Anda di halaman 1dari 496

Nuclear Physics B 613 (2001) 316

www.elsevier.com/locate/npe

A minimal S 1/(Z2 Z2 ) orbifold GUT


Arthur Hebecker a , John March-Russell a,b
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Theory Group, LBNL, University of California, Berkeley, CA 94720, USA

Received 28 June 2001; accepted 27 July 2001

Abstract
We investigate supersymmetric SU(5) grand unified theories (GUTs) realized in 5 space time
dimensions and broken down to the MSSM by SU(5)-violating boundary conditions on a S 1 /(Z2
Z2 ) orbifold with two 3-branes. The doublet-triplet splitting problem is entirely avoided by locating
the MSSM Higgs doublets on the brane on which SU(5) is not a good symmetry. An extremely
simple model is then described in which the MSSM matter is also located on this SU(5)-violating
brane. Although this model does not unify the MSSM matter within SU(5) multiplets, it explains
gauge coupling unification. A second model with MSSM matter in the SU(5)-symmetric bulk
preserves both the SU(5) explanation of fermion quantum numbers as well as gauge-coupling
unification. Both models naturally avoid problematic SU(5) predictions for the Yukawa couplings
of the first two generations and are consistent with proton decay constraints. We analyse the running
of gauge couplings above the compactification scale in terms of a 5d effective action and derive the
implications for the values of compactification scale, unification scale and of the scale at which the
bulk gauge theory becomes strongly coupled. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.10.Dm; 11.25.M; 12.60.Jv

1. Introduction
Since the pioneering work of Georgi and Glashow [1] (also [2]) the compelling concept
of a grand unified theory (GUT) of the standard model (SM) gauge interactions has
dominated our thinking about physics at high energies. The success of gauge coupling
unification [3] in minimal supersymmetric standard model (MSSM) extensions of these
theories [4,5] has further supported this idea [6]. However, despite these great successes
(and others, such as the generation of small neutrino masses) this GUT concept is not
without its faults. In particular we recall the problems of the Higgs structure of the highscale theory, especially the doublet-triplet splitting problem, the issue of too fast proton
decay, and the mismatch of the GUT scale with the naive scale of unification with gravity.
E-mail address: arthur.hebecker@cern.ch (A. Hebecker).
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 1 ) 0 0 3 7 4 - 1

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

Recently, a new possibility for the embedding of the SM into a form of GUT has been
suggested by Kawamura [79] and further extended by Altarelli and Feruglio [10] and
Hall and Nomura [11]. The basic idea is that the GUT gauge symmetry is realized in
5 or more spacetime dimensions and only broken down to the SM by utilizing GUTsymmetry violating boundary conditions on a singular orbifold compactification. Given
the success of the traditional supersymmetric gauge-coupling unification predictions,
the most attractive class of models are ones with both supersymmetry and (at least)
SU(5) gauge symmetry in 5 dimensions. In this case both the GUT group and 5d
supersymmetry are broken down to a N = 1 supersymmetric model with SM gauge group
by compactification on S 1 /(Z2 Z2 ) (a related idea was employed for EWSB in Ref. [12]).
This construction allows one to avoid some unsatisfactory features of conventional GUTs
with Higgs breaking, such as doublet-triplet splitting, while maintaining, at least at leading
order, the desired MSSM gauge coupling unification. Moreover, given the fact that string
theory requires additional dimensions, as well as branes located at orbifold fixed points,
the necessary (inverse-GUT-scale-sized) 5th dimension is not unreasonable. 1
In this paper we present two GUT models of this type, which we believe to be the
simplest yet discovered, our particular focus being the realization of the MSSM Higgs and
the MSSM matter. Specifically in Section 2 we review the basic physics of the construction
of Refs. [711], applying this in Section 2.1 to the gauge sector of the model. Section 2.2
contains the novel observation that the electroweak Higgs can be a localized field on
the SU(5)-violating brane, with no triplet partners whatsoever, thus solving the doublettriplet problem of traditional GUTs in a particularly simple way. Section 2.3 contains
a detailed discussion of two novel constructions of the MSSM matter in such theories.
The first of these two models locates the MSSM matter on the SU(5)-violating brane and
has both an exceptionally simple structure and is evidently consistent with low-energy
phenomenology. The model explains gauge coupling unification but, intriguingly, has no
SU(5) multiplet structure for the MSSM matter, so in itself does not explain the quantum
numbers of the MSSM matter. In our second model we present a novel way of realising
the MSSM spectrum from SU(5)-representations in the 5d bulk (thus explaining the matter
quantum numbers), which nevertheless leads to zero-modes which couple to the branelocalized Higgs in a way that does not imply the incorrect SU(5) relations for the Yukawa
matrices for the first two generations. Thus both models possess a number of appealing
features. In Section 3, we address the nature of gauge coupling unification in these models
(developing the treatment of [11]), with special emphasis on the corrections to exact
unification arising from SU(5)-violating brane kinetic terms. We analyse the running of
gauge couplings above the compactification scale in terms of a 5d effective action and
derive the implications for the values of the compactification scale, the unification scale
and of the scale at which the bulk gauge theory becomes strongly coupled. Most of the

1 We will take an effective-field-theory viewpoint of orbifold branes, namely we will impose all constraints

necessary for a consistent low-energy theory, but not require an explicit string theory (or other quantum gravity)
realisation.

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

discussion of Section 3 is generic and applies to other models with orbifold GUT breaking
as well. Section 4 contains our conclusions.

2. The structure of the minimal model(s)


Let us recall in greater detail the essential features of the S 1 /(Z2 Z2 ) model as
formulated in Refs. [8,10,11]. Even at this simple stage we will argue that there are more
possibilities for the construction of the theory than have been utilized so far, and that the
new model(s) we describe more minimally solve the problems of traditional 4d GUTs.
Consider a 5-dimensional factorized spacetime comprising a product of 4d Minkowski
space M4 (with coordinates x , = 0, . . . , 3), and the orbifold S 1 /Z2 Z2 , with
coordinate y x 5 . The circle S 1 will be taken to have radius R where 1/R MGUT
(so large extra dimensions are not being considered). The orbifold S 1 /Z2 is obtained by
modding out the theory by a Z2 transformation which imposes on fields which depend
upon the 5th coordinate the equivalence relation y y. Modding out the theory means
that we restrict the configuration space of fields in the functional integral to ones that are
invariant under the specified Z2 action. To obtain the orbifold S 1 /(Z2 Z2 ) we further mod
out by Z2 which imposes the equivalence relation y  y  , with y  y + R/2. Actually
this formulation of the S 1 /(Z2 Z2 ) orbifold slightly obscures one fact: The second
equivalence y  y  can be combined with the first, y y, to obtain the physically
identical equivalence y y + R, and this slight rewriting makes manifest that the second
equivalence introduces no new fixed points by itself. However, let us continue to work with
the basis of identifications
P : y y,

P  : y  y  .

(1)

It is important to the whole construction that under these equivalences there are two
inequivalent fixed 3-branes (or orbifold planes) located at y = 0, and y = R/2 ,
which we denote O and O  , respectively. It is consistent to work with the theory obtained
by truncating to the physically irreducible interval y [0, ] with the 3-branes at y = 0, 
acting as end-of-the-world branes, and henceforth we so do.
The action of the equivalences P , P  on the fields of a quantum field theory living on
M4 S 1 /(Z2 Z2 ) is not fully specified by the action Eq. (1) on the coordinates. One
must also define the action within the space of fields. To this end, let (x, y) be a vector
comprising all bulk fields, then the action of P and P  is given by
P : (x, y) P (x, y),


P : (x, y

) P (x, y  ).

(2)
(3)

Here, on the rhs, P and P are matrix representations of the two Z2 operator actions,
which necessarily have eigenvalues 1 when diagonalized. Let us from now on work in
this diagonal basis of fields, and classify the fields by their eigenvalues (1, 1). Then
these fields have the KK expansions:

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316


++ (x, y) =

+ (x, y) =

+ (x, y) =

(x, y) =

4 
1
2ny
(2n)

,
++ (x) cos

R
R
( 2) n,0

(4)

n=0

(2n+1)

(x) cos

(2n + 1)y
,
R

(5)

(2n+1)

(x) sin

(2n + 1)y
,
R

(6)

(2n + 2)y
4  (2n+2)
.
(x) sin
R
R

(7)

4
R
4
R

n=0

+
+

n=0

n=0

From the 4d perspective the KK fields (k) (x) acquire a mass k/R, so only the ++
possess a massless zero mode. Moreover, only ++ and + have non-zero values at
y = 0, while only ++ and + are non-vanishing at y = .
It is important to realize that the action of the identifications P , P  on the fields (namely
the matrices P and P ) can utilize all of the symmetries of the bulk theory. Thus P and P 
can involve gauge transformations, discrete parity transformations, and most importantly
in the supersymmetric case, R-symmetry transformations. This last feature allows one
to break the higher-dimensional supersymmetry to a phenomenologically acceptable, and
desirable N = 1 SUSY theory in 4d.
2.1. The gauge structure of the minimal model
To reproduce the good predictions of a minimal supersymmetric GUT, we start from a
5d SU(5) gauge theory with minimal SUSY in 5d (with 8 real supercharges, corresponding
to N = 2 SUSY in 4d). Thus, at minimum, the bulk must have the 5d vector superfield,
which in terms of 4d N = 1 SUSY language contains a vector supermultiplet V with
physical components A , , and a chiral multiplet with components , . Both V and
transform in the adjoint representation of SU(5).
Now choose the matrix representation of the parity assignments, expressed in the
fundamental representation of SU(5), to be
P = diag(+1, +1, +1, +1, +1),

P  = diag(1, 1, 1, +1, +1),

(8)

so that
P : V a (x, y)T a V a (x, y)P T a P 1 ,
P  : V a (x, y  )T a V a (x, y  )P  T a P 1 ,

(9)

where the action on the rhs is as matrices. For the same assignments are taken apart
from an overall sign for both P and P  , so that
P : a (x, y)T a a (x, y)P T a P 1 ,
P  : a (x, y)T a a (x, y)P  T a P 1 .

(10)
O

brane at
These boundary conditions then break SU(5) to the SM gauge group on the
y = , and 4d N = 2 SUSY to 4d N = 1 SUSY on both the O and O  branes. This can

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

Table 1
Parity assignment and KK masses of fields in the 4d vector and chiral adjoint supermultiplet. The
index a labels the unbroken SU(3) SU(2) U(1) generators of SU(5), while a labels the broken
generators
(P , P  )

4d superfield

4d mass

(+, +)
(+, )
(, +)
(, )

Va
V a
a
a

2n/R
(2n + 1)/R
(2n + 1)/R
(2n + 2)/R

be seen explicitly by examining the masses of the KK towers of the fields as displayed in
Table 1. Only the (+, +) fields possess massless zero modes, and at low energies the gauge
and gaugino content of the 4d N = 1 MSSM is apparent.
2.2. The electroweak Higgs
We now come to the first significant difference with respect to the treatments of Refs. [7
11]: our model does not include bulk Higgs multiplets. If we were to include such bulk
Higgs, then the unbroken SU(5) gauge invariance of the bulk would force them to transform
as 5s (and 5s) of SU(5). As such representations contain more than the SU(2) doublets
necessary for electroweak symmetry breaking, we are faced with a version of the infamous
doublet-triplet splitting problem. Of course a success of the models of Refs. [711] is that
by extending the action of P , P  to this bulk Higgs sector, the triplet components can
be made heavy. Nevertheless it is unarguably simpler (and phenomenologically different)
never to have had the Higgs triplet components in the first place!
How then are the Higgs doublets to be realized? To understand the possibilities it is
important to realize that after the orbifolding, Eqs. (8)(10), the amount of unbroken gauge
symmetry is position dependent. On the O brane at y = 0 and in the bulk the entire
SU(5) symmetry is a good symmetry. In particular at position y =  the allowed gauge


transformations are of the form U = exp(i a a (x, y)T a + i a a (x, y)T a ), with both
the gauge transformation parameters a (x, y) and a (x, y) non-vanishing. Only at the O 
brane are the a (x, ) = 0, and purely the SM gauge symmetry is defined. The implication
of this for the gauge sector of the theory is that the form of the 5d effective Lagrangian (at
scale ) is:
 


1
(y)
2
+
Leff () = d
tr W W
2g5 ()2 2g0 ()2

3

(y )

tr Wi W ,
(11)
2gi ()2
i=1

where W,i (i = 1, 2, 3) are the supersymmetric field strengths of the unbroken U(1)
SU(2) SU(3) SM gauge group on the O  brane, and W is the SU(5) field strength. In

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

other words, the effective action of the O  brane need only respect the SM gauge group,
while the bulk theory must respect the full SU(5) gauge symmetry.
The symmetries of the theory allow one to add interactions, and additional matter, on the
O  brane which only respect the SM gauge symmetry. Thus, unlike the models of Refs. [7
11], we choose to add a pair of weak-SU(2) doublet N = 1 chiral superfields Hu , Hd to
the O  3-brane theory at y = . This is a particularly natural and attractive definition of
the theory precisely because the Higgs is the one matter field in the MSSM that does not
have partners that effectively fill out a full SU(5) multiplet as far as quantum numbers are
concerned. As we do not have any Higgses which transform as a triplet of SU(3), there is
automatically no doublet-triplet-splitting problem either.
2.3. The MSSM matter
The next question is the location of the SM matter (and their 4d N = 1 superpartners).
In our framework it cannot be on the SU(5)-invariant O brane at y = 0 as it would not then
be able to interact with the electroweak Higgs. This leaves us with two possibilities, both
of which lead to attractive models.
I. SM matter on the SM brane
If, together with the Higgs, the matter is also located on the O  brane, then the Yukawa
couplings only need be invariant under the SM gauge group. Thus, the Yukawa couplings
are no more restricted than in the traditional 4d MSSM, and automatically there is enough
flexibility to accommodate fermion masses and mixings. Therefore the Yukawa terms in
the superpotential of the 5d effective Lagrangian for the light SM-charged fields of our
minimal theory are localized entirely on the O  brane at y = :



J + hd QI Hd D
J + hl LI Hd E
J ,
WYukawa = d 2 (y )
huIJ QI Hu U
IJ
IJ
IJ

(12)
where I, J = 1, 2, 3 are generation indices, and all fields depend only upon
This
theory has considerable advantages as compared to the set-ups of both Altarelli and
Feruglio [10] (AF) and Hall and Nomura [11] (HN). Specifically, the construction of AF
considers the Higgs to be a bulk field (which therefore must be extended to a full 5), and
locates the SM on the SU(5) invariant brane (the SM matter therefore transform as full
5 + 10 multiplets of SU(5)). On this brane AF then assume SU(5)-non-invariant Yukawa
interactions between the bulk Higgs and the SM matter. But such a set-up is not consistent
with unbroken gauge invariance on the O brane. In particular it does not allow the freedom
to move between unitary and covariant gauges to show the simultaneous unitarity and
Lorentz-invariance of the low-energy theory. (Lack of renormalizability, at least at high
scales, is not really the important issue after all the basic Kawamura set-up considers a
5d non-renormalizable theory. This is perfectly OK if the scale of the loss of calculability
is sufficiently high.) Indeed unless one includes further SU(5)-breaking mass terms for
the SU(5) gauge bosons, m2 Aa A,a (localized to the formerly SU(5)-invariant brane), the
theory is not unitary at low energy. The whole theory then has a rather different, and much
x.

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

more complicated character than that discussed by Altarelli and Feruglio. We therefore
consider this theory to be more complicated than necessary.
On the other hand the theory of HN, where again the Higgs is taken to be a bulk
field, and the SM matter is located on the SU(5)-invariant brane, assumes SU(5) invariant
Yukawa interactions localized to the SU(5)-invariant brane. This is a consistent set-up.
Unlike in our model where we place the SM matter on the O  brane, the HN model can
explain the b mass ratio [13] via SU(5) relations for the Yukawas. Unfortunately it also
leads to SU(5) predictions for the first and second generation fermion mass ratios, md /me
and ms /m , which are in strong disagreement with experiment. Thus HN are forced to
introduce additional fields in the bulk which mix with the first and second generations of
MSSM fields so as to correct the wrong predictions for the first two generations of quarks
and leptons. Although possible, this procedure is ad hoc.
Our model does not suffer from either of the above shortcomings. It is also physically
different in advantageous ways. First, since there are no Higgs triplets at all, it immediately
follows that there are no problems with triplet-higgsinomediated proton decay. (The
models of AF and HN, where triplet higgsinos exist as KK excitations, also avoid
dangerous higgsinomediated proton decay, but for different reasons.) There are no
dimension-5 proton decay mediating operators from higgsinos in our theory. Of course,
there are still proton-decay processes mediated by the heavy (+, ) KK mode X and Y
gauge bosons. These are present due to terms in the O  -brane action involving the (nonvanishing) 5 derivatives of Aa . (Note that such KK excitations in the X and Y gauge boson
directions carry hypercharge, which is correctly normalized because it sits inside the bulk
SU(5), so their interactions with the brane-localized SM states also correctly normalize the
hypercharge of these SM states.) Being dimension 6 operators these lead to much less of
a constraint on the effective GUT scale 1/R. In addition, however, there can in principle
be brane-localized dimension 5 baryon-number-violating operators from global-quantumnumber-violating quantum gravity effects [14]. Because of the existence of a relatively
large 5th dimension the quantum gravity scale is lower than the usual Planck mass, so these
operators, if they are there, are more problematic than usual. We leave a detailed discussion
of this issue to a future publication. Second, this model has several very nice features when
one considers the breaking of the remaining N = 1 supersymmetry down to the SM at
the TeV scale. An attractive solution of the supersymmetric FCNC problem follows if
the MSSM matter is sequestered from the SUSY-breaking sector. The localization of the
MSSM matter to the O  -brane naturally allows us to achieve this separation by having the
SUSY-breaking sector on the SU(5)-symmetric O-brane, and communicating the SUSY
breaking to the MSSM matter by the MSSM gauginos which live in the bulk. This is
similar but not identical to the models of [15] considered by HN [11] (see also the earlier
works [16]). For example, one difference arises from the fact that in our model the gauginos
necessarily pick up their soft SUSY-breaking mass from an SU(5)-symmetric interaction
(e.g., (S tr W W )|F cpt if there is a singlet chiral superfield S on O with FS  = 0).
Thus the gaugino masses will satisfy the traditional SUSY SU(5) relations to the same
high degree of accuracy as the gauge-couplings themselves unify. This and other details of
SUSY breaking in this (and the following) model will be discussed in a future publication.

