subsection, near Eq. (2.7). For n = 0, it reads
uE |Bi = 0. It means that, to satisfy Eq. (2.46), we can use only the states built on the
oscillator vacua with uE = 0. It can be shown that the condition (2.46) is satisfied by the
state

X

0,
E vE .
E vE = exp
(2.47)
aEnL aEnR 0,
n
This might be regarded as an Ishibashi state with respect to the U (c) current algebra. It is
a linear combination of (an infinite number of) Ishibashi states with respect to the Virasoro
algebra.
While each state (2.47) satisfies (2.46) and hence the conformal invariance, it does not
satisfy Cardys consistency condition. Let us consider a linear combination of (2.47)
X

N(E0 ) = gN
E vE ,
exp i vE E0 0,
(2.48)
vE
for a constant gN and constant vector E0 . The diagonal partition function on the strip is
given by

1 c X vE2
2
= gN
q 8 ,
(2.49)
ZNN (q)
(q)
vE
where the summation is over the the entire dual lattice . Modular transforming to the
open string channel, it reads

1 c X 2 uE 2
c/2
2
q
,
(2.50)
ZNN (q) = (4) V0 ( )gN
(q)
uE
where V0 ( ) is the volume of the unit cell of the compactification lattice . We note that
the volume of the unit cell of the dual lattice is given by:
V0 ( ) = 1/V0 ( ).
(2.51)
uE2
The factor q /((q))2 is the character of the U (c) current algebra, and is a superposition
of the Virasoro characters with non-negative integer coefficients. Choosing
gN =
,
V0 ( )
(4)c/4
(2.52)
the state (2.48) satisfies Cardys condition and thus is a physical boundary state. The
constant gD actually represents the (generally fractional) ground-state degeneracy due to
the boundary [28]. From the diagonal partition function (2.50) in the open string channel,
we can read off the scaling dimensions of boundary operators occurring with the boundary
condition. Namely, the scaling dimensions of the boundary operators are given by
N = 2 uE 2 + (non-negative integer),
where uE is an element of the lattice .
(2.53)
549
The physical meaning of the boundary state |N(E0 )i turns out to be the Dirichlet
boundary condition on E namely E = E0 at the boundary. This can be checked by the
calculation of the partition function in the open string channel imposing this boundary
condition. This b.c. is equivalent to the Neumann boundary condition on the dual field,
E
d /dx|
0 = 0. We label the state with respect to the dual field for convenience in the
following section. We note that it is not clear whether the N boundary state is the only
solution of the Cardys condition even within the linear combinations of the bosonic
Ishibashi states (2.47).
A similar construction of the boundary state is possible, starting from the condition
(E
nL + E nR )|Bi = 0,
(2.54)
instead of (2.46). Again we are imposing invariance under the Heisenberg algebra,
Eq. (2.7). The bosonic Ishibashi state is given by

X

uE, 0E .
uE , 0E = exp +
(2.55)
aEnL
aEnR
n
A physical boundary state is given

X

D E0 = gD
exp i uE E0 uE, 0E ,
(2.56)
uE
where the summation is over the entire lattice and

p
gD = c/4 V0 ( ).
The diagonal partition function reads, in the closed string channel,

1 c X uE2
= gD 2
q 2 ,
ZDD (q)
(q)
(2.57)
(2.58)
uE
and in the open string channel

1 c X vE2/(2)
q
.
ZDD (q) =
(q)
(2.59)
vE
The scaling dimensions of the boundary operators are

vE2
+ (non-negative integer),
(2.60)
2
where vE is an element of the dual lattice . The physical meaning of this boundary state
E or equivalently the Dirichlet boundary
is the Neumann boundary condition for the field ,
E
E
E
condition for the dual field , i.e., = 0 at the boundary.
In the simplest case, c = 1,
D =
= {nR|n Z},
= {m/R|m Z},
(2.61)
where R is the compactification radius. In this case the degeneracies are given by:

gD = 1/4 R.
(2.62)
gN = 1 1/4 2R ,
For the multicomponent (c > 1) case, there is a simple generalization of the above construction of N and D boundary states. Since the generators of the Virasoro algebra (2.42)
550
are bilinears in the creation/annihilation operator n , a sufficient condition for the conformal invariance (2.16) is given by
nR )|Bi = 0,
(E
nL RE
(2.63)
where R is an orthogonal c c matrix. This corresponds to another sort of covariance under

the Heisenberg algebra, Eq. (2.8). This includes the N case (2.46) which corresponds to
R = 1 (identity matrix) and the D case (2.54) which amounts to R = 1. Other choices of
the orthogonal matrix gives different boundary states. For example, when R is a diagonal
matrix with diagonal elements +1 or 1, it represents a mixture of the D and N b.c.s.
When R is a rotation matrix, it gives a b.c. corresponding to the QBM under an external
magnetic field [29]. We postpone the discussion of such a problem to a later publication.
The zero modes of the boson are restricted by the condition (2.63) with n = 0, namely
(2 uE + vE) = R(2 uE + vE).
Its solution is given as a linear combination of basic solutions
X
n (E
un , vEn ),
(E
u, vE) =
(2.64)
(2.65)
u, vE), 2 uE + vE form a new lattice 2 e. Only

where n s are integer coefficients. For such (E
the oscillator vacua with these zero modes can contribute to the boundary state.
A possible boundary state is given as
X
(E
u, vE)ii,
(2.66)
|Ri = gR
(E
u,E
v)
where the sum is taken over (2.65) and

!
E
anL RE
anR |(E
u, vE)i.
|(E
u, vE)ii = exp
(2.67)
n=1
The diagonal partition function for this state ZRR turns out to be identical to that for ZNN ,
if the original lattice is replaced by e. The scaling dimensions of the boundary operators
and g-factors are also easily given by this replacement.
For the multicomponent free boson, we have thus constructed an infinite variety of b.c.s
by the generalization of D/N as in Eq. (2.63), for the infinite possible different choices
of R. The diagonal partition function for these b.c.s has a structure similar to that for D
or N b.c. and can be written by a non-negative linear combination of bosonic (Heisenberg)
characters q h /((q))c . We emphasize that we have just given some simple examples of
physical boundary states for the free boson theory; we have not exhausted all the possible
conformally invariant b.c.s.
In fact, more nontrivial nonbosonic boundary states are possible in the multicomponent free boson, at least in some cases. In Section 5 we will construct such highly nontrivial
boundary states and will demonstrate that they are indeed not bosonic; they cannot be understood by a generalization of D and N as described above.
551
2.3. Dirichlet and Neumann boundary conditions from fusion: c = 1 case

The problem of finding out all possible conformal invariant boundary conditions is still
an open one for all but minimal theories. Even the case of the free boson c = 1 is not
fully understood yet. In this section, we discuss this problem further, and demonstrate how
the twist field (of dimension 1/16) allows one to connect D and N boundary conditions
in some simple rational cases. This gives some justification for the fusion procedure to be
implemented in the following sections in the case c = 2. Some of the following results are
related to the more abstract approach in Ref. [30].
We thus consider a single compact free boson, corresponding to the one-dimensional
limit of the model reviewed in the previous section with the compactification lattice
{nR|n Z}. We restrict to the case R 2 = p0 /2 , p0 an integer (p0 = 1 is the SU(2)
symmetric point and p0 = 2 is the square of the Ising model). The torus partition function
of the periodic boson can be expressed in terms of the generalized characters [31,32]
(M)
(M)
= KM
=
K(M) = K
1 X (nM+)2 /2M
q
,
n=
(2.68)
as
Zper =
M1
X
(M) 2
K ,
(2.69)
=0
with M 2p0 . This decomposition reflects the symmetry of the periodic boson at this
particular radius under a chiral algebra [31] (traditionally denoted by Ap0 ) which is larger
than the Virasoro algebra, and is generated by, besides the stress energy tensor T , the
current j and a pair of vertex operators of integer spin p0 . The primary fields of the algebra
Ap0 are vertex operators V , = 0, . . . , M 1, with fusion coefficients
= +, mod N
N
(2.70)
while charge conjugation takes V VM .

For the same periodic boson, we now consider partition functions in the open string
channel. Two kinds of conformal boundary conditions are known beforehand: the Dirichlet
boundary conditions, where the value of the field on the boundary is fixed modulo the
compactification lattice, = 0 + 2nR, and the Neumann boundary conditions, where
the value of the dual field on the boundary is fixed modulo the dual compactification lattice,
= 0 + m/R. These boundary conditions preserve the Virasoro symmetry, but do not, in
general, preserve the chiral symmetry Ap0 . In fact, only the set of M Dirichlet boundary
, = 0, . . . , M 1 do so, with
conditions with 0 = 2R
ZD(/2R)D(/2R) = K .
(2.71)
These Dirichlet boundary conditions are presumably the only Ap0 invariant boundary
conditions. Introducing the Ishibashi states |ii associated with the module of this
algebra, the corresponding boundary states read
552
=
|i
M1
X
=0
M1
X e2i/M
S
q |ii =
|ii,
M 1/4
S0
=0
(2.72)
where the matrix of modular transformations is given by

1 2i/M
e
.
(2.73)
M 1/2
The general DD partition functions ZD(/2R)D(/2R) can naturally be obtained by fusion
from the partition function ZD(0)D(0) using the fusion algebra coefficients (2.70).
It is possible to write some of the Neumann partition functions in terms of the K s. For
instance
S =
M1
X
ZN(0)N(0) =
(M)
(2.74)
=0, even
Similarly,
ZN(0)N(/2R) =
M1
X
(M)
(2.75)
=1, odd
All the Neumann boundary conditions do however break the symmetry Ap0 . This is most
clearly seen by writing the DirichletNeumann partition function, which is independent of
the radius, and well known to be
1 X (2n1)2/16
q
.
(2.76)
ZND =
4 n
This partition function cannot be expressed in terms of generalized characters. The
corresponding closed string partition function cannot either, but of course, it is expressible
in terms of the bulk c = 1 Virasoro characters, as it should be since both D and N are
possible conformal boundary conditions for the free periodic boson:
X
1
2
(1)n q n .
(2.77)
ZND =
2 (q)
n
To relate D and N boundary conditions by fusion, it is necessary to relax the constraint
on invariance under Ap0 . Of course, what one should really do, since one is only interested
in general in imposing conformal invariance on the boundary conditions, not any higher
symmetry, is to simply use the fusion algebra of the c = 1 theory for arbitrary radius; this
however poses technical problems in particular, continuous modular transformations
which are, for the moment, insurmountable. To proceed, the empirical strategy we use is to
consider variants of the theory, with different, finite, extended algebras, for which fusion
can be implemented and, hopefully, meaningful boundary conditions for our original free
boson found. The natural candidate that comes to mind is the Z2 orbifold, with partition
function (recall R 2 = p0 /2; M = 2p0 )
Zorb =

2
X (8) 2 1 X 0 2

(2p0 ) 2

0 ) 2
pn
n n2
K
+ 2 1 K (2p

+2
+
q
+ (1) q
K
2 p0
2 n
0 1
pX
=1
=1,3
553
X
2
1

p 0 n2
n n2

+
q
(1) q .
2
(2.78)
The operator content of this theory is quite different from that of the periodic free boson.
The most obvious feature is that, due to the identification, the multiplicity two of
vertex operators has now disappeared: we get instead operators with multiplicity one, the
2
first p0 1 terms, which correspond to operators which we call , of dimension 4p
0 . The
(i)
next term in the partition function stands for two degenerate operators p0 of dimension
p0 /4 (the two energy operators in the Ising square case), the next for the twice degenerate
twist (i) and excited twist (i) operators of dimensions 1/16, 9/16 (the spins and the
tensor product of spin and energy for the Ising square case). Finally the last two terms are
respectively the characters of the identity and the current (which we denote ), which is
primary in the orbifold algebra. We will denote the corresponding characters (which appear
as moduli squares in the foregoing formula) KI and K .
To proceed, observe that, presumably, the boundary states of the periodic boson and the
orbifold boson are not mutually compatible in general: for instance, a Dirichlet boundary
state for the orbifold |Do (0 )i should be a linear combination of Dirichlet boundary
states for the nonorbifold boson at values 0 , and the normalization will make |Do (0 )i
incompatible with |D(0 )i. There are thus various families of boundary states we can
envision constructing. To start and get quickly to the point we consider the particular
cases of |D(0 = 0)i, |N( 0 = 0)i for the periodic boson. The corresponding partition
functions, in the open string channel, can be fully written in terms of the generalized
characters of the orbifold algebra:
ZD(0)D(0) = KI + K ,
(2p0 )
(2p 0 )
(2p 0 )
+ K4
+ + 12 Kp0
,
ZN(0)N(0) = KI + K + 2 K2
ZDN = K1(8) + K3(8)
(2.79)
(p0 even). Formulas (2.79) suggest that it is now possible to connect D and N by fusion,
which we now demonstrate.
To do so, we need the fusion algebra of the orbifold theory: it can be found for instance
in Ref. [32] with the following results (p0 even):
(1) = (1) ,
(1)
(1) (1) = I + p0 +
(1) (2) =
even
odd
(1)
(1) even =
(1)
odd =
p(i)0
(i)
(i)
p0 p0
(1)
(2)
(1)
= I,
554

(2)
p(1)
0 p 0 = ,
= + + ,
= 1 + + 2 ,
(1)
(2)
2p0 = 2 + p0 + p0 ,
(2.80)
where the fusion of operators holds only for 6= , 2p0 , and is defined modulo p0 .
We now start with ZDD and fuse with the (1) operator. From (1) I = (1) and
(1)
= (1) we get the partition function ZND , since the two terms in the last equation
of (2.79) are the characters of the twist and excited twist operators respectively.
For p0 even, we fuse again with (1) . We need the basic fusion rule (1) (1) =
P
(1) (1) = +
I + p(1)
0 +
even , and also the rule deduced from associativity
P
(2)
p0 + even ; the fusion gives therefore ZNN . To summarize: by fusion with (1) we
thus found ZDD ZND ZNN .
Just to illustrate the difference with p0 odd, we remark that in the latter case, while the
first fusion is the same, the next one has to be with the other twist operator, (2) . Using the
equivalent of (2.80) in this case
X
X
,
(2) (1) = +
(2.81)
(2) (1) = I +
even
even
we recover ZNN indeed. This time therefore we have ZDD (1) ZND (2) ZNN .
To express the boundary states, it is better now to carry out a more complete analysis
within the boundary conditions of the orbifold model. We thus consider the proper orbifold
Dirichlet boundary states, which are generally defined by (0 6= 0, R)

Do (0 ) = 1 |D(0 )i + |D(0 )i .
(2.82)
2
The corresponding Dirichlet partition functions are
6= , 6= 2p0 ,
K + K+ ,
(2.83)
ZDo (/2R)Do(/2R) = KI + K + K20 , = ,
(2p )
K2 + 2 12 Kp0 , = 2p0 .
The only orbifold Neumann boundary states compatible with the orbifold algebra are the
. We will discuss them later.

end point ones: 0 = 0, 2R
We now need the modular transformation matrix, which reads:
1
S = p
8p0
1
1
2
p
p0
p
p0
1
1
1
2
p
pp 0
p0
1
1
1
2(1)
p
0
(1)i j pp0
00
(1)i j p0
2
2
2(1)
4 cos
2p 0
0
0
p
0
pp
0
pp
0
(1)ij p0
0
p
0
0
i j p2p0
i 00 j 0 2p0
0
pp
0
p p
00
(1)ij p0
,
p
i 0 j 00p 2p0
i 00 j 00 2p0
(2.84)
555
00
where the conventions are that the matrix maps (I, , p(i)0 , , (i ) , (i ) ) onto the same
quantities but with i, j exchanged.
After modular transformations
"
0)
1
1
e(2p
2 I + 2 + 4 (1) K
ZDo (/2R)Do(/2R) = p
p0
0
2
2p
#
0 1
pX
e(2p0 )
+4
cos
cos
K
.
(2.85)
2p0
2p0
=1
Expressions of the various boundary states in terms of Ishibashi states of the orbifold
algebra follows:
(
(1) (2)

1
Do (/2R) =
2 |I i + 2 | i + 2 (1) p0 + p0
0
1/4
(2p )
)
0 1
pX
cos
| i .
(2.86)
+2
2p0
=1
Setting = 0 gives the periodic Dirichlet boundary state, up to a factor of 2:

)
(
p 0 1
1
1 (1) (2) X
| i .
(2.87)
|D(0)i =
|I i + | i + p0 + p0 + 2
(2p0 )1/4
2
=1
Modular transformation gives the other partition functions in the closed string channel
p

1 p 0 e
e ,
p KI p0 K
ZD(0)N(0) = p
2p0
0)
2 e
e(2p
e + 2 1 K
KI + K
.
(2.88)
ZN(0)N(0) = p
0
2
p
2p0
From this, we obtain the Neumann boundary state
p
p

p
p0 (1) (2)
1
0
0
p 0 p 0
p |I i p | i +
.
|N(0)i =
(2p0 )1/4
2
(2.89)
Observe that, because of the particular radius we have chosen, the Ishibashi states which
appear in |Ni all appear in |Di also: this is a necessary condition for N to be obtained from
D by fusion, as follows from Eq. 2.27 and of course would not hold in general.
So far, the boundary states we considered are all in the periodic sector the Hilbert
space in the closed string channel is the one of the periodic boson. As discussed in Ref. [33]
(see also Ref. [30]), there are also eight boundary states which involve, in addition, the
twisted sector. These states are

Do (0 ) = 21/2 D(0 ) 21/4 D(0 )T ,
(2.90)
where 0 is at the fixed points of the orbifold, 0 = 0, R, and similar Neumann like states:

No (0 ) = 21/2 N(0 ) 21/4N(0 )T
(2.91)
556
with 0 = 0, /2R.
The partition functions found in Ref. [33] can be written as
ZDo (0 ),Do (0 ) = KI ,
ZDo (0 ),Do (0 ) = K ,
(2p 0 )
ZDo (0 ),Do (R0 ) 0 = 12 Kp0
(2.92)
Similarly for the Neumann sector one has

(2p 0 )
ZNo (0 ),No (0 ) = K2
(2p 0 )
ZNo (0 ),No (0 ) = K2
(2p 0 )
ZNo ( 0 ),No (R0 ) 0 = K1
(2p 0 )
+ K4
(2p 0 )
+ K4
(2p 0 )
+ K3
(2p 0 )
+ + 12 Kp0
(2p 0 )
+ + 12 Kp0
(2p 0 )
+ + Kp0 1 .
+ KI ,
+ K ,
(2.93)
Finally, combining Neumann and Dirichlet gives

ZDo (0 ),No ( 0 ) 0 = K1(8) , resp. K3(8) ,
(2.94)
where the choice depends on the signs , 0 and the values of 0 , 0 .

Corresponding boundary states can of course be written, giving rise to rather bulky
expressions we will avoid here. Rather, we would like to stress that the orbifold boundary
conditions are all connected by fusion indeed. Starting with the first partition function
in (2.92), we get to the second one by fusion with the field , thanks to = I , so
Do (0 ) follows from Do (0 ) by fusion with . Similarly, starting again with the first
(i)
partition function in (2.92) and fusing with p0 gives the third partition function using that
(j )
p(i)0 = p0 , so |Do (0 )1 i follows from |Do (r 0 )2 i by fusion with p(i)0 .

The Neumann and Dirichlet sectors are still related through (i) , or, also (i) , using
(i)
(i) = (i) . Finally, fusion with and p0 gives the remaining partition functions
in the Neumann sector. As for the continuous orbifold Dirichlet and Neumann boundary
conditions, they are obtained by fusion with the operators .
One of the lessons illustrated here, is that to obtain all possible conformal boundary
conditions for the free boson theory, one needs to consider its Z2 orbifold as well. It is not
clear in general what good variant of the bulk theory is needed to explore all possible
conformal boundary conditions for a given model. In the case c = 2 for example, we shall
see that fusion with a field of dimension 1/8 allows one to go from D to N. This is a
natural generalization of the c = 1 case, the 1/8 field being now a double twist field for
the bosons 1,2 . We will also see however that fusion with a field of dimension 2/5 is also
necessary to explore the possible boundary states. The origin of this field in the two boson
language is still mysterious. Is it some other sort of twist field?
The discovery of nontrivial boundary conditions should also lead to new bulk theories.
In the c = 1 case for instance, consideration of the possible D and N boundary conditions
should lead one to the discovery of the Z2 orbifold, were this theory not already known.
This is also related to the existence of a c = 1 theory the square of the Ising model
which is not a periodic boson. In the c = 2 case, the embedding we will use to
describe partition functions with boundaries leads similarly to a bulk theory the product
557
of Ising, tricritical Ising and Potts which does not seem to be any simple sort of two
boson orbifold. Understanding its meaning would be crucial to generalize our work to
other values of the triangular lattice parameter, a.
3. Quantum brownian motion and conformal field theory

We first discuss the one-dimensional case. The study of brownian motion of a classical
particle described by the simple Langevin equation
M
dQ V
d 2Q
+
= (t)
+
2
dt
dt
Q
(3.1)
is of course one of the basic topic in nonequilibrium statistical mechanics. In (3.1), is a

phenomenological friction coefficient, V an external potential, M the particle mass, and
is a random force that obeys

(t) = 0,

(3.2)
(t)(t 0 ) = 2T (t t 0 ).
More recently, there has been a great deal of interest in trying to describe the quantum
behavior of systems for which the classical motion would be described by the Langevin
equation (3.1). The most tractable approach has been to couple the particle (which might
actually represent something quite different, like the phase in a Josephson junction) to
a bath (an environment) with an infinite number of degrees of freedom, which provides
both the friction and the fluctuating force. The simplest example of that approach is the
CaldeiraLeggett model [4] where the coordinate Q is coupled linearly to an infinite set of
harmonic oscillators, with the Hamiltonian

1 X pk2
Qk 2
P2
2
+ V (Q) +
+ mk k qk
.
(3.3)
H=
2M
2
mk
mk k2
k
Here Q and P are the co-ordinate and momentum of the particle; V (Q) is an arbitrary
potential, with V (Q + a) = V (Q). The exact distribution of coupling constants k , masses
mk and frequencies k is important for the properties of the particle only through the
weighted density of states
J ()
X 2k
( k ),
2
mk k
> 0.
(3.4)
In the limit where the number of oscillators is infinite, J () becomes a continuous

function. In the following we will restrict to the so called ohmic case J () = ; with
this choice, it can be shown that the Hamiltonian (3.3) when treated classically yields,
after elimination of the bath degrees of freedom, the Langevin equation (3.1).
By a canonical transformation on the oscillator bath:
qk
k qk
,
mk k2
pk
mk 2
pk ,
k k
(3.5)
558
we may rewrite the Hamiltonian in the form:

1 X pk2
P2
2
2
+ V (Q) +
+m
ek e
k [qk Q] ,
H=
2M
2
m
ek
(3.6)
where:
m
ek 2k /mk k4 ,
e
k k ,
k 2k /mk k2 = m
ek e
k2
(3.7)
are the masses, frequencies and couplings of the transformed oscillators. Henceforth, we
drop the tilde notation on these quantities. The Hamiltonian of Eq. (3.6) is invariant under
the translation:
Q Q + a,
qk qk + a.
(3.8)
As the particle moves in the periodic potential it drags a cloud of other fictitious particles
with it which do not feel the periodic potential V but are bound to the real particle
by a quadratic interaction. Due to the translation symmetry of Eq. (3.8), there is a total
momentum operator, PT , which is conserved modulo 2/a:
X
pk .
(3.9)
PT P +
Explicitly, a basis of eigenstates of H may be chosen such that any eigenfunction gets
multiplied by the phase eiPc a under this translation. Without loss of generality, the
eigenvalue (crystal momentum), Pc , may be chosen to lie in the first Brillouin zone, |Pc | <
/a.
After the canonical transformation of Eq. (3.5) the weighted density of states becomes:
X
k k ( k ), > 0.
(3.10)
J ()
2
k
The properties of the original particle are unaffected by a rescaling of k (independently

for each value of k) provided that k is scaled by the inverse factor such that k k is held
fixed.
We note that an alternative way of writing the Hamiltonian is in terms of the total
momentum PT and transformed oscillator co-ordinates,
qk0 qk Q,
(3.11)
which define a transformed canonical set of co-ordinates. The Hamiltonian in these coordinates becomes:
P

1 X pk2
(PT pk )2
2 02
+ V (Q) +
+ mk k q k .
(3.12)
H=
2M
2
mk
A convenient choice for the oscillator bath is the Fourier modes of a one-dimensional
free massless boson quantum field. In order to begin with a discrete set of oscillators, we
may define the field theory with vanishing boundary conditions in a box of size 2l, so that:
k = k,
k = n/2l,
n = 1, 2, . . . .
(3.13)
Here we have set an arbitrary velocity of light to 1. Strictly speaking an ultraviolet cut
off must be applied, k < , but the universal critical behavior will be independent of the
cut-off.
559
Note that there is a zero-frequency oscillator in the limit l . The field couples to
the particle at x = 0 and the boundaries of the box are at x = l. The oscillator coordinate
and momenta can be expressed in terms of the boson creation and annihilation operators
as:
p

(3.14)
pk = ak + ak / 2k.
qk = i k/2 ak ak ,
We may then define the field and conjugate momentum field on a fictitious one-dimensional
space, x, completely unrelated to the physical space in which the particle moves, by:
r
r
1X
1X
pk sin k(x + l),
(x)
qk sin k(x + l).
(3.15)
(x)
l
l
k
We note that (x) corresponds to a noncompact boson field, so no term linear in x is

allowed in its mode expansion.
We see that, with this choice of oscillator bath, in order to obtain ohmic dissipation, we
must choose, k = 2/ l. From Eq. (3.7) this implies that the oscillator masses are given
by:
mk = k /k2 = 2/ lk 2 .
(3.16)
The fictitious particles become infinitely heavy at zero frequency!

In this case the Hamiltonian of Eq. (3.6) can be written:
1
P2
+ V (Q) +
H=
2M
2

Zl
dx
l
d
dx
2
+ +
2 Q(x)
2

.
(3.17)
It is convenient to separate (x) into its even and odd components since the odd part
decouples from the particle:

(3.18)
e,o (x) (x) (x) / 2.
The even field obeys a Neumann b.c., de /dx = 0 at x = 0. Thus we may equivalently
consider a field defined on the positive axis, 0 6 x < l, with a Neumann b.c. at x = 0 and
a Dirichlet b.c. at x = l coupled to the particle. This Hamiltonian is equivalent to Eq. (3.6)
with:
e
k = k,
k = 2/ l,
k = (n + 1/2)/ l, n = 0, 1, 2, . . . .
(3.19)
We might expect the translational symmetry to lead to the usual band structure in which
the discrete set of eigenstates of given crystal momentum depends in some smooth way
on Pc . However, when there is an oscillator, q0 , of infinite mass a dramatic simplification
occurs. This follows from observing that any eigenfunction of crystal momentum Pc , can
be written in the form:
= eiPc q0 ,
(3.20)
where (Q, qk ) is invariant under the translation of Eq. (3.8). Since p0 does not appear in
H , it follows that the periodic function, , obeys exactly the same Schroedinger equation
as the full wave-function, , independent of Pc . This implies that all eigenstates come in
560
degenerate sets, where the degenerate members are labelled by Pc . Effectively we can give
the infinite mass particle, q0 , all the momentum at no cost in energy. This degeneracy is
strictly infinite if Q is allowed to range over the entire real line. Alternatively, if we restrict
Q to an interval of size N a then we expect the eigenvalues to form nearly degenerate
sets of size approximately N , the number of allowed momenta in the Brillouin zone. In
a localized phase we can identify this degeneracy with the N minima of V at which the
particle can become localized. However we emphasize that this degeneracy is an exact
property of H (at infinite N ) and does not depend on which phase the system is in. We
have dwelt on this point here because we will find that, in the limit where the oscillator
frequencies form a dense set, the groundstate degeneracy can take the more general form
gN where g is a universal (generally noninteger) number of O(1) which does depends on
which phase the system is in. This rather unusual behavior of the model is related to its
essentially unphysical nature as a real model of QBM, mentioned in the Introduction. We
elaborate on this point further in Subsection 3.3.
In the field theory representation, the conserved total momentum operator and transformed oscillator co-ordinates are:
e0 (x) = e (x) + 2 Q(x).

(3.21)
PT = P + e (0),
The transformed Hamiltonian (keeping only the even part) is:
H=
(PT
e (0))2
1
+ V (Q) +
2M
2

Zl
dx
de
dx
2

+ e0 2 .
(3.22)
Much of the literature on QBM proceeds by integrating out the field, e (x) to obtain
a noninstantaneous effective action for the particle. However, we introduce a different
approach here in keeping with previous analyses of quantum impurity problems using
boundary CFT methods. In our approach we rather absorb the particle co-ordinate and
momentum into a b.c. on the field, in the effective low energy Hamiltonian. This has the
advantage that correlation functions of the field are immediately accessible. For the QBM
problem this allows us to study the back-reaction of the particle on the oscillator bath.
We note that this dichotomy of approaches also exists in other types of quantum impurity
problems. There are two limits in which the QBM Hamiltonian reduces to a boundary
sine-Gordon model: the strong and weak corrugation limits. We consider each in turn.
3.1. Strong corrugation limit
In the strong corrugation limit the potential is large compared to the dissipation.
More precisely, we require the energy bands produced by the potential V , ignoring the
dissipation, to have large band gaps compared to the energy scale of the dissipation, /M.
In this limit we expect that band-mixing effects due to the dissipation will be small and the
wave-functions for the particle-bath system will be linear combinations of wave-functions
in a particular band, n (Q; Pc ). Here Pc is the crystal momentum and the wave-functions,
n are periodic:
n (Q + a; Pc ) = n (Q; Pc ).
(3.23)
561
The solutions of the nondissipative Schroedinger equation are eiPc Q (Q; Pc ) and the
periodic functions obey the equation:

d
d2
1
2
+
P
2iP
(3.24)
+
V
(Q)
(Q; Pc ) = E(Q; Pc ).
c
c
2M
dQ2
dQ
Projecting the Hamiltonian of Eq. (3.22) into a fixed energy band reduces it to:
Hn n Pc
Zl
e (0) +
dx
1
2

de
dx
2

+ e0 2 .
(3.25)
Here n (Pc ) is the dispersion relation of the nth energy band for the nondissipative
problem. Note that e obeys a N b.c. at x = 0 and a D b.c. at x = l. The degeneracy
discussed above corresponds to shifting e (x) by Pc / . However, this is inconsistent

with the vanishing b.c. at x = l so the degeneracy is lifted by effects of O(1/ l) as expected.
Upon expanding n in harmonics:
n (Pc ) = An + Bn cos Pc a +
(3.26)
and ignoring the (less relevant) higher harmonics, we obtain the boundary sine-Gordon
model:

H = B cos aPc ae (0) +
Zl
1
dx
2

de
dx
2
+ e0 2

.
(3.27)
This result can be confirmed by considering a (single band) tight-binding model [2].
Introducing operators cj which annihilate a particle at site j , the position operator is
represented as:
X
j cj cj .
(3.28)
Q=a
j
The transformation to the total momentum operator then corresponds to a unitary

transformation by the operator:
U ei
2 Q(0)
Explicitly:
p

U (x) + 2 Q(x) U = (x).
(3.29)
(3.30)
The part of the Hamiltonian containing only the particle can be diagonalized in momentum
space. Defining:
1 X
cj eipj ,
(3.31)
cp
N
j
the particle part of the Hamiltonian is:

X
(p)cp cp .
H0 =
p
(3.32)
562
We find:
U cp U = cp+2 (0) ,
(3.33)
and hence the Hamiltonian is transformed into Eq. (3.25), using 2 (0) = e (0). In the
case of nearest neighbor hopping the dispersion relation is:
(P ) = 2t cos P a.
(3.34)
The above derivation generalizes this result slightly to the case of several well-separated
bands with arbitrary dispersion relation. Of course, for sufficiently weak dissipation and
sufficiently low temperature, we will only be concerned with the lowest band.
It is convenient to define the parameter:
1
R ,
a
(3.35)
so that the Hamiltonian is invariant under the translation:

e e + 2R.
(3.36)
However, as mentioned above, the boson e is not compact so we cannot identify the field
under this translation. We return to this point below.
The boundary interaction has scaling dimension
=
1
a 2
=
.
2
2R 2
(3.37)
This can be verified, for example, by using the N b.c. to replace 00 (0) by 2L (0) and
using the standard result that the scaling dimension of cos L is 2 /8 . The boundary
interaction is relevant for a 2 < 2 . For a 2 > 2 , we may think of the hopping term
as being irrelevant corresponding to a flow to the localized fixed point. For a 2 < 2
we expect a flow to the diffusing fixed point. Note that either increasing the distance
over which the particle must hop or increasing the coupling to the heat bath promotes
localization.
3.2. Weak corrugation limit
The momentum operator PT and co-ordinate Q, can be absorbed into the field, e , by
transforming to a dual oscillator bath, a transformation that was introduced by Fisher and
Zwerger from a different viewpoint [3]. This is done by first diagonalizing the oscillator
part of the Hamiltonian, whose even part is:
[e (0)]2 1
+
H0
2M
2

Zl
dx
de
dx
2

+ e0 2 .
(3.38)
This quadratic Hamiltonian can be diagonalized by solving the classical EulerLagrange

equations:

d2
2
(3.39)
(x) = 2 + 2 (x) (x).
dx
M
563
This is just the Schroedinger equation with a -function potential. The odd wave-functions
are unaffected by the potential and the even ones can be expressed in terms of a phase shift,
(k):

(3.40)
k (x) cos k|x| + (k) ,
with frequency (k) = |k|. (Here k runs over the wave-vectors of the even modes, given in
Eq. (3.19).) Substituting into Eq. (3.39) we obtain the phase shifts:
cot (k) =
kM
.
(3.41)
For small k, this becomes:

kM
+ O k2 .
(k) +
2
The allowed wave-vectors are given by:

kl + (k) = (n + 1/2),
n = 1, 2, 3, . . . .
(3.42)
(3.43)
For large l (l M/) and small k (k /M) we obtain:

k n/ l,
n = 1, 2, 3, . . . .
(3.44)
The mode expansion for e (x) now takes the form:

r

2X
pk cos k|x| + (k) ,
e (x) =
l
(3.45)
and similarly for e (x). (The pk s are the same operators, up to a minus sign, that occur
in Eq. (3.15) for wave-vectors corresponding to odd values of n in Eq. (3.13).) Noting that,
for small k,

cos k|x| + sin k(|x| + M/) ,
(3.46)
we see that the effect of the e (0)2 term in H0 at low energies, is to impose, approximately
a Dirichlet b.c., e = 0 on the even field. (More accurately, the low energy field vanishes
at x = M/ but, in the low energy effective theory (x) does not vary on such short
wavelengths so we ignore this shift.) It is now convenient to introduce a transformed even
field, 0 (x) obeying the D. b.c. at x = 0:
r
2X
0
sin kx pk .
(3.47)
(x)
l
k
The even part of H0 may be written in terms of 0 as simply:

1
H0 =
2

Zl
dx
d 0
dx
2

+ .
02
(3.48)
We now must consider the term in the Hamiltonian H of Eq. (3.22) that couples the field
to the particle:
(3.49)
Hint /M e (0)PT .
564
This may be expressed in terms of the pk operators of Eq. (3.45) using the explicit
expression for the phase shift in Eq. (3.41) as:
r
2X
kpk
p
.
(3.50)
e (0) = M
2
l
+ (kM)2
k
We see that, at low energies this is given by:

e (0)
M d 0
,
dx
(3.51)
but we do not make this approximation here, proceeding to a more exact analysis. The
entire Hamiltonian of Eq. (3.22) can now be written:
s

2

PT2
2
1X
2
P
+ V (Q).
(3.52)
+
q
kpk +
H
T
k +
2
2
2
l[ + (kM) ]
2l
k
In order to complete the square we have used a crucial identity:

"
#
X

1
1
1
2 1
coth(l/M) =
+ O e2l/M
+
=
2
2
2
l 2
+ (nM/ l)
M
M
(3.53)
n=1
and dropped the exponentially small term. [Note that we are being a bit cavalier and
using the small k result, k n/ l for all n. It can be shown that the corrections to this
approximation only affect the exponentially small term in Eq. (3.53).] It is now convenient
to Fourier transform to position space, using:
r
2X
d 0
=
kpk cos kx.
(3.54)
dx
l
k
To this end it is convenient to introduce the function:

#
"
X
2 1
cos kx
p
+
.
f (x)
l
2
2 + (kM)2
k
(3.55)
Here the sum is over k = n/ l with n = 1, 2, 3, . . . f (x) vanishes exponentially at large x.

Up to corrections which are exponentially small in l/M, we may write:

2
x
K0
.
(3.56)
f (x) =
M
M
Here K0 is a modified Bessel function. We may thus rewrite H as:
1
H=
2

Zl
dx
d 0
f (x)PT
dx
2
+
02

+ V (Q).
(3.57)
It is now possible to make a further transformation that absorbs PT into 0 :

00
Zl
(x) = (x) +
x
dx 0 f (x 0 )PT .
(3.58)
565
We may now also absorb Q by going over to the dual field. This is defined, as usual by:
d 00
= 00 .
(3.59)
dx
These equations only determine 00 (x) up to a constant term. This may be fixed by
e00 (x)
imposing the canonical commutation relations and EulerLagrange equations on
00
and (x). The finite momentum part of the dual field is:
r
2 X qk cos kx
0
.
(3.60)
(x) =
l
k
d 00
e00 ,
=
dx
The zero mode term in 00 (x), which we write 000 , must be chosen to commute with all
e00 (x) but have a nonzero commutator with the zero mode of
finite momentum modes of
00
e
(x):
e000
Zl
PT
f (x)
PT
dx
= .
l
l
(3.61)
Using the commutator:

0
x
0
(x y) ,
(x), (y) = i
l
which can be checked from the mode expansion of these fields, we find
000 =
Zl
dy f (y) 0 (y) Q.
(3.62)
(3.63)
Note that 00 (x) obeys N b.c.s at x = 0 and l. We see that Q has become (part of) the
zero-mode of the dual field. The Hamiltonian can be written:
#
" Zl
00 2
Zl
d
1
00 2
00
e
dx +
(3.64)
+ V dx f (x) (x) .
H
2
dx
0
Now consider integrating out the high wave-vector components of 00 (x) to obtain a low
energy effective Hamiltonian. Since f (x) vanishes exponentially on the length scale M/,
it follows that once we have reduced the wave-vector cut off to a value much less than
/M we may approximate 00 (x) by its value at x = 0 inside the integral in the last term
of Eq. (3.64), leading to the simplified expression:
1
H
2
Zl

00 2

d
00 (0)
e00 2 +
dx
.
+V
dx
(3.65)
The degeneracy mentioned above, corresponds to shifting 00 (x), or equivalently Q, by

a constant na where a is the lattice spacing. This degeneracy is lifted at finite l as expected.
Now let us consider a particular choice for V :
V (Q) = V0 cos(2Q/a).
(3.66)
566
Our Hamiltonian then becomes the standard boundary sine-Gordon model:

1
H=
2
Zl

00 2

d
00 2
e
dx +
+ V0 cos 2R 00 (0) ,
dx
(3.67)
where we have again introduced the parameter R defined in Eq. (3.35). Recall that we
obtained the boundary sine-Gordon model only after integrating out high frequency modes,
reducing the cut-off to /M. This process will, in general, introduce renormalization of the
interactions. In order to be able to ignore this renormalization, the original dimensionless
coupling constant must be small. We may estimate how small it needs to be by considering
the renormalization of the boundary interaction in Eq. (3.67). This boundary interaction
has a scaling dimension of
=
2
= 2R 2 .
a 2
(3.68)
Thus, upon reducing the cut-off from its bare value, to /M, the effective coupling is
renormalized to:
2

V0
1/a
V0
.
(3.69)
/M
Requiring the renormalized dimensionless coupling constant to be small after reducing the
cut off to /M gives the condition:
2
V0 (/M)1/a .
(3.70)
Thus, this boundary sine-Gordon model is only obtained in the weak corrugation limit. The
dimensionless parameter, a = 1/R may take any value, however.

We note that the weak corrugation formulation of the problem is actually equivalent
at l , under a dual transformation, to a tight binding model interacting with a heat
bath with a Lorentzian weighted density of states, J (). This can be seen by starting from
Eq. (3.52) and making the duality transformation:
pk qk /k,
qk kpk ,
PT Q,
Q PT .
(3.71)
The dual Hamiltonian is:
s
!2
"
#
1X
2
Q2
+ V (PT ) +
Q + k 2 pk2 .
qk +
H
2l
2
l[2 + (kM)2 ]
(3.72)
The last term has the same form as our original heat bath term in Eq. (3.3) with
s
1
2
mk = 2 ,
k =
k = k,
k
l[2 + (kM)2 ]
and hence a weighted density of states:
X
2k
( k).
J () = /2
2
l[ + (kM)2 ]
k
(3.73)
(3.74)
567
Taking the limit l this gives:
.
(3.75)
J () = 2
+ 2 M 2
This is ohmic below a cut off scale /M:
(3.76)
J () .
Note that the dissipation strength parameter, has been inverted. We may also drop the
first term in Eq. (3.72) in the limit l so that the particle part of the Hamiltonian
is diagonal in momentum space. This is then equivalent to a tight binding model with a
dispersion relation:
(PT ) V (PT ).
(3.77)
This gives another way of understanding why a boundary sine-Gordon model is obtained
in both the weak and strong corrugation cases. We note however, that the PT2 /2l term,
e00 (x) a zero mode, plays quite an important role in the following subsection.
which gave
3.3. Groundstate degeneracy and boson compactification
An interesting quantity in quantum impurity problems is the zero temperature impurity
entropy or its exponential, the groundstate degeneracy, g. More generally, we may define
an impurity free energy by subtracting off the bulk free energy (the term proportional
to l), then taking the length of the bulk system to infinity. The zero temperature impurity
entropy is defined this way, with the T 0 limit taken after the infinite length limit. Note
that if we take T 0 first, before taking l , the degeneracy is necessarily integer
valued since the spectrum of the finite size Hamiltonian is discrete. However, taking the
limits in the opposite order, we are effectively dealing with a continuous spectrum and it
is entirely possible to obtain a noninteger (even irrational) groundstate degeneracy g,
independent of system size. g has been argued to be universal and to always decrease
under renormalization group flow between boundary fixed points [28]. Thus it plays a role
in boundary critical phenomena analogous to that of the conformal anomaly parameter, c,
in two dimensional bulk critical phenomena. A new feature occurs when we consider g
in QBM; it turns out to be proportional to the length of the interval on which the particle
is allowed to move. We emphasize that we are taking the length, l, of the fictitious line
interval used to define the oscillator bath to . Thus the oscillator spectrum becomes
continuous. On the other hand, we are considering a long but finite physical line interval,
of length N a with N 1, on which the particle moves.
To orient ourselves let us first consider the partition function in the case = 0 where
the particle is decoupled from the heat bath and the periodic potential V is set to zero.
We only consider the even oscillator modes, with energies (n + 1/2)/ l, n = 0, 1, 2, . . . .
(Recall that e obeys a N b.c. at x = 0 and a D b.c. at x = l so that the allowed momenta
are proportional to the half-integers.) Thus the oscillator part of the partition function is:
Zosc =
Y

1
.
1 e(n+1/2)/ lT
n=0
(3.78)
568
Taking the limit lT 1 we obtain:

elT /6
Zosc ,
2
corresponding to an oscillator groundstate degeneracy of
gosc = 1/ 2.
(3.79)
(3.80)
This must be multiplied by the partition function of the particle. To make this well-defined,
we place the particle in a box of size N a, with periodic b.c.s. In this case the allowed
momenta are P = 2m/N a giving the full partition function:

lT /6
X
2m 2 1
e
exp
.
(3.81)
Z=
Na
2MT
2
m=
If we now take T 0, holding N fixed, the particle has a unique groundstate with P = 0
so it makes no contribution to g. We note, however, that in the opposite limit, (N a)2 MT
1, the partition function is proportional to N since the particle may be treated
classically:
r
Z
aN MT lT /6
dP P 2 /2MT
elT /6
e
e
=
.
(3.82)
Z aN
2
2
Remarkably, this factor of N will persist in the partition function as we take T 0 at

fixed N , once we couple the particle to the oscillators. If we now include the periodic
potential V , the partition function behaves the same way at temperatures small compared
to the bandwidth. This follows since we are only concerned with low momentum states in
the lowest energy band, whose dispersion may be approximated by:
P2
,
(3.83)
2M
for some effective mass, M .
Let us now couple the particle to the oscillator bath, > 0, but, for the moment ignore the
potential V . Thus we are considering a freely diffusing particle. From the mode expansion
of the dual field, 00 (x), which obeys N b.c.s at x = 0, l with the zero mode of Eq. (3.61),
or more directly from Eq. (3.52), we see that the spectrum is given by:
1 (P )
E=
X m
PT2
+
nm .
2l
l
(3.84)
m=1
Here the nm s are the occupation numbers of the finite momentum oscillator modes. In this
case the oscillator partition function is given by:
Zosc =
Y

m=1
1 em/ lT
1
elT /6

.
2lT
(3.85)
Thus the entire partition function, in the limit of large l, for the freely diffusing particle is
given by:

elT /6 X
2m 2 1
Zdiff =
exp
.
Na
2lT
2lT m=
569
(3.86)
Now taking the limit l first before taking T 0, we may replace the sum over m by
a classical integral over PT :
r
Z
elT /6
dP P 2 /2lT aN lT /6
aN
e
e
=
.
(3.87)
Zdiff =
2
2
2lT
We have thus obtained a groundstate degeneracy:

r
aN
.
(3.88)
gdiff =
2
Of course, the precise value of g depends on precisely how we have defined the oscillator
bath, in particular, the boundary condition at x = l on the original field (x). However,
ratios of g at different fixed points are expected to be independent of these choices,
depending on the oscillators only through the parameter . This is related to the fact that
the degeneracy may be regarded as a product of factors for each of the two boundaries of
the system. Comparing this calculation to the case discussed in the previous paragraph we
see that the degree of freedom associated with PT has become infinitely massive in the
limit l so that the associated partition function remains in the classical regime all the
way down to T = 0, yielding the factor of N .
Now let us consider the effect of the periodic potential, beginning with the weak
corrugation Hamiltonian of Eq. (3.67). This has a scaling dimension of 2/a 2 . If the
potential is irrelevant, a 2 < 2 , we expect that g remains unchanged. On the other hand,
if it is relevant, a 2 > 2 , we expect a RG flow to a different fixed point. The nature of this
fixed point may be deduced by assuming that V0 flows to , pinning 00 (0) at its minima.
On the other hand, 00 still obeys N b.c.s at x = l. Thus its mode expansion takes the form:
r

2X
00
sin (n + 1/2)x/ l pk .
(3.89)
(x) = am +
l
n=0
The oscillator part of Z is the same as in Eqs. (3.78), (3.79) for the case where the bath is
decoupled from the particle, and the sum over m simply gives a factor of N , the number
of minima of the potential where the particle can get localized. Thus we obtain, in the
localized phase:
N
(3.90)
gloc = .
2
Despite all the transformations and the RG flow that went into this result the answer is
intuitively obvious. The degeneracy is simply N times that of the decoupled bath, reflecting
the N locations where the particle can be localized. Note that for the decoupled bath, the
original field e obeys N b.c. at x = 0 and D at x = l. On the other hand, at the localized
fixed point the transformed dual field, 00 obeys D b.c. at x = 0 and N at x = l.
It is a useful check on our results to derive the low energy spectra using the strong
corrugation form of the Hamiltonian, Eq. (3.25). In this case the simple limit is the one
570
in which the dispersion relation n 0. This corresponds to the limit of zero hopping
between sites when the particle is localized. We expect to renormalize to this fixed point
whenever n of Eq. (3.26) is irrelevant. This has dimension a 2 /2 and so is irrelevant for
a 2 > 2 . It is very important that the hopping term is irrelevant in the strong corrugation
formulation whenever the potential is relevant in the weak corrugation formulation and
vice versa. This follows since the scaling dimensions are the inverse of each other. Thus
the diffusing fixed point is stable when the localized one is unstable and vice versa. This is
consistent with our assumption that the coupling constant flows to when it is relevant.
Ignoring the hopping term, n , the partition function is just that of the decoupled oscillator
bath, Eq. (3.78), multiplied by the result of integrating over Pc . Note that the crystal
momentum, Pc is restricted to lie in the first Brillouin zone. The number of momenta in
the first zone when the particle lives on a line of length N a is N , so we obtain a partition
function of N Zosc , the same partition function (and finite size spectrum) as we obtained
using the weak corrugation formulation. Now let us consider the case where the hopping
term is relevant. If n renormalizes to , e (0) gets pinned at one of the minima of n
which we can take to lie at e (0) = (2/a)(Pc + m/ ) for integer m. At x = l, e obeys

a simple D b.c., e (l) = 0. Thus the mode expansion for e (x) contains a winding mode in
this case:
r
(l x) Pc + 2m/a
2X
sin(nx/ l)pn .
(3.91)
e (x) =
l n
For fixed Pc , the partition function is:
elT /6 X (Pc +2m/a)2 /2lT
e
.
2lT m=
(3.92)
We must further sum Pc over the Brillouin zone. The result of the combined sums is a sum
over Pc = 2m/N a with m summed over all integers. Thus we obtain the same partition
function as in Eq. (3.87), which resulted from the weak corrugation formulation and hence
the same finite size spectrum.
We also note that the same ratio of g factors could be obtained using a compact boson
field with D and N b.c.s. In this case we would identify:
e (x) e (x) + 2R,
(3.93)
and
1
,
(3.94)
R
the smallest possible compactification radius consistent with the BSG interactions. In fact,
although the original field (x) is noncompact, as we remarked above, the transformed
field, 00 (x) can be thought of as being compact. This follows from the mode expansion
of Eqs. (3.61), (3.63) once we impose periodic boundary conditions on Q. Since Q is
identified with Q + N a it follows from Eq. (3.63) that
00 (x) 00 (x) +
N
00 (x) 00 (x) .
R
(3.95)
571
The effective compactification radius for the dual boson is increased by a factor of N . The
allowed values of the zero mode of the conjugate momentum given by Eq. (3.61) become
000 = 2R/N the correct values for the periodic boson radius R/N . The N b.c. that
we have been discussing corresponds to an N b.c. on a periodic boson. However the D
b.c. is not a simple D b.c. for the periodic boson. The boundary sine-Gordon interaction of
Eq. (3.67) has N inequivalent minima. Thus 00 (0) may take any of these N inequivalent
values corresponding to a multi-Dirichlet b.c.
The corresponding N boundary state is:
X
u
|(0, N 2m/R)ii
(3.96)
|N u ( = 0)i = gN
m
with
N 1/4
.
2R
The D boundary state for this compactification radius is:
X
u

D ( = 0 ) = g u
ei 0 nR/N |(nR/N , 0)ii
D
u
gN
(3.97)
(3.98)
with
r
u
= 1/4
gD
R
.
N
(3.99)
The multiple-Dirichlet boundary state, |MDi, is obtained by summing over the N lattice
points where the particle can be localized:
|MD u i =
N
X
u

D ( = mR) .
(3.100)
m=1
Using:
N
X
ei2mn/N = N n,N p ,
(3.101)
m=1
we thus obtain:
|MD u i =

N D( = 0) ,
(3.102)
proportional to the ordinary D boundary state with radius R. Thus:
u
= N gD = 1/4 N R.
gMD
(3.103)
Note that both degeneracy factors are simply increased by a factor of N compared to the
ordinary compact D and N values. We may use these degeneracy factors for each boundary
to calculated the total degeneracy associated with the partition functions. These give us:
gloc
2
N
,
=
4R
N
u u
= gN
gMD = .
2
u
gdiff = gN
(3.104)
(3.105)
572
These are exactly the same values obtained from an explicit calculation in Eqs. (3.88) and
(3.90).
We note that this infinite groundstate degeneracy, in the limit where the particle is
diffusing on an infinite line, is related to the unphysical nature of the CaldeiraLeggett
type model for QBM. The more physical applications of the model, to the quantum
wire problem, involve compact bosons from the beginning, and do not have this strange
behavior.
In general, the change in ln g due to a change in b.c.s at one end of the system appears
to be insensitive to the compactness or noncompactness of the boson. This arises from
the fact that the compactness only appears when one considers properties depending on
both boundaries whereas g is a property of each boundary separately. This can be checked
very explicitly in the boundary sine-Gordon case, where changes in g can be computed
perturbatively to a very large order, and matched against the thermodynamic Bethe ansatz
results [34]. In the perturbation theory, there is no dependence on the compactification at
all orders.
We note that, upon imposing periodicity of the Hamiltonian under +2R or
+ 1/R, the correct operator content at the D and N fixed points is given by the finite size
spectrum of the ordinary compact theory (radius R). Thus in order to deduce the operator
content (which is consistent with the imposed symmetry) from the finite size spectrum, it
is convenient to use compact bosons of radius R. The main effect of the noncompactness
seems to be to increase the groundstate degeneracy in the partition function by a factor
of N , a phenomena whose origin can be traced back to the infinite mass oscillator discussed
near the beginning of this section. In our discussion of the triangular lattice case we shall
use bosons compactified on this lattice. Most physical quantities, such as the change in
ln g, are not affected by this compactification.
Note that in the case when the N (diffusing) b.c. is stable, R > 1/ 2, gN < gD and
vice versa. Thus the g-theorem is obeyed; the RG flow always serves to decrease g.
4. Quantum brownian motion on a triangular lattice

In this section we first review the phase diagram of the boundary 3-state Potts model.
We then introduce the triangular lattice QBM problem and conjecture a phase diagram by
analogy with that of the Potts model. This is substantiated and studied in more detail in the
following sections.
4.1. Phase diagram of the boundary 3-state Potts model
The 3-state Potts model is a natural generalization of the Ising model to a discrete spin
variable that takes on three equivalent values. (We sometimes refer to these as A, B and C.)
As such it is naturally related to QBM on a triangular lattice, as will become clear in the
next subsection. Here we review the boundary Potts model phase diagram. The classical
Potts model Hamiltonian contains nearest neighbor interactions of these spins such that
the energy is (1) when two neighboring spins are in the same state and zero otherwise.
573
The critical behavior of the two-dimensional Potts model is equivalent to that of a one
dimensional quantum chain. In addition to the classical Potts interaction this also contains
a transverse field term which permutes the spins on each site between the three states
with equal amplitudes. A complete set of conformal invariant b.c.s for the quantum Potts
chain consists [10,35] of only four types of b.c.s. One of these is the fixed b.c. in which
the quantum spin at the end of the semi-infinite chain is constrained to take a fixed value
(either A, B or C). A second b.c. is a mixed b.c. where the quantum spin at the end of the
chain is hopping back and forth between two of the states (e.g., A and B). A third b.c. is
the free b.c. which results from simply terminating the bulk Hamiltonian without otherwise
modifying it near the boundary. Finally there is a fourth b.c. whose physical interpretation
is not obvious, and which we referred to as the new b.c.
In Ref. [10] a phase diagram was conjectured for the boundary Potts model in which
both the transverse and longitudinal fields acting on the spin at the end of the chain was
varied. As stated above, when the boundary transverse field is positive (the same sign as
in the bulk at the bulk critical point), the system is at the free fixed point. If a boundary
longitudinal field is now turned on which favors the A state, the system renormalizes to
the fixed b.c. On the other hand, if this boundary longitudinal field has the opposite sign
so as to equally favor B and C then the system renormalizes to the mixed b.c. It should be
emphasized that the mixed b.c. corresponds to the system dynamically jumping back and
forth between the B and C states. In particular, it is invariant under the Z2 subgroup of Z3
which interchanges B and C. On the other hand, the broken symmetry b.c. whose boundary
state is sum of B and C fixed boundary states is unstable against an RG flow to the mixed
b.c. If there is no longitudinal field and the boundary transverse field is negative, then the
system renormalizes to the new b.c. The special case where both types of boundary fields
vanish is a type of degenerate boundary condition for which the corresponding boundary
state is a linear combination of the three different fixed boundary states.
It turns out that all of these RG flows have direct analogies in the problem of QBM on a
triangular lattice, which thus provides a particular realization of the quantum Potts chain.
4.2. Back to quantum brownian motion
We consider a natural extension of the 1D QBM model discussed in Section 2 to two
dimensions. The Hamiltonian of Eq. (3.6) is extended by introducing a separate set of
oscillators, qka for each component of the particle co-ordinate vector, Qa , with an identical
set of masses and coupling constants:

X pE2
E k 2
mk k2
Q
PE 2
k
E +
+ V (Q)
+
qEk
.
(4.1)
H=
2M
2mk
2
mk k2
k
In the ohmic case we may again represent the oscillators by free massless relativistic boson
fields, however we now get a separate boson, a (x) for each component Qa . We emphasize
that the bosons still live on a fictitious one-dimensional line. The derivation of the boundary
sine-Gordon model in the strong and weak corrugation limits carries over directly to the
multi-dimensional case. In the strong corrugation limit the Hamiltonian is:
574
1
HS =
2
Zl

E 2

d
E
2
E
dx +
.
+ PEc (0)
dx
(4.2)
We have dropped extraneous 0 and e notation. Here (PE ) is the dispersion relation (in the
E
lowest band) for the dissipationless problem. (x)
obeys an N b.c. at x = 0 and a D b.c.
E
at x = l. The c-numbers Pc can take any values in the first Brillouin zone. Similarly in the
weak corrugation formulation this Hamiltonian is:
HW =
1
2

2

Zl
E
2
d E
(0)
e
E +
.
+V
dx
(4.3)
E
obeys N b.c.s at x = 0, l.
(We have again dropped the cumbersome 00 notation.) (x)
The observations about boson compactness made in Section 3 carry over to the twodimensional case. We henceforth set:
= 1,
(4.4)
by a rescaling of the lattice spacing. We label the physical lattice on which the particle
E so that
moves . It is convenient to impose periodic b.c.s on the particle co-ordinate, Q,
we identify:
E Q
E + N aE ,
Q
(4.5)
where aE is any primitive vector in and N is a large integer. It follows from the derivation
in Section 3 of the boundary sine-Gordon model for QBM that E may be regarded as
compactified on this coarse lattice with spacing bigger by the factor of N . Equivalently,
we may regard E as being compactified on the lattice with an extra factor of N
appearing in the degeneracy, g.
We wish to consider a model of QBM in a periodic potential with hexagonal symmetry.
The tight-binding model that we consider is a simple generalization of the one introduced
in Ref. [6]. We consider a triangular lattice generated by the vectors:

aE2 = a/2, 3 a/2 .
(4.6)
aE1 = (a, 0),
This model exhibits interesting behavior for any value of the dimensionless parameter a
but in this paper we focus primarily on the value:
a 2 = 4/3,
(4.7)
since, only then, can it be mapped into the Potts model. The simplest form of the tightbinding Hamiltonian, before adding dissipation, is:
X
ci cj ,
(4.8)
H = t
hi,j i
where the sum is over nearest neighbors.

In order to understand the connection with the 3-state Potts model it is useful to focus
on the symmetry transformations that leave fixed the center of one of the triangles on
575
Fig. 2. Triangular lattice with A,B and C sublattices marked. Note that the A points also form
a triangular lattice with spacing 3 a and orientation rotated by 90 . The B and C points form a
honeycomb lattice.
the lattice. These consist of three-fold rotations and reflections, referred to as 3m in the
international crystal nomenclature. There is actually a larger point group symmetry, 6mm,
when one considers transformations that hold fixed a lattice point. However, this is not
relevant for our purposes. It is convenient to decompose the lattice into three sublattices, A,
B and C such that the nearest neighbours of a point in one sublattice are in the other two, as
drawn in Fig. 2. Note that a 2/3 rotation of the lattice maps each point on the A sublattice
into a point on the B sublattice and similarly B C and C A. Simlarly a reflection
about a line passing through A points interchanges all B points with C points. Hence we
may think of this point group symmetry as the permulation group on three objects, S 3 . This
is the symmetry of the 3-state Potts model. We will also consider an on-site potential which
assigns different energies to the A, B and C sublattices, vA , vB and vC , thus breaking the
S 3 symmetry.
To map out the phase diagram let us begin with the strong corrugation formulation in
the case t > 0 and all vi = 0. t is relevant for this lattice spacing (of dimension x = 2/3),
corresponding to the boundary sine-Gordon Hamiltonian:
HS = H0 + Hint,
where
1
H0
2
Zl
0

2
d E
2
E
dx +
,
dx
(4.9)
576
Hint
r
r

r

2
1
1
t cos 2
1 + cos 2
1 +
2
3
6
2
r
r

1
1
1
2 .
+ cos 2
6
2
(4.10)
(The fields are all evaluated at x = 0 but we suppress this argument. We have set PEC
to 0.) We expect t to induce an RG flow to the perfect mobility phase, corresponding to a
E
D boundary condition, (0)
= 0. This fixed point is stable as we can see by considering
the lowest dimension potential term with the full symmetry of the lattice, in the weak
corrugation formulation, which is:
HW = H0 + Hint,
(4.11)
where

2
E
2
e
E + d
dx
,
dx
0

v cos 4 2 + cos 2 + 3 1 + cos 2 3 1 .
1
H0
2
Hint
Zl
(4.12)
This has dimension 2 and is irrelevant. Now consider turning on a potential vA < 0, which
favors the A sublattice. We do this in the most symmetric possible way, preserving a Z3
symmetry of rotation about a lattice point, as well as a mirror symmetry about a lattice
link, the 3m subgroup of the original 6mm point group.
Choosing the origin to lie on the A sublattice, this potential is:
r
r
r

2
1
1
1 + cos 2
1 +
2
H H vA cos 2
3
6
2
r
r

1
1
1
2 .
(4.13)
+ cos 2
6
2
This operator has dimension 2/3 at the perfect mobility fixed point, which corresponds to
E it is relevant. Thus, for v > 0, we expect a
a D b.c. on E or equivalently, an N b.c. on ;
A
E The stability of this fixed
flow to a localized fixed point, corresponding to a D b.c. on .
point can be checked by observing that, if vA flows to infinity, then only hopping between
A sublattice points is possible. The particle gets localized on
one of the A sublattice sites.
This is a stable fixed point since the intra-sublattice is now 3 a and hence the hopping
term gives:

(4.14)
Hint = t cos 4 2 + cos 2 + 3 1 + cos 2 3 1
of dimension 2 which is irrelevant. Note that the dimension of the nearest neighbor
hopping term is 2/3, smaller by a factor of 3 due to the reduced distance of the hop and
hence relevant. The effect of the symmetry breaking term, vA is to stabilize a localized
phase on one of the sublattices.
The analogy with the Potts model is quite transparent. The perfect mobility fixed
point corresponds to the free b.c. in the Potts model. The potential vA corresponds to a
577
longitudinal boundary field which produces a flow to the fixed (A) b.c. Now consider the
case vA < 0. If vA , the particle is localized on one of the B or C sublattice sites.
Together, these two sublattices define a honeycomb lattice. However, for finite negative vA ,
this is not a stable fixed point. This can be seen by considering hopping on this honeycomb
lattice. Since the nearest neighbor distance is again a, this hopping term has dimension 2/3
and is relevant. Thus, following the logic of Yi and Kane, there must be an intermediate
mobility fixed point. In the Potts model we add a boundary field which favors either the
B or C state. This leads to a flow to the mixed b.c. of the boundary Potts model. Thus we
see that the new fixed point found by Yi and Kane corresponds physically to the mixed
boundary fixed point in the Potts model. In the Potts model we think of the boundary spin
as fluctuating back and forth between B and C states. In the QBM problem we think of
the particle as hopping back and forth between B and C sublattices. As can be shown
explicitly, this state has an intermediate mobility. We refer to this at the Y fixed point,
after YiKane.
In fact, we can discover yet another intermediate fixed point in the QBM problem by
pursuing this analogy further. We now set the symmetry breaking potentials vi , to 0 but
consider a negative hopping term, t < 0. This would seem to correspond to a negative
transverse boundary field in the Potts model. This was argued to lead to a flow to another
fixed point, unimaginatively referred to as the new fixed point in Ref. [10]. Thus we
expect a new fixed point to occur in the QBM problem with t < 0. We refer to this
as the W fixed point. Such a new fixed point is possible due to the lack of particle
hole symmetry of this model. For t > 0 the lowest energy single-particle state has crystal
momentum pE = 0. However, for t < 0 the lowest
energy single-particle states occur at two
inequivalent crystal momenta: pE = (2/3a, / 3 a). These two lowest energy states for
the particle are dual to the B and C sublattices in the mixed phase. In fact, duality in QBM
is easily understood, simply corresponding to:
E
E .
(4.15)
Comparing Eqs. (4.10) and (4.13) we see that the flow from localized to perfect mobility
is dual to the flow from perfect mobility to localized on the A sublattice, induced by a
positive t or vA respectively. Similarly, the flow from localized to mixed is dual to the flow
from perfect mobility to new, induced by a negative t or vA respectively.
If both transverse and longitudinal boundary fields are set to 0 in the Potts model we get
a phase referred to as A + B + C signifying that the boundary spin remains fixed in any
of its 3 possible states. The corresponding boundary state is a sum of three boundary states
corresponding to the 3 possible fixed b.c.s. Turning on a positive transverse field produces
a flow to the free b.c. whereas a negative transverse field produces a flow to the new b.c.
The analogue in QBM is to start with t = 0, corresponding to the particle being localized
at any lattice site (A, B or C). We may, if we wish, include an irrelevant intra-sublattice
hopping, of dimension 2. Now turning on t induces a flow to either perfect mobility or new
fixed points. This phase diagram is summarized schematically in Fig. 3. Note that, since
only localized behavior is possible on the line t = 0, two copies of the Y fixed point must
occur, as drawn. Also note that, although the flow away from the localized on A, B or C
578
Fig. 3. Phase diagram and RG flows of the triangular lattice QBM model with hopping strength t
E As pointed out in
and potential vA . Neumann and Dirichlet b.c.s are imposed on the dual fields, .
Ref. [6] and discussed in Section 7, the localized on B or C phase corresponds to th weak coupling
fixed point in the 3-channel Kondo problem.
fixed point depends on the sign of t, that from the localized on B or C fixed point does
not. This is consistent with the fact that the sign of the hopping term cannot be changed by
a redefinition of the sign of the electron operators for a triangular lattice but can be for a
honeycomb lattice.
We also comment briefly on the model with general values of a. The strong corrugation
Hamiltonian of Eq. (4.10) becomes:

1
1
3 2
3 2
+
+ cos a
(4.16)
Hint = t cos a1 + cos a
2
2
2
2
and the symmetry breaking weak corrugation term in Eq. (4.13) becomes:

2 1 2 2
2 1 2 2
4 1
+ cos
. (4.17)
+ cos
+

H H vA cos
3a
3a
3a
3a
3a
The scaling dimension of the hopping term at the (symmetric) localized fixed point and of
the symmetry breaking potential at the N fixed point are:
D = a 2 /2,
(4.18)
N = 8/9a .
(4.19)
Thus the hopping term is relevant for a 2 < 2 and vA is relevant for a 2 > 8/9 indicating
that the perfect diffusion fixed point is unstable in this region. For vA > 0 we again expect
a flow to the fixed point where the particle gets localized on the A sublattice. The scaling
dimension of the intra-sublattice hopping term is 3a 2/2 so this localized phase is stable
in this region and we expect an RG flow from diffusing to localized on A. On the other
579
hand, if vA < 0 we should consider the stability of the localized on B or C phase. The
scaling dimension of the inter-sublattice hopping term (between the B and C sublattices)
is a 2 /2 , so neither diffusing nor localized phases are stable over the range:
8/9 < a 2 < 2.
(4.20)
In this entire range we expect a nontrivial fixed point. This can be studied using an
expansion [8] for a 2 = 8/9 + or a 2 = 2 for 0 < 1. It can be solved exactly
at a 2 = 4/3 as we show in the next section. In general, however, it remains an unsolved
problem.
It is also instructive to calculate the degeneracy (g) factors at the diffusing (vA = 0)
and localized on A phases. We may do this using the results in Section 2.2 for compact
bosons, taking into account the connection of the QBM problem with compact bosons
explained in Section 3 and earlier in this section. We assume that the space on which the
particle moves is periodic with
E Q
E + N aEi ,
Q
(4.21)
where the primitive vectors aEi are given by Eq. (4.6). The unit cell area for the fine lattice
with spacing a is:
(4.22)
V0 ( ) = 3 a 2 /2.
(Note that this is twice the area of a triangle on the lattice; each triangle contains 1/2 point
or three points, each of which is shared with six triangles.) Thus the degeneracy factor for
the D fixed point where the particle is localized on any of A, B or C sublattices is:
N 2
(4.23)
g3D = 1/4 .
3 a
This is the result for the compactified boson with dual lattice from Eq. (2.57) multiplied
by the factor of N . The degeneracy for the state where the particle is localized on the A
sublattice only is:
gD = g3D /3,
(4.24)
and that for the state where the particle is localized on the B or C sublattices is:
gDD = (2/3)g3D .
(4.25)
These factors of 1/3 and 2/3 can be understood as simply corresponding to the number
of sites on which the particle can be localized. Alternatively, we may think of the D fixed
point
by a factor
as being the same as 3D except that the lattice constant a is increased
of 3 and the factor N , defined in Eq. (4.5), is decreased by a factor of 3 in order that
the area on which the particle moves remains fixed. The degeneracy factor for the N fixed
point is:
gN =
N a31/4
.
2 2
(4.26)
580
This is obtained from Eq. (2.52) for a compact boson, multiplied by the factor of N .
Note that the total degeneracy for the phase where the particle is localized on A, B or
C sublattices is:
N2
.
(4.27)
2
Following the discussion in Section 3 for the one-dimensional case, we see that this
is simply the number of
lattice sites times the degeneracy for the decoupled oscillator
bath. (The factor of 1/ 2 that occurs in the one-dimensional case gets squared in two
dimensions.) We also see that the ratio between fully localized and freely diffusing fixed
points is:
gloc = g3D gN =
4
g3D
=
.
gN
3 a2
(4.28)
Thus the flow from fully localized to freely diffusing is consistent with the g-theorem only
for:
4
(4.29)
a2 < .
3
On the other hand the ratio of g-factors between the localized on A and freely diffusing
fixed points is:
4
gD
=
,
gN
3 3 a2
(4.30)
so the flow between freely diffusing and localized on A is consistent with the g-theorem
for:
4
(4.31)
a2 > .
3 3
It is instructive to consider a more general model with even less symmetry than that
of Eq. (4.13). The model of Eq. (4.13) in which we have changed the energy on the A
sites, relative to B and C sites still has a 3m point group symmetry when we consider
transformations holding fixed a lattice point (A, B or C). The less symmetric model that
we now consider has no rotational symmetry whatsoever. This model is obtained from Eq.
(4.13) (weak corrugation formulation) by letting the coefficients of the 3 terms be different:
r
r
r