10

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

An aesthetic disadvantage of this model is that it gives up on an explanation via SU(5)


multiplet structure of the quantum numbers of the SM matter fields. (The well-known
restrictions of anomaly cancellation for these quantum numbers of course still apply.)
Overall we find this model intriguing because of its extreme simplicity and because it
manages to unify the gauge couplings without a corresponding unification of the Higgs or
MSSM matter.
II. SM matter in the bulk
We now discuss an alternate model for the MSSM matter. The bulk of our theory
is invariant under both the full SU(5) gauge symmetry and the full N = 2 minimal
supersymmetry (8 real supercharges) of 5 dimensions. If we take the MSSM matter to
reside in the bulk, then the first of these symmetries immediately tells us that they must
come in complete SU(5) multiplets. Therefore the usual SU(5) matter multiplet structure,
10 + 5, is naturally incorporated, and the quantum numbers of the SM matter fields, in
particular the hypercharge assignments and the issue of charge quantization, are explained
for the same reasons as in traditional 4d GUTs with simple gauge group. In fact we will
soon argue that the correct situation with regard to quantum numbers is slightly more subtle
in a particularly interesting fashion.
On the other hand, the bulk N = 2 SUSY appears to pose a problem. The minimal
matter superfield representation for such a theory is a hypermultiplet, which in 4d N = 1
language decomposes in to a chiral multiplet R together with a mirror chiral multiplet in
the conjugate representation c . Thus, the choice of matter in the bulk would appear to
R
have problems reproducing the chiral structure of the SM.
Fortunately we are saved once again by the structure of the orbifold projections P and
P  acting on fields. The action of these projections on the N = 1 component fields and
c residing in a 5d hypermultiplet is inherited from the action on the 5d vector multiplet
Eqs. (8)(10). The result is that actions of both P and P  on the 4d chiral fields and c
have a relative sign:
P : P ,

P : c P c ,

(13)

and similarly for P  . This difference leads to a chiral spectrum for the zero modes. Indeed
the KK spectrum of 5d bulk hypermultiplets in the representation 10 + 5 (whose 4d chiral
+ Tc + F
c ) resulting from the P , P  actions is given in
components we denote T + F

+
Table 2. Note that since P does not commute with SU(5), components of the T + F
c fields in different SU(3) SU(2) U(1) representations have different KK mode
Tc+F
, D,
 L, E,
 etc., to indicate
structures. Thus in this table we label the expansions by Q, U
their SM transformation properties.
We immediately note from Table 2 that the zero mode structure (the n = 0 components
=
of the (+, +) fields) do not fill out a full generation of SM matter. We just get U

(3, 1)2/3 , E = (1, 1)1 , and L = (1, 2)1/2 left-chiral N = 1 superfields. This is rectified
by taking another copy of 10 + 5 in the bulk (with N = 1 chiral components denoted
c
T  + F + T c + F ), and using the freedom to flip the overall action of the P  parity on

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

11

Table 2
Parity assignments and KK masses of fields in the 4d chiral supermultiplets resulting from the
) representation. The subscript labels the SM
decomposition of 5d hypermultiplets in the (T + F
 = (3, 2)1/6 , U
 = (3, 1)2/3 , etc.
transformation properties, e.g., Q = (3, 2)1/6 , Q
(P , P  )

4d superfield

4d mass

(+, +)

L
TU
 , TE
, F

2n/R

(+, )


TQ , F
D

(2n + 1)/R

(, +)

c ,F
c
T
Q D
c
TUc , TEc , F
L

(2n + 1)/R

(, )

(2n + 2)/R

Table 3
Parity assignments and KK masses of fields in 4d chiral supermultiplets resulting from the
decomposition of 5d hypermultiplets in the (T  + F ) representation. The alteration of the action
of P  causes an effective interchange of the 1st and 2nd rows, and of the 3rd and 4th rows, relative
to Table 2
(P , P  )
(+, +)
(+, )
(, +)
(, )

4d superfield
 ,F

TQ

D
T  , T  , F L

 E

U
c
TUc , TEc , F L
c , F
 c
T
D
Q

4d mass
2n/R
(2n + 1)/R
(2n + 1)/R
(2n + 2)/R

+ T c + F
c . 2 This difference leads to a different
these multiplets by a sign relative to T + F
selection of zero mode components, the KK spectrum being given in Table 3.
Now a marvellous thing has happened! Combining the results of Tables 2 and 3 we have
zero modes which fill out the full matter content of a SM generation. Remarkably, this
occurs in such a way as to explain charge quantization and the hypercharge assignments
despite the fact that what we think of as a single generation filling out a 10 + 5, has
components that arise from different (10 + 5)s in the higher-dimension. This means that
when we couple three copies (for the three generations) of this combination of fields to
the Higgs on the O  brane at y = , the Yukawa couplings do not have to satisfy the usual
SU(5) relations. 3 Explicitly, the Yukawa couplings for the light zero-mode MSSM matter
fields now result from 3 different combinations of the 5d fields,





J
l
I J
huIJ TQI TUJ Hu + hdIJ TQI F D
WYukawa = d 2 (y )
 Hd + hI J TE
FL Hd ,
IJ

(14)

2 Hall and Nomura make this observation in the 2nd footnote of the final version of [11].
3 These results are reminiscent of those from studies of string theory carried out in the mid-to-late 1980s [17].

We also thank Lawrence Hall for discussions concerning Yukawa couplings.

12

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316


where we have indicated the zero mode components contained in the 5d T , T  , F and F
fields (using the same notation for the components as in Tables 2 and 3). As before, Hu
and Hd are 4d chiral fields localized to the O  brane at y = . Thus we have no problems
with the standard incorrect 1st and 2nd generation mass-ratio predictions of SU(5), and do
not require the contortions of other models to avoid these difficulties.
Having discussed how a successful understanding of quantum numbers and interactions
may be achieved, we now turn to the question of the nature of gauge coupling unification
in these models.

3. Gauge unification at tree level and beyond


At least in the domain of applicability of quantum (effective) field theory, the gauge
couplings in an S 1 /(Z2 Z2 ) orbifold GUT of the general type discussed here and in
Refs. [711] never exactly unify. The reason for this is that the theory on the GUT-breaking
brane O  never recovers the full SU(5) symmetry as the energy scale is increased. Its
symmetry remains limited to the SM gauge group. In particular, there are brane-localized
4d kinetic terms for the SM gauge fields with SU(5)-violating coefficients 1/gi2 (see
Eq. (11)). Nevertheless, a 4d observer performing experiments at or just below the energy
scale Mc = 1/R sees an approximate SU(5) unification of couplings. The reason for this
is that the bulk gauge kinetic term is SU(5)-symmetric and the wavefunction of the zeromode gauge boson is independent of y. Thus the physical gauge coupling, as measured
in such experiments, is obtained by integrating over the 5th dimension, and if the linear
extent of the 5th dimension is sufficiently large (we shall quantify this shortly), then the
SU(5)-violating brane kinetic terms are dominated by the bulk contribution, leading to an
approximate SU(5)-symmetry.
This can be made more explicit by developing the discussions of Hall and Nomura,
and Nomura, Smith and Weiner [11,18]. We begin by estimating the region of validity of
the 5d field-theoretic description. Disregarding, for the moment, the -function terms in
Eq. (11) and integrating over the fifth dimension, one finds that the 5d gauge coupling g5
2 = g 2 /. Let us first
is related to the 4d gauge couplings of the SM by g12 = g22 = g32 = gU
5
assume that this tree-level relation is approximately valid at scale 1/R. Since the couplings
gi run in a conventional way below that scale, this puts 1/R in the vicinity of MGUT
2 /4  1/25. We now show that the smallness of
1016 GeV. Phenomenologically, U  gU
this number allows a large region of validity of weakly coupled 5d gauge theory. By navedimensional analysis (NDA) (cf. Ref. [19]), the effective loop expansion parameter in D
dimensions is given by the reduced coupling g 2 = g 2 /(2D D/2 :(D/2)). This means,
in particular, that g 42 = g42 /(16 2 ) in 4d and g52 = g52 /(24 3 ) in 5d. The 5d gauge theory,
being non-renormalizable, has power-divergent loop corrections. With a cutoff , these
loop corrections are negligible as long as g 52  1. Thus, the 5d theory becomes strongly
interacting at the scale


12 1
1

R  103 R 1 .
2
U
g5

(15)

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

13

We see that the range of validity of weakly coupled 5d gauge theory can span up to three
orders of magnitude and may lead us directly to, e.g., string physics at the Planck scale.
We now turn to loop corrections to the values of gi at the scale Mc 1/R. One approach
is to integrate out the fifth dimension at the maximal scale M at which our model is valid
and to run the 4d effective theory down to Mc = 1/R in a conventional manner. (Note
that M can be significantly smaller than its obvious upper bound if, for example, the
internal structure of the brane is resolved at a lower energy.) The 4d gauge couplings are
obtained by applying gi2 = g52 / (for i = 1, 2, 3) at scale M and integrating out the KK
modes between M and Mc :


N
N


1
M
M
M
1
1
+ bi
ln
+ ci
ln
. (16)
i (Mc ) = U +
ai ln
2
Mc
(2n 1)Mc
2nMc
n=1

n=1

Here ai is the zero-mode contribution (corresponding to the usual MSSM running) while
bi and ci are the contributions of the higher KK modes. The masses of these modes are
even or odd multiples of Mc , depending on the boundary conditions at O and O  . The
upper limit of the sum is determined by the UV cutoff N M/2Mc . Rewriting Eq. (16) as



N
N


1
M
M
1
1
1
i (Mc ) = U +
ai ln
,
+ (bi + ci )
ln
bi
ln 1
2
Mc
2nMc
2n
n=1

n=1

(17)
it seems impossible to extract any useful information about low-energy gauge couplings
since the sum multiplying (bi + ci ) produces a power-like dependence on the cutoff M.
The reason is that a power-like cutoff dependence has an error which is not parametrically
smaller than the effect itself. Note however that this power-like behaviour is a manifestation
of the 5d SU(5)-symmetric bulk. Thus, even without a detailed calculation, we know the
sum (bi + ci ) to be independent of i. As a result, the physically important differences
ij (Mc ) i1 (Mc ) j1 (Mc ) are only logarithmically sensitive to the cutoff scale and
are therefore calculable:


1
1
M
.
ij (Mc ) =
(18)
aij + bij ln
2
2
Mc
Here aij = ai aj , bij = bi bj , and the sum multiplying bi in Eq. (17) has been
evaluated at leading logarithmic accuracy. The effect described by Eq. (18) has been called
differential running in [18], where it was discussed in the framework of GUT breaking
by an adjoint Higgs on the brane.
Let us now compare this to the running of the 5d effective action from M to Mc . Due to
unbroken gauge invariance in the bulk, the most general form of the effective Lagrangian
(disregarding SM matter) at scale (with M   Mc ) is given by Eq. (11). This
Lagrangian, which is obtained from the original Lagrangian Leff (M) by integrating out
all modes with momenta above , has to be evaluated on fields that are smooth on a
scale . Thus, the functions in Eq. (11) have to be understood as approximate functions,
localized on a scale .

14

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

An explicit calculation of Leff () on the basis of Leff (M) is not straightforward


since it requires a gauge-invariant separation of high- and low-momentum gauge fields.
Nevertheless, some information about Leff () can be obtained by deriving differences of
low-energy gauge couplings on the basis of Eq. (11). 4 For example, the analog of Eq. (18),
derived from Eq. (11), reads


1
1

.
aij + bij ln
ij (Mc ) = ij () +
(19)
2
2
Mc
At the same time, one can start from the original Lagrangian LM and derive


1
1
M
ij (Mc ) = ij (M) +
.
aij + bij ln
2
2
Mc

(20)

This implies



1
1
M
ij () = ij (M) +
aij + bij ln ,
2
2

(21)

thus determining the evolution of the SU(5)-breaking brane-terms in Eq. (11) with .
The differences ij (M) in Eq. (21) are parameters of the (for our purposes most
fundamental) action at scale M. Since Eq. (21) holds only in the leading logarithmic
approximation, we have to ascribe an uncertainty of approximately 1 to the expression
ln(M/). Equivalently, this uncertainty can be attributed to the value of ij (M). This
implies that it is unnatural for the quantity ij (M) (which corresponds to a coefficient of a
term in the effective Lagrangian at scale M) to be smaller than this renormalization scale
uncertainty. In the most optimistic scenario, where no new physics at scale M contributes
to ij (M), we have therefore to assume that


1
1
ij (M)
(22)
aij + bij ,
2
2
where means that an O(1) coefficient remains undetermined. 5
To develop some intuition for the numerical values of the relevant scales, focus first
on model I of Section 2.3. The relevant function coefficients are ai = (33/5, 1, 3)
(MSSM) and bi = (10, 6, 4), ci = (0, 4, 6) (KK modes) (cf. [11,18]). With
i1 (mZ ) = (59.0, 29.6, 8.4) and an effective SUSY breaking scale mZ , conventional
unification occurs at MGUT  2 1016 GeV. An obvious possibility is to have Mc only
slightly below MGUT , in which case unification (within the uncertainty specified by
Eq. (22)) occurs slightly above MGUT (since above Mc the ij run somewhat slower than
in the MSSM). In this scenario, even though is near the Planck scale, the scale M at
which couplings unify and our orbifold model breaks down is near Mc . In the opposite
4 A careful treatment using effective field theory with dimensional regularization is possible [20]. We thank
R. Rattazzi for discussions of this issue.
5 In Ref. [11] the size of these brane localized gauge-kinetic terms is estimated to be 1 1/(4 ) using NDA
i
and the assumption that the 4d gauge theory on the boundary becomes strongly interacting at the same scale as
the 5d bulk theory. Our result, which is similar numerically, does not rely on this assumption and is also valid in
the case of a weakly coupled 4d theory.

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

15

extreme, one can have Mc  2 1014 GeV and unification at M  2 1017 GeV. In this
scenario, the 5d orbifold model is valid over 3 orders of magnitude and the 5d Planck
scale is reached. In model II of Section 2.3, the zero-mode ai coefficients are identical to
model I, while the non-zero KK mode function coefficients are bi = (2, 6, 8), and ci =
(12, 8, 6). Note that not only is the difference aij = ai aj the same for the two models,
but so is bij = bi bj , namely (b12 , b23, b31 ) = (4, 2, 6). The reason for this is that the
alteration between the two models is that model II has, effectively, at every non-zero KK
level a number of complete SU(5) multiplets (as can be seen by combining the quantum
numbers at each non-zero-mode mass level from Tables 2 and 3). Thus the differences bij
are determined solely by the gauge structure of the bulk which is unchanged. Therefore,
apart from a somewhat reduced cutoff scale due to the faster SU(5)-invariant running of
the bulk theory, the two models are very similar in this regard.
Summarizing, the primary phenomenological result of this section is that in both
models I and II the uncertainty in the gauge coupling unification (quantified by Eq. (22))
is comparable to the GUT scale threshold corrections of a conventional GUT.

4. Conclusions
In this paper we have investigated supersymmetric SU(5) grand-unified theories (GUTs)
realized in 5 spacetime dimensions and broken down to the MSSM by SU(5)-violating
boundary conditions on a S 1 /(Z2 Z2 ) orbifold with two end-of-the-world 3-branes.
Because of the position dependence of the amount of gauge symmetry, the MSSM Higgs
doublets can be located on the brane on which SU(5) is not a good symmetry. Thus SU(3)
triplet partners are never required, and the doublet-triplet splitting problem is entirely
avoided. In Section 2.3 we constructed an extremely simple model in which the MSSM
matter is also located on this SU(5)-violating brane. In this case, despite the loss of the
SU(5) understanding of the multiplet structure of the MSSM matter, one maintains gauge
coupling unification up to small corrections. As we discuss in Section 3, this is due to the
y-independence of the SM gauge boson zero modes, which allow the contribution of the
SU(5)-symmetric bulk gauge kinetic term to dominate that of the SU(5)-violating branelocalized kinetic terms. We find it intriguing that such a simple model is consistent. In
Section 2.3 we also constructed a second model, this time with MSSM matter located
in the SU(5)-symmetric bulk. The model is slightly more involved, but preserves the
SU(5) predictions for fermion quantum numbers as well as gauge-coupling unification.
Both models naturally avoid problematic SU(5) predictions for the Yukawa couplings
of the first two generations. Finally, in Section 3, we analysed the running of the gauge
couplings above the compactification scale in terms of a 5d effective action and derived
the implications for the values of the compactification scale, the unification scale and of
the scale at which the bulk gauge theory becomes strongly coupled, finding values that are
phenomenologically attractive.

16

A. Hebecker, J. March-Russell / Nuclear Physics B 613 (2001) 316

Acknowledgements
We are very grateful to Guido Altarelli, Wilfried Buchmller, Lawrence Hall, and
Riccardo Rattazzi for helpful conversations. J.M.R. thanks the members of the Theory
Group, LBNL, for their kind hospitality during the completion of this work.

References
[1] H. Georgi, S. Glashow, Phys. Rev. Lett. 32 (1974) 438;
H. Georgi, Unified gauge theories, in: Proc. Coral Gables 1975, Theories and Experiments In
High Energy Physics, New York, 1975.
[2] J. Pati, A. Salam, Phys. Rev. D 8 (1973) 1240;
J. Pati, A. Salam, Phys. Rev. D 10 (1974) 275.
[3] H. Georgi, H. Quinn, S. Weinberg, Phys. Rev. Lett. 33 (1974) 451.
[4] S. Dimopoulos, S. Raby, F. Wilczek, Phys. Rev. D 24 (1981) 1681.
[5] S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150.
[6] N. Sakai, Z. Phys. C 11 (1981) 153;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 105 (1981) 439;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 110 (1982) 215;
M.B. Einhorn, D.R.T. Jones, Nucl. Phys. B 196 (1982) 475;
W.J. Marciano, G. Senjanovic, Phys. Rev. D 25 (1982) 3092;
P. Langacker, M.-X. Luo, Phys. Rev. D 44 (1991) 817;
J. Ellis, S. Kelley, D.V. Nanopoulos, Phys. Lett. B 260 (1991) 131;
U. Amaldi, W. de Boer, H. Furstenau, Phys. Lett. B 260 (1991) 447.
[7] Y. Kawamura, Prog. Theor. Phys. 103 (2000) 613, hep-ph/9902423.
[8] Y. Kawamura, hep-ph/0012125.
[9] Y. Kawamura, hep-ph/0012352.
[10] G. Altarelli, F. Feruglio, hep-ph/0102301.
[11] L. Hall, Y. Nomura, hep-ph/0103125.
[12] R. Barbieri, L. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[13] M.S. Chanowitz, J. Ellis, M.K. Gaillard, Nucl. Phys. B 128 (1977) 506;
A.J. Buras, J. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 135 (1978) 66.
[14] M. Kamionkowski, J. March-Russell, Phys. Lett. B 282 (1992) 137, hep-th/9202003;
M. Kamionkowski, J. March-Russell, Phys. Rev. Lett. 69 (1992) 1485, hep-th/9201063;
R. Holman et al., Phys. Lett. B 282 (1992) 132, hep-ph/9203206;
R. Holman et al., Phys. Rev. Lett. 69 (1992) 1489.
[15] D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko et al., JHEP 0001 (2000) 003, hep-ph/9911323.
[16] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155;
Z. Chacko et al., JHEP 0001 (2000) 003, hep-ph/9911323.
[17] For early works see, for example, P. Candelas et al., Nucl. Phys. B 258 (1985) 46;
E. Witten, Nucl. Phys. B 258 (1985) 75.
[18] Y. Nomura, D. Smith, N. Weiner, hep-ph/0104041.
[19] Z. Chacko, M.A. Luty, E. Ponton, JHEP 0007 (2000) 036, hep-ph/9909248.
[20] R. Contino, L. Pilo, R. Rattazzi, E. Trincherini, hep-ph/0108102.