2
1
1
1 v2 cos 2
1 +
2
HW = H0 v1 cos 2
3
6
2
r
r

1
1
1
2 .
(4.32)
v3 cos 2
6
2
The behavior of this model becomes obvious in the case v1 6= 0, v2 = v3 = 0. In this case
the (relevant) potential only depends on the co-ordinate Q1 . Thus we expect the motion of
the particle to be localized in the 1-direction but freely diffusing in the 2-direction. Now
there is only 1 minimum per unit cell for either sign of v1 . We may think of the particle
as diffusing freely along vertical lines on the lattice. To check the stability of this fixed
581
point we should consider the hopping process between vertical lines. Since the horizontal
distance between points on the A sublattice is 1/2 times the lattice spacing (of the A
sublattice) we expect this hopping term to have dimension 1/2 [smaller by 1/4 than the
dimension 2 hopping in Eq. (4.14)]. Thus this fixed point is unstable.
5. Construction of boundary states via a conformal embedding

5.1. Conformal embedding
Let us consider QBM on the triangular lattice discussed in Section 4, considering the
weak corrugation Hamiltonian of Eq. (4.17). Because the symmetry among the sublattices
A, B and C is broken by the applied potential, the generic translation symmetry of the
model is invariance under translations which do not interchange the sublattices. We can
impose this symmetry on the system, so that we can use the compactified formulation.
The
compactification lattice is thus a triangular lattice of nearest neighbor distance 3 a.
Namely, we introduce the compactification
E E + ,
(5.1)
where is a triangular lattice generated by

3a/2, 3 a/2 .
0, 3 a ,
(5.2)
Its dual is generated by (2/3a, 0) and (1/3a, 1/ 3 a). The allowed vertex operators
E where vE is an element of the lattice . As
made from E are of the form exp(i vE ),
mentioned in Section 3.3, the operator content upon compactification corresponds to the
possible boundary perturbations given the imposed symmetry. Note that the Hamiltonian
of Eq. (4.17) is the most general one, up to less relevant operators, respecting the 3m point
group symmetry.
E while it is
For vA = 0, the corresponding boundary condition is Neumann on ,
Dirichlet on E for v . However, there are other nontrivial boundary conditions
A
which are neither Dirichlet or Neumann, as discussed in Section 4. They correspond to

nontrivial phases in QBM. Our problem, then, is to construct and to classify the possible
boundary states of the c = 2 CFT. Possible boundary conditions may be restricted by higher
symmetries, such as current conservation at the boundary. However, in the case of QBM,
apparently the only requirement (at the RG fixed point) is the conformal invariance on the
boundary and Cardys consistency conditions. Even if there are extra symmetries in the
bulk, they are not necessarily respected at the boundary. This allows some extra boundary
conditions which would be forbidden if the higher symmetry is imposed on the boundary.
Unfortunately, at present we have no systematic understanding of the boundary states
in an irrational CFT, which has an infinite number of primary fields. While we do not
know the solution to this fundamental problem, in this paper we analyze some nontrivial
boundary conditions for the special compactification radius a 2 = 4/3. In this case, the
irrational c = 2 CFT admits a conformal embedding in terms of rational CFTs. For such a
582
rational CFT, we can invoke the fusion construction of boundary states, to find nontrivial
boundary states for the c = 2 CFT.
This value of a 2 is actually the point where Yi and Kane [6] obtained the exact value of
the mobility at a nontrivial fixed point by a slightly different approach. Our approach leads
to a somewhat more systematic description of the boundary states, and several additional
results including novel boundary states.
The present c = 2 boundary CFT admits a conformal embedding c = 2 = 1/2 + 7/10 +
4/5 = (Ising) + (Tricritical Ising) + (Potts). Namely, the partition functions on the strip
can be expressed (in the open string channel), as
P

I
T
3/2
0 + 3P
ZDD (q) = 0I 0T + 1/2
P

T
I
T
P
+ 1/2
1/10
2/5 + 7/5
,
(5.3)
+ 0I 3/5
P

I T
I
T
P
P
ZNN (q) = 0 0 + 1/23/2 0 + 3 + 22/3

T
I
T
P
P
P
+ 1/2
1/10
+ 2/5
+ 7/5
21/15
,
(5.4)
+ 0I 3/5
where hI,T ,P is the Virasoro character of the weight h for Ising, tricritical Ising and
Potts model, respectively. (The argument q is omitted.) The two amplitudes have the same
factors for Ising and tricritical Ising sector; the only difference is in the Potts sector. We
emphasize that these are the partition functions for compact bosons compactified on the
lattice, of Eq. (5.2). They do not quite correspond to the partition functions occurring
in the QBM problem. Nonetheless, as explained in Section 3.3, these are the partition
functions which are related to the boundary operator content. We remark that, since the 0
and 3 and also 2/5 and 7/5 partition functions only occur added together in Eqs. (5.3) and
(5.4), ZDD and ZNN can be expressed in terms of the W-characters of the 3-state Potts
model, invariant under W-symmetry.
In fact, we have no proof of these identities. However, we have verified, using
M ATHEMATICA, that the first several (40 to 100) terms in the series expansion agree
exactly. In the following, we assume this conformal embedding. Because of several
additional pieces of evidences, we believe it is indeed correct. Namely, as we will show
later, we can construct several reasonable boundary states based on the embedding.
We also note that, the torus partition function of the c = 2 theory, even at this particular
compactification, apparently cannot be written in terms of Ising, tricritical Ising and
Potts models. We are in a somewhat strange situation that the boundary CFT admits
the conformal embedding while the bulk CFT does not. Although this is difficult to
understand, it does not pose a real problem in our construction. The fusion construction
is a multiplication by constant factors of each Ishibashi state appearing in the reference
boundary state, Eq. (2.27). These constant factors are given by the matrix elements of the
matrix of modular transformations. Thus, by construction, the obtained boundary state is a
linear combination of the Ishibashi states. They are well contained in the Hilbert space of
the theory, as long as the original reference state is a physical boundary state.
In fact, a similar phenomenon already appeared for the Dirichlet and Neumann boundary
conditions in the simpler c = 1 free boson theory, as we have discussed in Section 2. There,
they are connected by the fusion with a twist operator, which is contained in the Z2 orbifold
583
theory. While the free boson theory is not equivalent to the Z2 orbifold, the latter appears
in the discussion of the boundary conditions in the former.
It is a valuable check to see that the dimensions of boundary operators at D and N
fixed points are reproduced by the conformal embedding of Eqs. (5.3) and (5.4). We see
that at the D fixed point we have only integer dimension operators. (The = 1 operators
E which are not allowed by
presumably correspond to the exactly marginal operators, ,
symmetry.) On the other hand, at the N fixed points we see that there are 6 relevant
operators with = 2/3. These must correspond to the 3 cosine operators in Eq. (4.32)
together with the corresponding sines. It is interesting to note that 2 of these operators are
purely in the Potts sector while the other 4 are products of operators in 2 or more sectors.
We expect that the 2 pure operators are the interaction term in the Hamiltonian of Eq. (4.17)
and the same operator obtained by replacing all cosines by sines, as argued below.
In fact, there is an alternative conformal embedding, also involving the Potts model,
(5)
in which the Ising and tricritical Ising models are replaced by the Z3 conformal field
(p)
theory [16]. (A detailed review of Z3 theory is given in Ref. [36].) This may be regarded
as a sort of tricritical 3-state Potts model. Again there is a W-algebra and only W characters
occur in the conformal embedding. This alternative conformal embedding is obtained
from Eq. (5.3) and (5.4) by replacing the Ising and tricritical Ising characters by the Z3(5)
characters. Here we use the (conjectured) identity
I
T
3/2
= 05 + 225 ,
0I 0T + 1/2
T
0I 3/5
I
T
+ 1/2
1/10
5
5
23/5
+ 8/5
.
(5.5)
(5.6)
Here a5 refers to characters in the Z35 theory. We note that, again, this identity is only
verified up to finite order with M ATHEMATICA. The equivalence between the Z3(5) model
and the product of Ising and tricritical Ising models is only a partial equivalence. Certain
sums of conformal characters in the two theories are presumably equal but there is not a
complete equivalence of all characters.
Further insight into the conformal embeddings can be obtained by rewriting the
interaction term in Eq. (4.17) in a more symmetric way, first by reexpressing everything in
terms of chiral components, using L = R at the fixed point, and by introducing the new
chiral fields j :
8
1L ,
3a

4
1L + 3 2L ,
2
3a

4
1L 3 2L
3
3a
with propagators
1

16
i (z)i (w) = 2 ln(z w),
9a

8
ln(z w), i 6= j.
i (z)j (w) =
9a 2
(5.7)
(5.8)
584
The (weak corrugation) interaction term in Eq. (4.17) now reads, recalling a 2 = 4/3,

(5.9)
Hint = vA (cos 1 + cos 2 + cos 3 ) = (vA /2) 1 + 1 ,
where:
1
3
X
eik .
(5.10)
k=1
1 corresponds to the fundamental Z3 parafermion field of the Potts model [37,38]. (This
generalizes the representation of a Majorana fermion-Z2 parafermion as cos .)
The relation between the c = 2 free bosonic theory and the Potts model is not
straightforward. It can be formulated in terms of cosets beginning with SU(3)1 SU(3)1 .
This conformal field theory arises from bosonization of critical SU(3) spin chains (the
two factors of SU(3)1 arising from left and right movers) and the associated boundary
critical phenomena is closely related to that of the triangular lattice QBM problem. It
will be discussed in a later paper [39]. It is natural to associate the SU(3)2 CFT with the
diagonal SU(3) symmetry of this model. The remaining coset, SU(3)1 SU(3)1 /SU(3)2 ,
with c = 4/5, gives the 3-state Potts model. This is also equivalent to SU(2)3 /U (1). Yi
and Kane first discussed the nontrivial critical behavior of the triangular lattice QBM
problem using results from the 3-channel SU(2) Kondo problem, which is, in turn, related
to the SU(2)3 CFT. The second factor in the conformal embedding, SU(3)2 can be further
factorized into two U (1) CFTs [the maximal abelian subalgebra of SU(3)] and a c = 6/5
coset, SU(3)2 /U (1)2 which is naturally regarded as the Z35 CFT. Since the SU(3)1 CFT,
with c = 2, is actually equivalent to two free bosons, this conformal embedding essentially
expresses 2 free bosons in terms of the Potts model and the Z35 CFT. One can certainly
write the stress energy tensor for 2 free bosons as T = T1 + T2 where:
"
#
3
X
1X
1
2
i(j k )
(j ) +
e
,
T1 =
5
2
j 6=k
j =1
"
#
3
X
1
3X
2
i(j k )
(j )
e
,
(5.11)
T2 =
5
4
j =1
j 6=k
such that the short distance expansion of T1 with T2 is trivial. Here denotes /z where
z = + ix and is imaginary time. T1 is a stress energy tensor with central charge c1 = 45
of the coset SU(3)1 SU(3)1 /SU(3)2 or SU(2)3 /U (1), and similarly for T2 with c2 = 65
of the SU(3)2 /U (1)2 coset. The parafermion 1 turns out to be primary both for T1 and
T1 + T2 , but it is not the case for most other operators of interest. Some of the most crucial
operators in the Potts model, like the field with conformal weight 25 , do not have any
known representation in the two boson theory, neither as vertex operators nor generalized
twist fields.
It follows that, strictly speaking, we do not have a conformal embedding. As mentioned
before, the situation is reminiscent of what happens at the Ising square point of the c = 1
theory: while c = 1 = 12 + 12 , and a decomposition of the c = 2 stress energy tensor
analogous to (5.11) exists [40], the periodic boson theory cannot be considered as the
585
product of two Ising theories; only the Z2 orbifold can. In the absence of an understanding
of the orbifold, one would nevertheless observe that, if the torus partition function of the
periodic boson cannot be expressed in terms of Ising model characters, boundary partition
functions can:
2
2
I
I 2
,
ZDD = 0I + 1/2
,
ZNN = 0I + 1/2

I
I
I
(5.12)
ZND = 1/16 0 + 1/2 .
Again, the existence of a boundary embedding in our c = 2 case suggests that the
situation is similar, and that there is some sort of orbifold version of the bosonic theory for
which there would be a genuine conformal embedding. Nevertheless, since the perturbation
in (4.17) makes sense as a pure Potts operator, it is natural to expect it to induce a flow
purely in the Potts sector of the theory. Thus we come to the important conclusion that the
interaction term in Eq. (4.17) is purely a Potts operator. In fact, it is precisely the same
boundary operator which corresponds to a boundary magnetic field, h,1 in the Potts
model where = 1, 2, 3 labels the Potts variable. The case vA < 0 then corresponds to
h < 0, inducing a flow from free to fixed boundary conditions in the Potts model, while the
case vA > 0 corresponds to h > 0, and should instead induce a flow from free to mixed.
This verifies the handwaving arguments of Section 4 relating QBM to the Potts model. Note
however that this connection is only established for the special choice of lattice spacing
a 2 = 4/3.
We also consider the more general QBM model of Eq. (4.32) in which the remaining 3m
point group symmetry is broken. Upon using the conformal embedding, we now expect
that products of Potts with Z3(5) (or Ising tricritical Ising) operators will appear in the
Hamiltonian.
In the remainder of this section we use the fusion technique to extract information about
the boundary critical points of the QBM model. By considering fusion in the Potts sector
we can study all the fixed points that occur in the 3m symmetric model. We expect that
we have obtained all fixed points with this symmetry yielding the phase diagram of Fig. 3.
On the other hand, by considering fusion in the other sectors we obtain a collection of
additional fixed points whose properties we understand in less detail. In Section 6 we
discuss the integrability of some of the RG flows between fixed points and thereby confirm
the corresponding ratios of g-factors.
5.2. Fusion in the Potts sector
P are
The Potts sector in the first term of the amplitudes 0P + 3P and 0P + 3P + 22/3
exactly the fixedfixed and freefree amplitudes in the Potts model. It suggests that the
Dirichlet and the Neumann boundary conditions correspond to the free and fixed boundary
conditions of the Potts model, respectively.
In the (pure) Potts model, we can construct the free boundary state from the fixed
boundary state by the fusion with the weight-1/8 primary operator O44 (which is absent
in the bulk spectrum of the Potts). In fact, the fusion with the same operator in the Potts
sector in our conformal embedding gives the Neumann b.c. from the Dirichlet b.c. The
586
Potts sector in the first term of (5.3) is the freefree amplitude of the Potts model, and is
thus transformed to the fixedfixed amplitude by double fusion:
P
P
P
+ 13/8
0P + 3P + 22/3
,
0P + 3P 1/8
(5.13)
where each means the fusion with 1/8. On the other hand, the Potts sector of the second
term of (5.3) is transformed as:
P
P
P
P
P
P
P
+ 7/5
1/40
+ 21/40
21/15
+ 2/5
+ 7/5
.
2/5
(5.14)
Thus, by double fusion with 1/8 in the Potts sector, the NeumannNeumann amplitude (5.4) is obtained from the DirichletDirichlet amplitude (5.3).
In this fusion construction of the Neumann boundary state, the ratio of the groundstate degeneracy gN /gD is given by the ratio of modular S-matrix elements of the Potts
model. Thus it automatically agrees with the ratio of Potts degeneracy gfree /gfixed . The
correspondence between Dirichlet/Neumann boundary conditions in the c = 2 theory and
fixed/free boundary conditions in the Potts model is also consistent with the integrable field
theory approach taken in Section 6.
In a similar way, we can construct another boundary state for the c = 2 theory, which
corresponds to the mixed boundary condition of the (pure) Potts model. In the Potts model,
the mixed boundary condition is obtained from the fixed boundary condition by fusion with
the 2/5-operator (). Thus we attempt a fusion with the same operator in the Potts sector
to construct a boundary state from the Dirichlet boundary state. The Potts sector of the first
term in (5.3) is transformed as:
P
P
P
P
+ 7/5
0P + 3P + 2/5
+ 7/5
,
0P + 3P 2/5
(5.15)
which is the same as the transformation of fixedfixed amplitude to mixedmixed one in

the Potts model. That of the second term is transformed as:
P
P
P
P
P
P
+ 7/5
0P + 3P + 2/5
+ 7/5
0P + 3P + 22/5
+ 27/5
.
2/5
Thus, the amplitude for the nontrivial boundary state (Y -state) is given by
P

I
T
P
P
3/2
+ 7/5
0 + 3P + 2/5
ZY Y (q) = 0I 0T + 1/2
P

T
I
T
P
P
+ 1/2
1/10
+ 27/5
0 + 3P + 22/5
,
+ 0I 3/5
(5.16)
(5.17)
where the argument q is omitted. By the modular transformation to the closed string
channel, it is expressed as
T

P

3 + 5 I
I
T
I
T
P
=
7/16
0 + 1/2
0 + 3/2
+ 21/16
0 + 3P + 22/3
ZY Y (q)
4 3
T

3 5 I
I
T
I
T
0 + 1/2
3/80
3/5 + 1/10
+ 21/16
+
4 3

P
P
P
+ 7/5
+ 21/15
,
(5.18)
2/5
whereqthe omitted argument is now q.
The ground-state degeneracy for this state is
p
gY = (3 + 5)/(4 3 ) = 2 cos (/5)/ 2 3. Again, by construction, the ratio of the

degeneracy is equal to corresponding one in the pure Potts model
gfree
gN
3
.
=
=
gY
gmixed 2 cos /5
587
(5.19)
It turns out that this Y -state is identical to the nontrivial fixed point found by Yi and
Kane [6] using a different mapping. This will be confirmed by the calculation of mobility,
and also by a physical consideration on the relation to the Potts model.
It has been known that the Potts model admits fixed, mixed and free boundary conditions
as conformally invariant boundary conditions. Recently [10], a new boundary condition
was found in the Potts model and related to the mixed one by the duality transformation.
The new boundary state is obtained by fusion with the operator O22 of dimension 1/40
from the fixed boundary state. Applying the fusion with the same operator to the Dirichlet
Dirichlet amplitude, we obtain a new boundary state, which we label W , for the present
problem:
P

I
T
P
P
P
3/2
+ 22/3
+ 21/15
0 + 3P + 2/5
ZW W (q) = 0I 0T + 1/2
P

T
I
T
P
P
P
+ 1/2
1/10
+ 22/3
+ 41/15
0 + 3P + 22/5
. (5.20)
+ 0I 3/5
This is another nontrivial boundary state, which corresponds to the Potts newqboundary
3/2.
state. The ground-state degeneracy of the W -state is given by gW = 2 cos (/5)
Thus, there is a corresponding boundary state in the present problem for each boundary
state in the Potts model. The ratios of g-factors between the boundary states are identical
to the corresponding ratios in the Potts model, by construction. Moreover, more boundary
states can be constructed in the Potts model by forming superposition of several boundary
states. Such boundary states contains dimension-zero boundary operator(s) other than
identity. Although they are often unphysical due to their instability, they can be relevant
for some physical situations with a first-order transition.
In the present problem, the potential minima form a triangular lattice isomorphic to for
vA > 0. However, for negative vA , the potential minima form a hexagonal lattice, which has
two points within each unit cell of . In the limit vA , the corresponding boundary
condition is Dirichlet, but the boundary field is pinned to one of two inequivalent minima.
This can be represented by double Dirichlet boundary state, which is the superposition
of two Dirichlet boundary states for the two minima. Considering the correspondence
Dirichlet fixed and Neumann free, it would be natural if the double Dirichlet
corresponds to the sum of two fixed, say A and B boundary states in the Potts model.
Indeed, the double Dirichlet amplitude can be expressed as follows:

I
T
P
3/2
20P + 23P + 22/3
ZDD,DD (q) = 0I 0T + 1/2

T
I
T
P
P
+ 1/2
1/10
+ 22/5
21/15
,
+ 0I 3/5
(5.21)
where the Potts part of the first term is the amplitude for the superposed state A + B.
To summarize, based on the conformal embedding, we find several boundary states
for the c = 2 problem constructed by fusion in Potts sector as follows. There is a c = 2
boundary state corresponding to each Potts boundary state.
588
c = 2 model
Potts model
Dirichlet
Fixed (A)
Neumann
Free
Y (i-Kane)
Mixed (AB)
Double Dirichlet
A+B
New
g-factor (for c = 2)
p
1/ 2 3
q
3/2
p
2 cos (/5)/ 2 3
q
2/ 3
q
2 cos (/5)
3/2
5.3. Fusion in Ising/tricritical Ising sector

It is also possible to construct other boundary states by fusion in Ising or tricritical-Ising
sectors.
The U -state is obtained from D-state by fusion with either 1/16 operator of Ising or
7/16 operator of tricritical Ising. Its diagonal partition function is given by
T
P

P
I
T
T
T
+ 3/5
0 + 3/2
0 + 3P + 1/10
2/5 ,
(5.22)
ZU U (q) = 0I + 1/2
with the degeneracy gU = 1/31/4 . Similarly, S-state is obtained by fusion with the same
operators from the N -state
P

T
P
0T + 3/2
0 + 3P + 22/3

T
T
P
P
+ 3/5
+ 2/5
21/15
,
+ 1/10
I
ZSS (q) = 0I + 1/2
(5.23)
with the degeneracy gS = 31/4 . It turns out that these state represents mixtures of Dirichlet
and Neumann: imposing Dirichlet in one component and Neumann in the other.
On the other hand, another boundary state, the T -state is obtained from D-state by fusion
with 3/80 operator of tricritical Ising. The diagonal amplitude is

T
T
T
+ 3/5
+ 3/2
0T + 1/10
I

T
T
T
+ 0T + 21/10
+ 23/5
+ 3/2
,
I
ZT T (q) = 0I + 1/2
(5.24)
with the degeneracy gT = (7 + 3 5/6)1/4 . Yet another state, R, is found by fusion with
the same operator from N -state:

T
T
T
+ 3/5
+ 3/2
0T + 1/10
(I + + )

T
T
T
T
+ 0 + 21/10 + 23/5 + 3/2 ( + + ) ,
I
ZRR (q) = 0I + 1/2
(5.25)
which has the degeneracy gR = [3(7 + 3 5)/2]1/4 . The calculation of the mobility, which
will be given later, suggests that they are mixtures of Y and W , namely taking the dual of
one component of the bosons in Y or W boundary states.
589
5.4. Fusion in the Z3(5) sector

(5)
We can apply similar lines of arguments as in the previous section to the Z3 sector in
the conformal embedding of the c = 2 model. As we will see, we obtain new boundary
states as well as those found using the Potts sector. Starting from the Dirichlet boundary
state, we have tried fusion with all primary fields in the Z3(5) model. As a result, we found
new boundary states, besides the same ones obtained by Potts fusion. For simplicity, here
we focus on the boundary states with single dimension-0 boundary operator.
A new boundary state F is obtained by fusion with Z3(5) operators of dimension 1/9,
7/9 or 13/9 from the D boundary state:

5
5
5
5
+ 8/5
+ 31/10
+ 23/5
.
(5.26)
ZF F = I 05 + 225 + 31/2
This turns out to represent free diffusion under a magnetic field [29]. While a detailed
discussion of this identification will be given in a later publication, we will see a signature
of the magnetic field, namely the Hall effect, in the next subsection.
Another one, X is obtained from the D by fusion with an operator of dimension 2/45,
17/45 or 32/45. The amplitude in the open string channel is given by

5
5
5
5
+ 8/5
+ 225 + 31/2
+ 31/10
ZXX (q) = I 05 + 23/5

5
5
5
5
+ 28/5
+ 225 + 31/2
+ 61/10
.
(5.27)
+ 05 + 43/5
By construction, the ground-state degeneracy gX satisfies
S0
gX
= c0 ,
gD
S0
(5)
where S is the matrix modular transformations of the Z3 model and 0 and c represents
q

the identity operator and the operator used in the fusion. S00 = (5 5)/10 6 and Sc0 =
q

(5 + 5)/10 3 gives gX = 4 cos (/5)gD = 2gY . Fusion with an operator of dimension
3/5 or 8/5 gives the same Y -state previously obtained by fusion in Potts sector.
On the other hand, starting from the Neumann boundary state, we obtain two additional
(5)
boundary states V and Z by fusion in the Z3 sector. No other states are found by fusion in
Z3(5) on other known states. To summarize, there are four new boundary states F , X, V and
(5)
Z obtained by fusion in Z3 sector. As we will see later, these four states exhibits finite Hall
mobility, and are presumably related to RG fixed points of QBM under a magnetic field.
5.5. Mobility and nature of the boundary states
Now let us calculate the mobility in the QBM problem. It helps us to identify the physical
natures of the constructed boundary states. The mobility is defined as the linear response
coefficient of the velocity of the particle to the external force. The external FE acting on
E The linear
the particle is represented by the additional term to the Hamiltonian, FE Q.
response of the velocity of the particle to this force is given by the Kubo formula

Z
dQa (0) b
dQa
b
Q (t) .
(5.28)
= F
dt
dt
dt
590
Thus the static mobility tensor can be expressed [6] as

dQb
dQa
()
() .
ab = lim
0
dt
dt
(5.29)
E
E with (0),
Using the weak corrugation formulation we can identify Q
and the velocity
E
E
operator d Q/dt
with d (0)/dt,
which corresponds to the current operators. Thus the
mobility is deduced from the two-point boundary correlation function of the current
ja = Ja + Ja (a = x, y), where Ja and Ja are a-component of the holomorphic and
antiholomorphic currents. The mobility corresponds to the conductance in the problem
of a tunneling between quantum wire [8,41], which can be also mapped to essentially the
same boundary CFT problem. Note that we now have a two-component boson, and thus
the current has two components.
The time dependence of the correlation function is basically insensitive to the boundary,
although some coefficients do depend on the boundary state. As in Ref. [41], the Ja Jb
and Ja Jb correlations are insensitive to the boundary conditions, and can be fixed by an
appropriate normalization of the current operator. On the other hand, while the Ja Jb
correlation obeys always the same power law, the amplitude depends on the boundary
condition. The amplitude Aab is determined by the boundary state [42] as
Aab =
h0|Ja,1 Jb,1 |Bi

,
h0|Bi
(5.30)
where |0i is the ground state, |Bi is the boundary state under consideration, and Ja,1 (Jb,1 )
is the level-1 annihilation part of the holomorphic (antiholomorphic) current. By a suitable
normalization of the current operator, the mobility ab can be written as
ab + Aab
.
(5.31)
2
Since the N boundary condition corresponds to the delocalized phase, we normalize the
mobility in the N phase to be unity, fixing AN
ab = ab for the Neumann boundary condition.
On the other hand, the D boundary condition corresponds to the localized phase and the
mobility should vanish, thus AD
ab = ab . In our conformal embedding approach, the two
level-1 states are identified as
ab =
Jx,1 |0i = |0iI |3/5iT |2/5iP ,
(5.32)
Jy,1 |0i = |1/2iI |1/10iT |2/5iP ,
(5.33)
since they are the only two combination of Virasoro primaries with weight 1.
The coefficients Aab of these states in the boundary state are changed by fusion. The
changes are given by the modular S-matrix elements. Thus we can calculate the coefficients
and consequently the mobility of the new boundary state constructed by fusion. Since the
both states contains the same Potts primary state |2/5iP , fusion in the Potts sector gives
Aab = Aab . Thus the mobility is also diagonal and isotropic: ab = ab .
The N -state is obtained by fusion with the operator of dimension 1/8 in the Potts sector
from the D-state, and the amplitudes is given by
591
2/5
S1/8 S00
AN
=
= 1,
2/5 0
AD
S0 S1/8
(5.34)
where S is the matrix of modular transformations for the Potts model. This is indeed
consistent with the physical consideration D = 0 and N = 1.
Similarly, the Y -state is obtained by fusion with 2/5-operator in the Potts sector and thus
the amplitude for the Y -state AY is given by
2/5
S2/5 S00
3 5
AY
.
(5.35)
= 2/5 0 =
AD
2
S S2/5
0
The mobility is then

Y = 5 5 /4 = 2 sin2 /5.
(5.36)
This agrees with the result obtained for the nontrivial fixed point by Yi and Kane [6],
implying that our Y -state is identical to their fixed point obtained by a different mapping.
The W -state is obtained by fusion with 1/40-operator in the Potts sector:
2/5
S1/40 S00
3 5
AW
.
(5.37)
= 2/5 0 = +
AD
2
S0 S1/40
Thus W = ( 5 1)/4. This is equal to 1 Y ; this is related to the duality between the
Y -state and W -state.
Next let us consider the fusion in the Ising/tricritical Ising sector. Since the two
components of the currents are related to the Ising/tricritical Ising sector in an asymmetric
manner, in general the amplitude Aab is asymmetric (Axx 6= Ayy ) although it is diagonal
(Axy = 0). For example, the U -state is obtained from D-state by fusion with the 1/16
primary in the Ising sector. Thus, the coefficients are given by
AU
xx =
0
S1/16
S00
0
S00 S1/16
AD
xx = 1,
(5.38)
AD
yy = 1,
(5.39)
1/2
AU
yy
S1/16 S00
1/2
S0
0
S1/16
where S represents the matrix of modular transformations of the Ising model. As a result,
U
U
the mobility in the U -state is anisotropic: U
xx = 0, yy = 1 (xy = 0). This represents
perfect mobility in y-direction and complete localization in the orthogonal x-direction.
This would naturally correspond to a mixed b.c. of D and N . Indeed, this is the case as
we demonstrate in the following.
A similar calculation gives the mobility for the S-state as Sxx = 1, Syy = 0 (Sxy = 0).
One might think that U and S are equivalent upon a space rotation, because their mobilities
are the same if we exchange x and y directions. However, they are not equivalent because
the underlying lattice is not invariant under /2 rotations. The inequivalence can be also
seen in the following construction of the mixed D/N boundary state.
592
As discussed in Section 3.2, the mixed D/N state can be constructed from (2.63) with
an orthogonal matrix R. Let us choose

1 0
R=
.
(5.40)
0 1
This would correspond to D (localized) in x-direction and N (free diffusion) in y, as
in the U -state. Then the allowed zero modes are given by the integer-coefficient linear
combinations of

1
(5.41)
, 0 , (0, 0) ,
[E
u1 , vE1 ] =
3

(5.42)
[E
u2 , vE2 ] = (0, 0), 0, 2 .
Thus, the new lattice e introduced in Section 3.2 is a rectangular lattice with the unit cell
of the size 1 1 . The diagonal amplitude for this state is given by
3

1 2 X vE2/(2)
q
.
(5.43)
Z(q) =
(q)
vEe
This actually agrees with the diagonal amplitude ZU U (q), implying that the U -state is
identical
to the constructed mixed D/N state. Of course, the g-factor also agrees: g =
p
V0 (e) = 31/4 = gU .
On the other hand, the S-state would correspond to

1 0
R=
.
(5.44)
0 1
In this case, the bases for the allowed zero-modes is given by

1
0, , (0, 0) ,
[E
u1 , vE1 ] =

[E
u2 , vE2 ] = (0, 0), 2 3, 0 .
(5.45)
(5.46)
This is inequivalent to the previous case (5.40), because the lattice is not invariant
under a rotation by /2. Now the new lattice e p
is again a rectangular lattice, but with
a different size 1/ 3/. This gives g = V0 (e) = 31/4 = gS . Moreover, its

diagonal partition function agrees with ZSS , implying that the S-state is in fact the mixed
D/N state.
We can construct other mixed D/N states, by considering other (infinitely many)
possible rotation matrices R. The fusion construction based on the present conformal
embedding does not give those states. It does not mean that these states do not exist. It
rather means that, only a part of the possible b.c.s can be accessible by the conformal
embedding which reduces the number of conformal towers effectively. In general, we do
not know how far we can reach by a particular conformal embedding. From physical
considerations, however, we think that the fixed points appearing in the QBM model
introduced in Section 4 have been exhausted.
The T -state is obtained from the D-state by fusion with 3/80 primary in the tricritical
Ising sector. The coefficients are given by
593
3 5
,
= 3/5 0
=
2
S0 S3/80
1/10
S3/80 S00 D
3 + 5
.
= 1/10 0 Ayy =
2
S0 S3/80
3/5
ATxx
ATyy
S3/80 S00
AD
xx
(5.47)
(5.48)
Thus the mobility is given by Txx = Y and Tyy = W = 1 Y . The T -state seems to
be a mixture of Y and W boundary states.
W
R
Y
By a similar calculation, we find the mobility in the R-state as R
xx = , yy = .
Namely, the mobility in the R-state is equivalent to the T -state after /2 rotation in the
xy-plane. They are again not equivalent, presumably reflecting the fact that the lattice is
not invariant. T - and R-states are apparently given by taking dual of Y or W for only one
component of the boson. However, we do not know how they can be realized in context
of QBM. Possibly they correspond to some anisotropic QBM. Below we summarize the
properties of all the boundary states obtained by fusion in Potts, Ising or tricritical Ising
sectors.
Boundary state
D
N
Y
W
R
g-factor
p
1/ 2 3
q
3/2
p
2 cos (/5)/ 2 3
q
2 cos (/5)
3/2
[3(7 + 3 5)/2]1/4
31/4
(7 + 3 5/6)1/4
1/31/4
xx
yy
(5 5)/4
(1 + 5)/4
(1 + 5)/4
(5 5)/4
(1 + 5)/4
(5 5)/4
(5
5)/4
0
5)/4
(1 +
In the embedding using Z3(5) , the level-1 current states corresponds to the product state
(5)
| 35 i5 | 25 iP , where | 35 i5 is the primary state of the Z3 theory with weight 3/5. In fact, there
(5)
are two such primary states in the Z3 theory, which are complex conjugates. Let us denote
one of them as | 35 i. In this approach, the natural identification would be

1
(Jx,1 + iJy,1)|0i = 35 5 25 P ,
2

1
(Jx,1 iJy,1)|0i = 35 5 25 P .
2
(5.49)
(5.50)
Assuming this correspondence, for a boundary state constructed from the D-state by fusion
with the Z3(5) operator , we find
Aab = ab Re A ab Im A,
(5.51)
594
where ab is the antisymmetric tensor xy = yx = 1, xx = yy = 0. Here A is defined
as
3/5
A=
S S00
3/5
S0 S0
(5.52)
where Sji is the matrix of modular transformations of the Z3(5) theory. Thus, when the
matrix of modular transformations acquires an imaginary part, there is a nonvanishing offdiagonal mobility. This corresponds to the Hall effect. Physically, such an effect is expected
in QBM under a magnetic field.
The Y -state is obtained by fusion with the 8/5-operator in the Z3(5) sector. Thus
3/5
S8/5 S00
3 5
.
(5.53)
AY = 3/5 0 =
2
S S8/5
0
This gives the same result as we have obtained by fusion in the Potts sector. The offdiagonal mobility vanishes in this case.
On the other hand, a similar calculation on the F -state reads
3/5
S1/9 S00
1 3 i
.
(5.54)
AF = 3/5 0 =
2
S S1/9
0
This complex amplitude gives a nonvanishing off-diagonal mobility:
3
3
xy = yx =
.
(5.55)
xx = yy = ,
4
4
By a fusion with the conjugate operators, we can also obtain the boundary state with
opposite sign of the off-diagonal mobility.
The other three states X, V and Z also exhibit the nonvanishing off-diagonal mobility.
Thus they should also correspond to a critical behavior under a magnetic field. The mobility
in these states is summarized as follows.
Boundary state
g-factor
q
F
X
V
Z
2
3
1/4
[ 14
3 + 2 5]
[6(7 + 3 5)]1/4
p
2 3
xx
xy
3
4
1+ 5
8
7 5
8
1
4
43
15
3 3
8
15
3 3
8
43
5.6. Structure of the nontrivial boundary state

We have obtained the explicit expression of the amplitude for the boundary state Y . It
gives some insight into the structure of the nontrivial boundary state.
595
In general, a conformally invariant boundary state should be a linear combination of

Ishibashi states. Each Ishibashi state is constructed from a spinless primary field that
appears in the bulk theory. Thus, the closed string channel amplitude (5.18) must be
consistent with the bulk spectrum.
In c = 1 free boson theory with a generic compactification radius, each bosonic Fock
space build on a vacuum corresponds to an irreducible representation of the Virasoro
algebra (except for zero winding-number sector). However, in c = 2, a bosonic Fock space
is always reducible and contains an infinite number of irreducible representations of the
Virasoro algebra. Thus, there are infinitely more Ishibashi states other than those built
on a bosonic vacuum. Nevertheless, the c = 2 Dirichlet and Neumann boundary states
have simple expressions in terms of bosons, and their (self-)amplitudes are written as
P h
2 . Roughly speaking, the factor 1/(q)
2 means the boundary state is made
h q /(q)
of whole bosonic Fock space.
However, the nontrivial boundary state is not made of such bosonic boundary states.
This can be seen as follows. Expressing the diagonal amplitude (5.18) as a sum of the
2 of the Heisenberg algebra,
character q h /(q)

1/6 9(3 5) 1/2
gY 2
q + 16 6 5 q 2/3
=
ZY Y (q)
1 + 21 9 5 q +
(q)
2

7/6

+ 5 3 5 q + 9 3 5 q + 3 7 3 5 q 3/2

5/3
+ 5 3 5 q + .
(5.56)
This involves negative coefficients at least for q and q 5/3 . While any given partition
function can be written as an infinite sum of Heisenberg characters, the coefficients reveal
the nature of the boundary state. If the boundary state is a linear combination of the bosonic
boundary states, all the coefficients must be positive.
Thus, this is a highly nontrivial boundary state which cannot be constructed by a
generalization such as Eq. (2.63) or as in Ref. [29] of the Dirichlet or Neumann. To our
knowledge, this is the first proof that such a nonbosonic boundary state does exist in a free
boson field theory. Similar nontriviality can be shown for W , R, T , X and V states.
On the other hand, since the boundary state is a linear combination of Ishibashi states,
the diagonal amplitude must be a linear combination of c = 2 Virasoro characters with
positive coefficients. In fact, the amplitude (5.18) is expressed as

2
2
= 3 + 5 02 + 6 3 5 1/6
+ 181/2
4 3 ZY Y (q)
2
+ 18 2 5 2/3
+ ,
(5.57)
with positive coefficients, where the c = 2 Virasoro characters for weight h are given by
Q
Q
n
2
1/12 q 11/12)/ (1 q n ) for
h2 = q h1/12/
n=1 (1 q ) for h > 0 and 0 = (q
n=1
h = 0. Moreover, all the primary weights correspond to those of spinless fields in the bulk
spectrum. These are guaranteed by the fusion construction.
Thus this nontrivial boundary state is certainly a consistent boundary state for the
c = 2 free boson theory, although it appears to be very complicated in terms of the
596
bosons. A possible direction to extend the construction of the nontrivial state for different
lattices is to write down the boundary state in the bosonic language and then guess a
generalization of the boundary state for other lattices. However, so far we have been unable
to develop this approach.
6. Integrable field theory analysis

6.1. Integrable flows away from the perfect mobility (Neumann) fixed point
Here we consider the Hamiltonian of Eq. (4.32) for small v1 , v2 , v3 . This model
turns out to be integrable for general values of v1 , v2 , v3 as discussed in detail in
Ref. [43]. This integrability is, in a distant sense, related to the existence of the conformal
embedding; more precisely, it is the underlying parafermionic algebra satisfied by the
vertex operators of (4.17) at that particular radius that ensures the existence of nonlocal
conserved currents [43].
For general values of v1 , v2 , v3 the scattering theory describing (4.32) is simply made
up of three left moving and three right moving massless particles with mass parameters
m1 , m2 , m3 , energy and momenta being parametrized by rapidities e = p = mj e .
Scattering between left and right movers is trivial since the theory is scale invariant in the
bulk, while scattering among left or among right movers is described by pure CDD factors:
Sk,k1 = i tanh( 2 i
4 ), S1,3 = 1. Only the first particle then scatters nontrivially on the
i
B
impurity, with a boundary R matrix again given by a CDD factor, R = i tanh(
2 4 ).
3
Here, B parametrizes the impurity energy scale TB e B vi . The exact dependence of

TB , as well as the dependence of the mass parameters upon the vi s are complicated and
will not be needed in the following, except for special cases.
The main result of the integrable approach is that the impurity free energy can be
computed via the thermodynamic Bethe ansatz (TBA) [43]. If we parametrize the filling
1
fractions of the particles by pseudo energies j as fj =
/T , the latter are solutions of
1+e j
a set of integral equations
Z
X

1
d 0
0
ln 1 + ek ( )/T .
Nj k
(6.1)
j = mj e T
2 cosh( 0 )
k
Here, j, k take values 1, 2, 3 and can be represented as the nodes of an A3 Dynkin diagram
with incidence matrix Nj k (crossed nodes indicate the existence of a source term in the
TBA equations)
m1
N
m2
N
m3
N
The impurity free energy reads then

Z

1
d
ln 1 + e1 ()/T .
(6.2)
Fimp = T
2 cosh( B )
Formula (6.2) gives us quick access to the change of boundary entropies between the
UV fixed point (Neumann boundary conditions, T /TB = ) and the possible infrared
597
(IR) fixed points (T /TB = 0) our model can flow to. Rather than entropies, we will use the
ground state degeneracy, g = eS/T . The change of g-factor is then neatly expressed as

1 + e1 ()/T 1/2
gUV
=
.
(6.3)
gIR
1 + e1 ()/T
When , the source terms just drop from the equations, the s go to constants that
obey the system, setting xj = ej ()/T ,
Y
xj = (1 + xk )Njk /2 .
k
This is solved right away by x1 = x3 = 2, x2 = 3.

Setting similarly yj = ej ()/T , the system obeyed by the yj s depends on the source
terms: different possibilities can arise. If m1 6= 0, the source term for 1 diverges, and
therefore y1 = 0. If m1 = 0 and m2 6= 0, in the IR the system is simply 1 = 0 so y1 = 1.
Finally, if m1 = 0 and m2 = 0, the IR system is
y1 = y2 = (1 + y1 )1/2
with solution y1 = 1+2 5 . We thus obtain the possible ratios of degeneracies

gUV
= 3,
m1 6= 0
gIR
r
3
gUV
,
m1 = 0, m2 6= 0
=
gIR
2
3
gUV
=
, m1 = m2 = 0
(6.4)
gIR
2 cos 5
q
with 2 cos 5 = 3+2 5 .
We obtained the first ratio in Sections 3 and 4 when discussing the flow from N boundary
conditions to D boundary conditions (i.e., freely diffusing to localized) in the case v1 =
v2 = v3 > 0. For general vi , the potential of Eq. (4.32) has a unique minimum (within each
unit cell) and the TBA equations in the first case describe the flow from N to D in the
general case where the three-fold symmetry is broken. This is confirmed by the analysis of
the central charge associated with (6.1) which gives generically c = 2, hence indicating a
flow within the whole two boson theory (i.e., not purely within the Potts sector). The third
ratio was also obtained previously when we discussed the flow from N to Y in the case
v1 = v2 = v3 < 0. Since this ratio is obtained only when m1 = m2 = 0, it follows that the
flow from N to Y occurs only in the three-fold symmetric case. Meanwhile, the analysis
of the central charge associated with (6.1) gives only c = 45 , indicating that the flow takes
place within the Potts sector of the theory only.
Finally the second ratio is somewhat trivial: it corresponds to a flow from Neumann to
Dirichlet boundary conditions for the field 1 while the field 2 remains at the Neumann
fixed point. This corresponds to the U boundary state discussed in Section 4. This is
obtained by setting v2 = v3 = 0, while v1 can take either sign. The analysis of the central
charge associated with (6.1) gives c = 1, confirming that the flow takes place in the c = 1
sector only.
598
6.2. W3 integrable flows

Integrable flows can be used to explore more thoroughly the phase space and the possible
fixed points in our problem.
The integrability of the flow from free to fixed or mixed boundary conditions in the Potts
model, and therefore, from Neumann to Dirichlet or the Y fixed point in the c = 2 theory,
is ultimately related with the fact that the Z3 parafermionic theory (the three state Potts
model) can be represented as the coset SU(2)3 /U (1). The field 1 is then the adjoint
operator in this construction. 2 We observe that this theory can also be represented as the
coset SU(3)1 SU(3)1 /SU(3)2 . The adjoint operator for this coset now has dimension
= 25 , and physically corresponds to the energy operator of the Potts model. It can be
shown to define an integrable perturbation that preserves the W3 symmetry: more precisely,
there is a natural deformation of the current W3 that is conserved, at least perturbatively,
away from the UV fixed point, if this fixed point itself preserves W3 symmetry. It is natural
to assume that this extends all the way to the IR fixed point, which therefore should have
W3 symmetry too.
In our problem, the operator with = 25 is present in the spectrum for mixed boundary
conditions in the Potts model, or, equivalently, at the Y fixed point in the c = 2 theory. In
the Potts model it corresponds to applying a magnetic field at the mixed (AB) fixed point
which breaks the remaining Z2 symmetry and favors the A state. This corresponds to a
flow from mixed to fixed. In the QBM problem we expect this operator to produce an RG
flow from Y to localized fixed points; if Y corresponds to the particle being on A and B
sublattice then the flow is to a state where the particle is localized on an A site. This RG
flow can also be studied using integrability.
For the more general coset SU(3)1 SU(3)k /SU(3)k+1 , perturbed by the operator
3
, which preserves W3 symmetry, S matrices and R matrices
of weight = 1 k+4
are known [45], and the impurity free energy can be computed once again using the
thermodynamic Bethe ansatz. It reads
Z

d
G1 ( B ) ln 1 + e11 /T
Fimp = T
2

(6.5)
+ G2 ( B ) ln 1 + e21 /T .
Here, as before, TB is an impurity energy scale, TB eB , G1 and G2 are known kernels,
and the are pseudo energies, solutions of the equations:
X

Nij,ik G1 ln 1 + eik /T
ij = j 1 me T
T

Nij,lj G2 ln 1 + elj /T .
(6.6)
Here, ij are line and column labels on the following diagram (the cross on the (1, 1) node
stands for the mass term in (6.6))
2 In a general coset construction, one usually calls adjoint [44] the operator obtained by taking the identity
representation for the algebras in the numerator, and the adjoint for the algebra in the denominator.
1

N
...
...
599
i = 1, 2, j = 1, . . . , k, and Nij,kl its incidence matrix.

We will not need the exact form of the kernels in what follows, and shall concentrate
on the ground state degeneracies in the UV and in the IR. From the previous formulas, we
find the simple result (the yj 1 all vanish here)

gUV 1
= ln(1 + x11 ) + ln(1 + x21 ) ,
(6.7)
gIR
2
where xij = eij /T is solution of the limit of the TBA system:

Y
Y
1 Nij,lj /2
.
1+
xij = (1 + xik )Nij,ik /2
xlj
k
(6.8)
For the case k = 1 of interest here, the system reduces to x = x11 = x21 , x = x or
1+x
x = 2 cos 5 . Hence
gUV
(6.9)
= 2 cos .
gIR
5
This is the ratio gmixed /gfixed in the Potts model, or the ratio gY /gD for the two boson
system. We thus expect that perturbation by the operator = 25 does not lead to a new
fixed point, but induces a flow from Y to D in the two boson system. Let us stress that
this is a rather abstract statement: although we know the operator with = 25 is present in
the spectrum (because of the partition functions analysis), we are not able to represent it
in terms of bosonic operators, since we do not have an entirely clear picture of what the Y
boundary conditions mean for the bosons.
7. Generalization to higher dimensions and relations with the Kondo model

7.1. Generalization to higher dimensions
Many of our results can be generalized to the case of n 1 bosons, corresponding to
QBM on an (n 1)-dimensional lattice. Yi and Kane found a generalization of the Y
fixed point for all n > 3 using a mapping onto the n-channel SU(2), overscreened Kondo
model. They also observed that the 3-dimensional (n = 4) case corresponds to QBM on
a diamond lattice. The conformal embedding is now naturally understood in terms of an
SU(n) spin chain. We may decompose the product of left and right moving SU(n)1
factors into SU(n)2 a Zn parafermion. We may then divide out the maximal abelian
subalgebra, U (1)n1 from SU(n)2 . Taking into account that SU(n)1 is equivalent to n 1
free bosons, we see that we have constructed a conformal embedding:
2(n 1) n(n 1)
+
.
(7.1)
n1=
n+2
n+2
600
The first component of the embedding is the theory of Zn parafermions, that can also
be considered as the coset SU(2)n /U (1), with c = 2(n 1)/(n + 2). The fundamental
parafermion in this theory has weight (n 1)/n, and can be bosonized using a system of
n bosons satisfying (for the chiral components)

n1
ln(z w),
i (z)i (w) = 2
n
2
i (z)j (w) = ln(z w), i 6= j,

n
(7.2)
as
1 =
n
X
eik .
(7.3)
k=1
The embedding then follows from the decomposition T = T1 + T2 , where

"
#
n
X
1X
1
2
i(j k )
(j ) +
e
,
T1 =
n+2
2
j 6=k
j =1
"
#
n
X
X
n
1
(j )2
ei(j k )
T2 =
n+2
4
j =1
(7.4)
j 6=k
such that the short distance expansion of T1 with T2 is trivial, T1 is a stress energy tensor
with central charge c1 = 2 n1
n+2 of the coset SU(n)1 SU(n)1 /SU(n)2 or SU(2)n /U (1), and
n1 coset. T = T + T of course is
similarly for T2 with c2 = n(n1)
1
2
n+2 of the SU(n)2 /U (1)
the stress energy tensor of the initial (n 1) boson theory.
It turns out that the n 1 boson theory with boundary perturbation
n
X
vj cos j ,
(7.5)
j =1
is integrable [43], the particular case where all the vj s are equal corresponding again to a
perturbation of the form 1 + 1 . The impurity free energy reads, generalizing (6.2)
Z

1
d
ln 1 + en ()/T ,
(7.6)
Fimp = T
2 cosh( B )
where the satisfy the same TBA equations as (6.1), but the incidence matrix is the one of
the following diagram
N
n1
1
N
2
N
nN
3
n2
N
Here again, TB vjn , and the
mj are dimensionless numbers functions of the couplings vj .
601
In the UV, the system of equations for the xs is solved easily with xj = (j + 1)2 1,
j = 1, . . . , n 2; xn1 = xn = n 1. If mn does not vanish, one has yn = 0, and then we
obtain the ratio
gUV
= n.
(7.7)
gIR
This generalizes the first ratio in (6.4), and corresponds to a flow from Neumann to
Dirichlet boundary conditions in the n 1 bosons problem, where at the D fixed point,
the field lies on a (n 1)-dimensional lattice generalizing .
When mn vanishes, the ratio of degeneracies can take various values depending on which
of the other mj s vanish. A particularly interesting case is when all mj but mn1 vanish.
This corresponds to a perturbation with all aj = 0, j 6= 1, and a1 > 0. In that case, one gets
a ratio
n
gUV
=
(7.8)
,
gIR
2 cos n+2
generalizing the case n = 3; we denote the corresponding fixed point by Yn . These fixed
points were discovered by Yi and Kane [6].
Let us discuss the other cases. If mn1 still does not vanish, and in addition some of the
),
other mj s do not vanish (j 6 n 2), the ratio is of the form gUV /gIR = n/(2 cos n+2k
indicating a flow to a fixed point encountered previously for a lower value of n, Ynk . When
mn1 does vanish,
qwhile some of the other mj s do not (j 6 n 2), one gets a ratio of the
n
. This corresponds to an IR fixed point where k components of the
form gUV /gIR = nk
field have D boundary conditions, the other ones having N. We see therefore that, as in the
n = 3 case, no other fixed point is reached, besides the various D, N combinations, and the
Yk fixed points.
As before, the perturbation of N boundary conditions by the adjoint operator in the
coset SU(2)n /U (1) does not preserve the Wn symmetry. A perturbation preserving this
symmetry is the one with the adjoint in the coset picture SU(n)1 SU(n)1 /SU(n)2 . The
n
, and the perturbation is integrable. The
perturbing field now has dimension = 1 n+2
boundary free energy reads as in (6.5), with now a sum over n 1 pseudo energies. The
TBA diagram is as in the SU(3) case, but the base is the Dynkin diagram of SU(n), i.e.,
An1 instead of A2 . For more general cosets SU(n)1 SU(n)k /SU(n)k+1 , the diagram
looks as follows
...
k
n1
...
...
with the obvious generalization for Eqs. (6.6) and (6.5) that the row labels now run from 1
to n 1. In particular the ratio of boundary entropies in the UV and IR now reads
602
gUV 1 X
=
ln(1 + xi1 ).
gIR
2
n1
(7.9)
i=1
For k = 1 we have

Y
1 Ni1,j1 /2
.
1+
xi1 =
xj 1
(7.10)
The solution of this system is

sin j /n + 2 2
1
=
1.
xj
sin /n + 2
By some simple manipulations one finds then
gY
gUV
.
=
= 2 cos
gIR
gD
n+2
(7.11)
This corresponds to a flow from Yn to D in the n 1 bosons language, generalizing (6.9).

As in the n = 3 case, the key ingredient to understand the phase diagram is the set
of boundary conditions and flows in the Zn parafermions theory SU(2)n /U (1). These
theories have known lattice model or quantum spin chains realizations [47], and it is
possible to show, using results in Ref. [48], that the N and D fixed points correspond
to free and fixed boundary conditions, while the operator 1 + 1 is the most relevant
order parameter, and corresponds again to a longitudinal boundary field. The generalization
of the mixed boundary conditions for the (Z3 ) Potts model to the Zn case has not been
investigated, to our knowledge.
7.2. Relation to the Kondo problem
The emergence of parafermions is also closely related with the n-channel Kondo
problem. Recall [11] that this problem involves originally 2n Dirac fermions, which can be
bosonized in terms of the current algebras SU(2)n SU(n)2 U (1). Only the spin currents
interact with the boundary spin, leading to a flow entirely within the SU(2)n sector, from
the weak coupling fixed point in the UV to the non-Fermi liquid Kondo fixed point in the
).
IR. The ratio of degeneracy factors is gUV /gIR = 2/(2 cos n+2
As is well known, the strong coupling Kondo fixed point does not depend on the
anisotropy, and can be reached as well starting from an anisotropic Kondo interaction.
There is a particularly simple choice of this anisotropy which corresponds to the Toulouse
limit in the n = 1 case, or the Emery Kivelson limit in the n = 2 case, where an additional
U (1) decouples, and the flow takes place entirely within the SU(2)n /U (1) sector, i.e., the
parafermionic theory. The perturbation Hamiltonian then reads

(7.12)
Hint = v 1 + 1 + .
One can then argue from the integrable analysis that the IR fixed point the strong
coupling Kondo fixed point coincides with the Yn fixed point identified previously,
the generalization of the mixed boundary conditions of the Potts model to the Zn case.
603
The UV fixed point the weakly coupled Kondo fixed point does not have a simple
interpretation in terms of the Zn variables. For the case n = 3, it coincides with the
localized on B or C fixed point discussed in Section 4. This correspondence was used
extensively by Yi and Kane [6].
7.3. Another relation to the Kondo problem: a remark on the marginal case
Returning to the two boson model (n = 3), we note that the case a 2 = 8
9 can also be
studied by integrability techniques, using an unexpected mapping onto the four-channel
Kondo model. To do this, recall that it is possible to represent the currents of the SU(2)4
algebra (that has c = 2) with two bosons as follows: 3

Jx = x cos 2 1L + 6 2L ,

Jy = y cos 2 1L 6 2L ,

(7.13)
Jz = z cos 8 1L .
A similar representation was used by Fabrizio and Gogolin [46] recently; these authors
however did not explicitly consider the operators in their representation (cocycles),
which will play a crucial role in what follows, though it does not affect their results. The
four channel anisotropic Kondo problem has a boundary interaction

(7.14)
Hint = J x Jx (0) + y Jy (0) + J// z Jz (0),
and is integrable for any value of the anisotropy.
Now let us use our bosonization. The boundary action reads then, using the boundary
conditions iL = iR in the UV to introduce the nonchiral fields,
r
r
r
r

3

3
1 +
2 (0) + y y cos
1
2 (0)
J x x cos
2
2
2
2

(7.15)
+ J// z z cos 2 1 (0) .
Consider now the expansion of the boundary free energy in powers q
of J , J// . For every
q
insertion of cos 2 1 , we need one insertion of cos 2 1 + 3

2 2 and one of
q
q
cos 2 1 3
2 2 , contributing a term x x y y z z = 1 to the spin part. It follows
that all the spin terms actually disappear, and that the boundary free energy is the same as
the one of the model (7.15) with no spin variables:
r
r

3
(7.16)
1 cos
2 J// cos 2 1 .
Hint = 2J cos
2
2
This is nothing but the weak corrugation Hamiltonian of Section 4 (Eq. (4.17)), this time
for a 2 = 8
9 , where the perturbation is marginal.
3 H.S. thanks P. Fendley and N. Warner for discussions on this point.
604
Using the TBA for the anisotropic Kondo model, the impurity free energy is simply
expressed as
Z

d
1
ln 1 + e1 ()/T ,
(7.17)
Fimp = T
2 cosh( B )
where TB eB is the Kondo temperature, and 1 is a pseudo energy, solution of the TBA
system
Z
X

1
d 0
0
ln 1 + ek ( )/T .
Nj k
(7.18)
j = m 4,j e T
2 cosh( 0 )
k
Here, the labels k run over the nodes of the diagram

t 1
1

4
N
t 3

t 2
t
and we have restricted to the simplest values of the anisotropy such that t 1
t = 1 J// .
Provided Jk > 0, the perturbation generates a flow (the situation here is quite different
from the one boson theory, where the boundary cosine interaction is truly marginal).
Independently of the value ofJk , one finds x1 = e1 ()/T = 3, while y1 = e()/T
= 2. The ratio gUV /gIR = 2/ 3, the well known value for the 4 channel Kondo model.
This coincides with the ratio gN /gD (4.30) for our values of the coupling constant. For
Jk 6 0 (the equivalent of vA < 0 in Section 4), meanwhile, the Kondo perturbation is
irrelevant, and no flow is generated. Hence, we conclude that for N = 1 (4.19), there is
no new fixed point, and presumably the Y fixed point becomes identical with the N one in
the limit a 2 = 8/9. An -expansion around this marginal point was developed by Kane
and Fisher [8].
8. Conclusion
We have established the general connection between a simple model for QBM and
boundary CFT with considerable attention paid to the issue of boson compactness. For
a simple Hamiltonian we have conjectured a complete phase diagram, which is analogous
to that of the boundary Potts model. These results were obtained by using a sort of
boundary embedding involving the Potts model and the fusion approach to generating
boundary states. We have also mentioned other fixed points which occur with more general
Hamiltonians and which arise from fusion in the other sectors of the conformal embedding.
We have discussed the integrability of some of the RG flows.
We note that the Y b.c. is more dynamical or quantum than D or N. Within the weak
corrugation formulation, while the potential term vA of Eq. (4.17) is relevant, it does not
605
simply pin the boson fields at one of its minima; rather they fluctuate between the two
inequivalent minima. This is only possible for a range of lattice spacing, 8/9 < a 2 < 2
where a semiclassical analysis fails.
We have given an explicit demonstration that other, highly nontrivial b.c.s are possible
for 2 free periodic bosons besides D and N and their variants. However, since this
demonstration rests on the conformal embedding and fusion, and since we do not have
a conformal embedding for general values of the radius parameter, a, our construction
only works at one special value of a. A more straightforward understanding of these b.c.s,
directly in terms of the bosons eludes us. Thus this general problem, known since 1992
[7,8] remains open.
Acknowledgements
We would like to thank M.P.A. Fisher, A.W.W. Ludwig, C. Schweigert, A. Zawadowski
and J.B. Zuber for helpful discussions. This research was begun while all three authors
were visitors at the Institute for Theoretical Physics, Santa Barbara, in July, 1997. H.S.
also acknowledges the hospitality of the SPhT Saclay. This research is supported in part
by the NSF under grant No. PHY-94-07194, NSERC of Canada (I.A.), Grant-in-Aid from
Ministry of Education, Science, Sports and Culture of Japan (M.O.), and by the DOE and
the NSF under the NYI program (H.S.).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
A. Schmid, Phys. Rev. Lett. 51 (1983) 1506.

F. Guinea, V. Hakim, A. Muramatsu, Phys. Rev. Lett. 54 (1985) 263.
M.P.A. Fisher, W. Zwerger, Phys. Rev. B 32 (1985) 6190.
A.O. Caldeira, A.J. Leggett, Ann. Phys. 149 (1983) 374.
C. Callan, L. Thorlacius, Nucl. Phys. B 329 (1990) 117.
H. Yi, C.L. Kane, Phys. Rev. B 57 (1998) R5579.
A. Furusaki, N. Nagaosa, unpublished;
A. Furusaki, N. Nagaosa, Phys. Rev. B 47 (1993) 4631.
C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233.
J.L. Cardy, Nucl. Phys. B 324 (1989) 581.
I. Affleck, M. Oshikawa, H. Saleur, J. Phys. A 31 (1998) 5827.
For a review see, I. Affleck, Acta Physica Polonica 26 (1995) 1869, cond-mat/9512099.
K. Itai, Phys. Rev. Lett. 58 (1987) 602.
G.T. Zimnyi, K. Vldar, A. Zawadowski, Phys. Rev. B 36 (1987) 3186.
For a one-dimensional model of a heavy particle in a Luttinger liquid see, A.H. Castro Neto,
M.P.A. Fisher, Phys. Rev. B 53 (9713) 1996.
C. Callan, L. Lovelace, C. Nappi, S. Yost, Nucl. Phys. B 293 (1987) 83.
V.A. Fateev, A.B. Zamalodchikov, Nucl. Phys. B 280 (1987) 644.
N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 251;
T. Onogi, N. Ishibashi, Nucl. Phys. B 318 (1989) 239.
V.G. Kac, M. Wakimoto, Adv. Math. 70 (1988) 156.
J.L. Cardy, Nucl. Phys. B 270 (1986) 186.
606
[20] G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 356 (1995) 230.
[21] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Phys. Lett. B 444 (1998) 163;
R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Nucl. Phys. B 579 (2000) 707.
[22] J. Fuchs, C. Schweigert, Phys. Lett. B 414 (1997) 251.
[23] E. Verlinde, Nucl. Phys. B 300 (1988) 360.
[24] F. Lesage, H. Saleur, P. Simonetti, Phys. Lett. B 427 (1998) 85.
[25] L. Chim, J. Mod. Phys. A 11 (1996) 4491.
[26] A. Recknagel, D. Roggenkamp, V. Schomerus, Nucl. Phys. B 588 (2000) 552.
[27] I. Affleck, J. Phys. A 33 (2000) 6473.
[28] I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
[29] C. Callan, I. Klebanov, J. Maldacena, A. Yegulalp, Nucl. Phys. B 443 (1995) 444, and
references therein.
[30] J. Fuchs, C. Schweigert, Nucl. Phys. B 558 (1999) 419;
J. Fuchs, C. Schweigert, Nucl. Phys. B 586 (2000) 543.
[31] R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, Commun. Math. Phys. 123 (1989) 485.
[32] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Berlin, Springer, 1996.
[33] M. Oshikawa, I. Affleck, Nucl. Phys. B 495 (1997) 533.
[34] P. Fendley, F. Lesage, H. Saleur, J. Stat. Phys. 79 (1995) 799.
[35] J. Fuchs, C. Schweigert, Phys. Lett. B 441 (1998) 141.
[36] V.A. Fateev, S.L. Lukyanov, Sov. Sci. Rev. A 15 (1990) 1.
[37] T. Eguchi, K. Higashijima, Prog. Theor. Phys. Suppl. 86 (1986) 192.
[38] P. Griffin, D. Nemeschansky, Nucl. Phys. B 323 (1989) 545.
[39] I. Affleck, M. Oshikawa, H. Saleur, cond-mat/0011454.
[40] E.B. Kiritsis, Phys. Lett. B 198 (1987) 379.
[41] E. Wong, I. Affleck, Nucl. Phys. B 417 (1994) 403.
[42] J.L. Cardy, D. Lewellen, Phys. Lett. B 259 (1991) 274.
[43] H. Saleur, P. Simonetti, Nucl. Phys. B 535 (1998) 596.
[44] T. Eguchi, S.K. Yang, Phys. Lett. B 235 (1990) 282;
C. Ahn, D. Bernard, A. Leclair, Nucl. Phys. B 346 (1990) 409.
[45] C. Ahn, C. Rim, J. Phys. A 32 (1999) 2509.
[46] M. Fabrizio, A.O. Gogolin, Phys. Rev. B 50 (1994) 17732.
[47] V. Fateev, A.B. Zamolodchikov, Sov. Phys. JETP 62 (1985) 215.
[48] C. Vanderzande, J. Phys. A 20 (1987) L549;
H. Saleur, J. Phys. A 22 (1988) L41.