Nuclear Physics B 613 (2001) 1733


www.elsevier.com/locate/npe

A theorem on the power of supersymmetry


in Matrix theory
Y. Kazama, T. Muramatsu
Institute of Physics, University of Tokyo, Komaba, Meguro-ku, Tokyo 153-8902, Japan
Received 11 July 2001; accepted 14 August 2001

Abstract
For the so-called source-probe configuration in Matrix theory, we prove the following theorem
concerning the power of supersymmetry (SUSY): let be a quantum-corrected effective
SUSY

transformation operator expandable in powers of the coupling constant g as = n0 g 2n (n) ,
where (0) is of the tree-level form. Then, apart from an overall constant, the SUSY Ward identity
= 0 determines the off-shell effective action uniquely to arbitrary order of perturbation theory,
provided that the SO(9) symmetry is preserved. Our proof depends only on the properties of the treelevel SUSY transformation laws and does not require the detailed knowledge of quantum corrections.
2001 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.30.Pb

1. Introduction
Ordinarily, the role of a global symmetry with a finite number of parameters is to
a large extent kinematical. It provides conservation laws for the currents and charges,
relates the Greens functions, restricts the possible form of counter terms and so on, but
does not control the dynamics completely. The associated Ward identity is of an integrated
form unlike the more powerful SchwingerDyson equation and from group-theoretic point
of view it only organizes a set of terms into invariants for which independent coupling
constants can be assigned. Especially, when one deals with the effective action, which can
contain arbitrary number of derivatives of fields, there may be many such invariants and
global symmetries are not expected to be able to fix their relative strengths.
Despite these well-known facts, certain theories with a large number of global
supersymmetries (SUSY) might prove to be exceptions under some favorable conditions.
Notably, for the so-called Matrix theory for M-theory, proposed by Banks, Fischler,
E-mail addresses: kazama@hep3.c.u-tokyo.ac.jp (Y. Kazama), tetsu@hep1.c.u-tokyo.ac.jp (T. Muramatsu).
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 1 ) 0 0 3 9 6 - 0

18

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

Shenker and Susskind [1,2], evidence has been accumulating [310] that its high degree of
SUSY, namely, with the maximally allowed 16 supercharges, may be powerful enough to
determine the effective action of a D-particle in interaction with an aggregate of coincident
source D-particles, i.e., for the so-called source-probe configuration. Judged from our
common wisdom this may sound rather surprising but on the other hand impressive
agreement [1114] with 11-dimensional supergravity lends support to such a conjecture,
as the latter is considered to be unique.
A clear-cut settlement of this issue is hampered by the lack of unconstrained superfield
formalism for this system. One is forced to deal with the component formalism, where
SUSY is intertwined with gauge (BRST) symmetry and its algebra does not close without
the use of the full equations of motion. A related difficulty is that the SUSY transformation
laws for the effective action receive complicated quantum corrections and one cannot easily
identify possible SUSY invariants.
In order to gain insight into the issue of the power of SUSY in Matrix theory, in a
series of papers [9,10,15] we have (i) derived the SUSY Ward identity for the background
gauge in the form where the quantum-corrected SUSY transformation laws can be read
off in closed forms, (ii) computed the effective action and the SUSY transformation
operator explicitly for arbitrary off-shell trajectory of the probe D-particle including all
the spin effects at 1-loop at order 4 in the derivative expansion, 1 (iii) checked that they
indeed satisfy the SUSY Ward identity = 0, and (iv) finally demonstrated that, given
such , the solution to the Ward identity regarded as a functional differential equation
for is unique to the order specified above. All of these calculations were extremely
cumbersome and elaborate codes had to be developed for the algebraic manipulation
program M ATHEMATICA, including a new fast algorithm for generating SO(9) Fierz
identities of considerable complexity. Although restricted to the source-probe situation, the
fact that the fully off-shell effective action was uniquely determined from the knowledge
of the SUSY transformation laws was remarkable and strongly suggested that this feature
would persist to higher orders.
Indeed our result was not an accident that occurred at a particular low order. In
this article, we shall be able to prove the following theorem valid for the source-probe
configuration:
Theorem. Let be a quantum-corrected effective SUSY transformation operator expand
able in powers of the coupling constant g as = n0 g 2n (n) , where (0) is of the treelevel form. Then, apart from an overall constant, the SUSY Ward identity = 0 determines the off-shell effective action uniquely to arbitrary order of perturbation theory,
provided that the SO(9) symmetry is preserved.
Actually the main effort will be devoted to proving the following proposition, from
which our theorem follows straightforwardly:
1 As usual, the concept of order in the derivative expansion is defined as the number of derivatives plus half
the number of fermions.

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

19

Proposition. Let (0) be the tree level SUSY transformation operator. Then assuming that
SO(9) is a good symmetry, the solution to the functional differential equation (0) = 0 is
unique up to an overall constant, and is given by the tree level action.
We wish to emphasize that our proof of the Proposition and hence the Theorem will
depend only on the properties of the tree-level SUSY transformation laws and will not
require the detailed knowledge of quantum corrections.
To avoid any possible misconception, however, let us state clearly that our theorem does
not yet prove that supersymmetry (together with SO(9) symmetry) completely determines
the source-probe dynamics in Matrix theory. What it states is that given an appropriate ,
which, for example, can be computed independently of by the method developed in [9]
in some gauge, the Ward identity fixes completely. In order to claim that SUSY fully
determines the dynamics, one must show that it is capable of determining both and
simultaneously in a self-consistent manner, up to field redefinitions, without assuming the
knowledge of the underlying action of the Matrix theory. How this more ambitious program
should be formulated and attempted will be discussed in the final section.
Nevertheless, we believe that our theorem discloses another aspect of remarkably
powerful features of maximal supersymmetry: it clears a highly nontrivial necessary
requirement for SUSY to be able to determine everything, under certain conditions.
The rest of the article is organized as follows: in Section 2, after setting up our problem,
we derive what will be called the basic equation on which the proof of our Proposition
will be based. It is a reformulation
of the consistency
or integrability condition for the
 (0)
(0)
(0) (0) 
tree-type Ward identity, namely,

= 0. Although it provides only a
necessary condition, it will prove to be sufficiently restrictive. Proofs of the Proposition
and the Theorem will be given in Section 3. As the proofs are somewhat involved, we will
first illustrate our basic ideas using a simple example. These ideas are then sharpened and
extended into a number of Lemmas which apply for general situation. These Lemmas step
by step reduce the possible solutions to be of more and more restricted form and lead to the
proof of the Proposition. This in turn immediately yields our main Theorem, by a recursive
argument with respect to the powers of the coupling constant. Finally, in Section 4 we will
discuss and indicate the line of attack for the important remaining problems, namely, the
investigation of the multi-body case and of the possibility of complete determination of the
dynamics by SUSY.

2. Derivation of the basic equation


2.1. Preliminaries
Let us begin by briefly recalling the relevant features of Matrix theory and set up the
problem.
The classical action for the U (N + 1) Matrix theory in the Euclidean formulation is
given by

20

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

1
g2
[D , Xm ]2 [Xm , Xn ]2
2
4

1 T
1
T m
+ [D , ] g [Xm , ] ,
2
2
D igA.
S0 = Tr

(2.1)
(2.2)

In this expression, Xijm ( ), Aij ( ) and ,ij ( ) are the (N + 1) (N + 1) Hermitian matrix
fields, representing the bosonic part of the D-particles, the gauge fields, and the fermionic
part of the D-particles, respectively. m are the real symmetric 16 16 SO(9) -matrices,
and the vector index m runs from 1 to 9.
This action is known to possess a number of important symmetries. Among them, the
main focus in this article will be the supersymmetry, which carries 16 spinorial parameters

and transforms the basic fields in the manner

A =
T ,

Xm = i
T m ,


g
m
mn

= i [D , Xm ] + [Xm , Xn ]

.
2

(2.3)
(2.4)

Although the algebra closes only on-shell up to field-dependent gauge transformations,


S0 itself is invariant without the use of equations of motion, i.e., off-shell.
We will also make use of the SO(9) symmetry throughout for restricting the forms of
possible terms to be considered. Its role, however, is relatively minor as we will not need to
resort to a host of Fierz identities of considerable complexity, which were heavily utilized
in our previous works.
In this article, we shall concentrate on the so-called source-probe situation, namely,
the configuration of a probe D-particle interacting with a large number, N , of the source
D-particles all sitting at the origin. This is expressed by the splitting
Xm ( ) =

1
Bm ( ) + Ym ( ),
g
N

( ) =




Bm ( ) = diag rm ( ), 0, 0, . . . , 0 ,

1
( ) + ( ),
g

(2.5)
N




( ) = diag ( ), 0, 0, . . . , 0 ,

(2.6)

where Bm ( ) and ( ) are the bosonic and the fermionic backgrounds expressing the
positions and the spin degrees of freedom of the D-particles respectively and Ym ( ) and
( ) denote the quantum fluctuations around them. We will be interested in the general
case where rm ( ) and ( ) are arbitrary functions of not satisfying equations of motion.
At the tree level, the action and the SUSY transformations for the probe D-particle are
of the form

 2

r

(0)
= d
(2.7)
,
+
2g 2 2g 2
1

(0) rm =
m ,
(2.8)
i

(0) = i(/r
) ,
(2.9)

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

21

where the dot means differentiation with respect to and stands for , etc. After a
gauge is fixed and the functional integral over the fluctuations is performed, one obtains
the quantum effective action

= d L[rm , ].
(2.10)
Although a general argument does not exist, explicit calculations performed so far in the
so-called background gauge strongly indicate that L[rm , ] is a sum of local products
of derivatives of rm ( ) and ( ). As it will be emphasized again at the beginning of
Section 3, our proof of the Theorem and the Proposition will operate at each order in g 2
and in the derivative expansion, so that effectively we will be dealing with a finite number
of local expressions at each step regardless of whether the entire L is fully local or not.
satisfies the SUSY Ward identity

= 0,

(2.11)

where
is the quantum-corrected effective SUSY transformation operator. In a previous
article [9], we have derived the explicit form of this Ward identity in the background gauge,
where
is given in closed form in terms of expectation values of certain products of fields.
Further in [15] a complete calculation of and
at 1-loop at order 4 in the derivative
expansion was performed and the validity of (2.11) was explicitly verified at that order.
As was stated in the introduction, an interesting question is, for a given
how unique
is, without the use of the equations of motion. For the simplest case, namely at 1-loop
at order 2, it was found that is completely determined by
[9]. Moreover, a highly
nontrivial calculation at order 4 again showed that is unique [10]. In fact, these results
were not accidental: we shall prove that this feature persists to an arbitrary order of
perturbation theory.
2.2. Consistency equation
Since our main Theorem can be shown to follow rather easily from the Proposition, we
will first concentrate on the proof of the latter.
(0)
What we wish to show is that the solution to the equation
= 0 is unique up
(0)
to an overall constant, where the action of
is as defined in (2.8) and (2.9). This
equation, despite its simple appearance, is a highly complicated partial differential equation
involving Grassmann variables for the integrand L of . In fact, written out in full, it reads




+
(0) ( )

(0) = d
(0) rm ( )
(2.12)
= 0,
rm ( )
( )
where, in terms of L, the functional derivative /rm means (suppressing -dependence)

= D[L, rm ],
rm
L
L
L
D[L, rm ]

+ 2
+ ,
rm
rm
rm

(2.13)
(2.14)

22

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

and similarly for / . Later we will often use the symbol DN [L, rm ] to denote
(N)
(N)
D[L, rm ] truncated at (and including) (1)N N (L/rm
), where rm
is the N th
derivative of rm with respect to .
Our basic idea for analyzing (2.12) is to look at its consistency or integrability
equation, namely,

 (0) (0)


(0) (0) = 0.
(2.15)
Although this only gives a necessary condition, it will prove to be powerful enough to fix
the fermionic part of completely. The original equation
(0) = 0 then immediately
determines the bosonic part as well.
Now computing the left hand side of (2.15) by using the functional derivative as in
(2.12), one finds that the parts containing the second functional derivatives of cancel
and the result is

 (0) (0)


(0) (0)



(0)
(0)



= d ,
(0) rm ( )
(2.16)
+ ,
(0) ( )
.
rm ( )
( )


The closure of the algebra (0) ,
(0) is easily calculated. On rm it simply gives the usual
time-translation
(0) (0)
drm
,
,
rm = 2(
)
(2.17)
d
while on one finds
(0) (0)
,
= (m )(m
) (
m )(m ) .
(2.18)
By using a well-known SO(9) -matrix identity
m m

+ (cyclic in , , ) = + (cyclic in , , ),

it can be rewritten as
(0) (0)
d
,
= 2(
) + (,
),
d
where


(,
) m
(m ) +
( ) (
) + (
) .

(2.19)

(2.20)

(2.21)

Let us put these results back into (2.16). Then, by using the definition of the functional
derivative (2.14) and integration by parts, one finds that the contributions from the
canonical time-translation part of the closure relation add up to an integral of a total
derivative as




d ( )
dL
drm ( )
+
= 2(
) d
2(
) d
(2.22)
d rm ( )
d ( )
d
and vanish. Thus, the consistency equation is reduced to

d (,
)F = 0,

(2.23)

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

23

where for simplicity we defined F / . This notation will be used throughout the
rest of the article. Further, since this equation must hold for arbitrary and
, we may
remove these global spinor parameters and obtain

d F = 0,
(2.24)
m
( m F ) + F + F ( F ).
F

(2.25)

There are two ways that this equation may be satisfied; either (a) F vanishes identically
or (b) F is a total derivative.
Let us begin with case (a). First, take the trace over the index set (, ). Since m
is traceless, we immediately get F = 0. Putting this back into the equation, we have
m ( F ) + F + F = 0. Now contract this with n . Then, using Tr n m =

n F = 0. In this way, the consistency equation is reduced to the following


16 mn , we find
simple form:
F + F = 0.

(2.26)

n F = 0, as well.
Note that this guarantees the previous two relations, F =
There are two types of possible solutions to this algebraic equation. One is of the form
F = G,

(2.27)

where G is some SO(9) scalar. Due to the Grassmann nature of , this obviously satisfies
(2.26). The other is the case where F contains the product of 16 s, namely,

16


(2.28)

=1

which is easily checked to be SO(9) invariant. In this case, F vanishes identically for
any and and hence (2.26) is trivially satisfied.
Now let us turn to case (b), for which F = dG /d for some nonconstant G .
Going through exactly the same procedure as in case (a), we find

dG
,
(2.29)
d
m
 = G 1 Tr G 1
G
(2.30)
Tr(m G).
14
14
d
( F + F ) ( F + F ), we must demand that the
Rewriting F + F = d

expression F + F either (b-1) vanish identically or (b-2) be a total derivative itself.


Let us quickly dispose of the possibility (b-1). Since F , being a time derivative, cannot


be proportional to the product 16


=1 , the case (b-1) can occur only if F = Z for
some Z. But clearly it is impossible to produce the structure like Z by a time derivative
of a local 2 expression. Hence case (b-1) is excluded.
F + F =

2 Remember that implicitly we are dealing with expressions order by order in the double sense and hence
locality argument is valid.

24

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

Now consider case (b-2). For this to be possible at all, F must contain the derivative
of , namely, it must be of the form F = f for some SO(9) invariant f . Then, we
have F + F = ( + )f . As this does not contain f, needed to form a
total derivative, f can only be a constant. This reduces the possibility to the expression
+ . But this is clearly not a total derivative and hence case (b-2) is also
excluded.
Summarizing, we have shown that only case (a) can be possible and the consistency
equation is reduced to

= D[L, ] = G + H ,

(2.31)

where is as defined in (2.28) and G and H are, respectively, an SO(9) scalar


(0)
and a spinor. Note that compared to the original equation
= 0, which contains
functional derivatives with respect to both rm and , the consistency equation above
only involves the one with respect to the latter and hence is more tractable. Hereafter we
shall call (2.31) our basic equation. In the next section, we shall solve this basic equation
completely to prove our Proposition, which will then straightforwardly lead to our main
Theorem.

3. Proof of the theorem


From the preceding analysis, we have learned that as a necessary condition a solution
to the tree-type Ward identity
(0) = 0 must satisfy the basic equation (2.31). Below
we shall prove that, up to an overall constant and -independent structures, the solution to

(2.31) is unique and is given by d , which is nothing but the fermionic part of the tree
level action (0) . Then imposition of
(0) = 0 immediately fixes the -independent part
to coincide with the bosonic part of (0) .
Since the proof is rather involved, we shall first illustrate our basic strategy by treating
a simple example. Then, we shall develop several useful lemmas, which are basically the
generalizations of the logic used for that example. Finally with the aid of these lemmas the
proof of the Proposition and the Theorem will be given.
Before we begin our proof, let us make some important remarks. (i) Since we shall work
to an arbitrary but finite order in g 2 and at each such order the space of solutions can
be further classified by the order in the sense of derivative expansion, we may assume,
without loss of generality, that L consists of a finite number of local expressions. In
particular, this implies that the number of -derivatives acting on has a maximum
in L. Hereafter we shall refer to such a spinor with the largest number of derivatives as
SPLD. (ii) As the actual proof does not depend on which order of g 2 one is working, we
shall not display g 2 -dependence. (iii) Lastly, SO(9) invariance will be taken for granted
throughout.

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

25

3.1. Illustration of a simple case


Consider the simple case where L consists of bilinears of and its derivatives up to
only. The general form of such L is
1
1
1
+ D + E + F ,
L = A + B + C
(3.1)
2
2
2
where A F are bispinors made out of rm , its derivatives, SO(9) -matrices and the unit
matrix. Clearly, A, C and F must be antisymmetric. In this case, the special product
cannot occur and the equation to be solved is of the form
D2 [L, ] = G,

(3.2)
2 (L/ ).

where D2 [L, ] is given by L/ (L/ ) +


We now proceed in
steps.
1. First, focus on the F term, containing more than one SPLD ( in this example).
In D2 [L; ], look at the term containing (4) . Such a term can only be produced from the
F term through 2 (L/ ) and it cannot be canceled by anything else. Hence F must
vanish.
2. Next, consider the term which is linear in and contains a spinor with one less
derivative, namely, . In this example it is the E term. Look at the terms containing (3)
in D2 [L; ]. They come from (L/ ) and from 2 (L/ ) and add up to give
(E + E T ) (3) . As this must vanish, E must be antisymmetric. Then, the E term can
be rewritten as
+ E T E .
E = ( E )

(3.3)

On the right-hand side (RHS), the first term is a total derivative and can be dropped, while
due to antisymmetry the second term is actually the negative of the left-hand side (LHS).
Thus we get
1

E = E .
(3.4)
2
This does not contain any more and can be absorbed in the C term.
3. Having disposed of the E term, look at the remaining term with , namely, the
D term. It can be rewritten as
D = ( D ) D D .

(3.5)

Thus, arises only from the total derivative term, which we may discard, and the D term
can be absorbed by the other terms with less number of derivatives on the spinors. At this
stage all the terms containing SPLD have disappeared.
4. Next consider the C term, with two powers of , and look for terms with in
D2 [L; ]. It is produced only from (L/ ) and is given by C . This must vanish
and hence C = 0.
5. We are now left with L = (1/2) A + B . The B term can be rewritten as

B = ( B ) + B T B.

(3.6)

26

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

Contributions from the symmetric part of B, to be denoted by BS , cancel between LHS


and RHS. Writing the antisymmetric part as BA , we get

1
( BA ) B A .
(3.7)
2
Thus, dropping the total derivative, this can be absorbed by the A term. So we only need
to keep BS , and L is reduced to (1/2) A + BS .
6. Finally, look at the basic equation (3.2). It becomes
BA =

(A ) + 2(BS ) + (B S ) = G.

(3.8)

Since A is antisymmetric whereas BS is symmetric, the terms proportional to must


vanish separately. Thus, we find A = 0 and BS = constant. Now, since the only symmetric
constant bispinor is , we must have BS = c , with c a constant. This in turn dictates
G = 2c. In this way, we find the unique solution
L = c( ).

(3.9)

This completes the proof for this simple case.


3.2. Useful lemmas
In this subsection, we shall prove a number of useful lemmas to pave the way for
the proof of the Proposition. The proofs of these lemmas may look rather involved but
essentially they are refined generalization of the logic used for the simple case described
above.
The first lemma eliminates the possibility of the appearance of the special product
under certain conditions.
Lemma 1. Let (N) be the SPLD in L. Then for N  2, the special product cannot
appear in L.
Proof. The product can be written as
= 1 2 16 E1 2 16 ,
1

.
E1 2 16
16! 1 2 16
Then, if L contains , it must be of the form
(N) (N)

(3.10)
(3.11)

(N)

L = 1 2 n Z1 2 n
(N) (N)
(N)
= 1 2 16 1 2 n E1 2 16 Z1 2 n ,

(3.12)

where Z1 2 n may contain derivatives of the spinor up to (N1) except for s, which
are explicitly exhibited. Now look at the terms of the form ( )15 (N+1) ( (N) )n1 in
DN [L, ]. There are only two sources for such a structure. One is produced as

L
 16n2 16 (N+1)
(N)
(N)
E2 16 Z1 2 n .
n
1
2

(3.13)

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

27

The other comes from


L
(1)N N (N)  (1)N+16 16n1 15 (N+1)
(N)
(N)
E1 16 Z2 n .
16
n
2

(3.14)
Since the product E1 2 16 Z1 2 n cannot be totally antisymmetric in its entire index
set (1 16 , 1 n ), (3.13) and (3.14) are independent and cannot cancel each other.
Moreover, neither term can be proportional to due to the total antisymmetry of E1 16 .
(For (3.13) it is obvious. For (3.14), if there exists 1 , then there must also be 16 ,
(N+1)
instead of .) Therefore, they must vanish separately and we find
which produces
Z1 n = 0, which proves the lemma.
The second lemma will severely restrict the way a product of more than two SPLDs can
appear in the effective Lagrangian L.
Lemma 2. Let (N) , with N  2, be the SPLD in L and assume that there are at least two
of them. Then, they can only be contained through the special combination (N) .
Proof. In general such structures with different number of (N) s may appear in L.
However, as they lead to algebraically independent expressions in the basic equation, we
may treat each term separately. Thus, we consider a generic structure of that type and write
it as
(N)
(N)
Z1 2 m ;1 2 n ,
L = 1 2 m (N)
n
1
2

(3.15)

where Z1 2 m ;1 2 n does not contain nor (N) and is separately antisymmetric in the
indices {1 2 m } and in {1 2 n }. Now look at the structure containing (2N) in
DN [L, ]. Such a structure can only be produced from N (L/ (N) ) and is proportional
to
(N)
(N)
(2N) Z1 2 m ;1 2 n1 .
1 2 m (N)
1
2
n2 n1

(3.16)

Since the appearance of is excluded by Lemma 1, for this to be allowed it must be


proportional to . This requires the existence of i in Z1 2 m ;1 2 n1 and similar
Kronecker s required by its antisymmetry property. This means that m  n and all the
indices {1 n1 , } must be paired off with (a part of ) i s through such Kronecker
s. Hence, (N) in L can only exist through the combination (N) .
The third lemma will further restrict the possible dependence on SPLD to the linear
level.
Lemma 3. Let (N) , with N  2, be the SPLD in L. Then, L can depend on (N) only
linearly.
Proof. Assume that L has more than two (N) s. According to Lemma 2, its dependence
on (N) must be through the combination (N) and therefore L must be of form

28

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

L=


m
(N) Zm ,

(3.17)

m2

where Zm contains derivatives of only up to (N1) . We now look at the part involving
(2N1) and in DN [L, ]. We will distinguish two cases.
First suppose Zm does not contain a term of the form ( (N1) )n . Then, the only way
(N)
that such a structure is produced is through N L/ and is given by
N

L
(N)


 (N) m2

 Nm(m 1) (2N1)
Zm .

(3.18)

Since this is not proportional to nor to (whose existence has already been excluded
by Lemma 1) and since m  2, this must vanish and hence Zm = 0.
On the other hand, if Zm contains ( (N1) )n , there will be another source. In such a
case, we may write L in the form
 
n
m 
(N) (N1) Zmn .
L=
(3.19)
m2,n

Suppressing the summation symbol, the contribution of the previous type takes the form
N

L
(N)


 (N) m2  (N1) n

Zmn ,
 Nm(m 1) (2N1)

(3.20)

while the additional contribution is


N1

L
(N1)

 (2N1)  (N) m 1  (N1) n 1

 m n
Zm n .

(3.21)

These contributions must add up to vanish. Therefore, we must have


m = m 1,

n = n + 1,
 

Nm(m 1)Zmn = m n Zm n .

(3.22)
(3.23)

From this we get


NmZmn = (n + 1)Zm1,n+1 .

(3.24)

Now let the largest m and n be mmax and nmax , respectively. From the above equation,
we find that Zmmax ,nmax must be proportional to Zmmax 1,nmax +1 , which, however, does not
exist. Thus Zmmax ,nmax must vanish. This contradicts the assumption that mmax , nmax are
the maximum values for m, n. Applying this logic recursively, we conclude that Zmn must
vanish for all m, n and L may depend on (N) only linearly.
The last lemma will completely dispose of SPLD with more than two time derivatives.
Lemma 4. Let the SPLD in L be (N) with N  2 and assume that L depends on (N)
linearly. Then, the dependence of L on (N) can only be through a total derivative.

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

29

Proof. We shall distinguish three cases.


Case 1: First assume that L contains more than two (N1) . Then it can be written as
 (N1) (N1)
L=
(3.25)
1
2
(N1)
(N) Z1 2 m ; ,
m
m2

where Z1 2 m ; contains the derivative of only up to (N2) . By integration by parts,



this can be rewritten as (suppressing the summation symbol m )


(N1)
(N1)
(N1) Z1 2 m ;
L = (N1)
m
1
2

m(N)
(N1)
(N1)
(N1) Z1 2 m ; + Z,
m
1
2

(3.26)

 is (N1) . Now if Z1 2 m ; is not totally antisymmetric with


where the SPLD in Z
respect to the indices (2 , 3 , . . . , m , ), then the second term vanishes and the lemma is
proved. If it happens to be totally antisymmetric, we shall write it as Z1 2 m without
the semicolon and the second term on the RHS is seen to have exactly the same structure
as L. In fact, it is precisely mL. Solving for L, we get

1
,
() + Z
(3.27)
m
which proves the lemma.
Case 2: If there is no (N1) in L, then L can be trivially rewritten into the form
 and the lemma is proved.
() + Z
Case 3: Now we come to the remaining case where there is one power of (N1)
present in L. In this case, we have
L=

(N1) (N)
Z; ,

L =

(3.28)

and a priori there is no symmetry property required of Z; . Now look at the terms

(N) 
and
containing (2N1) in DN [L, ]. Such terms can only be produced by N L/

(N1) 
N1
L/
and never arise by acting on Z; . Explicitly, these two contributions

add up to give
(1)N1 (2N1) (Z; + Z; ).

(3.29)

For N  2, obviously this cannot be proportional to . It cannot contain either, as was


already shown in Lemma 1. Hence it must vanish and Z; has to be antisymmetric. In
such a case, L can be rewritten as


L = (N1) (N1) Z; L (N1) (N1) Z; .
(3.30)
As in the previous two cases, we may solve for L and the lemma follows.

3.3. Proof of the Proposition


We are now ready to give a proof of the Proposition.
Proof. Through the previous lemmas, the problem of finding the solution to our basic
equation has been reduced to the case where L contains and only. The general form of

30

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

L is then
L=

1 2 n 1 2 m Z1 2 n ;1 2 m .

(3.31)

First consider the terms with more than two s. Then from L/ , we will get a term
proportional to
1 2 n 2 3 m Z1 2 n ;2 m ,

(3.32)

which cannot be produced in any other way. Such a term is allowed if and only if it is
proportional to . For this to happen, Z1 2 n ;2 m must contain 3 and similar
Kronecker s. However, due to the total antisymmetry of the index set (, 2 , . . . , m ),
this is not possible. Hence such terms cannot be present in L.
We are left with the possibility with zero or one and the most general form of L is


1 2 n Z1 2 n + 1 2 n 1 Z1 2 n ;1 .
L=
(3.33)
The terms with a can be rewritten as
L1 1 2 n 1 Z1 2 n ;1
= (1 2 n 1 Z1 2 n ;1 )
n1 2 n 1 Z1 2 n ;1 1 2 n 1 Z 1 2 n ;1 .

(3.34)

Consider first the case where n  2. Then if Z1 2 n ;1 is not totally antisymmetric, the
second and the third term vanish and hence L1 is a total derivative, which can be dropped.
On the other hand if Z1 2 n ;1 is totally antisymmetric, then the second term is equal to
nL1 . Thus we may solve for L1 and get

1 
(1 2 n 1 Z1 2 n ;1 ) 1 2 n 1 Z 1 2 n ;1 .
n+1
(3.35)

Thus, up to a total derivative, L1 is reduced to an expression without .


The important exception occurs for n = 1, namely, when L1 = Z; . In this case
there is no a priori symmetry properties required of Z; . Rewriting this as before, we get
L1 =

Z; = ( Z; ) + Z; .

(3.36)

If Z; is antisymmetric, then just as before Z; is a total derivative. On the other


hand if Z; has a symmetric part, then such a term can survive without contradiction.
At this point we have reduced the possible form of L down to

L=
1 2 n Z1 2 n + ZS ,
(3.37)
where ZS is a symmetric bispinor. Then, our basic equation becomes

n2 n Z2 n + 2(ZS ) + (Z S ) = G.

(3.38)

Look at the terms containing only. Since Z2 n is totally antisymmetric whereas Z S is


symmetric, these coefficient functions must vanish separately. In particular, it implies that

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

31

ZS is a constant and, since the only symmetric constant bispinor is , it must be of the
form (ZS ) = c , with c a constant. Then the basic equation is satisfied with G = 2c.
Thus, adding the -independent contribution [r] which has hitherto been suppressed,
we have found that the solution to the basic consistency equation can only be of the form

=c

d ( ) + [r].

(3.39)

Finally, requiring that the original Ward identity


(0) = 0 be satisfied, is fixed
completely to be proportional to the tree-level action. This completes the proof.
3.4. Proof of the main Theorem
Having established the Proposition, the proof of the Theorem follows immediately.
Proof. Let be a quantum-corrected SUSY transformation operator and let be a
solution to the SUSY Ward identity = 0. Now let us write the most general solution
as  = + - . With being a solution, - must by itself satisfy - = 0. Since the
tree-level solution is unique up to a constant, we fix its normalization so that the tree-level
correction - (0) vanishes. By assumption, we can expand and - in powers of the
coupling constant g as
=

g 2n (n) ,

- =

n0

g 2m2 - (m) .

(3.40)

m1

In our convention, rm , , and g 2 carries dimensions 1, 32 , 1 and 3, respectively, and


hence the dimension of - (m) is 3 3m. The Ward identity for - then becomes, for
each k  1,
0=

k1


(n) - kn .

(3.41)

n=0

Explicitly, the first few equations are


0 = (0)- (1) ,
0 = (0)- (2) + (1) - (1) ,
0 = (0)- (3) + (1) - (2) + (3) - (1).

(3.42)

Our Proposition dictates that the only possible solution to the first equation is - (1) =

with c a constant independent of g. Such an expression does not have the
c d (r 2 + ),
right dimension and hence - (1) must vanish. This in turn makes the second equation to
be of the form (0) - (2) = 0 and for the same reason we get - (2) = 0. Repetition of this
reasoning clearly shows that - = 0 altogether. This completes the proof.

32

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

4. Discussions
As the results of our present investigation have already been well-summarized, we shall
not repeat them. Below we shall discuss two important future problems and indicate how
they should be attacked.
One immediate question is whether similar results hold for the multi-body case.
Replacing rm and by rm,i and ,i , where the extra index i refers to different D-particles,
it is easy to see that our consistency equation (2.31) is generalized to

( ,i F,i + ,i F ,i ) = 0.
(4.1)
i

Apart from the special case where F,i contains a product of 16 s for each particle, this
equation can be algebraically satisfied if and only if
F,i = Sij ,j ,

(4.2)

where Sij is a symmetric matrix. There are many such matrices, examples of which are
rim rjm , i mn j rm rn , etc. Therefore, the analysis of its solution becomes more involved
compared to the single D-particle case treated in this work. Nevertheless, it would be quite
interesting and important to see, first, if one can construct a nontrivial solution, thereby
proving that SUSY alone cannot determine the structure of the effective action uniquely
for the multi-body case. If serious efforts in this direction should all fail, that would then
indicate a possibility to be able to construct a uniqueness proof.
The second and more ambitious enterprise is, as was mentioned in the introduction,
to study if SUSY alone truly determines the dynamics of the theory. In other words,
one would like to know whether the requirement of SUSY determines both the effective
transformation law and the effective action at the same time, up to some field redefinitions.
To formulate this problem more precisely, we must define what we mean by the effective
SUSY transformation to begin with. In the absence of off-shell superfield formalism, this
is already nontrivial since the closure relations, only by which one can guarantee that the
transformation is SUSY, themselves must depend on the form of the full effective action.
Consider, for example, the simplest case of the source-probe situation. Assuming that the
noncanonical parts of the closure relations are linear in the equations of motion, 3 the set
of equations characterizing SUSY would be of the form

= T
,

(4.3)

rm = m
,

(4.4)

+ B n

rn

[
, ]rm = 2(
)rm + Cm

+ Dm n

rn

= 0,

[
, ] = 2(
) + A

(4.5)
(4.6)
(4.7)

3 Although we have no proof, judging from the results at tree and 1-loop levels, nonlinear dependence is highly
unlikely.

Y. Kazama, T. Muramatsu / Nuclear Physics B 613 (2001) 1733

33

where
and are the global SUSY parameters and A D are some, in general field
dependent, coefficients. To show that SUSY determines the dynamics completely, one must
solve this set of nonlinear functional differential equations, with the tree-level information
as the only input, and demonstrate that all the quantities are fixed uniquely, up to field
redefinitions. A preliminary investigation indicates that at least at low orders this might be
feasible.
We hope to report on these and related matters in future communications.

Acknowledgement
Y.K. is grateful to T. Yoneya for a useful discussion. The research of Y.K. is supported in
part by Grant-in-Aid for Scientific Research on Priority Area #707 Supersymmetry and
Unified Theory of Elementary Particles No. 10209204 and Grant-in-Aid for Scientific
Research (B) No. 12440060, while that of T.M. is supported in part by the Japan Society
for Promotion of Science under the Predoctoral Research Program No. 12-9617, all from
the Japan Ministry of Education, Culture, Sports, Science and Technology.

References
[1] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hepth/9610043.
[2] L. Susskind, hep-th/9704080.
[3] S. Paban, S. Sethi, M. Stern, Nucl. Phys. B 534 (1998) 137, hep-th/9805018.
[4] S. Paban, S. Sethi, M. Stern, J. High Energy Phys. 06 (1998) 012, hep-th/9806028.
[5] D.A. Lowe, J. High Energy Phys. 11 (1998) 009, hep-th/9810075.
[6] S. Sethi, M. Stern, J. High Energy Phys. 06 (1999) 004, hep-th/9903049.
[7] S. Hyun, Y. Kiem, H. Shin, Nucl. Phys. B 558 (1999) 349, hep-th/9903022.
[8] H. Nicolai, J. Plefka, Phys. Lett. B 477 (2000) 309, hep-th/0001106.
[9] Y. Kazama, T. Muramatsu, Nucl. Phys. 584 (2000) 171, hep-th/0003161.
[10] Y. Kazama, T. Muramatsu, hep-th/0106218.
[11] K. Becker, M. Becker, J. Polchinski, A. Tseytlin, Phys. Rev. D 56 (1997) 3174, hep-th/9706072.
[12] Y. Okawa, T. Yoneya, Nucl. Phys. B 538 (1999) 67, hep-th/9806108.
[13] Y. Okawa, T. Yoneya, Nucl. Phys. B 541 (1999) 163, hep-th/9808188.
[14] W. Taylor, M.V. Raamsdonk, J. High Energy Phys. 04 (1999) 013, hep-th/9812239.
[15] Y. Kazama, T. Muramatsu, Class. Quantum Grav. 18 (2001) 2277, hep-th/0103116.