One-point functions in integrable coupled minimal

models
P. Baseilhac
Department of Mathematics, University of York Heslington, York YO105DD, UK
Abstract
We propose exact vacuum expectation values of local fields for a quantum group restriction of
(1)
the C2 affine Toda theory which corresponds to two coupled minimal models. The central charge
of the unperturbed models ranges from c = 1 to c = 2, where the perturbed models correspond to
two magnetically coupled Ising models and Heisenberg spin ladders, respectively. As an application,
in the massive phase we deduce the leading term of the asymptotics of the two-point correlation
functions. 2001 Elsevier Science B.V. All rights reserved.
PACS: 10.10.-z; 11.10.Kk; 11.25.Hf; 64.60.Fr
Keywords: Massive integrable field theory; Coupled minimal models; Perturbed conformal field theory
1. Introduction
The vacuum expectation values (VEV)s of local fields play an important role in quantum
field theory (QFT) and in statistical mechanics [1,2]. In statistical mechanics the VEVs
determine the generalized susceptibilities, i.e., the linear response of a system to external
fields. Furthermore, the VEVs provide all the information about correlation functions in
QFT defined as a perturbed conformal field theory (CFT) that is not accessible through a
direct calculation in conformal perturbation theory [3]. A few years ago, some important
progress was made in the calculation of such quantities in integrable (1 + 1) QFT. In
Ref. [4], an explicit expression for the VEVs of the exponential field in the sinh-Gordon
and sine-Gordon models was proposed. In Ref. [5] it was shown that this result can be
obtained using the reflection amplitude [6] of the Liouville field theory. This method was
also applied in the so-called BulloughDodd model with real and imaginary coupling. It is
known that c < 1 minimal CFT perturbed by the operators 12 and 13 can be obtained
E-mail address: pb18@york.ac.uk (P. Baseilhac).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 2 - 5
608
P. Baseilhac / Nuclear Physics B 594 [FS] (2001) 607624
by a quantum group (QG) restriction of the sine-Gordon [7] and imaginary Bullough
Dodd model [8] with special values of the coupling. The VEVs of primary fields were then
calculated. The same method was applied later to integrable QFTs involving more than
one field. For instance, the VEV for a two-parameter family of integrable QFTs [9] gave
rise to the VEV of local operators in parafermionic sine-Gordon models and in integrable
perturbed SU(2) coset CFT [10]. The VEV of simply-laced affine Toda field theories
(ATFT)s is known for a long time [11] and the case of non-simply laced dual pairs was
recently studied in [12,13] for which a general expression for the VEVs was derived.
Such perturbed CFTs have recently attracted much attention in condensed matter
physics, such as in the context of point contacts in the fractional quantum Hall effect
and impurities in quantum wires [14,15]. In such cases the property of integrability has
provide a non-perturbative answer for experimentally important strongly interacting solid
state physics problems [15,16]. Particularly, on-shell results were obtained using exact
relativistic scattering and related form factor techniques [17,18].
In this paper, we are interested in exact off-shell results for two coupled conformal field
theories 1 for which the inter-layer coupling preserves integrability. The on-shell dynamics
of these models were studied in [1921].
Let us briefly recall the ideas of [1921]. We consider two planar systems corresponding
to two coupled minimal models Mp/p0 which interact through a relevant operator which
leads to an integrable theory. The resulting action can be written:
Z
(1) (2)
12
(1)
A = Mp/p0 + Mp/p0 + d 2 x 12
or
A = Mp/p0 + Mp/p0 +
(1) (2)
d 2 x 21
21 ,
(2)
(1)
(1)
(2)
(2)
(21
) and 12
(21
) as two specific primary
where we denote, respectively, 12
operators of each unperturbed minimal models and where the parameters and
characterize the strength of the interaction. Here, we will be interested in exact one-point
functions in such system.
In Section 2 we introduce the notations and those known results which are useful for
our purpose. In Section 3 using the exact result for the VEV of the C2(1) ATFT derived
(1)
(2)
in [13], we deduce the exact VEV hrs (x)r 0 s 0 (x)i for any values of (r, s), (r 0 , s 0 ) in
the model with action (1). To do it, we relate the parameter in (1) to the masses
(1)
of the particles and we perform the QG restriction of the C2 ATFT with imaginary
0
0
coupling which leads to the model (1). For (r, s) = (r , s ) = (1, 2) this VEV can be
calculated exactly as well as the bulk free energy. The specific case M3/4 + M3/4
(1) (2)
coupled by 12 12 is considered in details. It corresponds to two layer Ising models
coupled by their magnetization operator (1) (2) . The previous approach is also extended
to the model described by action (2). In Section 4, we extract some limited information
about the asymptotics of two-point correlation functions between any pairs of primary
1 In literature, the first example of such integrable coupled models was studied in [19].
609
Fig. 1. Two coupled two dimensional models. Short distance results are obtained by taking the limit
0.
operators which belong to the same or different unperturbed models. More precisely, we
will distinguish four different cases:

(1)
(2)
(x)r 0 s 0 (y) for |x y| 0,
(a): rs

(1)
(x)r(2)
for |x y| ,
(b): rs
0 s 0 (y)

(i)
(i)
(c): rs (x)r 0 s 0 (y) for |x y| 0 and i {1, 2},

(i)
(x)r(i)
for |x y| and i {1, 2}
(d): rs
0 s 0 (y)
which are depicted in Fig. 1. We finally give some numerical results for various examples of
coupled minimal models such as two energy-energy coupled tricritical Ising, two coupled
A5 -RSOS models and two energy-energy coupled 3-state Potts model. Perspective and
conclusions follow in this final section.
(1)
2. Coupled minimal models as restricted C2 ATFT

The ATFT with real coupling b associated with the affine Lie algebra C2(1) is described
by the action in the Euclidean space:

Z
1
( )2 + 0 e2b1 + 0 e2b2 + eb(1 2 ) ,
(3)
A = d 2x
8
where we chose the convention that the length squared of the long roots is four. As the
different vertex operators do not renormalize in the same way, we introduced two scale
parameters and 0 . The fields in Eq. (3) are normalized such that:
hi (x)j (y)i = ij ln |x y|2 .
(4)
This model possess two fundamental particles with mass Ma which depends on one
parameter m:

a
sin
, for a = 1, 2,
(5)
Ma = 2m
H
610
where we introduced the deformed Coxeter 2 number [22] H = h(1 B) + h B with

b2
B = 1+b
2 and h = 4, h = 6 are, respectively, the Coxeter and dual Coxeter numbers. The
exact relation between m,
and 0 was found in [12] and is given by:

1 + b 2 0 1 + 2b2
H

4
((1 B)/H ) (1 + B/H ) 1B
.
(6)
= 2 1B m
(1/H )
In ATFT approach to perturbed CFT, one usually identifies the perturbation with the
affine extension of the Lie algebra G. Instead, the perturbation will be associated here with
the standard (length 2) root of C2(1) . Removing the last term in the action (3) leaves a
D2 = SO(4) = SU(2) SU(2) model, i.e., two decoupled Liouville models. To associate
the two first terms of the C2(1) Toda potential to two decoupled conformal field theories, we
first introduce for each one a background charge at infinity. Then, the total stress-energy
tensor T (z) is written T (z) = T (1) (z) + T (2)(z) where:
1
(7)
T (i) (z) = (i )2 + Qi 2 i , for i {1, 2}
2
ensures the local conformal invariance of the D2 model for the choice Q2 = Q1 =
b + 1/2b. With our conventions, 3 the exponential fields
eai i (x),
for i {1, 2}
(8)
are spinless conformal primary fields of each Liouville model with conformal dimensions

a2
eai i (x) = i + ai Qi
2
and i {1, 2}.
(9)
In particular, the exponential fields e2b1 and e2b2 have conformal dimensions 1. As is
0 )2
can be
well known, the minimal model Mp/p0 with central charge c = 1 6 (pp
pp 0
obtained from the Liouville case. Consequently, the D2 CFT can be identified with two
decoupled minimal models by the substitution:
b i,
0 0 ,
(10)
and the choice:

2
= p/2p0
2 = +
or
2
2 =
= p0 /2p
with p < p0 .
(11)
Similarly, the primary operators of each minimal model Mp/p0 are obtained through
the substitution aj ij for j = 1, 2 in (8). With these substitutions, the conformal
dimension of the perturbing operator becomes:

(12)
pert = ei(1 2 ) = 3 2 1.
2 Differently to the simply laced case for which the mass ratios take the classical values, the mass ratios for
non-simply laced case get quantum corrections.
3 Here, the length of the longest root is chosen to be 4.
611
As long as we consider a relevant perturbation, we are restricted to choose 2 < 2/3. In the
following we will consider only the cases for which this condition is satisfied, in particular
2.
2 = +
(1)
(2)
We define, respectively, {rs } and {r 0 s 0 } as the two sets of primary fields with
conformal dimensions:
(p0 r ps)2 (p p0 )2
, for 1 6 r < p, 1 6 s < p0 and p < p0 .
(13)
4pp0
They are simply related to the vertex operators of each minimal model through the relation:

(1 r)
(i)
(i)1
(1 s),
(14)
(x) = Nrs
exp iirs i (x) with 1rs = 2rs =
rs
2
rs =
(i)
where we have introduced the normalization factors Nrs
for each model. These numerical
factors depend on the normalization of the primary fields. Here, they are chosen in such a
way that the primary fields satisfy the conformal normalization condition:

(i)
1
(i)
(y) CFT =
, for i {1, 2}.
(15)
rs (x)rs
|x y|4rs
(i)
= N (i) (irs ) where:
For further convenience, we write these coefficients Nrs

/2
N (1) () = 0 2 2

(2 2 + 2) (1/2 2 /) (2 2 2 ) (2 1/2 2 ) 1/2
(2 2 2 2) (2 1/2 2 + /) (2 2 ) (1/2 2)
and N (2) () = N (1) ().

For imaginary values of the coupling b = i, the C2(1) ATFT possesses complex soliton
solutions which interpolate between the degenerate vacua. This QFT possesses the QG
(2)
symmetry associated to Uq (D3 ) as we will recall in the next section. In [23] the
S-matrix of the B2(1) ATFT was constructed using the Uq (A(2)
3 ) QG symmetry of this QFT.
In particular, using the bootstrap procedure, the authors deduced the breatherbreather
S-matrix. It was also shown that a breather-particle identification holds by comparing the
S-matrix elements of the lowest-breathers (breathers with lowest mass) with the S-matrix
elements for the quantum particles in real ATFT. Following the conventions of [12], the
(1)
sin(/H ) and M2 = m.
Using
particle spectrum in real B2 ATFT is given by M1 = 2m
the results of [23] we find the identification:

2
,
(16)
with =
m
= 2M sin
4 2
1 2
where we denote M as the mass of the lowest kink. Eq. (16) holds for our case due to the
(2)
identification B2(1) C2(1) and A(2)
3 D3 .
3. Expectation values in coupled minimal models
In Refs. [12,13] we derived the exact VEV G(a) = hexp(a.)(x)i for all non-simply
laced ATFTs using the so-called reflection relations which relate different fields with the
612
same quantum numbers. We refer the reader to these papers for more details. Although
the model (1) is very different from the C2(1) ATFT (3) in its physical content (the
model (1) contains solitons and excited solitons), there are good reasons to believe that
the expectation values obtained in the real coupling case provide also the expectation
values for imaginary coupling. The calculation of the VEVs in both cases (b real or
imaginary) within the standard perturbation theory agree through the identification b = i.
Following the analysis done for the BulloughDodd model [5], it is then straightforward to
obtain the VEV of primary operators which belong to different minimal models. With the
substitutions (10) one gets G() = G(i):

(1+) (1 +2 )

2
1
1
2
0
G() =
1+
1+
(1+) (1 2 )
2

2
1
1
0
1+
1+
"Z
#
(, t)
dt
2 2t
e
exp
(17)
t sinh((1 + )t) sinh(2t ) sinh((4 2 )t)
0
with

(, t) = 2 sinh (1 2 )(1 + )t sinh (1 2 )(1 + ) + 2 2 t

+ sinh2 (1 + 2 )(1 + )t sinh(t) cosh(t )

+ sinh 22 (1 + )t sinh 22 (1 + ) 2 t

+ sinh 21 (1 + )t sinh 21 (1 + ) + 2 t sinh (1 )t .
As usual we denote (x) = (x)/ (1 x). The integral in (17) is convergent if:
2
2
< (1 + 2 ) <
(1 + )
(1 + )
and
1 < (1 2 ) <
3
,
(1 + )
(18)
and is defined via analytic continuation outside this domain. In (17) we defined 1 and
2
2 , the conformal dimensions in the imaginary coupling case by 1 = (ei1 1 ) = 21 +

1 1
2 ( ) and similarly for 2 with the change 1 2 .
If we want to express the final result for the VEV in terms of the parameter in the
(1)
action (1), we need the exact relation between and the parameters , 0 in the C2
ATFT with imaginary coupling. We obtain:
=

0
4 2 2 1 2 2 ,
2
2
(4 1)
(19)
which corresponds to = N (1) ()N (2) ().

Like any other ATFT, the model (3) for imaginary values of the coupling has non-local
Sk } for k = 0, 1, 2 generated, respectively, by the purely chiral
conserved charges {Qk , Q
and anti-chiral components:
Jek (z) = e
i e
k .(z)
e .(
z)
and Jek (z) = e k
,
(20)
613
where the fundamental vector field (z, z ) = (z) + (

z) and {e
k }, k = 0, 1, 2, is the set
of dual simple roots of the non-simply laced affine Lie algebra C2(1) . We also define the
topological charge:
Z
(21)
d 2 x e
Hk =
k .x (x, t).
2i
Using the equal-time braiding relations for all x, y:
Jek (x, t)Jel (y, t) = qlakl Jel (y, t)Jek (x, t)
with ql = e
2i
|ek |2 2
(22)
where akl = 2ek .el /|ek |2 denotes the extended Cartan matrix of C2(1) , one can show that
Sk , Hk } for k = 0, 1, 2 satisfy the quantum universal enveloping algebra
the charges {Qk , Q
Uq (D3(2) ) with deformation parameter q q0 = e 2 2 . If we express these generators in

terms of the standard Chevalley basis {Ek+ , Ek , Hk } by:
Qk Ek+ q Hk
Sk E q Hk ,
and Q
k
for k = 0, 1, 2,
(23)
then we have:

[Hk , Hl ] = 0,

Hk , El = al,k El ,

q 2Hl q 2Hl
.
Ek+ , El = kl
ql ql1
(24)
The Uq (D3(2) ) has two subalgebras Uq (D2 ) as well as Uq 2 (C2 ) [23] where the
S0 , Q2 , Q
S2 , H0 , H2 }. As was found in [23] the
subalgebra Uq (D2 ) is generated by {Q0 , Q
S-matrix in the unrestricted form acts as an intertwinner on the modules of the Uq (D2 )
representations:
Sa,b ( ) : Va Vb Vb Va ,
(25)
where V is the module with highest weight . The scattering of two solitons of species
a and b is then described by Sa,b (12 ) with 12 being their relative rapidity. There are two
such fundamental multiplets denoted {4} and {6} in Ref. [21]. In addition, there are also
scalar bound states and excited solitons depending on the values of (p, p0 ) chosen.
(1)
(1)
To understand the restricted C2 (denoted RC2 below), we use the general framework
of superselection sectors (for details see [2426]). The model (1) is a perturbation of the
(1) (2)
12 . Each minimal model contains a finite
two minimal models by the operator 12
number of primary fields (13), (14). Using the superselection criterion for the present
case: 4
(1)

(1)
(2)
(2)
(z) 12
(w)12
(w)
2j
+1 1 (z)
2j +1 1
1
(1) (w) (2)
(w),
2j +1 2
(z w)l 2j +1 2
lZ
4 We only consider the holomorphic part of the primary operators but keep the same notation.
(26)
614
we find j + j Z where j j denote the representations of Uq (D2 ) with j the spin-j

representation of SU(2) with dimension 2j + 1 (and similarly for j). If q is a root of
unity, i.e, if Eq. (11) is satisfied then using the Coulomb gas representation condition (13)
1 6 2j + 1 6 p 1 and 1 6 2j + 1 6 p 1 one has:
0 6 j 6 p/2 1
and 0 6 j 6 p/2 1.
The superselection sectors H
(1)
RC2
j j
(1)
of the RC2 model (1) are thus:
(1)
HRC2 =
(27)
(j,j){0,1/2,...,p/21}
(1)
RC2
j j
with j + j Z.
(28)
As shown in [20,21] for the unitary series (p0 = p + 1), after the quantum group
restriction (11) for p > 3 the fundamental solitons in the {4} and {6} representation of
j j
. These kinks interpolate between different vacua
j2 j2 j1 j1
i
connected using the Uq (sl2 ) fusion ring at q = e p :
Uq (D2 ) become the RSOS kink K

|0j1 j1 i and |0j2 j2 i which are
j1 j =
min(j1 +j,p2j
X 1 j )
j2
(29)
j2 =|j1 j |
and similarly for j. However, for p = 3, (j, j) {0, 1/2} and then the {6} is projected out
of the spectrum, leaving only the {4}. We refer the reader to [20,21] for more details.
From the previous remarks and the identification D2 = SO(4) = SU(2) SU(2), by
(1)
(2)
analogy with [5] the primary fields 1s and 1s 0 commute with the generators in
(24) (for k = 0, 2) of the subalgebra Uq (D2 ) Uq (D3(2) ). If one interpret the fields
(1)
(2)
(z) as the highest component fields in the multiplet, it can be shown that
2j +1 1 (z)
2j +1 1
(1)
and r(2)
the primary operators rs
0 s 0 are not invariant with respect to Uq (D2 ). Together
with some non-local fields they form finite-dimensional representation of this algebra.
j j
Consequently, the VEV in (1) should take into account the factor drs,r 0 s 0 coming from
the QG restriction of the QFT (3). Following the conjecture of [5] it takes the form:

sin (2jp+1) |p0 r ps| sin (2pj+1) |p0 r 0 ps 0 |

j j
(30)
drs,r 0s 0 =

.
sin (2jp+1) (p0 p)

sin (2pj+1) (p0 p)
Using the notations introduced in the previous section, and Eqs. (14), (17), the outcome for
the VEV between different primary operators is:
(1)
(x)r(2)
h0j j |rs
0 s 0 (x)|0j j i
(1+)
1 (1 + ) 42
1+ 2 (rs +r 0 s 0 )
1+
j j
= drs,r 0s 0
1
1
3
1+ 1+

exp Q12 (1 + )r 2 s, (1 + )r 0 2 s 0 .
The function Q12 (, 0 ) for | 0 | < 4 and >
1
3
is given by the integral:
(31)
Q12 (, ) =
dt
t
615
12 (, 0 , t)
sinh((1 + )t) sinh(2t ) sinh((4 2 )t)
2 + 0 2 2(1 )2 2t
e
4( + 1)
with

12 (, 0 , t) = cosh ( + 0 )t cosh ( 0 )t cosh (2 2 )t

sinh (1 )t cosh (4 2 )t

+ cosh ( + 0 )t + cosh ( 0 )t cosh (2 2 )t 1
sinh(t) cosh(t )
and defined by analytic continuation outside this domain. Notice that Eq. (31) satisfies:
(1)
(1)
(x)r(2)
h0j j |rs
0 s 0 (x)|0j j i = h0j j |pr
(2)
p 0 s (x)pr 0 p 0 s 0 (x)|0j j i.
(32)
A particular case of Eq. (31) is the expectation value of the perturbing operator which
can be calculated explicitly to give the result:
(1)
(2)
h0j j |12 (x)12 (x)|0j j i

# 1+ 32
"
1
2
1+
2 2 (1 + )
1
=

3
1
1

( 2)
1+
1+

1
42 42
.
1+
42
(33)
By using the exact relation between the mass parameter m

and the mass of the kink M (16)
and Eqs. (6), (19) we immediately obtain the relation between M and :
M=
"
#

1
1
42
1+
42 42
.

1+
1
1
42
3
1+ 1+
1+
2 2
(34)
Consequently, according to Eqs. (19), (33), (34) and 2 < 2/3, the perturbed CFTs (1)
develop a massive spectrum for (Im() = 0):
(i)
0 < < 1/3,
> 0,
(ii)
1/3 < < 1,
< 0,
1
p
< ,
p0 2
p
1
i.e., < 0 < 1,
2 p
i.e., 0 <
(35)
where = 2pp0 p . In particular, the condition (ii) is always satisfied for the coupled unitary
minimal models defined by (1).
Finally, the expectation value (33) can be used to derive the bulk free energy:
1
ln Z,
V V
f12 = lim
(36)
where V is the volume of the 2D space and Z is the singular part of the partition function
associated with action (1). Using
616

(1) (2)
f12 = h0j j |12
12 |0j j i,
(37)
and Eqs. (33), (34), the result for the bulk free energy follows:

sin 42
M 2 sin 42
.
f12 =
)
2
sin (1+
42
(38)
As was suggested in [20,21], a particular case of the model (1) can be related to the SG
theory at the reflectionless point as well as the D8(1) ATFT with imaginary coupling. Let us
first consider the SG theory with action:

Z
1
2
2
( ) + cos() .
(39)
ASG = d x
16
This model is integrable and its on-shell solutions, i.e., spectrum of particles and S-matrix
are well-known [27]. This theory contains soliton, antisoliton and solitonantisoliton
2
bound-state (breathers) Bn , n = 1, . . . , 1/ , where = . The lightest bound-state
1 2
B1 coincides with the field in the perturbative treatment of the QFT (39). Its mass is
given by m1 = 2MSG sin( 2 ) where MSG denotes the soliton mass of the SG model. For
p = 3, p0 = 4 the model (1) corresponds to two magnetically coupled Ising models. It is a
(1) (2)
c = 1 CFT perturbed by the operator 12 12 with conformal dimension 1/8. Indeed, the
VEV of this operator must be simply related to the VEV of hcos()i

in the SG model for
2
= 1/8, which is also a c = 1 CFT with perturbing operator of conformal dimension 1/8.
For this latter value, = 1/7 and the SG model possesses 6 neutral excitations:

a
, a = 1, . . . , 6;
ma = 2MSG sin
14
m1 < m2 < MSG < m3 < < m6 .
(40)
Similarly, for 2 = 3/8, = 3/5 the spectrum of the model (1) was described in [21] and
summarize here. Here, we denote the lowest mass by M. If we compare the spectrum of
each model, using the notations of [23] we have:
m1 M,
m2 mB (1) ,
1
MSG mB (2) , etc.
(41)
Furthermore, the bulk free energy of the SG model (39) is well known [4,28] and given by:

M2

.
(42)
fSG = SG tan
4
2
By expressing it in terms of the lightest particle m1 using (40) and the relation
8 sin(/14) sin(3/14) cos(/7) = 1 we find the same result as in Eq. (38) for 2 = 3/8,
i.e., = 3/5. Also, using the results of [4] for 2 = 1/8, one has:

2 (1/8) MSG (4/7) 7/4
.
(43)
=
2
(1/14)
Similarly, the VEV of the perturbing operator is given by:

8/7

8 (1/7) (6/7)
1/7
hcos()i 2 =1/8 =
.
7 2 (4/7) 2 (13/14) (1/8)

2
617
(44)
Comparing Eqs. (43), (44) with Eqs. (33) and (34), respectively, for = 3/5 and using
Eq. (41), one find the relations:
1
22
and
2 (1) (2)
cos()|
12 12
2 =1/8 2
1
(45)
in perfect agreement 5 with Eq. (B.21) in [4].

Let us now turn to the relation with the D8(1) ATFT. The calculation of the bulk free
(1)
energy for the simply-laced D8 ATFT gives [28]:
sin 14
m
2
(46)
fD (1) | =

,
8 b
8 sin B sin (1B)
14
14
where B is defined in Section 2 and the mass of the particles are related with the mass
2 , m2a = 8m
2 sin2 ( a
parameter m
by: m28 = m27 = 2m
14 ) for a = 1, . . . , 6. For imaginary
coupling, one expects the particle-breather identification:
2

2
2
= 2M sin
,
(47)
8m
sin
14
14
where is defined as in Eq. (16). This yields:

sin
M2
14
=
fD (1) |

,
8 b=i
16 sin sin (1+ )
14
(48)
14
which for = 7 is in perfect agreement with Eq. (38) evaluated at = 3/5.

It is also interesting to study the behaviour of the model (1) for p0 = p + 1 in the limit
p . If we consider the sine-Gordon model with action (39) there is an equivalent
description in terms of the massive Thirring model [30]:

Z

g

MSG
2 ,
(49)
AMT = d 2 x i
2
where , is a Dirac field. The four fermion coupling constant g relates to in (39) by
g
1
2
= 2 1. Also the free fermion point is reached for 1/2. Using the results of
2
Ref. [4] concerning the one-point function heia i in the SG model one gets:
hcos()i
lim MSG () and
0
MSG
,
1
for 2 .
2
(50)
cos()/
Using the bosonfermion correspondence, one has the identification

MSG
0
and hi lim0 (). Let us now return to the case p = p + 1 and the limit
p in (1). It corresponds to 2 12 , i.e., 1 and similarly to Eq. (50) the VEV
(1) (2)
12 also becomes singular. Consequently, the perturbing
of the perturbing operator 12
5 At 2 = 1/8, the operator can be expressed [4,29] in terms of two independent Ising spin fields, (i) = (i)
12
for i = 1, 2.
618
operator in (1) can be conveniently rewritten in terms of two Dirac fields (with the same
mass Mf ) of two independent massive-Thirring models at the free fermion point:
Mf 0 0 .
12 12 Mf
(1)
(2)
(51)
In fact, this situation is not really surprising: when p the coset algebra of each
Mp/p+1 reduces to a level-1 SU(2) current algebra. The two models are coupled via their
primary fields in the spin 12 representation of dimension 14 [21]. As expected, in this limit
the model becomes the free field theory of two complex massive fermions with undeformed
SO(4) symmetry.
Finally, let us consider the case p0 = p + 1 and p = 2 in the action (1). Each minimal
model is then identified to M2/3 with central charge c = 0. For each model, the unitary
representation is the vacuum one with conformal weight 11 = 12 = 0, for which cases
(i)
= I(i) for i {1, 2}. This corresponds to the choice 2 = 13 ,
we have the identifications 12
1
i.e., = 2 and, using Eqs. (33), (34), (38), one can check that
(1) (2)
12 |0i = 1
h0|12
and f12 = ,
(52)
as expected.
Let us now turn to the model associated with action (2). Due to Eq. (11) it corresponds
2 . The condition 4p > 3p0 guarantees that the perturbing operator is
to the choice 2 =
relevant. Then, the vacuum structure is expected to be similar to that of (1). From Eqs. (30)
and (31) and the substitutions:
p p0 ,
(r, r 0 ) (s, s 0 ),
1+
3 1
(53)
the result for the VEV immediately follows:

(1)
(x)r(2)
h0j j |rs
0 s 0 (x)|0j j i
53 # 4

"
( + )
341 (2 ) 2 53 rs r 0 s 0
j
j
= drs,r 0s 0

1
1 2

exp Q21 (1 + )r 2 s, (1 + )r 0 2 s 0 .
The function Q21 (, 0 ) is given by the integral:

0
Q21 (, ) =
0
dt
t
21 (, 0 , t)
sinh((1 )t) sinh(2t ) sinh((3 5 )t)
2 + 0 2 2(1 )2 2t
e
4( + 1)
with

21 (, 0 , t) = cosh ( + 0 )t cosh ( 0 )t cosh (2 2 )t

sinh (1 )t cosh (3 5 )t

cosh ( + 0 )t + cosh ( 0 )t cosh (2 2 )t 1
(54)
619

sinh (3 1)t/2 cosh (1 + )t/2
and defined by analytic continuation outside this domain. The prefactor associated with the
j j
j j
QG restriction d 0 0 = d 0 0 |pp0 . For (r, s) = (r 0 , s 0 ) = (2, 1) (54) becomes:
rs,r s
sr,s r
(1)
(2)
(x)21
(x)|0j j i
h0j j |21
"
1
=

1
3 1 # 53
4
1
2
4
53
2 53 (4 )
(3 5 )
1+
3 1
10 6 10 6

523
(55)
with the relation between the mass of the lightest kink M and :
1+
M=
2 53
#
1+
3 1 "
3 1 53

10 6 10 6
4
.

1
523
1 2
2
(56)
For the coupled minimal models defined by (2), the massive phase corresponds to the
domain:
(iii)
3
< < 1,
5
< 0,
i.e.,
3
p
< 0 < 1.
4 p
(57)
One also obtains the bulk free energy associated with action (2):
1)
(1+ )
M 2 sin (3
10 6 sin 10 6
.
f21 =
)
2
sin (2
5 3
(58)
4. Application and concluding remarks

Accepting the conjectures (31) and (54), one can easily deduce interesting predictions 6
for the short and long distance asymptotic of two-point correlation functions in the model
(1) or (2). For each case depicted in Fig. 1, we can express the result in terms of:
(2)
j j
(1)
(x)r 0 s 0 (x)|0j j i = Grs,r 0 s 0
h0j j |rs
(59)
given by Eq. (31) or (54), respectively. First, using the short distance approximation
(operator product expansion) and Eq. (4) we get eia1 (x)eib2 (y) ei(a1 (x)+b2 (y)) . Then,
using the Coulomb gas representation of each primary operator (14) one has
(2)
(2)
(1)
(1)
(x)r 0 s 0 (y)|0j j i h0j j |rs
(x)r 0 s 0 (x)|0j j i.
h0j j |rs
|xy|0
(60)
In this limit case (a) , the two-point functions then become:

j j
(1)
(x)r(2)
h0j j |rs
0 s 0 (y)|0j j i Grs,r 0 s 0 .
|xy|0
(61)
6 The same method was applied in [4] to obtain a prediction about the long distance asymptotic of two-point
correlation function h0 n in in the XXZ spin chain.
620
Secondly, in the long distance approximation the asymptotic two-point function simply
reduces to the product of two one-point functions as heia.(x)eib.(y) i heia.(x) iheib.(y)i
when |x y| . Then we obtain case (b) and (d)
(1)
(x)r(2)
h0j j |rs
0 s 0 (y)|0j j i
(j )
(1)
(x)I(2)|0j j ih0j j |I(1) r 0 s 0 (x)|0j j i,
h0j j |rs
|xy|
(1)
(1)
(x)r 0 s 0 (y)|0j j i
h0j j |rs
(1)
(1)
(x)I(2)|0j j ih0j j |r 0 s 0 (x)I(2) |0j j i.
h0j j |rs
|xy|
Indeed, it gives:
(i)
(1)
(x)r 0 s 0 (y)|0j j i
h0j j |rs

j j
j j
j j
Grs,11 i1 Gr 0 s 0 ,11 + i2 G11,r 0 s 0 ,
|xy|
for i {1, 2}
(62)
with ii 0 the Krnecker symbol. Obviously similar results are obtained for the two-point
(i)
(2)
function h0j j |rs (x)r 0 s 0 (y)|0j j i using the Z2 symmetry (1 2).
Finally, using the fusion rules in the short distance approximation (the two primary
(i)
(i)
r(i)
fields belong to the same space of states) rs
0 s 0 r 00 s 00 , the two-point
(2) |0 i or
function is expanded in terms of the one-point functions h0j j |r(1)
00 s 00 (x)I
j j
h0j j |I(1) r(2)
00 s 00 (x)|0j j i for i {1, 2}, respectively. Then, we have for i = 1 case (c)
(1)
(1)
(x)r 0 s 0 (y)|0j j i
h0j j |rs
X 00 00
r s
2(r 00 s 00 rs r 0 s 0 ) j j
Crs,r
Gr 00 s 00 ,11
0 s 0 |x y|
|xy|0
(63)
r 00 s 00
00 00
r s
and similarly for i = 2 where Crs,r
0 s 0 are the structure constants of the minimal model
operator algebra [33].
Example 1 Two magnetically coupled Ising models. It is now straightforward to compute

different two-point correlation functions in one of the simplest (non-trivial) cases: twomagnetically coupled Ising models in the massive phase. It corresponds to 2 = 3/8,
(i)
= (i) with conformal
i.e., = 3/5 in (31). In this case we have the identification 12
(i)
dimension = 1/16 the spin operator and 13 = (i) with conformal dimension
= 1/2 the energy operator. There are two degenerate ground states [20,21] denoted
|000i |i and |01/2 1/2 i |+i. For simplicity, we write h| |i h i . Sometimes,
the reflection relation:
G(1 , 2 ) = SL (2 )G(1 , 2 + 2 1/)
with

2 +2
SL (2 ) = 0 2 2
2 1/2
(64)
621
(22 + 2 2 ) (2 / + 1/2 2 )
(2 + 22 2 2 ) (2 2 / 1/2 2)
= d1s,11
for all s we obtain
is useful for analytic continuation. Using Eq. (31) and d11,1s
for instance:
(i)
= 1.297197220 . . .()1/14 ,
= G11,12
(i)
= 2.278284275 . . .()4/7 ,
= G11,13

(1)
= 1.698928047 . . .()1/7 ,
(0) (2)(0) = G12,12

(1)
)2 = 1.682720628 . . .()1/7 ,
(0) (2)() = (G11,12

(1)
= 3.311880669 . . .()9/14 ,
(0) (2)(0) = G12,13

(1)
G11,13
= 2.955384028 . . .()9/14 ,
(0) (2)() = G11,12

(1)
)2 = 5.160349412 . . .()8/7 ,
(0) (2)() = (G11,13
where the parameter is related to the mass of the lowest kink by:
= 0.2379062104 . . .M 7/4.
(65)
Notice that h (1) (0) (2)(0)i and h (1) (0) (2)()i differ by less than 0.7% as
expected. 7
Example 2 Two energyenergy coupled tricritical Ising models. The case p = 4, p0 = 5
in (1) describes two tricritical Ising models which interact through their leading energy
(i)
density operators 12 = (i) of conformal dimension = 1/10. It corresponds to
2 = 2/5, i.e., = 2/3 in (31). Beside (i) and the identity operator, each minimal
(i)
= 0 (i) with 0 = 3/5
model contains the sub-leading energy density operator 13
(i)
(i)
(i)
(vacancy operator), two magnetic operators 22 = with = 3/80, 21 = 0 (i)
(i)
j j
j j
j j
with 0 = 7/16 and 14 . Due to the obvious property drs,r 0s 0 = d11,r 0s 0 drs,11 , similarly to
the previous case we find for instance for any vacuum |j ji:
(1)
j j
j j
j j = G22,11 = d22,11 1.144656674 . . .()3/64,
(1)
j j
j j
j j = G12,11 = d12,11 1.529866659 . . .()1/8 ,

(1)
j j
j j
(0) (2)(0) j j = G22,22 = d22,22 1.315726811 . . .()3/32,

(1)
j j
j j
j j
(0) (2)() j j = G22,11 G11,22 = d22,22 1.310238901 . . .()3/32 ,

(1)
j j
j j
(0) (2)(0) j j = G12,12 = d12,12 2.419476973 . . .()1/4 ,

(1)
j j
j j
j j
(0) (2)() j j = G12,11 G11,12 = d12,12 2.340491994 . . .()1/4 ,
7 For instance, it is known that form factors are able to reproduce with high accuracy the UV behaviour of the
correlation functions [3,31,32].
622
= 0.2566343706 . . .M 8/5.
(66)
Notice that h (1) (0) (2)(0)ij j and h (1) (0) (2)()ij j , h (1) (0) (2)(0)ij j
h (1) (0) (2)()ij j differ by less than 0.5% and 3%, respectively.
and
Example 3 Two coupled A5 RSOS models. The case p = 5, p0 = 6 in (1) describes

(i)
two A5 RSOS models coupled by their primary operators 12 with conformal dimension
12 = 1/8. It corresponds to 2 = 5/12, i.e., = 5/7 in (31). Each minimal model also
(i)
contains the primary operator 22 with 22 = 1/40. As before we obtain:
(1)
22
(1)
12
j j
j j
j j
j j
j j
= G22,11 = d22,11 1.090446894 . . .()1/30,
j j
= G12,11 = d12,11 1.726352342 . . .()1/6 ,

j j
j j
(1)
(2)
(0)22
(0) j j = G22,22 = d22,22 1.191588988 . . .()1/15,
22

(1)
j j
j j
j j
(2)
() j j = G22,11 G11,22 = d22,22 1.189074429 . . .()1/15,
22 (0)22
= 0.2697511940 . . .M 3/2.
(1)
(2)
(67)
(1)
(2)
Notice that h22 (0)22 (0)ij j and h22 (0)22 ()ij j differ by 2%.
Example 4 Two energyenergy coupled 3-state Potts models. The case p = 5, p0 = 6
in (2) describes two 3-state Potts models coupled [19] by their energy density operator
(i)
= (i) with conformal dimension 21 = 2/5. It corresponds to = 5/7 in (54). Each
21
(i)
minimal model also contains the primary operator 23 = (i) the spin operator with
23 = 1/15. We obtain for instance:
(1)
j j
j j
j j
1/3 ,
= G23,11 = d23,11 1.9079 . . .()

j j
j j
(1) (0) (2)(0) j j = G23,23 = d23,23 4.50 . . .( )2/3 ,

(1)
j j
j j
j j
2/3 ,
(0) (2)() j j = G23,11 G11,23 = d23,23 3.64 . . .()
= 0.2612863655 . . .M 2/5.
(68)
To conclude, we would like to mention that here we studied only a special case of a
much more general class of integrable coupled models [1921]. There exist many other
examples which can be worked out along the same lines. Let us also note that the exact
form factor techniques as well as the truncated conformal space approach may be used and
similar numerical analyses can be performed for the correlation functions.
623
Acknowledgements
I am grateful to C. Ahn, G. Delfino, G. Delius, D. Reynaud, P. Simon, C. Kim, C. Rim,
L. Zhao and especially to V.A. Fateev and E. Corrigan for useful discussions. Work
supported by the EU under contract ERBFMRX CT960012 and Marie Curie fellowship
HPMF-CT-1999-00094.
References
[1] A. Patashinskii, V. Pokrovskii, Fluctuation Theory of Phase Transitions, Pergamon Press,
Oxford, 1979.
[2] M. Shifman, A. Vainstein, V. Zakharov, Nucl. Phys. B 147 (1979) 385.
[3] Al. Zamolodchikov, Nucl. Phys. B 348 (1991) 619.
[4] S. Lukyanov, A. Zamolodchikov, Nucl. Phys. B 493 (1997) 571.
[5] V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Phys. Lett. B 406 (1997) 83;
V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Phys. Lett. B 516 (1998) 652.
[6] A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 477 (1996) 577.
[7] A. LeClair, Phys. Lett. B 230 (1989) 423;
N. Reshetikhin, F. Smirnov, Commun. Math. Phys. 131 (1990) 157;
D. Bernard, A. Leclair, Nucl. Phys. B 340 (1990) 721;
C.J. Efthimiou, Nucl. Phys. B 398 (1993) 697.
[8] F. Smirnov, Int. J. Mod. Phys. A 6 (1991) 1407.
[9] V.A. Fateev, Nucl. Phys. B 473 (1996) 509.
[10] P. Baseilhac, V.A. Fateev, Nucl. Phys. B 532 (1998) 567.
[11] V.A. Fateev, Mod. Phys. Lett. A 15 (2000) 259.
[12] C. Ahn, P. Baseilhac, V.A. Fateev, C. Kim, C. Rim, Phys. Lett. B 481 (2000) 114.
[13] C. Ahn, P. Baseilhac, C. Kim, C. Rim, in preparation.
[14] C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233;
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. Lett. 74 (1995) 3005;
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934;
P. Fendley, A.W.W. Ludwig, H. Saleur, Stat. Phys. 19 (1996) 137.
[15] F. Lesage, H. Saleur, P. Simonetti, Tunneling in quantum wires I: exact solution of the spin
isotropic case, cond-mat/9703220;
F. Lesage, H. Saleur, P. Simonetti, Tunneling in quantum wires II: a new line of IR fixed points,
cond-mat/9707131.
[16] F.P. Milliken, C.P. Umbach, R.A. Webb, Solid State Commun. 97 (1996) 309.
[17] M. Karowsky, P. Weisz, Nucl. Phys. B 139 (1978) 445.
[18] F.A. Smirnov, Forms Factors in Completly Integrable Models of Quantum Field Theories,
World Scientific, 1992.
[19] H.J. de Vega, V.A. Fateev, J. Phys. A 25 (1992) 2693.
[20] I. Vaysburd, Nucl. Phys. B 446 (1995) 387.
[21] A. LeClair, A.W.W. Ludwig, G. Mussardo, Nucl. Phys. B 512 (1998) 523.
[22] G.W. Delius, M.T. Grisaru, D. Zanon, Nucl. Phys. B 382 (1992) 365;
E. Corrigan, P.E. Dorey, R. Sasaki, Nucl. Phys. B 408 (1993) 579.
[23] G.M. Gandenberger, N.J. MacKay, G.M.T. Watts, Nucl. Phys. B 465 (1996) 329.
[24] J. Frlich, Commun. Math. Phys. 47 (1976) 269.
[25] K. Fredenhagen, K.H. Rehren, B. Schroer, Commun. Math. Phys. 125 (1989) 201.
[26] G. Felder, A. LeClair, Int. J. Mod. Phys. A 7 (1992) 239.
[27] A. Zamolodchikov, Al. Zamolodchikov, Ann. Phys. (NY) 120 (1979) 253.
624
[28]
[29]
[30]
[31]
[32]
[33]
C. Destri, H.J. de Vega, Nucl. Phys. B 358 (1991) 251.

T.T. Wu, M.B. McCoy, C.A. Tracy, E. Barouch, Phys. Rev. B 13 (1976) 316.
S. Coleman, Phys. Rev. D 11 (1975) 2088.
V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 3419.
G. Delfino, G. Mussardo, Phys. Lett. B 324 (1994) 40.
V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312;
V.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691.

One-point functions in perturbed boundary

conformal field theories
P.E. Dorey a,b, , M. Pillin c , R. Tateo d , G.M.T. Watts e
a Department of Mathematical Sciences, University of Durham, Durham DH1 3LE, UK
b SPhT, CEA-Saclay, F-91191 Gif-sur-Yvette Cedex, France
c ETAS, PTS-A, Borsigstr. 10, D-70469 Stuttgart, Germany
d Universiteit van Amsterdam, Inst. voor Theoretische Fysica, 1018 XE Amsterdam, The Netherlands
e Mathematics Department, Kings College London, Strand, London WC2R 2LS, UK
Abstract
We consider the one-point functions of bulk and boundary fields in the scaling LeeYang model
for various combinations of bulk and boundary perturbations. The one-point functions of the bulk
fields are analysed using the truncated conformal space approach and the form-factor expansion.
Good agreement is found between the results of the two methods, though we find that the expression
for the general boundary state given by Ghoshal and Zamolodchikov has to be corrected slightly.
For the boundary fields we use thermodynamic Bethe ansatz equations to find exact expressions for
the strip and semi-infinite cylinder geometries. We also find a novel off-critical identity between the
cylinder partition functions of models with differing boundary conditions, and use this to investigate
the regions of boundary-induced instability exhibited by the model on a finite strip. 2001 Elsevier
1. Introduction
The purpose of this paper is to make a detailed analysis of the one-point functions in
a particular integrable boundary quantum field theory, the boundary scaling LeeYang
model. Such a study is of intrinsic interest, allowing one to check the consistency of various
different approaches to the computation of these quantities. It also serves to illuminate
some issues that have arisen in recent investigations of the spectra and boundary entropies
(g-functions) of integrable models with boundaries.
E-mail addresses: p.e.dorey@durham.ac.uk (P.E. Dorey), mathias.pillin@etas.de (M. Pillin),

tateo@wins.uva.nl (R. Tateo), gmtw@mth.kcl.ac.uk (G.M.T. Watts).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 2 - 2
626
P.E. Dorey et al. / Nuclear Physics B 594 [FS] (2001) 625659
After a brief review of the pertinent features of the boundary scaling LeeYang model in
Section 2, Sections 3 and 4 of the paper are devoted to the one-point functions of bulk fields
in the presence of a boundary. Since these generally have a dependence on the distance
from the boundary, their exact determination is a non-trivial problem, of similar difficulty
to the calculation of two-point functions in the absence of boundaries. For this reason,
apart from some simple cases associated with the Ising model [13], no exact formulae for
these boundary one-point functions are known. Instead, one has to resort to approximate
methods, two of which we investigate in some detail. The first is the truncated conformal
space approach, or TCSA. This has previously been used to study integrable models both
without [12,23] and with [5,6] boundaries, but this particular application of the technique
is new. The second method is based on a form factor (FF) expansion, and makes use of the
expansion of the boundary states in the basis of infrared multi-particle states given in [11].
Previous work on this topic includes Refs. [13,14,17,19]. The main novelties arising in
the discussion of the scaling LeeYang model presented below are that this model has a
non-trivial bulk S-matrix, and that the boundary one-point functions receive contributions
from both even- and odd-particle-number components of the boundary state. This allows
the relative normalisations of the two sectors to be checked, and we find evidence that the
prescription given in [11] is out by a factor of 2. Section 3 introduces and discusses the
TCSA and FF methods, culminating in a numerical comparison between the two which,
modulo the small correction to the boundary state just mentioned, shows an excellent
agreement.
Both the TCSA and the FF methods can be pushed a little further, allowing expectation
values in states other than the ground state to be accessed. These generalisations are
explained in Section 4.
The expectation values of boundary fields are also of interest, and for these there is more
hope to find exact results, at least for certain geometries. This is the subject of Section 5.
First, in Section 5.1, we describe the so-called R-channel approach, which allows us
to obtain the expectation values of boundary fields at the edge of a strip of finite width,
both in the ground state and, in principle, in any excited state. By relating the expectation
values to the derivative of the energy with respect to the boundary field, we find that they
can be expressed in terms of the solutions to TBA equations, and compare these results
with data from the TCSA (obtained by adapting the method of [12]), for various different
combinations of boundary conditions and for varying strip width. Then in Section 5.2
we switch to the L-channel. Making use of results from [6], we are able to relate the
expectation value of a boundary field placed at one end of a semi-infinite cylinder to the
so-called Y-function of the LeeYang model. An interesting relationship between the
spectra of certain different models placed on the same strip emerges as a by-product of this
discussion.
Finally, in Section 6, we make use of the results in the preceding sections to examine
the RG flow from the (h) boundary to the 1 boundary, first discussed in [5], and to treat
some previously obscure features of the model on a strip of finite width, with simultaneous
perturbations on both boundaries. Section 7 contains our conclusions, and indicates some
directions for future work.
627
As in previous papers [5,6,8], we will be using the boundary scaling LeeYang model
as our example. We have included a brief review of the features of this model in the next
section, but the earlier works should be consulted for some more detailed explanations.
2. The boundary LeeYang model

2.1. The conformal field theory description of the critical LeeYang model
The LeeYang model is the simplest non-unitary conformal field theory, M2,5 , and has
central charge 22/5 and effective central charge 2/5. There are only two representations
of the Virasoro algebra of interest, of weight 0 and 1/5, and consequently only two bulk
primary fields, the identity 1 of weight 0, and of weight x = 2/5; equally there are
only two conformally-invariant boundary conditions, which we denote by 1 and .
()
There are three non-trivial boundary fields 1/5 interpolating the various boundary
conditions and , of weight h = 1/5. Two of these (, ) interpolate the two different
conformal boundary conditions, 1 and one () lives on the boundary:
(1)
,
1/5
(1)
1/5
,
()
1/5
.
(2.1)
The OPEs of interest are the bulk OPE,

+ ,
(z, z )(w, w)
= C 1 |z w|4/5 + C |z w|2/5 (w, w)
(2.2)
the boundary OPEs,

(z) (w) = C 1 |z w|2/5 + C |z w|1/5 (w) + ,
(z)(w) = C |z w|1/5 (w) + ,
(z) (w) = C |z w|1/5 (w) + ,

(z) (w) = C 1 |z w|2/5 + ,
(z)(w) = C 1 |z w|2/5 + C |z w|1/5 (w) + ,
and the two bulk-boundary OPEs
2/5

(z)|1 = (1)B1 2(z w) + ,
2/5
1/5

(z)| = ()B1 2(z w) + ()B 2(z w) (w) + .
A suitable choice for the structure constants is
1+ 5
,
C = 1,
C =
C = C = 1,
2

1/2
1/2
2
2
(1) 1
B =
C =
2 ,
,
1+ 5
1 + 5

1 + 5 1/2
1 + 5 3/2
()
1
,
,
B =
C = C =
2
2
1
1 In [6] we did not distinguish these fields and denoted them both by .
(2.3)
(2.4)
628
= C

2 1/2

=
,
1 + 5

5 + 5 1/2
,

B =
2
()

(1/5) (6/5) 1/2
.

=
(3/5) (4/5)
(2.5)
There are three possible choices for pairs of conformal boundary conditions on a strip:
(1, 1), (, 1) and (, ). We shall take the strip to be of width R with coordinates 0 6
x 6 R across the strip and y running along the strip, and normalise all our correlation
functions so that the expectation value of the identity operator on a strip is always one.
The strip correlation functions between states h| and |i can be found by mapping
the strip to the unit disc and inserting the appropriate fields and . Since the ground
states on the strips with boundary conditions (1, 1), (1, ) and (, ) correspond to the
fields 1, and respectively, one needs to include the appropriate field insertions to find
the ground state expectation values on these strips. These insertions lead directly to the
particular chiral blocks and structure constants in (2.7) and (2.8).
The one-point functions of the field (x) on such a strip are best expressed in terms of
the four strip chiral block functions fi ( ),

2 sin 2/5
2
4 8 11
( ) =
f1 ( ) = f1
2 F1 10 , 10 ; 10 ; tan ,
2
cos

2 sin
cos3
1/5
( ) =
11
( ) = (2 sin )2/5 ,
( ) = (2 sin )1/5 .
f2 ( ) = f
f3 ( ) = f1
f4 ( ) = f
2
3 8 9
2 F1 10 , 10 ; 10 ; tan
In terms of these functions, we have

2/5

x 2/5
R
x
2R
(1) 1
= (1)B1
sin
B f3
,
(x, y) (1,1) =
R
2/5

R
x
x
() 1
B f1
+ ()B C f2
,
(x, y) (,1) =
R
R

2/5

R
x
x
() 1
+ ()B C f2
.
B f1
(x, y) (,) =
R
R
(2.6)
(2.7)
It is only for the latter two pairs of boundary conditions that the boundary field (y) exists
on the boundary x = 0. For these cases the one-point functions are simply
1/5
1/5

R
R
C ,
(y) (,) =
C .
(2.8)
(y) (,1) =
Of the five expectation values (2.7), (2.8), only the first has a finite limit as the strip
width tends to , the others all diverging. None of them depend on y, and so when no
629
confusion can arise this variable will often be omitted, even inside the vacuum expectation
values.
2.2. The scaling LeeYang model
The scaling LeeYang (SLY) model can be described as a perturbation of the critical
LeeYang model by the term
Z
(2.9)
(w, w)
d2 w.
This leads to a massive scattering theory, comprising a single particle with two-particle
S-matrix [4]

sinh 2 + ix
6
(2.10)
S( ) = (1)(2),
(x) =
.
sinh 2 ix
6
The mass M of the particle is related to the bulk perturbation parameter by [24,26]

5/12
35 45
5/12
19/12
(2.11)
, =2
M =

= 2.642944 . . . .
55/16 23 56
We will also need the form factors of the bulk model. These are the matrix elements of
the elementary field (x) in the asymptotic n-particle states which can be formally written
in terms of the ZF operators A( ) as |1 , . . . , n i = A(1 ) A(n )|0i. The form-factor
Fn (i n ) is then given by

X
cosh i Fn (1 , . . . , n ).
(2.12)
h0|(x, 0)|1, . . . , n i = exp Mx
i
The form factors of the SLY model were first computed in [21]; we, however, adopt the
conventions of [25], modulo the fact that for us (x) is a real field. The function Fn can be
parametrised as [25]:
Fn (1 , . . . , n ) = Hn Qn (x1 , . . . , xn )
n
Y
f (i j )
i<j
xi + xj
(2.13)
where xi = exp(i ), i = 1, . . . , n. The terms in (2.13) can be determined through the form
factor bootstrap [22], with the result
f ( ) =
cosh 1
cosh +
1
2
v(i )v(i + ),
(2.14)
where we take the function v in a form suitable for numerical evaluation [25]
"

#n
N
1
1
1
Y
2i + n + 2 2i + n 6 2i + n 3
v( ) =

1
1
1
2i + n 2 2i + n + 6 2i + n + 3
n=1
Z
exp 2
dt
0
sinh(t/2) sinh(t/3) sinh(t/6)

t sinh2 (t)
N + 1 Ne
2t
!
e
2Nt +it /
630
v(0) = 1.111544045 . . .
(2.15)
(with N arbitrary) and

i 31/4 n
.
Hn =
2 v(0)
(2.16)
This overall normalisation of the form factors is taken from the results of [10], where (in
our conventions) the expectation value hi in the bulk is

"
#5/12
9
1/6 1 2 1
5
6
10
= hi = ||1/6 7/4 1/2 3 7/6

1
2 3
56 10
= (0.840184 . . .)||1/6 .
(2.17)
Using the relation between M and in (2.11), this boils down to

9
36
M 5
2/5
.
=

= (1.239394325 . . .)M
14 1
(2) 5 5 4 15 25
3 10
1
3
(2.18)
The functions Qn (x1 , . . . , xn ) in (2.13) are symmetric polynomials of degree n(n 1)/2
and partial degree n 1. These polynomials have been determined via the form factor
bootstrap approach for arbitrary particle numbers n. They can be nicely written in the form
of a determinant of a matrix in symmetric polynomials [21,25], for a related formulation
see [20]. For our purposes it will be sufficient to list the first few:
Q0 = 1,
Q2 = e1(2),
Q1 = 1,
(3) (3)
(4) (4) (4)
Q3 = e2 e1 ,
Q4 = e3 e2 e1 ,
(2.19)
where the elementary symmetric polynomials in n variables er(n) are defined by

n
Y
(1 + txi ) =
i=1
n
X
t r er(n) .
(2.20)
r=0
The integrable boundary conditions for the model were discussed in detail in [5]. The
allowed boundary conditions are the 1 conformal b.c., and the perturbation (h) of the
conformal boundary by the integral along the boundary
Z
h (x) dx.
(2.21)
The boundary reflection factors corresponding to these two boundary conditions are
R(h) ( ) = Rb ( ),
where
Rb ( ) =
R1 ( ) = R0 ( ),
1

1
3
4
b + 3 1
b+3
.
S i
S + i
2
2
2
6
6
(2.22)
(2.23)
The relation between b and h was conjectured in [5] to be

h(b) = |hcrit | sin b + 12 /5 .
631
(2.24)
Sometimes we shall find it useful to consider instead the dimensionless quantity

h crit = hcrit M 6/5,
(2.25)
a constant which was found in [6]:

sin 2
5
h crit = 2 5
4 1/2
3
5 5
= 0.68528998399118 . . . .
3/5 4/5 1/4
!6/5
2
3

1
6
Combining (2.11) and (2.26), we obtain the more convenient formula

53/8 sin b + 12 /5 1/2
.
h = 1/2
2
sin(/5)
(2.26)
(2.27)
Finally, we will need the boundary-particle couplings ga for the various boundary
conditions. In [11] these were defined in two different ways, either via the residue at =
i/2 of the reflection factor Ra ( ) for a particle of type a on the boundary :
Ra ( )
i (ga )2
,
2 i/2
(2.28)
or, in models where bulk fusings occur, via the residues of certain other poles, divided by
the corresponding bulk couplings. (If the boundary scattering is non-diagonal, the formulae
become a little more complicated see [11].) For the LeeYang model, the consistency of
the two definitions was shown to follow from the bootstrap equations and crossing in [8].
Since for this case there is only one particle type, the particle index a can be dropped, and
we have
for the 1 boundary,
q
(2.29)
g1 = i 2 2 3 3 ;
for the (h(b)) boundary,
g (b) =
tan((b + 2)/12)
g1 .
tan((b 2)/12)
(2.30)
Notice that while the reflection factors for the 1 and (h(0)) are identical, the
corresponding boundary-particle couplings differ by a sign. This can be traced to the fact
that when b = 0, the residue of Rb at i/6 for the (h(0)) boundary receives additional
contributions from intermediate states containing boundary bound states. The net effect is
to negate the coupling; indeed, this is the only option given that the residue at i/2 must
remain unchanged. In terms of the boundary states to be discussed in the next section, this
means that the only difference between the (infinite-volume) boundary states for these two
boundary conditions is in the sign of the contributions from states of odd particle number.
632
3. The one-point functions of the bulk field

3.1. h(x)i from form factors
In this section we consider the one-point function h(x)i in the presence of a boundary
using the form factor approach. We will restrict consideration to a theory with a single
scalar particle, as in the LeeYang model. Schematically, the idea of the form-factor
approach is to evaluate the one-point function on the upper half plane as
X

h0|(x)|nihn|B i,
(x) = h0|(x)|B i =
(3.1)
n=0
where |B i is a boundary state corresponding to the boundary condition and the sum
over asymptotic states has been split into the contributions from n = 0, 1, 2, . . . particles
(here |nihn| represents the projection onto asymptotic states with n particles). We have also
taken the boundary state to be normalised such that h0|B i = 1.
This boundary state can be expanded in terms of multi-particle states on the infinite line,
using the ZamolodchikovFaddeev (ZF) operators A( ) which create single particles of
rapidity . When the state contains no zero-rapidity particles, it can be written as [11]
" Z
#
d
K ( )A( )A( ) |0i,
(3.2)
|B i = exp
4
where K ( ) is related to the reflection amplitude R ( ) for the boundary condition by

i
.
(3.3)
K ( ) = R
2
In general there may also be contributions to the boundary state involving zero-momentum
particles, which can be associated with couplings of single bulk particles to the boundary.
Up to the three-particle contribution, the appropriate boundary state was given in [11] as
"
Z
d
K ( )A( )A( )
|B i = 1 + g A(0) +
4
Z
+ g A(0)
#
d
K ( )A( )A( ) + |0i,
4
and it is natural to suppose that the full expression is

"
#
Z
d
K ( )A( )A( ) |0i.
|B i = exp g A(0) +
4
(3.4)
(3.5)
In [11], the factor g was identified with the boundary-particle coupling g , as defined at
the end of the last section. However, our numerical results (and also an examination of
the reflection factor and boundary state for the Ising model with free boundary conditions
633
given in [11]) cast doubt on this suggestion. As will be explained in Section 3.3 below, we
found that we had rather to set
g = g /2
(3.6)
in order to obtain a successful match with TCSA data.

The form factors for the LeeYang model were given in the previous section, and
substituting these into (3.4), we find the leading large-x behaviour of h(x)i is

(x) = h0|(x)|B i
Z
31/2
31/4
d
Mx
K ( )f (2 )e2Mx cosh()
e
= 1 i g
2
4
2v(0)
2 v(0)
33/4
+ i g
2 2 v(0)3
d
K ( )
4
!
f ( )f ( )f (2 )(1 + 2 cosh )2 Mx(1+2 cosh())
e
+ . (3.7)
8 cosh cosh(/2)2
This will be compared with TCSA data in Section 3.3 below.
3.2. Estimating h(x)i using the TCSA
The first step in the calculation of h(x)i using the TCSA is the numerical evaluation of
of the perturbed Hamiltonian with dimensionless strip width r MR.
the ground state |0i
b(R, , hl , hr )
H
12/5 Z
R

c
+
exp(i ) d
L0
=
R
24
=0
!
6/5
6/5
R
R
+ hl l (1) + hr r (1) ,
(3.8)
b represents the operator O on the upper half plane restricted to the conformal space
where O
truncated to level N . The parameter determines the bulk mass and we have allowed the
possibility of boundary fields on the left and right edges of the strip, with strengths hl
and hr . The second step is then to estimate the expectation value of the (dimensionless)
operator M 2/5 (x) in terms of the matrix elements of the operator as
2/5
2/5
r
h0|(exp(ix/R))|
0i
.
(3.9)
M (x)
0i
h0|
can be expanded in Virasoro primary and descendent states; by repeatedly
The state |0i
commuting Virasoro algebra elements through ,
the general matrix elements of can
be expressed in terms of the matrix elements between Virasoro primary states, and their
derivatives. In the rest of this section we discuss the TCSA method in more detail;
634
a comparison of the TCSA and FF results is given in Section 3.3. All the TCSA results
in this paper were calculated on a workstation in Mathematica with truncation levels up to
18 and on spaces with up to 161 states.
3.2.1. The strip with (1, 1) boundary conditions
The simplest case is the strip with boundary conditions 1 on both sides. The unperturbed
conformal field theory expectation value is given in (2.7). If we denote the scaled position
of the field by and the normalised strip width by r, where
= xM,
r = MR,
so that 0 6 6 r,
(3.10)
then the TCSA estimate of the expectation value in the bulk perturbed model truncated to
level N takes the form

N

2/5 X (N,r)
2r
2n
sin
. (3.11)
fn
cos
G (N,r) ( ) M 2/5 (x) (1,1) N,r =
r
r
n=0
(N,r)
and have
The coefficients fn
are determined by the expansion of the ground state |0i
lies in the h = 0 representation, the
to be calculated numerically. Since the state |0i
matrix elements of are given in terms of the chiral block f3 ; furthermore, since this
representation has a null state at level 1, one can eliminate all states containing L1 , and
so one does not need any terms with derivatives of this chiral block in (3.11). To compare
with the form-factor calculation we will need to take the simultaneous limits N ,
r while keeping fixed. We shall often drop the labels (N, r) and write simply G( )
for the TCSA estimates.
Table 1
(N,r)
The TCSA coefficients fn
for r = 8 and 4 6 N 6 12
N
n
0
1
2
3
4
5
6
7
8
9
10
11
12
4
0.695639
0.074584
0.015439
0.0004456
4.406105
10
12
0.695205
0.072820
0.013388
0.004562
0.0001517
2.009105
3.59106
0.695043
0.072397
0.003932
0.009036
0.001821
6.377105
9.89106
2.30106
5.60107
0.694975
0.072162
0.012698
0.003828
0.001572
0.002017
3.15105
5.29106
1.43106
4.42107
1.29107
0.694979
0.072053
0.012618
0.003715
0.001524
0.0007604
0.000480
1.74105
3.06106
8.95107
3.20107
1.17107
3.84108
635
Table 2
(N,r)
The TCSA coefficients fn
for N = 12 and 2 6 r 6 12
r
n
10
12
0
1
2
3
4
5
6
7
8
9
10
11
12
0.780435
0.0049452
0.0005305
0.0001405
5.579105
2.796105
1.642105
3.10108
4.13109
1.15109
4.011010
1.441010
4.301011
0.759482
0.0226611
0.0027410
0.0007379
0.0002942
0.0001469
8.691105
8.01107
1.14107
3.20108
1.12108
4.02109
1.22109
0.728325
0.0476133
0.0068911
0.0019225
0.0007746
0.0003855
0.0002332
4.98107
7.85107
2.24107
7.88108
2.85108
8.93109
0.694979
0.0720535
0.0126176
0.0037146
0.0015242
0.0007604
0.0004801
1.73105
3.06106
8.95107
3.20107
1.17107
3.84108
0.663779
0.0923617
0.0192286
0.0060376
0.0025408
0.0012814
0.0008648
4.44105
8.76106
2.64106
9.66107
3.64107
1.26107
0.635945
0.108094
0.026052
0.0087501
0.0037897
0.001952
0.001443
9.48105
2.07105
6.45106
2.44106
9.50107
3.54107
0.786151
0.786151
0.786151
0.786151
0.786151
sum 0.786151
Fig. 1. Plots of G( ) and I( ) against from TCSA truncated to level 12. The upper set of lines are
the (symmetric) functions G( ), and the lower set of lines are the (asymmetric) functions I( ) for
b = 1/2, h = 0. These are plotted for r = 2n/2 with 2 6 n 6 8. As r increases, the functions G( )
and I( ) both approach universal functions which we take to be the expectation values of M 2/5 (x)
on a half-line with boundary conditions 1 and (0) respectively. For large r, truncation effects start
to intervene, as can be seen in the slight ripple discernible for n = 8.
636
To show the typical behaviour of these quantities, we give the values of fn(N,r) for r = 8
and 4 6 N 6 12 in Table 1, and for N = 12 and 2 6 r 6 12 in Table 2. In Fig. 1 we plot
G (12,r)( ) for fixed truncation level N = 12 and for varying values of r. We see that on
increasing r, G (12,r)( ) approaches a universal form until truncation effects take over and
the TCSA approximation breaks down.
There are two sources of error in the TCSA estimate G (N,r) ( ). Firstly, TCSA gives
the function G (N,r) ( ) ((2l/) sin(/ l))2/5 in Eq. (3.11) as a Fourier series truncated
at the 2N -th term, which leads to the usual errors associated with truncation of Fourier
(N,r)
appearing in the truncated Fourier series are only
series. Secondly, the coefficients fn
approximately calculated.
3.2.2. The strip with ((h), 1) boundary conditions
This case is only slightly more complicated. The unperturbed expectation value is again
given in (2.7). The TCSA estimation of the expectation value in the massive model with
boundary perturbations then takes the general form

I (N,r) ( ) M 2/5 (x)

((h),1) N
2N 2

2r 2/5 X X (N,r)
n
=
cos
gnj (h)fj
r
r
n=0 j =1

n
0
sin
,
(h)f
+ h(N,r)
j
nj
r
r
(3.12)
where h is related to the reflection factor parameter b by h = h(b), and where the functions
(N,r)
of the
(h) and h(N,r)
(h) have to be evaluated numerically. Since the ground state |0i
gnj
nj
TCSA Hamiltonian lies in the h = 1/5 representation, the matrix elements of are given
in terms of the two chiral blocks f1 , f2 ; furthermore, since this representation has a null
state at level 2, one can eliminate all states with more than one mode L1 , which leads to
the fact that one does not need to use higher than the first derivative of the chiral blocks
(N,r)
(N,r)
, hn , we shall not give any
in (3.12). Since there are rather many coefficients gn
explicit examples.
In Fig. 1 we also plot I( ) against for various values of l between 0.5 and 16, for the
fixed value h = 0, b = 1/2. The excellent agreement between I( ) and G( ) that can
be seen on the half of the strip r/2 < < r for the larger values of r is a good sign that
the TCSA estimates of the functions are converging to their correct values. Table 3 below
includes results on the convergence in N of I (N) ( ) for values of between 0.01 and 1.0 .
3.2.3. The strip with ((hl ), (hr )) boundary conditions
The calculation of the expectation value of on the strip with two perturbed boundary
conditions ((hl ), (hr )) is in principle the same, except that the functional form is rather
more involved. The unperturbed expectation value is again given in (2.7),
2/5

R
x
x
() 1
()
B f1
+ B C f2
(3.13)
(x) ((0),(0)) =
R
R
637
where the functions fi ( ) are the strip chiral blocks (2.6). However, the massive
perturbation introduces terms proportional to the other two chiral blocks and their
derivatives, so that the TCSA estimation of the expectation value in the massive model
with boundary perturbations takes the general form

H(N,r) ( ) = M 2/5 (x) ((h ),(h )) N
r
l

2/5 X

4
2N X
n
r
(N,r)
cos
=
jkn (hl , hr ) fk
r
r
n=0 k=1
(N,r)
+ kkn (hl , hr ) fk0
n
sin
r

,
(3.14)
(N,r)
(N,r)
(hl , hr ) and kin
(hl , hr ) are
where hl = h(bl ), hr = h(br ) and the functions jin
evaluated numerically. Again, since there are rather many coefficients, we shall not give any
explicit examples. For large values of r, the two boundaries are essentially non-interacting
this was already seen in Fig. 1. We therefore see no new phenomena over those seen
already; the TCSA estimates of hM 2/5 (x)i near the left boundary from the system with
boundary conditions ((hl ), (hr )) are barely distinguishable from those from the system
with boundary conditions ((h), 1). Further confirmation is contained in Table 3, which
presents data from the two situations.
However, for small r, the presence of two perturbed boundaries can destabilise the
vacuum even for values of the parameters hl and hr which are less negative than |hcrit |,
the value for which a single boundary destabilises the bulk vacuum on a half line [5].
For bl + br = 0, |bl | < 2, the ground state and first excited state have an exact crossing
at a finite value of r, while for bl + br > 0 there is a finite range of r for which they
become complex. We discuss this in Section 6.2, where we give examples of the spectra,
and plots of the boundary field expectation values hM 1/5 i, for systems with two perturbed
boundaries.
3.3. The comparison of the TCSA and form-factor results

The form-factor method gives the expectation value of (x) on the half-plane. While
we can think of the half-plane the infinite-width limit of a finite strip, this is not accessible
directly using the TCSA method, which is limited to strips of finite width r and is expected
to perform best near to r = 0. To enable a comparison of the two methods, we shall simply
take the TCSA results on a strip of width r = 12, rather than extrapolating finite width
TCSA results to infinite width. (The error for small values of from taking TCSA results
at r = 12 should be much less than the one-particle FF contribution in the middle of the
strip, which is 0.2%.)
In Fig. 2 we show results for two cases: the boundary conditions (1) and ((0)). We
give the form-factor expansion (with g = g/2) truncated to one-, two- and three-particle
states, and the TCSA data from truncation to level 16 (with r = 12), and they are clearly
in excellent agreement. We also show the form-factor expansion assuming g = g, and it is
638
Fig. 2. Comparisons of hM 2/5 (x)i on a half-plane with boundary condition (1) (upper lines) and
((0)) (lower lines). The points are the TCSA data, the dashed lines the FF result up to 1 particle
with g = g/2, the dot-dashed line the FF result up to 2 particles, the solid line the FF result up to 3
particles. The dotted lines are the FF results up to 3 particles with g = g.
Table 3
The TCSA and FF estimates of h( )i(0) , for varying distance from the boundary from the formfactor approach truncated at particle number n and from TCSA truncated to level N with r = 12 on
strips with b.c.s ((0), 1) (upper line) and ((0), (0)) (lower line)
TCSA truncation level N
0.01
1.19134
1.16788
0.03
FF truncation level n
12
16
1.16743 1.16362
1.16261 1.16234
1.16306
1.16239
1.61719
1.21289
1.14851
1.35236
1.32317
1.32259 1.31786
1.31662 1.31627
1.31716
1.31633
1.60971
1.34554
1.31311
0.1
1.48965
1.45280
1.45199 1.44604
1.44456 1.44414
1.44517
1.44421
1.58467
1.45438
1.44451
0.3
1.52882
1.48563
1.48386 1.47722
1.47625 1.47586
1.47633
1.47592
1.52209
1.47810
1.47657
1.0
1.40584
1.37786
1.37064 1.37158
1.37380 1.37424
1.3738
1.37447
1.37977
1.37549
1.37546
639
Fig. 3. Plots of log |hM 2/5 (x)i1 | vs. log( ). The points are TCSA data, the dashed, dot-dashed and
solid lines are from the FF expansion truncated to one, two and three particles respectively, and the
dotted line is (3.16).
clear that this is wrong. In Table 3 we give some numbers illustrating the convergence
in TCSA truncation level and in form-factor truncation level for various values of
0.01 6 6 1.
The boundary condition (0) (b = 1/2) was chosen because it is for this value of b
that the accuracy of the TCSA is highest; comparisons between FF and TCSA results for
the further values b = 1 and b = 0 can be found in Fig. 10, in Section 6.1 below. The
curve there for b = 0 is particularly interesting, as this is the case, mentioned at the end of
Section 2, for which the only difference between the (h) and the 1 boundary states is the
sign of the odd-particle-number contributions.
The two- and three-particle form factor expressions in Fig. 2 are barely distinguishable
from the TCSA data, and so in Figs. 3 and 4 we plot log(hM 2/5 (x)i) against log( ) for
the 1 and (0) boundary conditions. We also show the leading behaviour

2/5
M (x) (0) = (2 )1/5 ()B hi(0) + (2 )2/5 ()B1 h1i(0) + O 12/5 , (3.15)

2/5
(3.16)
M (x) 1 = (2 )2/5 (1)B1 h1i1 + O 12/5 ,
of small- expansions obtained from a perturbative treatment of the structure functions.
(We intend to report on this approach elsewhere [9].) We see that the three-particle FF
approximation already agrees very well with these expansions for 3 . log . 2.
4. Excited states and energy density
The TCSA and FF methods are not restricted to the ground state expectation values of
the field (x, y). In this section we give a couple of examples of their wider applicability.
640
Fig. 4. Plots of log |hM 2/5 (x)i(0) | vs. log( ). The points are TCSA data, the dashed, dot-dashed
and solid lines are from the FF expansion truncated to one, two and three particles respectively, and
the dotted line is (3.15).
4.1. The expectation value of in the first excited state

as to find the
It is just as easy in the TCSA to find the first excited eigenstate |1i
ground state and to find the corresponding expectation value of . The result is somewhat
less accurate than in the ground state, and this accuracy decreases as higher and higher
excited levels are considered. In Fig. 5 we show (as points) the TCSA result near
the (0) boundary of the strip with boundary conditions ((0), 1) with r = 14 and
truncation level 14. This
state corresponds to the boundary bound state of energy e1 =
M cos((b + 1)/6) = 3M/2, and so we expect that near the (0) boundary the
expectation value is approximately given by the expectation value in the first excited state
for the semi-infinite geometry. It turns out that this expectation value can also be obtained
using FF techniques, with results that are shown in the various curves on the figure. The
calculation relies on an idea of analytic continuation between states; similar methods were
used to find TBA equations for excited states in [7].
The fact that the reflection factors Rb ( ) obey a curious continuation property was
already observed in [8]. While the physical parameter h(b) is invariant under under b
4 b, the reflection factor Rb is not; instead, it is interchanged with the reflection factor of
the first boundary bound state. This suggests that the expectation value of the field in the
first excited state can be obtained by continuing the FF expression (3.7) from the domain
3 < b < 2 to the region 2 < b < 5.
In Fig. 5 the result of the substitution of b = 9/2 in (3.7) is shown as a dotted line.
Clearly, it is a long way from the corresponding TCSA results for hi(1) . The explanation
is simple: there are poles in Kb ( ) whose positions depend on b, in particular at =
i(2b)/6 and = i(4b)/6. These two pairs of poles cross the integration contour
641
Fig. 5. Comparisons of hM 2/5 (x)i(0) in the first excited state. The points are the TCSA data from
truncation level 14 and r = 14. The dotted line is the FF expression (3.7) with b = 9/2; the dashed
line is the correction by the terms in (4.2) from the poles at = (2b)/6, and the solid lines are
the results of including all corrections by the addition of the terms up to (4.2), up to (4.3) and up
to (4.5), respectively.
(the real axis) at b = 2 and b = 4 respectively, and contributions from both pairs must
be added in explicitly to recover the correct analytic continuation in b of hi to b = 9/2,
corresponding to h = 0.
We can regard the contributions from these poles as directly affecting the exponential
in (3.5). We denote the positions of the active poles (those which have crossed the
integration contour during the continuation) by i (b), and the contribution to the contour
integral from the pole in K ( ) at i (b) by ki (b) this will be 1 times the relevant
residue, depending on whether the contour was crossed from above or below when the
pole became active. Then we can associate the following state in the full-line Hilbert space
with the first excited state on the half-line:
"
X
0
B = exp g A(0) + i
ki (b)A(i (b))A(i (b))
2
i
#
Z
d
K ( )A( )A( ) |0i.
(4.1)
+
4
Expanding this out and inserting the appropriate form factors, the first corrections to the
expectation value coming from the residue terms are
iX
ki (b)h0|(0)|i (b), i (b)ie2Mx cosh i (b) ,
2
(4.2)
i g X
ki (b)h0|(0)|0, i (b), i (b)ieMx(1+2 coshi (b)) ,
2
(4.3)
642
1X
ki (b)kj (b)h0|(0)|i (b), i (b), j (b), j (b)ie2Mx(coshi (b)+cosh j (b)),
4
i,j
(4.4)
Z
d
iX
K ( )h0|(0)|i (b), i (b), , ie2Mx(cosh i (b)+cosh ) , (4.5)
ki (b)
2
4
where the last two terms can be seen as coming from poles in the four-particle contribution
to the boundary state which hitherto we have neglected. Also, it turns out that the term (4.4)
always gives zero, due to the particular relative positions of the poles at 1 and 2 .
We have shown the result of correcting (3.5) by the dominant correction (the term in (4.2)
coming from the poles at = (2 b)/6) as a dashed line on Fig. 5, and the result of
adding all terms up to (4.2), up to (4.3) and up to (4.5) as solid lines. It is clear that these
are converging rapidly to the TCSA value.
4.2. The expectation value of the energy density
As a second example, we use the TCSA to find the expectation values of the energy
density (x) on the strip for which
ZR
H=
(x) dx,
(x) =

1
T (x) + TS(x) + (x).
2
(4.6)
One has to be rather careful about the specification of the operators in this expression. Here,
we mean by T (x) and TS(x) the bare TCSA quantities in other words, their expectation
values are computed in any given state using the matrix elements of the CFT operators T
and TS between the eigenstates |ni
of the perturbed Hamiltonian, themselves expanded in
the basis of CFT states. This is the same procedure as was used for computations of h(x)i
earlier, but some new issues arise in this case, to which we hope to return in [9].
Leaving these questions to one side, the operator allows us to see directly that
boundary bound states are indeed localised at the boundary. In Fig. 6 we plot the
difference in the energy density between the first few excited states and the ground state for
the system with boundary conditions ((0), 1) for r = 8, calculated using TCSA truncated
to levels 12, 14 and 16. The result of truncating the Fourier-like expansions is very evident
here, the TCSA estimates having distinct high-frequency ripples (varying with truncation
level) superposed on the overall function.
From the analysis in [5,8], for large r the first excited state is a boundary bound state,
and the next several excited states are single-particle scattering states. This is borne out
by Fig. 6, where we see very clearly that the first excited state corresponds to a particle
trapped on the left ((0)) boundary and decaying exponentially across the strip, while the
higher excited modes are well spread across the strip, attracted to the (0) boundary and
repelled by the 1 boundary.
It is impossible to decide whether a boundary (B ) is repulsive or attractive purely given
the reflection factors R ( ). The (h(0)) and 1 boundaries have the same reflection factor
but from Fig. 6, the (h(0)) boundary is attractive and the 1 boundary repulsive. This is the
Fig. 6. Plots of M 2 (h( )in h( )i0 ) vs. for the first

few excited states on the system with boundary conditions
((0), 1) and r = 8. The solid line is the 1st excited state,
the dash-dotted line the 2nd, and the dotted line the 3rd.
For each line, TCSA results truncated to levels 12, 14 and
16 are superimposed to give an idea of the errors.
643
Fig. 7. An attractive and a repulsive scattering process with the

same classical time delay.
quantum analogue of the inability in a classical theory to determine whether a boundary is

attractive or repulsive given only the time-delays the two processes illustrated in Fig. 7
have the same time delay but clearly one describes attraction to the boundary and the other
repulsion.
5. The one-point functions of the boundary field

We now turn to the one-point functions involving the boundary perturbing field . In
preparation for the main calculations, consider first the partition function Z of the model
on a cylinder of length R and circumference L, with boundary conditions and imposed
at the two ends. Formally, we can write
Z
(5.1)
Z = [D ]eABLY ,
R
where [D ] implies that a functional integral over all bulk and boundary degrees of
freedom is taken, and ABLY denotes the combined bulk and boundary action. For the pair
of boundary conditions (, ) = ((hl ), (hr )), this can be written as
ZR
ABLY = ABCFT +
ZL
dy (x, y) + hl
dx
0
ZL
ZL
dy (0, y) + hr
dy (R, y), (5.2)

0
where ABCFT is an action for the M2,5 conformal field theory on the cylinder with
conformal boundary condition at the two ends. (For the other two pairs of boundary
conditions the expression is similar, but lacks one or both perturbing boundary fields.)
The behaviour of Z as a function of R and L is complicated, but if both are much
larger than all bulk and boundary scales, then, up to exponentially-small corrections,
log Z RLEbulk Lf Lf ,
(5.3)
644
Fig. 8. The L-channel decomposition: states

|n i live on the dotted line segment across
the cylinder.
Fig. 9. The R-channel decomposition: states |n i

live on the dotted circle around the cylinder.
where f and f are the extensive parts of the boundary free energies, and Ebulk
the extensive
part of the bulk free energy. For the scaling LeeYang model, Ebulk =
M 2 /(4 3 ) [24].
Given Z , the (normalised) one point functions of the field can be simply obtained
by differentiation:
hl/r icyl =
1
log Z .
L hl/r
(5.4)
However, while it was shown in [5] that the partition function was numerically accessible
via the TCSA, at the current state of technology Z (and hence hicyl ) is not directly
computable by means of the TBA. Contact with this exact method can instead be made
in certain limits, and these are best discussed using a Hamiltonian formulation.
In fact, there are two alternative Hamiltonian descriptions of the partition function. In
the so-called L-channel representation the rle of time is taken by L:
Z = TrH(,) eLH (M,R)
X

strip
exp LEn (M, R) ,
=
(5.5)
(5.6)
En spec(H )
while in the R-channel representation the role of time is taken by R:

Z = h| exp RHcirc (M, L) |i
X

h|n ihn |i
exp REncirc (M, L) .
=
hn |n i
(5.7)
En spec(Hcirc )
In (5.7) we have used the boundary states |i and |i, and the eigenbasis {|n i} of
Hcirc , the Hamiltonian propagating states living on a circle of circumference L. The two
decompositions are illustrated in Figs. 8 and 9.
For the rest of this section we will focus on results for the boundary field l on the lefthand end of the cylinder, but with a trivial relabelling it is clear that analogous results for
r can be found.
645
5.1. L-channel decomposition

First we consider the L-channel representation, depicted in Fig. 8. This will enable us to
find exact formulae for the expectation value of on strips of finite width. Introducing the
eigenbasis {|n i} of H(,) we differentiate inside the trace (5.5) to find

1 X hn |l |n i
strip
exp LEn (M, R) .
(5.8)
hl icyl =
Z n hn |n i
Comparing with (5.6),
hn |l |n i
strip
=
En (M, R),
hn |n i
hl
(5.9)
where hn |l |n i/hn |n i is the expectation value of the field l on the left boundary,
taken in the nth excited state.
strip
It was shown explicitly in [5] that, at least for small n, En can be computed using
generalisations of the boundary TBA equations of [16]. For the scaling LeeYang model
this involves a non-linear integral equation for a single function ( ):
X
S( p )
KL( ),
(5.10)
log
( ) = 2r cosh log ( ) +
S(
p )
p
and an associated set of equations for the (possibly empty) set {p , p } of so-called active
singularities (cf. [5,7]):
e(p ) = e(p ) = 1
(p).
(5.11)
Here, r = MR as in earlier sections,

L( ) = log 1 + e
()
1
f g( ) =
2
d 0 f ( 0 )g( 0 ),
(5.12)
and, for the ((h(bl )), (h(br ))) boundary conditions,
log S( ),
(5.13)
with Kb ( ) defined in (3.3), and S( ) the bulk S-matrix (2.10). The number of active
singularities depends on the particular energy level under consideration; for some pairs of
boundary conditions on the strip it is nonzero even for the ground state [5], in contrast to
the situation for the more familiar TBA equations for periodic boundary conditions.
The solution to (5.10) for a given value of r determines the function cn (r):
( ) = Kbl ( )Kbr ( ),
6
cn (r) = 2
Z
d r cosh L( ) + i
strip
in terms of which En
strip
En
K( ) = i

12r X
sinh p sinh p ,
p
(5.14)
(M, R) is
(M, R) = EbulkR + fbl + fbr
cn (r),
24R
(5.15)
646
and, rewriting (5.9) in terms of bl ,

1
dh
5
hn |l |n i
strip
strip
=
En =
En .