Nuclear Physics B 613 (2001) 3448


www.elsevier.com/locate/npe

Protected operators in N = 2, 4
supersymmetric theories
Nicola Maggiore a , Alessandro Tanzini b
a Dipartimento di Fisica, Universit di Genova, via Dodecaneso 33 and INFN, Sezione di Genova,

I-16146 Genova, Italy


b LPTHE, Universit Pierre et Marie Curie (Paris VI) et Denis Diderot (Paris VII), Place Jussieu,

75252 Paris cedex 05, France


Received 2 May 2001; accepted 14 August 2001

Abstract
The anomalous dimension of single and multi-trace composite operators of scalar fields is shown to
vanish at all orders of the perturbative series. The proof hold for theories with N = 2 supersymmetry
with any number of hypermultiplets in a generic representation of the gauge group. It then applies
to the finite N = 4 theory as well as to nonconformal N = 2 models. 2001 Published by Elsevier
Science B.V.
PACS: 03.70.+k; 11.10.Gh; 11.30.Pb
Keywords: Extended supersymmetry; BRS quantization; AdS-CFT correspondence

1. Introduction
It has been known for a long time that quantum field theories with extended
supersymmetry display a set of remarkable nonrenormalization properties. The -function
of the gauge coupling was argued to vanish to all orders of perturbation theory for the
N = 4 Super YangMills theory (SYM) [1] and to receive only one-loop corrections
for the N = 2 case [2]. The conjectured AdS/CFT correspondence [3] has renewed the
interest on the finiteness properties of these theories; in fact, many of the tests of the
correspondence have relied on nonrenormalization properties, which are crucial in order
to ensure a meaningful comparison between the strong coupling regime, accessible by
type IIB supergravity computations, and the weak coupling one, where the usual field
theory techniques are reliable.
In particular, the prototype example of the correspondence [3] establish a duality
between type IIB superstring on AdS5 S 5 and N = 4 SYM in the superconformal
E-mail addresses: maggiore@ge.infn.it (N. Maggiore), tanzini@lpthe.jussieu.fr (A. Tanzini).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 9 8 - 4

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

35

phase. In this context, it has been shown that chiral gauge invariant operators belonging
to short multiplets of the superconformal group SU(2, 2|4) have vanishing anomalous
dimensions [4]. The analysis of correlation functions of Chiral Primary Operators (CPO),
which are the lowest scalar components of such short superconformal multiplets, provided
a highly nontrivial check for the AdS/CFT correspondence [5,6]. Moreover, it recently led
to discover new double-trace operators which are protected despite they do not obey any
of the known shortening conditions [7,8].
We remark that the nonrenormalization properties of some CPOs does not necessarily
rely on superconformal algebra, and it can actually be shown also for theories with less
number of supersymmetries. This is indeed the case for the gauge-invariant operator Tr 2 ,
where in the N = 2 formalism corresponds to the complex scalar field of the vector
multiplet. The anomalous dimension of this operator has been shown to vanish to all orders
of the perturbative series in pure N = 2 SYM [9]; this allowed to provide a rigorous proof
of the celebrated nonrenormalization theorems for the beta functions of N = 2 and N = 4
theories [9,10].
In this work we extend the analysis of [9] to higher rank gauge-invariant polynomials
including both single and multi-trace operators. It is worth
of the scalar field and ,
to underline that our proof holds for N = 2 theories with any number of hypermultiplets
in a generic representation of the gauge group. It then applies to the finite N = 4 theory
as well as to nonconformal models, which are at present object of intensive research in
extensions of the AdS/CFT duality to phenomenologically more interesting gauge theories
with a nontrivial renormalization group flow [11,12].
A geometrical interpretation of such nonrenormalization properties can be given by
resorting to the topologically twisted formulation of the N = 2 theory. In fact, in this
context, the operators we consider can be written as components of the Chern classes of
the universal bundle defined in [22].
This work is organized as follows. In Section 2 we sketch the basic features of the
twisting procedure. In Section 3 we give the proof of finiteness of the operators Tr k ,
k  2 (single-traced) and i Tr ki , i ki = k (multi-traced). Our conclusions are drawn in
Section 4, while we confined into appendices the unavoidable technicalities of our proofs
and the explicit examples k = 2, 3, 4.

2. The twist
It is well known that N = 2 SYM theory can be given a twisted formulation [13],
which on flat manifolds is completely equivalent to the conventional one. Nonetheless,
the use of the twisted variables makes more evident and easily understandable some
important features, like the topological nature of a subset of observables [13] and
the nonrenormalization theorem concerning the beta-function of the gauge coupling
constant [9]. In view of these advantages, we choose to work in the twisted version of
N = 2 SYM theory. We address the interested reader to the existing literature [13,15] for

36

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

what concerns the many and deep aspects of the twisting procedure. In this section, we
prefer rather to set up the field theoretical background in which we shall work.
j ) are characterized by an index i = 1, 2, which
The usual N = 2 supercharges (Qi , Q
counts the number of supersymmetries, and a Weyl spinor index , which runs on the same
values as i : = 1, 2. At the origin of the twist, is Wittens idea of identifying the indices
i .

(2.1)

 can then be rearranged, with usual


The resulting twisted supercharges
and Q
conventions [16], into a scalar , a vector and a selfdual tensor :

1
Q ,
(2.2)
2
 
1

Q ,
(2.3)
2
1
( ) Q .
(2.4)
2
The N = 2 supersymmetry algebra, written for the twisted supercharges, contains a
subalgebra, formed by , and by the translations :
2 = 0,

{, } = ,

{ , } = 0.

(2.5)

We call the subalgebra (2.5) topological, since it appears to be common to all known
topological field theories [17], both of the Witten and Schwartz type [18]. The three twisted
supercharges turn out to be redundant, since they do not play any role either in the
classical definition or in the quantum extension of the theory [14]. For this reason, we shall
discard them throughout the rest of our reasoning.

Analogously to supercharges, the fields of the N = 2 gauge multiplet (A , i , i , , ),


which belong to the adjoint representation of the gauge group G, twist to


A , , , , , ,
(2.6)
where
[] ,

( ) ,

( ) () ,

(2.7)

are the anticommuting fields resulting from the twist of the spinor fields (i , i ) into
and the gauge boson A are not altered by
( , ). Notice that the scalar fields (, )
the twisting operation.
The N = 2 SYM pure gauge action twists to the TYM action



N =2 
A , i , i , , STYM A , , , , , ,
SSYM
(2.8)
which, in the 4D flat Euclidean spacetime reads [14]:


1
+ +
STYM = 2 Tr d 4 x 12 F
F
(D D )+ + D
g




D + 1 , 1 ,
12 D
2
2



1
18 [, ] 32
, , ,

(2.9)

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

37

1
+ =F
, F
+ = F+ = 1
+ and

where F
+ 2 F

2 F



1
(D D )+ = (D D ) + D D .
2
The symmetries of the action (2.9) are the usual gauge invariance
gaugeSTYM = 0,

(2.10)

and the relevant twisted supersymmetries


STYM = STYM = 0.

(2.11)

It is apparent from (2.7), that the 4D flat Euclidean twist simply corresponds to a linear
change of variables, therefore, it is not felt by the partition function

Z = DeS[] .
(2.12)
Moreover, the stress-energy tensor of the theory is modified by the twist only by a
total derivative term which do not affect the translation generators [15]. This observation
underlies the equivalence of the two, twist-related, theories N = 2 SYM and TYM when
formulated on flat manifolds. The theory has been extended to the quantum level both in
its untwisted [19] and twisted [14] formulation. Moreover, in [9] has been demonstrated
that a renormalization scheme does exist, in which the -function of the gauge coupling
constant g receives one loop contributions only.

3. Protected gauge invariant operators


Let us analyze the class of composite operators of the type Tr k , k  2 and i Tr ki ,
i ki = k. The aim of this section is to show that these operators are protected, in the
sense that are perturbatively finite, or, equivalently stated, they have vanishing anomalous
dimensions.
In order to prove this statement, let us consider the operator Q defined as
Q = s + ,

(3.1)

where s is the nilpotent BRS operator, is the twisted scalar supercharge (2.2), and is a
global commuting ghost.
The operator Q describes an invariance of the action STYM
QSTYM = 0,

(3.2)

and it is nilpotent
Q2 = 0,
since s 2 = 2 = {s, } = 0.

(3.3)

38

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

For our purposes, the Q-variations of the scalar field and of the FaddeevPopov ghost
c are sufficient
Q a = f abc cb c ,

1
Qca = f abc cb cc 2 a ,
2

(3.4)

where the index a = 1, . . . , dim G counts the adjoint representation matrices of the gauge
group G, and f abc are the structure constants of the corresponding Lie algebra.
The operators Tr k and i Tr ki , i ki = k are the observables of TYM (hence N = 2
SYM) theory [13,14]; we can write them as
D a1 ak a1 ak ,

(3.5)

where the D a1 ak are completely symmetric invariant tensors of rank k


k = 2,
k = 3,

D a1 a2 = a1 a2 ,
D a1 a2 a3 = a1 a2 a3 ,
..
.

In particular, the k = 2 operator Tr 2 can be thought to as a kind of prepotential of the


theory, since the whole action can be written as

2
1
d 4 x Tr ,
2
(3.6)
3g
2
where this relation holds modulo BRS trivial cocycles. Eq. (3.6) reflects the fact that Tr 2
contains all the physical information of the theory, as it constitutes its inner bulk (for
a detailed discussion of this point, see [9,14]).
A crucial point is that, in the space of local field polynomials which are not necessarily
analytic in the constant parameter , the operators (3.5) can be written as Q-variations,
i.e., in this space the cohomology of Q is empty. We stress that quantum field theory
rules require that the 1PI generating functional must be an analytic function in as
well as in any other parameter of the theory, whereas this condition can obviously be
relaxed for quantum insertions, like for instance (3.5), which can be rendered analytic
by a multiplication for a suitable power of , as we shall see.
The following result holds (see proof in Appendix A)

 (k)
(k)
D a1 ak a1 ak = Q D a1 ak f0 ca1 sca2 scak + f1 ca1 sca2 scak1 ak

(k) a1 a2
+ + fk1
c ak


k1

a1 ak
(k) a1 a2
akp akp+1
ak
=Q D
(3.7)
fp c sc sc

,
p=0

or, in shorthand notation,


(k) () = QP (k) (c, ).

(3.8)

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

39

The coefficients fp(k) are given by


fp(k) =

(k 1)!k!
(1)kp
,
2(kp) p!(2k p 1)!

0  p  k 1,

and sca is the ordinary BRS variation of the FaddeevPopov ghost c



sca = Qca =0 = 12 f abc cb cc .

(3.9)

(3.10)

Notice that the coefficient of the last term on the r.h.s. of (3.7) is universal
(k)

fk1 =

1
.
2

(3.11)

It is worth to remark that the operators (3.5) we are considering have an interesting
geometrical interpretation [22]. Let us consider for the moment k = 2; by redefining the
scalar field as 2 , (3.8) can be written as


Tr 2 = Q Tr cQc 23 ccc .
(3.12)
If we define the universal connection A = A + c, (3.12) can be seen as the (p = 0, q = 4)
component of the equation




F
 = dA
CS = d Tr A dA 2 A A A ,
Tr F
(3.13)
3
where by (p, q) we indicate the grading of the forms with respect to the extended exterior
derivative d = d + Q, p being the spacetime form degree and q the ghost number. Thus
P (2)(c, ) can be understood as the (0, 3) component of the ChernSimons form ACS
while as the (0, 2) component of the corresponding
of the universal connection A,
. Analogously, higher rank single trace operators (3.5) can be expressed in
curvature F
k .
terms of (0, k) components of higher Chern classes Tr F
We now come to the proof of finiteness; what is actually important in this sense is that
operators (3.5) can be expressed as the Q-variations of polynomials in the fields and
c, and, in particular, that they are related to D a1 ak ca1 sca2 scak (the first term on the
r.h.s. of (3.7)), which belongs to the cohomology of the ordinary BRS operator, and is
known to be finite to all orders of perturbation theory [20]. It is then natural to expect
that the nonrenormalization property of the BRS invariant odd polynomials in the ghost c
translates to (3.5) as well. As we shall see, this is indeed the case.
Let us consider now the classical action
= STYM + Sgf + Sext + Sk .

(3.14)

STYM is the twisted N = 2 SYM action (2.9), Sgf is the (Landau) gauge fixed term

A ,
Sgf = Q Tr d 4 x c
(3.15)
c being the FaddeevPopov antighost. According to the usual BRS procedure to implement
at the quantum level classical field theories, in Sext the nonlinear Q-variations of the

40

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

fields are coupled to external fields , sometimes called antifields


 
c,
Tr d 4 x Q, = A , , , , , ,
Sext =

(3.16)

and finally Sk is given by





2k2
 a1 ap1
2k
4
(k)
(k)
(k)
.
2kp a1 ap1 P
d x 0 + 2k1 P +
Sk =

(3.17)

p=2

In the above expression for Sk , the s are external sources coupled to (k) (), P (k)(c, )
(defined by (3.8)), and its derivatives with respect to the ghost c (a /ca ). They are
antisymmetric in their color indices as, and are commuting or anticommuting objects
depending from the number of ghosts c contained in the composite operators they are
coupled to. Such a number is represented by the lower indices, hence their statistics is given
by (1)2kp . Notice that, thanks to the overall multiplying factor 2k , the functional Sk
is, as it should for being part of the action, an analytic expression in .
The whole action satisfies the SlavnovTaylor (ST) identity
S() = S old () + S (k) = 0,

(3.18)

where the usual ST operator




 

S old () =
Tr d 4 x
+
c

(3.19)

has been implemented to the -sector as follows (see Appendix B)





2k3
 p(1 p)

(k)
4
mnap1 mna1 ap2
S = d x 2k1
f
+
2kp1
a1 ap1 .
0
2
2kp
p=2

(3.20)
Notice that the -sources, transforming one into the other, enter the ST identity (3.18) as
BRS doublets, which have the nice property of keeping unaltered the cohomology of the
(linearized) ST operator.
For the proof of finiteness, it is crucial that an additional constraint on the theory is
satisfied: the ghost equation, peculiar of the Landau gauge [21]. Its main consequence is
a nonrenormalization theorem, which states that, in the Landau gauge, the ghost field c
enters the quantum theory only if differentiated ( c). It is precisely for the purpose of
being able to take advantage of this additional symmetry of the theory, that in Sk external
sources have been coupled not only to (k) () and P (k) (c, ), but also to the operators
a1 ap1 P (k)(c, ). The ghost equation reads (see Appendix C)
F a = a ,
where


Fa =

(3.21)

d 4x


2k3


a
a
1
p1
+ f abc cb c
(1)2kp 2kp
aa1 ap1 ,
ca
b
2kp1
p=1

(3.22)

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

41

and a is a classical breaking, linear in the quantum fields, given by






a
a
f abc b c + 2 1 2k3 a a1 a2k3 P (k) ,
a = d 4 x

(3.23)

indeed, the operator a a1 a2k3 P (k)(c, ) is independent from the field , and it is
linear in the ghost field c.
Once observed that both the ST identity (3.18) and the ghost equation (3.21), being
anomaly-free, can safely be extended to the quantum level [14]
S( ) = 0,

F a = a ,

(3.24)

where is the 1PI generating functional = + O(h ), we are able to prove our main
result, i.e., that the quantum insertions Tr k = Tr k + O(h ) have vanishing anomalous
dimensions, namely, satisfy the CallanSymanzik (CS) equation


C Tr k (x) = 0,
(3.25)
where C is the CS operator
C

+ h g
h
N

(3.26)

satisfying
C = 0.

(3.27)

In (3.26), g is the -function of the gauge coupling constant g, is the anomalous

dimension of the generic field , the operator N is the counting operator Tr d 4 x


,
and is the renormalization scale.
Let us take the derivative of the quantum ST identity (3.18) with respect to the external
source 2k1 , coupled in (3.17) to P (k)(c, )



(3.28)
S( )
= 0,

2k1

=0

where we put to zero the set of external sources appearing in Sk (3.17). From the
expression of the ST operator, Eq. (3.28) writes




b
(3.29)
=
.



2k1 =0

0 =0

The operator b appearing in the above equation is the (off-shell nilpotent) linearized
quantum ST operator





4
+
+ c
b = d x Tr
+ 2k1



b
0


2k3
 p(1 p)

mnap1 mna1 ap2


f
+
2kp1
(3.30)
a1 ap1
2
2kp
p=2

and it can be recognized as the quantum functional translation of the operator Q. The
relation (3.29) is the quantum extension of the classical one (3.8) between the composite

42

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

operators Tr k and P (k) (c, ):



D a1 ak a1 ak = b P (k) (c, ) .

(3.31)

Owing to the ghost equation (3.21), the quantum insertion P (k)(c, ) has vanishing
anomalous dimension, since the undifferentiated ghost field c does not get quantum
corrections


C P (k) (c, ) = 0.
(3.32)
Now, since the CS operator commutes with the linearized quantum ST operator b
[C, b ] = 0,

(3.33)

from Eqs. (3.32) and (3.33) we derive our result (3.25): the composite operators Tr k are
finite to all orders of perturbation theory. This result can be trivially extended to the class
of operators Tr k .
We would like to stress once again that the presence of matter does not alter in any way
our result (3.25). Indeed, we needed only to consider the symmetry transformations (3.4)
of the scalar field of the N = 2 gauge supermultiplet and of the FaddeevPopov ghost
field c. These transformations are not altered by the presence of matter fields, hence the
matter sector is completely decoupled in our proof of existence of protected operators.
Under this respect, matter is only an avoidable graphical charge, which we preferred to
omit. The class of protected operators discussed in this paper refers, therefore, to N = 2
SYM theories coupled to matter in an arbitrary representation of the gauge group, hence to
N = 4 theory as well, since for the particular, choice of matter in the adjoint representation,
N = 4 SYM is recovered from the N = 2 case.

4. Conclusions
We have shown the vanishing to all orders of the perturbative expansion of the
anomalous dimension of local single and multi-trace composite operators of scalar fields
in theories with N = 2 supersymmetry. Although the proof has been given explicitly for
the pure gauge case, it holds unaltered for N = 2 SYM coupled to matter belonging to
a generic representation of the gauge group G, hence in particular it remains valid for
the N = 4 case, obtained by putting matter multiplets in the adjoint representation of G.
Nonrenormalization properties of these operators can be better displayed using the twisted
formulation of the theory, where they have a nice geometrical interpretation as components
of the Chern classes of a suitable universal bundle [22]. The study of correlation functions
of such operators play a relevant role in tests of the AdS/CFT correspondence [5,6]. In
addition, it recently led to conjecture the existence of a new class of gauge invariant
operators which are protected despite they do not obey any of the known shortening
conditions [7,8].
We stress that our analysis does not rely on the superconformal algebra. It would then
be interesting to extend it to the above mentioned operators. Moreover, the approach we

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

43

followed can be applied also to nonconformal N = 2 theories which have been recently
considered in extensions of the AdS/CFT duality [11,12]. Even if a precise map between
bulk supergravity fields and boundary gauge-invariant field theory operators, as that of
the N = 4 superconformal theory, is not presently known for these cases, the appearance
of protected operators in theories with a nontrivial renormalization group flow is a very
interesting feature by itself, and could motivate further analysis on the correlation functions
of these operators.

Acknowledgements
We acknowledge helpful discussions with L. Baulieu, M. Bianchi, S. Kovacs and
C. Schweigert. A.T. is supported by a fellowship of the MURST project. This work
is partially supported by the European Commission, RTN programme HPRN-CT-200000131 and by MURST.

Appendix A. (k) () = QP (k) (c, )


The fields and c and the parameter are assigned the following quantum numbers

dim .
R-charge
-charge
R +

1
2
0
2

0
0
1
1

1/2
1
1
0

(A.1)

Therefore, the most general expression for (k) () = D a1 ak a1 ak having homogeneous R + quantum number is the following


k1

a1 ak a1
ak
a1 ak
(k) a1 a2
akp akp+1
ak
= Q D
fp c sc sc

.
D
(A.2)
p=0
(k)

In (A.2), the coefficients fp depend on the global ghost appearing in the definition of
the symmetry operator Q = s + (3.1). The first term of the sum (A.2) is proportional to
D a1 ak ca1 sca2 scak , which is the c(2k1) cocycle, belonging to the cohomology of the
BRS operator [23], and known to be finite to all orders of perturbation theory [20]. The
properties of the invariant polynomials (k) () are mostly determined by the relationship
existing between them and the BRS cocycles c(2k1) .
Our first goal is to find the explicit expression for the coefficients fp(k) in (A.2). In order
to obtain it, let us observe that
Qsca = 2 s a ,

(A.3)

44

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

and that
D a1 an ca1 sca2 scam am+1 an1 s an
2
D a1 an sca1 scam am+1 an .
=
(A.4)
nm
Eq. (A.4) is a consequence of the nilpotency of the BRS operator s and of the fact that the
completely symmetric rank-k tensor D a1 ak is invariant, hence it satisfies the Jacoby-like
identity
D p1 a2 ak f p1 qa1 + D a1 p2 a3 ak f p2 qa2 + + D a1 ak1 pk f pk qak = 0.

(A.5)

Performing the Q-variation in (A.2) and exploiting the identities (A.3), (A.4), we get
D a1 ak a1 ak
=

k1


D a1 ak fp(k) sca1 scakp akp+1 ak

p=1

k2


a1 ak

fp(k) 2

p=0



2k
sca1 scakp1 akp ak
1
p+1

fk1 2 D a1 ak a1 ak .
(k)

Hence, reorganizing terms, we have


D a1 ak a1 ak
= fk1 2 D a1 ak a1 ak
(k)

k1

p=1


 

2k
(k)
fp1
D a1 ak sca1 scakp akp+1 ak 2 1
fp(k) . (A.6)
p

We must, therefore, impose


1
,
2



2k
(k)
(k)
2
fp = 1
fp1 ,
p
(k)
fk1
=

(A.7)
1  p  k 1.

(A.8)

Eqs. (A.7) and (A.8) can be solved by a closed formula


fp(k) =

(k 1)!k!
(1)kp
,
2(kp) p!(2k p 1)!

0  p  k 1,

(A.9)

as can be verified by induction. This proves the relations (3.7) and (3.9).
For future reference, it may be useful to write (3.7) for the lowest values k = 2, 3, 4.
For k = 2, the invariant tensor D ab is just the Kronecker delta ab , and from (3.7)
and (3.9) we immediately get



 (2)
1 a a
1 a a
(2)
c
sc

a a = Q f0 ca sca + f1 ca a = Q
(A.10)
.
34
2

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

45

For k = 3, there is only one completely symmetric tensor D abc = d abc , and


d abc a b c = Qd abc f0(3) ca scb scc + f1(3) ca scb c + f2(3) ca b c


1 a b c
1 a b c
1 a b c
abc
= Qd
c sc sc + 4 c sc 2 c .

106
2

(A.11)

For k = 4, there are more than one symmetric tensors D abcd


D abcd a b c d
 (4)
(4)
= QD abcd f0 ca scb scc scd + f1 ca scb scc d


+ f2(4) ca scb c d + f3(4) ca b c d

1 a b c d
1
= QD abcd
c sc sc sc 6 ca scb scc d
8
35
5

3 a b c d
1 a b c d
+ 4 c sc 2 c .
5

(A.12)

Appendix B. The SlavnovTaylor identity


Our task is to extend the action of Q on the set of external -sources introduced in



2k2
 a1 ap1
2k
4
(k)
(k)
(k)
Sk =
(B.1)
,
2kp a1 ap1 P
d x 0 + 2k1 P +
p=2

by demanding
QSk = 0.

(B.2)

We have


QSk =

2k

d x (Q0 ) (k) +
4

2k2


a1 ap1 
Q2kp
a1 ap1 P (k)

p=1
a1 ap1 
Qa1
+ (1)2kp 2kp

ap1 P

(k)


,

(B.3)

where we taked into account the Q-invariance of the polynomials (k) () and the statistics
of the -sources, given by (1)2kp .
In order to evaluate (B.3), we observe that the anticommutator between the derivative
with respect to the ghost field c (a ) and the symmetry operator Q, gives the generators of
the rigid gauge transformations Ha
{a , Q}Xb = Ha Xb = f abc Xc ,

(B.4)

where Xa is a generic function of the fields and c and belongs to the adjoint
representation of the gauge group. From the obvious relations
Ha (k) () = Ha P (k)(c, ) = a QP (k) (c, ) = 0,

(B.5)

46

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

we derive
Qa P (k) = 0

(B.6)

and, in general,
a1 am Qa1 am P (k) = (1)m+1

m(m 1) a1 a2 p
a3 am p P (k) a1 am .
f
2
(B.7)

Hence (B.3) writes





QSk =

2k

d x (Q0 ) (k) 2k1 (k)


4

2k2


a1 ap1 
Q2kp
a1 ap1 P (k)

p=1


(p 1)(p 2) a1 ap1 a1 a2 q
(k)
2kp f
+
a3 ap1 aq P
2


2k
4
=
d x (Q0 ) (k) 2k1 (k) + (Q2k1 )P (k)
 a a 
+ Q2 1 2k3 a1 a2k3 P (k)


2k3
  a1 ap1  p(p 1) rsa1 ap2
2kp1 f rsap1 a1 ap1 P (k) .
Q2kp
+
+
2
p=2

(B.8)
In order that the whole expression QSk vanish, by identifying terms containing equal
number of ghost fields c, we have
Q0 = 2k1 ,
a a
Q2 1 2k3

(B.9)

= 0,

(B.10)

Q2k1 = 0,

(B.11)

and
a a

1
p1
Q2kp
=

p(p 1) rsap1 rsa1 ap2


f
2kp1 ,
2

2  p  2k 3.

(B.12)

Hence, the implementation of the ST identity to the -sector is given by





2k3
 p(1 p)

(k)
4
mnap1 mna1 ap2
f
+
2kp1
S = d x 2k1
a1 ap1 ,
0
2
2kp
p=2

(B.13)
which coincides with (3.20).

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

47

Appendix C. The ghost equation


To extend the ghost equation, let us take the functional derivative with respect to ghost
field c of the action functional Sk
a S (k) =

2k2


a a

1
p1
(1)2kp 2kp
a a1 ap1 P (k)

p=1

2k3


a a

1
p1
(1)2kp 2kp

p=1

a a2k3

d 4 x 2 1

aa1 ap1
2kp1

a a1 a2k3 P (k).

(C.1)

Notice that, according to the expression (3.7), the maximum number of ghosts c present
in P (k)(c, ) is 2k 1, in the term c(2k1), whereas the maximum number of its ghost
derivatives in the last term at the r.h.s. of (C.1) is 2k 2, hence this term is linear in c and
independent from , i.e., it represents a classical breaking.
Therefore, the complete ghost equation reads
F a = a ,
where


F =
a

(C.2)

d x


2k3


abc b
2kp a1 ap1
+ f c

(1)
2kp
aa1 ap1 ,
ca
bc
2kp1

(C.3)

p=1

and a is the classical breaking given by






a1 a2k3
a
4
abc b c
(k)
= d x
.
f + 2
a a1 a2k3 P

(C.4)

References
[1] M.F. Sohnius, P.C. West, Phys. Lett. B 100 (1981) 245;
S. Mandelstam, Nucl. Phys. B 213 (1983) 149;
P.S. Howe, K.S. Stelle, P.K. Townsend, Nucl. Phys. B 236 (1984) 125.
[2] P.S. Howe, K.S. Stelle, P.C. West, Phys. Lett. B 124 (1983) 55.
[3] J. Maldacena, Supergravity, Adv. Theor. Math. Phys. 2 (1998) 231;
J. Maldacena, Int. J. Theor. Phys. 38 (1998) 231, hep-th/9711200.
[4] S. Ferrara, C. Fronsdal, A. Zaffaroni, Nucl. Phys. B 532 (1998) 153, hep-th/9802203;
L. Andrianopoli, S. Ferrara, Phys. Lett. B 430 (1998) 248, hep-th/9803171;
L. Andrianopoli, S. Ferrara, Lett. Math. Phys. 46 (1998) 265, hep-th/9807150;
L. Andrianopoli, S. Ferrara, Lett. Math. Phys. 48 (1999) 145, hep-th/9812067.
[5] M. Bianchi, S. Kovacs, Phys. Lett. B 468 (1999) 102, hep-th/9910016;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Nucl. Phys. B 584 (2000) 216, hep-th/0003203;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, JHEP 9908 (1999) 020, hep-th/9906188.

48

N. Maggiore, A. Tanzini / Nuclear Physics B 613 (2001) 3448

[6] B. Eden, P.S. Howe, P.C. West, Phys. Lett. B 463 (1999) 19, hep-th/9905085;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 472 (2000) 323, hepth/9910150;
P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 571 (2000) 71, hep-th/9910011;
B.U. Eden, P.S. Howe, E. Sokatchev, P.C. West, Phys. Lett. B 494 (2000) 141, hep-th/0004102.
[7] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, hep-th/0104016.
[8] G. Arutyunov, B. Eden, A.C. Petkou, E. Sokatchev, hep-th/0103230.
[9] A. Blasi, V.E. Lemes, N. Maggiore, S.P. Sorella, A. Tanzini, O.S. Ventura, L.C. Vilar,
JHEP 0005 (2000) 039, hep-th/0004048.
[10] V.E. Lemes, M.S. Sarandy, S.P. Sorella, A. Tanzini, O.S. Ventura, Operator, JHEP 0101 (2001)
016, hep-th/0011001;
V.E.R. Lemes, N. Maggiore, M.S. Sarandy, S.P. Sorella, A. Tanzini, O.S. Ventura, Nonrenormalization theorems for N = 2 super YangMills, in: J. Bagger, S. Duplij, W. Siegel (Eds.),
Concise Encyclopedia of Supersymmetry, Kluwer, Dordrecht, hep-th/0012197;
V.E. Lemes, M.S. Sarandy, S.P. Sorella, O.S. Ventura, L.C. Vilar, hep-th/0103110.
[11] J. Polchinski, Int. J. Mod. Phys. A 16 (2001) 707, hep-th/0011193;
M. Bertolini, P. Di Vecchia, M. Frau, A. Lerda, R. Marotta, I. Pesando, JHEP 0102 (2001) 014,
hep-th/0011077;
M. Petrini, R. Russo, A. Zaffaroni, hep-th/0104026.
[12] N. Evans, C.V. Johnson, M. Petrini, JHEP 0010 (2000) 022, hep-th/0008081;
K. Pilch, N.P. Warner, Nucl. Phys. B 594 (2001) 209, hep-th/0004063;
A. Brandhuber, K. Sfetsos, Phys. Lett. B 488 (2000) 373, hep-th/0004148.
[13] E. Witten, Commun. Math. Phys. 117 (1988) 353.
[14] F. Fucito, A. Tanzini, L.C. Vilar, O.S. Ventura, C.A. Sasaki, S.P. Sorella, Algebraic renormalization: perturbative twisted considerations on topological YangMills theory and on N = 2
supersymmetric gauge theories, lectures given at the First School on Field Theory and Gravitation, Vitria, Esprito Santo, Brazil, hep-th/9707209;
A. Tanzini, O.S. Ventura, L.C. Vilar, S.P. Sorella, J. Phys. G 26 (2000) 1117.
[15] E. Witten, Supersymmetric YangMills theory on a four-manifold, J. Math. Phys. 35 (1994)
5101, hep-th/9403195;
M. Marino, The geometry of supersymmetric gauge theories in four dimensions, hepth/9701128.
[16] J. Bagger, J. Wess, Supersymmetry and Supergravity, Princeton Univ. Press.
[17] D. Birmingham, M. Rakowski, Phys. Lett. B 269 (1991) 103.
[18] D. Birmingham, M. Blau, M. Rakowski, G. Thompson, Phys. Rep. 209 (1991) 129.
[19] N. Maggiore, Int. J. Mod. Phys. A 10 (1995) 3781, hep-th/9501057;
N. Maggiore, Int. J. Mod. Phys. A 10 (1995) 3937, hep-th/9412092.
[20] O. Piguet, S.P. Sorella, Nucl. Phys. B 381 (1992) 373.
[21] A. Blasi, O. Piguet, S.P. Sorella, Nucl. Phys. B 356 (1991) 154.
[22] L. Baulieu, I.M. Singer, PAR-LPTHE-88/18 Presented at Annecy Mtg. on Conformal Field
Theory and Related Topics, Annecy, France, March 1416, 1988, Nucl. Phys. B (Proc. Suppl.) 5
(1988) 12.
[23] J. Manes, R. Stora, B. Zumino, Commun. Math. Phys. 102 (1985) 157;
B. Zumino, Nucl. Phys. B 253 (1985) 477.

Nuclear Physics B 613 (2001) 4963


www.elsevier.com/locate/npe

Two-loop Higgs mass in supersymmetric


KaluzaKlein theories
A. Delgado, G.v. Gersdorff, M. Quirs
Instituto de Estructura de la Materia (CSIC), Serrano 123, E-28006 Madrid, Spain
Received 25 July 2001; accepted 2 August 2001

Abstract
We perform an explicit two-loop computation of the Higgs mass in a five-dimensional supersymmetric theory compactified on the orbifold S 1 /Z2 , with superpotential localized on a fixed-point
brane and supersymmetry broken in the bulk of the extra dimension by ScherkSchwarz boundary
conditions. At one-loop, the Higgs mass is finite and proportional to 1/R 2 , R being the radius of the
extra dimension. The two-loop corrections contain finite ( 1/R 2 ) and linearly divergent ( /R)
terms, where is the cutoff of the five-dimensional theory. The divergent terms are cancelled by
diagrams containing one-loop wave-function renormalization counterterms, i.e., by linear threshold
effects of Yukawa (and gauge) couplings, at the scale 1/R. As a result there is no counterterm for the
Higgs mass, which is finite and calculable to the considered order. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 12.60.Jv; 04.50.+h; 11.30.Pb; 11.10.Kk

1. Introduction
There has been recently a lot of effort in order to provide an explanation for electroweak
symmetry breaking (EWSB) by means of supersymmetric theories with one compact
(radius R) extra dimension [18]. Such theories arise naturally as low energy limits of
string theories [9] and are to be understood as effective theories, when all heavy string
modes are integrated out, with an ultraviolet (UV) cutoff of the order of the string
scale Ms .
A common claim of these models is that electroweak symmetry breaking (EWSB) is
triggered by radiative corrections which are insensitive to the scale . Some one loop

Work supported in part by CICYT, Spain, under contract AEN98-0816, and by EU under contracts HPRNCT-2000-00152 and HPRN-CT-2000-00148.
E-mail address: mariano@makoki.iem.csic.es (M. Quirs).

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 1 ) 0 0 3 8 0 - 7

50

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

calculations of the effective potential, or the Higgs mass, have been performed in order to
establish this result in field [18], and string [10] theory, and also some general arguments
have been given [5,1113].
One uses typically the so-called KaluzaKlein (KK) regularization which exploits the
fact that sums over KK modes (integration over the fifth momentum) turn out to be finite
(this is already known from finite temperature field theories). Since the summation goes
over modes with masses far beyond the cutoff of the effective theory, there have been some
doubts on the validity of such a regularization [14]. However, it has been shown that several
cutoffs which respect all symmetries of the theory lead to a -insensitive result at one-loop,
which equals the one obtained in the KK regularization [15,16]. This is a very interesting
conclusion since it means that EWSB does not depend on the microscopic theory at all and
we have an effective theory with an unexpected amount of predictability. However, since
no rigorous proof has yet been given which guarantees the independence of EWSB on the
high energy theory, it seems worthwhile to explicitly check this in a two loop calculation.
In this article we will perform a two-loop calculation of the Higgs mass from the
top/stop sector of the model presented in Ref. [8]. It displays the way how supersymmetric
cancellations survive the soft breaking. There remains a linear sensitivity to the cutoff
which is, however, entirely absorbed by the supersymmetric running of the Yukawa
coupling. 1 As a consequence there is no mass counterterm to the considered order and
the Higgs mass is calculable and finite. These features would remain in the two-loop
calculation of the Higgs mass when the gauge sector is included.
The outline of this paper is as follows. In Section 2 we will summarize the main features
of the five-dimensional (5D) supersymmetric model that will be used in the two-loop
calculation. In particular the four-dimensional Lagrangian, after integration over the fifth
dimension, with all relevant interactions will be explicitly provided. The corresponding
Feynman rules can be easily deduced from the Lagrangian. In Section 3 we review, for
completeness, the diagrammatic calculation of the one-loop Higgs mass, and the two-loop
contribution is provided in Section 4, where the relevant one-loop counterterms due to the
wave-function renormalization of the superfields, and their contribution to the two-loop
Higgs mass, is computed. Our conclusions are drawn in Section 5.