1
hn |n i
db
bl
b
|hcrit | cos bl + 2 /5
l
(5.16)
strip
In the expression (5.15), fbl and fbr are R-independent contributions to En (M, R)
from the two boundaries, and Ebulk is the bulk energy per unit length. If the equations for
the particular state under consideration contain active singularities {p , p } whose positions
do not tend to zero as r , then there will be further R-independent contributions
strip
to En (M, R) coming from the second term on the RHS of (5.14) these will be
described shortly. However, for ground states such contributions are always absent, and
so (as anticipated by the notation) fbl and fbr can be identified with the extensive parts of
the boundary free energies as defined in (5.3). The exact values of these quantities were
extracted from the (ground state) TBA equations in [5], both for the (h(b)) and the 1
boundaries, and are

b
31
+ sin
M,
f1 = fb=0 .
(5.17)
fb =
4
6
States |n i lying above the ground state |0 i will generally be separated from |0 i by a
finite energy gap, even at large R. At the level of the TBA, this gap is seen in the presence
of the active singularities, mentioned in the last paragraph, whose positions do not tend to
zero as r . These give cn (r) a linear growth in r = MR, which via (5.15) yields an
strip
extra constant term in the large-R asymptotic of En (M, R). Physically, the gap arises
from two sources: the state |n i may contain a number, k(n) say, of bulk particles bouncing
between the two edges of the strip, and in addition there may be a boundary bound state
sitting at one or both of the boundaries. Suppose that the possible boundary bound states
for the left-hand boundary are indexed by k = 0, 1, . . . , kmax (bl ) with k = 0 the boundary
ground state, and likewise for the right boundary, and that for the state |n i the left and right
boundaries are in states kl (n), kr (n) respectively. Then the general behaviour for R
is as follows:
strip
En
(M, R) EbulkR + Mk(n) + fbl + fbr + ekl (n) (bl ) + ekr (n) (br ),
(5.18)
where ek (b) is the energy of the kth boundary bound state of the (h(b)) boundary
condition (with e0 (b) = 0, since k = 0 is just the boundary ground state).
Taking the limit R of the ground state expectation value allows the boundary
expectation values on the semi-infinite plane to be recovered. For the state 0 , k(0) =
kl (0) = kr (0) = 0, and substituting (5.18) into (5.16) using (5.17), we obtain the exact
value of the dimensionless expectation value hM 1/5 i in a semi-infinite geometry:

cos(b/6)
5
M 1/5 =

.

6 hcrit cos b + 12 /5
(5.19)
For now we take b to be restricted to the fundamental region 3 < b < 2, in which
case the above formula is indeed correct for the ground state. Its interpretation as b moves
outside this region will be described in Section 6.1 below.
647
Table 4
The TCSA estimates, with truncation level N = 14, of hM 1/5 i, for = (0), = 1 in the small
r region compared with the numerical prediction from the TBA
log(r)
8
TCSA
TBA
0.19684202180 0.29365378867 0.4380797960 0.65350591 0.969216

0.19684202181 0.29365378869 0.4380797961 0.65350592 0.969217
Returning to finite values of R, the asymptotic (5.18) no longer suffices to obtain the
expectation value of the boundary field, and one has rather to differentiate the exact
formula (5.15), so that (5.19) becomes
1
cos(bl /6) 4r

5
b c(r)
M 1/5 l (r) =

l ,
1
cos bl + 2 /5
6h crit
(5.20)
where b l c(r) b l c0 (r) can be obtained by differentiating the full TBA equation (5.10) (a
similar idea was applied to the case of periodic boundary conditions in [24]). Restricting,
for simplicity, to the ground state energy in the region where no active singularities are
present, the terms involving the p in (5.10) are absent, and
6
c(r) = 2
bl
Z
d r cosh
( )
,
1 + exp(( ))
where ( ) solves (5.10), while ( ) =

equation
( ) =
bl ( )
(5.21)
can be obtained from the linear integral
( ).
log ( ) + K
bl
1 + exp()
(5.22)
In this way the final estimates for ( ) and ( ) have roughly the same accuracy.
We also estimated hi using the TCSA. For this, we used a strip geometry with
specific boundary conditions (, ) on the two edges. We then calculated the dimensionless
expectation value h M 1/5 l i, (r), as a function of the strip width r for finite truncation
level N . While hM 1/5 l i, can depend strongly on for small r, as r increases, this
dependence decreases, and as r it approaches the half-plane value hM 1/5 i . In
Table 4 we compare the TCSA estimates for hM 1/5 i, for = (0), = 1 and r small
with the numerical solution of the ground state TBA equations (5.9), (5.10), (5.11) and
(5.21), (5.22). A similar agreement was found for other pairs of boundary conditions.
In Table 5 we also give hM 1/5 i for the same boundary conditions for larger values of r
to show the convergence to the IR value, and include several plots of hM 1/5 i for various
boundary conditions in Fig. 14.
648
Table 5
The TCSA estimates of hM 1/5 i(0),1 (r) (upper data) and hM 1/5 i(0),(0) (r) (lower data).
These can be compared with the exact IR value of 1.17459499975 . . .
log(r)
N
3/2
5/2
0.969212
1.321108
1.129791
1.202314
1.166399
1.179004
1.175226
1.174604
1.182737
1.175250
0.969214
1.321116
1.129788
1.202350
1.166291
1.179061
1.174404
1.174612
1.178406
1.174563
10
0.969215
1.321120
1.129795
1.202370
1.166281
1.179099
1.174229
1.174643
1.176402
1.174357
12
0.969216
1.321122
1.129801
1.202382
1.166281
1.179124
1.174121
1.174679
1.175346
1.174337
14
0.969216
1.321124
1.129806
1.175194
1.166288
1.179142
1.174104
1.174709
1.175196
1.174342
5.2. R-channel decomposition

We now turn to the R-channel representation, depicted in Fig. 9. This will ultimately
lead to expressions for expectation values when the boundary field is placed at one end of
a semi-infinite cylinder. We shall use the notation of [6] and denote the boundary states |i
and |i as |(hl )i and |(hr )i, respectively.
If R is taken to infinity in (5.7) with all other variables held fixed, then the contribution
of the ground state |0 i |i will dominate the spectral sum. Thus

(5.23)
Z A (M, L) exp RE0circ (M, L) ,
where E0circ (M, L) is the ground state energy of Hcirc and
A (M, L) =
h|ih|i
.
h|i
(5.24)
If we now let L grow as well and compare with (5.3), we see that the inner products
appearing in (5.24) will, in general, contain a term corresponding to a boundary free energy
per unit length:

h|i
(5.25)
= Lf + log g (M, L) .
log
h|i1/2
On the other hand, this linear term can be extracted from the small-L behaviour of the
functions log(g (M, L)), for which L-channel TBA equations 2 were proposed in [16].
This is explained in [6], where a precise match with the earlier result (5.17) was found.
2 The terminology is unfortunately, but unavoidably, a little confusing these equations are called L-channel
because their derivation proceeds via the L-channel representation of the partition function.
649
The consistency between these two determinations of fb is in some respects a mystery,

since, from other results reported in [6], there are good reasons to doubt the ability of
the L-channel TBA equations of [16] to describe the full variation of log(g (M, L)) as a
function of ML.
Returning to the R-channel decomposition of the full partition function, we have
hn |(hr )i

1 X d
exp REncirc (M, L) .

(5.26)
(hl )n
hl icyl =
LZ n dhl
hn |n i
The following identification was made in the massless [1] and massive [6] cases:

(h(b))n = Yn i b+3
6 h1|n i,
(5.27)
where Yn ( ) = exp(n ( )), and n ( ) is the solution of the excited-state TBA equation
(with periodic boundary conditions) for the state |n i (see [2,7]). Taking the limit R
one deduces that

1 dh 1
h(h)|l |i
=
b log Y i b+3
(5.28)
hl i =
6
h(h)|i
L db
which gives the 1-point expectation value of acting on the end of an infinite cylinder
of circumference L in terms of the function Y . One check on this formula is easily made:
from the large-L limit of the TBA equation for Y , we have

(5.29)
ML sin b
log Y i b+3
6
6 = L(fb f1 ) (L )
using (5.17) in the second equality; differentiating, the expectation value on the upper halfplane quoted in (5.19) is recovered.
There is evidence (see [7]) that the full set of excited Yn s can be obtained from Y
via a process of analytic continuation in the bulk perturbing parameter . It then seems
reasonable to suppose that (5.28) can also be continued, leading to the following general
relation:

1 dh 1
h(h)|l |n i
(n)
=
b log Yn i b+3
(5.30)
hl i =
6 .
h(h)|n i
L db
(Alternatively, one can obtain this simply by differentiating (5.27).)
It turns out that there is a further consequence of the R-channel decomposition. Note
first that (5.27) can be used to write the partition function as
Z(h(br )),(h(bl ))
X h1|n ihn |1i

Yn i br 6+3 Yn i bl 6+3 exp R Encirc (M, L) .
=
hn |n i
n
Now, the Ys satisfy the functional relation [27]

Yn + i 3 Yn i 3 = 1 + Yn ( ),
(5.31)
(5.32)
and this suggests the following identity

Z(h(b+2)),(h(b2)) = Z1,1 + Z1,(h(b)) .
(5.33)
650
This should hold for all M, R and L; from the L-channel decomposition (5.6), it is
equivalent to the following relation between the spectra of models on strips of equal widths
but different boundary conditions:

strip
En (M, R) (h(b+2)),(h(b2))
strip

strip

= En (M, R) 1,1 En (M, R) 1,(h(b)).
(5.34)
Preliminary numerical work confirms this rather surprising identity, but as yet we do not
have a good physical understanding of its origin. However, in Section 6.2 below it will be
used to formulate an exact conjecture concerning the regions of the (b, b0 ) plane for which
the model with ((h(b)), (h(b0 ))) boundary conditions develops a boundary-induced
vacuum instability.
6. Applications
In this section we apply some of the results obtained above to elaborate a few further
aspects of the boundary scaling LeeYang model. We start with the boundary flows of the
semi-infinite system, and then turn to the way that the boundary-induced vacuum instability
of the model is affected when the system is confined to a finite strip.
We first recall the way the parametrisations of the TCSA, TBA and FF calculations are
related. The physical parameter describing the (h) boundary is, of course, h, which is
related to b by

(6.1)
h = |hcrit | sin b + 12 /5 .
This physical parameter h is periodic in b with period 10, and we choose the fundamental
domain to be 3 6 b 6 2. Thus for real b, h is restricted to the range |hcrit | 6 h 6 |hcrit |.
To reach real values of h > |hcrit | we can continue in b to values b = 3 + i b with b real;
to reach real values of h < |hcrit | we can continue in b to values b = 2 + i b with b real.
As we pointed out in Section 4.1, if we formally continue in b outside the fundamental
region we have to be careful, as quantities may not continue naively in particular
the continuation of the fundamental reflection factor to 2 < b < 5 in fact describes the
reflection properties of the boundary with the addition of (the lowest) boundary bound
state.
6.1. Boundary flows on the semi-infinite system
It is quite straightforward to see the boundary flow, at the level of the one-point functions,
using the TCSA. In [5] the spectra for the strip with boundary conditions ((h), 1) were
calculated for several values of h, and they are consistent with the idea that this spectrum
is real for all real h > |hcrit |. Thus we can calculate the expectation values of (x) and
on the boundary (h) by looking at the large r limit of calculations on the strip with bcs
((h), 1). The only restriction is that the TCSA errors increase sharply with |h|, so that
TCSA results are restricted to a small range of b values centred on b = 1/2.
651
The FF and exact (TBA) results have more interesting properties. Recall that the FF
calculations are formally functions of

tan((b + 2)/12)
g1 ,
Kb ( ) = Rb i 2 .
(6.2)
g (b) =
tan((b 2)/12)
The continuation to large positive value of h through b = 3 + i b works well, as one can
easily verify that none of the b-dependent poles in Kb ( ) cross the integration contours in
the FF integrals (3.7), and so none of the subtleties described in Section 4.1 above arise.
The formulae
= g1 ,
lim g (3 + i b)
lim K(3+i b)
( ) = K0 ( ) = K1 ( )
(6.3)
can therefore be substituted directly into the form factor expansion, establishing the result

(6.4)
lim M 2/5 (x) (h) = M 2/5(x) 1
h+
as an exact identity. One can similarly discuss the continuation of the expectation value of
the boundary field, finding, as expected,

1/5
1.
(6.5)
lim M 1/5 (h(3+i b))
=0= M
To illustrate the result (6.4), in Fig. 10 we plot hM 2/5 (x)i close to the (h) boundary
for various values of h. The FF results smoothly interpolate between the 1 and (h)
boundaries, and give good agreement with the TCSA results for small values of |h|. The
h limit is not directly accessible in TCSA, but we have extrapolated our results in
h and we can see that (modulo an amplification of the Fourier-type truncation errors) it
shows every sign of converging on the expectation value in the 1 bc.
As was discussed in [5], for h > h(1), in the massive system on the half-plane the
(h) boundary has no boundary bound states, and flows to the 1 boundary condition as
h +. For h(1) < h < h(1) this boundary has one bound state, and for |hcrit | =
h(2) < h < h(1) there are two boundary bound states; at h = |hcrit | the ground state and
first excited states become degenerate and for h less than this critical value |hcrit |, the
system does not have a real vacuum. The presence of the bound states can be understood
as particles being trapped near the boundary, as we have seen in Section 4.2.
The continuation of the FF results beyond b = 2 was already discussed in Section 4.1.
The boundary expectation values can also be continued, and here we find a repetition of
the reflection factor results the ground state and first excited state expectation values
swap under b 4 b, and the second excited state is invariant. To recall, the boundary
free energy and the excitation energies of the two lowest lying states (for 1 6 b 6 2) are

31
+ sin(b/6) M,
fb =
4

e2 (b) = M cos (b 1)/6 .
(6.6)
e1 (b) = M cos (b + 1)/6 ,
This gives (cf. Eqs. (5.16) and (5.18)) the expectation value hM 1/5 i(n) in these three
lowest lying states as
652
Fig. 10. Plots of hM 2/5 (x)i(h(b)) with boundary condition (h(b)). The solid lines are the
form-factor results truncated at three particles with b taking values (from second line from top to
bottom) 3 + 3i, 2, 1, 1/2, 0, 1/2. The points are the TCSA results I( ) with truncation level
14 and r = 8. We have also included a plot of hM 2/5 (x)i1 (top line), the extrapolation of the TCSA
data to h = (long dashed line) and the bulk value (short dashed line) for comparison.

cos(b/6)
5
,
6|h crit | sin((b 2)/5)

1/5 (1)
5
cos((4 b)/6)
d
,
= M 1/5 (fb + e1 ) =
M
dh
6|h crit| sin((b 2)/5)

1/5 (2)
5 3 sin((b 2)/6)
1/5 d
(fb + e2 ) =
.
=M
M
dh
6|h crit| sin((b 2)/5)
M 1/5
(0)
= M 1/5
d
fb =
dh
(6.7)
Note that hi(0) = hi(1) at the threshold for the first excited state b = 1, and hi(0) =
hi(2) at the threshold for the second excited state b = 1.
Unfortunately the TCSA does not give very good results for the expectation values in
the excited states for large r, and while extrapolations in r and truncation level indicate
that these results are indeed correct, there seems little point in showing any plots.
6.2. RG flows on the finite size strip
In [5] the spectrum of the model with (1, ) boundary conditions was described in some
detail, but results for the (, ) system were not presented. The analysis of these results
provides a nice application of the spectral identity found in Section 5.2 above, and in this
section we describe how this goes. We begin with some results from the TCSA.
In Figs. 1113 the finite size spectrum is plotted for the system on a strip with boundary
conditions ((h(bl )), (h(br ))) with bl = br = b taking values 1/2, 0, 1/2. Observe
that the ground and first excited states cross for b = 0, and for b = 1/2 an interval appears
in which the ground state has left the real spectrum. However, so long as h > |hcrit | the
large-r spectrum stays real.
653
Fig. 11. The spectrum of the model on a strip with boundary conditions bl = br = 1/2, plotted
against r.
Fig. 12. The spectrum of the model on a strip with boundary conditions bl = br = 0, plotted against r.
In Fig. 14 we plot the log of the expectation value of the boundary field hM 1/5 i,
against log(r), for the two pairs of boundary conditions (, ) = ((h(b)), 1) and (, ) =
((h(b)), (h(b))) for b taking the values 1/2, 0, 1/2. For large r, the expectation
values in the two boundary conditions converge to the same value, as we expect; for large r
the influence of the right boundary on the left boundary decreases and the expectation value
tends to the half-plane value, hl i, hi , independent of (provided that the system
r
is not destabilised by the boundary condition ). Conversely, for small r, hi, tend to the
654
Fig. 13. The spectrum of the model on a strip with boundary conditions bl = br = 1/2, plotted
against r. The dashed lines indicate the real part of a pair of complex conjugate eigenvalues.
Fig. 14. Plots of log(hM 1/5 i, ) against log r for pairs of boundary conditions ((h(b)),
(h(b))) and ((h(b)), 1) for b = 1/2 (long dashed lines), b = 0 (solid lines) and b = 1/2
(short dashed lines). Also shown (dotted lines) are the conformal expressions (2.8). See text for
more details.
conformal limits (2.8) which are governed by the UV fixed point of the boundary flow. On
Fig. 14 the conformal limits are shown as dotted straight lines the expectation values
for (, ) = ((h(b)), 1) all converge to the lower straight line in the UV, and those for
(, ) = ((h(b)), (h(b))) all converge to the upper line.
655
The most interesting behaviour is that shown by hM 1/5 i(h(b)),(h(b)) for intermediate
values of r. For b = 0 we saw in Fig. 12 that the ground state and first excited states cross
at one point; at the same point hi diverges with a characteristic behaviour. For b = 1/2
the ground state drops out of the real spectrum for a finite range of r; hi diverges at the
end points of this range.
The appearance of level crossing can be understood by making use of the spectral
identity (5.34). Shifting b, and using the symmetry of the model about b = 2, the identity
can be written as

strip
En (M, R) (h(b)),(h(b))
strip

strip

= En (M, R) 1,1 En (M, R) 1,(h(2b)) .
(6.8)
Consider now the two ground state energies which appear on the RHS of this equation.
Depending on their relative values, one or other will correspond to ground state of the
model on the LHS. Two limits are easily analysed. As R 0, the fact that the (1, )
conformal Hilbert space contains the field while the (1, 1) does not shows that
strip
R 0: E0 (1,1) +,
strip
(6.9)
E0 (1,(h(2b))) .
In the opposite, large-R, limit, the behaviours follow from (5.18):

strip
R : E0 (1,1) EbulkR + 3 1 M,

strip
M.
E0 (1,(h(2b))) EbulkR + 3 1 + sin (2b)
6
(6.10)
In the latter formula, without loss of generality, we took b to be positive. Comparing (6.9)
with (6.10) shows that the relative values of the two ground states on the RHS of (6.8)
swap over when going from small to large values of R whenever b is less than 2. Since
these states can be identified with the two lowest-lying levels of the (, ) model on the
LHS of (6.8), we see that a level-crossing in this model is inevitable in all such cases.
Thus the point (0, 0) in the (bl , br ) plane, shown in Fig. 12, belongs to a whole line of
points (b, b), |b| < 2 which also exhibit a level crossing. The fact that the relevant two
states are taken from the spectra of distinct models on the RHS of (6.8) prohibits their
mixing and ensures that the crossing will be exact. Once the line bl + br = 0 is left, the
identity (6.8) can no longer be invoked and the exact level crossing is lost, as can be
seen in Figs. 11 and 13. Observe in the first of these the lowest two levels remain real,
while in the second there is an intermediate range of R for which their energies become
complex. Physically this can be explained as follows: in the first situation, bl + br < 0
and the boundary fields are less strong than in the marginal case of bl + br = 0, and
hence have less chance to destabilise the model; and in the second the story is reversed,
the boundary fields are stronger, and there is therefore a possibility of a vacuum instability
for some finite values of R. Once this has been understood it is reasonable to conjecture
that for all points (bl , br ) in the fundamental domain 3 6 bl , br 6 2 with bl + br > 0
the model with ((h(bl )), (h(br ))) boundary conditions exhibits a vacuum instability
656
Fig. 15. The fundamental region in the (bl , br )

plane, showing (shaded) the region within which,
for at least one value of the strip width, the model
exhibits a boundary-induced instability.
Fig. 16. As Fig. 15, but plotted on the (hl , hr )

plane. The inner dashed line, a symmetrically-placed square of size 2hcrit , indicates the
extent of the region covered for real values of
bl and br .
for some range of system sizes, while for the points below this line, the spectrum remains
entirely real at all values of R.
This picture is confirmed by TCSA plots analogous to Figs. 1113 taken at various
other values of bl and br , and leads to the phase diagram for the model shown in
Figs. 15 and 16. The change in coordinates from (bl , br ) to (hl , hr ) in passing between
the two figures transforms the line segment (b, b), |b| < 2 into the portion of the ellipse
+(hl hr )2 / cos2 10
= h2crit on Fig. 16 which touches the (shaded) region
(hl +hr )2 / sin2 10
of instability. (Along the rest of the ellipse, the spectral identity (6.8) also holds, but does
not imply a level crossing.) Note also that whenever either hl < hcrit or hr < hcrit , the
vacuum is already unstable in infinite volume and so this region can immediately be shaded
in, without the need to appeal to more subtle arguments. No markedly new features emerge
in the spectra for bl 6= br , so we will not show any further plots. However, we note that the
value of R at which the level-crossing occurs for models on the line bl + br = 0 diverges
as bl (or br ) tends to 2. This is as one would expect, since in this limit the value of one of
the boundary fields is approaching |hcrit |, and the corresponding boundary bound state is
becoming degenerate with the vacuum in infinite volume.
7. Conclusions
In this paper we have given a rather detailed analysis of the one-point functions of both
bulk and boundary fields in a simple but nonetheless non-trivial integrable quantum field
theory, the scaling LeeYang model. A number of different techniques have been explored,
with results which have been shown to be in good accord. Previous work on this topic
657
has tended to use just one method (most usually the form-factor expansion) and thus the
consistency that we have found with between the different approaches is an important
confirmation that the earlier studies have been well-founded. For this purpose the use of
the scaling LeeYang model as a testing-ground is very natural, but we feel that it would
be very worthwhile to extend this work to further theories, and work on such matters is
currently in progress.
There are a number of open questions that arise. The modification to Ghoshal and
Zamolodchikovs boundary state that we were forced to make in order to reconcile FF and
TCSA results remains a numerical observation, for which we have no terribly compelling
physical argument. Furthermore, in Section 4.2 we noted that the boundary states could
sometimes be analytically continued, yielding excited boundary states associated with
the boundary bound states of the semi-infinite line. A systematic understanding of offcritical boundary states, both in infinite and finite geometries, is still lacking and would
presumably shed some light on the questions raised by our observations. It also might help
to understand at a more fundamental level how the modifications to the boundary TBA
equations of [16], found in [5] by an indirect method of analytic continuation, arise.
It would also be worth investigating boundary-to-boundary correlation functions such as
Z Z

1
(7.1)
hl ihr i.
l
r
L2
These are accessible by a simple generalisation of the techniques explained in Section 6.1
above, and might be amenable to comparison with, for example, the results of lattice simulations. We also remark that the one-point functions at the end of an infinite cylinder calculated in Section 6.2 are essentially the analogues, for the boundary fields of a semi-infinite
system, of the finite-temperature expectation values discussed in [15,18]. In the model we
discussed these were relatively easy to obtain as the field under consideration was also the
boundary perturbing operator. For more general fields in more complicated models it would
presumably be necessary to develop the discussion more along the lines of the work in [18].
This seems to us an important open problem which should certainly be investigated further.
Finally, we remark that the partition function identity discovered in Section 6.2 merits
further study. In some senses it can be considered as a first off-critical extension of
the identities between (sums of) conformal partition functions that can be observed on
examining the lists of such objects provided in, for example, [3]. A physical understanding
of why such identities should exist is lacking, even in the conformal cases, and perhaps
the broader perspective provided by the off-critical results will help towards this end. This
alone should motivate the extension our work on off-critical boundary integrable models
to further examples.
Acknowledgements
The work was supported in part by a TMR grant of the European Commission, contract
reference ERBFMRXCT960012, in part by a NATO grant, number CRG950751, and in
part by EPSRC grants GR/K30667 and GR/L26216.
658
P.E.D. and G.M.T.W. thank the EPSRC for Advanced Fellowships, MP thanks the
EPSRC for support and RT thanks the Universiteit van Amsterdam for a post-doctoral
fellowship.
We would like to thank Ph. Di Francesco, A. Leclair, G. Mussardo, I. Runkel, G. Takcs,
Al. Zamolodchikov and J.-B. Zuber for helpful discussions. G.M.T.W would also like to
thank R. Guida and N. Magnoli for many discussions of TCSA methods, Brian Davies for
very helpful discussions on the numerical analysis of the results, and SPhT for hospitality
during the final stages of this work.
References
[1] V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable structure of conformal field
theory, quantum KdV theory and thermodynamic Bethe ansatz, Commun. Math. Phys. 177
(1996) 381398, hep-th/9412229.
[2] V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable quantum field theories in finite
volume: excited state energies, Nucl. Phys. B 489 (1997) 487531, hep-th/9607099.
[3] R.E. Behrend, P.A. Pearce, J.-B. Zuber, Integrable boundaries, conformal boundary conditions
and A-D-E fusion rules, J. Phys. A 31 (1998) L763L770, hep-th/9807142.
[4] J.L. Cardy, G. Mussardo, S matrix of the YangLee edge singularity in two dimensions, Phys.
Lett. B 225 (1989) 275278.
[5] P. Dorey, A. Pocklington, R. Tateo, G. Watts, TBA and TCSA with boundaries and excited
states, Nucl. Phys. B 525 (1998) 641663, hep-th/9712197.
[6] P. Dorey, I. Runkel, R. Tateo, G. Watts, g-function flow in perturbed boundary conformal field
theories, Nucl. Phys. B 578 (2000) 85122, hep-th/9909216.
[7] P. Dorey, R. Tateo, Excited states by analytic continuation of TBA equations, Nucl. Phys. B 482
(1996) 639659, hep-th/9607167;
P. Dorey, R. Tateo, Excited states in some simple perturbed conformal field theories, Nucl.
Phys. B 489 (1998) 575623, hep-th/9706140.
[8] P. Dorey, R. Tateo, G. Watts, Generalisations of the ColemanThun mechanism and boundary
reflection factors, Phys. Lett. B 448 (1999) 249256, hep-th/9810098.
[9] P. Dorey, R. Tateo, G. Watts, in preparation.
[10] V.A. Fateev, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Expectation values of
local fields in BulloughDodd model and integrable perturbed conformal field theories, Nucl.
Phys. B 516 (1998) 652674, hep-th/9709034.
[11] S. Ghoshal, A.B. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional
integrable quantum field theory, Int. J. Mod. Phys. A 9 (1994) 38413886, erratum;
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353, hep-th/9306002.
[12] R. Guida, N. Magnoli, Vacuum expectation values from a variational approach, Phys. Lett.
B 411 (1997) 127133, hep-th/9706017.
[13] R. Konik, A. Leclair, G. Mussardo, On Ising correlation functions with boundary magnetic
field, Int. J. Mod. Phys. A 11 (1996) 2765, hep-th/9508099.
[14] A. Leclair, F. Lesage, H. Saleur, Exact Friedel oscillations in the g = 1/2 Luttinger liquid, Phys.
Rev. B 54 (1996) 1359713603, cond-mat/9606124.
[15] A. Leclair, G. Mussardo, Finite temperature correlation functions in integrable QFT, Nucl.
Phys. B 552 (1999) 624642, hep-th/9902075.
[16] A. Leclair, G. Mussardo, H. Saleur, S. Skorik, Boundary energy and boundary states in
integrable quantum field theories, Nucl. Phys. B 453 (1995) 581618, hep-th/9503227.
659
[17] F. Lesage, H. Saleur, Form-factors computation of Friedel oscillations in Luttinger liquids,

J. Phys. A 30 (1997) L457L463, cond-mat/9608112.
[18] S. Lukyanov, Finite temperature expectation values of local fields in the sinh-Gordon model,
hep-th/0005027.
[19] G. Mussardo, Sprectral representation of correlation functions in two-dimensional quantum
field theories, Talk given at the international colloquium on modern QFT II, Tata Institute,
Bombay, January 1994, hep-th/9405128.
[20] M. Pillin, The form factors in the sinh-Gordon model, Int. J. Mod. Phys. A 13 (1998) 4469
4486, hep-th/9712033.
[21] F.A. Smirnov, The perturbated C < 1 conformal field theories as reductions of sine-Gordon
model, Int. J. Mod. Phys. A 4 (1989) 42134220;
F.A. Smirnov, Reductions of the sine-Gordon model as a perturbation of minimal models of
conformal field theory, Nucl. Phys. B 337 (1990) 156180.
[22] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory,
Adv. Series in Math. Phys., Vol. 14, World Scientific, Singapore, 1992.
[23] V.P. Yurov, Al.B. Zamolodchikov, Truncated conformal space approach to the scaling Lee
Yang model, Int. J. Mod. Phys. A 5 (1990) 3221.
[24] Al.B. Zamolodchikov, Thermodynamic Bethe ansatz in relativistic models. Scaling 3-state Potts
and LeeYang models, Nucl. Phys. B 342 (1990) 695720.
[25] Al.B. Zamolodchikov, Two point correlation function in scaling LeeYang model, Nucl. Phys.
B 348 (1991) 619.
[26] Al.B. Zamolodchikov, Mass scale in sine-Gordon model and its reductions, Int. J. Mod. Phys.
A 10 (1995) 11251150.
[27] Al.B. Zamolodchikov, On the thermodynamic Bethe ansatz equations for the reflectionless
ADE scattering theories, Phys. Lett. B 253 (1991) 391394.

The scaling supersymmetric YangLee model with

boundary
Changrim Ahn a , Rafael I. Nepomechie b,
a Department of Physics, Ewha Womans University, Seoul 120-750, South Korea
b Physics Department, P.O. Box 248046, University of Miami, Coral Gables, FL 33124 USA
Received 2 October 2000; accepted 2 November 2000
Abstract
We define the scaling supersymmetric YangLee model with boundary as the (1, 3) perturbation
of the superconformal minimal model SM(2/8) (or equivalently, the (1, 5) perturbation of the
conformal minimal model M(3/8)) with a certain conformal boundary condition. We propose
the corresponding boundary S matrix, which is not diagonal for general values of the boundary
parameter. We argue that the model has an integral of motion corresponding to an unbroken
supersymmetry, and that the proposed S matrix commutes with a similar quantity. We also show by
means of a boundary TBA analysis that the proposed boundary S matrix is consistent with massless
flow away from the ultraviolet conformal boundary condition. 2001 Elsevier Science B.V. All
rights reserved.
1. Introduction
A (1 + 1)-dimensional massive integrable quantum field theory without boundary (i.e.,
on the full line x (, )) is characterized by its factorizable bulk scattering (S)
matrix [1]. It can also be characterized as a perturbation [2] of a bulk conformal field
theory (CFT) [3]. For example, a perturbed minimal model is the renormalization group
infrared trivial fixed point of the action
Z
A = ACFT +
Z
dy
dx (,) (x, y),
(1.1)
where ACFT is the action of a c < 1 minimal model M(p/q), (,) is a spinless
degenerate primary field with (right, left) conformal dimensions (, ) which is relevant
E-mail address: nepomechie@physics.miami.edu (R.I. Nepomechie).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 6 8 - 4
C. Ahn, R.I. Nepomechie / Nuclear Physics B 594 [FS] (2001) 660684
661
( < 1) and integrable, and is a parameter of dimension [length]22 . One link

between these two descriptions is provided by the thermodynamic Bethe Ansatz (TBA),
by means of which the central charge of the CFT can be computed from the S matrix [4,5].
The integer-spin and fractional-spin [6,7] integrals of motion of an integrable field theory
are manifested in both its S matrix and perturbed CFT descriptions. These features of
integrable field theory are by now relatively well understood, due to the great number of
examples which have been worked out in detail. (See, e.g., [8] and references therein.)
For an integrable field theory with boundary (say, on the half-line x (, 0]),
the above framework has a nontrivial generalization [9]. The theory is characterized by
a factorizable boundary scattering matrix, together with the bulk S matrix. It can also be
described as a perturbation of a boundary CFT. The boundary generalization of (1.1) is
given by
Z
A = ACFT+CBC +
Z0
dy
Z
dx (,) (x, y) + B
dy () (y).
(1.2)
The boundary CFT is specified [10] by a conformal boundary condition (CBC), which
for c < 1 minimal models corresponds to a cell (n, m) of the Kac table. A CBC is also
characterized by the so-called boundary entropy or ground-state degeneracy (g) factor [11],
which (roughly speaking) is a measure of the number of bulk vacua which are compatible
with a given CBC. This is well illustrated in the unitary minimal models [9,12]. As can
be seen from (1.2), the boundary CFT in general has perturbations by both bulk ((,) )
and boundary (() ) relevant primary fields. The boundary parameter B has dimension
[length]1 . Note that the boundary perturbation has the same conformal dimension as
the bulk perturbation, and therefore, presumably it is integrable [9]. Furthermore, the
CBC and the boundary perturbation must be compatible [9,10]. By means of a boundary
TBA [1315], ratios of g factors of the boundary CFT can be computed from the bulk
and boundary S matrices. (See also [16,17].) Fractional-spin integrals of motion should
be manifested in both the boundary S matrix and the perturbed CFT descriptions [18].
These features of integrable field theory with boundary have been studied in relatively few
examples and are less well understood, in comparison to the case without boundary.
In an effort to provide more such examples, we consider here the boundary version of
the bulk scaling supersymmetric YangLee (SYL) model [1921]. This model is arguably
the simplest nontrivial supersymmetric quantum field theory. Its spectrum consists of one
Boson and one Fermion of equal mass, and the bulk S matrix is factorizable and has N = 1
supersymmetry. This model is the supersymmetric generalization of the scaling YangLee
(YL) model [4,22,23], which describes the scaling region near the YangLee singularity of
the two-dimensional Ising model [24,25]. The SYL model is the first member of an infinite
family of integrable models with N = 1 supersymmetry [19].
In particular, we define the boundary SYL model as a perturbed boundary CFT, and we
propose the corresponding boundary S matrix, which is not diagonal for general values
of the boundary parameter. We support this picture by identifying a supersymmetry-like
integral of motion, and by studying massless boundary flow using the boundary TBA.
662
Some related work was done by Moriconi and Schoutens in [26]. These authors proposed
two diagonal boundary S matrices for the boundary SYL model, without reference to any
specific boundary conditions. For a special value of the boundary parameter, our boundary
S matrix differs from one of theirs by a CDD factor.
The outline of this article is as follows. In Section 2, we briefly review some necessary
results about the YL model, and we clarify a few subtleties of the boundary TBA. In
Section 3, we review some necessary results about the bulk SYL model. We also recall
the useful observation [27] that the critical SYL model can be formulated as either the
superconformal minimal model SM(2/8) or the conformal minimal model M(3/8). This
is completely analogous to the well-known fact that the tricritical Ising model can be
formulated as either SM(3/5) or M(4/5). One consequence of this fact is that the SYL
model can be regarded, following [28,29], as a restriction of the ZMS model [3032], as
we discuss in an appendix. Section 4 is the heart of the paper. There we first define the
boundary SYL model as a perturbed boundary CFT, and we argue that it has an integral
of motion corresponding to an unbroken supersymmetry. We then propose the boundary S
matrix for the boundary SYL model. Our approach is to restrict the boundary S matrix of
the boundary supersymmetric sinh-Gordon model [15], by imposing the various boundary
bootstrap constraints [9]. We then show that the proposed boundary S matrix commutes
with a supersymmetry-like charge. Finally, we perform a boundary TBA analysis, and
show that the proposed boundary S matrix is consistent with massless flow away from the
ultraviolet conformal boundary condition. In Section 5 we present a brief discussion of our
results.
2. The YL model
We now briefly recall the basic results of the scaling YangLee model which we shall
need in subsequent sections to formulate the supersymmetric generalization. We also
clarify a few subtleties of the boundary TBA.
2.1. Bulk
The critical behavior of the YangLee singularity is described [33] by the minimal
model M(2/5). This is a (nonunitary) CFT with central charge c = 22/5. There are only
two irreducible representations of the Virasoro algebra, and the corresponding conformal
dimensions (n,m) of the primary fields are organized into a Kac table in Table 1.
The scaling YangLee model (without boundary) is defined [22] by the perturbed action
(1.1), where the CFT is M(2/5), and = (1,3) = 1/5. Arguments developed by
Table 1
Kac table for M(2/5)
0
15
15
663
Zamolodchikov [2] imply that this model is integrable. The spectrum consists of a single
particle of mass m, with energy E = m cosh and momentum P = m sinh , where is the
rapidity. The two-particle S matrix for particles with rapidities 1 and 2 is given by [22]

sinh + i sin 2
3
(2.1)
SYL ( ) =
,
sinh i sin 2
3
where = 1 2 . This S matrix has a direct (s) channel pole at = i2/3, since the
particle is a bound state of itself. Hence, the S matrix obeys the bootstrap equation

i
i
SYL
= SYL ( ).
(2.2)
SYL +
3
3
The TBA analysis [4] demonstrates that this S matrix correctly reproduces the central
charge of the unperturbed CFT. The YL model can be regarded [23] as a restriction of
the sine-Gordon model in which the solitons are projected out and only the first breather
remains. Indeed, the S matrix (2.1) coincides with that of the first sine-Gordon breather [1,
34], with = 16/3.
2.2. Boundary
Following [14,35], we consider the boundary YL model which is defined by the
perturbed action (1.2), where the CFT is M(2/5), the CBC corresponds to the cell (1, 3)
of the Kac table, and = (1,3) = 1/5. The (1, 3) conformal boundary condition and
(1,3)
the (1, 3) boundary perturbation are compatible, since the fusion rule coefficient N(1,3)(1,3)
is nonvanishing. The boundary S matrix SYL ( ; b) is given by [35] 1

1
3
4
5b
5 + b 1
1
1 b 1 1 + b
SYL ( ; b) =
,
(2.3)
2
2
2
2
2
2
2
where
(x)
sinh
sinh
ix
6
,
ix
6
(2.4)
and b is a parameter which is related to B . This S matrix obeys the boundary bootstrap
equation [9]

i
i
; b SYL (2 )SYL
; b = SYL ( ; b).
SYL +
(2.5)
3
3
This model can be regarded as a restriction of the boundary sine-Gordon model. Indeed,
the boundary S matrix (2.3) coincides with that of the first sine-Gordon breather [36] with
= 16/3, and with the parameters , of [9] taking the values [14] = 4 (b + 4), i =
4 (b + 2).
This picture is supported by the boundary TBA, which implies that the boundary entropy
is given (up to an additive constant) by
1 We make an effort to distinguish boundary quantities from the corresponding bulk quantities by using sans
serif letters to denote the former, and Roman letters to denote the latter.
664
1
ln g =
4

1
d YL ( ; b) YL (2 ) YL ( ) L( ),
2
(2.6)
where
YL ( ) =
and
1
ln SYL ( ),
i
YL ( ; b) =
1
ln SYL ( ; b),
i

L( ) = ln 1 + e() .
(2.7)
(2.8)
Moreover, ( ) is the solution of the bulk TBA equation [4]

1
(YL L)( ),
2
where denotes convolution
Z
d 0 f ( 0 )g( 0 ),
(f g)( ) =
( ) = r cosh
(2.9)
(2.10)
and r = mR, with R the inverse temperature. Note that our expression (2.6) for the
boundary entropy differs in the third term in the brackets from the one given in Refs. [13]
and [14]. This term originates from the exclusion [37,38] of the Bethe Ansatz root at zero
rapidity.
For simplicity, let us consider the case of massless boundary flow. 2 That is, we consider
the bulk massless scaling limit
n
(2.11)
m = n,
= ln , n 0,
2
where and are finite, which implies E = e , P = e . Moreover, we consider

n
i6
B ln
, n 0,
(2.12)
b = 3
where the boundary scale B is finite. For the sign in the limit (2.11), the boundary S
matrix reduces to S( B )1 [14], and we obtain
2
ln g =
4
),
d YL ( B )L(
(2.13)
) = ln(1 + e() ). Note the factor of 2 appearing in

where (
) ( ln n2 ), and L(
(2.13), which accounts for the contribution from the sign + in the limit (2.11). That is, it
can be shown that right-movers and left-movers give equal contributions to the boundary
entropy. In the UV limit B , the integrand is nonvanishing for ; similarly,
= ln( 1+2 5 ), L()

=0
the IR limit B requires . Using the results L()
which follow from the TBA equation (2.9), we obtain
2 The bulk-massive case seems to have several complicated issues which remain to be resolved [14].

g UV
1+ 5
ln IR = ln
.
g
2
665
(2.14)
This is precisely the ratio of g factors corresponding to the conformal boundary

conditions (1, 3) and (1, 1)

1+ 5
g(1,3)
= ln
,
(2.15)
ln
g(1,1)
2
which have been computed [14] from the M(2/5) modular S matrix. Hence, the boundary
S matrix (2.3) is consistent with massless flow away from the UV conformal boundary
condition; namely, from the CBC (1, 3) to the CBC (1, 1). In Section 4.3 we shall find a
generalization to the supersymmetric case.
3. The bulk SYL model

We turn now to the supersymmetric generalization of the scaling YangLee model,
which was first defined in [19] as a perturbation of the superconformal minimal model
SM(2/8). This (nonunitary) CFT has central charge c = 21/4; and the corresponding
dimensions (n,m) of the primary superconformal fields are given in Table 2. These fields
are of NeveuSchwarz (NS) or Ramond (R) type if n m is even or odd, respectively. We
recall [3] that the superconformal symmetry is generated by the right and left supercurrents
S z) of dimensions ( 3 , 0) and (0, 3 ), respectively. The NS fields are local with
G(z) and G(
2
2
S z), while the R fields are semi-local with respect to these currents.
respect to G(z) and G(
The action of the SYL model is given by [19]
Z
A = ASM(2/8) +
Z
dy
S 1 (,) (x, y),

dxG 1 G
(3.1)
Sn ) are operators appearing in the operator expansion

where = (1,3) = 14 , and Gn (G
S z)) with (,) (z, z ). An interesting feature of this model
of the supercurrent G(z) (G(
1
is that it has fractional ( 2 ) spin integrals of motion. Indeed, the perturbation preserves
supersymmetry, since [6,19]
S 1 (,) ,
S,
S = (2 1)G
z G = z
S = z , = (2 1)G 1 (,) .
z G
(3.2)
The corresponding integrals of motion are given by

Table 2
Kac table for SM(2/8)
0
3
32
14
7
32
14
3
32
666
Table 3
Kac table for M(3/8)
3
2
Z
Q =
S=
Q
25
32
7
32
1
4
14
3
32
14
7
32
3
32
1
4
25
32
3
2

S (x, y) ,
dx G(x, y) +

S y) + (x, y) .
dx G(x,
(3.3)
We now recall the important observation [27] that there is an equivalent formulation
of the SYL model as a perturbation of the ordinary minimal model M(3/8). 3 Indeed,
M(3/8) also has central charge c = 21/4. The corresponding dimensions of the primary
fields are given in Table 3. Note that these dimensions either coincide with those for
SM(2/8) or else correspond to their super-descendants. Indeed, the fields of dimension
1
3
4 and 2 correspond to G 1 (1,3) and G 1 L1 (1,1) respectively; and the field of
2
dimension 25
32 corresponds to G1 (1,4) .
The SYL model can therefore also be formulated by the action (1.1), where the
CFT is the minimal model M(3/8), and = (1,5) = 14 . This is an integrable
perturbation, since [39] the (1, 5) perturbation of M(p/q) is integrable if 2p < q. There is
a corresponding formulation of the conservation laws (3.2), with the supercurrents G and
S replaced by the chiral primary fields (2,1),(1,1) and (1,1),(2,1) respectively, etc.
G
The spectrum of the SYL model consists of one Boson and one Fermion of equal mass
m. Following
[1,9], it is convenient to introduce the Zamolodchikov operators Aa ( ) =
b()
which
create
the corresponding Boson and Fermion asymptotic particle states,
f ()
|Aa1 (1 )Aa2 (2 ) AaN (N )i = Aa1 (1 )Aa2 (2 ) AaN (N )|0i.
(3.4)
This is an in state or out state if the rapidities are ordered as 1 > 2 > > N or
1 < 2 < < N , respectively.
The two-particle S matrix is defined by
Aa1 (1 )Aa2 (2 ) = Sab11ab22 (1 2 )Ab2 (2 )Ab1 (1 ).
(3.5)
For the SYL model, the S matrix is given by [19]

S( ) = SYL ( )SSUSY ( ),
(3.6)
where SYL ( ) is given by (2.1). Moreover,

SSUSY ( ) = Y ( )R( ),
(3.7)
3 As mentioned in introduction, this is completely analogous to the well-known fact that the tricritical Ising
model can be formulated as either SM(3/5) or M(4/5).
667
where R( ) is the 4 4 matrix 4
a+ ( )
0
0
d( )
0
b
c( )
0
,
R( ) =
0
c( )
b
0
d( )
0
0
a ( )
(3.8)
with
a ( ) = 1 +
2i sin 3
,
sinh
b = 1,
c=
i sin 3
sinh 2
d=
sin 3
cosh 2
(3.9)
The scalar factor Y ( ) is given by

Y ( ) =
sinh 2
sinh 2 + i sin 3
Z
exp
0
!
dt sinh(it/) sinh 2t3 sinh 3t
,
t
cosh t cosh2 2t
which we find has the following infinite-product representation:

1

i
i
0 2 2
0 12 + 2
Y ( ) =

i
i
0 2
0 1 + 2
(

i 2
i 2
Y
0 1 + k + 2
0 32 + k 2

1
i 2
i 2
k=0 0 2 + k + 2 0 1 + k 2
5
1
i
i
0 23 + k 2
0 6 + k + 2
0 +k
5
34
5
i
i
0 3 + k + 2 0 6 + k 2 0 3 + k +
(3.10)
i
2 0

i
2 0
7
6
7
6
+k+
+k
)
i
2

i
2
(3.11)
It is convenient to denote the total scalar factor by Z( )
Z( ) = SYL ( ) Y ( )
=
sinh 2
sinh 2 i sin 3
Z
exp
0
!
dt sinh(it/) sinh 4t3 sinh 3t
.
t
cosh t cosh2 2t
(3.12)
Hence, the SYL bulk S matrix is given by

S( ) = Z( )R( ),
(3.13)
where the matrix R( ) is given by Eqs. (3.8), (3.9). TBA analysis [20,21] shows that this
S matrix correctly reproduces the central charge of the unperturbed CFT.
In analogy with the YL model, the SYL model can be regarded as a restriction of the
supersymmetric sine-Gordon (SSG) model in which the solitons are projected out and
only the first breather multiplet remains. Indeed, the S matrix is that of the first SSG
4 Our conventions are such that if A and B are matrices with matrix elements A
a1 a2 and Bb1 b2 , then the tensor
a b
product C = A B has matrix elements Ca 2b 2 = Aa1 a2 Bb1 b2 .
1 1
668
breather [40,41] with = 1/3. In particular, it coincides with the expression for the S
matrix of the supersymmetric sinh-Gordon model given in [15] with B = 1/3.
In view of the alternative formulation of SYL as the (1, 5) perturbation of M(3/8), the
SYL model can also be regarded as a restriction [28,29] of the ZhiberMikhailovShabat
model [3032]. Details of this identification are given in Appendix A.
We recall [6,19] that the supersymmetry charges are assumed to act as follows: on oneparticle states,

i
0
e4
i
,
q( ) = m e 2
QAa ( ) = qab ( )Ab ( ),
e 4
0

i

4
0
e
S
i
q(
) = me 2
(3.14)
QAa ( ) = qab ( )Ab ( ),
e4
0
and on multiparticle states,
Q|Aa1 (1 ) AaN (N )i
!
N
l1
X
Y
Fak
(1)
|Aa1 (1 ) Aal1 (l1 )(QAal (l ))Aal+1 (l+1 ) AaN (N )i,
=
l=1
k=1
S a1 (1 ) AaN (N )i
Q|A
!
N
l1
X
Y
Fak
S al (l ))Aal+1 (l+1 ) AaN (N )i,
(1)
|Aa1 (1 ) Aal1 (l1 )(QA
=
l=1
k=1
(3.15)
where (1)F is +1 for a Boson and 1 for a Fermion. These charges obey the
supersymmetry algebra
S2 = E P ,
Q

S (1)F = 0.
Q, (1)F = Q,
Q2 = E + P ,

S = 0,
Q, Q
(3.16)
It can be shown [19] that the SYL S matrix commutes with the supersymmetry charges Q
S as well as with (1)F .
and Q,
To conclude this section, we demonstrate that the above S matrix satisfies the bulk
bootstrap equations. We do this in preparation for our investigation in Section 4.2 of the
boundary bootstrap equations, which will help determine the boundary S matrix. Near the
direct-channel pole at = i2/3, the bulk S matrix is given by

3
3 0 0
2
i 3c
0
1
1
0
,
(3.17)
S( ) '
i2 0
1
1
0
3
3 0 0 1
where
!
Z
1 dt sinh2 (2t/3) sinh(t/3)
.
c = exp
2
t cosh t cosh2 (t/2)
0
Hence, the nonvanishing three-particle couplings are given by
b
fbb
= ic 3 3,
4
f
f
b
fff
= fbf = ff b = ic 3,
669
(3.18)
where b and f denote Boson and Fermion, respectively. Using the infinite-product
representation for the scalar factor Y ( ) (3.11), one can prove the identity

i
i
2 sinh 2 i
Y + i
3 Y 3
6 cosh 2 + 6
=
.
(3.19)
Y ( )
sinh
Recalling the YL bootstrap relation (2.2), it follows that the total scalar factor Z( ) (3.12)
satisfies

i
i
2 sinh 2 i
Z + i
3 Z 3
6 cosh 2 + 6
=
.
(3.20)
Z( )
sinh
With the help of this identity, it is now straightforward to verify the bulk bootstrap
equations

i c2 c3
i
c
bb3
b
c1 b3
Sa2 a3
.
(3.21)
fa1 a2 Sca3 ( ) = fc1 c2 Sa1 c3 +
3
3
4. The boundary SYL model
We now address the main problems of defining the boundary SYL model and
determining its boundary S matrix.
4.1. Definition of the model as a perturbed CFT
As in the bulk case, we can define the boundary SYL model in either of two ways.
One way is to define the model as a perturbation of the superconformal minimal model
SM(2/8) (cf., Eq. (3.1))
Z
A = ASM(2/8)+SCBC(1,3) +
Z0
dy
S 1 (,) (x, y)
dx G 1 G
Z
+ B
dy G 1 () (y),
2
(4.1)
where = (1,3) = 14 . Indeed, the arguments of [9] suggest that this boundary
perturbation is integrable. Following [10], we observe that for the boundary CFT,
superconformal invariance requires that the stress-energy tensors and supercurrents obey
the boundary conditions

S = 0.
GG
(4.2)
T TS x=0 = 0,
x=0
We assume that for a superconformal minimal model, a superconformal boundary
condition (SCBC) corresponds to a cell of the Kac table, which in (4.1) we take to be
(1, 3). (See below.)
670
Although for the case with boundary the supersymmetry charges

Z0
Q=

S (x, y) ,
dx G(x, y) +
S=
Q
Z0

S y) + (x, y)
dx G(x,
(4.3)
(cf., Eq. (3.3)) are not conserved, it is plausible that some combination of these charges
(plus a possible boundary term) survives. Indeed, following [9], let us first consider
the massless case = 0, and compute the operator product expansion [G(y + ix)
G(y ix)]G 1 () (y 0 ). We conclude that the quantity
2
Z0
Q=

S y) + (y),
dx G(x, y) + G(x,
(4.4)
with (y) B (1 2)() (y) is an integral of motion. It is plausible that, for the
general massive case 6= 0, this becomes
S + ,
Q=Q+Q
(4.5)
S are given in (4.3).

where Q and Q
Alternatively, we can define the boundary SYL model as a perturbation of the minimal
model M(3/8). That is, we can define the model by the action (1.2), where the CFT is
M(3/8), = (1,5) = 14 , and the CBC is either (1, 3), (1, 4), or (1, 5). Indeed, these
three conformal boundary conditions are compatible with the (1, 5) boundary perturbation,
(1,5)
(1,5)
(1,5)
since the corresponding fusion rule coefficients N(1,3)
(1,3) , N(1,4) (1,4) and N(1,5) (1,5) are
all nonvanishing, as can be seen from Table 4. Presumably, only the CBC (1, 3) preserves
superconformal invariance, since only for this CBC does the corresponding dimension
(1,3) = 14 appear in the SM(2/8) Kac Table 2. Hence, here we shall consider only the
CBC (1, 3), for which case the corresponding action is presumably equivalent to (4.1).
We have obtained the M(3/8) fusion rule coefficients given in Table 4 using the
corresponding modular S matrix. Indeed, we recall (see, e.g., [42]) that for M(p/q) the
modular S matrix elements are given by
Table 4
Fusion rule coefficients for M(3/8). Here we list all the triplets (i, j, k) with i 6 j 6 k for which
(1,i)
(1,i)
N(1,j )(1,k) is nonvanishing, and in fact, equal to 1. Note that N(1,j )(1,k) is symmetric under the
interchange of any pair of indices (i, j, k)
(1, 1, 1)
(1, 2, 2)
(1, 3, 3)
(1, 4, 4)
(1, 5, 5)
(1, 6, 6)
(1, 7, 7)
(2, 2, 3)
(2, 3, 4)
(2, 4, 5)
(2, 5, 6)
(2, 6, 7)
(3, 3, 3)
(3, 3, 5)
(3, 4, 4)
(3, 4, 6)
(3, 5, 5)
(3, 5, 7)
(3, 6, 6)
(4, 4, 5)
(4, 4, 7)
(4, 5, 6)
(5, 5, 5)
s
S(r,s) (r 0 ,s 0 ) = 2
671
pss 0
qrr 0
2
0
0
(1)rs +r s+1 sin
sin
,
pq
p
q
1 6 r, r 0 6 p 1,
1 6 s, s 0 6 q 1.
(4.6)
Setting r = r 0 = 1, for M(3/8) we obtain the result

1
1
1
2 sin 3
12 21 sin 8
8
2 sin 8
2 2
1
1
1
1
0

22
2
2 2
2 2
1
1
1
1
1
3
3
sin
8
2 sin 8
2
2 sin 8
2
2 2
1
S = 12
0
0
12
2
1
1
1
1
3
1
sin

2 2 sin 3
8
2 sin 8
8
2
2 2
1
1
1
12
0

2 2
12 sin 3
8
2 2
1
2
2 2
(s, s 0 )
sin 8
2 2
1
2
1
2
sin 8
2 2
12
1
2 2
2 2
1
2
1

2 2
12 sin 3
8
2 2
1
2 sin 8
1
2
1
2 sin 8
1

2 2
12 sin 3
8
, (4.7)
corresponds to S(1,s) (1,s 0 ) . This matrix is real, symmetric,

where the matrix element
2
and unitary, S S = S = I. Finally, the Verlinde formula [44] implies that the fusion rule
coefficients are given by
(1,i)
N(1,j )(1,k) =
7
X
S(1,i)(1,l)S(1,j )(1,l)S(1,k)(1,l)
S(1,1)(1,l)
l=1
(4.8)
We close this subsection with the computation of g factors for the various conformal
boundary conditions, which also relies on the modular S matrix. As shown in [10,14], the
g factor for the CBC (1, s) is given by
S (1,s)
,
g(1,s) =
|S 0 |
(4.9)
i
where 0 denotes the conformal vacuum (which has the property N0(1,i)
(1,j ) = j ), and is
the state of lowest dimension. For M(3/8), 0 is (1, 1) and is (1, 3). In this way, we
obtain
g(1,4) = q
1
2 sin 8
1
,
g(1,2) = g(1,6) = q
2 sin 8
sin 3
g(1,3) = g(1,5) = q 8 ,
2 sin 8
r
1
sin .
g(1,1) = g(1,7) =
2
8
(4.10)
It should also be possible to compute g factors from the SM(2/8) modular S matrix.
However, we do not attempt this here. 5
4.2. Boundary S matrix
The boundary S matrix S( ) is defined as [9]
5 It is not clear how to compute the SM(2/8) modular S matrix directly from the coset su(2)
2
su(2)m /su(2)2+m with m = 4/3 [43], since an additional coset field seems to be required.
672
Aa ( )B = Sba ( )Ab ( )B,
(4.11)
where here B is the so-called boundary creation operator. We now try to determine S( )
for the boundary SYL model (4.1). By analogy with the bulk SYL model, as well as with
the boundary YL model, we expect that the boundary S matrix of the boundary SYL
model should be some reduction of that of the boundary supersymmetric sine-Gordon
model [45], or equivalently, the boundary supersymmetric sinh-Gordon model [15]. We
therefore consider
S( ) = SYL ( ; b)SSUSY ( ; ),
(4.12)
where the scalar factor SYL ( ; b) is given by (2.3), and SSUSY ( ; ) is given by
SSUSY ( ; ) = Y( ; )R( ; ),
(4.13)
where R( ; ) is the 2 2 matrix

A+ B
R( ; ) =
,
B A
(4.14)
with matrix elements

A = cosh
G+ i sinh G ,
2
2
where

e sinh2 2
,
G+ = r sinh +
1 sin 3

2(1 sin 3 ) 1/2
.
r=
sin 3
B = i sinh ,
(4.15)

e sinh2 2
G = r cosh +
,
1 sin 3
(4.16)
Moreover, Y( ; ) is a scalar factor given by

Y( ; ) = Y0 ( )Y1 ( ; )F ( ; ),
(4.17)
where
Y0 ( ) =
2 sinh( 2 + i
4 )
!
Z
1 dt sinh(2it/) sinh(2t/3) sinh(t/3)
,
exp
2
t
cosh2 t cosh2 (t/2)
0
Y1 ( ; ) =
sin( 12
2 ) sin( 12
+ 2 )
1
r sinh sin( i ) sin( + i )
12
2
2
12
2
2
!
Z
dt sinh(it/) sinh(t/3) cosh(t /)
,
exp 2
t
sinh t cosh(t/2)
0
and is a function of defined by
(4.18)

cos = 1 e2 1 sin
.
3
673
(4.19)
The exponential factors of Y0 ( ) and Y1 ( ; ) do not have zeros or poles in the physical
strip 0 6 Im 6 2 , provided | | < 2/3. Finally, F ( ; ) is a CDD-like factor obeying

i
i
+ ; = F
; ,
(4.20)
F ( ; )F ( ; ) = 1,
F
2
2
which is still to be determined.
The above expression for SSUSY essentially coincides with the one for the supersymmetric sinh-Gordon model given in [15] with B = 1/3, = +1, = + i
2 with real, and
r = ir. The only differences lie in the CDD factor F ( ; ) (which is absent from [15])
and the factor Y1 : the expression given here is an analytic continuation of the one given
in [15]. The former does not diverge for = i
3 , which is important for implementing
the boundary bootstrap equations, as we shall see below (4.27).
The alert reader will have noticed that, while the boundary SYL action (4.1) contains
only one boundary parameter (namely, B ), the above boundary S matrix seems to
contain two parameters, namely, b and . The key point to realize is that these two
parameters are not independent. By demanding that the boundary S matrix satisfy the
various constraints [9] arising from the existence of boundary and bulk bound states, we
shall determine the relation between and b (4.26), as well as the CDD factor F ( ; )
(4.33).
We begin by considering the constraints due to boundary bound states. In general [9],
be the position of a pole of the boundary S matrix in the physical strip associated
let iv0a
with the excited boundary state |iB , which can be interpreted as a boundary bound state
of particle Aa with the boundary ground state |0iB . Near this pole, the boundary S matrix
has the form
Sba ( ) '
b0
g
i ga0
,
2 iv0a
(4.21)
are boundary-particle couplings.

where ga0
We assume that (as in the bulk) the SYL boundary S matrix inherits its pole structure
from the YL boundary S matrix (2.3). Therefore, it has [46] two boundary bound state
poles, corresponding to excited boundary states |1iB , |2iB , with 6
v01 =
(b + 1)
,
6
v02 =
(b 1)
.
6
(4.22)
It follows from the condition (4.21) and the form (4.14) of the S matrix that for = iv0 ,
2
2
(4.23)
A+ gb0 , A gf 0 , B gb0 gf 0 ,
where the indices b and f again denote Boson and Fermion, respectively. Hence, we arrive
at the important constraint
6 The subscript a of v can be dropped, since YL has only one type of particle.
0a
674

A+ A
B2
=iv0
= 1.
(4.24)
This equation gives a relation between the boundary parameter and v0 . As shown in
Appendix B, the relation can be expressed most succinctly in terms of the parameter
defined in (4.19):
(4.25)
= v0 .
6
The above relation can hold for both poles (4.22) only if
b
.
(4.26)
6
Eq. (4.26) is the desired relation between and b. The restriction | | < 2/3 which we
found above implies |b| < 4.
We now consider the constraints due to bulk bound states. In view of the direct-channel
pole of the SYL bulk S matrix at = i2/3, the following boundary bootstrap relations
must hold [9]

i b2 a
i
Sb1 a2 (2 )Sbb2
.
(4.27)
fdab Sdc ( ) = fcb1 a1 Saa21 +
3
3
Using infinite-product representations for the scalar factors Y0 ( ), Y1 ( ; ), and Y ( ), one
can prove the identities

i
Y0 ( + i
i 2 sinh sinh 2 i
3 )Y0 ( 3 )Y (2 )
4
=

,
i
Y0 ( )
sinh + i
3 cosh 3

i
Y1 ( + i
1
3 ; )Y1 ( 3 ; )
=
sin
+
sin
Y1 ( ; )
r sinh
12 2
12 2
2(1 + 2 cos 2 2 cosh 2 4i cos sinh )
.
cos 3 + i sinh 3
(4.28)
=
With the help of these identities, together with (2.5), one can show that the SYL boundary
bootstrap relations (4.27) are satisfied, provided that the CDD factor obeys

i
F + i
cos 3 + i sinh 3
3 ; F 3 ;
=
.
(4.29)
F ( ; )
cos 3 i sinh 3
In addition to the boundary bootstrap relation, another constraint due to bulk bound
states is stated in [9]. Namely, let iucab be the position of the pole of the bulk S matrix
associated with the direct-channel bound state of Aa Ab which can be interpreted as the
particle Ac . If the particles Aa and Ab have equal mass, then the boundary S matrix must
have a pole at = i u cab /2, where u cab = ucab . Furthermore,
Sab ( ) '
i fcab g c
c ,
2 i u ab
(4.30)
where g c describes the coupling of Ac to the boundary. The SYL boundary S matrix indeed
has such a pole at = i/6. It follows from the condition (4.30) that for = i/6,
A+ fbbb g b ,
675
ff
A fb g b ;
(4.31)
and hence [26]

f bb
A+
= bff = 3.

A = i f
6
b
(4.32)
However, this equation is satisfied for arbitrary values of , and so does not provide any
further constraints on the S matrix.
The scalar factor Y( ; ) (4.17) should not have zeros or poles in the physical strip. In
view of the relation (4.26), we see that the factor Y1 ( ; ) has poles at = i( 6 ) =
i(b 1)/6. The pole at = i(b 1)/6 is undesirable, since it is physical for
1 < b < 4. 7 Fortunately, we can arrange for this pole to be canceled by a corresponding
zero of the CDD factor. Indeed, a solution to the CDD constraint equations (4.20)
and (4.29) which has a zero at = i(b 1)/6 is given by

5+b
1b
,
(4.33)
F ( ; ) =
2
2
where we have again used the notation (2.4).
In short, the boundary S matrix which we propose for the boundary SYL model (4.1)
is given by Eqs. (4.12)(4.19), (4.26), (4.33). This is one of the main results of our paper.
Note that our proposed boundary S matrix depends on a single independent boundary
parameter b. The relation of this parameter to the boundary parameter B in the action
(4.1) is not yet known.
One check on this proposal is provided by supersymmetry. We have suggested that the
SYL model (4.1) has the integral of motion Q given by (4.5). We now demonstrate that
our proposed boundary S matrix commutes with a similar quantity. Indeed, let us assume
S act on states according to (3.14), (3.15). It is
that the supersymmetry charges Q and Q
straightforward to show that the matrix R( ; ) (4.14) commutes with
S + (1)F ,
Q=Q+Q
(4.34)
q
where here = m 1 + 2/ 3 e . Note that Q does not anticommute with (1)F ,

unlike usual supersymmetry charges. The appearance of (1)F in Q should not be too
surprising, since similar topological charges also appear in the fractional-spin integrals of
motion of the boundary sine-Gordon model [18]. Presumably the operator in (4.5) can be
identified with (1)F . Note that B = 0 (for which vanishes) corresponds to = ,
and hence b = 0. For this value of b, the boundary S matrix S( ) is diagonal. We recall that
Moriconi and Schoutens proposed [26] two diagonal boundary S matrices for the boundary
SYL model (although without reference to any specific boundary conditions), which they
[1]
[1]
[1]
and R(2)
. Our boundary S matrix for b = 0 differs from R(2)
by the CDD
designated R(1)
factor, i.e.,

S( )
[1]
= R(2)
( ).
(4.35)
F ( ; )
b=0
7 The pole at = i(b 1)/6 is canceled by a corresponding zero in the factor
1+b
2

from SYL .
676
4.3. Boundary TBA and massless boundary flow

We have defined the boundary SYL model in Section 4.1, and we have proposed the
corresponding boundary S matrix in Section 4.2. We shall now demonstrate that this picture
is supported by the boundary TBA. Our analysis is a generalization of the one for the
boundary YL model, which we briefly reviewed in Section 2.2. For simplicity, we again
focus our attention on the case of massless boundary flow.
We begin by determining the massless scaling limit. We set
m = n,
i
n
(b + a) = B ln ,
6
2
n
= ln ,
2
n 0,
(4.36)
with , , and B real and finite. Our objective is to determine the value(s) of a (also real
and finite) for which the boundary S matrix, in the above limit, remains finite and unitary.
After some computation, we find that a = 6; and the resulting massless boundary S matrix
is given by
S( ) = Z( B )R( B ),
(4.37)
where
Z( ) =
sinh
sinh
i5
12
and
R( ) =
sinh
i 23
i
4
i
12
sinh
Z
+
i5
12
exp
0
!
, (4.38)
i 23
sinh
dt sinh 3t sinh t i
1
t
sinh t cosh 2t
i
4
.
(4.39)
Indeed, S( ) satisfies the unitarity condition, since

Z( )R( ) Z( )R( ) = I.
(4.40)
In order to formulate the TBA equations, we consider N particles with real rapidities
1 , . . . , N in an interval of length L, with bulk S matrix S( ) (3.13) and boundary S
matrices S( ; b ) (4.12), where the subscripts here denote the left and right boundaries.
As already discussed, the bulk and boundary S matrices of the SYL model essentially
coincide with those for the supersymmetric sinh-Gordon model given in [15] with B = 13 ,
= +1, = + i
2 , and r = ir. Hence, the Bethe Ansatz equations and the transfer
matrix eigenvalues for SYL can be easily obtained from [15], to which we shall henceforth
refer as I. From Eq. (I 4.14) we obtain the Bethe ansatz equations for zk+

N
Y
tanh 12 (zk+ + j )
tanh 12 (zk+ j )

1 +
i
tanh 12 (zk+ j ) + i
3 tanh 2 (zk + j ) + 3
j =1

+
sinh2 1 i z+
e sinh2 i
sinh2 12 i5
k
6 + zk
12 +e
2 6
=

[ + ],
+
sinh2 1 i5 + z+
sinh2 12 i
e sinh2 i
k
6 zk
12 +e
2 6
(4.41)
677
and from (I 4.15) we obtain a similar result for zk . In view of the massless scaling limit
(4.36), we set
n
j = j ln ,
2
and we obtain
N
Y
j =1
i
n
(b + 6) = B ln ,
6
2
n
zk = z k ln ,
2

1 +
zk j )
2 (

1 +
zk j ) + i
2 (
3
(4.42)
tanh
tanh

1 +
1 +
zk B ) i
zk
2 (
12 cosh 2 (
1 +
i5
1
cosh 2 (zk B ) + 12 cosh 2 (zk+
setting z k+ = xk i
3 , we obtain
cosh
Finally,
Eq. (I 4.19)),
N
Y
tanh
j =1
n 0,
tanh
1
2 (x k
1
2 (x k
j )
j ) +
i
6

i
6
cosh
cosh
1
2 (x k
1
2 (x k
B+ )
+ ) +
B
i
12
.
i5
12
(4.43)
the Bethe ansatz equations for xk (cf.,
B ) +
)
B
i
4 cosh

i
4 cosh
1
2 (x k
1
2 (x k
B+ ) +
+ )
B
k = 0, 1, . . . , N.
i
4

i
4
= 1,
(4.44)
The transfer matrix eigenvalues ( |1 , . . . , N ) can be deduced from Eqs. (I 4.12),

(I 4.17), (I 4.24). In the scaling limit (4.42) (with = ln n2 ), we obtain
+
Z( B+ )Z( B )ex0 2 (B +B )
1
N
N
Y
Z( k )exk k Y
k ( k ), (4.45)
1
sinh( k )
k=1
k=0
where Z( ) and Z( ) are given by (3.12) and (4.38), respectively;

i
i
cosh
,
( ) = sinh
2
6
2
6
(4.46)
k = 1 (see Eq. (I 4.20)), and xk satisfy (4.44).

We introduce the densities P ( ) of magnons, i.e., of real Bethe ansatz roots {xk } with
) of particles {k } and holes,
k = 1, respectively; and also the densities 1 ( ) and (
8
respectively. The Bethe ansatz equations (4.44) imply
P+ ( ) + P ( ) =
where

1
1
(1 )( ) +
B+ + B ,
2
2L

tanh
1
ln
( ) =
i
tanh

cosh
1
ln
( ) =
i
cosh
(4.47)
i
4 cosh sin 3
2 6
=
i
cosh 2 cos 2
2 + 6
3

i
1
2 + 4
,
=
i
cosh
2 4
= YL ( ),
(4.48)
for negative values of to be equal to 1 (||).
and we have defined 1 ()

8 The counting function should be monotonic increasing, in order that the corresponding density (defined as
the derivative of the counting function) be nonnegative.
678
The Yang equations (I 5.7) and the expression (4.45) for the eigenvalues imply
)
1 ( ) + (
1
1
1

(1 Z )( ) +
(P+ + )( ) +
(P )()
= e +
2
2
2

1
Im ln Z B+ +
Im ln Z B ,
+
2L

(4.49)
where
Z ( ) =
Im ln Z( ),
( ) =
Im ln ( ),
(4.50)
Using the fact

and we have defined P ( ) for negative values of to be equal to P (||).
1
( ) = 2 ( ), and using (4.47) to eliminate P+ , we obtain

1
1
1
(P )( ) +
( )
) = e +
1 Z
1 ( ) + (
2
2
4

1
1
( ) B+
+
Im ln Z B+
2L
4

1
( ) B . (4.51)
Im ln Z B
+
4

With the help of the identities
1
( )( ) = ( ),
4
1
Im ln Z( )
( )( ) = 0,
4
we obtain the simple result
Z ( )
) =
1 ( ) + (

1
1
e +
(P )( )
(1 )( ).
2
2
(4.52)
(4.53)
Proceeding as in I, we obtain the TBA equations 9

1
(L1 L2 ) ( ),
2
1
( L1 )( ),
0 = 2 ( ) +
2
re = 1 ( )
(4.54)
where

Li ( ) = ln 1 + ei () ,

,
1 = ln
1
r = R,

P+
2 = ln
.
P
(4.55)
9 This set of TBA equations is the same as for the case of periodic boundary conditions, which was first
conjectured in [47] (see also [48]) and later derived from the SYL S matrix in [20] and generalized in [21].
679
Moreover, the boundary entropy of one boundary is given (up to an additive constant)
by
2
ln g =
4
d ( B )L2 ( ),
(4.56)
where we have included the factor 2 in order to account for contributions from both rightmovers and left-movers. In the UV limit B , the integrand is nonvanishing for
;
similarly, the IR limit B requires . Using the results L2 () =
ln(2 + 2), L2 () = ln 2 which follow from the TBA equations (4.54), we obtain

1+ 2
g UV 1
.
ln IR = ln
g
2
2
(4.57)
This is precisely the ratio of g factors corresponding to the M(3/8) conformal boundary
conditions (1, 3) and (1, 2)
ln

1+ 2
g(1,3) 1
,
= ln
g(1,2) 2
2
(4.58)
as one can verify from Eq. (4.10). Hence, the proposed boundary S matrix is consistent with
massless flow away from the UV conformal boundary condition; namely, from the CBC
(1, 3) to the CBC (1, 2). In the SM(2/8) description, this corresponds to the flow from
the SCBC (1, 3) to the SCBC (1, 4). A plot of ln g as a function of B is given
in Fig. 1.
1
ln(1
+
2) and the
For convenience,
a
constant
has
been
added
so
that
the
UV
value
is
2
IR value is 12 ln 2.
Fig. 1. Boundary entropy: ln g vs. B .
680
5. Discussion
We have proposed the boundary S matrix (4.12)(4.19), (4.26), (4.33) for the boundary
SYL model defined by the action (4.1). Some support for this conjecture is provided by
the fractional-spin integral of motion (4.5), (4.34), and by the massless boundary flow
(4.57), (4.58). Several important problems remain to be solved, including the relation of
the parameter B in the action to the parameter b of the boundary S matrix; and the
identification of the operator in (4.5) with the operator (1)F in (4.34). It would
also be interesting to consider other conformal boundary conditions, as well as extend the
present study to the full family of integrable models with N = 1 supersymmetry [19,26].
Acknowledgements
We thank O. Alvarez, P. Dorey, D. Kastor, P. Pearce, F. Ravanini, R. Tateo and M. Walton
for helpful discussions and/or correspondence. C.A. thanks Universit di Bologna and
University of Miami for hospitality where part of this work was done. This work was
supported in part by KOSEF 1999-2-112-001-5 (C.A.) and by the National Science
Foundation under Grant PHY-9870101 (R.N.).
Appendix A. SYL model as restriction of ZMS model
Here we show that the scaling supersymmetric YangLee model is a restriction of the
ZhiberMikhailovShabat model [30,31], whose action is given by
Z

m2
(A.1)
A = d 2 x ( )2 + 2 ei 8 + ei 2 .
(2)
This is the A2 imaginary coupling affine Toda field theory, whose S matrix was found by
Izergin and Korepin [32]. We follow closely the paper [29] of Takcs, to which we shall
refer as II. (See also [49,50].)
It is useful to first recall the related work [28] of Smirnov. There it is observed that, for
r
,
(A.2)
=
s
the ZMS model is the (1, 2) perturbation of the minimal model M(r/s). Indeed, one can
regard the first two terms in the action (A.1) as the action for M(r/s), and the third term as
the (1, 2) perturbation. The S matrix of the perturbed model can be obtained as the RSOS
2
(2)
restriction of the A2 S matrix, using the models Uq (sl(2)) symmetry, where q = ei / .
In II, it is observed that, for
4r 0
,
(A.3)
s0
the ZMS model is the (1, 5) perturbation of the minimal model M(r 0 /s 0 ). Indeed, one
can regard the first and third terms in the action (A.1) as the action for M(r 0 /s 0 ), and the
second term as the (1, 5) perturbation. The S matrix of the perturbed model can be obtained
= 4 0 =
681
0
as the RSOS restriction of the A(2)
2 S matrix, using the models Uq (sl(2)) symmetry, where
2 / 0
0
i
4
=q .
q =e
We have suggested in Section 3 that the SYL model can be regarded as the (1, 5)
perturbation of M(3/8). We now proceed to compute the latters S matrix following II,
and we shall find that it coincides (up to a scalar factor) with Eq. (3.8). For M(3/8) we
have r 0 = 3, s 0 = 8; hence, q 0 = q = e2i/3 . The first positive integer p for which q 0p =
1 is p = 3. Hence, the maximum spin is jmax = p2 1 = 12 . Thus, the model contains
charged kinks K0 1 = K 1 0 which we denote by c, and neutral kinks K0 0 = K 1 1
2
2
2 2
which we denote by n. Since (II 23)

2
= 2,
(A.4)
=
3 2
the model contains neither breathers nor higher kinks. The S matrix is expressed in terms of
the rapidity variable y = e/ = e/2 . The c c c c amplitude is given by (II 43)(II 45)
@ 2
0 @
0
1@
2 @
q
y2 q5 1
1
y2
2 + q + 5 2 + q = 2i 3 2 sinh .
q
y
q
q
y
q
(A.5)
The n n n n amplitude is given by (II 46), (II 40)

@0
0 @ 0
@
0 @
q 6 y 2 + y 2 q 8 q 8 q 4 y 2 + y 2 q 10 y 2 + y 4 q 2 y 2 q 2
y2q 5
= 2i 3 + 2 sinh .
(A.6)
The c c n n and n n c c amplitudes are equal, are are given by (II 48)
@0
0 @
0
1@
2 @
=i
(q 4 1)(y 2 1)
=
2
3 sinh .
2
q 2y
(A.7)
Finally, the n c forward scattering and reflection amplitudes are given by (II 46), (II 40)
1
@ 2
1
0 @ 2
@
0 @
(y 2 + q 6 )(y 2 1)
= 2 sinh
y 2q 3
(A.8)
and
@
1
2
1
2
0
1@
2 @
(q 4 1)(y 2 + q 6 )
=
2i
3 cosh ,
2
yq 5
(A.9)
respectively. Identifying n and c as the Boson and Fermion (respectively) of the SYL
model, we see that the above amplitudes coincide with those in 2(sinh )R( ), where R( )
is the matrix (3.8). That is, the SYL model is indeed a restriction of the ZMS model,
corresponding to the (1, 5) perturbation of M(3/8).
682
Appendix B. Solution of constraint Eq. (4.24)

Here we solve Eq. (4.24), which for simplicity we now write as

A+ A B 2 =iv = 0.
(B.1)
Using the definitions of A+ , A , and B given in (4.15), (4.16), and introducing the variable
t sin2 v2 , Eq. (B.1) can be brought to the form

3
3
1
2
2 3
t + t 1 +
+e
t
2
2
4
2

3
3 7
3
7
2
4 7
+e
+e
= 0.
(B.2)
+
16
4
2
8
16
4
to a fixed value of v (and hence, b). The
We discard the solution t = 12 , which corresponds
1
two remaining solutions are t = 4 ( ), where

3
,
3
= 2 3 + e2
2

7
3
.
(B.3)
= e2 (2 3) + e4
4
In terms of the parameter defined by

3
2
,
1
cos = 1 e
2
(B.4)
we have
=2
3 cos ,
= sin ;
and therefore,

1
.
t = 1 cos
2
6
(B.5)
(B.6)
Finally, recalling the definition t = sin2 v2 , we arrive at the remarkably simple result
=v
,
6
which is quoted in text (4.25).
References
[1] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253;
A.B. Zamolodchikov, Sov. Sci. Rev. A 2 (1980) 1.
[2] A.B. Zamolodchikov, JETP Lett. 46 (1987) 160;
A.B. Zamolodchikov, Adv. Stud. Pure Math. 19 (1989) 641.
(B.7)
683
[3] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333;
A.B. Zamolodchikov, R.G. Pogosyan, Sov. J. Nucl. Phys. 47 (1988) 929;
A.B. Zamolodchikov, Al.B. Zamolodchikov, Sov. Sci. Rev. A 10 (1989) 269;
A.B. Zamolodchikov, Rev. Math. Phys. 1 (1990) 197.
[4] Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695.
[5] Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497.
[6] A.B. Zamolodchikov, Fractional-spin integrals of motion in perturbed conformal field theory,
in: H. Guo, Z. Qiu, H. Tye (Eds.), Fields, Strings and Quantum Gravity, Gordon and Breach,
1989.
[7] D. Bernard, A. LeClair, Commun. Math. Phys. 142 (1991) 99.
[8] G. Mussardo, Phys. Rep. 218 (1992) 215.
[9] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
[10] J.L. Cardy, Nucl. Phys. B 324 (1989) 581.
[11] I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
[12] L. Chim, Int. J. Mod. Phys. A 11 (1996) 4491.
[13] A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.
[14] P. Dorey, I. Runkel, R. Tateo, G. Watts, Nucl. Phys. B 578 (2000) 85.
[15] C. Ahn, R.I. Nepomechie, Nucl. Phys. B 586 (2000) 611.
[16] F. Lesage, H. Saleur, P. Simonetti, Phys. Lett. B 427 (1998) 85.
[17] C. Ahn, C. Rim, J. Phys. A 32 (1999) 2509.
[18] L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. A 13 (1998) 2747.
[19] K. Schoutens, Nucl. Phys. B 344 (1990) 665.
[20] C. Ahn, Nucl. Phys. B 422 (1994) 449.
[21] M. Moriconi, K. Schoutens, Nucl. Phys. B 464 (1996) 472.
[22] J.L. Cardy, G. Mussardo, Phys. Lett. 225 (1989) 275.
[23] F.A. Smirnov, Int. J. Mod. Phys. A 4 (1989) 4213;
F.A. Smirnov, Nucl. Phys. B 337 (1990) 156.
[24] C.N. Yang, T.D. Lee, Phys. Rev. 87 (1952) 404;
T.D. Lee, C.N. Yang, Phys. Rev. 87 (1952) 410.
[25] M.E. Fisher, Phys. Rev. Lett. 40 (1978) 1610.
[26] M. Moriconi, K. Schoutens, Nucl. Phys. B 487 (1997) 756.
[27] E. Melzer, Supersymmetric analogs of the GordonAndrews identities and related TBA
systems, hep-th/9412154.
[28] F.A. Smirnov, Int. J. Mod. Phys. A 6 (1991) 1407.
[29] G. Takcs, Nucl. Phys. B 489 (1997) 532.
[30] R.K. Dodd, R.K. Bullough, Proc. Roy. Soc. Lond. A 352 (1977) 481.
[31] A.V. Zhiber, A.B. Shabat, Sov. Phys. Dokl. 24 (1979) 607;
A.V. Mikhailov, Pisma Zh. Eksp. Theor. Fiz. 30 (1979) 443.
[32] A.G. Izergin, V.E. Korepin, Commun. Math. Phys. 79 (1981) 303.
[33] J.L. Cardy, Phys. Rev. Lett. 54 (1985) 1354.
[34] I.Ya. Arefeva, V.E. Korepin, JETP Lett. 20 (1974) 312.
[35] P. Dorey, A. Pocklington, R. Tateo, G. Watts, Nucl. Phys. B 525 (1998) 641.
[36] S. Ghoshal, Int. J. Mod. Phys. A 9 (1994) 4801.
[37] P. Fendley, H. Saleur, Nucl. Phys. B 428 (1994) 681.
[38] M. Grisaru, L. Mezincescu, R.I. Nepomechie, J. Phys. A 28 (1995) 1027.
[39] V. Fateev, S. Lukyanov, A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys. B 516
(1998) 652.
[40] R. Shankar, E. Witten, Phys. Rev. D 17 (1978) 2134.
[41] C. Ahn, Nucl. Phys. B 354 (1991) 57.
[42] A. Cappelli, C. Itzykson, J.-B. Zuber, Nucl. Phys. B 280 (1987) 445.
[43] P. Mathieu, D. Senechal, M. Walton, Int. J. Mod. Phys. A 7 (1992) 731.
684
[44] E. Verlinde, Nucl. Phys. B 300 (1988) 360.

[45] C. Ahn, W.M. Koo, J. Phys. A 29 (1996) 5845;
C. Ahn, W.M. Koo, Nucl. Phys. B 482 (1996) 675.
[46] P. Dorey, R. Tateo, G. Watts, Phys. Lett. B 448 (1999) 249.
[47] F. Ravanini, R. Tateo, A. Valleriani, Int. J. Mod. Phys. A 8 (1993) 1707.
[48] P. Dorey, C. Dunning, R. Tateo, Nucl. Phys. B 578 (2000) 699.
[49] M.J. Martins, Phys. Lett. B 262 (1991) 39.
[50] C.J. Efthimiou, Nucl. Phys. B 398 (1993) 697.

Random defect lines in conformal minimal models

M. Jeng, A.W.W. Ludwig
Physics Department, University of California, Santa Barbara, CA 93106-4030, USA
Received 19 September 2000; accepted 25 October 2000
Abstract
We analyze the effect of adding quenched disorder along a defect line in the 2D conformal minimal
models using replicas. The disorder is realized by a random applied magnetic field in the Ising model,
by fluctuations in the ferromagnetic bond coupling in the tricritical Ising model and tricritical threestate Potts model (the 12 operator), etc. We find that for the Ising model, the defect renormalizes to
two decoupled half-planes without disorder, but that for all other models, the defect renormalizes to
a disorder-dominated fixed point. Its critical properties are studied with an expansion in 1/m for
3
3
the mth Virasoro minimal model. The decay exponents XN = N2 1 9(3N4)2 + O m+1
of the
4(m+1)
Nth moment of the two-point function of 12 along the defect are obtained to 2-loop order, exhibiting
3
9
3
multifractal behavior. This leads to a typical decay exponent Xtyp = 12 1 +
+ O m+1
.
(m+1)2
One-point functions are seen to have a non-self-averaging amplitude. The boundary entropy is larger
than that of the pure system by order 1/m3 .
eN = N 1
As a byproduct of our calculations, we also obtain to 2-loop order the exponent X

2 (3N 4)(q 2)2 + O(q 2)3 of the Nth moment of the energy operator in the q-state Potts
9 2
model with bulk bond disorder. 2001 Elsevier Science B.V. All rights reserved.
PACS: 64.60.Ak; 64.60.Cn; 64.60.Fr; 64.60.Kw

Keywords: Conformal; Defect; Disorder; Multifractal; Potts
1. Introduction
Conformal symmetry tends to emerge at critical points of pure (homogeneous and
rotationally invariant) 2D statistical mechanics models. This high degree of symmetry
severely constrains these theories, so that these critical points are well understood.
Many models, including the Ising model, tricritical Ising model, and tricritical threestate Potts model, are part of a class of conformal field theories known as Virasoro
minimal models [13]. The Ising model [4], tricritical Ising model [46], and tricritical
three-state Potts model [7,8] have all been realized experimentally in adsorbed monolayer
systems.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 9 - 8
686
M. Jeng, A.W.W. Ludwig / Nuclear Physics B 594 [FS] (2001) 685712
Because physical systems always have impurities, it is important to consider the effect
of quenched disorder on the critical behavior of these theories. When disorder is added
to the bulk, random fields are usually relevant (in the renormalization group [RG] sense),
but random bonds may or may not be (see, e.g., Refs. [9,10]). It is possible to show, for
certain systems where disorder is relevant, that with the addition of disorder the system
renormalizes into a new infrared fixed point, governing the long distance properties of
the disordered system. In fact, a rigorous theorem shows that when disorder in the order
parameter is added to a system undergoing a first-order phase transition in two dimensions,
the latent heat vanishes [11], and a 2nd order transition can be expected.
However, one generally does not know a priori whether the new critical point is disorderdominated. Indeed, a number of studies have reported cases in which the addition of
quenched disorder to a pure 2D model at its critical point resulted in another pure (that
is, non-random) critical model [1217]. Other studies found quenched disorder to result
in new disorder-dominated fixed points [1822]. One can often see that a critical point
is disorder-dominated by showing that various universal quantities are not self-averaging
(that measurements for a specific typical sample may differ substantially from those
obtained upon sample-averaging). One particularly interesting manifestation of this is
multifractal behavior, which occurs when an infinite hierarchy of independent scaling
dimensions are associated with a single operator [2326]. (See also Ref. [27] for a related
discussion.)
All the above studies have focused on the effects of adding quenched disorder to the
bulk of a system. However, it is also interesting to consider the case where the 2D
model has a defect along which impurities have clustered. In this paper we consider the
effects of adding quenched disorder only along a defect line, in each of the minimal
models; these models are labelled by an index m, m > 3. Each m represents a different
model: the Ising model (m = 3), tricritical Ising model (m = 4), tetracritical Ising model
(m = 5), tricritical three-state Potts model (m = 6), etc. In Section 2 we introduce our
defect model, adding quenched disorder in the coupling to 12 (an operator in the Kac
Table [1]) using replicas. The physical meaning of this disorder varies from model to
model, representing a random magnetic field for the Ising model, but fluctuations in the
ferromagnetic coupling (or chemical potential) in the tricritical Ising model and tricritical
three-state Potts model. We find that the Ising model renormalizes to a non-random
model (consisting of decoupled half-planes with free boundary conditions and no random
magnetic field), while all other models renormalize to disorder-dominated fixed points.
Just as with bulk disorder, disorder on the defect can result in either a pure or disorderdominated fixed point.
In Section 3 we calculate the renormalization group equation of the strength of the
disorder , to 2-loop order by dimensional regularization [21,28]. We find that arbitrarily
weak disorder grows under the RG, and flows to a new fixed point at which is of order
1/m. This justifies a 1/m expansion, where critical quantities are calculated perturbatively
in 1/m. The boundary entropy [32] at the random fixed point is found to be O(1/m3 ),
and larger than the entropy of the pure fixed point.
687
In Section 4 the same scheme is used to find the moments of correlation functions
of the operator 12 along the defect. (Technical details associated with the irreducible
representations of the symmetric group [23] are delegated to Appendices B and C.) It is
found that these moments fall off as a sum of power laws, the dominant term decaying as 1
h12 (x1 )12 (x2 )iN |x1 x2 |2XN
with
XN =

2
3

1
3
N
3
1 (3N 4)
+O
.
2
4
m+1
m+1
(1.1)
(1.2)
These exponents satisfy XN < NX1 , so that the operator 12 exhibits multifractal behavior.
Our results lead to a typical decay exponent [23]:

2
3

3
1
3
1+
+O
.
(1.3)
Xtypical =
2
m+1
m+1
In Section 5 we calculate even moments of one-point functions of the same operator,
12 , as a function of the distance from the defect line. (Odd moments are seen to vanish.)
The universal (normalized) amplitudes of the moments of these one-point functions are
found to be non-self-averaging, whereas the power law of the decay away from the defect
is self-averaging.
In Section 6, the Ising model is analyzed, which requires special considerations due to
the presence of an additional marginal (boundary) operator. It is argued that when a random
magnetic field is added along the defect line, the system renormalizes to two decoupled
half-plane Ising models with free boundary conditions and no disorder.
We finally note that the manipulations needed to obtain the exponents in Eq. (1.2)
are similar to those needed to obtain the decay exponents of moments of the two-point
functions of the energy operator, , for the q-state Potts model with bulk disorder in the
bond strength. This model has been analyzed elsewhere by expanding about q = 2 (the
Ising model) [2123,28,29]. As a byproduct of our calculations here we find in Section 4,
eN of the energy correlation function
Eq. (4.11)(4.13), the dominant 2 scaling exponents X
e
h(x1 )(x2 )iN |x1 x2 |2XN ,
(1.4)
where

eN = N 1 2 (3N 4)(q 2)2 + O(q 2)3 .
X
9 2
(1.5)
e1 ), showing up in two loop order, lead to the typical

eN < N X
The multifractal features (X
etypical of Eq. (4.13) in the bulk random q-state
decay exponent for energy correlation X
Potts model.
1 The N th moment [G ]N of the connected correlation function G = h (x ) (x )i h (x )ih (x )i
c
c
12 1 12 2
12 1
12 2
decays with a single power XN [29].
2 As in the preceeding footnote, there are no subleading powerlaws when the moments of the connected
correlation functions are considered.
688
Recent numerical calculations by Jacobsen [30] for the random-bond q-state Potts model
agree well with our results here. Independent one-loop analyses of this model were done
by Davis and Cardy [29], using different methods than in this paper, and their results agree
with ours.
2. The defect model

We start with a Virasoro minimal conformal field theory [13] labelled by an integer m,
m > 3, and perturb it along a defect line. The perturbed action is
Z
S = Sm +
dx h(x)12 (x, y = 0),
(2.1)
Sm is the action of the unperturbed conformal field theory. h(x) is a random coupling, is
picked from a Gaussian probability distribution with zero mean and variance 0 , and is
uncorrelated along the defect:
h(x)h(x 0 ) = 20 (x x 0 ).
h(x) = 0,
(2.2)
The overbar indicates the disorder average. The operator 12 is located at position (p, q) =
(1, 2) in the Kac Table [1], and exists for any minimal model. For the Ising model (m = 3)
it is the spin operator, while for the tricritical Ising model (m = 4) and tricritical three-state
Potts model (m = 6) it is the energy operator. The scaling dimensions 2hpq (twice the
conformal weight, not to be confused with the random coupling h(x)) of operators pq in
minimal models are known [1]:
2hpq =
[(m + 1)p mq]2 1

,
2m(m + 1)
(2.3)
3
1
.
2 2(m + 1)
(2.4)
which gives
2h12 =
The replicated [31] effective action is

replica
Sm
n
X
=1
Sm
+ 0
dx
n
X
12
12 (x, y = 0),
(2.5)
6=
where n is the number of replicas, and we take n 0 at the end of the calculation. Note
that the scaling dimension of the operator perturbing the non-random action is 4h12 =
1 3/(m + 1) so that the RG eigenvalue of the disorder strength 0 near the non-random
fixed point 0 = 0 is 14h12 = 3/(m+1). For large values of the index m the disorder
perturbation is barely relevant since 1, and a new infrared fixed point at disorder
strength of order exists. We study universal properties of this fixed point perturbatively
in the small parameter .
689
We have ignored higher cumulants of the probability distribution of h(x), because power
counting shows that they are irrelevant 3 in the RG sense, for large m, and we will be
expanding about large m.
We have also dropped the terms with = in Eq. (2.5). These terms produce a nonrandom 13 by the conformal fusion rules [1]. Upon renormalizing they will generate other
non-random 1q with q > 3. However, from Eq. (2.4) we can see that all these terms are
irrelevant except when q = m = 3. So when m > 3 the perturbation in Eq. (2.5) is the only
relevant one. The following analysis in Sections 35 will assume this, and will thus only
hold for m > 3. This will not be too restrictive, since all quantities will be calculated by
3
, and will thus be based on large m. But if m = 3 (the Ising model),
expansion in = m+1
then 13 (the energy operator) is marginal, and we need to include the effects of a constant
coupling to 13 . This is done in Section 6.
3. Renormalization of the disorder strength

We calculate the renormalization group equation for to 2-loop order by minimal
subtraction [21,28]. To calculate the renomalization of 0 , we want to know, given a
microscopic disorder strength 0 , what effective disorder strength (r) is seen on large
replica
} in powers of 0 . (0 )p ,
length scales r. In a region of size r, we expand out exp{Sm
with p > 0, will come with p disorder operators at various points in the region of size r.
p = 2, 3 or more disorder operators may look, using repeated operator product expansions,
like a single disorder operator on larger length scales (or size r), and thus create a new
effective disorder strength (r). This can be represented schematically as:
!2
Z
Z
n
n
X
X
1

0
dx
12 12 (x) +
dx
12 12 (x)
0
2
6=
1
0
+
6
dx
n
X
6=
!3

12
12 (x)
6=
Z
+
dx
n
X
12
12 (x).
(3.1)
6=
This analysis will of course generate numerous terms besides the disorder operator.
However, as noted above, these terms are all irrelevant in the RG sense, so we will not
calculate these terms.
For each power of 0 , the integrals generated above are regulated at short distances
3
, and at large distances by an infrared cutoff r. The
by analytic continuation in m+1
calculation, which uses the method of [28], is done in Appendix A, where we obtain
3 For the lower values of m = 4 or 5, the 4th order cumulants are, respectively, relevant and marginal, but the
2nd order cumulant is more relevant, and presumably will be physically dominant. While the 4th order cumulant
is not irrelevant at the unperturbed fixed point for m = 4 or m = 5, it is irrelevant for large m, and we expect it to
be irrelevant for m = 4 or 5 at the new disordered fixed point. This would be similar to the WilsonFisher fixed
point, where the 6 operator becomes relevant at the Gaussian fixed point below three dimensions, wheras it is
in fact irrelevant at the new fixed points obtained by an expansion about four dimensions.
690
(r) = r 0 + 4(n 2)

r 2 2
1 r 3 3
0 4(n 2) 2 4(n 2)
0 + O 40 . (3.2)

When we calculate the beta function by taking a derivative with respect to log(r), we find
that the poles in cancel, as they must for any physical quantity. The result is

d
= 82 + 323 + O 4 ,
(3.3)
() =
d(log(r))
where we have taken the replica limit n 0 (and have ceased writing the r dependence
of (r)). The RG flows take the unperturbed theory to a new infrared fixed point with
disorder . Solving for by putting ( ) = 0 gives

2

3
+ O 3
.
(3.4)

= +
8 16
m+1
The new fixed point is at a distance of order from the unperturbed theory, and so we
can calculate physical quantities by expanding in powers of , which is small for minimal
models with large index m. This is analogous to the WilsonFisher epsilon expansion in
d = 4 dimensions. We will see below that the new fixed point is disorder-dominated.
3
, we really only show that
Because the analysis is based on expansion in powers of = m+1
there is a new disorder-dominated fixed point for large m but we expect no qualitative
changes for lower m, such as m = 4. However, we again note that m = 3, the Ising model,
is qualitatively different, and will be treated separately.
The boundary entropy (the universal constant independent of the system size appearing
in the disorder-averaged free energy) associated with the defect line, calculated as in
Ref. [32], is found to be

g 2 3
=
+ O 4 .
(3.5)
n
96
Note that the entropy has increased from that of the unperturbed system (where it vanishes).
This is to be contrasted with the case of a pure (non-random and unitary) system, where
the entropy [32] is expected to only decrease upon renormalization.
4. Scaling dimensions of moments of 12

We now look at h12 (x1 )12 (x2 )iN , the disorder-averaged N th moment of the 2-point
function for points x1 and x2 which lie both near the defect and far from each other (Fig. 1).
We want to see how these moments fall off at large distances. As formulated in replicas we
have
* N
+
N
Y
Y
h12 (x1 )12 (x2 )iN =

12
(x1 )
12 (x2 ) .
(4.1)
=1
=1
As explained in [23], we cannot simply calculate the dimension of N

=1 12 , because
this operator is, in general, not multiplicatively renormalizable. Instead, it is, in general,
a sum of independent scaling operators with different scaling dimensions. In the limit
691
Fig. 1. A two-point correlation function for operators lying along the defect.
where the pure and disordered fixed points collide (m = ), we have numerous operators
1 2 and 3 7 have the
with the same dimension. For example, looking at N = 2, 12
12
12 12
same scaling dimension and are equally good operators (note that the replica limit n 0
is not taken until after all calculations are completed). In general, for the N th moment, and

with n replicas, we have Nn operators with the same dimension at m = ; because they
all have the same scaling dimension at m = , any linear combination of them also has
this scaling dimension at m = .
However, when we move to m < , only appropriate linear combinations of these
Q
operators will have well-defined scaling dimensions. The entity N

=1 12 which appears in
Eq. (4.1) will contain all of these scaling operators, so that the average h12 (x1 )12 (x2 )i0 N
will decay as a mixture of power laws, and at large distances will be dominated by the
scaling operator with the lowest dimension. We thus need to calculate the dimension of
each of these scaling operators. The appropriate multiplicatively renormalizable operators
transform in irreducible representations of the symmetric group [23]:
X

n(M1) M+1
1
M
n
12
12
(4.2)
12
12
12 12N ,
ONMn =
i 6=j
16i 6(nM)
where 0 6 M 6 N .
R To calculate the scaling dimensions of ONMn , we will add the term NMn,0
dx ONMn (x) to the action. As in Section 3, we calculate the renomalization of
NMN,0 to NMn (r) = ZNMn (r)NMn,0 on length scales of size r, by expanding the
action in powers of 0 :
!
Z
dx ONMn (x)
NMn,0
Z
dx ONMn (x)
+ NMn,0 0
1
+ NMn,0 20
2
Z
dx
!
dx ONMn (x) .
ZNMn (r)NMn,0
!

12
12 (x 0 )
6=
dx
n
X
dx ONMn (x)
n
X
!2

12
12 (x 0 )
6=
(4.3)
In Appendix B we check that to 2-loop order we do not need to worry about mixing
with other operators. ZNMn (r) is calculated with the same type of integrals as in the last
section (they are again regulated at short distances by analytic continuation in and at large
692
distances by an infrared cutoff r). However, technical combinatorial complexities arise in

counting the number of contractions associated with various irreducible representations of
the symmetric group. They are delegated to Appendices B and C. The result is
r 2 2
r
0 4 N(n N) + (N 1)bNMn

0
r 2

+ 2(bNMn )2 + 4(n 2)bNMn 2 20 + O 30 ,


bNMn 2 (N M)n N 2 + M(M 1)
ZNMn (r) = 1 + 2bNMn
(4.4)
(4.5)
bNMn is a solely combinatorial factor which arises from counting the number of replica
contractions associated with ONMn . We then use Eq. (3.2) to rewrite the series in 0 as a
series in , yielding NMn ().
d(log(ZNMn (r)))
d(log(r))

= 2bNMn 8 N(n N) + (N 1)bNMn 2 + O 3 .
NMn ()
(4.6)
Note that the poles in again cancel. To obtain the scaling dimensions XNM of ONMn at
the disordered fixed point, we take n 0 and (from Eq. (3.4)), yielding
XNM = 2Nh12 NM ( ) =

N
+ N(N 1) M(M 1)
2
2

2
N(3N 4) + 2(N 2)[N(N 1) M(M 1)]
8

+ O 3 .
(4.7)
In the unperturbed (non-random) theory, for all M, hONMn (x1 )ONMn (x2 )iN will of course
fall off as |x1 x2 |4Nh12 . But with the defect it will fall off as |x1 x2 |2XNM , where
Q
2XNM = 4Nh12 2NM ( ). Because N

=1 12 is a linear combination of the scaling
operators ONMn , the moment h12 (x1 )12 (x2 )iN will be a sum of terms decaying with
powers XNM . For |x1 x2 | large, this will be dominated by the smallest power, and it is
easy to see that this is XN XNN , giving our main result:
h12 (x1 )12 (x2 )iN |x1 x2 |2XN ,

2
N
3
1 (3N 4) + O
.
XN =
2
4
(4.8)
(4.9)
Note that XN < NX1 , so that the asymptotic scaling is not self-averaging. Instead, we
have an infinite number of independent scaling dimensions all associated with the single
operator 12 . Note that XN1 /N1 > XN2 /N2 for N2 > N1 , as required by convexity. The
result above also yields the typical [23] exponent:
Xtypical =

1
1 + 2 + O 3 .
2
(4.10)
693
Bulk random bond 2D Potts model

The combinatorial problems encountered here are the same as those for the q-state Potts
model with disorder in the bulk ferromagnetic couplings. This is related to the fact that the
correlation function of an odd number of (subtracted) fields (12 for the defect problem,
or for the bulk energy operator in the random Potts ferromagnet), vanish in the nonrandom theory for both problems. (For the defect problem this is a consequence of the
conformal fusion rule, whereas for the bulk random Potts ferromagnet it is a consequence
of KramersWannier duality under which is odd.) The specific integrals are different
but have already been done in [21,28]. The only new difficulty is that to two loop order,
the operators ONMn are no longer always multiplicatively renormalizable, but instead mix
with other (descendent) operators. However, luckily, the leading and subleading operators,
ONNn and ON,N1,n , remain multiplicatively renormalizable to this order (see Appendix
B for details). So the combinatorics in Appendices B and C also give the 2-loop result for
the decay exponent of the N th moment of the two-point function of the energy operator ,
in a Potts model with bulk bond disorder. We have:
e
h(x1 )(x2 )iN |x1 x2 |2XN ,

2
2
3
e
(3N 4)(q 2) + O(q 2) .
XN = N 1
9 2
(4.11)
(4.12)
This also gives us the typical decay exponent:

etypical = 1 + 8 (q 2)2 + O(q 2)3 .
(4.13)
X
9 2
Earlier numerical transfer matrix calculations on the random-bond q-state Potts
model [22] had difficulty calculating dimensions of energy moments, and resulted in
exponents violating the Chayes bound [33]. However, more recent numerical results by
Jacobsen [30] fix earlier problems, and find decay exponents in agreement with our results
here.
5. Moments of the one-point function of 12

We now calculate the disorder-averaged moments of the one-point function of 12 ,
evaluated at a distance y from the defect line (Fig. 2). The method is similar to that in [34].
We no longer need to worry about the various irreducible representations of the symmetric
Fig. 2. A one-point function for an operator in the bulk.
694
group, because the one point function of ONMn vanishes by symmetry except when M = 0.
The disorder-averaged odd moments of 12 vanish because the disorder-averaged system
is symmetric under 12 12 . The even moments are calculated perturbatively:
* 2N
+
Y
h12 (x = 0, y)i2N =
12 (0, y)
*"
=
=1
2N
Y
12
(0, y) e
dx
=1
=y
(1)N
"
X
i=N
i
1
0 y
i!
i
Y
n
X
j =1
j 6=j
y (1)N
X
i=M
*"
Pn
6= 12 12
0,cutoff a
2N
Y
#"
12
(0, 1)
=1
12j (xj , 0)12j (xj , 0)

i
1
a
.
0 y Ii(N)
i!
y
!#"
i Z
Y
#
dxj
j =1
i
Y
(k<l)=1

#+
a
|xi xj |
y
(5.1)
In the 2nd line we have introduced a cutoff a to regulate the short-range divergences. In
the 3rd line we have expanded out the effect of the defect perturbatively in 0 , and used
the conformal symmetry to rescale each expectation value by y. In the lowest order term,
(N)
IN , to get a nonzero expectation value, the 2N operators on the defect must lie in different
replicas so nothing special happens when two defect terms get close together, and we
can drop the cutoff a.
" Z
#N

dx
a
(N) a
N
= (2N)!
,

.
(5.2)
=
(2N)!
1
+
O
IN
y
y
(x 2 + 1)1
We can now use the operator product expansion (OPE)

#" n
#
" n
X
X
0
0
12
(x + , y = 0)12(x + , y = 0)
12
(x, y = 0)12 (x, y = 0)
0 6= 0
6=
n
4(n 2) X
12 (x, y = 0)12 (x, y = 0),

1
for 0
(5.3)
6=
to get the leading poles in for the higher-order terms, by taking derivatives of Ii(N) ( ay )
with respect to ay . When we take derivatives of the step function (|xi xj | ay ), we bring
two operators close together, and so can use the OPE. We get
(N)
1+

Ii ( ay )
a
(N) a
= 4i(i 1)(n 2)
Ii1
, for i > N.
(5.4)
(a/y)
y
y
(N)
(N)
(N)
for i > N . Ii
We already have IN , so we can now repeatedly integrate to get Ii
must be finite as 0 for a 6= 0, because with a finite cutoff a, there are no ultraviolet
695
singularities in the integrals for Ii(N) this requirement determines all the constants of
(N) a
( y ) is
integration. So, for example, the leading term in IN+1

1 a
a
a
(N)
(N)
= 4N(N + 1)(n 2)IN (0)
+ constant 1 + O
IN+1
y
y
y
( a ) 1

a
y
(N)
1+O
= 4N(N + 1)(n 2)IN (0)

y

a
0
.
(5.5)
4N(N + 1)(n 2)IN(N) (0) log
y
(Note that this does diverge as a 0.) More generally, by repeatedly integrating Eq. (5.4)
and using the requirement that all terms be finite as 0 for a 6= 0, we find

iN

i!(i 1)!
a
(N) a
IN(N) (0), i > N. (5.6)
=
4(n 2) log
Ii
y
(i N)!N!(N 1)!
y
Now taking the leading divergences of each term in Eq. (5.1) gives

iN
(N)
(0 y 1+2 )N IN (0) X (i 1)!
a
2N
h12 (y)i =
4(n 2) log
y 0
N!(N 1)!
(i N)!
y
i=N

N
(N)
I (0) (1)N
0 y
y
= N
.
(5.7)
N!
1 + 4(n 2) log( ay )y 0
In the front we have extracted a constant and the expected power-law dependence
y (1)N = y 2h12 N . The remaining terms are written as a function F [a, y, 0] of the
large distance y and the microscopic variables a and 0 . However, we know from the
CallenSymanzik equation that this can be rewritten in terms of a single scaling function
dependent only on the renomalized coupling for length scales of order y, (log( ya )) [34].
Explicitly,

= G log ya .
(5.8)
F [a, y, 0] = F ae`, y, (`) = F y, y, log ya
We rewrite 0 in terms of (log( ya )) by integrating Eq. (3.3) to 1st order, getting

2
3
+ O log ya
+ .
0 y = log ya + 8 log ya log ya
(5.9)
Substituting this into Eq. (5.7), we indeed get that the leading divergences at all orders of
perturbation theory sum up to give a function which depends only on (log( ya )):

y
(2N)! (log( a )) N
h12 (y)i2N =
,
h12 (y)i2N+1 = 0.
(5.10)
N!
y 2h12
We see that the amplitudes of one-point functions are not self-averaging: h12 (y)i2N 6=
N
h12 (y)i2 . That is, the average of the N th power is different than the N th power of the
average. The amplitude ratios are universal properties of the random defect fixed point.

Also note that N11 log h12 (y)i2N1 < N12 log h12 (y)i2N2 for N1 < N2 , as required
by convexity.
696
6. Ising model
We now look at the defect introduced in Eq. (2.1) for the special case of the Ising model.
As noted at the end of Section 2, the analysis in the three sections above fails when m = 3,
because along with the perturbation in Eq. (2.5), we generate the marginal non-random
operator 13 . The minimal model with m = 3 corresponds to the Ising model, and in this
model 12 is the spin operator ( ) and 13 is the energy operator (). So in this case our
defect consists of a random applied magnetic field and a non-random energy operator. Our
action is
#
Z " X
n
n
X
dx
(x, 0) (x, 0) +
(x, 0) .
(6.1)
S = SIsing +
6=
=1
In the lattice formulation of the Ising model, perturbing with the energy operator is the
same as changing the bond strength. Defects where the bond strength is changed along a
single line have been studied and solved exactly by Bariev and McCoy et al. [35,36]. They
dealt with cases where the bond strength was changed only in the bonds perpendicular to
the defect (the ladder geometry Fig. 3), or only in the bonds parallel to the defect (the
chain geometry Fig. 4). Our defect, for the Ising model, is thus a perturbation with a
random magnetic field, of these exactly solved defects. The scaling dimension of the spin
operator along the defect is x = 12 g()2 , where

21
2
1
for the ladder geometry,
(6.2)
g() = tan
tanh(Kc + )
and

1
cos1 tanh(2) for the chain geometry.
(6.3)
Here we use the results of Bariev and McCoy

et al. [35,36], and along the defect have
changed the bond coupling from Kc = log( 12 (1 + 2)) to Kc + . This is not quite correct,
because is the coefficient of the energy operator in the continuum formulation, while
K = Kc + is the coupling in the lattice formulation. However, taking this difference into
account will only give a (-dependent) rescaling of our renormalizaion group flows, which
will not affect the qualitative results. Some subtleties regarding the branch of the arctangent
g() =
Fig. 3. The ladder defect in the Ising model.
Fig. 4. The chain defect in the Ising model.
697
in Eq. (6.2) for antiferromagnetic ladder couplings are dealt with in Appendix D. We get
the OPE coefficient b = g() from [37,38]:

2
(6.4)
x+
= g [1 + g + ].
x
2
2
We can now get the renormalization group equations to 1-loop order solely from the
OPEs [21,34]:

d
= 1 g()2 82 ,
(6.5)
d`
d
= 4g()2 .
(6.6)
d`
The flows for the ladder and chain cases are shown in Figs. 5 and 6. The flows for
the ladder case show that perturbations about the point = = 0 (the defect-free point)
eventually flow to the point with (, ) = (0, Kc ). This value of corresponds to
vanishing bond strength along the ladder. So the renormalization group flow takes us to
a point with two decoupled half-plane Ising models and no random magnetic field.
Fig. 5. Renormalization group flows for the ladder defect in the Ising model. is the strength of the
disordered magnetic field along the defect line, and is the bond strength along the line.
Fig. 6. Renormalization group flows for the chain defect in the Ising model.
698
We can check our result by looking at the flows around the decoupled point. At this point
the spins can be represented by free fermions, and we can calculate the renormalization
group equations about this point. These results agree with the flows around the decoupled
point obtained from Eq. (6.5) and Eq. (6.6). In making the comparison, it is important to
note that in the replica formalism, disorder in the magnetic field corresponds to an operator
Pn Pn
=1
=1 (x, 0) (x, 0), but in Eq. (6.1), is the coefficient of this operator with
the = terms removed. This means that is not really the strength of the random
magnetic field, but a linear combination of the strength of the random magnetic field and
the bond strength (). So, in Fig. 5, adding a random magnetic field to the decoupled point,
(, ) = (0, Kc ), moves us to a point with > 0 and > Kc , and the renormalization
group flows from this new point eventually go back to the original decoupled point.
The flows in the chain case take perturbations about the defect-free point to (, ) =
(0, ). Again, the random magnetic field has vanished at the new fixed point. To
interpret this value of , we look at the spin operator. The dimension of the spin operator
at = is given by Eq. (6.3) as x = 12 g()2 = 12 , which is the same as the
dimension of the spin operator along the edge of an Ising model with free boundary
conditions. The possible boundary conditions of the Ising model have been completely
classified [39] and the only boundary condition where the spin operator has dimension 1/2
is the free boundary condition. This makes sense physically, because = corresponds
to an infinitely antiferromagnetic coupling that produces alternating spins along the defect,
which upon coarse-graining gives net magnetization zero everywhere along the line. We
conclude that in the chain case we also flow to two decoupled Ising models with free
boundary conditions.
If is not hard to see that adding higher cumulant terms of the magnetic field to Eq. (6.1)
will not change the qualitative results of our 1-loop renormalization group calculations.
So far, we have represented the perturbation in the energy operator as changing the bond
strength either in the vertical or the horizontal direction, but not in both. More generally, we
should represent the perturbation as changing bond strengths in both directions. However,
analogously to the ladder and chain cases, we expect that a more isotropic treatment would
only give a different monotonically decreasing function g(), and that the point with
(, ) = (0, g 1 (1)) would still be a stable fixed point with a large basin of attraction
(including the defect-free model at = = 0). And by the classification of Ising model
boundary states in [39], the point with (, ) = (0, g 1 (1)) will always correspond to two
decoupled half-plane Ising models with free boundary conditions and no random magnetic
field.
7. Conclusions
We have found a new universality class of disordered defect lines. The defect lines
exist in various two-dimensional statistical mechanical models, such as the tricritical Ising
model and tricritical 3-state Potts model. The large-distance behavior of these defect lines
has been shown to be disorder-dominated. Two-point functions along the defect exhibit
699
multifractal behavior, and universal (normalized) amplitudes of one-point functions are

non-self-averaging. We have also argued that when a random magnetic field is applied
along a single line of the Ising model, it causes the two sides of the defect to decouple, and
to turn into two half-plane Ising models with free boundary conditions and no disorder.
Results for the defect line in the mth Virasoro minimal model were obtained by a 1/m
expansion. However, the physically most interesting models are at low m. For example,
m = 4 corresponds to the tricritical Ising model, with a random bond strength (or a
random chemical potential) along a line. Our calculations show that this results in disorderdominated long-distance behavior. It would be interesting to understand this model in a
more fundamental and non-perturbative fashion. The random boundary/defect fixed points
that we have found in this paper are, besides the bulk random bond q-state Potts models,
a rare case where detailed analytic information about random critical behavior is available.
In particular, it would be most interesting to compare our results for the random defect
lines in minimal models with future numerical results, such as those obtained by Jacobsen
and Cardy for the bulk random Potts models [22].
As a byproduct of our calculations for the defect line, we have also obtained multifractal
energyenergy correlations in the bulk random bond q-state Potts model. Our results are
in good agreement with recent numerical results by Jacobsen [30].
Finally, we comment on the RG analysis performed under the assumption of broken
replica symmetry. Such a calculation was done in [40] for the q-state Potts model with bulk
disorder in the bond strength, where it was found that the replica symmetric disordered
fixed point is unstable to a new fixed point with broken replica symmetry. However,
numerical tests show that the Potts model with random bonds is best described by the
replica symmetric fixed point [41,42]. An identical analysis to that of [40] for the random
defect problem that is the subject of the present paper shows that, again, the fixed point
considered in this paper is unstable, and flows to a new stable fixed point with broken
replica symmetry. Thus, our random defect problem may provide further insights into the
significance of the replica broken fixed point.
Acknowledgements
We thank Tom Davis for useful discussions. This work was supported in part by a
Regents Fellowship of the University of California (M.J.), and in part by the A.P. Sloan
Foundation (A.W.W.L.).
Appendix A. Renormalization of
We calculate the renormalization of to second order. The structure of the calculations
closely parallels that in [28]. We get the coefficients in

(r) = r 0 + A2 (r, )20 + A31 (r, ) + A32(r, ) + A33 (r, ) 30 + , (A.1)
700
where, as explained below, the different terms come from the different types of contractions
at each order of perturbation theory. All integrals are regularized at short distances by
3
and at large distances by an infrared cutoff r.
analytic continuation in m+1
A.1. First order
To lowest order we bring two perturbation terms together
Z
n
n
X
X
20
dx1 dx2
12
(x1 )12 (x1 )
12 (x2 )12
(x2 ).
2
6=
|x1 x2 |<r
(A.2)
6=
We get the same perturbation back again (and thus a contribution to ) when = and
6= . This gives us the A2 term in Eq. (A.1)
Z

r
(A.3)
dx2 12 (x1 )12 (x2 ) 0 = 4(n 2) .
A2 = 2(n 2)

|x1 x2 |<r
The 0 subscript on the correlator indicates that it is calculated in the defect-free theory.
A.2. Second order
To second order we have
Z
Z
n
n
X
X
30
dx1 dx2 dx3

12
(x1 )12 (x1 )
12 (x2 )12
(x2 )
3!
|x1 x2 |<r
6=
|x1 x3 |<r
n
X
6=
12 (x3 )12
(x3 ).
(A.4)
6=
This gives us several possible contractions. We get one possible contraction when = ,
= , 6= , 6= , 6= :
A31 = 4(n 2)(n 3)I1 ,
Z
dx2
I1
|x2 x1 |<r
dx3 12 (x1 )12 (x2 ) 0 12 (x2 )12 (x3 ) 0
|x3 x1 |<r
dx2
|x2 x1 |<r
= 2r
dy|y|
0

1+

x2 x3 1+
dx3 x1 x2
|x3 x1 |<r
Z1
2
(A.5)
1+2
dz|z|1+ |1 z|1+ .
(A.6)
|z|<1/y
1
1
and z = xx32 x
We have transformed to coordinates y = x2 x
r
x1 . We can now extend the
integral over z to go from to , since this will only change I1 to O( 0 ). This gives
us two integrals which we can do exactly:

212
(1 + cos ) ()

+ O 0 .

I1 = 2r 2
1
2
1
2
701

+ O 0
r 2
(A.7)
2
Note that in the integrals we have picked out one point x1 as special, and integrated
over x2 and x3 , but in the combinatorial factor 4(n 2)(n 3) we have treated all
points symmetrically (i.e., have treated = , = the same as = , = ). This
is permissible to this order in perturbation theory, since we are only concerned with the
poles in , which result from short range divergences and are the same for any permutation
of the contractions. We can see this explicitly by considering a different arrangements of
the same contractions:
Z
Z

dx2
dx3 12 (x1 )12 (x2 ) 0 12 (x1 )12 (x3 ) 0
I01
= 4
|x2 x1 |<r
= 4
r 2
2
|x3 x1 |<r

= I1 + O 0 .
(A.8)
We get a second possible set of contractions when = = and = 6= :

A32 = 4(n 2)I2 ,
Z
dx2
I2 =
|x2 x1 |<r
(A.9)
Z
|x3 x1 |<r
dx3 12 (x1 )12 (x2 )12 (x3 )12 () 0
2
12 (x2 )12 (x3 ) 0 12 (x2 )12 (x3 ) 0 .
(A.10)
The four-point function gives the coefficient of three 12 s projecting to a single 12 . We

have subtracted off (h12 (x2 )12 (x3 )i0 )2 this term corresponds to the contribution from
two perturbation terms (disorder operators) getting close, and does not affect the dimension
of the operator at x1 . This term is only a contribution to the free energy, so subtracting it off
simply corresponds to normalizing the correlation functions. We get the four-point function
from the Coulomb gas formalism [43,44]:

12 (0)12 (1)12(z)12 () 0 = |z|1+ |1 z|1/3

2

3k1 24/3
+ F 1 , ; 2 ; z 2 , (A.11)
|z|
F 1 3 , 2 ; 2 2
;
z
3
3 3 3
4
k1

2
4( ( 2
3 )) (2 ) (1 3 )
3 ( 3 ) (1 + ) (2
2
3 )
= 1 + O()
(A.12)
and F is the hypergeometric function. If we take the limit as 0 at fixed z for the
four-point function, we get
702

12 (0)12 (1)12(z)12 () 0
sign(z) sign(z 1)
.
(A.13)
|z 1|
|z|
We now go back to I2 , which we can calculate by using the symmetries x2 x2
and x2 x3 to cut the integration region down by a fourth, and then transforming to new
1
1
and z = xx32 x
coordinates y = x2 x
r
x1 . The integral now exactly factorizes into two onedimensional integrals:
0
sign(z) sign(z 1) +
Z1
I2 = 4r
2
dy y
1+2
Z1

dz 12 (0)12 (1)12(z)12 () 0
The y-integral gives
Z1
1
2 ,
2
12 (1)12(z) 0 12 (1)12(z) 0 .
(A.14)
while the z-integral can be rewritten as

dz 12 (0)12(1)12 (z)12 () 0 12 (1)12(z) 0
2
12 (0)12 (z) 0 12 (1)12 (z) 0
Z1
+

dz 12 (0)12(z) 0 .
(A.15)
The second integral is exactly 2/. It is straightforward to use Eq. (A.11) to check that
as z 1, the four-point function goes to |z 1|1+ + O(z 1). So the first integral
converges everywhere, and we can get its value to O( 0 ) by simply replacing with zero
inside the integral:

Z1
dz
1

sign(z) sign(z 1)
1
sign(z) sign(z 1) +
|z 1|
|z 1|
|z|

1
1
+ O()
|z| |z 1|2
= 1 + O().
(A.16)
Putting this all together gives

2
2
1
1
1 + O() = 2r 2
2 .
I2 = 4r 2
2

Again, as with A31 , other possible permutations of this contraction like
Z
Z

dx2
dx3 12 (x1 )12 (x2 )12 (x3 )12 () 0
I20 =
|x2 x1 |<r
|x3 x1 |<r

12 (x1 )12 (x2 ) 0
(A.17)
(A.18)
703
give the same result up to terms of O( 0 ).