2. The model
The model we will use for the calculation is the one of Ref. [8]. The 5D spacetime is
compactified on the orbifold M S 1 /Z2 which has two fixed points at y = 0, ( = R
is the segment length) where two 3-branes are located. There are two types of fields: those
living in the bulk of the extra dimension (U -states), organized in N = 2 supermultiplets,
and those living on the branes and localized at the fixed points (T -states), organized in
N = 1 supermultiplets. U -states are: gauge fields, which appear in N = 2 vector multiplets
V = (V , 1 ; + iV5 , 2 ) and transform under the adjoint representation of the SU(3)
1 In the supersymmetric limit there is only a wave-function renormalization for the fields which is absorbed by
the running of the Yukawa coupling. This running is of course linear for D = 5.

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

51

SU(2)L U (1)Y , and SU(2)L singlets, which appear in N = 2 hypermultiplets, U =


(U, u; U  , u ), D = (D, d; D  , d  ) and E = (E, e; E  , e ). Even and odd N = 1 superfields
under Z2 are separated by a semicolon as (even; odd). T -states localized at the y = 0
brane are the SU(2)L doublets, H1,2 = (H1,2, h1,2 ), Q = (Q, q) and L = (L, ). 2 This
field assignment allows for a superpotential involving the top/stop sector as
(5)

W = ht QH2 U (y)

(2.1)

which triggers EWSB at one-loop as was shown in Ref. [8]. In the rest of this paper we
will consider the leading contribution from the top-stop sector and neglect, in explicit
calculations, the contributions from gauge couplings. The latter can be easily included
and will not modify qualitatively the result of this paper.
Let us express the top/stop sector of the theory in terms of four-dimensional N = 1
superfields, using the formalism of Ref. [17]


L = U U + U  U  + Q Q(y) + H2 H2 (y)


+ U y U  + h(5)
(2.2)
t U QH2 (y) + h.c. .
The superpotential term is given by the second expression and depends on the chiral
superfields Q, H2 , U and y U  . The inclusion of the derivative term, on top of the
superpotential given in (2.1), is pure convention and it is motivated by the fact that,
from a four-dimensional viewpoint, it constitutes a mass term for the modes. Notice
that we have not included a H2 H1 (y) term in the superpotential. The parameter
is a supersymmetric mass and thus its presence is completely decoupled from the
actual calculation of the possible soft supersymmetry Higgs mass counterterms, while
it makes the calculation more involved. For that reason we have decided to work in the
(phenomenologically unrealistic) case = 0. In Section 5 we will make some comments
about the impact, in the present calculation, of the -term. We can anticipate, not
unexpectedly, that the final result (in fact the absence of a mass counterterm) will be
unchanged by the presence of the -term.
In terms of component fields we have
L = U U + U  U  + Q Q(y) + H2 H2 (y)
+ i u u + i u  u + i q q(y) + i h 2 h2 (y)
|FU |2 |FU  |2 |FQ |2 (y) |FH2 |2 (y)
W
W
W
W
FU +
FQ +
+
FU  +
FH
U
U 
Q
H2 2
2W
2W
2W
2W


uu

qh

qu

h2 u + h.c.
2
U U 
QU
QH2
H2 U
The auxiliary fields are computed to be:
FU =

W
(5)
= ht (y)QH2 y U  ,
U

(2.3)

(2.4)

2 We are using the same notation, upper case symbols, for the N = 1 chiral superfields and their scalar
components, while lower case symbols are reserved for the fermionic components, e.g., U = (U, u).

52

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

W
= y U,
U 
W
(5)
= ht (y)U H2,
(y)FQ =
Q
W
(5)
= ht (y)U Q.
(y)FH 2 =
H2
FU  =

(2.5)
(2.6)
(2.7)

The interaction and mass terms become


 (5)
 (5)
 (5)
2
2
2
Lint = ht (y)QH2 + y U   |y U |2 (y)ht U H2  (y)ht U Q
uy u ht (y)H2 qu ht (y)U qh2 ht (y)Qh2 u + h.c.
(5)

(5)

(5)

(2.8)

Supersymmetry breaking is performed using the ScherkSchwarz (SS) [18] boundary


conditions based on the subgroup U (1)R which survives after the orbifold S 1 /Z2 action.
Using the R-symmetry U (1)R with parameter to impose different boundary conditions
for bosons and fermions inside the 5D N = 1 multiplets, one obtains for the nth KK mode
of gauge bosons and chiral fermions living in the bulk the compactification mass n/R,
while for the nth KK mode of gauginos and supersymmetric partners of chiral fermions
living in the bulk, the mass (n + )/R. Of course, the chiral multiplets localized on the
brane do not acquire any tree-level supersymmetry breaking mass.
After SS-breaking, the bulk hypermultiplets have the following Fourier expansions:
u=

cos(ky/R)u(k),

(2.9)


1
u =
sin(ky/R)u(k),
R k=

(2.10)




1
cos (k + )y/R U (k) ,
U=
R k=

(2.11)




1
sin (k + )y/R U (k) ,
U =
R k=

(2.12)

R k=

where we chose to combine the two half-towers to give a single tower.


Plugging these expressions into the Lagrangian and performing the y-integration from
0 to R we get

 2 (k) (k) 
U (k) U (k) mB
+ Q Q + H2 H2
U
L=
k U
k



1 F (k) (k) 1 F (k) (k)
(k)
(k)
i u u mk u u mk u u
+
2
2
k

+ i q q + i h 2 h2



U (k) U (l) H2 H2 + U (k) U (l)Q Q
h2t
Q QH2 H2 h2t
k

k,l

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

ht

53



(k)
QH2 + h.c. ,
H2 qu(k) + U (k) qh2 + Qh2 u(k) + mB
kU

(2.13)

where we have defined h2t = h(5)2


/R, mFk = k/R and mB
t
k = (k + )/R and made use

1
of the representation (0) = (R)
1. Had we kept two separate half towers of
fermions one could combine the nonzero modes to give a Dirac fermion.

3. One-loop contribution to the Higgs mass


In this section we compute, for the sake of completeness and to fix some notations, the
one-loop contribution to the Higgs mass. We define the propagators of the bulk fields with
end points at y = 0 as:
D (p) =


n=

p2




R  
1
n+ 2 = 2p cot (pR + ) + cot (pR ) ,
( R )
(3.1)

n+
R

(p) =
D
2 ( n+ )2
p
R
n=




R  
cot (pR + ) cot (pR ) .
=
2
(3.2)

For Euclidian momenta (p ip)






R 
coth (pR + i) + coth (pR i) ,
2p





R

(p) =
coth (pR + i) coth (pR i) ,
D
E
2i

DE
(p) =

(3.3)

the propagators have the asymptotic expansions

DE
(p) =


R 
1 + 2e2Rp cos(2) ,
p

(3.4)

E
D
(p) = 2Re2Rp sin(2),

(3.5)

(p) has a leading term 1/p which is independent


From Eq. (3.4) we can see that DE
of while the remaining part decays exponentially. This reflects the soft nature of the
supersymmetry breaking by the SS mechanism.
At one loop there are four diagrams which are shown in Fig. 1. They contribute [2] to
the Higgs mass as



2

 1
(mB
d 4q
k)
2
2
0


+
mH = iNc ht
+ D (q)
2D (q) +
2
(2)4
q2
q 2 q 2 (mB
k)
k

Nc h2t
4 2



0
dq DE
(q) DE
(q) .

(3.6)

54

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

Fig. 1. The diagrams contributing to the Higgs mass at one loop.


(q) D 0 (q) decays exponentially (as can be seen from Eq. (3.4)) so
The difference DE
E
the result is finite. The integral has been evaluated in Refs. [1,2,8]:




Nc h2t   2i 
Li3 e
+ Li3 e2i 2(3) .
(3.7)
4
2
16 R
There is a cancellation of tree level infinities between the second and third diagrams. This
cancellation always occurs according to the diagrammatical rule
m2H =

= ih2t q 2 D (q).

(3.8)

The corresponding rule will be to drop all diagrams containing (0) and replace in the
remaining ones


m2k

q 2 m2k

q 2 D (q).

(3.9)

4. Two-loop contribution to the Higgs mass


In this section we compute the leading two-loop contributions to the Higgs mass coming
from the top/stop sector, as we did at one-loop in Section 3, i.e., we consider O(h4t )
corrections. We use for that the Lagrangian (2.13). Inclusion of either gauge couplings
or a = 0 term will not alter the qualitative conclusions of our analysis. We will comment
more on that in Section 5.
In order to display explicitly the various cancellations we group the genuine two-loop
diagrams as shown in Figs. 28, and the one-loop diagrams with counterterm insertions in
Figs. 9 and 10. In all cases external legs correspond to H2 fields.
4.1. Two-loop diagrams
First there are the three diagrams shown in Fig. 2. The third one does actually not exist
since there is no corresponding contraction providing the corresponding propagators (fields
are complex). In all diagrams containing fermions, the terms proportional to the masses
cancel, since the masses are zero for q and h and for the case of the u(k) cancellation

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

55

Fig. 2. Diagrams which vanish identically.

occurs due to symmetric summation over an odd function of k. For this reason diagrams
0a and 0b are zero when summed over.
Contractions are done using Weyl fermions. The relevant propagators are:

 
pm m
T (p) = 2 2 ,
p m

  
pm
T (p) = 2 m 2
p m

(4.1)
(4.2)

and the contraction rules:


Tr m n = 2mn ,



Tr m n r s = 2 mn rs mr ns + ms nr

(4.3)

are used.
There will be two kinds of cancellation between bosons and fermions:
If a loop involves only particles which are localized on the brane, there will be
the usual supersymmetric cancellations of quadratic divergences. Supersymmetry
breaking does not affect these particles (both fermions and bosons remain massless).
If a loop involves particles moving in the bulk, one expects cubic divergences by
power counting. In the supersymmetric theory ( = 0) these terms are cancelled exactly by the sfermion. For = 0 fermions and bosons get different masses. However, after summation over the modes (integration over the loop fifth momentum) the
-dependence lies in the propagators D and D 0 whose difference is exponentially
suppressed for large Euclidian four-momentum (see Eq. (3.4)). So the cancellation is
no longer exact but leaves over a UV-finite piece.
We use in the following the definitions k = p + q, l = p q and the identities
2p q = k 2 p2 q 2 = l 2 + p2 + q 2 ,
2p k = q 2 + p2 + k 2 ,

2q k = p2 + q 2 + k 2 ,

(4.4)

to eliminate the scalar products and express all integrands in terms of p2 , q 2 and k 2 . We
will suppress the integrations and the common factor of h4t . The diagrams of the first group
are shown in Fig. 3.
The calculation of the diagrams in Fig. 3 leads to expressions d1 = d1a + d1b + d1c + d1d :


 2 p k
 2
q2
1
1
= iNc D (q) 2 2 + 2 + 2
d1a = 2iNc D (q)
p2 k 2
p k
p
k


2

2
2
q
= iNc D (q) 2 2 + 2 .
(4.5)
p k
p

56

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

The possibilityof the renaming k p becomes obvious if one writes the integration over
k explicitly as d 4 k (p + q k) and observes that the sharp cutoff has to be introduced
symmetrically for all three momenta:
 2 1

(q)
,
d1b = iNc D
k 2 p2

(4.6)


2 1
.
d1c = d1d = iNc D (q)
p2

(4.7)

Assembling the diagrams of Fig. 3 we get



 2 1
2 q 2

(q)
d1 = iNc D (q)
+ iNc D
.
2
2
p k
k 2 p2

(4.8)

Note the supersymmetric cancellation of the quadratic divergences in the p-integration.


The cancellation involves only loops in the brane and it is identical to the one that would
occur in an unbroken theory.
Now consider the class of diagrams of Fig. 4:


 2 p k
 2
1
1
1

(q)

=
iN
(q)

+
D
d2a = 2iNc D
(4.9)
c
q 2 p2 k 2
q 2 p2 p2 k 2 q 2 k 2


 2
1
2

(q)
= iNc D
(4.10)

,
q 2 p2 p2 k 2

Fig. 3. Two-loop diagrams computed in Eqs. (4.5)(4.7). Diagrams 1c and 1d have the same value.

Fig. 4. Two-loop diagrams computed in Eqs. (4.9)(4.12). Diagrams 2b and 2c are equal. The last
three diagrams combine according to the rule (3.8). Note there is indeed a combinatorial factor of 2
for diagram 2e, as needed.

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

 2 1

(q)
d2b = d2c = iNc D
,
q 2 p2

2 q 2
.
d2d + d2e + d2f = iNc D (q)
p2 k 2

57

(4.11)
(4.12)

Adding these we get a cancellation of the quadratic divergence in the p-integration:



 2 1
2 q 2

(q)
d2 = iNc D (q)
+ iNc D
.
2
2
p k
k 2 p2

(4.13)

Again this cancellation is the unmodified supersymmetric cancellation.


The next contribution comes from correction of the u propagator (see Fig. 5):
2 q p q 2

d3a = d3b = 2iNc D 0 (q)
p2 q 2 l 2




2
2 q 2
1
q2
1
= iNc D 0 (q) 2 + 2 + 2 2 = iNc D 0 (q)
.
p
k
p k
p2 k 2

(4.14)

Adding d1 to d3 we get

2 
2  q 2
 2 1

(q)
+ 2iNc D
. (4.15)
d1 + d2 + d3 = 2iNc D (q) D 0 (q)
2
2
p k
k 2 p2
Now the q-integration is completely finite. The remaining logarithmic divergence of the
p-integration will be associated to the wave-function renormalization (WFR) of the bulkfield.

Fig. 5. Two-loop diagrams computed in Eq. (4.14).

Fig. 6. Two-loop diagrams computed in Eqs. (4.16)(4.18). They have to be combined according to
rule (3.8), i.e., 4a + 4b, 4c + 4d, 4e + 4f + 4g + 4h.

58

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

Next we consider the diagrams from Fig. 6. All of them are infinite by themselves
(having (0) vertices or divergent sums) but combining them as below these divergences
cancel:
pk
d4a + d4b = 2iNc D (q)D 0 (p) 2 2
q k


1
p2
1
= iNc D (q)D 0 (p) 2 + 2 2 + 2 ,
(4.16)
k
q k
q
d4c + d4d = iNc D (q)D (p)

1
,
q2

d4e + d4f + d4g + d4h = iNc D (q)D (p)

(4.17)
p2
.
q 2k2

(4.18)

Summing these up we find a supersymmetric cancellation of a cubic divergence in the


integration over p:



 p2
1
1

0
+ 2 + iNc D (q)D 0 (p) 2 .
d4 = iNc D (q) D (p) D (p)
(4.19)
2
2
q k
q
k
We also have the corrections to the fermionic (q) propagator (see Fig. 7):


q k
p2
1
1
d5a = 2iNc D (p)D 0 (q) 2 2 = iNc D (p)D 0 (q) 2 2 + 2 + 2 ,
q k
q k
k
q
(4.20)
d5b = 2iNc D 0 (q)D 0 (p)



pq
1
1
p2
0
0
=
iN
D
(q)D
(p)

+
+
.
c
q 2l2
q 2 k2 q 2k2

(4.21)
They combine to give a supersymmetric cancellation of the cubic divergence in the
integration over p:




p2
1
0
0
d5 = iNc D (q) D (p) D (p) 2 2 + 2
q k
q


1
iNc D 0 (q) D (p) + D 0 (p) 2 .
(4.22)
k
Summing up the two contributions we get:


1
d4 + d5 = iNc D (q) D 0 (q) D (p) D 0 (p) 2
q

Fig. 7. Two-loop diagrams computed in Eqs. (4.20)(4.21).

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963


 p2

+ iNc D (q) + D 0 (q) D (p) D 0 (p) 2 2
q k


 1
+ iNc D (q)D 0 (p) D 0 (q) D (p) + D 0 (p) 2 .
k

59

(4.23)

There is a linear divergence in the p-integration coming from the last term. It is
associated with the WFR of the matter fields living on the brane and will be cancelled
by the usual supersymmetric counterterm.
To make the q-integration finite we look at the three remaining diagrams (see Fig. 8)
1
,
k2

(q) 1 ,

(p)D
d7 = iNc2 D
q 2 p2

(q) 1 ,

(p)D
d8 = 2iNc D
k 2 p2
d6 = iNc D (q)D (p)

(4.24)
(4.25)
(4.26)

d7 and d8 are UV finite according to the expansion (3.5), while d6 is exactly the missing
piece in Eq. (4.23).
We end up with:

Nc h4t

d 4p d 4q
(2)4 (2)4

 
2  q 2
 2 1
 2  0

E (q)
(q) DE (q)

2
D
2 DE
p2 k 2
k 2 p2



1
0

0
+ DE (q) DE
(q) DE
(p) DE
(p) 2
q



p2
0

0
+ DE (q) + DE
(q) DE
(p) DE
(p) 2 2
q k



1
0

0
+ DE (q) DE
(q) DE
(p) + DE
(p) 2
k

1
1

Nc DE (p)DE (q) 2 2 2DE (p)DE (q) 2 2 .


q p
k p

(4.27)

One sees that the only remaining divergences are in the p-integration, more specifically
there is the logarithmic divergence in the first line and the linear divergence in the fourth.
The expression vanishes for = 0.

Fig. 8. Last three two-loop contributions, computed in Eqs. (4.24)(4.26). Diagrams 7 and 8 are
UV-finite.

60

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

4.2. The counterterms


As in any (exact or softly broken) supersymmetric theory the superpotential couplings
run due to the wave-function renormalization of the superfields. We would like to rescale
the fields which couple to the brane in a manifestly supersymmetric way. To this end we
rescale the fields as follows:




Q
Q
Q,
q 1
q,
Q 1
(4.28)
2
2




H
H
h2 1 2 h2 ,
H2 1 2 H2 ,
(4.29)
2
2




U
U
R(y) U,
u 1
R(y) u,
U 1
(4.30)
2
2




U
U
y U  1 +
(4.31)
R(y) y U  ,
R(y) y u ,
y u 1 +
2
2
y U and U  do not couple to the brane and thus will not receive a brane rescaling. We
define also a rescaled (renormalized) coupling as


U
Q H2

R (0) hbare
ht = 1
(4.32)
t .
2
2
2
By inspection of Eq. (2.2) it is obvious that this leaves the Lagrangian unchanged if
expressed in rescaled quantities. The different sign for U and y U  guarantees that the first
term of the superpotential remains unaltered.
The following counterterms now appear in the 4D Lagrangian:


LCT = Q Q Q H2 H2 H2 U R U U + y U  y U  y=0


Q q q H2 h2 h 2 U R u u y=0




Q h2t R U U y=0 H2 H2 H2 h2t R U U y=0 Q Q

2




U h2t R(0) Q QH2 H2 + U ht R(0) QH2 y U  y=0 + h.c.