Finally, the last possible contraction in Eq. (A.4) comes from = = and = = .
This is
Z
Z
2

4
dx2
dx3 12 (x1 )12 (x2 )12 (x3 )12 () 0
A33 =
3
|x2 x1 |<r
|x3 x1 |<r
2
12 (x2 )12 (x3 ) 0
16 2
r
3
Z1
dy
y 12
Z1
dz
1
2
12 (0)12 (1)12(z)12 () 0
2
12 (1)12 (z) 0 .
(A.19)
The y-integral gives 2/. The z-integral is evaluated similarly to Eq. (A.15). We rewrite it
as
Z1
2
dz 12 (0)12 (1)12(z)12 () 0
1
2
2

12 (0)12 (z) 0 12 (1)12 (z) 0 1
Z1
+
2
dz 1 + 12 (0)12(z) 0 .
(A.20)
The second integral can be evaluated exactly, and is O(). The first integral is nowhere
divergent, so we can get its value to O( 0 ) by taking to 0 inside the integral. Then
squaring Eq. (A.13) gives
2
12 (0)12(1)12 (z)12 () 0

sign(z) sign(z 1) 2
0
sign(z) sign(z 1) +
|z 1|
|z|
=1+
1
1
+
.
2
z
(z 1)2
(A.21)
So both z-integrals are O(), and A33 is O( 0 ). Since A33 has no pole in , it can be
dropped to this order. Combining all these results gives Eq. (3.2).
Appendix B. Renormalization of NMn

To get the dimensions of the moments of 12 , we need the coefficients of 0 and 20
in ZNMn (r), where NMn (r) = ZNMn (r)NMn,0 . The integrals in this section will be the
same as the integrals in Appendix A. The only difference will be in the combinatorial
factors which precede them. In this section we will use hN, M, n| and |N, M, ni to
represent the bra and ket forms of the operator ONMn defined in Eq. (4.2). However, before
704
we start, we should make sure that no problems occur with mixing, either for the defect
line considered in this paper, or for the parallel calculation in the q-state Potts model with
bulk disorder in the bond strength.
B.1. Mixing with other operators
Random defect
Q
.
The combinations ONMn of Eq. (4.2) diagonalize the operators of the form 12
However, we also need to check that these operators do not mix with other operators that
have the same scaling dimensions at m = . We only need to consider operators of the
form 1q , because these operators form a closed subalgebra in the minimal models. To 2
loops, only 4 12 operators can be affected by the perturbation, and 4 dim(12 ) = 2 at m =
. It easy to see that the only operator, or descendent of an operator, or combination of
operators, in the 1q subalgebra with dimension 2, is 13 (therefore, to this order, no mixing
occurs with derivative operators). 13 does in fact mix with (12 )4 , and so 13 (12 )N4
mixes with (12 )N . However, the 2-loop overlap integral is
Z
Z

dx2
dx3 12 (x1 )12 (x2 )13 () 0
|x2 x1 |<r
|x3 x1 |<r
2
12 (x1 )12 (x2 ) 0 12 (x1 )12 (x3 ) 0 .
(B.1)
This has no poles in , so we would only need to take this mixing into account to do 3-loop
calculations.
Bulk random q-state Potts model
Mixing is potentially more of a problem when we use these calculations to get moments
of the energy operator in the q-state Potts model with bulk bond disorder this model
is analyzed by expanding about the Ising model [23,28], so we need to look for mixing
with other operators that have the same scaling dimension at q = 2. Denoting the energy
operator in the replica by , we see that and 2 have the same dimension
at q = 2, and that in general we can replace any two s with a 2 without changing the
dimension. These operators mix to two loops, with overlap integrals such as
Z
d 2 x2 d 2 x3 (0) (x2 ) (x3 ) 2 ()
Z

= d 2 x2 d 2 x3 (0)(x2 )(x3 ) 2 () 0

(0)(x2 ) 0 (0)(x3 ) 0 .
(B.2)
This means at the combinations ONMn of Eq. (4.2) (with 12 replaced by ) are no
longer generally multiplicatively renormalizable, but mix with derivatives of ON2,M,n
e .
Luckily, however, no problem with mixing occurs for the operators ON,N,n and ON,N1,n ,
which we know from the one-loop calculations to provide the leading and subleading decay
exponents. No problem with mixing occurs for ON,N,n because the overlap

n
X
6=
n
X
|N, N, ni 2 |N 2, M, ni
705
(B.3)
6=
is 0 for all M, where the 2 is understood as acting on only one of the s. It is not hard to
see this by counting the connected contractions of
hN, N, n|
n
X
6=
n
X
|N 2, M, ni.
(B.4)
6=
If we expand |N 2, M, ni into monomials, each monomial term gives a contraction of

0 this is because after contracting all the s in the monomial term, we will be left
with at least two antisymmetric terms from the hN, N, n|, and thus positive and negative
contractions will cancel. So the contractions give a total of 0 (even before the replica limit
n 0 is taken). This is true even if the |N 2, M, ni has a 2 on it, and even if hN, N, n|
is replaced with hN, N 1, n|. So mixing is not a problem for the ON,N,n and ON,N1,n
operators in the q-state Potts model with bulk bond disorder.
Note that this argument for the q-state Potts model works not just for descendent
operators with 2 = terms in them, but also for operators containing
terms.
Since the above argument for the absence of mixing to two loop order is crucial for
the validity of our main result, Eq. (4.13), for the bulk random bond Potts model, we will
present it again in more explicit form: the bras hN, N, n| or hN, N 1, n|, to be used on
the l.h.s. of a correlator such as Eq. (B.4), consists of a sum of terms
0
0
0
(for hN, N, n|),
(B.5)
1 1 2 2 N N
where i {1, 2, . . . , (n N)} and 0 i {(n N) + 1, . . . , n}, or

0
0
0
1 1 2 2 N1 N1 N (for hN, N 1, n|),
(B.6)
where i {1, 2, . . . , (n N) 1} and 0 i {(n N), . . . , n}. The expressions in

Eqs. (B.5), (B.6), are sums of monomials of the general form
a1 a2 aN
(ai 6= aj for i 6= j ).
(B.7)
Therefore, in the relevant matrix element of the second order perturbation between such
a bra and a potentially mixing operator in the ket with two derivatives (bi 6= bj for i 6= j ),

(B.8)
b1 b1 b1 b2 bN2
or

b1 b1 b2 b2 bN2
(B.9)
any of the (N 2) pairwise unequal indices bi must be equal to at most one of the (also
pairwise unequal) indices ai (from the bra), of which there are N in number. The crucial
point is that this leaves at least two indices ai and aj uncontracted with (i.e., unequal to)
the set of indices bi . Because the forms in Eqs. (B.5), (B.6) have all but perhaps one of the
706
operators in an antisymmetric term, it follows that either ai or aj is in an antisymmetric

term. This term must get contracted with one of the indices (say, ) of the perturbation
X
X

(B.10)
(x2 )
(x3 ).
6=
6=
This gives

0
i i = (i i0 ) ,
(B.11)
where neither i , i0 , or is equal to any of the bs, and is summed over all replica
indices except the bs. The sum over will thus give a factor of (1 1)h i = 0 (the
remaining combinatorial factors being the same regardless of whether = i or = i0 ).
We finally note that, on the other hand, bras hN, N M, n| with 2 6 M 6 N are sums
of terms as in Eqs. (B.5), (B.6), but now with only (N M) 6 (N 2) antisymmetric
factors. After (N 2) pairwise unequal indices are contracted with those of kets of the
form of Eqs. (B.8), (B.9) as above, non-antisymmetric terms do in general survive. Hence,
the above argument proving the absence of mixing to two loop order now breaks down.
This completes our discussion of mixing for the bulk random bond Potts model.
B.2. First order
Returning to our defect model, the 1st order correction comes from
!
n
X

|N, M, ni bNMn |N, M, ni,
12 12
(B.12)
6=
P
with one
where we contract one 12 operator in the n6= 12
12 operator in the
12
|N, M, ni. The integral is the same integral done in Eq. (A.3), giving 2r /. We can get
the combinatorial factor bNMn by contracting with hN, M, n| on both sides:
P
|N, M, ni
hN, M, n| n6= 12
12
bNMn =
.
(B.13)
hN, M, n| M, N, ni
This combinatorial factor is found in Appendix C.2. Leaving our result in terms of bNMn ,
the order 0 term in ZNMn is

r
(B.14)
0 .
2bNMn

B.3. Second order
To second order, we need to look at the number of ways that we can take
hN, M, n|
n
X
6=

12
12
n
X
12 12
hN, M, n| .
(B.15)
6=
The index in hN, M, n| is a label used to make discussion easier, and does not signify an
independent variable. The simply indicates that for discussing contractions, the dummy
indices i in Eq. (4.2) will all be called i for i = 1, 2, . . . , n.
707
There are now a number of possible contractions to consider. The one with i = =
= j and k = = = l (for some i,j ,k,l) gives an integral over a squared four-point
function. We did this integral for A33 in Appendix A and found that it had no singularities.
So we can ignore this term.
We now consider the contraction with i = = = j and = , which gives a
combinatorial factor that we call Q(2)
NMn , and the contraction with i = = = j , k =
, and = l , which gives a combinatorial factor that we call Q(3)
NMn . It is to be understood
that no replica indices other than the specified ones are equal to each other. In both cases,
we get the same integral as in Eq. (A.10) (or Eq. (A.18)), which was found to be I2 =
2r 2 ( 1 22 ).
To get the combinatorial factor for Q(2)
NMn , we look at
Pn
Pn

hN, M, n|
6= 12 12
6= 12 12 |N, M, ni
hN, M, n | N, M, ni
(B.16)
and only allow the contractions described above. For the contractions allowed in Q(2)
NMn ,
the replica indices that appear in a monomial term must be the same in the bra and ket
sides. The = replica index can be the same as any of the these N monomial terms. The
= replica index must be something different, so has (n N) choices. We also get an
overall factor of 4 for the different pairs of terms that we could have chosen to contract. So
Q(2)
NMn = 4N(n N).
left and the index conQ(3)
NMn can be calculated similarly. The index contracts to the
Pn
|N, M, ni =
tracts to the right this gives a combinatorial factor hN, M, n| 6= 12
12
bNMn hN, M, n | N, M, ni. The = index can then be any of the (N 1) replica indices
in the right (or left) |N, M, ni (excluding the two which are equal to or ). And there is
(3)
again an overall factor of 4, so QNMn = 4(N 1)bNMn .
Now consider the contraction with i = , = and = j , which gives a
(4)
combinatorial factor QNMn , and the contraction with i = , j = , i = , and j = ,
(5)
which gives a combinatorial factor QNMn . Both of these terms give the same integral as in
Eq. (A.6) (or Eq. (A.8)), and thus a factor of I1 = 4r 2 / 2 .
(4)
(3)
(4)
QNMn can be evaluated in the same manner as QNMn , giving QNMn = 4(n
(5)
N 1)bNMn . The term QNMn is more complicated and is found in Section C.3 to be
(5)
QNMn = 4N(n N) 2(n 2)bNMn + (bNMn )2 . Putting this all together, we get the
result

r 2
1 (2)
r 2
r
(3)
+4 2
ZNMn = 1 + 2bNMn 0 + QNMn + QNMn 2

2

2

r
1 (4)
+ O 30
+ QNMn + Q(5)
NMn 4 2
2

r 2 2
r
= 1 + 2bNMn 0 4 N(n N) + (N 1)bNMn

0
r 2

+ 2(bNMn )2 + 4(n 2)bNMn 2 20 + O 30 .
(B.17)

708
Appendix C. Combinatorial factors

In this appendix we calculate the combinatorial factors used in Appendix B. We want
the number of possible contractions between the irreducible representations of SN , given
by ONMn in Eq. (4.2), and copies of the disorder operator. The trivial spatial dependence
is suppressed in the equations below.
C.1. Normalization
First, we calculate the normalization of |N, M, ni
A[N, M, n]
hN, M, n | N, M, ni
X
X
=
i 6=j
i 6=j
16i 6(nM) 16i 6(nM)

n(M1) M+1

1
M
n
12
12
12
12
12 12N

n(M1) M+1
n
12M 12
12 12N .
121 12
(C.1)
M
If we expand out the two terms (12
n(M1)
12
)
and (12M
n(M1)
12
) into monomials,
the cross terms are 0, and the remaining terms are of the same form A[ ] as before, but
with different values of M, N and n. We get
A[N, M, n]
= A[N, M 1, n 1] + (n M N + 1)2 A[N 1, M 1, n 1].
(C.2)
We can calculate A[N, M, n] explicitly when M = 0:

X

X

1 2 N 1 2 N
A[N, 0, n] =
12 12
12
12 12
i 6=j
i 6=j
16i 6n 16i 6n
N
Y
(n N + i).
= N!
12
(C.3)
i=1
Here the N! comes from the number of ways to contract the left side with the right side, and
the factors of n N + i come from the different ways to pick the i once the contractions
have been done. Given the value of A[N, M, n] when M = 0, and the recursion relation
above, we can show by induction that the general solution is
#" M
#
" N
Y
Y
(n M N + i)
(n 2M + 1 + i) .
(C.4)
A[N, M, n] = (N M)!
i=1
i=1
C.2. OPE 1st order term

Here we want to calculate
B[N, M, n] hN, M, n|
n
X
6=
12
12 |N, M, ni.
(C.5)
709
Note that three-point functions of 12 are 0 by the 12 12 duality symmetry.
n(M1)
n(M1)
M
12
) and (12M 12
) terms into
As before, we expand out the (12
monomials. The direct terms again give contractions of the form B[ ], but we now have
a cross term in which either or must be equal to n (M 1). We get
1
B[N, M, n]
2
1
= B[N, M 1, n 1]
2
1
+ (n M N + 1)2 B[N 1, M 1, n 1]
2
2(n M N + 1)
n
X
X
X
i 6=j
i 6=j
=1
16i 6(nM) 16i 6(nM)
1
n(M2) M M+1
N
n
12
12
12
12M1 12
12 12 12

M1
1
N
n(M2) M+1
n
12 12 12 12
12 12 .
(C.6)
The contraction in the last term looks just like the contraction A[N, M 1, n 1], with
taking the place of M . The only difference between the sum above and A[N, M 1,
n 1], is that the sum over above includes terms with (n M + 2) 6 6 n, and terms
where is the same as some other term i , for 1 6 i 6 (M 1). However, these two
extra contributions are equal and opposite, being equal to (M 1)(n M N + 1)
A[N 1, M 2, n 2]. So they cancel, and we have
1
1
B[N, M, n] = B[N, M 1, n 1]
2
2
1
+ (n M N + 1)2 B[N 1, M 1, n 1]
2
2(n M N + 1)A[N, M 1, n 1].
(C.7)
As with A[N, 0, n], it is easy to calculate B[N, 0, n]:

#
" N
Y
(n N + i) = 2N(n N)A[N, 0, n].
B[N, 0, n] = 2N(N!)
(C.8)
i=0
Given the recursion relation for B[N, M, n], the initial condition B[N, 0, n], and the result
for A[N, M, n] in the previous section, we can show by induction that

B[N, M, n]
= 2 (N M)n N 2 + M(M 1) .
(C.9)
bNMn
A[N, M, n]
C.3. 2nd order term
We want to calculate
C[M, N, n] 4hN, M, n|
n
X
6=
12
12
n
X
6=
12 12
|N, M, ni,
(C.10)
710
where we require that and contract to the left, the and contract to the right, and
, , , are all distinct from one another (other possible directions of contractions give
the factor of 4 in front). This combinatorial factor arises in the contraction for Q(5)
NMn in
n(M1)
M
) and
Appendix B.3. As in the previous subsection, expanding out the (12 12
M
n(M1)
) terms into monomials gives back two terms of the form C[ ], and a
(12 12
more complicated cross-term. The cross term almost has the same form as B[N, M 1,
n 1], but we also get some extra terms because the sums over , , and are different
than we would have in a B[ ] term. Counting all the ways in which our cross term differs
from B[N, M 1, n 1] is tedious but straightforward, and we get
1
B[N, M 1, n 1] (M 1)(n M N + 1)2 A[N 1, M 2, n 2]
2
(M 1)A[N, M 2, n 2] (n M N)A[N, M 1, n 1]
1
(C.11)
= B[N, M 1, n 1] (n N 1)A[N, M 1, n 1].
2
The recursion relation is
1
C[N, M, n]
4
1
= C[N, M 1, n 1]
4
1
+ (n M N + 1)2 C[N 1, M 1, n 1] 4(n M N + 1)
4

1
B[N, M 1, n 1] (n N 1)A[N, M 1, n 1] .
(C.12)
2
Combined with the initial condition,
C[N, 0, n] = 4N(N 1)(n N)(n N 1)A[N, 0, n],
(C.13)
we can find the value of C[N, M, n] for all M and N by induction. The result is
(5)
QNMn
C[N, M, n]
= 4N(n N) 2(n 2)bNMn + (bNMn )2
A[N, M, n]
(C.14)
where bNMn was defined in Eq. (C.9).
Appendix D. Ising ladder defect for K < 0

The branch of the arctangent used in Eq. (6.2) for antiferromagnetic (K = Kc + < 0)
ladder couplings requires some explanation. We take the value of the arctangent to be in
(0, 2 ) if its argument is positive (i.e., K > 0) and to be in ( 2 , ) if its argument is negative
(i.e., K < 0). This makes the slope of g() continuous through K = 0. On the other hand,
the results of [35,36] have a slope discontinuity at K = 0, and have g symmetric under
K K. This slope discontinuity results from a level crossing in the lowest scaling
dimension for operators on the boundary [39]. If we let t be the spin on one side of
the defect, and b the spin on the other side, we see that the operator with the lowest
711
scaling dimension changes from t + b to t b as K goes through 0, so that while the

dimension of each operator changes smoothly through K = 0, the dimension of the lowest
scaling operator does not. However, if we take our random applied magnetic field to not
vary across the defect, it couples to t + b only, and we want the lowest scaling dimension
for K > 0, but the 2nd lowest scaling dimension for K < 0. We can get these from [39],
thus justifying the branches of arctangent chosen above.
Note that if we had used the other branch of the arctangent, corresponding to a magnetic
field uncorrelated across the defect, the flow picture in the ladder case would have been
symmetric under K K, and the entire (, ) plane would have flowed into the
decoupled point.
References
[1] P. Ginsparg, in: E. Brezin, J. Zinn-Justin (Eds.), Fields, Strings and Critical Phenomena, Les
Houches XLIX, North-Holland, Amsterdam, 1990.
[2] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B. 241 (1974) 333.
[3] D. Friedan, Z. Qiu, S. Shenker, Phys. Rev. Lett. 52 (1984) 1575.
[4] M.J. Tejwani, O. Ferreira, O.L. Vilches, Phys. Rev. Lett. 44 (1980) 152.
[5] O. Ferreira, He/Kr/Grafoil measurement, Ph.D. dissertation, Instituto de Fisica, Universidade
de Campinas, Campinas, Brasil, 1978 (unpublished).
[6] M.J. Tejwani, 4 He/Kr/graphite foam measurement, Ph.D. dissertation, University of Washington, 1979 (unpublished).
[7] M. Sokolowski, H. Pfnur, Phys. Rev. B 49 (1994) 7716.
[8] A.N. Berker, S. Ostlund, F.A. Putnam, Phys. Rev B. 17 (1978) 3650.
[9] J. Cardy, Scaling and the Renormalization Group in Critical Phenomena, Cambridge University
Press, Cambridge, 1996, Chapter 8.
[10] A.B. Harris, J. Phys. C 7 (1974) 1671.
[11] M. Aizenmann, J. Wehr, Phys. Rev. Lett. 62 (1989) 2503;
M. Aizenmann, J. Wehr, Comm. Math. Phys. 130 (1990) 489.
[12] V.S. Dotsenko, J. Phys. A 18 (1985) L241.
[13] P. Simon, Nucl. Phys. B 515 (1998) 624, cond-mat/9710024.
[14] V. Dotsenko, M. Picco, A. Pujol, Phys. Lett. B 383 (1996) 287, hep-th/9512087.
[15] S. Chen, A. Ferrenberg, D. Landau, Phys. Rev. E 52 (1995) 1377.
[16] S. Wiseman, E. Domany, Phys. Rev. E 51 (1995) 3074.
[17] J.L. Cardy, J. Phys. A 29 (1996) 1897, cond-mat/9511112.
[18] G. Grinstein, A. Luther, Phys. Rev. B 13 (1978) 1329.
[19] T. Lubensky, Phys. Rev. B 11 (1975) 3573.
[20] V.J. Emery, Phys. Rev. B 11 (1975) 329.
[21] A.W.W. Ludwig, Nucl. Phys. B 285 (1987) 97.
[22] J. Cardy, J.L. Jacobsen, Phys. Rev. Lett. 79 (1997) 4063;
J.L. Jacobsen, J. Cardy, Nucl. Phys. B 515 (1998) 701, cond-mat/9711279.
[23] A.W.W. Ludwig, Nucl. Phys. B 330 (1990) 639.
[24] C. Castellani et al., J. Phys. A 19 (1986) L429.
[25] T.C. Halsey, M.H. Jensen, L.P. Kadanoff, I. Procaccia, B.I. Shraiman, Phys. Rev. A 33 (1986)
1141.
[26] B. Huckestein, Rev. Mod. Phys. 67 (1995) 357, cond-mat/9810319.
[27] M. Janssen, M. Metzler, M. Zirnbauer, Phys. Rev. B 59 (1999) 15836, cond-mat/9810319.
712
[28] Vl.S. Dotsenko, M. Picco, P. Pujol, Nucl. Phys. B 455 (1995) 701, hep-th/9501017;
Vl.S. Dotsenko, Nucl. Phys. B 314 (1989) 687.
[29] T. Davis, J. Cardy, Nucl. Phys. B 570 (2000) 713, cond-mat/9911083.
[30] J.L. Jacobsen, Phys. Rev. E 61 (2000) R6060, cond-mat/9912304.
[31] T. Lubensky, in: B. Bailan et al. (Eds.), Ill Condensed Matter, Les Houches XXXI, NorthHolland, Amsterdam, 1979.
[32] I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161;
I. Affleck, A.W.W. Ludwig, Phys. Rev. B 48 (1993) 7297.
[33] J.T. Chayes, L. Chayes, D.S. Fisher, T. Spencer, Phys. Rev. Lett. 57 (1986) 2999;
J.T. Chayes, L. Chayes, D.S. Fisher, T. Spencer, Commun. Math. Phys. 120 (1989) 501.
[34] A.W.W. Ludwig, J.L. Cardy, Nucl. Phys. B 285 (1987) 687.
[35] R.Z. Bariev, Sov. Phys. JETP 50 (1979) 613.
[36] B.M. McCoy, J.H. Perk, Phys. Rev. Lett. 44 (1980) 840.
[37] Kadanoff, Phys. Rev. B 3 (1971) 3918.
[38] Kadanoff, Phys. Rev. B 24 (1981) 5382.
[39] M. Oshikawa, I. Affleck, Nucl. Phys. B 495 (1997) 533, cond-mat/9612187.
[40] Vik.S. Dotsenko, Vl.S. Dotsenko, M. Picco, P. Pujol, Europhys. Lett. 32 (1995) 425, hepth/9502134.
[41] Vik. Dotsenko, Vl.S. Dotsenko, M. Picco, Nucl. Phys. B 520 (1998) 633, hep-th/9709136.
[42] M. Lewis, Europhys. Lett. 48 (1999) 655, cond-mat/9905401.
[43] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312;
Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691.
[44] Vl.S. Dotsenko, Adv. Stud. Pure Math. 16 (1988) 123.

Notes on infinite layer quantum Hall systems

J.D. Naud a , Leonid P. Pryadko b, , S.L. Sondhi a
a Department of Physics, Princeton University, Princeton, NJ 08544, USA
b School of Natural Sciences, Institute for Advanced Study, Olden Lane, Princeton, NJ 08540, USA
Abstract
We study the fractional quantum Hall effect in three-dimensional systems consisting of infinitely
many stacked two-dimensional electron gases placed in transverse magnetic fields. This limit
introduces new features into the bulk physics such as quasiparticles with non-trivial internal structure,
irrational braiding phases, and the necessity of a boundary hierarchy construction for interlayer
correlated states. The bulk states host a family of surface phases obtained by hybridizing the edge
states in each layer. We analyze the surface conduction in these phases by means of sum rule and
renormalization group arguments and by explicit computations at weak tunneling in the presence of
disorder. We find that in cases where the interlayer electron tunneling is not relevant in the clean
limit, the surface phases are chiral semi-metals that conduct only in the presence of disorder or at
finite temperature. We show that this class of problems which are naturally formulated as interacting
bosonic theories can be fermionized by a general technique that could prove useful in the solution of
such one and a half dimensional problems. 2001 Elsevier Science B.V. All rights reserved.
PACS: 73.40.Hm; 71.10.Pm; 11.25.Hf; 11.10.Hi
1. Introduction
The quantum Hall effect arises from physics that is special to two dimensions.
Nevertheless, it was demonstrated in an early experiment [1] that it survives a small
amount of three dimensionality; more precisely, that an infinite layer system with interlayer
tunneling weak compared to the single layer (mobility) gap would continue to exhibit
dissipationless transport and a quantized Hall resistance per layer. The quantized Hall
phases in the organics also rely on the same effect [2]. In the presence of tunneling the
chiral edge states which exist in each layer hybridize and yield a family of one and a half
* Corresponding author. Present address: Department of Physics, University of California, Riverside,
CA 92521, USA.
E-mail address: leonid@landau.ucr.edu (L.P. Pryadko).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 7 9 - 9
714
J.D. Naud et al. / Nuclear Physics B 594 [FS] (2001) 713746
dimensional phases that live on the surfaces of the three-dimensional systems and exhibit
interesting transport in the direction transverse to the layers.
Following the pioneering theoretical work [3,4], a flurry of work [5] has focused on the
integer Hall effect in infinite layers systems with a particular emphasis on the properties of
the chiral metal that forms at their surface in the quantum Hall phases and some of this has
found support in experiments [6]. Recently, interaction effects and magnetic order at = 1
(per layer) have been discussed as well [7]. All of this work has antecedents in work on
multi-component systems such as double layer devices [8], but the limit of infinitely many
layers sharpens quantitative differences into qualitatively new features, e.g., a different
universality class for the transitions out of the quantum Hall phases [3].
In this paper we extend the study of three-dimensional quantum Hall phases in two
directions. First, we offer a systematic analysis of the simplest fractional quantum Hall
phases in infinite layer systems with a special focus on the phases that exhibit interlayer
correlations. Here we find several novel bulk properties, most notably a non-trivial structure
for the quasiparticles. This part of our work builds on the pioneering and early work of Qiu,
Joynt and MacDonald (QJM) [9], who studied the energetics of multilayer states, and noted
the irrational partitioning of the quasiparticle charge. The second axis of our work is the
analysis of the one and a half dimensional phases that arise at the surfaces of fractional
quantum Hall phases. In this regard we report a general analysis of the surface conduction
by combining renormalization group arguments (which have strong consequences for the
ground state structure in chiral systems) and a conductivity sum rule. We also carry out
computations of the surface conductivity of the disordered system in a weak tunneling
expansion for a large class of states. Together, these show that the surfaces of states
with marginal or irrelevant electron tunneling are semi-metallic, in that they are perfect
insulators at all frequencies in the clean limit but begin to conduct at finite temperature
and disorder. Some of the results derived here were announced previously in a companion
paper [10].
This last theme, of analyzing the surface phases, is interesting from a more theoretical
standpoint in that the surface phases are half a dimension up from one dimension where
exact solutions are often possible. We have not made very much progress on this front.
What we have accomplished is to solve the case of bilayers in some generality [11,12],
find special solutions for three and four layers, and recast the infinite layer problems
in algebraic and non-standard fermionized formulations that appear potentially fruitful.
We report these here in the hope that some readers will find them useful in carrying this
program to completion. We should also note that the construction of more complex bulk
states than those considered in this paper could well lead to a richer class of such problems.
The outline of this paper is as follows. We begin in Section 2 with a discussion of the bulk
properties of multilayer fractional states, in particular the novel features of their quasihole
excitations which include a non-trivial internal charge structure and irrational statistics.
In Section 3 we discuss the edge theory of general multilayer states in the presence of
nearest-neighbor single-electron tunneling. We specialize to the case where the tunneling
is marginal in Section 4, first considering the clean limit and then adding disorder and
interactions. In Section 5 we extend our analysis to states where tunneling is irrelevant. We
715
conclude in Section 6 by presenting a method for fermionizing the multilayer edge theory
via the addition of auxiliary degrees of freedom. The appendices contain a discussion of
an exactly soluble model problem with marginal tunneling (Appendix A), exact solutions
for the spectra of two edge theories for the special case of four layers (Appendix B), and
some technical details including Klein factors (Appendix C), and the proof of an assertion
made in the main text (Appendix D).
2. Bulk properties
Consider a system which consists of N parallel layers of 2DEGs in a strong
perpendicular magnetic field. We assume there is a confining potential which restricts the
electrons in each layer to a region with the topology of a disc. The phase diagram of this
system was first investigated by Qiu, Joynt, and MacDonald (QJM) [9]. At low electron
densities they found a variety of solid phases, while at higher densities they found the
ground state to be an incompressible fluid described by the N -layer generalization of the
Laughlin wavefunction:
0 ({zi, }) =
Ni
N Y
Y
(zi zi )Kii
i=1 <
Nj
Ni Y
N Y
Y
(zi zj )Kij e
P
,i
|z,i |2 /4
(1)
i<j =1 =1
Here zi is the coordinate of electron in layer i, and Ni is the number of electrons

in layer i. The exponents are specified by a symmetric, N N matrix K. The diagonal
elements of this matrix determine the electron correlations within each layer and the offdiagonal elements specify the correlations between layers. For the wavefunction to describe
electrons the diagonal elements of K must be odd integers. The filling factor in layer i is
given by
i =
N
X
j =1
K 1

ij
(2)
Note that it is possible to construct more complicated multilayer states by beginning with
the states in Eq. (1) and carrying out a generalization of the single-layer HaldaneHalperin
hierarchy construction [13,14]. For example, at the first level of the hierarchy the effective
K matrix is given by
K(1) = K + AP 1 ,
(3)
where A is a diagonal N N matrix with elements 1, and P is a symmetric N N matrix

with even integer elements along the diagonal and integer elements everywhere else. The
sign of the element Ajj determines whether the excitations in layer j are quasiholes (+1)
or quasielectrons (1) and the matrix P specifies the (bosonic) multilayer state into which
these excitations condense. Using the matrix K(1) in Eq. (2) gives the filling factors at the
first level of the hierarchy.
We will restrict ourselves to the case where the matrix K is tridiagonal
Kij = mij + n(i,j 1 + i,j +1 ),
(4)
716
where m and n are non-negative integers. Since we are interested in the limit of large
N , we will impose periodic boundary conditions along the direction perpendicular to the
layers by the identification N + 1 1. One of the difficulties encountered when working
without periodic boundary conditions is discussed briefly below. The standard convention
is to refer to the state whose K matrix is of the form (4) as the nmn state. Using Eqs. (2)
and (4), the filling factor per layer in the nmn state is
=
1
.
m + 2n
(5)
From this expression we see that there are multiple states with the same filling factor per
layer. For example, the states 050 and 131 both have filling factor = 1/5 per layer.
QJM found that as the interlayer separation d was decreased there is a phase transition
between the 050 state in which there are no interlayer correlations and the 131 state in
which electrons in neighboring layers are correlated.
Note that Eq. (5) applies to the case N or to the case of finite N with periodic
boundary conditions. For a physically realizable multilayer system, i.e., one with N
finite but without periodic boundary conditions, one finds that the filling factor j is
not independent of the layer index j . If one solves Eq. (2) for finite N without periodic
boundary conditions, using the K matrix given in Eq. (4) one finds

1
y j + y N+1j
1
,
(6)
j =
m + 2n
1 + y N+1
where
m
y = +
2n
m
2n
2
1.
(7)
From this form we find that for large N the filling factor near the middle of the multilayer
stack is given approximately by Eq. (5), but exhibits oscillations about this value as
one approaches the end layers. Such non-uniformities in the electron density would cost
electrostatic energy, and could be mitigated by the formation of quasihole excitations near
the outermost layers. Therefore, one would expect that the true ground state for a finite
multilayer with interlayer correlations would be determined by balancing the creation
energy of quasiholes against the electrostatic energy of the density oscillations. It is
interesting to note that to construct a state in a finite stack with uniform filling by carrying
out the hierarchy construction in only the outermost layers one must go to an infinite level
in the hierarchy. Henceforth we will avoid these complications by working with periodic
boundary conditions, arguing that in the limit of large N they can safely be ignored.
2.1. Quasihole excitations
In the last section we learned that for a range of electron densities the ground state of the
multilayer system is an incompressible state described by the wavefunction (1). It is natural
to ask about the properties of the excitations of this state. By analogy with the single-layer
717
quantum Hall effect we write the wavefunction for a single quasihole at point in layer j
as:
(j, ) ({zi }) =
Nj
Y
(zj )0 ({zi }).
(8)
=1
If one interprets |(j, ) |2 eV ({z}) as the Boltzmann factor for a classical 2D generalized
Coulomb plasma at inverse temperature with an impurity at , the perfect screening
conditions give:
(j )
Kik Qk = ij ,
(9)
where Qk = (K 1 )kj is the local deviation of the charge in layer k from its ground state
value, due to the quasihole in layer j [9]. The total charge of the quasihole is
X (j ) X

1
,
(10)
Qk =
K 1 j k = j =
Q=
m + 2n
(j )
where we have used Eq. (2). Thus, the relation between the total quasihole charge and the
filling factor is the same as in the single-layer quantum Hall effect. However, if there are
interlayer correlations, i.e., n 6= 0, the quasihole charge is spread over many layers. Indeed,
in the limit N the individual charges are irrational:
!|kj |
1
m2 4n2 m
(j )
.
(11)
Qk =
2n
m2 4n2
The same Coulomb plasma analogy, combined with the mean field approximation,
allows us to find the spatial distribution of the extra charge excited by the hole. The Debye
(j, )
(z) acting on an electron at the point z
screening equation for an average potential Vi
in layer i can be written as
X
(j, )
= 4ij (z ) 2 + 4
Kik nk ,
(12)
2 Vi
k
where the single-particle average charge density

X
1
Kij
eVi ,
ni = 2
(13)
is chosen so that at Vi = 0 the unperturbed density would be restored. Clearly, the screening
in each successive layer is limited by the Debye screening length which is on the order of
the magnetic length (l = 1 in chosen units). Therefore, as the amount of charge induced
by the hole is reduced with the separation along the stack, this charge also spreads over
a wider and wider area, further reducing the charge density. The shape of the induced
charge distribution can be found by linearizing the density (13) in Eq. (12) and solving the
resulting set of linear partial differential equations by Fourier transformation. In particular,
for the r.m.s. spread of the distribution created in layer j by the quasihole in layer i we
obtain
718
2
r|ij
|
Z2
= 2I2 (|j i|)/I1 (|j i|),
where Ip (j )
dq
0
cos(qj )
p ,
Kq
(14)
and Kq = m + 2n cos qis the Fourier transform of Eq. (4). Specifically, for the 131 state
we find hrj2 i = 4 (3 + 5 |j |)/5. If we were to color in the perturbed charge density, the
hole would produce a characteristic hourglass shape.
To compute the statistics of these excitations we follow the method of Arovas, Schrieffer,
and Wilczek [15] and consider a state with two quasiholes, one at point in layer j and
one at point in layer k:
(j, ;k,) =
Nj
Y
(zj )
=1
Nk
Y
(zk )0 .
(15)
=1
The Berry phase for moving the quasihole at (k, ) around a closed loop of radius R
containing the quasihole at (j, ) is
I
B = i d h(j, ;k,) | (j, ;k,) i
I
=i
k
X
1
h(j, ;k,) |
(2)(z zk )|(j, ;k,)i.
z
d 2z
(16)
=1
The expectation value appearing in this equation is just the electron density at the point z
in layer k for the two-quasihole state. We can write this as
h(j, ;k,) |
Nk
X
(k,)
(2) (z zk )|(j, ;k,) i = k(0) (z) + k
(j, )
(z) + k
(z),
(17)
=1
(k,)
where k(0) (z) is the density in the ground state, k (z) is the change in density due to
(j, )
the quasihole at (k, ) and k (z) is the change in density due to the quasihole at (j, ).
If we substitute Eq. (17) into Eq. (16), the k(0) term gives the AharonovBohm phase, the
(k,)
term gives zero by symmetry, and hence
k
Z
I
1
(j, )
(j )
(stat)
= 2Qk ,
= i d 2 z k (z) d
(18)
B
z
(j )
where once again Qk is the charge deviation in layer k due to a quasihole in layer j , which
we found above to be equal to (K 1 )kj , see Eq. (9). The statistical angle for interchanging
two quasiholes in the same layer is
1 (stat)
= K 1 jj =
.
1 = B
2
2
m 4n2
(19)
We find the statistical angle of the quasiholes is an irrational multiple of in the infinite
layer system.
719
3. Edge theory
In this section we consider the edge theory of the multilayer nmn state in the presence
of interlayer single-electron tunneling. The low-energy effective Hamiltonian of the edge
theory in the absence of tunneling is [16]:
ZL/2
H0 =
dx
L/2
1
Vij : x ui x uj : ,
4
(20)
where V is a symmetric, positive definite, N N matrix which depends on the interactions

and confining potentials at the edge, and we take the x-axis along the edges of length
L. The N chiral bosons appearing in this Hamiltonian obey the equal-time commutation
relations

(21)
ui (x), uj (x 0 ) = iKij sgn(x x 0 ).
The electron charge density and the electron creation operator in layer i are given by

1
K 1 ij x uj (x),
i (x) eiui (x) .
(22)
i (x) =
2
Note that because of the factor of K 1 in the expression for the density operator (22)
there is a non-trivial relation between the matrix V appearing in the Hamiltonian and the
matrix determining the densitydensity couplings, which we will call U . If we rewrite the
Hamiltonian (20) as
ZL/2
H0 =
dxUij : i (x)j (x) : ,
(23)
L/2
which serves as our definition of U , we find by equating these two forms that
1
(24)
V = K 1 U K 1 .
We assume that the system is translationally invariant along the direction perpendicular
to the layers, which we shall take to be the z-axis. We shall often refer to this as the
vertical direction. Specifically, we assume the matrices Kij and Uij depend only on the
difference |i j |. The tridiagonal form of K given above in Eq. (4) certainly obeys this
constraint, provided we recall that we are assuming periodic boundary conditions along z.
From Eq. (24) we see that if both K and U are translationally invariant, so is V . Hence
to diagonalize K and V we perform a discrete Fourier transformation in the z direction,
defining:
N
1 X iqj
e
uj (x),
uq (x) =
N j =1
(25)
where
q {2n/N | n = 0, 1, . . . , N 1},
and we identify q and q + 2k for any integer k.
(26)
720
Using transformation (25), the commutation relations (21) become

uq (x), uq 0 (x 0 ) = iq+q 0 ,0 Kq sgn(x x 0 ),
(27)
where
Kq m + 2n cos q,
(28)
are the eigenvalues of the matrix (4). We will largely restrict our analysis to those states for
which all the edge modes move in the same direction, i.e., the maximally chiral case. This
requires that K be positive definite. From Eqs. (26) and (28) this implies n < m/2.
Finally, we rescale the fields, defining
1
uq (x),
q (x) = p
Kq
which have conventionally normalized commutation relations

q (x), q 0 (x 0 ) = iq+q 0 ,0 sgn(x x 0 ).
(29)
(30)
Using the transformations (25) and (29) to express the Hamiltonian (20) in terms of the
fields q we find
ZL/2
H0 =
L/2
dx
1
vq : x q x q : .
4
(31)
The velocities vq are given by vq = Vq Kq where Vq are the eigenvalues of the V matrix.
We now specialize to the case where the densitydensity coupling matrix is of the form
Uij = vij + g(i,j 1 + i,j +1 ),
(32)
where v is a velocity determined by the confining potential at the edge and g parameterizes the nearest-neighbor densitydensity interactions between the layers. Using the
relation between U and V (24) we find that the eigenmode velocities appearing in the
Hamiltonian (31) are
v + 2g cos q
.
(33)
vq =
m + 2n cos q
Next we consider adding interlayer single-electron tunneling to the multilayer edge
theory. The electron annihilation operator in layer i is given in Eq. (22) in terms of the
original bosonic field ui (x). Using the transformations (25) and (29) the tunneling operator
between layers j and j + 1 can be written
j (x)j+1 (x) + h.c. eijq q (x) + h.c.,
(34)
where is the tunneling amplitude and we have defined the coefficients

p
1
(35)
j q eiqj 1 eiq Kq .
N
As defined in Eq. (22) the electron operators in different layers do not obey proper
anticommutation relations. In Appendix C we demonstrate that this can be remedied
without significantly altering the form of the tunneling operator (34).
721
Fig. 1. The (mn)-plane. States to the right of the solid line are maximally chiral. The dashed line
separates states with irrelevant tunneling (to the right) from states with relevant tunneling (to the
left). The shaded region contains all maximally-chiral states with non-irrelevant tunneling: 010, 020,
and 131 (dark circles). The states 121 and 242 which lie on the boundary of the maximally-chiral
region are also shown (open circles).
The lowest-order perturbative RG flow of the tunneling amplitude , assumed to be

uniform along the edge, is controlled by the scaling dimension, , of the tunneling operator
(34):
d
= (2 ),
d`
(36)
where the short-distance cutoff increases as ` increases. Since the fields q obey
canonically normalized commutation relations, (30), we can read off the scaling dimension
of the tunneling operator from Eqs. (28), (34), and (35):
=
1X
j q j q = m n,
2 q
(37)
where there is no sum on j . Using Eqs. (36) and (37) we see that tunneling is relevant for
m < n + 2, irrelevant for m > n + 2, and marginal for m = n + 2.
If we combine this result with the condition of maximum chirality, n < m/2, we obtain
the diagram in Fig. 1. The only maximally-chiral state for which the tunneling is relevant
is 010, i.e., uncorrelated = 1 layers. There are two maximally-chiral states for which
tunneling is marginal: 131 and the bosonic state 020. For all other maximally-chiral
multilayer states interlayer electron tunneling is irrelevant.
There are two bosonic states, 121 with relevant tunneling and 242 with marginal
tunneling, which lie on the boundary of the maximally-chiral region. The corresponding
K matrices have zero eigenvalues in the limit of an infinite number of layers. The
complication this zero eigenvalue adds is the possibility of soft modes which are not
strictly confined to the edge [17]. The 121 state is the first of a sequence of pyramid
states: 121, 12321, 1234321, . . ., where in an obvious extension of the nmn notation the
state onmno has next-nearest-neighbor correlations specified by the exponent o, in addition
to nearest-neighbor (n) and intralayer correlations (m), and similarly for the ponmnop
722
states, etc. 1 All of the pyramid states lie on the boundary of the maximally chiral region
(i.e., their K matrices are positive semi-definite) and have relevant tunneling. Were it not
for the complication from the soft modes, their edge theories would be exactly soluble by
fermionization as discussed in Section 6.
The full Hamiltonian of the multilayer edge theory with tunneling can be written using
Eqs. (31), (34), and (37):
ZL/2
H=
L/2

X

1
ijq q (x)
vq : x q x q : +
dx
+ h.c. ,
e
4
(2a)
(38)
where a is a short distance cutoff. In the following sections we will be concerned with the
z-axis conductivity of this theory, so we will need the form of the current operator along
z. If j(2D) (x, t) is the two-dimensional charge density operator in layer j , the continuity
equation can be written

1
(2D)
z
j (x, t) = Jjz1,j (x, t) Jj,j
(x, t) x Jjx (x, t),
(39)
+1
t
d
z
is the current density
where Jjx is the current density along the edge of layer j , Jj,l
flowing from layer j to layer l, and d is the layer separation. Introducing the discrete
Fourier transforms
XZ
(2D)
(2D)
(k, q, t) = d
(40)
dx eiqj eikx j (x, t),
with an analogous definition for J x (k, q, t), and

XZ
J z (k, q, t) = d
dx eiq(j 1/2)eikx Jjz1,j (x, t),
(41)
where the allowed values of the dimensionless z-momentum q are given in Eq. (26), the
continuity equation (39) becomes
2i
(2D)
(k, q, t) = sin(q/2)J z (k, q, t) ikJ x (k, q, t).

t
d
(42)
(2D)
Note j (x, t) (1/d)j (x, t), where j is given in Eq. (22). Therefore, at k = 0 we
can write Eq. (42) as

d
H, (2D)(k = 0, q, t) .
(43)
J z (k = 0, q, t) =
2 sin(q/2)
Evaluating the commutator with the help of Eqs. (21), (22), and (34), we find
ZL/2

id X
dx eijq 0 q 0 (x,t ) h.c. .
I (t) J (k = 0, q = 0, t) =
(2a)
z
(44)
j L/2
This expression will be used to calculate the z-axis conductivity of the multilayer system.
1 In their Monte Carlo calculations QJM found that for some values of the layer separation d the ground state
had a non-zero value for the next-nearest-neighbor correlation exponent [9].
723
4. Marginal tunneling cases

We know there are only two maximally-chiral states with marginal tunneling: 131 and
the bosonic state 020. In this section we begin by showing that in the absence of disorder
these states exhibit insulating behavior in the vertical direction, i.e., zz () is identically
zero for all frequencies at T = 0. We will then find that adding disorder to the edge theory
increases the vertical conductivity, and, therefore, the surface behaves like a chiral semimetal. We conclude this section by considering how the transport is modified by nearestneighbor densitydensity interactions.
4.1. Clean limit
Our discussion of the system in the absence of disorder will proceed in two steps.
First, we derive a sum rule which relates the frequency integral of the conductivity to
the expectation value of the tunneling term in the Hamiltonian. Next we prove that there
exists a finite range of values for the tunneling amplitude , and the interaction strength g,
for which the ground state in the presence of tunneling and interactions is identical to
the ground state in the absence of tunneling and interactions. We then combine these two
results to arrive at our conclusions about the zero temperature conductivity. We emphasize
that this argument is non-perturbative in and g. This is an important point since we do not
have a quadratic form of the Hamiltonian for the multilayer state with marginal tunneling.
The conductivity sum rule is derived by considering the double commutator of the
Hamiltonian with the charge density operator. Specifically, we calculate:
i
h

(45)
H, (2D) (k = 0, q, t) , (2D) (k = 0, q, t) ,
where the Hamiltonian is given in Eq. (38), and from Eqs. (22), (25), (29), (40), and
j(2D) (x, t) = (1/d)j (x, t) we have
(2D)
1
(k = 0, q, t) =
N
2
ZL/2
L/2
1
dx p x q (x, t).
Kq
(46)
Using the commutation relations for the fields q (x) (30) it is a straightforward exercise to
evaluate the commutator (45) with the help of Eqs. (38) and (46). One finds:
i
h

(47)
H, (2D) (k = 0, q, t) , (2D) (k = 0, q, t) = 4 sin2 (q/2) H,
where H is the tunneling part of the Hamiltonian.
We next relate the expectation value of this double commutator to the integrated
conductivity. The continuity equation (39) introduced in the previous section can be written
Z
Z
0
d
2
ei(t t ) z
dt 0
J (k = 0, q, t 0 ).
(48)
(2D) (k = 0, q, t) = sin(q/2)
d
2
Suppressing the argument k = 0, using [H, ] = i/t, and taking the expectation value
of Eq. (48) we find
724
iE
(2D)
(q, t) , j (q, t)
Z
Z

d
4
dt eit J z (q, 0), J z (q, t) .
= 2 sin2 (q/2)
2
d
Dh
(2D)
H, j
(49)
If we take the expectation value of Eq. (47), and use it to eliminate the l.h.s. of Eq. (49) we
find in the limit q 0:
Z
Z

4
d
(50)
dt eit I z (0), I z (t) ,
hH i = 2
2
d
where we have used I z (t) = J z (q = 0, t). The Kubo formula relates the commutator
appearing in this expression to the z-axis conductivity:
Z

1 1
(51)
dt eit I z (t), I z (0) .
<e zz () =
2 NLd
Thus we arrive at the final form of the sum rule
Z
d
hH i.
d <e zz () =
NL
(52)
This is an exact relation valid at any temperature. It applies to all multilayer states, even in
the presence of disorder, regardless of the relevancy of the tunneling term.
Next we turn our attention to proving the stability of the ground state in the presence
of interlayer tunneling and interactions. We will describe in detail the case of the bosonic
020 state. A proof along the same lines can be constructed for the 131 state. We begin by
writing the Hamiltonian of the 020 edge theory in terms of the original radius R = 1 Bose
fields uj (x):
ZL/2
H020 =
dx
X v
g
: (x uj )2 : +
x uj x uj +1
16
8
j
L/2

i(uj uj+1 )
+
h.c.
.
e
(2a)2
(53)
Instead of performing the discrete

Fourier transformation along z (25)
we simply rescale
the fields, defining uj (x) = 2 j (x), where j (x) are radius R = 1/ 2 chiral bosons.
We can then define a set of N su(2)
b 1 KacMoody (KM) currents:
Jjz (x) =
1
x j (x),
2 2
Jj (x) = Jjx iJj =

y
1 i 2 j (x)
e
,
2a
(54)
in terms of which the Hamiltonian (53) can be written

ZL/2
H020 =
dx
L/2
2
v
: Jj (x) : + gJjz (x)Jjz+1 (x)
6
+

,

Jj (x)Jj++1 (x) + Jj+1 (x)Jj+ (x)
(55)
725
where we have employed the identity

ZL/2
ZL/2
2
dx : J (x) : =
z
L/2
L/2
2
1
dx : J(x) : .
3
(56)
Consider the theory without interactions, g = 0, and without tunneling = 0. The zeroenergy ground state of this Hamiltonian, |0i, is formed by taking the tensor product of the
highest-weight state of the spin-0 irreducible representation of each su(2)
b 1 algebra. Thus,
the ground state satisfies
a
|0i = 0,
Jj,n
for j,
a = x, y, z,
n > 0,
(57)
where the Fourier components of the currents are defined by

ZL/2
a
Jj,n
dxJja (x) e2inx/L,
n Z,
(58)
L/2
and obey the algebra

n
a
b
c
.
(59)
= ij ab n+m,0 + iij abc Jj,n+m
Ji,n , Jj,m
2
Using Eqs. (55) and (58) we can write the Hamiltonian as

v a a
g z z
+
+
: Jj,n Jj,n : +
Jj,n Jj +1,n +
Jj,n Jj +1,n + Jj+1,n Jj,n
.(60)
H020 =
6L
L
L
For later reference we also record the expression for the current operator in terms of the
KM generators
id +
+
Jj,n Jj +1,n Jj+1,n Jj,n
.
(61)
Iz =
L
From Eqs. (57) and (60) we see that not only is |0i the zero-energy ground state of the
Hamiltonian with g = = 0, it is a zero-energy eigenstate of the Hamiltonian for all g
and :
H020 |0i = 0.
(62)
Of course, this does not mean that |0i is the ground state of the Hamiltonian with
non-zero g and . To establish this we now demonstrate that for a range of g and the
Hamiltonian is positive semi-definite. For > 0, g > 0, we rewrite Eq. (55) as

ZL/2
H020 =
dx
L/2
2
g
: Jjz + Jjz+1 :
2

i
h
: Jj + Jj+1 Jj+ + Jj++1 : + J + J
2

2
v
g : Jjz (x) :
+
2

+
+
:Jj (x)Jj (x): + :Jj (x)Jj (x): .

+
(63)
726
The two terms in the first line are clearly non-negatively defined, while the terms in the
second line can be shown to be non-negatively defined for v > 2g + 8/ using Eq. (56)
and the basic identity
(Jj )2 = Jjz
2

1 +
J J + Jj Jj+ .
2 j j
(64)
An analogous construction can be done for < 0 or g < 0, and we conclude that the
Hamiltonian (55) is positive semi-definite provided
2|g| + 8||/ < v.
(65)
This is also the sufficient condition for the stability of the original ground state, as follows
from our previous result that the Hamiltonian annihilates the original ground state |0i.
Note that the normal ordering does not present a complication because it amounts to
subtracting off the unit operator times the expectation value of the Hamiltonian in the
state |0i, which is zero. Therefore, since the Hamiltonian is positive semi-definite and we
know the state |0i is a zero-energy eigenstate, it follows that |0i is the ground state of the
Hamiltonian provided Eq. (65) is satisfied.
We can now combine the sum rule with our knowledge of the exact ground state to say
something about the conductivity. At zero temperature the expectation value on the r.h.s.
of the sum rule (52) is a ground state expectation value. Since we know the ground state
|0i is unchanged by tunneling and interactions and h0|H1 |0i = 0 we can conclude
Z
(66)
d <e zz () = 0.
The real part of zz () is dissipative and hence it cannot be negative. Therefore, from
Eq. (66) we have
zz () = 0.
(67)
We reiterate that this is an exact statement for the clean 020 multilayer at T = 0 provided
g and satisfy Eq. (65). As we mentioned above, a similar result can be shown to hold for
the 131 state, the other maximally-chiral state with marginal tunneling. The condition for
the stability of the ground state of the 131 edge theory is also given by Eq. (65). In the next
section we will explore what happens when we add disorder to these systems.
4.2. Disordered case
In the previous section we found that the clean multilayer with marginal tunneling
exhibits insulating behavior. In this section we study the marginal tunneling case in the
presence of disorder. We perform a perturbative calculation of the vertical conductivity via
the Kubo formula, finding that disorder increases the conductivity.
In Section 3 we discussed the general edge theory of clean multilayer systems. Our first
task is to add disorder to this model. We will then use the Kubo formula to find the z-axis
conductivity.
727
The disorder we consider is a random scalar potential Vj (x) in each layer j , which
couples to the Bose fields via the charge density
ZL/2
HV =
ZL/2
dxVj (x)j (x) =
L/2
dx
L/2
1
Vj (x)Kj k x uk (x).
2
(68)
We assume Vj (x) is a Gaussian random variable, uncorrelated between different layers,

but for now we will not specify how it is correlated within the same layer.
Using the Fourier transformation (25) and rescaling (29) of the Bose fields we can write
Eq. (68) as
ZL/2
HV =
dx
L/2
1 1
p Vq x q (x),
2 Kq
(69)
where
1 X iqj
e
Vj (x).
Vq (x) =
N j
(70)
Recalling the form of the clean multilayer edge theory from Section 3, our full Hamiltonian
is

ZL/2
H =
dx
X

1
vq : x q x q : +
eijq q (x) + h.c.
2
4
(2a)
j
L/2

1 1
p Vq x q (x) .
+
2 Kq
(71)
We move the disorder term into the phase of the tunneling amplitude by shifting the bosons:
1
7 q + p
q
vq Kq
Zx
dyVq (y).
(72)
If we define the phases

Zx

1 X iqj
iq 1
e
1e
dyVq (y),
j (x)
vq
N q
(73)
then with Eq. (72) the Hamiltonian (71) is

ZL/2
H=
L/2
dx

1
X ij (x) ijq q (x)
,
vq : x q x q : +
e
+
h.c.
e
4
(2a)2
(74)
and the current operator (44) becomes

id X
I =
(2a)2
ZL/2
j L/2

dx eij (x) eijq q (x) h.c. .
(75)
728
We are now in a position to calculate the z-axis conductivity. We will first evaluate the
Matsubara two-point function of the current operator (75). Since Eq. (74) is a theory of N
interacting chiral bosons, the best we can do is a perturbative calculation in the tunneling
amplitude . Although we will work perturbatively in , we will be treating the disorder
exactly. We thus expect the O(2 ) result to be reliable, since the disorder essentially
randomizes the phase of the electrons between tunneling events. More formally, if the
tunneling amplitude is a coordinate-dependent, delta-correlated random variable, then the
RG flow (36) is modified to
d
= (3 2 ),
(76)
d`
where is the variance of the tunneling. Thus for = 2 we would conclude that the
tunneling is irrelevant. The combination eij (x) that appears in the Hamiltonian (74)
is in general not delta-correlated, as we will discuss below, but since the case of xindependent tunneling is marginal we expect any departure from uniformity makes the
tunneling irrelevant and therefore justifies a perturbative expansion in .
Using the expression for the current operator (75) we evaluate the Matsubara two-point
function

C( ) = T I z ( )I z (0)
Z
X

d 2 2
0
T eij (x) : eijq q (,x) : eij (x) : eijq q (,x) :
dx
dx
=
4
L
j,k
0
0
0
0
eik (x ) : eikq q (0,x ) : eik (x ) : eikq q (0,x ) : ,
(77)
where we have normal-ordered the vertex operators. Note that because the current operator
(75) has a factor of out front, there is an explicit factor of 2 in Eq. (77), and, therefore,
to find this function to order 2 we can evaluate the expectation value using the = 0
Hamiltonian. One finds that only the cross terms in the product of the current operators
give non-vanishing contributions, and only when j = k. The terms have the form

0
T : eijq q (,x) : : eijq q (0,x ) :
=
Lq
[(2ivq ) sinh( v
(x x 0 + ivq + ia sgn ))]q
q
2
2
(78)
where q2 j q j q , with no sum on j . This is a very complicated function; a great

simplification occurs if the eigenmode velocities, vq , are identical for all q. From the
expression for vq (33) we see that this occurs when g = nv/m. For the two multilayer states
with marginal tunneling, 020 and 131, this condition gives g = 0 and g = v/3, respectively.
We will refer to the case where vq is independent of q as the solvable point. For 020 it
corresponds to no densitydensity couplings between neighboring layers while for 131 it
occurs at a non-zero value of the densitydensity interaction strength. Note that for more
general forms of the matrices K and U , i.e., not tridiagonal, the solvable point corresponds
to U K.
729
Working at the solvable point, writing vq v0 , and using our knowledge of the scaling
dimension of the tunneling operator (37),
X
X
q2 =
j q j q = 2 = 4,
(79)
q
we find upon disorder averaging Eq. (77)

Z
Wj (x, x 0 )
2 d 2 2 X
0
C( ) =
,
dx
dx
(2v0 )4
[sinh( v
(x x 0 + iv0 + ia sgn ))]4
0
(80)
where we have defined the function

0
Wj x, x 0 ei[j (x)j (x )] .
(81)
If we take the correlation function of the scalar potential to be

Vj (x)Vk (x 0 ) = j k Z(x x 0 ),
(82)
then using the definition of the phase j (x) (73) we find that the function Wj
written
"
#
Zx 0
Zx 0
1
Wj (x, x 0 ) = exp 2 dy1 dy2 Z(y1 y2 ) .
v0
x
(x, x 0 )
can be
(83)
Note that Wj (x, x 0 ) W (x x 0 ) is independent of j and depends only on the magnitude of

the relative coordinate |x x 0 |. Therefore, the sum over j in Eq. (80) just gives a factor of
the number of layers, N , and we can replace the integration over x and x 0 by an integration
over the average coordinate, which gives a factor of L, and an integration over the relative
coordinate y = x x 0 . If we define the Fourier transform of the function W (x):
Z
e (k) dx eikx W (x),
(84)
W
then we can write Eq. (80) as
Z
Z
eiky
2d 2 NL2
dk e
W
(k)
dy
C( ) =
.
4
(2v0 )
2
[sinh( v0 (y + iv0 + ia sgn ))]4
(85)
The factor of L in this expression will cancel against the factor of 1/L in the Kubo formula
(51), and in anticipation of this we have extended the range of the y integral to include the
entire real axis since we are interested in the result in the limit where the system size L
. The integration over y can now be performed by the method of residues. The integrand
has fourth-order poles on the imaginary y-axis at the points yn = i[v0 (n ) a sgn ]
for integer n, and is exponentially small in the upper (lower) half plane for k > 0 (k < 0).
One finds
Z
eiky
dy
[sinh( v0 (y + iv0 + ia sgn ))]4

v0 k
v0 k 3
ev0 k
4
+
.
(86)
=
3
ev0 k 1
We now Fourier transform Eq. (85) with respect to the imaginary time ,
730
Z
C(n ) =
d ein C( )
0
d 2 2 NL
48 4 v04
e (p/v0 )
dp W
p3 + 4 2 p/ 2
,
in + p
(87)
where we have changed the integration variable to p v0 k. Analytically continuing

(in 7 + i0+ ) we find
Z
C() = i
dt eit h[I z (t), I z (0)]i

0
d 2 2 NL
=
48 4 v04
e (p/v0 )
dp W
p3 + 4 2 p/ 2
.
+ p + i0+
(88)
Using this result in the Kubo formula (51) with the help of the identity
1
1
(89)
= P i(x),
+
x i0
x
e (k) is real since W (x) is an even function of x, we arrive finally
and the observation that W
at
<e zz () =
2 d
48 3 v04

e (/v0 ).
2 + 4 2 T 2 W
(90)
This is our central result. It is the O(2 ) term in the vertical conductivity at the solvable
point for both the 020 and 131 states at a finite temperature T , including potential disorder
e (k). The imaginary part of the conductivity can of
which enters through the function W
course be found from Eq. (90) by the usual dispersion relations [18].
Let us first consider the limit of a clean system, Vj (x) 0. From Eqs. (82) and (83), we
e (k) = 2(k). Hence from Eq. (90) we find
see that in this case W (x) = 1 and therefore W
<e zz () =
2 d
6v03
T 2 ().
(91)
This is consistent with our result in Subsection 4.1 that the z-axis conductivity in the clean
system vanishes at T = 0. An interesting aspect of this result is that it is proportional to
(), i.e., the conductivity vanishes at all 6= 0, and the DC conductivity is infinite. One
question to ask is whether these features are a consequence of the fact that Eq. (91) is only
the first non-vanishing term in a perturbative expansion in .
Although we cannot definitively answer the question about the finiteness of the DC
conductivity beyond O(2 ), we can prove that at higher orders in the conductivity
cannot vanish at all 6= 0. First we note that in the clean system the current operator
commutes with the Hamiltonian in the absence of tunneling: [I z , H0 ] = 0. For example,
for the 020 state this can be seen from Eqs. (60) and (61), using the commutation relations
for the KacMoody currents given in Eq. (59). Since the current operator commutes with
H0 , the expectation value of the currentcurrent commutator evaluated using the = 0
731
Hamiltonian vanishes, and, therefore, by the Kubo formula (51) the vertical conductivity
must vanish for all non-zero frequencies. This explains why the conductivity is zero except
at = 0 in Eq. (91). By the same reasoning we see that if the current operator commutes
with the full Hamiltonian, [I z , H] = 0, then zz () would be zero for all 6= 0 to all orders
in . The converse of this is also true, i.e., if zz () is zero for all 6= 0 then [I z , H] = 0.
This is proven in Appendix D. For the multilayer edge theory with marginal tunneling:
[I z , H] 6= 0. For the 020 case this can be established using Eqs. (60) and (61). Therefore,
since the current operator does not commute with the full Hamiltonian, we can conclude
that zz () cannot be zero for all 6= 0 for the clean 020 or 131 multilayers. Hence, we
would expect that the () factor in Eq. (91) would be broadened by the higher-order
corrections in . Recalling that is dimensionless and the conductivity should not depend
on the sign of , one plausible scenario is:
<e zz () =
2 d
6v03
T (/T )
7
2 d
6v03
2
.
(/T )2 + O(4 )
(92)
This conjectured form has several interesting features. First, we find that the DC
conductivity goes to zero with temperature as T . We expect this to be a generic feature
of the exact result in for the clean system, provided the DC conductivity is finite. This
differs from the disordered system (90) where we find the DC conductivity vanishes as
T 2 . In addition, note that the 0 limit of Eq. (92) is independent of the tunneling
amplitude .
We now return to the result for zz () in the presence of disorder (90). If the disorder is
delta-correlated, the function Z(x) defined in Eq. (82) is
Z(x) = (x),
(93)
where is the variance of the disorder. For this case W (x) = exp(|x|/v02 ), which has
e (k) = (2/v 2 )/(k 2 + (/v 2 )2 ), and hence (90) yields
a Fourier transform W
0
0
<e zz () =
2 + 4 2 T 2
.
24 3 v04 2 + (/v0 )2
2 d
(94)
Note that at zero temperature the conductivity vanishes in the DC limit. This conclusion
e (0) is finite. We
is independent of the details of the disorder and holds as long as W
also see that the conductivity approaches a constant as . This second feature is
a consequence of taking the disorder to be delta-correlated. We see that zz () goes to a
e (k), which is in turn caused by the
constant at large because of the slow fall off of W
non-analyticity of W (x) at x = 0. We shall now investigate how Eq. (94) is modified if the
disorder has a finite correlation length.
As a first step, we determine the asymptotic form of the disorder-averaged phase, W (x),
at small and large x when the scalar potential disorder has a finite correlation length. We
return to the expression for the correlation function of the random potential (82) and write
Z(x) 2 A(x/`0 ),
(95)
732
where is a parameter with the dimensions of energy that sets the strength of the disorder,
`0 is the correlation length of the disorder, and A(z) is a dimensionless function which we
take to have the properties: A(0) = 1, A(z) = A(z), and
Z
dz A(z) = 1.
(96)
Now consider the disorder-averaged phase, W (x), given in Eq. (83). For |x| `0 , i.e.,
for distances small compared with the correlation length of the random potential, we can
replace Z(y1 y2 ) by its value at y1 y2 = 0 and obtain:

`0 z 2
2
, for |z| 1.
(97)
W (`0 z) e(`0z/v0 ) 1
v0
In the opposite limit |x| `0 we change integration variables in Eq. (83) to yc = (y1 +
y2 )/2 and yr = y1 y2 :
"
1
W (x) = exp 2
v0
Z0
!#
Z|x|
dyr (|x| + yr )Z(yr ) + dyr (|x| yr )Z(yr ) .
|x|
(98)
Since |x|/`0 1 we can extend the integration over yr to infinity with small error. Using
the symmetry of A and defining
Z
=
dz zA(z),
(99)
we then find

`0 2
(|z| ) ,
W (`0 z) exp 2
v0
for |z| 1.
(100)
We see that at short distances W (x) is parabolic (97) while at long distances it decays
exponentially (100). The presence of a finite correlation length for the random potential
does not change the long distance behavior of W (x), but it does round out the nonanalyticity at x = 0 present in the case of delta-correlated disorder. Generally, this is
sufficient to make the function rapidly decaying as .
Having found the generic behavior of the function W (x), we now choose a specific form:
W (x) =
1
,
cosh( x)
(101)
characterized by the parameter which has the dimensions of an energy. This choice of
W (x) has several appealing features. First, it has the correct asymptotic forms at large and
small values of x. Indeed, if we match the asymptotics of Eq. (101) to Eqs. (97) and (100)
we find
1
(102)
= ,
`0 = .
v0
2
733
Fig. 2. The real part of the vertical conductivity for the 020 or 131 multilayer with disorder (104).
The horizontal axis is frequency measured in units of the parameter v0 . The temperatures of the
curves, starting with the uppermost, are: T /v0 = 0.25, 0.2, 0.15, 0.1, 0.01.
We see that for the choice of W (x) in Eq. (101), the parameter is proportional to the
strength of the disorder and inversely proportional to the correlation length. Hence this
form allows us to explore the effect of a finite correlation length in the region `0 1.
A second advantage of Eq. (101) is that its Fourier transform can be evaluated in closed
form. A straightforward computation gives
Z
ikx
/
e (k) = dx e
=
.
(103)
W
cosh( x) cosh(k/2 )
Using this result in the general form above for the conductivity (90) we find
<e zz () =
2 d
2 + 4 2 T 2
.
4
2
48 v0 cosh(/2 v0 )
(104)
Comparing Eq. (104) with the expression for the conductivity in the case of deltacorrelated disorder (94) we see that the finite correlation length gives a conductivity which
exponentially decays to zero as rather than approaching a non-zero value. This
result for zz (104) is shown in Fig. 2 at various temperatures. For large temperatures the
conductivity is peaked around = 0, while at smaller temperatures the maximum occurs
at a finite frequency.
4.3. Effect of interactions
In the preceding section we calculated the order 2 term in the z-axis conductivity of
the multilayer states with marginal tunneling, 020 and 131, at the solvable point where the
eigenmode velocities vq are independent of q. In this section we discuss how the zerotemperature result is modified away from the solvable point.
Since we will only consider T = 0 in this section we shall work directly in real time t.
The real-time zero-temperature analog of the expression for the Matsubara two-point
function of the current operator (80) is
734
C(t) =
2id 22 NL
(2)4
Z
dy W (y)
(y vq t + ia sgn t)q
(105)
where from the definitions (73) and (81) we see that the expression for W (x) in Eq. (83) is
changed to
#
"
Zx
Zx
2 X sin2 (q/2)
dy1 dy2 Z(y1 y2 ) .
(106)
W (x) = exp
N q
vq2
0
The corresponding retarded correlation function is

Z
2id 22 NL
C R (t) =
(t)
dy W (y)
(2)4

Y
Y
1
1
q2
q2
q (y vq t + ia)
q (y vq t ia)
(107)
We would now like to perform the y-integration. From Eq. (35) we see that the exponents
appearing in Eq. (107) are

2
(108)
3 cos q 2 cos2 q ,
q2 =
N
and, therefore, as a function of complex y, the first integrand in Eq. (107) has N branch
point singularities at the points vq t ia and similarly for the second integrand at vq t + ia.
This is a complicated singularity structure, but observe that in the first term these branch
points are all below the real axis, while in the second term they are all above the real axis.
P
Hence, since q q2 is an integer, we can certainly choose branch cuts that lie entirely
on one side of the real axis for each term. When we considered the solvable point in the
previous section we were able to evaluate the conductivity for any disorder W (y). We
shall not attempt the same feat away from the solvable point. The evaluation of the yintegral in Eq. (107) is greatly simplified if W (y) is a meromorphic function of y. One
such function was discussed in the previous section (101), and for the remainder of this
section we specialize to this form.
Using W (y) = 1/ cosh(y) in Eq. (107), we can close the first term in the upper-half
plane and the second term in the lower-half plane, avoiding all the complicated branch
point singularities, and picking up the residues of the poles of W (y). We find
C R (t) = i (t)h[I z (t), I z (0)]i
X
Y
1
2id 22 NL
3
r
(t)
(1)
.
=
q2
(2)3
q [vq t i(r + 1/2)]
r=
Therefore, from the Kubo formula (51) we arrive at
Z
Y
2 d 3 X
1
r
it
zz
(1)
.
dt
e
<e () =
3
q2
(2) r=
q [vq t i(r + 1/2)]
(109)
(110)
This is certainly far from a closed form result, but it is a convenient form for extracting
the high and low frequency asymptotics of the conductivity. At large (positive) the
735
asymptotic form is determined by the closest singularity to the real axis in the upper half
of the complex t plane. From Eq. (110) this clearly occurs a finite distance away from the
real-t axis and we find

zz
,
(111)
<e () exp
2 vmax
where vmax is the maximum of vq over q. For the 020 state we see from the expression
for vq (33) that at the solvable point (g = 0) v0 = v/2 while away from the solvable point
(g > 0) vmax = v/2 + g. Comparing Eq. (111) with the asymptotic form of Eq. (104):

<e zz () exp
,
(112)
2 v0
we find that one effect of interactions at T = 0 is to soften the exponential decay of the
conductivity at large frequencies.
Returning to Eq. (110), we now consider the behavior at small frequencies. If we
assume > 0 we can close the t integral in the upper-half plane because the integrand
is exponentially small there. Hence all the terms with r < 0 give no contribution. For
each r > 0 there remains N branch point singularities in the upper-half plane enclosed by
the contour. Since we are interested in small we stretch the integration contour into a
large circle so that everywhere along the contour |t| is very large. We can then expand the
integrand in powers of 1/t and integrate term-by-term:
Z
Y
1
dt eit
q2
q [vq t i(r + 1/2)]
!

I
2
Y 1
1
1
i(r + 1/2) X q
it
= dt e
+O 2
1+
q2 ( t)4
t
v
t
q
q vq
q
!

X q2
Y 1 3
2
(r + 1/2)
+O .
1
(113)
=
3 q q2 4
4
vq
q
vq
If we retain only the lowest term in in this expansion, we note that upon substituting this
into Eq. (110) we would encounter the divergent sum
(1)r .
(114)
r=0
By considering the solvable point, where vq = v0 independent of q, it is clear that the

proper regularization of this sum is
X
r=0
(1)r = lim
X
r=0
r
1
1
= ,
0 1 + e
2
= lim
and thus from Eqs. (110), (113), and (115) we have

!
2d
Y 1 2
0
.
<e zz ()
48 2 q q2
vq
(115)
(116)
736
Comparing this with the small frequency asymptotics of Eq. (104) at T = 0:

0
<e zz ()
2 d 1 2
,
48 2 v04
(117)
we see that the interactions do not modify the result that the conductivity goes to zero in
the DC limit at zero temperature. While the power of is the same in Eqs. (116) and (117),
the interactions do modify the prefactor.
To summarize, we have considered how interactions away from the solvable point
modify the vertical conductivity at T = 0. It should be noted that in arriving at Eq. (110) the
interactions were treated exactly. We find that at large frequencies the interactions soften
the decay of the conductivity, but it remains exponential, while at small frequencies the
interactions do not change the frequency exponent of the conductivity. We remark that the
difficulty encountered when attempting to apply the method presented in this section to
investigate the effect of interactions at a finite temperature is that in the finite-T version of
Eq. (107) there are complicated branch point singularities on both sides of the real-y axis,
see Eq. (78). Nevertheless, we believe that even at a finite temperature the asymptotics of
zz () are not modified by interactions.
5. Irrelevant tunneling
In the previous section we have been concerned exclusively with the case of marginal
tunneling, namely the 020 and 131 multilayer states. The calculation presented in
Subsection 4.2 can readily be extended to the more general case of the nmn state.
Essentially the only change in the computation is that the scaling dimension of the
tunneling operator, = m n, differs from its marginal value of 2. Therefore, the poles
in the complex y-plane for the integrand in Eq. (86) are now of order 2(m n). One finds
that for the solvable point g = nv/m:
zz
() =
<e nmn
2 d
(2)2 1 v02
1
e (/v0 ) Y
2

W
+ (2kT )2 .
(2 1)!
(118)
k=1
For the case of relevant tunneling, = m n = 1, the product over k is absent in

e (/v0 ) = (2)/(2 + (/v0 )2 ) appropriate for delta-correlated
Eq. (118), and if we use W
disorder we reproduce the behavior of the chiral metal.
In the remaining cases, > 2, we see at T = 0 the conductivity vanishes as 0 as
2( 1) , and at = 0 it vanishes with temperature as T 2( 1) . In the absence of interlayer
correlations, i.e., for the 0m0 state, this means the DC conductivity obeys zz T 2m2 .
This disagrees with the result of Balents and Fisher, who find by a scaling argument that
zz T 2m3 [4]. While it is unclear whether Balents and Fisher have in mind the clean
system or the disordered system, we believe the reconciliation of this discrepancy is that, as
we found for the marginal case, the temperature scaling of the DC conductivity is different
in the clean and disordered systems. If we consider the clean limit of Eq. (118) by writing
737
e (/v0 ) = (v0 /T )(/T ), and then conjecture that the interactions broaden the delta
W
function, we find:
zz
()
<e nmn
Y
1
2

g
2 d
+ (2kT )2 ,
2
2
T (/T ) + O(g )
(119)
k=1
where g is the dimensionless interaction strength. The broadening of the delta function by
interactions is suggested by the fact that the nearest-neighbor densitydensity interaction
Hamiltonian does not commute with the current operator I z . For the case of the 0m0 state
the DC conductivity of Eq. (119) goes to zero as T 2m3 , in agreement with Balents and
Fisher.
Finally, returning to the disordered case (118), we see that for the 050 state the DC
conductivity scales as T 8 , while for 131 it scales as T 2 (90). Recall from Eq. (5) that both
of these states occur at the same electron density, = 1/5 per layer. In their numerical
work, QJM found a phase transition between these states as the layer separation d was
varied. The significant difference in the temperature scaling of the DC vertical conductivity
in these states suggests a way to experimentally detect this phase transition.
6. Fermionization
In certain special cases the multilayer edge theory can be solved exactly. The case of
N = 2 layers is discussed extensively elsewhere [11,12], and examples of exact solutions
for N = 4 layers are given in Appendix B. A potentially useful first step to solving the
edge theory for arbitrary N is to find a fermionic representation. In this section we discuss
the fermionization of the edge theory for an arbitrary number of layers N . In some cases
the procedure involves the introduction of auxiliary degrees of freedom as discussed in
Ref. [11].
We begin with the observation that the scaling dimension (37) of the tunneling operator
is always an integer. This underlies the fact that the chiral Hamiltonian (38) can be
fermionized exactly, with the tunneling term expressed as a product of Fermi operators
whose number equals twice the scaling dimension.
The usual mapping between chiral radius-one bosons and fermions requires independent
bosonic fields. Even though we are interested in correlated states with non-trivial
commutators (21), these can be readily diagonalized by a linear transformation. For
example, for the 131 state we can write
uj = j 1/2 + j + j +1/2 ,
j = 1, 2, . . . , N,
(120)
where the 2N bosons j with integer and half-integer indices have the usual commutation
relations [i (x), j (x 0 )] = i ij sgn(x x 0 ). Clearly, Eq. (120) is not a canonical
transformation because the number of degrees of freedom has been doubled. This formal
difficulty can be circumvented if we introduce an independent set of auxiliary bosons with
half-integer indices u i+1/2 , i = 1, 2, . . . , N , with identical commutation relations,

ui (x), u j +1/2 (x 0 ) = 0. (121)
u i+1/2 (x), u j +1/2 (x 0 ) = i Kij sgn(x x 0 ),
738
Then, we can further write for the 131 state

u j +1/2 = j + j +1/2 j +1 ,
(122)
and the transformation ui , u i+1/2 7 j becomes canonical.

At the end of a calculation the auxiliary degrees of freedom need to be projected out.
Such a projection has been explicitly carried out by us in Ref. [11] for the simpler case of
correlated quantum Hall bilayers. However, since the auxiliary degrees of freedom are
independent of the physical ones, all quantum-mechanical averages involving physical
quantities will automatically be correct, independent of the Hamiltonian chosen for the
additional bosons u i+1/2 . In the following we select this Hamiltonian in the form (20), i.e.,
identical to that for the physical bosonic fields sans the tunneling term. In this case, the
transformations (120) and (122) give

ZL/2
H131 =
dx
N

1 X e
Vij :x i x j : + :x i+1/2 x j +1/2 :
4
i,j =1
L/2
X
j

i(j1/2 +j j+1 j+3/2 )

:e
: + h.c. ,
(2a)2
eij Kil Vlj . Now

where the modified interaction matrix is given by the matrix product V
we can use the standard fermionization prescription:
1
eij+ ,
(123)
j + =
2a
where = 0, 1/2. In particular, in the fermionic representation the original electron annihilation operator is written as a product of three fermion operators, j j 1/2 j j +1/2 ,
while the complete fermionized Hamiltonian for the surface of the 131 system becomes
ZL/2
H131 =
L/2
X
X
ejj : ix j + : + 1
eij : i+ : : j + :
dx
V
V
j +
i+
j +
2
j,
X
,i6=j

j1/2 j j +1 j +3/2 + h.c. .
(124)
A curious feature of this transformation is that the introduced auxiliary degrees of freedom
can be thought of as located at the layers sandwiched between the physical layers this
is the reason for assigning half-integer indices to them.
Similar fermion representations also exist for other correlated multilayer states. For the
bosonic state 020, the interlayer correlations are absent, and the commutation relations can
be diagonalized by the transformation
uj = j + j ,
(125)
where the auxiliary degrees of freedom labeled by the pseudospin index are located in the
same layers. The fundamental boson operator can be written as j = j j , and the full
Hamiltonian is
ZL/2
H020 =
L/2
739
X
X
ejj : ix j : + 1
eij : i : : j :
V
V
dx
j
i
j
2
j,
X
,i6=j

j j j +1 j +1 + h.c. ,
= , .
(126)
Despite the similarity with its counterpart for the 131 state [Eq. (124)], this Hamiltonian
involves subtly different correlations.
In the case of 020 state, one can also avoid doubling the number of degrees of freedom by
using the alternative fermionization introduced in Ref. [11]. Specifically, the neighboring
layers are combined into pairs,
u2n+1 = 2n+1 + 2n+2 ,
u2n+2 = 2n+1 2n+2 ,
(127)
and the tunneling operators are fermionized alternatingly as anomalous two- or fourfermion combinations,
1
1
ei(u1 u2 ) = 2 ix 2 ,
ei(u2 u3 ) = 1 2 3 4 , . . . ,
(128)
(2a)2
(2a)2
where j is given in Eq. (123). In cases where the number of layers N is small, e.g.,
N = 2, 3, 4, alternative fermionizations exists for the 020 and 131 states, which in some
cases allow exact solutions, see Ref. [11] and Appendix B.
The fermionization is also very simple for the pyramid states. As an example consider
the 12321 state where the commutation relations are satisfied if we write
uj = j 1 + j + j +1 ,
(129)
and the corresponding tunneling operator is bilinear,

1 i(uj uj+1 )
e
= j1 j +1 .
(130)
2a
The edge theory with tunneling is quadratic and, therefore, exactly soluble, except for the
complication of the soft modes mentioned in Section 3.
7. Summary
We have investigated the fractional quantum Hall effect in infinite layer systems. In
the bulk we find quasiparticles with several unusual (irrational) features and at the edge
we find surface phases which can be considered chiral semi-metals when the interlayer
tunneling is not relevant. In addition we have presented fermionizations of the edge theory
which may prove useful in finding exact solutions.
Acknowledgements
We would like to acknowledge support by an NSF Graduate Research Fellowship (JDN),
DOE Grant DE-FG02-90ER40542 (LPP) as well as NSF grant No. DMR-99-78074,
740
USIsrael BSF grant No. 9600294, and fellowships from the A. P. Sloan Foundation and
the David and Lucille Packard Foundation (SLS). LPP and SLS thank the Aspen Center
for Physics for its hospitality during the completion of part of this work.
Appendix A. Model problem with marginal tunneling

Here we consider a model problem with marginal interlayer tunneling. The Hamiltonian
is chosen in analogy with the Hamiltonian (74),
ZL/2
H=
L/2

1 ij (x)
dx vj i x j + e
j i x j +1 ix j j +1 + h.c. , (A.1)
2
with the marginal tunneling operator in parenthesis of a form reminiscent of the twofermion-fermionized version of the marginal tunneling operator (128). The great advantage
of the Hamiltonian (A.1) is its linearity, which allows us to calculate the conductivity for
this model exactly. In order to do this, we perform a gauge transformation j eij (x)j ,
with j +1 j j . This gives, in obvious matrix notation,
ZL/2
H=
L/2

1
b+V
b ,
dx i x + V
2
(A.2)
where (1 2 )T , the tri-diagonal matrix ll 0 vll 0 + (l,l 0 +1 + l+1,l 0 ) is

b diag(V1 , V2 , . . .) with Vj x j . The corresponding
coordinate-independent, while V
transverse current operator has the form
ZL/2
I =
z
L/2

1 b b
dx t ix + tV
+ V t ,
2
(A.3)
where tij i (i,j +1 i+1,j ).

The formal solution of this model (for a given realization of disorder) can be written
in the limit of infinite system size (L ) using the transfer matrix approach described
in Appendix A of Ref. [12]. Using the Kubo formula for the vertical conductance, we
obtain
L zz
()
Nd
ZL/2

1 X
dx(nk nk+ ) Tr e
V(y)Sk+ (y, x) , (A.4)
=
V(x)Sk (x, y)e
2N
Gzz () =
k L/2
where nk = (evk + 1)1 , the matrix in the vertex

1 b b 1/2
1/2
e
,
V(x)
t ix + tV + V t
2
(A.5)
741
and the transfer matrix

Zb
Sk (a, b) = Tx exp i

!

1
b+V
bt 1/2 .
dx k1 1/2 tV
2
(A.6)
To compute the disorder averages we assume that Vj (x) are independent delta-correlated
Gaussian random variables with zero average, Vi (x) Vj (y) = ij (x y). Then, a
calculation at a finite temperature T gives
Gzz

L 2
+ 4 2 T 2
=
24
2(q)
dq t 2 (q)
,
2 3 (q) 2 + 2 (q)
(A.7)
where (q) = v + 2 cos q, t (q) = 2 sin q, and

2 (q)
2(q) + 1 +
.
(q)
=
4
v 2 2
As in the physical marginal cases, 020 and 131, at zero temperature and in the absence
of disorder the conductivity vanishes. In the leading order in Eq. (A.7) is identical to
the result for 020 with delta-correlated disorder (94). However, in the clean limit we find
Gzz () for the model problem, which is not the correct behavior for the 020 and 131
states. The difference arises because, in contrast to 020 and 131, in the model problem the
current operator I z (A.3) commutes with the full Hamiltonian (A.2).
Appendix B. Exact solutions for N = 4

In this appendix we describe exact solutions for the spectrum of the edge theory for the
special case of N = 4 layers with periodic boundary conditions. We begin with a discussion
of the 010 state with both tunneling and interactions and then consider the 020 state. We
remark that similar solutions exist for the 010, 020, and 131 states in N = 3 layers.
B.1. 010 state

The Hamiltonian for the 010 state is

ZL/2
N=4
H010
=
L/2
dx

v
g
: (x ui )2 : +
: x ui x ui+1 : +
ei(ui ui+1 ) + h.c. ,
4
2
2a
(B.1)
where u5 u1 . A straightforward fermionization i = eiui (x) / 2a would yield a

quartic Hamiltonian because of the densitydensity interaction term. We thus perform the
orthogonal transformation ui (x) = ij j (x) where
742
1
1
1
=
2 1
1
1
1
1 1
1 1
1 1
1
1
.
1
1
(B.2)
The Hamiltonian (B.1) then reads

ZL/2
N=4
H010
dx
L/2
vi :(x i )2 : +
ei(2 3 ) + ei(3 4 )
4
2a

+ ei(2 +3 ) + ei(3 +4 ) + h.c ,
(B.3)
where v1,3 = v 2g and v2 = v4 = v. The original fields, ui , are all radius R = 1

chiral bosons. We restrict ourselves to the sector where the total topological charge of the
fields ui is constant, which corresponds to the conservation of the total fermion number
in the fermionic form of (B.1). Then the requirement that the topological charges of
the original (ui ) and transformed (i ) bosons be integral is consistent with taking the
fields
i to all have unit radius as well. We can therefore fermionize according to i (x) =
eii (x)/ 2a
and then express the fermions in terms of their Fourier components i (x) =
P ikx
e
c
/
L. In terms of these mode operators the Hamiltonian (B.3) is
ik
k
Xh
i

N=4
=
cik + c2k
c3k + c3k
c4k + c2k c3k + c3k
c4k + h.c. ,
(B.4)
vi kcik
H010
k
where the momentum k takes values in (2/L)(Z + 1/2). This form of the edge theory is
quadratic and hence exactly soluble via a Bogoliubov transformation. We find:
X
N=4
=
i (k)bik
bik ,
(B.5)
H010
k
{bik , bjk 0 }
where
dispersions
= ij kk 0 are four independent branches of fermionic oscillators with

q
3 (k) = (v g)k (gk)2 + 42 ,
q
4 (k) = (v g)k + (gk)2 + 42 .
1 (k) = (v + 2g)k,
2 (k) = vk,
(B.6)
We find that two of the branches develop curvature if and only if both g and are non-zero.
B.2. 020 state

Now consider the 020 state whose bosonic Hamiltonian is given above in Eq. (53). The
fields ui all have unit radius, but their commutators are not conventionally normalized,
see Eq. (21). From our solution above for the 010 state we know that we can perform an
orthogonal transformation (B.2)
and still have four radius one bosons i . If we then define
rescaled fields i (x) i (x)/ 2 the 020 Hamiltonian becomes
ZL/2
N=4
H020
dx
L/2
743
vi
i 2 (2 3 )
i 2 (3 4 )
+
e
e
: (x i )2 : +
8
(2a)2
+ ei
2 (2 +3 )
+ ei
2 (3 +4 )

+ h.c. ,
(B.7)
where i are radius R = 1/ 2 chiral bosons with conventionally normalized commutators,

and the velocities vi are the same as those given above in the solution of the 010 state. We
can define a quartet of su(2)
b 1 KM currents as in Eq. (54). The Hamiltonian becomes

ZL/2
N=4
H020
dx
L/2

2

vi
: Ji (x) : +4 J2x J3x + J3x J4x ,
3
(B.8)
where we have again used the identity (56). Next, we define a new set of currents Jia (x) =
b 1 algebras, where R ab az bx + ay by
R ab Jib (x), which also obey independent su(2)
ax
bz
. This rotation leaves the first term in the Hamiltonian (B.8) unchanged and in the
second term gives Jix 7 Jiz . If we express the rotated KM currents, Jia , in terms of four
new radius R = 1/ 2 chiral bosons i via Eq. (54), we find
ZL/2
N=4
H020
dx
L/2
where
1
Mij : x i x j : ,
4
v/2 + g
0
M
0
0
0
v/2
/
0
0
/
v/2 g
/
0
0
.
/
v/2
(B.9)
(B.10)
We have succeeded in finding a quadratic representation of the Hamiltonian which can be

readily diagonalized by performing an orthogonal transformation from the fields i to new
fields i . The final form of the Hamiltonian is
ZL/2
N=4
H020
=
L/2
dx
1
vi : (x i )2 : ,
4
(B.11)
where the velocities are

v1 = v/2 + g,
v2 = v/2,
q
v3 = v/2 g/2 + (g/2)2 + 2(/)2 ,
q
v4 = v/2 g/2 (g/2)2 + 2(/)2 .
(B.12)
The exact spectrum of the 020 state contains four independent free chiral bosons whose
velocities are renormalized by both interactions and tunneling.
744
Appendix C. Klein factors

In this appendix we discuss Klein factors which are needed to produce the proper
anticommutation relations between different species of fermions. One place where Klein
factors are needed is in the definition of the physical electron operators at the edges of each
layer. We modify the definition of the electron operator at the edge of layer i (22) to read
i (x) eiAij Nj eiui (x),
(C.1)
where A is an N N matrix and Nj is the topological charge of the Bose field uj (x)
defined by
1
Nj =
2
ZL/2
dx x uj (x).
(C.2)
L/2
Using the commutation relations of the chiral bosons (21) one can readily show
0
i (x)j (x 0 ) = j (x 0 )i (x)eiKij sgn(xx ) ei(Aik Kkj Ajk Kki ) ,
(C.3)
and the combination appearing in the tunneling operator (34) can be written
(x) ei(Aik Ai+1,k )Nk ei[ui (x)ui+1 (x)] ei(Ai+1,k Kk,i+1 Ai+1,k Kki ) .
i (x)i+1
(C.4)
If we write A = BK 1 , for some matrix B, then we have

i (x)j (x 0 ) = j (x 0 )i (x)eiKij ei(Bij Bji ) ,
(C.5)
(x) ei[ui (x)ui+1 (x)] ei(Bi+1,i+1 Bi+1,i ) ei(Bil Bi+1,l )(K )lk Nk .
i (x)i+1
(C.6)
From Eq. (C.5) we see that for the electron operators in different layers to anticommute
we must have (Bij Bj i ) Zeven if Kij is odd and (Bij Bj i ) Zodd if Kij is even. From
Eq. (C.6) we see that if the elements of the matrix B are integers and (Bil Bi+1,l )(K 1 )lk
is an integer for all i and k, then the tunneling operator will be the same as in the absence
of the Klein factors up to a sign which depends on the topological charge sector.
As a concrete example, consider the multilayer 131 state. If the number of layers N is
such that det(K) is odd (i.e., N mod 3 6= 2) then the choice

det(K) for j i > 2,
(C.7)
Bij =
0
otherwise,
gives {i (x), j (x 0 )} = 0 for all i, j and modifies the tunneling term by a factor of 1.
Note that in correlation functions, such as the current two-point function used to compute
the conductivity (77), the tunneling operator is always accompanied by its hermitian
1
conjugate and, therefore, factors such as ei(Bil Bi+1,l )(K )lk Nk will cancel.
In the above discussion we have constructed Klein factors out of the topological charge
operators of the Bose fields in order to give the correct anticommutation relations between
the physical electron operators without significantly modifying the tunneling Hamiltonian.
When considering the fermionization of the multilayer edge theory, as discussed in
Section 6, we have a different goal. In addition to ensuring that the electron operators
745
anticommute we want the different species in the fermionized Hamiltonian to obey proper
anticommutation relations. In this case an alternative approach to the one described above
proves advantageous. This procedure is best described by considering a specific case, for
example the 131 state. One can introduce 2N unitary operators Fi+ , where = 0, 1/2
and i = 1, . . . , N , by formally enlarging the Hilbert space. These operators commute with
the Bose fields ui and obey

, Fj+ 0 = Fi+ , Fj+ 0 = 0,
(C.8)
{Fi+ , Fj + 0 } = Fi+
for i + 6= j + 0 . The electron operators are modified according to
i (x) Fi1/2 Fi Fi+1/2 eiui (x).
(C.9)
It is a straightforward exercise to demonstrate that Eqs. (C.8) and (C.9) imply that electron
operators in different layers properly anticommute. The tunneling operator is now
j (x)j+1 (x) ei[uj (x)uj+1 (x)] Fj 1/2 Fj Fj+1 Fj+3/2 .
(C.10)
If we then follow the procedure described in Section 6 with the modification that the new
fermion species each absorb a factor of F :
1
Fj + eij+ (x) ,
(C.11)
j +
2a
we arrive at the same fermionized Hamiltonian (124), but now with the different fermion
species properly anticommuting.
Appendix D. Proof that <e zz () () [I z , H ] = 0
In this appendix we prove that the real part of the z-axis conductivity is proportional
to a delta function in frequency if and only if the current operator commutes with the full
Hamiltonian of the edge theory. The basic connection is of course the Kubo formula (51):
1
1
<e () =
2 NLd
zz

dt eit I z (t), I z (0) .
(D.1)
One of the directions is trivial: if [I z , H ] = 0, then I z (t) is independent of time, and hence
the commutator in Eq. (D.1) vanishes identically implying that for all 6= 0, <e zz () =
0 and, therefore, <e zz ().
We now prove the converse, i.e., we assume <e zz (). Inverting the Fourier
transform in Eq. (D.1) we find
I (t), I (0) = 2NLd

z
d eit <e zz ().
(D.2)
Using our assumption in this equation gives

X E
z

z
n
e
n I (t), I z (0) n = 0,
I (t), I z (0) =
n
(D.3)
746
where |ni and En are eigenstates and eigenvalues of H . Note that the sum in Eq. (D.3)
is a linear combination of terms hn|[I z (t), I z (0)]|ni with non-negative coefficients which
depend on . Since the sum must vanish for arbitrary we conclude that

z
(D.4)
n I (t), I z (0) n = 0, n.
Writing I z (t) = exp(iH t)I z (0) exp(iH t) and inserting a summation over a complete set
of states {|mi}, Eq. (D.4) is equivalent to
X
2

sin (En Em )t nI z (0)m = 0, n t.
(D.5)
m
The quantities |hn|I z (0)|mi|2 are clearly non-negative. Since the sum in Eq. (D.5) must
vanish for all n and at all times t we can conclude

z 2
nI (0)m = 0, n 6= m.
(D.6)
This equation says that the current has no off-diagonal matrix elements in the basis of
energy eigenstates, i.e., the current operator I z is diagonal in the basis of eigenstates of
H . Since this means I z and H can be simultaneously diagonalized it implies that they
commute: [I z (0), H ] = 0, completing the proof.
References
[1] H.L. Stormer, J.P. Eisenstein, A.C. Gossard, W. Wiegmann, K. Baldwin, Phys. Rev. Lett. 56
(1986) 85.
[2] S. Uji, C. Terakura, M. Takashita, T. Terashima, H. Aoki, J.S. Brooks, S. Tanaka, S. Maki,
J. Yamada, S. Nakatsuji, Phys. Rev. B 60 (1999) 1650, and references therein.
[3] J.T. Chalker, A. Dohmen, Phys. Rev. Lett. 75 (1995) 4496.
[4] L. Balents, M.P.A. Fisher, Phys. Rev. Lett. 76 (1996) 2782.
[5] J.J. Betouras, J.T. Chalker, cond-mat/0005151, and references therein.
[6] D.P. Druist, P.J. Turley, K.D. Maranowski, E.G. Gwinn, A.C. Gossard, Phys. Rev. Lett. 80
(1998) 365.
[7] L. Brey, Phys. Rev. Lett. 81 (1998) 4692.
[8] S.M. Girvin, A.H. MacDonald, in: S. DasSarma, A. Pinczuk (Eds.), Novel Quantum Liquids in
Low-Dimensional Semiconductor Structures, Wiley, New York, 1995.
[9] X. Qiu, R. Joynt, A.H. MacDonald, Phys. Rev. B 40 (1989) 11943;
X. Qiu, R. Joynt, A.H. MacDonald, Phys. Rev. B 42 (1990) 1339.
[10] J.D. Naud, L.P. Pryadko, S.L. Sondhi, Phys. Rev. Lett. 85 (2000) 5408.
[11] J.D. Naud, L.P. Pryadko, S.L. Sondhi, Nucl. Phys. B 565 (2000) 572.
[12] J.D. Naud, L.P. Pryadko, S.L. Sondhi, cond-mat/0006173, Phys. Rev. B (in press).
[13] M. Greiter, I.A. McDonald, Nucl. Phys. B 410 (1993) 521.
[14] Y.J. Chen, J. Phys. Condens. Matter 10 (1998) 2437.
[15] D. Arovas, J.R. Schrieffer, F. Wilczek, Phys. Rev. Lett. 53 (1984) 722.
[16] X.-G. Wen, A. Zee, Phys. Rev. B 46 (1992) 2290.
[17] A. Karlhede, K. Lejnell, S.L. Sondhi, Phys. Rev. B 60 (1999) 15948.
[18] G.D. Mahan, Many Particle Physics, 2nd edn., Plenum, New York, 1990.

The renormalization-group approach for Fermi

systems in the presence of singular forward
scattering
C. Castellani, S. Caprara , C. Di Castro, A. Maccarone
Dipartimento di Fisica Universit di Roma La Sapienza, and Istituto Nazionale per la Fisica della
Materia Unit di Roma 1, Piazzale Aldo Moro, 2-00185 Roma, Italy
Abstract
Within the framework of perturbative renormalization group, we analyze the problem of spinless
fermions either coupled to a transverse gauge field or interacting through a longitudinal singular
potential, in generic spatial dimensions d. The model of fermions interacting through a singular
potential due to the proximity to a phase-separation instability is also discussed. We show how to
develop a unified renormalization-group treatment making use of the so-called loop cancellation,
coming from the peculiar symmetry properties of models with forward scattering. This allows to
solve the renormalization-group equations and to compute the asymptotic properties of the singleparticle propagator, also providing exact scaling results. 2001 Elsevier Science B.V. All rights
reserved.
PACS: 71.10.Hf; 71.45.Gm; 72.10.Di
Keywords: Renormalization group; Fermions in reduced dimensions; Non-Fermi-liquid ground states
1. Introduction
In recent years the issue of the validity of the Fermi-liquid (FL) description for
strongly correlated electron systems has been extensively studied in connection with many
relevant physical problems such as high-temperature superconductivity and the fractional
quantum Hall effect (FQHE). It has been shown that a continuum system of fermions
with a short-range forward interaction in generic dimension d > 1 is a FL [1], contrary
to the one-dimensional case where the quasiparticle picture breaks down leading to a
different metallic phase called Luttinger liquid (LL) [2]. A non-Fermi-liquid (NFL) state
E-mail address: sergio.caprara@roma1.infn.it (S. Caprara).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 4 - 4
748
C. Castellani et al. / Nuclear Physics B 594 [FS] (2001) 747768
in d > 1 could instead arise in the presence of a sufficiently singular interaction. Several
realizations of this scenario have been proposed in the literature in different contexts.
The implementation of RVB ideas [3] leads to a low-energy spin-liquid model, describing
fermions (spinons) coupled with a transverse gauge field [4,5] and a very similar picture
arises in the description of the = 1/2 state in the FQHE theory [6]. These two models
are referred to as gauge-theory (GT) models. More recently it has been shown [7] that
fermions near a charge instability (a phase separation (PS) or an incommensurate chargedensity wave) scatter via a singular potential mediated by the density fluctuations. In the
case of PS this interaction can be described by a model which is strictly related to the GT
model. Finally, also long-range (LR) interactions, with a singular momentum dependence
have been studied in the last years [8,9]. 1 From a theoretical point of view the treatment
of these problems, sharing the common feature of singular forward scattering, calls for
some nonperturbative method, due to the absence of an obvious expansion parameter and
to the presence of infrared divergences that plague the second-order perturbation theory
[10]. Bosonization and Ward-identity (WI) approach have been applied both to the LR [9]
and GT [11] problems. Even though the soundness of bosonization for GT is still debated,
there is generic agreement on the outcomes of this method in the GT and LR problems.
By contrast the renormalization group (RG), which has been widely considered for the GT
problem [1216], has had only limited applications to the LR case [17,18]. Moreover, for
GT, the RG approach has led to some conflicting results in the literature and some points
have been left unclear.
In this paper we shall carry on the RG program for fermions in the presence of singular
interaction on a quite general basis, pointing out the common features of all the models
considered. We shall systematically exploit the conservation laws to identify the proper
renormalization of the coupling constants in each model and to establish that the Random
PhaseApproximation (RPA) effective interaction is robust with respect to higher order
corrections. Specifically we shall:
Reconsider RG for the GT problem, to provide a consistent scenario, correcting minor
mistakes and clarifying the meaning of the main results in the literature [1216].
Extend the above results to deal with systems in the proximity to a PS.
Consider LR forces with generic power-law singularities, for which RG has never
been successfully applied.
2. The GT and LR models: a critical comparison

We consider here two models of fermions defined in a continuum d-dimensional space:
the fermions interact with a transverse gauge field and with a longitudinal boson field
respectively. The GT interaction term reads
Z d d
d k d q d d
(k + q/2, + /2)(k q/2, /2)vF A(q, ), (1)
HIGT = g GT
(2)2d+2
1 Notice that an effective singular scattering amplitude was first proposed by Anderson [3] as due to orthogonality effects in two dimensions.
749
The fluctuations of the gauge field A can be described for both the spinwhere vF = vF k.
liquid and the FQHE case with the RPA propagator [5,6]
D GT (q, ) =
1
q 2 i ||
q
(2)
where one has = 0 in the spin-liquid case and = 1 in the (unscreened) FQHE case,
q = |q|, and the coefficients and are determined by the underlying microscopic
model and can be regarded as constants. The expression (2) is derived by selecting the
leading contributions of the currentcurrent polarization bubble 2 which are present in
the kinematic region vF q. This propagator exhibits a damped mode iq 3
which, being strongly mixed with the fermion particle, is found to be responsible for the
corrections to the FL description of the metallic state.
In the case of a longitudinal LR potential we deal with a densitydensity interaction term
of the form
Z d d 0 d
(g LR )2
d k d k d q d d 0 d 1
H4 =
2
q
(2)3d+3
+ q/2, + /2)(k q/2, /2)
(k

k0 q/2, 0 /2 k0 + q/2, 0 + /2 ,
that can be decoupled by introducing a properly normalized auxiliary scalar boson field
(q, ) via a HubbardStratonovich transformation leading to
Z d d
d k d q d d
+ q/2, + /2)(q, )(k q/2, /2).
(k
HILR = ig LR
(2)2d+2
The bare boson-field propagator, D0LR (q, ) = 1/q , has to be dressed with a density
density RPA series. Expanding the result in the limit of small momentum and energy, in
the regime vF q the propagator becomes short-ranged (contrary to the GT case), due
to the usual Debye screening, while it stays singular in the region vF q , where
D LR (q, ) =
1
2
,
2
q c2 q 2
(3)
q
with c = g LR vF Sd kFd1 /d(2)d . Here and in the following Sd = 2 d/2/ (d/2) is the
surface of the d-dimensional unit sphere. The poles of (3) signal the existence of an
undamped plasmon mode = cq 1/2 (gapless in any spatial dimension provided
< 2, as we shall assume in the following) which couples to fermions, leading to NFL
corrections for > 2d 2 [8,9]. The second-order selfenergy contributions to the fermion
propagator computed with the potentials (2) and (3) are indeed affected by infrared
divergences [5,14,19] and the problems have to be dealt with nonperturbative methods.
In this paper we cure the infrared divergences of these models by means of RG.
2 We must notice that the expression (2) takes contribution also from a diamagnetic term of the form A
2 .
For simplicity this term, which produces sub-leading contributions to the quantities involved in RG calculations,
is not explicitly included in (1).
750
In view of this, one has first to assess the validity of the RPA expressions of the boson
propagators (2) and (3) with respect to selfenergy and vertex corrections which dress the
polarization bubble. These corrections determine the renormalization constant of the boson
field that mediates the interaction among the fermions. This issue has been addressed in the
past years by many groups [12,20,21] and formalized in the so-called loop-cancellation
theorem for forward scattering processes [10,22]. According to loop cancellation, the sum
of the corrections to the RPA form of the boson propagator involving fermion loops with
more than two insertions, is less singular than one would expect on the basis of naive power
counting. However, in the presence of singular interactions, this result leaves the question
open whether or not the corrections are strong enough to dominate over the RPA result. In
Ref. [12] this point was addressed through an effective boson action derived in the spinliquid case (GT model with = 0). This approach can be easily generalized to the case
of generic . By integrating out the fermion degrees of freedom one obtains the effective
action of the boson fields in the form
Z
Z

Seff (A) = S2 A2 + 3 A3 + 4 A4 + ,
where
1
S2 A2 =
2
d d q d 1
A D Aj .
(2)d+1 i ij
The coefficient Dij1 of the quadratic term is the inverse of the RPA propagator of the
gauge field whose transverse part is given by (2). This propagator sets up the relative
scaling between frequency and momentum z = 3 . Aiming to keep the form of the
gauge propagator invariant under scaling, we are forced to choose the dimensions of
momentum and frequency to be respectively [q] = 1 and [] = 3 . The engineering
dimensions of the boson field can then be calculated, yielding [A] = (d + 5)/2.
From the dimension of the field the scaling of the n-th coefficient can be obtained by
considering the n 1 momentum and frequency integrations, yielding n s xn , with xn =
d + 3 n(d + 1)/2. In d > 1 the terms with n > 4 are irrelevant in the RG sense for
s and can be regarded as zero. On the contrary, the constant part of 3 (k, ), which
scales as s (3d2 )/2 , it is apparently relevant in any dimension d < 3 2 . However, the
explicit calculation in the static limit (with the frequency going to zero first) shows that it
vanishes [12], keeping the theory massless, as it is indeed requiredR by gauge invariance.
The first non-vanishing cubic contribution is in fact proportional to q m 3(1) A3 with m >
2 [23], so that the coefficient 3(1) is irrelevant in any dimension greater than one [12]. The
GT problem is thus reduced to deal with a Gaussian fixed-point effective action, which is
the counterpart of the RPA form derived with the diagrammatic technique. The irrelevance
of higher-order terms in the effective boson action is mirrored by the absence of radiative
corrections to the RPA propagator (2). An explicit two-loop calculation [21] shows indeed
that the divergent parts of the vertex and selfenergy corrections cancel out each other in the
currentcurrent polarization bubble.
We can carry out the same analysis for the LR problem, writing an effective Lagrangian
for the field

Seff () = S2 2 +
751
Z
3 3 +
4 4 + ,
where the quadratic part takes the form

Z d
1
d q d
c2 q 2
(q,
)q
1
(q, ).
S2 2 =
2
(2)d+1
2
The dynamical exponent is here z = 1 /2. Assuming [q] = 1 and [] = 1 /2, the
scaling dimension of the field , as deduced from the kinetic term S2 , is [] = (2d +
2 + )/4. Given this scaling of the boson field the coefficient of the n-th term scales as
n s xn , with xn = d + 1 + [(3n 2) 2n(d + 1)]/4. This scaling form suggests that
the n-th operator is relevant for > (n,
d) (2n 4)(d + 1)/(3n 2). Since (n,
d) is
smaller than two (except for large d) one would conclude that there exist a sensible value
of for which the coefficients with n > 2 must be taken into account in the expansion.
However we have found that the explicit form of the coefficients heals such a difficulty
in this case also. In Appendix A we calculate the asymptotic behavior of the coefficient
n which is given by the n-leg fermion loop. This calculation obviously connects our
effective-action approach to the loop-cancellation theorem. We show in fact, by means of
a proper gauge transformation, that the first non-vanishing contribution to the coefficient
n for n > 2 vanishes in the dynamical limit at least as (q/)n , being thus irrelevant for
any physical value of and in any spatial dimension d. From a physical point of view this
result amounts to say that the plasmon mode does not acquire any screening from radiative
corrections beyond RPA.
We emphasize that the stability of the RPA form of the effective potential supports the
idea that the building blocks of RG calculations are the fermion propagator, the RPA boson
propagator and the bosonfermion vertex [18]. We also notice that the absence of radiative
corrections to the RPA boson propagator implies that the boson fields do not acquire an
anomalous dimension.
3. RG constants and Ward identities

Systems with forward scattering have been extensively studied in the literature by
exploiting their symmetry properties through the WI formalism [10]. In the onedimensional case (TomonagaLuttinger model), for example, the separate conservation
of the number of right-moving and left-moving particles (which is peculiar to the onedimensional case) leads to a WI that expresses the full interaction density vertex in terms
of the full fermion propagator [2,24,25]. This WI can be used to show in a nonperturbative
way that the -function of the model is exactly zero [25]. The same WI allows also to
exactly solve the model [2,24], recovering the same solution also found by bosonization
[2]. This so-called WI approach has been extended to deal with short-range interactions
in higher-dimensional systems, by assuming an approximate (yet controlled) conservation
of the number of particles at any point of the Fermi surface. Within this assumption the
current vertex is simply proportional to the density vertex and the long-distance form of
the propagator can be evaluated similarly to the one-dimensional case [1].
752
We want now to show how conservation laws can be used to establish the relations
among the various singularities of the theory, thus reducing the number of independent
renormalization constants also in the present problem. As it was shown in the previous
section, the boson propagator does not undergo any renormalization. It is therefore only
necessary to control the behavior of the interaction vertex and, if possible, relate it to the
singularities of the fermion propagator.
3.1. Total number conservation
The first conservation law to be exploited is the conservation of the total number of
particles which is generically valid. To write the corresponding WI we define the three-leg
irreducible interaction vertex

+ q/2, + /2)(k q/2, /2)
,
(k, q; , ) = j (q, )(k
amp
where the subscript amp indicates that the external fermion legs have been amputated, the
index = 0, . . . , d runs over time and space components, and j = (, j) is the densitycurrent operator. 3 The WI then reads
q (k, q; , ) 0 (k, q; , )
= G1 (k q/2, /2) G1 (k + q/2, + /2),
(4)
which holds for the bare and for the fully renormalized vertices and Green functions and
allows to relate the renormalization constants of the vertices with those of the propagator.
Since frequency and momentum are inequivalent variables, we have to introduce two
different renormalization constants for the frequency and the momentum part of the
propagator G1 (k, ) = Z 1 (ZZvF )1 k , where k (k 2 kF2 )/2m is the free-particle
dispersion that can be approximated near the Fermi surface as k vF (k kF ). The
renormalization constants Z and ZvF are defined as

1
,
(5)
Z= 1

1 1
,
(6)
ZZvF = 1
vF k
and have to be related with the singular parts of the current and density vertices in the
left-hand side of (4). More explicitly, one has to extract the coefficient of the momentum
q in the right-hand side of (4) and compare it with the static limit (frequency going to
zero first) of the three-leg vector (current) vertex in the left-hand side. In that way, defining
1
vF
the renormalization of the current vertex (k, q; , ) in the static limit, from Zstat
(k, q 0; , = 0) one finds
Zstat = ZZvF .
(7)
3 More specifically, in the notation of Ref. [10], j(q) = P v p
p F pq/2 p+q/2 is the quasiparticle current
operator.
753
Analogously, to extract the divergences of the frequency part of the propagator, one defines
1
0 (k, q = 0; , 0)
the dynamical limit of the three-leg scalar (density) vertex Zdyn
and finds
Zdyn = Z.
(8)
Eqs. (7), (8), describe the renormalization of the vertices, respectively in the static and
dynamic limit. We shall see in the following that Eq. (7) has to be used in the GT problem,
while the form (8) pertains to the LR case.
3.2. Conservation at each point of the Fermi surface
The second conservation law to be considered is the conservation of the number of
particles at each point of the Fermi surface, which is often used in forward-scattering
problems. This conservation is obviously only approximate in d > 1 and holds if forward
scattering dominates at low energies, as it is the case in 1 < d < 2 [1]. From a diagrammatic
point of view this conservation law amounts to say that in a quadri-current insertion
(k, q; , ) the transferred momentum q is very small, and the current vF = vF k is
conserved all along the fermion line. This implies that for low momenta the current vertex
is proportional to the density vertex
(k, q; , ) = vF 0 (k, q; , ).
(9)
Assuming the above approximate equation and combining it with (4) we can write down
an equation expressing the density vertex as a function of the single-particle propagator
only
0 (k, q; , ) =
G1 ( + /2, k + q/2) G1 ( /2, k q/2)

.
vF q
(10)
By inserting (10) in the expression for the polarization bubble, it can be shown [10]
that the latter does not get dressed, since the vertex corrections exactly cancel the
selfenergy corrections. In our model this amounts to say that the boson propagator does
not renormalize beyond RPA. Thus, in the language of RG, the validity of (9) is equivalent
to fix the renormalization of the boson fields to unity, ZA = Z = 1. We notice that this is
the same result obtained from loop cancellation.
The approximate nature of Eq. (9) is mirrored by the presence of corrections to the
loop-cancellation result. In the discussion of the previous section and in Appendix A these
corrections have been explicitly shown to be irrelevant at low energies in the cases of
interest. We recall that in one-dimensional systems with linear dispersion, where Eq. (9)
is exact, the loop-cancellation theorem is exactly valid. In higher dimensions it is apparent
that the importance of the corrections to the proportionality relation between the density
and the current vertices, depends on the problem we are interested in. For example in
the GT and LR models these corrections generate, in the effective fermion-integrated
Lagrangian, terms which are irrelevant in the infrared regime. By contrast, in the model
of fermions interacting through an effective singular potential in the proximity to a PS
instability, which we shall examine in the following, corrections to loop cancellation
754
produce a relevant operator which must be finely tuned to zero to reach the second-order
transition point (see below). This last example shows that the absence of renormalization
of the boson propagator cannot be inferred only from the approximate relation (9), but
corrections have to be controlled. Similarly, the exploitation of Eq. (9) to get the asymptotic
properties of the fermion propagator could be questionable in the presence of long-range
forces. In these problems the RG treatment is more appropriate and under control.
3.3. Selfenergy and renormalization constants
In this subsection we use the perturbative analysis to get insight into the independent
renormalization constants of the models discussed above. The renormalization constants Z
and ZvF (which also control, via WI, the singular behavior of the vertices) are calculated
through a perturbative expansion of the selfenergy both in the GT and in the LR problem.
These constants are defined in terms of at a proper normalization point in momentum
space. We shall carry out the actual calculations in the next section and present here only
a comparative preliminary analysis. It has been shown [5], via explicit calculations to one
loop (extended to the two-loop level by Altshuler et al. [14]), that in the GT case the
selfenergy depends only weakly on the momentum

1

.
(11)
vF k
The relation (11) is valid in any kinematic regime. Moreover, it is found that the derivative
of the selfenergy with respect to the momentum is not singular in the infrared limit.
On the contrary, in the LR case we shall show that the one-loop selfenergy contribution
yields

1
,
(12)
vF k
i.e., the derivative of the selfenergy with respect to the momentum and to the energy are
of comparable magnitude and have the same singular behavior in the infrared dynamical
regime.
In the GT problem the relation (11) implies that the singular part of the selfenergy is a
function of the frequency only, so that the momentum part of the fermion propagator does
not undergo any renormalization and we can write
ZZvF = 1,
(13)
i.e., the velocity renormalizes with the inverse of the wave-function renormalization [16].
On the other hand, in the LR problem the selfenergy is a singular function of the momentum
and of the frequency as well (Eq. (12)), and the renormalization of the momentum part of
the fermion propagator is equal to the renormalization of the frequency part,
ZZvF = Z,
(14)
i.e., the velocity does not acquire divergent corrections, as in the one-dimensional Luttinger
model.
755
4. The structure of RG equations

In this section we show how the different nature of the coupling (vector and scalar) and
the different relevant kinematic regimes in the two problems we are studying, strongly
affect the structures of the RG equations, leading to peculiar fixed-point actions, whose
properties we are going to determine. We first notice that both the boson propagators (2)
and (3) exhibit a polar structure. In particular the gauge propagator (2) is peaked around
q 3 , while the longitudinal propagator (3) is peaked around q 1/2 . In view of
this, one would naively assume that when the boson propagator is inserted in a perturbative
expression, it would force the integrated momenta and energies to obey one of the above
vs q relations, thereby restricting the relevant kinematic region for the GT and the LR
problems to be in the static and in the dynamic limit respectively. Actually, a more careful
analysis is needed to support this assumption, since the interaction vertex appearing in the
Feynman diagrams depends, in general, at least on d + 1 independent variables (i.e., the
energy and the d components of the momentum) that are not fixed only by the pole in
the boson propagator [26], and it is not known a priori which relation between and q
discriminates between the static and the dynamic limits. This analysis is crucial, since it
dictates the relevant WI that controls the structure of the RG equations.
4.1. The GT problem
We examine first the GT problem [12,15,16] and introduce the renormalization constants
for the Fermi field and for the Fermi velocity (hereafter we denote the bare quantities with
a subscript 0, while the renormalized ones carry no index)
vF0 = ZvF vF .
(15)
0 = Z ,
Our purpose is to identify the proper kinematic regime in which the interaction vertex takes
the main contribution. This will, in turn, define the renormalization of the fermionboson
coupling constant via the WI (7) and (8) for the current vertex and charge vertex, implied by
the total charge conservation. In order to do this we first consider the one-loop perturbative
result for the selfenergy [5,6,14,19], which is discussed in detail in Appendix B. The real
part reads
d
0 (k, ) || 3 sgn()
for || |k kF |3 ,
0 (k, ) |k kF |d3+
for || |k kF |3 .
(16)
We then use (16) in the generic form of the full selfenergy with dressed vertices, picking
up only the most singular contribution to the propagators
Z d
1
1
d q d
(k, ) = ig GT
d
2
(2)d+1 A| + | 3
i ||
sgn( + ) k+q q
q
viF Tij (q)j (k + q, q; + , ),
where the complex constant A may be deduced from (B.1), and (B.2), reported in
Appendix B, and Tij (q) (ij qi qj /q 2 ) selects the transverse component (with respect
756

j
to q), so that viF Tij (q)vF = vF2 sin2 , where is the angle between vF and q. From this
expression, using the relation k+q vF qk , it can be easily shown, following a remark of
Altshuler and coworkers [14], that the functional form of the fermion propagator fixes the
relation between the energy and the parallel component of the boson momentum (with
d
respect to k) qk as qk 3 , while the pole of the boson Green function imposes the
scaling q
1
3
. In view of this, the relevant limit of the vertex for the GT model is
(k, qk 3 , q 3 ; , ). We can thus define the renormalization constant of the

vertex ZGT as
d
1
vF0 = (k, qk 3 , q 3 ; , ).
ZGT
(17)
In Appendix C we show that the vertex is a finite constant in this limit. We consider
specifically the case d = 2, = 0 (spin-liquid case), where an explicit expression of
the corrections to the forward-scattering form (10) have been obtained [14], and give
arguments for this result to be valid for generic dimension d and . This implies that
ZGT does not contain any divergence and can be considered as a finite number. Indeed,
ZGT behaves as Zstat , which is also finite according to Eqs. (7) and (13). Therefore, we
can identify ZGT = Zstat , up to a constant factor, in agreement with the considerations at
the beginning of this section. Since the interaction vertex is proportional to g0GT vF0 , the
bare coupling constant g0GT can now be expressed in function of the renormalized coupling
constant g GT as
g0GT = g GT
Zstat
,
ZZvF
(18)
where a factor Z
is introduced to subtract the divergence of the two fermion fields (each one
accounts for the subtraction of the divergent part
carrying a factor Z ), and the factor Zv1
F
of the Fermi velocity. In (18) we use the above result that in the GT problem the vertex
is picked up in the limit of Eq. (17), thus renormalizing with ZGT = Zstat . Finally, the
expression (18) can be simplified and its physical content can be made apparent by using
the WI (7), thus obtaining
g0GT = g GT ,
(19)
i.e., the coupling constant g0GT does not acquire any correction [16].
The explicit study of the RG properties of the model proceeds now by finding the actual
skeleton structure of the perturbative expansion. It is seen, from perturbation theory, that
the effective coupling constant u consists of two vertices g GT vF and one factor vF1 coming
from the fermion propagator, so that it is given by (g GT )2 vF . To evaluate the Fermi-field
renormalization constant in terms of the effective coupling constant we consider again
Eq. (16). We see from this expression that the derivative of the selfenergy with respect
to the momentum is nonsingular in both kinematic regimes. This is obvious in the region
|| |k kF |3 . In the region || |k kF |3 this can be seen by writing the
derivative with respect to the momentum as

0
|k kF |d1
.
k
|k kF |3
757
In d > 1 this quantity vanishes in the infrared due to the kinematic constraint || |k
kF |3 . The divergences of the fermion propagator are thus only found in the derivative
with respect to the frequency. The one-loop renormalization constant of the fermion field
is obtained 4 from Eqs. (5) and (16)
Z 1 () = 1 + u0
3
3 d
|| 3 ,
3 d
(20)
where we have defined an effective bare coupling constant u0 (g0GT )2 vF0 for the
perturbative expansion, as previously discussed. All numerical factors included in the
definition of u and u0 can be deduced from (B.2), reported in Appendix B, and we shall not
make them explicit in the forthcoming discussion, for the sake of simplicity. We emphasize
that, thanks to the absence of renormalization of the gauge-field propagator, the factors
and , which also appear in (B.2), do not acquire any singular correction. Moreover, due
to the relative scaling between momentum and energy, their physical dimension is clearly
zero, so that we can regard them as constants.
In order to write down the RG equations, one has now to determine the bare dimensions
of the variables and parameters of the Lagrangian. The scaling dimension of the effective
d
coupling constant u0 is, from (20), [u0 ] = 3
3 []. Recalling that the dimension of the
frequency has been fixed to [] = 3 , we have [u0 ] = 3 d in units of momentum.
We want to stress here that, due to the transverse nature of the gauge interaction,
D GT (q, ) gives the dominant contribution to the selfenergy when the momentum q is
perpendicular to the incoming fermion velocity. It would thus be sufficient to impose the
3
scaling relation q between the frequency and the transverse component of the
momentum, leaving the way open for an anisotropic scaling prescription [14,15]. So one
could in principle write q sq , s 3 , qk s xk qk , thus defining an independent
scaling dimension for the parallel component of the momentum, qk . In the framework of
this anisotropic scaling the dimensions of the fields, as inferred from the Hamiltonian, are
[] = 3 (d 1 + xk)/2, [A] = 2 (d + xk )/2. These quantities can then be used
to evaluate the scaling dimension of u0 = g02 vF0 , obtaining once again [u0 ] = 3 d .
We notice that [u0 ] is remarkably independent of xk and the use of anisotropic scaling turns
out to be an unnecessary complication within our RG treatment.
From the above discussion we can introduce the renormalized dimensionless coupling
constant u, via u0 = ZvF uxu that, through the relation (13), gives
u0 = Z 1 uxu ,
(21)
where we have introduced the infrared momentum scale to account for the bare scaling
dimension of the running coupling constant, xu = 3 d . Eq. (21) shows that:
For d > dc 3 , one has xu < 0 and a stable infrared free fixed point u = 0 exists.
Z is non critical and remains finite.
4 We adopt the definition 0 u sgn()|| /(1 ) in the NFL regime < 1. This definition determines
0
the effective coupling constant u0 appearing in the expressions (B.2) and (27) for the one-loop selfenergy
in the GT and LR cases respectively. Then Z 1 = 1 + u0 ||1 /(1 ). An alternative definition 0
u0 sgn()|| /(1 ), to be associated with Z 1 = 1 0 /, yields an effective coupling constant u0 which
differs from the former by a trivial factor .
758
For d < dc a nontrivial fixed point exists, provided Z xu .

Eq. (21) allows, indeed, to determine the asymptotic form of the renormalization constant
Z when a nontrivial fixed point exists. Defining t = log , we write a differential
equation for the renormalization constant Z(t) at a given scale. By differentiating (21)
with respect to t one has
du
+ uxu .
= Z 1 Zu
dt
At the nontrivial fixed point (u = 0), the renormalization constant must then satisfy the
relation
u
u=u = xu .
Z 1 Z|
(22)
We would like to stress here that the result (22) is a consequence of the WI leading to (19)
and it is valid to all orders in perturbation theory. The absence of renormalization of the
coupling constant g GT guarantees indeed that only one renormalization is required for the
effective coupling constant u in the form given by (21).
The integration of (22) yields that, asymptotically near the fixed point, the renormalization constant Z behaves exactly as Z xu , which when translated into an dependence
leads to the exact relation
Z() ||xu /x = ||
3d
3
(23)
since x = 3 . We notice that a nontrivial fixed point can be calculated explicitly from
second-order perturbation theory. Indeed, inserting the one-loop result for Z given by (20)
into (21) one can write down the one-loop -function and u = (u) u(xu x u), which
gives the nontrivial fixed point u = xu /x . According to the previous analysis, higher
order corrections, if present, would only modify the value of u , leaving the exponent in
(23) unchanged.
At the critical dimension d = 3 the fixed-point value u goes to zero as 1/ log .
Correspondingly one has Z 1/ log ||.
We finally comment on some previous results found in the literature on the same
problem. The result (23) is in agreement with the derivation by Altshuler et al. [14], that
higher-order corrections in the 1/N expansion at d = 2 do not modify the dominant powerlaw behavior obtained from second-order perturbation theory.
In Ref. [12], which analyzes the spin-liquid problem, the calculation of the asymptotic
behavior of the one-particle selfenergy is carried out overlooking the flux of the Fermi
velocity which is taken as a constant. They thus obtain an incorrect non-universal powerlaw behavior at the critical dimension d = 3, while our approach yields the correct result
of a marginal FL behavior (logarithmic corrections with respect to the FL functional form).
In Ref. [16] the RG equations are correctly derived, in the spin-liquid case, by exploiting
the WI at one-loop order. However, the same physical dimension is assigned to momentum
and frequency. Given this choice, the problem would arise that the momentum and the
energy part of the gauge boson propagator would flow with two different power laws under
scaling, thus modifying the form of the gauge boson propagator itself at each step of the RG
process. This is in contrast with our present choice which preserve, as it should, the form
759
of the gauge boson propagator under the action of the RG. For this reason the exponent
found in the work of Ref. [16] differs from our Eq. (23) by a factor 3.
A last comment concerns Ref. [15], which analyzes the GT in d = 2 for 1. In
that paper the electronboson vertices are improperly assumed in the dynamical regime.
A further change of sign in the RG equation results however in the correct scaling behavior,
Eq. (21), at order 1 .
4.2. The phase-separation problem
The above results obtained for the gauge interaction also apply when the singular
potential between quasiparticles is due to the proximity of the system to a PS region.
In fact, starting from a model with short-range interactions among electrons, a singular
effective scattering amplitude between quasiparticles arises at the instability, which reads
[7]
1
.
(24)
D PS (q, ) =
q 2 i ||
q
Eq. (24) has the same form of the gauge propagator for = 0, apart from a different
expression of the constants and in terms of the microscopic parameters. Moreover it
can be shown that perturbation theory gives for the second-order selfenergy the same result
as GT (apart from irrelevant factors), even though in this case the critical fluctuations are
coupled to the density and not to the current.
We describe now this problem through a model of fermions interacting with a scalar
boson field ,
Z
HI = g0PS
d d k d d q d d
+ q/2, + /2)(q, )(k q/2, /2),
(k
(2)2d+2
where the RPA propagator of the field is given by (24). The question of the existence
of corrections to the propagator (24) can now be addressed within the framework of the
effective-action analysis. The scaling exponent of the n-th coefficient n of the fermionintegrated action is xn = d + 3 n(d + 1)/2 as in the GT problem with = 0, so
that the terms with n > 4 can now be neglected in d > 1, as they are irrelevant in the
infrared limit. The coefficient 3 is instead relevant and drives the system to a firstorder phase transition. On the first-order line the effective interaction among fermions
is finite in both the coexisting phases. In order to stay at the second-order critical point,
where the interaction is singular, 3 must be tuned to zero. Once 3 is set to be zero, no
anomalous dimensions appears in the boson propagator in the PS case also. Therefore,
1
= 0 (k, qk
the coupling constant g PS renormalizes as g0PS = g PS ZPS /Z, where ZPS
d
3 , q 3 ; , ). We must use here the limit of the density vertex since in this case
the fermion is coupled to a scalar field. Recalling now the asymptotic WI (9), coming from
the conservation of the particles at each point of the Fermi surface we assume that the
current vertex is proportional to the density vertex, so that, according to the discussion in
Subsection 4.1, we can write ZPS = ZGT = Zstat .
760
The skeleton structure of the perturbation theory, carried out with the potential (24) is the
same as in the spin-liquid case except for the effective coupling constant which now, due
to the scalar nature of the interaction, is u0 = (g0PS )2 /vF0 (up to a multiplicative constant).
The renormalized dimensionless coupling constant u is thus defined by

(g PS )2
Zstatg PS 2 1
Zstat 2 1
=
uxu .
u0 0
vF0
Z
ZvF vF
Z
Z vF
The above equation can be simplified using the WI (7) and the fact that the relation (13)
between Z and ZvF is valid in the present case also, leading to u0 = Z 1 uxu , which is
precisely the result of Eq. (21). Since the RG equations are the same both in the PS and in
the GT models we can extend to the former the results obtained for the latter (at = 0) in
the previous subsection.
4.3. The LR problem
In this section we develop the RG approach in the case of LR interaction. First we
consider the two different kinematic regimes for the RPA potential between two electrons.
If vF q the effective interaction assumes the usual short-ranged expression due to the
screening of the interaction by the particle-hole fluctuations within the medium, while
in the opposite regime vF q it assumes the form (3) which is highly singular in the
infrared limit and signals the formation of the collective density mode with dispersion
q q 1/2 . We consider this regime restricting ourselves to the case < 2 where the
collective mode is propagating and gapless.
Analogously to the GT case we compute the one-loop selfenergy for this model,
Z d
2
1
1
2
d q d
.
(25)
(k, ) = i ig0LR
d+1
2
(2)
+ k+q q c2 q 2
The imaginary part of expression (25), evaluated with the external momentum on the Fermi
surface is
2(d)
(g0LR )2 cSd
|| 2
sgn()
.
(26)
00 (k = kF , ) =
d
2(2) (2 )
c
The corresponding real part can be obtained via Hilbert transformation of 00 sgn(),
giving 5

(d )
0
00
(k = kF , ) = (k = kF , ) tan
.
(27)
2
The calculation for finite arbitrary momentum in the limit of zero energy shows that,
contrary to the case of GT, this selfenergy has also a momentum-depending singular part

2(d)
(g0LR )2 Sd
vF0 |k kF | 2
0
,
(28)
(k, = 0) =

sgn(k kF )
c
2(2)d sin 2(d)
2
5 In general, if 00 sgn()|| , with < 1, then, via Hilbert transformation, one finds 0 = tan( ) 00 .
2
761
while, of course, 00 (k, = 0) 0. Analogously to the GT case, we use now the results
(26), (27) and (28) to construct the contribution to the selfenergy with dressed vertices
Z d
2
d q d 1
0 (k + q, q; + , )
(k, ) = ig LR
d+1
2
q c2 q 2
(2)
1
,
2d2
2d2
A| + | 2 sgn( + ) B[vF ||k + q| kF |] 2 sgn(|k + q| kF )
where the complex constant A can be deduced form (26) and (27), and the real constant B
can be deduced from (28). It can be seen that the most singular contribution to this integral
comes from the pole 2 q 2 of the boson propagator, so that the relation between
momentum and energy is fixed as q . We define the renormalized quantities as in
1
= 0 (k, q =
(15). The renormalization constant which pertains to this case is thus Zdyn
0; , 0). The renormalization of the coupling constant g LR is now g0LR = g LR Zdyn/Z,
which, using the WI (8), becomes g0LR = g LR , i.e., the divergences in vertex and selfenergy
diagrams cancel out thanks to the WI coming from the particle number conservation law.
On the other hand, concerning the renormalization of the Fermi velocity, one gets from
Eqs. (27) and (28) that both / and /k yield the same singular behavior in
the infrared. This situation is translated into the relation (14) among the renormalization
factors, which implies that the Fermi velocity stays unrenormalized.
We now determine the effective coupling constant which appears in the skeleton
structure of the perturbative expansion. In order to do this we consider again the structure
of the one-loop selfenergy (25). The leading contribution comes from the pole of the
boson propagator = cq 1/2 , i.e., from the kinematic region where the energy of the
internal fermion propagator is + cq 1/2 . We notice that, contrary to the GT case,
where the damped mode, which is mixed to the fermions, is embedded in the particle
hole continuum, in the LR case the mode with dispersion q = cq 1/2 represents an
undamped collective excitation, with an energy which is much larger than the energy of
the fermion, for small momenta, since in the infrared limit k+q k cq 1/2 for < 2.
In view of this, for k = kF the leading contribution to (25) in the infrared region is obtained
by substituting the fermion propagator (evaluated at the pole of the boson propagator)
( + cq 1/2 k+q )1 with the approximate expression ( + cq 1/2)1 . One important
consequence of this is that the fermion propagator in (25) does not contribute with a factor
vF1 as it was the case in the GT model. The effective coupling constant thus coincides with
the coupling constant (g LR )2 , apart from a constant multiplicative factor coming from the
integration in the diagram [see Eqs. (26) and (27)].
Defining u0 (g0LR )2 we can write the one-loop wave-function renormalization constant
as 6
2+2d
2
|| 2 .
Z 1 () = 1 + u0
2 + 2d
Since this quantity must be dimensionless, we can compute the scaling dimension of u0 ,
thus obtaining, for the dimensionless coupling constant, the expression
6 See footnote 4.
762
u=
u0
,
x u
(29)
with xu = 2+2d
2 [] = 1 d + /2. Contrary to the GT case, in the LR problem the
flux of the coupling constant is given only by its bare dimensions since u does not acquire
any anomalous correction. Notice that the scaling dimension of the coupling constant
can also be determined from the condition that the RPA propagator (3) must retain the
same structure at all scales. In view of this we choose [q] = 1 and [] = 1 /2. This
assures that the coefficient c is dimensionless at the tree level. Moreover the absence of
renormalization of the boson propagator guarantees that the constant c does not acquire
anomalous dimensions from loop corrections. This gives the scaling dimension of the fields
[] = ( d)/2 1, [] = (2 + + 2d)/4 from which the scaling exponent of the
fermionboson coupling constant is computed as xg = (2 + 2d)/4 and xu = 2xg .
Eq. (29) shows that a line of non-universal fixed points exists only for xu = 0, i.e., for
d = dc 1 + /2, while for xu > 0 or xu < 0 (i.e., for d < dc or d > dc ) the effective
coupling constant either scales to strong coupling or to the free fixed point u = 0. In the
strong-coupling and nontrivial-fixed-point cases one obtains a NFL behavior, in agreement
with previous results that NFL behavior could only be produced by a sufficiently singular
potential, > 2d 2 [8,9,11].
A differential equation for the renormalization constant Z can be written in this case
only through one-loop perturbation theory obtaining Z 1 Z = x u. The integration of
this equation corresponds to the exponentiation of the one-loop results and yields:
For d < dc 1 + /2 the renormalization constant Z vanishes in the infrared as

xu
x
(30)
Z exp u0 || x ,
xu
while the system scales to strong coupling.
At the critical dimension dc , the effective coupling constant u has a non-universal
fixed point u = u0 as in the one-dimensional LL and the renormalization constant Z
goes to zero with the non-universal power
Z ||u0 .
(31)
We notice that Eqs. (30) and (31) are in agreement with the results obtained through the
WI method [9,10]. We can thus argue that even in the case of a LR interaction, the RG
approach based on a RPA expansions gives the exact result.
5. Conclusions
We have developed a RG approach specific of models with dominating forward
scattering. The conservation of particles makes it possible to control, via the WI, the flow
of the group equations and to compute the asymptotic properties of the single-particle
propagator.
We have thus settled the gauge-field case, by confirming in general the 1/N result of
Ref. [14] in d = 2 that higher-order corrections do not modify the one-loop result for the
power-law behavior of the selfenergy below the critical dimensionality dc 3 .
763
The scattering potential due to charge fluctuations near phase separation finds its group
characterization within the same scheme with = 0, even though the interaction of the
electron is now mediated by a scalar field.
In the LR problem (V (q) q ), we find a line of fixed points at a critical dimension
dc 1 + /2, with a LL behavior. For d < dc the system scales to strong coupling and a
NFL behavior is obtained with the fermion field renormalization scaling exponentially to
zero.
Acknowledgements
S.C. thanks P. Werner for discussions and careful reading of the manuscript. A.M. is
grateful to J.B. Marston and P. Kopietz for useful discussions.
Appendix A. Non-renormalization of the RPA boson propagator

We present here an estimate of the degree of scaling divergence of the n-th coefficient
of the effective action for the boson fields. An important issue in the evaluation of these
coefficients, which are given by the n-leg fermion loops, is the so called loop-cancellation
theorem. In d = 1 it is known [10,24] that, assuming a linear dispersion for the fermion
the full sum of the loops involving n internal fermion propagators vanishes for n > 2.
This cancellation is only partial in d > 1 and the evaluation of the leading corrections
to the loop-cancellation result has been considered in various papers, in connection with
problems involving only forward scattering processes near the Fermi surface.
Here we want to consider these n-leg loops in the kinematic regime q (dynamic
limit), which is relevant for the LR problem. To do this let us consider an action describing
a fermion field minimally coupled to a (d + 1)-dimensional gauge field A = (A, ),
Z
Z
+ q, + )(k, )(q, )
S(, , A) =
(k, )( k )(k, ) + (k
Z
1
+ q, + )(k, )(2k + q) A(q, )
(k
+
2m
Z
1
+ q + p, + + )(k, )A(q, ) A(p, ).
(k
2m
We perform the gauge transformation 0 ei so that the gauge fields change into
0 = + i,
A0 = A iq.
(A.1)
We choose = i/. In this way the gauge-transformed field is 0 = 0, A0 = A q/.

Since in our LR problem the initial vector potential is zero, the application of the gauge
transformation (A.1) amounts to eliminate the scalar potential in favour of a new vector
potential A0 = q/. Starting from S(, ) the final action is thus
Z
0 (k, )( k ) 0 (k, )
S( 0 , A0 ) =
764
Z
1
+
0 (k + q, + ) 0 (k, )(2k + q) A0 (q, )
2m
Z
1
0 (k + q + p, + + ) 0 (k, )A0 (q, ) A0 (p, ). (A.2)
2m
We notice that, since we are interested in the behavior in the dynamical limit q , A0 =
q/ is well-behaved in the infrared.
Due to gauge invariance the n-leg loop calculated with the action (A.2) must yield
the same result of that calculated with the initial scalar interaction. We find it more
advantageous to carry on the calculations with gauge-transformed fields because each
interaction vertex now automatically gives an explicit q/ or (q/)2 factor. The
calculation of the loop with n (amputated) external boson lines has contributions from
the diagrams with all three-leg interaction vertices (bosonfermion current vertices)

Z
q2
qn
q1
q2
qn
q1
k + q1 + + qn1 +
k + q1 +
k+
2
1
2
2
2
n
G(k, )G(k + q1 , + 1 ) G(k + q1 + + qn1 , + 1 + + n1 )
X
X

n
n
qi
i
i=1
i=1
+ (permutations)
(A.3)
and from the diagrams in which up to n (for even values of n) or n 1 (for odd values of
n) boson lines are coupled in four-leg vertices (bosonbosonfermion density vertices)

Z
q3
q3
q1 q2
k + q1 + q2 +
1 2
2
3

qn
qn
k + q1 + + qn1 +
2
n
G(k, )G(k + q1 + q2 , + 1 + 2 )
G(k + q1 + q2 + q2 , + 1 + 2 + 3 )
G(k + q1 + + qn1 , + 1 + + n1 )
X
X

n
n
qi
i
i=1
i=1
+ (permutations)

Z
q5
q5
q3 q4
q1 q2
k + q1 + + q4 +
+
1 2
3 4
2
5

qn
qn
k + q1 + + qn1 +
2
n
G(k, )G(k + q1 + q2 , + 1 + 2 )
G(k + q1 + + q4 , + 1 + + 4 )
G(k + q1 + + q5 , + 1 + + 5 )
G(k + q1 + + qn1 , + 1 + + n1 )
X
n
i=1
765
X

n
qi
i
i=1
+ (permutations) +
(A.4)
We notice that each contribution in (A.4) has one fermion propagator and two current
insertion less than the preceding one, as long as the current insertions are replaced by the
vertices generated by the interaction term 0 (A0 )2 0 .
It is apparent from the expressions (A.3) and (A.4) that each contribution to the nleg loop explicitly carries out a factor which scales as (q/)n , so that we can write
n (q, ) = n (q, )(q/)n . The leading contribution in q/ in the dynamic limit can
then be evaluated computing n (0, ).
It is then easy to see from expressions (A.3) and (A.4) that the factor n (q, ) evaluated
with all momenta equal to zero is vanishing for any choice of the in-going frequencies.
The first nonzero contribution has to be calculated by expanding the function n (q, )
for small momenta with finite frequencies this will yield a contribution to n that will
obviously scale as (q/)m , with m > n.
From the above discussion we can conclude that the n-th term of the effective action
goes to zero in the infrared dynamic limit more rapidly then (q/)n . Assuming now an
n-th term of the form
Z
n1
qn
,
cn n n d d q d
the coupling constants cn would have a scaling dimension equal to [(n 2) + (4

2n)(d + 1)]/4 which is positive for > 2(d + 1). These coupling constants would thus
be relevant only in the unphysical regime > 2. We recall that the actual n-th coupling
constant appearing in our problem will have an even smaller scaling dimension and thus it
will scale to zero in the infrared even more rapidly.
We can thus conclude that all terms with n > 2 scale to zero in the infrared dynamical
limit and the RPA form of the boson propagator keeps its form at all orders in perturbation
theory, according to the discussion given in the text.
Appendix B. Perturbative calculations in the GT model

The one-loop selfenergy for the GT model has been calculated in several papers [5,6,
14,19]. We report here its expression for generic d and . For || |k kF |3 the
imaginary part is

d
|| 3
,
(B.1)
00 (k, ) = g02 vF0 1 C1 (d, ) sgn()
where
C1 (d, ) =
Sd1
2d(2)d sin
d
62
,
766
is a dimensionless numerical factor, while for || |k kF |3 one has

2

|k kF |d6+2 ,
00 (k, ) = g02 vF0 1 C2 (d, ) sgn()
with

Sd1 d2 3 d2
.
C2 (d, ) =
4(2)d (3 )
The real part can be calculated via Hilbert transformation of 00 sgn(), yielding [5,19]
d

|| 3
0
2
1
,
(B.2)
(k, ) = g0 vF0 C3 (d, ) sgn()
with

d
,
C3 (d, ) = C1 (d, ) tan
6 2
for || |k kF |3 , and
(k, ) = g02 vF0 1 C4 (d, )

|k kF |d3+ ,
with
C4 (d, ) =
2
C2 (d, ),
for || |k kF |3 . We notice that for d = 3 the expression (B.2) acquires

logarithmic corrections i.e., one has 0 (k, ) log || instead than 0 (k, ) . This
dimension turns out to be the critical dimension below which a non-trivial fixed point of
the RG equations appears [16].
From (B.2) one can also read the numerical factors included in the definition of u0 7
d
Sd1
3 3 d 2
u0 =
g0 vF0 .
d
2
d
2 (2) (3 )
cos 62
Notice that u0 stays well defined at the marginal dimension dc = 3 .
Appendix C. Estimate of the vertex in the GT case

As we mentioned in the main text it is necessary to evaluate the fermionboson current
vertex in the kinematic region relevant to the GT problem. Neglecting at the beginning the
contribution of transverse scattering processes we start from Eqs. (9) and (10), and insert
in the latter the asymptotic expressions (B.1) and (B.2) for the imaginary and real part of
the electron selfenergy. The numerator of (10) has then the leading behavior d/(3 ) , for
d < 3 . Observing that the denominator is vF qk , and that the relevant kinematic
7 See footnote 4.
767
regime is qk d/(3 ) , we find that 0 stays finite as 0. Eq. (9) implies that
the fermionboson current vertex is also finite.
Next, we consider the effect of corrections due to transverse scattering processes. We
consider specifically the explicit estimate of the fermionboson current vertex obtained in
Ref. [14] for d = 2 and = 0 (spin-liquid case). The same results we present here have
been obtained by an independent numerical evaluation of the vertex [27]. Eq. (9) is still
valid, but Eq. (10) has to be substituted by Eq. (37) of Ref. [14],
0 (k, q; , ) =
G1 ( + /2, k + q/2) G1 ( /2, k q/2)

,
vF0 qk vF0 |q |B(qk , q ; )
where
qk
B(qk , q ; ) =
2|q |
"s
#
2|||q |
1
1 ,
(vF0 qk )2
(C.1)
(C.2)
and is a constant expression, involving the bare Fermi velocity and the bare coupling
constant that do not change under renormalization, due to the above-discussed nonrenormalization of the boson propagator.
The function B(qk , q ; ) describes the corrections to the density vertex coming from
the transverse scattering processes, not included in (10). However, even taking these
corrections into account (i.e., keeping B 6= 0), the density vertex computed in the limit
of Eq. (17) still does not yield any relevant contribution and ZGT can be still regarded as
a constant. Indeed, we calculated the expression (C.1) in the kinematic region (, qk
d
3 , q 3 ), making again use of the expressions (B.1) and (B.2) for the electron
selfenergy, specializing to the case d = 2 and = 0, when the expression (C.2) holds. We
found that the leading contribution to the density vertex is
2/3
= const.
(C.3)
2/3
An important property of the correction (C.2), which ultimately leads to the finiteness
of the density vertex in the infrared limit as in the case B = 0, is its scaling form in the
0 (k, q; , )
relevant kinematic region (, qk 3 , q 3 ). While we were not able to explicitly

extend the result (C.2) for generic d and , we suggest that the scaling form of (C.2) is not
accidental, and that the result (C.3) is generic. Together with Eq. (9), this implies that
the current vertex (17) is finite and ZGT can be consistently regarded as a number in the
derivation of the RG equations.
References
[1] C. Castellani, C. Di Castro, W. Metzner, Phys. Rev. Lett. 72 (1994) 316.
[2] J. Slyom, Adv. Phys. 28 (1979) 201.
[3] P.W. Anderson, Science 235 (1987) 1196;
P.W. Anderson, Phys. Rev. Lett. 64 (1990) 1839;
P.W. Anderson, Phys. Rev. Lett. 65 (1990) 2306.
768
[4] G. Baskaran, P.W. Anderson, Phys. Rev. B 37 (1988) 580;

L.B. Ioffe, A.I. Larkin, Phys. Rev. B 39 (1989) 8988.
[5] P.A. Lee, N. Nagaosa, Phys. Rev. B 46 (1992) 5621.
[6] B.I. Halperin, P.A. Lee, N. Read, Phys. Rev. B 47 (1993) 7312.
[7] C. Castellani, C. Di Castro, M. Grilli, Phys. Rev. Lett. 75 (1995) 4650.
[8] P. Bares, X.G. Wen, Phys. Rev. B 48 (1993) 8636.
[9] A. Maccarone, Thesis, University of Rome, 1994;
C. Castellani, C. Di Castro, Physica C 235240 (1994) 99.
[10] W. Metzner, C. Castellani, C. Di Castro, Adv. Phys. 47 (1998) 317.
[11] A. Houghton, H.-J. Kwon, J.B. Marston, R. Shankar, J. Phys. C 6 (1994) 4909.
[12] J. Gan, E. Wong, Phys. Rev. Lett. 71 (1993) 4226.
[13] J. Polchinski, Nucl. Phys. B 422 (1994) 617.
[14] B.L. Altshuler, L.B. Ioffe, A.J. Millis, Phys. Rev. B 50 (1994) 14048.
[15] C. Nayak, F. Wilczek, Nucl. Phys. B 430 (1994) 534.
[16] S. Chakravarty, R.E. Norton, O.F. Syljusen, Phys. Rev. Lett. 74 (1995) 1423.
[17] D. Schmeltzer, Phys. Rev. B 52 (1995) 7939.
[18] C. Castellani, C. Di Castro, A. Maccarone, Phys. Rev. B 55 (1997) 2676.
[19] B. Blok, H. Monien, Phys. Rev. B 47 (1993) 3454.
[20] J. Frlich, R. Gtschmann, P.A. Marchetti, Commun. Math. Phys. 173 (1995) 417.
[21] Y.B. Kim, A. Furusaki, X.-G. Wen, P.A. Lee, Phys. Rev. B 50 (1994) 17917.
[22] P. Kopietz, J. Hermisson, K. Schnhammer, Phys. Rev. B 52 (1995) 10877.
[23] H. Fukuyama, H. Ebisawa, Y. Wada, Prog. Theor. Phys. 42 (1969) 494.
[24] I.E. Dzyaloshinskij, A.I. Larkin, Sov. Phys. JETP 38 (1974) 202.
[25] C. Di Castro, W. Metzner, Phys. Rev. Lett. 67 (1991) 3852;
W. Metzner, C. Di Castro, Phys. Rev. B 47 (1993) 16107.
[26] P. Kopietz, Bosonization of Interacting Fermions in Arbitrary Dimensions, Lecture Notes in
Physics, Vol. M 48, Springer-Verlag, Berlin, 1997.
[27] A. Maccarone, J.B. Marston, unpublished.

Slave bosons in radial gauge:

the correct functional integral representation
and inclusion of non-local interactions
Raymond Frsard a,1 , Thilo Kopp b,
a Institut de Physique, Universit de Neuchtel, A.-L. Breguet 1, 2000 Neuchtel, Switzerland
b Experimentalphysik VI, Elektronische Korrelationen und Magnetismus, Universitt Augsburg, D-86135
Augsburg, Germany
Received 9 June 2000; accepted 2 November 2000
Abstract
We introduce a new path integral representation for slave bosons in the radial gauge which is
valid beyond the conventional fluctuation corrections to a mean-field solution. For electronic lattice
models, defined on the constrained Fock space with no double occupancy, all phase fluctuations of
the slave particles can be gauged away if the Lagrange multipliers which enforce the constraint on
each lattice site are promoted to time-dependent fields. Consequently, only the amplitude (radial
part) of the slave boson fields survives. It has the special property that it is equal to its square in the
physical subspace. This renders the functional integral for the radial field Gaussian, even when nonlocal Coulomb-type interactions are included. We propose (i) a continuum integral representation
for the set-up of further approximation schemes, and (ii) a discrete representation with an Ising-like
radial variable, valid for long-ranged interactions as well. The latter scheme can be taken as a starting
point for numerical evaluations. 2001 Elsevier Science B.V. All rights reserved.
PACS: 71.27.+a; 11.10.-z; 11.15.Me; 11.15.Tk
Keywords: Strongly correlated electrons; Field theories; Non-perturbative methods
1. Introduction
The significance of strong electronic correlations in metals has long been acknowledged
when intractable enigmas of solid state physics such as itinerant magnetism, the
correlation-induced metalinsulator transition, unconventional superconductivity, or spin
liquid states and non-Fermi liquid behavior of electronic excitations were embraced.
E-mail addresses: raymond.fresard@ismra.fr (R. Frsard), thilo.kopp@physik.uni-augsburg.de (T. Kopp).

1 Permanent address: Laboratoire CRISMAT-ISMRA, 14050 Caen, France.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 5 7 - X
770
R. Frsard, T. Kopp / Nuclear Physics B 594 [FS] (2001) 769789
The scope of the strongly correlated physics is nowadays enormous; it extends to the
fractional quantum Hall effect and high-Tc superconductivity, and also in mesoscopic
systems an increasing number of investigations take this route since certain scenarios with
strong local interactions can be modeled more directly in experiments on a mesoscopic
scale. The efforts to fathom the underlying mechanisms resulted only in partial success.
Especially, models with strong local interactions of itinerant electrons resisted a thorough
understanding beyond certain limiting, and often unphysical, cases.
The celebrated Hubbard model epitomizes the two opposing characters of narrow band
electrons: electron hopping (with energy scale t) supports the itinerant character, suitably
expressed in momentum space, and favors a metallic character. In contrast to this, the
local on-site Coulomb interaction U may drive the electrons, depending on band-filling,
into a Mott insulating state, with strong magnetic correlations. Standard perturbation
theory is not applicable in the transition regime (for a recent discussion see [1]), and
whether it even was valid for any finite U in two space dimensions [2] was a longstanding
problem. On the other hand, perturbation theory from the atomic limit, with no kinetic
terms, is not feasible in a straightforward manner since the tremendous degeneracy for low
energy states of the fully localized limit renders the usual perturbation theory inapplicable.
Furthermore, a hopping expansion cannot build on Wicks theorem since the zeroth order
Hamiltonian is not bilinear in the canonical electron fields, though progress has been
achieved recently [3,4]. Nonetheless, for t U a SchriefferWolff transformation may
be performed to the appropriate order in t/U . This may be visualized as a rotation of
the states in Hilbert space such that transitions between different subspaces with fixed
number of doubly occupied sites are suppressed to the corresponding order in t/U .
Equivalently, one usually applies a canonical transformation to the Hamilton operator in
order to generate an effective Hamiltonian of which the low energy part is explicit. The
Hubbard model is reduced in this way 2 to a tJ model (plus pair-hopping) up to linear
order in t/U or to a t-model in zeroth order. The t-model is just the kinetic term,
however it is now a correlated hopping in the sense that double occupancies are excluded.
This constrained hopping can only be envisioned in real space. The tJ model includes,
beyond the constrained hopping, a Heisenberg spin-exchange and a short range densitydensity correlation term of scale J t 2 /U . Although all these strong coupling models
present the local spin and charge excitations in an appealing way, it is still a long way to
extract the true (delocalized) low-energy excitations.
In the past decades considerable effort has been invested into the construction of
techniques which work in the constrained Fock space. Already Hubbard introduced
projection operators onto local (site-) states with no electron (|0i), one electron with
spin (| i) or double occupancy (|2i). These so-called Hubbard X-operators belong to
a graded Lie-algebra, and a canonical many-particle Greens function formalism is not
applicable (for a detailed review of functional integral representations and a suitable choice
of coherent states for X-operators see Ref. [6]). The evaluation scheme by Hubbard shaped
2 In a different approach to these effective models, a degenerate perturbation theory is set up ab initio so that a
Hamilton operator in the projected Fock space, restricted to low-energy excitations, can be determined [5].
771
the appropriate picture of lower and upper Hubbard bands for low and high energy (in-)
coherent excitations, respectively. Yet it suffered from inconsistencies, e.g., the failure to
satisfy the Luttinger theorem.
An alternative approach, which also implements a projection onto local states, is the
slave boson formalism [7,8]. Electron creation and annihilation operators are represented
by composite operators whereby an electron is annihilated either by the creation of a hole
site, accompanied by the annihilation of a singly occupied site with corresponding spin
or the annihilation of a doubly occupied site, accompanied by the creation of a singly
occupied site with opposite spin. Several choices exist how to ascribe fermionic and
bosonic degrees of freedom to these operators, for a discussion see [9]. Although the
operators are now canonical fermionic and bosonic operators, they live in a constrained
Fock space since the sum of their respective particle numbers (the pseudo charge) has to
be unity for each site. Mean field or saddle point approximations satisfy the constraint only
on the average and, furthermore, lead to Bose condensation of the slave bosons. 3 Actually,
the saddle point approximation becomes exact in the unphysical limit N , where N
is the spin degeneracy. Corrections to the mean field result were calculated in the first order
of the expansion parameter 1/N the constraint is thereby replaced by a soft constraint
with pseudo charge equal to N /2. For N = , the bosons condense, that is, they attain
a finite complex expectation value, the square of which is the density of empty sites.
In the Barnes representation, only the charge degree of freedom is represented by the
slave boson. In contrast, both spin and charge degrees of freedom are carried by bosons
(and fermions) in the Kotliar and Ruckenstein representation [8], while in the spin and
charge rotation invariant representation [9] they are carried by bosons only. The meanfield approach to the Kotliar and Ruckenstein representation has been quite successful
when compared to numerical simulations. Indeed, the paramagnetic mean-field solution is
equivalent to the Gutzwiller approximation[8]. Even though this is the saddle point of an
action, it turns out to obey a variational principle in the limit of large spatial dimensions,
where the Gutzwiller approximation and the Gutzwiller wave function are identical [11].
When the dimension is reduced, they are not known to differ markedly. Comparison of
ground state energies [12,13], and charge structure factors [14] with Quantum Monte
Carlo data show excellent agreement. A variety of saddle points have been investigated,
in particular in the vicinity of the Mott transition [1518]. The Mott gap can be calculated
as well [19], and comparison with experimental data in Lax Y1x TiO3 [20] displays a very
good agreement [17]. Nevertheless the computation of fluctuations, in particular in the
magnetic channel, remains problematic.
Moreover, such a finite expectation value of the complex slave boson field violates
Elitzurs theorem according to which, in a theory with local gauge invariance, only locally
gauge invariant operators may acquire a non-zero expectation value. Correspondingly, for
finite N , the phase fluctuations of the boson field suppress the condensation. A zero
expectation value of the bosonic field, however, is problematic as a starting point since
3 However, Bose condensation may be suppressed, even on a mean field level, and bond pairing of bosons
emerges instead if the constraint is recast into an interaction term between the various slave particles [10].
772
the bosonic field has to connect to the number of empty sites, in the end. It was early
on recognized that these difficulties are avoided in a polar decomposition where the zero
mode of the phase fluctuations may be singled out and the expectation value of the radial
amplitude is, in general, non-vanishing [21,22] (a more recent, detailed discussion of
1/N -corrections in the radial gauge is found in the review by Arrigoni et al. [23]).
Although subtleties are associated with the measure and the correct translation of the time
discretized functional integral into the continuum limit version, the result in 1/N is not
different from that of an X-operator approach [24].
Despite the progress in the understanding of the fluctuations around the mean field
saddle point, the physically most interesting case with the largest possible spin fluctuations,
N = 2, remains inaccessible. Moreover, the saddle point solution is generically a Fermi
liquid solution and, therefore, non-Fermi liquid behavior is necessarily non-perturbative
in 1/N . This motivated the Karlsruhe group [25] to device a slave-boson formalism which
builds on canonical many-particle techniques, is locally gauge-invariant by construction
and, most importantly, allows for Fermi liquid and non-Fermi liquid behavior for the
considered (multi-channel) Anderson impurity problem. It permits to recover the physical
quantities over the complete temperature range, from the high temperature local moment
regime to the low temperature correlated many-body state. In spite of this major success
of slave boson theory, it is still a formidable problem to set up a similar diagrammatic
evaluation for strongly correlated itinerant electrons.
It is quite instructive to compare the implementation of the constraint in this diagrammatic formalism with that in the functional integral radial gauge representation. In the first
approach, local gauge invariance is guaranteed through Ward identities in a conserving
approximation. This enforces the constraint only in so far as transitions between sectors
of the Fock space with different pseudo charge are suppressed during the time evolution
of a considered many particle state. The projection onto the physical sector of the Fock
space is achieved in a last step, when physical expectation values are calculated, using the
Abrikosov procedure of sending the Lagrange multiplier of the constraint to infinity. This
may be contrasted with the functional integral calculation in the radial gauge: here the
phase is fixed and the amplitude of the boson will in turn strongly fluctuate. The pseudo
charge conservation will not be satisfied a priori. This is precisely the reason why one has
to project onto the physical subspace at each time step (and, of course, at each site) which
is accomplished by promoting the Lagrange multiplier of the constraint into a field.
In this paper we reexamine functional integrals for strongly correlated electrons with
N = 2 in the radial gauge. We restrain our considerations to U , that is the limit
of strong local interaction between electrons so that the number of doubly occupied sites
is zero. The Hubbard model is reduced to either the t- or tJ model. In this limit, the
bosonic fields in radial representation reduce to their respective (real) amplitude since
all time derivatives of the corresponding phases can be absorbed in a (time-dependent)
Lagrange multiplier field.
It is possible to construct the functional integral either as an integration over continuous
values of the boson radial amplitudes and the Lagrange multiplier, or, equivalently one
can implement the constraint first and thereby reduce the integration over the slave
773
boson amplitude to a discrete sum over the values of hole occupancy, {0, 1}. The first
method is more appropriate for approximation schemes which further manipulate the path
integral analytically, and the latter method is to be taken as a starting point for numerical
calculations in the strong coupling limit which do not build on a Hirsch decoupling [26].
In general, there are no ab initio advantages of one functional integral representation
over another representation. It depends on the approximation scheme if a concrete
representation is suitable. For example, when 1/N -fluctuations are considered the radial
gauge consistently separates the zero-mode fluctuations. Moreover, for any N , the square
of the slave boson amplitude is equal to the amplitude itself since the constraint projects
the amplitude onto the values {0, 1}. In the radial gauge, this renders the integrations over
the real fields Gaussian, albeit the bosons are still coupled to the fermionic Grassmann
fields. This projection property also ensures that the measure does not produce the
cumbersome logarithmic terms in the action and, more significantly, non-local density
density interactions turn out to be Gaussian terms in the functional integral. The latter is not
valid in the Cartesian gauge. These observations should suffice as motivation to readdress
the construction of the slave boson functional integral in the radial gauge.
In Section 2 we introduce the correct functional integral representation in the radial
gauge, in the limit U . In Section 3 we propose an alternative discrete representation
which may serve for numerical simulations. In Section 4 we extend these representations
to lattice models, with (possibly) long-ranged densitydensity interactions. Several
calculations for simple models are carried out to illustrate this new method.
2. Functional integral representation in the radial gauge

In any slave boson representation of strongly correlated electron systems there is a
gauge symmetry group as a consequence of the redundancy of the representation of the
physical electron operator. The first slave boson representation is due to Barnes [7]. It
was introduced to tackle the single-impurity Anderson model. The local physical electron
operator c is, in the limit of infinitely strong on-site interaction, rewritten as a product of
auxiliary fields f and b:
c = bf .
(1)
Here f
is a doublet of canonical fermionic fields, and b a canonical bosonic field. We stay

in the physical sector of the Fock space provided the auxiliary fields fulfill the constraint
relation:
X
f f + b b = 1.
(2)
Q
In a functional integral representation of the partition sum Z, the constraint is enforced via
the integration over a Lagrange multiplier (for a standard reference see [27]):
Z/
Z=
/
d i
e
2
Z
Z Y

R
0
D f , f
D b, b e 0 d L ( )
(3)
774
with
L0 ( ) = Lf ( ) + Lb ( ) + Lt ( ) + Lnloc ( )
and
Lf ( ) =
(4)
f ( )( + i)f ( ),
Lb ( ) = b ( )( + i)b( ).
(5)
The last two contributions to the Lagrange density are hopping of electrons (Lt ) and
the non-local part of the electronelectron interaction (Lnloc ), respectively. The chemical
potential, which fixes the average number of electrons, is . The space dependence (lattice
site index i) of the fields and of is suppressed in Eq. (3), for simplicity. Since the pseudo
charge Q, defined in Eq. (2), is a conserved quantity, the action is invariant under a group
of local U (1) gauge transformations [22]. In the continuum limit, it reads:
f ( ) f ( ) = f ( )ei( ) ,
) = (
(
).
) = b( )ei( ) ,
b( ) b(
(6)
The physical electron operators are invariant under this transformation. It is well known
that has to be continued into the complex plane, and the integration contour has to
be shifted into the lower half-plane in order to ensure convergence of the functional
integral [28]:
i0 .
(7)
Moreover, the phase of the bosonic field may be gauged away by promoting the constraint
into a field. The remaining degree of freedom of the boson is consequently its amplitude.
Having gauged away the phase of the slave boson brings us to the radial gauge.
Certainly, the expectation value of any physical observable has to be gauge invariant.
In the conventional Cartesian representation with the integration over both, the real
and imaginary components of the bosonic field the particle number of bosons (holes) is
fixed within a gauge-invariant, conserving approximation [25]. In the radial representation,
however, the gauge is usually fixed in order to reduce the bosonic field to a real field,
and the number of bosons is fluctuating. Only in the final integration over the Lagrange
multiplier field ( ), the particle number conservation is ensured at each time step. This
situation is somewhat reminiscent of the theoretical modeling of superconductivity where,
in BCS theory, the phase of the condensate is fixed, but gauge invariance is restored
in RPA [29]. Nonetheless the situation for superconductivity is distinct because there,
a global gauge is broken, whereas here a broken local gauge has to be restored.
In a specific gauge, the correctness of the representation can be verified through the
evaluation of the partition sum, and any correlation function, in the atomic limit. In this
case the calculation is straightforward. For example, the partition sum reads:
Z/
Zat =
/
d (i+0 )
e
Zf Zb = 1 + 2e ,
2
(8)
775
where Zf = [1 + e(i+0 ) ]2 and Zb = [1 e(i+0 ) ]1 . Note that the projection

R
onto the physical subspace ( d) follows from a contour integral in the complex z-plane,
with z ei , and it only picks up the residue at z = 0.
One may try to carry out this program in the polar representation, namely by introducing
the amplitude rn and the phase n of the bosonic field:
Z Z

dbn0 dbn00
Z
d
rn2
Z2
dn
.
2
(9)
The bosonic part of the partition sum can be calculated [30], and the result is again
correctly Zb . Nevertheless, in this way one does not promote into a field, and the phases
n are not gauged away and, more importantly, the action to lowest order in the time step,
is not a bilinear form in r and [30].
We now start to set up the path integral representation in the radial gauge. In the first
place we observe that, contrary to speculations that the radial representation simply results
from a straightforward coordinate transformation, it needs to be set up on a discretized time
mesh from the beginning. A translation which respects the requirement that the integrals
converge, irrespective of the sequence of the various integrations, is the following:
Z
Zat = lim lim
D rn2
lim
N 0+
0
Z
D
dn
2
Z Y

S S + P (i + )
n
0 ,
n
D fn, , fn,
e f b
(10)
where
Sf =
N X
X

fn,
fn, fn1, 1 (in + 0 ) ,
n=1
Sb =
N
X

rn2 (in + 0 ),
(11)
n=1
and = /N with N the number of time steps.

We proceed with the evaluation of the atomic partition sum, step by step, in order
to convince the reader of the correctness of this representation: integration over the
Grassmann variables [f , f ] yields (we implicitly understand that the limit N has
to be taken at the end of the calculation):
Zat = lim
lim
0+
"
1+
N
Y
m=1
!
Z
N Z
Y

dn
2
d rn2 e(in +0 )(rn 1)
2
n=1
1 (im + 0 )
#2
( > 0 > 0).
(12)
776
Fig. 1. Contributions to the atomic partition sum for infinite on-site interaction U .
The path integral representation, and in particular the TrotterSuzuki decomposition, only
makes sense when |x| 1, where x denotes anything that multiplies in the course of the
calculation. Clearly, a finite |n | 6 is required in order to reexponentiate the fermionic
contribution, or, in other words, to introduce a small parameter, allowing to neglect terms
of order O( 2 ), for our purpose. The square bracket in Zat therefore may be written as
"
#2
N
Y
(in +0 )
e
(13)
Zf = 1 +
n=1
and we now obtain

Zat = lim
lim
0+
PN
!
Z
N Z
Y

dn
2
d rn
2
n=1
+ e2 e
= lim
lim
0+
"
2
n=1 (in +0 )(rn 1)
PN
N
Y
n=1 (in +0
n=1
dn
2
+ 2e e

)(r 2 +1)
PN
2
n=1 (in +0 )(rn )
#
N
N
N
Y
Y
Y
e(in +0 )(1+)
e(in +0 )
e(in +0 )(1)
2
+ 2e
+e
.
(in + 0 )
(in + 0 )
(in + 0 )
n=1
n=1
n=1
(14)
The origin of is now clear: it prescribes how to close the contour for the second term
in Eq. (14). A motivation, how to find and the bosonic Lagrange density in radial
representation, is presented in Appendix A.
In each term, the existence of the integrals is guaranteed by the shift of the -integration
contour into the lower complex plane induced by a positive 0 . The residue of the pole at
= i0 of the contour integration in the first two terms is finite and yields exactly the two
contributions of Zat : the hole line and the two particle lines with single occupancy of Fig. 1,
respectively. The third term in Eq. (14), corresponding to double occupancy, is annihilated
through the -integration, as expected from the implementation of the constraint.
It is instructive to invert the sequence of the integrations.

N Z
Y
Zat = lim
0+
rn2
n=1 0
lim
777
dn
2

2
2
2
e(in +0 )(rn 1) + 2e(in +0 )(rn ) e + e(in +0 )(rn +1) e2
N Z
Y

d rn2
= lim
0+
n=1 0

2
n2 1) + 2e0 (rn2 ) (r
n2 )e + e0 (rn2 +1) (r
n2 + 1)e2 ,
e0 (rn 1) (r
(15)
where is the Dirac -function. The last line is equivalent to:
Zat = 1 + 2e .
In this way it appears that, after the projection has been performed in the first step, rn
may only take two integer values, 0 or 1 in the last line of Eq. (15). We thus arrive at the
conclusion that the integrations over the bosonic and the constraint fields can be replaced
by the simple procedure demonstrated in Eq. (15).
To summarize, we have introduced a new path integral representation for slave bosons in
the radial gauge. It is defined on a discretized time mesh, and has a well-defined continuum
limit. Furthermore, we can take a more conventional approach (Eq. (10)), based on a
coherent state functional integral for both fermions and bosons, or use an Ising-like variable
representing the amplitude of empty sites (Eq. (15)). The extension to non-local terms in
the Lagrangian will be discussed in Section 4 where we also present a technique how to
handle the integration over the amplitude field r 2 .
3. Discrete representation
Besides the above integral representation with continuous fields, which may be
used to set up various approximation schemes, one can alternatively obtain a discrete
representation. Indeed the above calculation of the partition sum already suggests that
it can be obtained without having to integrate over continuous values of rn . Moreover,
integration over n mostly has the effect of picking up the appropriate value of rn . This
is achieved more elegantly by extending Colemans projection scheme [31] to the case of a
time-dependent constraint. For this purpose one introduces n e(in +0 ) , and rewrites
the partition sum in the following compact form:

!Z
N

X
Y
Y

rn
Sf
n
D f , f e
(16)
Zat =

n
{rn =0,1}n
n=1
1 ==N =0
with
Sf =
N X
X
n=1

fn,
fn, fn1, n e .
(17)
778
P
In Eq. (16) and all following relations, {rn =0,1}n denotes the path integral over the field
r which now is reduced to the set of discrete values rn = 0, 1 for n = 1, . . . , N . We easily
check that the atomic limit is reproduced:
!
!2
N
N

X
Y
Y
rn

n
m e
1+
Zat =

n
{rn =0,1}n
= 1 + 2e
n=1
m=1
1 ==N =0
(18)
This discretized representation is certainly the most direct way to construct the partition
sum and later on the Greens function as a sum of free fermionic paths, controlled
by a set of Ising variables r1 to rN .
3.1. Representation of the Grassmann fields for a physical electron
In the radial gauge we have the following assignment of electron fields to the previously
introduced auxiliary fields:
= rn fn,
,
cn,
cn, = rn+1 fn, .
(19)
The product composition is just the expression Eq. (1) only the choice of the time steps
has to be confirmed. This is easily derived from the requirement that in order to attain
a non-zero result for the Greens function we need complete chains in the Grassmann
integration. We first clarify the term complete chain in the following paragraph, and then
convince ourselves in Subsection 3.2 that the choice of time steps in Eq. (19) is necessary
to obtain a finite electron propagator.
Complete chain denotes a product of Grassmann numbers over all time steps, either
QN
Q
of the form N
n=1 fn, fn, or, equivalently, as
n=1 fn, fn1, which both result from
the expansion of the exponentiated action, and only the integration of such complete
chains results in a non-zero value of the path integral. The first form can be characterized
as a hole chain since it corresponds to the hole line in Fig. 1 (top line). Actually,
this hole contribution to Zat with value 1 is formed by the product of two hole chains
QN
Q
( N
n=1 fn, fn, )( n=1 fn, fn, ). In the presented slave boson scheme, this double hole
chain is matched by a complete chain of rn variables with value 1 (the first contribution
Q
in the last line of Eq. (15)). The second arrangement N

n=1 fn, fn1, then should be
interpreted as a particle chain which carries spin . Indeed, the dashed lines with value
QN
Q
exp() in Fig. 1 are formed by ( N

n=1 fn, fn, )( n=1 fn, fn1, e ) with =,
for the two lines, respectively. Finally, the third contribution in the last line of Eq. (15) is
QN
Q
generated from a double particle line ( N

n=1 fn, fn1, e )( n=1 fn, fn1, e ) but
n2 + 1), as we work in the constrained Fock space with no double
it is projected out via (r
occupancies.
3.2. Calculation of the Greens function
For the Greens function G (f i ) with electron creation at time step i and
annihilation at f the complete chains again have to exist. However, they are built from a
779
Fig. 2. Time evolution of the fields r and f in the imaginary-time local Greens function, G (m1).
The dashed line is a particle chain (see text for details). Hole chains are not depicted.
distinct arrangement of Grassmann variables from the Lagrangian and the physical electron
creation and annihilation operators. In this case, the hole chain runs only from time step
0 to the time step when the -electron is created (visualized in Fig. 2, where the electron
creation is at time step 1). The -hole chain is then replaced by a -particle chain in
between time steps 1 and m. This product of hole and particle chains still results in a finite
value for the Grassmann integrations since the electron creation and annihilation operators
at time steps 1 and m provide the missing Grassmann numbers in order to produce a
complete chain from 0 to N .
In order to verify this arrangement and to understand the shift of the time arguments by
one time step in the relation for cn, in Eq. (19), we explicitly carry out the evaluation of
the Greens function

= rm+1 fm, r1 f1,
(20)
ZG (m 1) cm, c1,
in the atomic limit. Obviously the -fermion exists for m1 time steps. When we integrate
out the Grassmann fields as above for Zat we have:

!
!2
N
N

X
Y
Y
rn
1
n rm+1 r1 1 +
m e Dm,1
.
Zat Gat, (m 1) =

n
{rn =0,1}n
n=1
m=1
1 ==N =0
QN
(21)
Here the integration over the -fermion produces one factor (1 + m=1 m e )
1
det [D], and the integration over the -fermion yields the propagator matrix-element Dm,1
multiplied by the second factor det [D]. This propagator is:
1
Dm,1
m
Y
i e

det [D].
(22)
i=2
Next we calculate the derivatives at each time step and obtain

X
rm+1 r1 (r1 ,1 r2 ,0 . . . rm ,0 rm+1 ,1 . . . rN ,1 )e(m1) .
Zat Gat, (m 1) =
{rn =0,1}n
(23)
780
As anticipated above, we produced a chain of r-values, with ri = 0 for 2 6 i 6 m. We find

the correct expression for this projected atomic Greens function
Zat Gat, (m 1) = e(m1) .
(24)
If we had erroneously taken cn, = rn fn, in Eqs. (19) and (20), we would have to replace
rm+1 by rm in Eq. (23), and the result would be incorrectly zero, due to the rm ,0 in the
r-chain. Higher order correlation functions can be calculated in a similar way.
We summarize that the expressions for the decomposition of the c-operators (Eq. (19)),
together with the expression of the local part of the action (Eq. (11)) and the measure of
Eq. (10) form a complete representation of the slave boson functional integral in the radial
gauge in the limit of infinitely strong local interaction. In the following chapter we will
discuss non-local terms in the Lagrangian, using exactly this representation.
4. Non-local Lagrangian
Two types of non-local terms have to be included for a full discussion of physical models
for correlated electrons: kinetic terms, i.e., hopping of electrons, and non-local interactions.
ci, ):
We first consider kinetic terms of the type (with ni, = ci,
X
ti,j (1 ni, )ci,

cj, (1 nj, )
(25)
Ht =
i,j,
which constrain the hopping of electrons (with hopping amplitude ti,j ) to the low energy
sector of the Fock space, i.e., doubly occupied sites are not created. In the language of
radial-gauge slave bosons we have
XX
ti,j ri,n+1 fi,n+1,

fj,n, rj,n+1
(26)
St =
n i,j,
using the relations Eq. (19). A formal solution of a lattice model with arbitrary hopping is
possible since the action is bilinear in f() and it is found in analogy to the atomic limit
of the previous section. The -derivatives, which enforce the constraint, are now taken at
each time step and each lattice site:

!

Y ri,n
X
2
i,n det [D[i,n ]]
(27)
Z=

i,n
{ri,n =0,1}i,n
i,n
i,1 ==i,N =0
with the fermionic inverse propagator matrix

(0)
Dij,nn0 [i,n ] = i,j Di,nn

0 + ti,j rj,n+1 ri,n+1 n,n0 +1
(28)
and its local part:

(0)
Di,nn0 = n,n0 i,n e n,n0 +1 .
(29)
It is understood that the anti-periodicity is taken care of implicitly. Although the sum over
{ri,n = 0, 1}i,n runs over all possible values of hole occupancies on each site and at each
time step, unphysical paths will be projected out due to the product of -derivatives.
781
Hereby, unphysical paths are the disconnected paths of the variable ri,n . 4 This observation
simplifies the calculation considerably. An explicit example is presented in Appendix B,
where the partition sum of the two-site cluster is calculated.
Secondly we consider non-local interactions. The infinite-U Hubbard model, that
we considered so far, is a highly idealized model which is unlikely to be realized
in nature. In the attempt to set up a lattice model for interacting electrons, one has
to realize that many other non-local interaction terms may contribute, (for a recent
discussion see Vollhardt et al. [32]). On top of the expected densitydensity interactions,
we also encounter spin-exchange terms, correlated hopping terms, four-site terms, etc.
Here, using our representation, it turns out that the densitydensity interaction term,
P P
Snloc = n i,j,, 0 Vi,j ni,n, nj,n, 0 , can be included without introducing a fermionic
interaction term. Rather, the contribution to the action may be rewritten in the form:
XX
Vi,j (1 ri,n )(1 rj,n ),
(30)
Snloc =
n
i,j
2 = r , valid in the physical subspace. This Gaussian form of

where we used the identity ri,n
i,n
the interaction term has to be contrasted with the canonical fermionic representation and
the Cartesian representation for slave bosons where the action remains a true interaction
term. In both latter cases, it cannot be treated in a straightforward fashion. However, here
it appears as additional Boltzmann factors in the expansion of the partition sum Eq. (27).
The last detail that prevents us from taking advantage of both the radial representation
and a conventional propagator expansion is the integration range for the bosonic field.
Indeed, it would be very tempting to extend it from [0, ] to [, ], such that one
could work in momentum space. However, such a step is unlikely to yield a meaningful
answer, since double occupancy is related to the value rn2 = 1 (see Eq. (15)), which would
then be inside the integration range. This difficulty may nevertheless be circumvented by
(i) introducing a new integration variable xn rn2 , (ii) making use of the fact that rn2 = rn
in the physical subspace, (iii) adding a potential term W xn (xn 1) that precisely vanishes
in the physical subspace, (iv) extending the integration range for xn to [, ] and,
(v) sending W to infinity at the end of the calculation that would definitely project out any
occupancy different from zero and one. This transforms the bosonic part of the action Sb
in Eq. (11) into:
X

(31)
(in + 0 )(xn 1) + W xn (xn 1) .
Sb =
n
We skipped here and in the following equations since xn now extends to negative values.
Integration over xn with measure one, now from to +, yields, in combination with
Eq. (13), the partition sum in the atomic limit:
4 A connected path is a temporal succession of r-values in which an initial value of r equal to 1 at a site is
continuously transported through the lattice until the final time step. For example, a transport process for
which the value of rj at site j jumps from 0 at time step n to 1 at time step (n + 1) is valid only when it is
connected to via a matrix element ti,j a site i where the value of ri jumps from 1 at time step (n + 1) to
0 at time step (n + 2). For disconnected paths such jumps are not related, and they are excluded since such paths
cannot generate complete chains (see previous section).
782
r
N Z
Y
dn
(in +0 ) W
e
Zat = lim
e
W
2
W
n=1

= lim 1 + 2e + e2(W ) .
!"
in +0 W 2
2W
1+
N
Y
#2
e
(im +0 )
m=1
(32)
As anticipated above double-occupancy is annihilated by taking the limit W while,

in the opposite limit W 0, i.e., hadnt we included this potential term as is customary in
slave boson calculations [23], the third contribution could well carry a substantial weight. 5
This scheme can be extended to the calculation of the Greens function, and in contrast to
the above calculation (Eq. (32)), W does not enter the result for Zat Gat, ( ), as explained
in Appendix D. Consequently, in a hopping expansion of the partition sum, W would be
related only to those paths where there is no change in occupancy on one or several sites
over the entire imaginary time interval.
Finally, we apply this scheme to lattice models. The measure of the functional integral
is the same as for the single site problem except that all fields now carry a site index.
The potential term, introduced above to enforce physical values for the amplitude of the
slave bosons, may be implemented as a global term. It is independent of the site index.
In summary, we find for spin- 21 fermions on a lattice, interacting through an arbitrary
Coulomb-like interaction, which is locally infinitely strong, the following expressions for
the action in the language of radial-gauge slave bosons:
S = Sf + Sb + St ,
where:
Sf =
(33)

fi,n,
fi,n, fi,n1, e(ii,n )
i,n,
Sb =
ii,n (xi,n 1) + W xi,n (xi,n 1) + Vi,j (1 xi,n )(1 xj,n )
i,n
St =
XX
ti,j xi,n+1 fi,n+1,

fj,n, xj,n+1 ,
(34)
n i,j,
and the partition sum is given by:

Z = lim
!
Z
Z
YZ Y

di,n
D fi,n, , fi,n,
dxi,n eS .
2
i,n
(35)
The measure is now trivial, and the interaction terms included in Sb are bilinear. As an
example of how this method works for a lattice problem we solve the Ising chain in
Appendix C.
5 For the atomic limit, this third term in Eq. (32) with W 0 in fact is equal to the contribution from the
state with double occupancy for zero on-site interaction energy. However, this observation does not imply that
we would have the option to recover the free limit without on-site interactions for an itinerant model by applying
W 0. The phase relation between the various local states would still be missing. Therefore, the third term in
Eq. (32) is spurious and we always have to take the valid limit W .
783
5. Conclusions
Models of strongly interacting lattice fermions have been the focus of many publications
on the MottHubbard metalinsulator transition, itinerant magnetism and strange metallic
phases in recent years. Apart from the thorough analysis of the Anderson impurity
case [25], slave boson approaches have been mainly restricted to saddle point evaluations.
However low energy spin fluctuations, for example, cannot be implemented adequately, as
quantum spin fluctuations are only included as perturbation. Furthermore, this technique
is not suitable to discuss non-Fermi liquid behavior, as corrections of higher order in 1/N
have to be included.
This necessitates to further pursue non-perturbative slave boson approaches. Here we
considered the construction of a functional-integral formalism in the radial gauge which is
not restricted to fluctuations around the N = saddle point. It is defined on a discretized
time mesh, and has a well-defined continuum limit. We restricted our considerations to
models in the limit of large (positive) on-site energy, that is, we work in the constrained
Fock space with no double occupancy.
These models allow, in the radial gauge, to perform the functional integrals with
real bosonic fields. The phase(s) can be absorbed in the constraint Lagrange multipliers
which are thereby promoted to real (time-dependent) fields. Dynamics of the radial fields
themselves does not exist (as discussed in Appendix A). They serve to enforce the
constraint for each time step by keeping track of the motion of empty sites, as exemplified
for the local problem in Eqs. (15) and (23), and for the itinerant problem in Eq. (28). This
procedure can be extended to other slave boson representations as, for example, to the
KotliarRuckenstein slave bosons, the representation of which is given for the Ising chain
in Appendix C. 6
Non-local terms in general render the functional integral unsolvable. Yet Coulomb-like
terms with non-local densitydensity correlations can be rewritten in bilinear form with
radial fields, that is, they represent Gaussian terms in the functional integral (Eq. (30) in
Section 4, and also other non-local interactions may be rewritten in bilinear form, e.g.,
the Ising spin coupling within the KotliarRuckenstein representation, see Appendix C).
This is neither possible in the canonical fermionic representation nor in the Cartesian
representation for slave bosons. These bilinear terms in the Lagrangian generate a finite
dispersion for the (radial) slave particles. In order to keep the bosonic part Gaussian and to
allow for a momentum space representation, one has to extend the range of integration for
the square of the radial field to minus infinity. It can be implemented with the observation
that the functional integration stays correct if we add an additional global Gaussian term,
that is, a potential term W (xi,n 1)xi,n , where xi,n is the square of the radial part of the
slave boson, and send W at the end. This proved to be a valid procedure to constrain
the amplitude of the slave boson to the physical values 0 and 1, without introducing any
6 The original Kotliar and Ruckenstein representation has been introduced for problems with finite on-site
interaction. In that case an additional field representing double occupancy needs to be introduced. Direct, but
lengthy, calculations show that this field needs to be complex in order to get the correct result for the partition
sum and correlation functions in the atomic limit.
784
additional complications (see Appendices C and D, and the paragraph with Eqs. (31)
and (32) in Section 4).
In this article we advanced two mechanisms for the solution of a functional integral
for constrained electronic problems: either we keep continuous fields which are originally
introduced on a discrete time mesh but have a well-defined continuum limit, or we enforce
the constraint in the first step and thereby reduce the slave boson radial field to an Isinglike variable in a discrete representation. The first approach is summarized in the final
expressions Eqs. (34) and (35) and for KotliarRuckenstein slave bosons in Eq. (C.2)
in which the bosonic fields are real fields and the bosonic action is of bilinear form.
The second approach is discussed in Section 3, and extended to itinerant problems in
Section 4, Eqs. (27) and (29), which can be taken as a starting point for a numerical
evaluation in the strong coupling limit that does not build on a Hirsch decoupling.
Furthermore, non-local interactions can be included without the need of any additional
decoupling.
Acknowledgements
We gratefully thank M. Dzierzawa and P. Wlfle for several stimulating discussions,
and H. Beck and J.-P. Derendinger for interesting discussions. R.F. is grateful for the warm
hospitality at the Institut fr Theorie der Kondensierten Materie of Karlsruhe University,
and the EKM of Augsburg University where part of this work has been done. T.K. greatly
enjoyed the hospitality at the Universit de Neuchtel. We acknowledge the financial
support by the fonds national suisse de la recherche scientifique, the BMBF 13N6918/1
and Sonderforschungsbereich 484 of the Deutsche Forschungsgemeinschaft.
Appendix A
Below we motivate why the radial amplitude field rn has no dynamics and, furthermore,
why the limit of integration is shifted to the negative real axis by an infinitesimal amount .
We start from Sb Eq. (5) in Cartesian representation.
X

bn (bn bn1 ) + ibn bn1
(A.1)
Sb =
n
and substitute bn = rn ein .

X
n1 n

rn (rn rn1 ) + rn rn1 1 ei (1 i)
Sb =
n

i( n + ) rn2 rn rn + rn rn + further terms in O 2 + O 3 .
(A.2)
dr 2
In the last line we observe that n rn rn 2 0 d d = 0 (due to periodic boundary

conditions for the bosonic field). The further terms in O( 2 ) can all be neglected
785
with respect to the linear terms. Here we introduced the time-dependent constraint field
n n + . The included term quadratic in is special: if rn rn is positive, it is always
a negligible correction to rn2 . However, for rn rn > 0 it may interpreted as an infinitesimal
shift of rn2 to negative values for rn 0. Consequently, the leading term in Sb now reads:
X

in rn2 .
(A.3)
Sb =
n
There is no dynamic term of the field rn as there is one for bn in Eq. (A.1) since
this amplitude field is a real field. On account of the constraint we avoid the essential
difficulty that one encounters when one uses polar coordinates as discussed by Edwards
and Gulyaev [30]. Particularly, the part of the action in polar coordinates, that corresponds
to the Gaussian part of the action in Cartesian coordinates, cannot be written in a Gaussian
form. This is true even to first order in , and in [30] higher order terms in the polar fields
must be kept as well.
Appendix B
As an illustration of how our discrete representation yields the partition sum we
explicitly calculate it for the two-site cluster. An instructive warm up exercise consists
in discretizing the imaginary time axis into six time steps. In order to evaluate Z Eq. (27)
we first choose the physical path:
r3 = r4 = 0;
r5 = r6 = 1 on site 1
r1 = r2 = 1;
(B.1)
r2 = = r5 = 1;
r6 = 0
on site 2
r1 = 0;
Here and in the following we use xn xi=2,n for anything on site 2. To built up the inverse
propagator matrix we observe that all s for which r is equal to 1 can be set to zero, since
the derivatives with respect to those s have to be taken from the prefactor. As a result the
inverse propagator matrix is:
1
0
0
0
0 0
0
0 0
0
0
0
0
1
0
0
0 0 t 0 0
0
0
0
0
1
0
0 0
0
0 0
0
0
0
3 e
0
1
0 0
0
0 0
0
0
0
0
4 e
0
0
0
0
1 0
0
0 0 t
0
0
0
0
0
0 1
0
0 0
0
0
0
0
D=
.
0
0
0
0 0
1
0 0
0
0
1 e
0
t
0
0
0
0 0
0
1 0
0
0
0
0
0
0
0
0 0
0
0 1
0
0
0
0
0
0
0
0 0
0
0 0
1
0
0
0
0
0
t 0 0
0
0 0
0
1
0
1
0
0
0
0
0 0
0
0 0
0
6 e
(B.2)
Along this path the hole hops from site 1 to site 2 at time 2, and back to site 1 at time 5.
and (4,
5). But
Accordingly one would expect that t only enters D on two entries: (2, 1)
786
it also appears at two other entries. Nevertheless, the latter two entries, which do not
correspond to physical processes, are easily seen not to contribute to the determinant. The
latter reads:
det [D] = 1 + (t)2 e4 3 4 1 6 .
(B.3)
If we now let the number of time steps to be N and gather all second order processes in t
(their number being N(N 1)), we obtain their contribution to the partition sum as
Z2 = 2(t)2 e(N2)
(B.4)
as it should. With some patience one may extend this procedure to an arbitrary number of
hopping processes, and we checked that they can all be summed up to
Z = 1 + 2e 2 cosh (t)
(B.5)
which is the correct answer.
Appendix C
In this appendix we illustrate the method described in Section 4 for the Ising chain.
To that aim we make use of the Kotliar and Ruckenstein representation of the Hubbard
model [8] in the limit U . We thus introduce two auxiliary fermions f and f ,
and three auxiliary bosons e, representing empty sites, and p and p representing singly
occupied sites. On each site i they are subject to three constraints:
X
pi, pi, = 1,
ei ei +
pi,
pi,
= fi,
fi, ,
=,
(C.1)
which are respectively enforced by three Lagrange multipliers denoted by i and i, .

The phase of all three bosons can be gauged away [9,33,34], and the three Lagrange
multipliers are promoted to fields. Since, for the Ising model, we are working at exactly
one electron per site, the e-field, representing empty sites, has to be fixed to zero. We also
could implement this half-filling condition in the standard way by introducing a chemical
potential for the electrons. However here, it is more convenient to introduce a Kronecker
-function eKi,n ,0 in the measure (see Eq. (C.3) below). Just setting the e-field to zero from
the outset would introduce a spurious divergence in the measure. The action is thus:
X

fi,n, fi,n, fi,n1, eii,n,
S =
i,n,
Wpi,n, (pi,n, 1) ii,n, pi,n,
i,n,
X
i,n
X
i,j,n
i,n
X

pi,n, + ei,n 1
Ji,j (pi,n, pi,n, )(pj,n, pj,n, ).
(C.2)
787
It consists of a fermionic part, the first contribution, and the remaining ones form the
bosonic part Sb . Here, Ji,j is the Ising-spin coupling, and W was introduced in Eq. (31)
and the preceding discussion. After the integration over the fermions we obtain the partition
sum as:
( " N Z
#)
Z Y
Z Y
Z
Y Y
di,n
di,n,
dpi,n,
dei,n eKi,n ,0
Z = lim
W
2
2
"
Y
i,
n=1
1+
N
Y
!#
eii,n,
eSb .
(C.3)
n=1
We first integrate over the constraint fields , then over and finally over e. As a result
the term with W drops out since the -integrations already enforce the constraint to
the point that pi,n, can take only the discrete values 0 and 1. Furthermore, we observe that
pi,n, loses its time-dependence since the previous integrations resulted in the two straight
Q
Q
in )(p
in 1)(p
in 1) + n (p
in ) for each site i. Thus, setting
world lines n (p
pi, pi,1, = = pi,N, , we get:
#
"
Z
YZ

i, )(p
i, 1) + (p
i, 1)(p
i, )
dpi,
dpi, (p
Z =
i
P
i,j Ji,j (pi, pi, )(pj, pj, )
(C.4)
which holds for any coupling matrix Ji,j and topology. If we now restrict ourselves to
nearest neighbor interaction and a chain of length L with open boundary conditions, we
can recursively integrate over p1, , p1, , . . . , pL, , and pL, to obtain the known result
L1
.
(C.5)
Z = 2 2 cosh (J )
Appendix D
We calculate the Greens function in the atomic limit using the action Eq. (34) in which
a global constraint was introduced so that the radial amplitudes can be integrated from
to . In combination with the expression of the Greens function Eq. (20) and the
propagator Eq. (22) we get:
!
Z
N Z
Y
dn (in (xn 1)+W xn (xn 1))
e
dxn
x1 xm+1
Zat Gat, (m 1) = lim
W
2
n=1
P
Pm (i )
N
(i
i )
i
i=1
i=2
1+e
(D.1)
This relation is easily verified either by straightforward integration over the Grassmann
fields or, more directly, by realizing that in Eq. (21) one has to replace n ein in
the
D 1 and in det D, and (nrn /n ) is replaced by the bosonic path integral
R propagator
dn (in (xn 1)+W xn (xn 1))
. Integration over yields:
2 e
788
Zat Gat, (m 1)
N Z
Y
= lim e(m1)
W
!
dxn eW xn (xn 1) x1 xm+1
n=1
"
1 1)
(x
m
Y
n)
(x
n=2
1)
+ e (x
m
Y
n=2
=e
(m1)
N
Y
n + 1)
(x
!
n 1)
(x
n=m+1
N
Y
!#
n)
(x
n=m+1
(D.2)
The second contribution above, which is related to double occupancy, cancels due to the
additional projection in the external electron operators, represented by the factors x1 xm+1
in Eq. (20). For this reason W drops out of the calculation of the Greens function, in
contrast to the calculation of the partition sum. This holds for the calculation of higher
correlation functions as well.
References
[1] Y. Vilk, A.-M.S. Tremblay, J. Phys. I France 7 (1997) 1309.
[2] P.W. Anderson, The Theory of Superconductivity in the High-Tc Cuprates, University Press,
Princeton, 1997.
[3] S. Pairault, D. Snchal, A.-M.S. Tremblay, Phys. Rev. Lett. 80 (1998) 5389, condmat/9905242.
[4] Ph. Brune, A.P. Kampf, cond-mat/0001210.
[5] M. Takahashi, J. Phys. C 10 (1977) 1289.
[6] E.O. Tngler, T. Kopp, Nucl. Phys. B 443 (1995) 516.
[7] S.E. Barnes, J. Phys. F 6 (1976) 1375;
S.E. Barnes, J. Phys. F 7 (1977) 2631.
[8] G. Kotliar, A.E. Ruckenstein, Phys. Rev. Lett. 57 (1986) 1362.
[9] R. Frsard, P. Wlfle, in: G. Baskaran, A.E. Ruckenstein, E. Tosatti, Y. Lu (Eds.), Proceedings
of the Adriatico Research Conference and Miniworkshop Strongly Correlated Electron
Systems III, Int. J. Mod. Phys. B 6 (1992) 685.
[10] T. Kopp, F.J. Seco, S. Schiller, P. Wlfle, Phys. Rev. B 38 (1988) 11835.
[11] W. Metzner, D. Vollhardt, Phys. Rev. Lett. 62 (1989) 324;
W. Metzner, Z. Phys. B 77 (1989) 253;
W. Metzner, D. Vollhardt, Phys. Rev. B 37 (1988) 7382.
[12] L. Lilly, A. Muramatsu, W. Hanke, Phys. Rev. Lett. 65 (1990) 1379.
[13] R. Frsard, M. Dzierzawa, P. Wlfle, Europhys. Lett. 15 (1991) 325.
[14] W. Zimmermann, R. Frsard, P. Wlfle, Phys. Rev. B 56 (1997) 10097.
[15] A. Tandon, Z. Wang, G. Kotliar, Phys. Rev. Lett. 83 (1999) 2046.
[16] G. Seibold, E. Sigmund, V. Hizhnyakov, Phys. Rev. B 57 (1998) 6937.
[17] R. Frsard, G. Kotliar, Phys. Rev. B 56 (1997) 12909.
[18] H. Hasegawa, J. Phys. Soc. Jpn. 66 (1997) 1391, cond-mat/0005271.
[19] M. Lavagna, Phys. Rev. B 41 (1990) 142.
[20] Y. Okimoto, T. Katsufuji, Y. Okada, T. Arima, Y. Tokura, Phys. Rev. B 51 (1995) 9581.
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
789
P. Coleman, Phys. Rev. B 35 (1987) 5072.

N. Read, D.M. Newns, J. Phys. C 16 (1983) 3273.
E. Arrigoni, C. Castellani, M. Grilli, R. Raimondi, G.C. Strinati, Phys. Rep. 241 (1994) 291.
L. Gehlhoff, R. Frsard, J. Phys: Cond. Matter 8 (1996) L13.
J. Kroha, P.J. Hirschfeld, K.A. Muttalib, P. Wlfle, Solid State Commun. 83 (1992) 1003;
J. Kroha, P. Wlfle, T.A. Costi, Phys. Rev. Lett. 79 (1997) 261;
J. Kroha, P. Wlfle, Acta Phys. Pol. B 29 (1998) 3781.
J.E. Hirsch, R.M. Fye, Phys. Rev. Lett. 56 (1986) 2521.
N.E. Bickers, Rev. Mod. Phys. 59 (1987) 845.
N.E. Bickers, Ph.D. Thesis, Cornell University, 1986.
P.W. Anderson, Phys. Rev. 110 (1958) 827;
P.W. Anderson, Phys. Rev. 112 (1958) 1900.
S.F. Edwards, Y.V. Gulyaev, Proc. Royal Phys. Soc. A 279 (1964) 229;
K. Ito, Mem. Am. Math. Soc. 4 (1951).
P. Coleman, Phys. Rev. B 29 (1984) 3035.
D. Vollhardt, N. Blmer, K. Held, M. Kollar, J. Schlipf, M. Ulmke, J. Wahle, Z. Phys. B 103
(1997) 283.
Th. Jolicur, J.C. Le Guillou, Phys. Rev. B 44 (1991) 2403.
Y. Bang, C. Castellani, M. Grilli, G. Kotliar, R. Raimondi, Z. Wang, in: G. Baskaran, A.E.
Ruckenstein, E. Tosatti, Y. Lu (Eds.), Proceedings of the Adriatico Research Conference and
Miniworkshop Strongly Correlated Electron Systems III, Int. J. Mod. Phys. B 6 (1992) 531.

CUMULATIVE AUTHOR INDEX B591B594
Affleck, I.
Ahn, C.
Akhmedov, E.T.
Armoni, A.
Arnowitt, R.
Asakawa, T.
B594 (2001) 535

B594 (2001) 660
B592 (2000) 234
B593 (2001) 229
B592 (2000) 143
B591 (2000) 611
Bagger, J.A.
Bakas, I.
Bandelloni, G.
Baseilhac, P.
Belitsky, A.V.
Beneke, M.
Beneke, M.
Berezhiani, Z.
Bialas, A.
Bialas, P.
Bigi, I.I.
Bilal, A.
Bill, M.
Bonneau, G.
Boos, H.E.
Borlaf, J.
Broadhurst, D.J.
Brodsky, S.J.
Bruckmann, F.
Buchalla, G.
Bunder, J.E.
Buras, A.J.
Burda, Z.
B594 (2001) 354

B593 (2001) 31
B594 (2001) 477
B594 (2001) 607
B593 (2001) 289
B591 (2000) 313
B592 (2000) 3
B594 (2001) 113
B593 (2001) 438
B592 (2000) 391
B592 (2000) 92
B593 (2001) 31
B591 (2000) 139
B593 (2001) 398
B592 (2001) 597
B593 (2001) 243
B592 (2000) 247
B593 (2001) 311
B593 (2001) 545
B591 (2000) 313
B592 (2001) 445
B592 (2000) 55
B592 (2000) 391
Capella, A.
Capitani, S.
Capitani, S.
Caprara, S.
Carena, M.
Casteill, P.-Y.
Castellani, C.
Catani, S.
Cervera, A.
Chalmers, G.
Chaudhuri, S.
B593 (2001) 336

B592 (2000) 183
B593 (2001) 183
B594 (2001) 747
B592 (2000) 164
B591 (2000) 491
B594 (2001) 747
B591 (2000) 435
B593 (2001) 731
B591 (2000) 39
B591 (2000) 243
Chen, B.
Chen, G.-H.
Chernodub, M.N.
Chizhov, M.V.
Chkareuli, J.L.
Choudhury, D.
Cleaver, G.B.
Costa, M.S.
B593 (2001) 505

B593 (2001) 562
B592 (2000) 107
B591 (2000) 457
B594 (2001) 23
B592 (2000) 35
B593 (2001) 471
B591 (2000) 469
Dalmazi, D.
Datta, A.
Del Debbio, L.
Derendinger, J.-P.
Di Castro, C.
Di Giacomo, A.
Dobado, A.
Dokshitzer, Yu.L.
Dolan, F.A.
Dolcini, F.
Donini, A.
Dorey, P.E.
Dotti, G.
Duan, Z.
Drr, S.
Dutta, B.
B592 (2001) 419

B592 (2000) 35
B594 (2001) 287
B593 (2001) 31
B594 (2001) 747
B594 (2001) 287
B592 (2000) 203
B593 (2001) 729
B593 (2001) 599
B592 (2001) 563
B593 (2001) 731
B594 (2001) 625
B591 (2000) 636
B592 (2000) 371
B594 (2001) 420
B592 (2000) 143
Ecker, G.
Ennes, I.P.
Ermolaev, B.I.
Evans, N.
B591 (2000) 419

B591 (2000) 195
B594 (2001) 71
B592 (2000) 129
Fabbri, D.
Faraggi, A.E.
Feldmann, Th.
Ferreiro, E.G.
Fioravanti, D.
Frste, S.
Forte, S.
Fr, P.
Frsard, R.
Froggatt, C.D.
B591 (2000) 139

B593 (2001) 471
B592 (2000) 3
B593 (2001) 336
B591 (2000) 685
B593 (2001) 127
B594 (2001) 46
B591 (2000) 139
B594 (2001) 769
B594 (2001) 23
Gabrielli, E.
Gambino, P.
B594 (2001) 3
B592 (2000) 55
792
Ganor, O.J.
Gavela, M.B.
Ghezelbash, A.M.
Gckeler, M.
Gogoladze, I.G.
Gomez Cdenas, J.J.
Gomis, J.
Gorbahn, M.
Govindarajan, S.
Grazzini, M.
Greco, M.
Groot Nibbelink, S.
Gubarev, F.V.
Guhr, T.
Gunion, J.F.
Gutierrez, T.D.
Gyulassy, M.
B591 (2000) 547

B593 (2001) 731
B592 (2000) 408
B593 (2001) 183
B594 (2001) 23
B593 (2001) 731
B591 (2000) 265
B592 (2000) 55
B593 (2001) 155
B591 (2000) 435
B594 (2001) 71
B594 (2001) 441
B592 (2000) 107
B593 (2001) 361
B591 (2000) 277
B591 (2000) 277
B594 (2001) 371
Han, T.
Hannah, T.
Harmark, T.
Heinzl, T.
Hernndez, P.
Honecker, G.
Horsley, R.
Huitu, K.
Hwang, D.S.
B593 (2001) 415

B593 (2001) 577
B593 (2001) 76
B593 (2001) 545
B593 (2001) 731
B593 (2001) 127
B593 (2001) 183
B592 (2000) 164
B593 (2001) 311
Isidori, G.
Itoyama, H.
B591 (2000) 419

B593 (2001) 505
Jger, S.
Jayaraman, T.
Jeng, M.
Jones, H.F.
B592 (2000) 55
B593 (2001) 155
B594 (2001) 685
B594 (2001) 518
Kaidalov, A.B.
Kataev, A.L.
Kehrein, S.
Khalil, S.
Kharraziha, H.
Kiem, Y.
Kim, J.E.
Kim, Y.J.
King, S.F.
Kirilova, D.P.
Kishimoto, I.
Klebanov, I.R.
Kniehl, B.A.
Kobakhidze, A.B.
Kobayashi, T.
Konechny, A.
Kopp, T.
Kugo, T.
Kyae, B.
B593 (2001) 336

B592 (2000) 247
B592 (2001) 512
B594 (2001) 3
B592 (2000) 321
B594 (2001) 169
B591 (2000) 587
B593 (2001) 415
B591 (2000) 3
B591 (2000) 457
B591 (2000) 611
B591 (2000) 26
B591 (2000) 296
B594 (2001) 23
B592 (2000) 164
B591 (2000) 667
B594 (2001) 769
B594 (2001) 301
B591 (2000) 587
Lazzarini, S.
Lechtenfeld, O.
Lee, H.M.
Lee, S.
Levai, P.
Linart, S.
Likhoded, A.
Lozano, C.
Lucenti, A.
Lucini, B.
Ludwig, A.W.W.
B594 (2001) 477

B591 (2000) 39
B591 (2000) 587
B594 (2001) 169
B594 (2001) 371
B592 (2001) 479
B593 (2001) 415
B591 (2000) 195
B593 (2001) 729
B594 (2001) 287
B594 (2001) 685
Ma, B.-Q.
Maccarone, A.
Magnea, L.
Magnea, L.
Mangazeev, V.V.
Marchesini, G.
Maroto, A.L.
Matsuo, T.
Maul, M.
Maxwell, C.J.
Mayr, P.
McKenzie, R.H.
McKeon, D.G.C.
Mehen, T.
Melnikov, K.
Mena, O.
Merlatti, P.
Mikhailov, A.Yu.
Mintchev, M.
Mohapatra, R.N.
Montorsi, A.
Mora, P.
Moroi, T.
Mller, D.
Mller, G.
Mller-Kirsten, H.J.W.
Murakami, K.
B593 (2001) 311

B594 (2001) 747
B593 (2001) 269
B594 (2001) 46
B592 (2001) 597
B593 (2001) 729
B592 (2000) 203
B593 (2001) 505
B594 (2001) 89
B592 (2000) 247
B593 (2001) 99
B592 (2001) 445
B591 (2000) 591
B591 (2000) 265
B591 (2000) 515
B593 (2001) 731
B591 (2000) 139
B591 (2000) 547
B592 (2000) 219
B593 (2001) 451
B592 (2001) 563
B594 (2001) 229
B594 (2001) 354
B593 (2001) 289
B591 (2000) 419
B594 (2001) 243
B593 (2001) 505
Naculich, S.G.
Nanopoulos, D.V.
Narison, S.
Naud, J.D.
Navelet, H.
Nekrasov, N.A.
Nepomechie, R.I.
Neubert, M.
Neufeld, H.
Niedermeier, L.
Niemeyer, B.
Nimai Singh, N.
Nyawelo, T.S.
B591 (2000) 195

B593 (2001) 471
B593 (2001) 3
B594 (2001) 713
B593 (2001) 438
B591 (2000) 26
B594 (2001) 660
B591 (2000) 313
B591 (2000) 419
B593 (2001) 289
B591 (2000) 39
B591 (2000) 3
B594 (2001) 441
Osborn, H.
Oshikawa, M.
B593 (2001) 599

B594 (2001) 535

Palisoc, C.P.
Pallante, E.
Paniak, L.D.
Park, D.K.
Park, J.
Parkin, P.
Parvizi, S.
Pasti, P.
Penati, S.
Prez-Lorenzana, A.
Perlt, H.
Perry, M.J.
Peschanski, R.
Petersson, B.
Petrini, M.
Piccione, A.
Pich, A.
Pich, A.
Pilch, K.
Pillin, M.
Pilo, L.
Polikarpov, M.I.
Polyakov, A.M.
Poppitz, E.
Pryadko, L.P.
B591 (2000) 296

B592 (2000) 294
B593 (2001) 671
B594 (2001) 243
B594 (2001) 169
B594 (2001) 518
B592 (2000) 408
B591 (2000) 109
B593 (2001) 651
B593 (2001) 451
B593 (2001) 183
B591 (2000) 469
B593 (2001) 438
B592 (2000) 391
B592 (2000) 129
B594 (2001) 46
B591 (2000) 419
B592 (2000) 294
B594 (2001) 209
B594 (2001) 625
B592 (2000) 219
B592 (2000) 107
B594 (2001) 272
B594 (2001) 354
B594 (2001) 713
Rakow, P.E.L.
Rasmussen, J.
Ridolfi, G.
Rigolin, S.
Rodrigues da Silva, P.S.
Rooman, M.
Rossi, A.
Ruelle, P.
Rychkov, V.S.
B593 (2001) 183

B593 (2001) 634
B594 (2001) 46
B593 (2001) 731
B592 (2000) 371
B594 (2001) 329
B594 (2001) 113
B592 (2001) 479
B594 (2001) 272
Sachrajda, C.T.
Salam, G.P.
Saleur, H.
Salgado, C.A.
Sannino, F.
Santambrogio, A.
Sarkar, T.
Saulina, N.
Schfer, A.
Schierholz, G.
Schiller, A.
Schmidt, I.
Schnitzer, H.J.
Schreyer, R.
Schwarz, A.
B591 (2000) 313

B593 (2001) 729
B594 (2001) 535
B593 (2001) 336
B592 (2000) 371
B593 (2001) 651
B593 (2001) 155
B591 (2000) 547
B593 (2001) 289
B593 (2001) 183
B593 (2001) 183
B593 (2001) 311
B591 (2000) 195
B593 (2001) 127
B591 (2000) 667
793
Sfetsos, K.
Shatashvili, S.L.
Silvestrini, L.
Simonov, Yu.A.
Sirlin, A.
Smolin, L.
Sondhi, S.L.
Sorokin, D.
Spindel, Ph.
Stanishkov, M.
Sugiyama, K.
Szabo, R.J.
B593 (2001) 31
B591 (2000) 26
B592 (2000) 55
B592 (2000) 350
B591 (2000) 296
B591 (2000) 227
B594 (2001) 713
B591 (2000) 109
B594 (2001) 329
B591 (2000) 685
B591 (2000) 701
B593 (2001) 671
Tabaczek, J.
Tamaryan, S.N.
Tateo, R.
Tonin, M.
Torrente-Lujan, E.
Troyan, S.I.
B592 (2000) 391

B594 (2001) 243
B594 (2001) 625
B591 (2000) 109
B594 (2001) 3
B594 (2001) 71
Uehara, S.
Uraltsev, N.G.
B591 (2000) 77
B592 (2000) 92
Valencia, G.
Valent, G.
van Holten, J.W.
van Ritbergen, T.
Vassilevich, D.V.
Vekua, T.
Verbaarschot, J.J.M.
Verhoeven, O.
Vitev, I.
Vogt, R.
B593 (2001) 415

B591 (2000) 491
B594 (2001) 441
B591 (2000) 515
B594 (2001) 501
B593 (2001) 545
B592 (2001) 419
B592 (2001) 479
B594 (2001) 371
B591 (2000) 277
Walker, J.W.
Warner, N.P.
Watts, G.M.T.
Wilke, T.
Wipf, A.
Wu, Y.-S.
B593 (2001) 471

B594 (2001) 209
B594 (2001) 625
B593 (2001) 361
B593 (2001) 545
B593 (2001) 562
Yamada, S.
Yamaguchi, S.
Yoshioka, K.
B591 (2000) 77
B594 (2001) 190
B594 (2001) 301
Zaffaroni, A.
Zakharov, V.I.
Zanon, D.
Zelnikov, A.
Zhang, J.-Z.
B591 (2000) 139

B592 (2000) 107
B593 (2001) 651
B594 (2001) 501
B594 (2001) 243

Nucl - Phys.B v.594

Diunggah oleh

Informasi Dokumen

Deskripsi Asli:

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Nucl - Phys.B v.594

Diunggah oleh

Hak Cipta:

Format Tersedia

Nuclear Physics B 594 (2001) 322

The b s decay in SUSY models

28049 Madrid, Spain

Received 16 June 2000; accepted 28 September 2000

E-mail addresses: emidio.gabrielli@cern.ch (E. Gabrielli), shaaban.khalil@uam.es (S. Khalil),

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

2. String inspired models with non-degenerate A-terms

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

(Au,d )ij = 3 m3/2 sin m3/2 cos (3 + ni + nj + nHu,d ),

xu,d xu,d yu,d

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

where for the up-sector

and for the down-sector

= Md2 RL = mD tan + SDL YdA? SD

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

3. The b s decay in SUSY models

In addition the ALEPH collaboration at LEP reported a compatible measurement of the

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

4. Numerical results and discussions

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

where mq is the average squark mass in the down sector and M 2

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

the sensitivity of the b s branching ratio to the non-universality parameters and i

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

E. Gabrielli et al. / Nuclear Physics B 594 (2001) 322

J.L. Hewett, Phys. Rev. Lett. 70 (1993) 1045.

Nuclear Physics B 594 (2001) 2345

From prototype SU(5) to realistic SU(7)

Received 24 March 2000; accepted 2 November 2000

E-mail address: c.froggatt@physics.gla.ac.uk (C.D. Froggatt).

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

Sc ) exchange [2], is largely

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

2. Motivations for SU(7) GUT

This condition uniquely leads to a missing VEV pattern of the type

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

which breaks the starting SU(N) symmetry to

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

of an SU(7) GUT symmetry (N k = 3, k/2 = 2); this SU(7) model is considered

S with the second adjoint is then forbidden by the above Z2

3. Basics of the SU(7) model

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

terms. In fact this orthogonality condition is precisely one

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

couplings inevitably appear as well. Similarly, in the general SU(N) case,

terms, destroying the missing VEV solution,

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

Sc6 , which are partners of the light

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

J.L. Chkareuli et al. / Nuclear Physics B 594 (2001) 2345

Here the parameters k (k = 1, 2, 3) include some known combination of the primary