= Q Q Q H2 H2 H2 U
U (k) U (l)
U

k,l

mk U

(k)

ml U

(l)

k,l

Q q q H2 h2 h 2 U
Q h2t


k,l

U h2t

u(k) u (l)

k,l

(k)

(l)

H2 H2

H2 h2t

U (k) U (l) Q Q

k,l





2

(k)
mk U QH2 + h.c. .
R(0) Q QH2 H2 + U ht R(0)
k

(4.33)

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

61

The values of U and Q are determined by the infinite parts of the one-loop two point
functions at nonvanishing momentum q. One finds:

B
2
 
d 4 p mB
k ml + q
Uk Ul = h2t
(2)4
p2 k 2
 2
 B B


2
2 1
+ finite,
= iht mk ml + q
log
(4.34)
8 2
2



p2
q2 0
d 4p 
0
1
+
+
(p)

D
(p)
D (p)
D
(2)4
k2
k2
R
+ finite,
= ih2t q 2
(4.35)
8
where is an infrared cutoff. The infinities can be absorbed if one chooses
 2

2 1
U = ht
(4.36)
,
log
8 2
2
R
.
Q = h2t
(4.37)
8
There are now the additional diagrams shown in Figs. 9 and 10 (U counterterms are
denoted by and Q counterterms by ).
The diagrams in Fig. 9 contribute:


2Q D (q) D 0 (q) .
(4.38)



Q Q = h2t

Fig. 9. Counterterm diagrams, computed in Eq. (4.38), proportional to Q .

Fig. 10. Counterterm diagrams, computed in Eq. (4.39), proportional to U .

62

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

The diagrams of Fig. 10 give (observe the factor of two for the third diagram):
 1

m2k
mk
ml  2
1
U
+
m
m
q
+
2
k l
q 2 q 2 m2k q 2 m2k
q 2 q 2 m2k
k,l

 2


2
1
1
1
q + mk ml + 2 2U q 2 D 0 (q)
+
2
2
2
2
q
q mk q mk

 1



1
1
q 2 + mk ml mk ml + 2 q 2 m2l m2k
= U
2
2
2
2
2
q q mk q mk
k,l
2

 



+ q 2 q 2 + mk ml + q 2 m2k q 2 m2l 2U q 2 D 0 (q)
 
2
2  2 


(q) .
= 2U q 2 D (q) q 2 D 0 (q) + D

(4.39)

So the total counterterm contribution is:





d 4q
0
Nc h2t
2Q DE
(q) DE
(q)
4
(2)
 
2
2  2 
 0

E (q) .
2U q 2 DE
(q) q 2 DE
(q) D

(4.40)

Adding them to the divergent terms of (4.27) we get



d 4q
Nc h2t
(2)4
 

 2
 0
2  2  2
2
2

2 q DE (q) q DE (q) DE (q)


ht


d 4p 1
U
(2)4 k 2 p2



1

d 4p 
0
0
(4.41)
(q) DE
(q) h2t
(p) 2 + 2Q
DE (p) + DE
+ DE
4
(2)
k

which is finite if the counterterms are chosen according to (4.36), (4.37).

5. Conclusions
In this paper we have made an explicit two-loop calculation of the Higgs mass in the
five-dimensional model of Ref. [8], where the fifth dimension is compactified on the
orbifold S 1 /Z2 and supersymmetry is broken in the bulk by means of ScherkSchwarz
boundary conditions. We have considered only (top) Yukawa couplings in the calculation
(ht ) and disregarded gauge couplings (g). We have also considered O(0 ) results, which
amounts to disregard the -parameter in the superpotential. Our result is that no Higgs
mass counterterm is generated at two-loop and the only divergent terms in the two-loop
correction can be absorbed by the renormalization group running of ht between the cutoff
and the scale where the theory becomes four-dimensional, 1/R. The latter can be
interpreted as threshold effects at 1/R when matching the 4D theory below it with the
complete KK theory above 1/R. As a result our explicit calculation shows that the Higgs
mass is finite and calculable to the considered order in perturbation theory.

A. Delgado et al. / Nuclear Physics B 613 (2001) 4963

63

Inclusion of gauge couplings has been done at one-loop, O(g 2 ) terms, for an
arbitrary gauge group [1,2]. At two-loop we would have O(g 2 h2t ) and O(g 4 ) terms. We
expect divergent terms which are cancelled by supersymmetry or by counterterms from
supersymmetric wave-function renormalization, which can be absorbed by the running of
gauge and Yukawa couplings, yielding again a finite result.
Inclusion of the -term (a supersymmetric mass) in the superpotential would introduce
divergences, even at the one-loop level, that can be interpreted as the supersymmetric
running of the tree level Higgs mass terms, while the analysis becomes much more involved
since the Higgs mass M2 becomes a two-by-two matrix. All infinities, in this case, are
expected to be absorbed by supersymmetry as well as by the running of gauge and Yukawa
couplings, and of the -parameter. The resulting absence of Higgs mass counterterms
should hold in this case. At one-loop, we have checked that the field rescaling H2 = Nc Q
is indeed sufficient to render M2 finite.

Acknowledgements
The work of A.D. was supported by the Spanish Education Office (MEC) under an FPI
scholarship. The work of G.G. was supported by the DAAD.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]

I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quirs, Nucl. Phys. B 554 (1999) 503.


A. Delgado, A. Pomarol, M. Quirs, Phys. Rev. D 60 (1999) 095008.
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007.
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0106190.
R. Barbieri, L.J. Hall, Y. Nomura, hep-th/0107004.
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81.
A. Delgado, M. Quirs, Phys. Lett. B 484 (2000) 355.
A. Delgado, M. Quirs, hep-ph/0103058.
For recent reviews see: I. Antoniadis, K. Benakli, Int. J. Mod. Phys. A 15 (2000) 4237;
I. Antoniadis, hep-th/0102202.
I. Antoniadis, K. Benakli, M. Quirs, Nucl. Phys. B 583 (2000) 35.
Y. Nomura, hep-ph/0105113.
N. Weiner, hep-ph/0106021.
A. Masiero, C.A. Scrucca, M. Serone, L. Silverstrini, hep-ph/0107201.
D. Ghilencea, H.P. Nilles, Phys. Lett. B 507 (2001) 327.
A. Delgado, G. von Gersdorff, P. John, M. Quirs, hep-ph/0104112.
R. Contino, L. Pilo, hep-ph/0104130.
N. Arkani-Hamed, T. Gregoire, J. Wacker, hep-th/0101233;
N. Arkani-Hamed, L. Hall, D. Smith, N. Weiner, Phys. Rev. D 63 (2001) 056003.
J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
P. Fayet, Phys. Lett. B 159 (1985) 121;
P. Fayet, Nucl. Phys. B 263 (1986) 649.

Nuclear Physics B 613 (2001) 6486


www.elsevier.com/locate/npe

F 5 contributions to the non-Abelian BornInfeld


action from a supersymmetric YangMills
five-point function
Andrea Refolli, Alberto Santambrogio, Niccolo Terzi, Daniela Zanon
Dipartimento di Fisica dellUniversit di Milano and
INFN, Sezione di Milano, Via Celoria 16, 20133 Milano, Italy
Received 26 June 2001; accepted 25 July 2001

Abstract
We consider the N = 4 supersymmetric YangMills theory in four dimensions. We compute the
one-loop contributions to the effective action with five external vector fields and compare them with
corresponding results in open superstring theory. Our calculation determines the structure of the F 5
terms that appear in the non-Abelian generalization of the BornInfeld action. The trace operation on
the gauge group indices receives contributions from the symmetric as well as the antisymmetric part.
We find that in order to study corrections to the symmetrized trace prescription one has to consistently
take into account derivative contributions not only with antisymmetrized products [ ] but also
with symmetrized ones ( ) . 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.15.-q; 11.30.Pb

1. Introduction
Scattering amplitudes of massless modes in string theories can be described in terms
of effective Lagrangians. For the open string case the Abelian BornInfeld action [1]
represents a remarkable example of an effective field theory which contains string
corrections to all orders in  [24]. The contributions can be summed up to all orders in the
case of a constant, Abelian field strength. As soon as these two conditions are relaxed the
problem becomes complicated and a complete action for such fields has not been obtained.
A non-Abelian generalization of the BornInfeld action can be defined as suggested in [5]
using a symmetrized trace operation over the gauge group matrices. However there are
E-mail addresses: andrea.refolli@mi.infn.it (A. Refolli), alberto.santambrogio@mi.infn.it
(A. Santambrogio), niccolo.terzi@mi.infn.it (N. Terzi), daniela.zanon@mi.infn.it (D. Zanon).
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 1 ) 0 0 3 6 8 - 6

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

65

indications that this prescription might not be sufficient to include all the contributions that
one would obtain from an open string approach [68].
The best available way to construct the effective action for the non-Abelian, non constant
curvature case, is to proceed order by order. One has to compute corrections with increasing
number of derivatives and of field strengths. A well established result is up to the order
 2 [9,10]. As soon as one focuses on higher orders the calculations become difficult.
The inclusion of supersymmetry seems to be quite useful to gain insights and it might
set enough constraints to fix uniquely the form of the action. Several attempts in different
directions are under consideration [7,8,11].
In this paper we attack the problem as follows: since we want to make contact
with the ten-dimensional open superstring we consider its four-dimensional field theory
limit, i.e., the N = 4 supersymmetric YangMills theory. Then we compute at one-loop
perturbatively: the idea is that in this way we construct an effective action which is
supersymmetric and generalizes the YangMills theory. If supersymmetry determines the
form of the allowed deformations [12] it should correspond to the non-Abelian Born
Infeld theory. We set the external background on-shell and compute one-loop n-point
vector amplitudes. These functions are in general non-local expressions that contain a loopmomentum integral. For a given n-point amplitude we perform a low-energy expansion
in the external momenta: this is achieved through the introduction of an infrared mass
regulator that in the expansion plays the same role as  , i.e., it keeps track of the order of
the derivative term under consideration. The net result is that each n-point amplitude gives
rise to an infinite series of local terms with higher and higher number of derivatives. The
leading term is automatically gauge invariant. The subleading contributions which contain
derivatives are not gauge-invariant by themselves. The correct covariantization is obtained
including corrections from amplitudes with a higher number of external background fields
through a mechanism similar to the one in [13] .
We present the calculation of the four- and five-point functions, and of the part of
the six-point function which is needed for the above mentioned covariantization of the
lower order results. These computations will allow us to evaluate the complete gauge
invariant structures for the F 4 and the F 4 , F 5 contributions. Since the symmetric trace
prescription would rule out F 5 terms, the nonvanishing result that we find confirms that
the actual form of the non-Abelian BornInfeld action is definitely richer. In fact the trace
operation on the gauge group indices receives contributions from the symmetric as well as
the antisymmetric part.
In the next section we briefly recall the N = 1 superspace formulation of the N = 4
YangMills action and describe the main ingredients that enter the quantization via the
background field approach. Moreover we outline the general procedure that, starting from
the one-loop YangMills amplitudes, allows us to determine corrections to the non-Abelian
BornInfeld action. In Section 3 we compute the one-loop amplitude with four external
vector fields and study its low-energy expansion. We extract from the superfield result its
component content and in particular we study the bosonic contributions which contain the
field strengths F . In Section 4 we compute in the same manner the five-point function.
The calculation is quite complicated, on one hand because of the gauge group structure,

66

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

on the other hand because of the existence of several on-shell identities which makes it
difficult to express the result in a canonical form. Finally in Section 5 we compute the part
of the six-point function that we use to complete the covariantization of the terms with
four field strengths and two derivatives. In Section 6 we collect all our results and compare
them with corresponding results from open superstring theory [14].

2. The N = 4 YangMills action and its quantization


The N = 4 supersymmetric YangMills classical action written in terms of N = 1
superfields (we use the notations and conventions adopted in [15]) is given by



1
1
1
4
4
V  V i
4
2


 W
d x d e i e +
d x d W W +
d4 x d2 W
S = 2 Tr
g
2
2



 j k 1
1
4
2
i
4
2 ij k   
+
(2.1)
d x d iij k , +
d x d i i [j , k ] ,
3!
3!
 2 (eV D eV )
where the i with i = 1, 2, 3 are three chiral superfields, and the W = i D
are the gauge superfield strengths. All the fields are Lie-algebra valued, e.g., i = ai T a ,
in the adjoint representation of the gauge group, with [Ta , Tb ] = ifabc Tc .
At the component level the three chiral superfields i contain six spin-0 particles and
three spin- 21 Weyl spinors, while W describes one spin- 21 spinor and one gauge vector
particle. Since we are interested in computing field strength corrections to the BornInfeld
action, we will consider amplitudes with vector fields as external background. Moreover
we will extract from them only the gauge field bosonic components using the relations
1
D( W) |=0 = f = ( ) F ,
2
( W
 |=0 = f = 1 ( ) F
D
(2.2)

)
2
with and defined as in (A.3). In this way we will be able to isolate the relevant
contributions.
In order to evaluate one-loop amplitudes we need to quantize the theory and this is
most efficiently done using the background field method [16,17] which guarantees a gauge
invariant result for the effective action. For the gauge multiplet one performs a non linear
splitting between the quantum prepotential V and the background superfield, via covariant
derivatives (in quantum chiral representation [15])
= eV D eV eV B eV ,

B .
 = D

(2.3)

In this way the external background enters the quantum action implicitly in the background
covariant derivatives through the connections
B = D i ,

B = D
 i
 ,

aB = a i a

(2.4)

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

67

B , {
B , B }]. Background
and explicitly in the background field strength W = 2i [

covariant gauge-fixing introduces additional terms





1
B2 B2 V + SFP + SNK
B2 +
2 Tr d4 x d4 V B2
(2.5)
2g
with FaddeevPopov action



1 
4
4



SFP = Tr d x d c c c c + (c + c )[V , c + c]
+
2
and NielsenKallosh ghost action


SNK = Tr d4 x d4 bb.

(2.6)

(2.7)

 c =
The three ghosts c, c and b are background covariantly chiral superfields, i.e.,
B




B c = B b = 0. In the following we drop the suffix B from the covariant derivatives.
After gauge-fixing the quadratic quantum V -action becomes


 1
1
i
4
4
S 2 Tr d x d V  i
2
2
2g

 (D
 i ) V .
iW (D i ) i W
(2.8)
As mentioned above we want to compute one-loop contributions to the effective action
with external vector fields. The N = 4 theory is particularly simple since this type of terms
are produced only by quantum vector loops. The loops with the three chiral matter fields
are cancelled by the three ghosts: in fact each ghost contributes to the one-loop effective
action exactly as a standard chiral superfield would do, the only difference being an overall
opposite sign because of the statistics.
Thus we focus on the action in (2.8). The interactions with the background are at most
 are needed in the loop
linear in the D-spinor derivatives and at least two Ds and two Ds
in order to complete the D-algebra. Thus the first non-vanishing result is at the level of
the four-point function [18]. Here we describe the general procedure we have adopted in
order to extract from the super YangMills effective action the wanted contributions to the
non-Abelian BornInfeld action.
A one-loop n-point amplitude will contain n external background fields from the
 in order to complete the D-algebra.
interactions in (2.8), with at least two W and two W
We label the external momenta p1 , p2 , . . . , pn and introduce a general notation for the
dependence on the gauge group (with structure constants fabc )
g(a1 , a2 , . . . , an ) = fx1 a1 x2 fx2 a2 x3 fxn an x1 .

(2.9)

In the following we treat the gauge fields as matrices in the adjoint representation, i.e.,
Wac fabc Wb . We define
TrAd (AB ) Aa Bb g(a, b, . . .).

(2.10)

68

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

Moreover we set the external fields on-shell, i.e., we freely use the equations of motion
W = 0,

 = 0.
 W

(2.11)

In this way a n-point function gives rise to a background field dependence which contains
 of
a total of n fields, with a given number of connections , and of field strengths W, W,
the general form



 (pi ) W (pj ) (pk ) W
 (pl )

 W
d4 TrAd W (p1 ) W
(pn ) .
(2.12)
From (2.12) we want to extract the relevant structures that might produce F s at the
component level. To this end we need convert


1
 D

d4 d2 d2  D D D
(2.13)
4
 using the relations in (2.2).
and act with these four spinor derivatives on the W and W
One can reconstruct gauge covariant expressions also from the connections , but this
can happen only at the level of the six-point function and beyond. Since in this paper we
consider structures up to the five-point functions we leave this point for a future discussion.
In conclusion we obtain the structures we are looking for if our one-loop diagram
.
( W
has produced products of fields D( W) and their hermitian conjugate ones D
)
In addition we have to deal with a loop-momentum integral which contains n scalar
propagators and momentum factors directly from the vertices in (2.8) and/or from
commutators of spinor derivatives produced while performing the D-algebra

hn (k, pi )
.
In = d4 k 2
(2.14)
k (k + p2 )2 (k + p2 + p3 )2 (k + p2 + + pn )2
We can rewrite (2.14) as an infinite series of local terms in a low-energy expansion with
higher derivatives. We introduce an IR mass M and expand the propagators keeping the
external momenta small as compared to M. First we Feynman combine the propagators in
(2.14)
1
= (n)
A1 A2 An

1

x1
dx2

dx1
0

xn2

dxn1
0

[A1xn1 + A2 (xn2 xn1 ) + + An (1 x1 )]n .


Then shifting the k-momentum we obtain
1
k 2 (k + p2 )2 (k + p2 + p3 )2 (k + p2 + + pn )2
xn2

1
x1
4
 (n 1)! d k dx1 dx2
dxn1
0


k 2 + M 2 + (1 x1 )(p2 + + pn )2 + + (xn2 xn1 )(p2 )2

2 n
(1 xn1 )p2 + + (1 x1 )pn

(2.15)

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

1
n!
2
(k + M 2 )n


d4 k

dxi

Hn (xi , pj )
+ ,
(k 2 + M 2 )n+1

69

(2.16)

where
Hn (xi , pj ) = (1 x1 )(p2 + + pn )2 + + (xn2 xn1 )(p2 )2
[(1 xn1 )p2 + + (1 x1 )pn ]2 .
All the integrals over the k-momentum are finite and give

1
2
1
d4 k 2
=
,
(k + M 2 )n (n 2)(n 1) (M 2 )n2

k k
1
2
=
C C ,
d4 k 2
(k + M 2 )n (n 3)(n 2)(n 1) (M 2 )n3
..
.

(2.17)

(2.18)

The non-local expression In has been traded by an infinite sum of local terms. In the next
two sections we apply this procedure to the computation of the four- and the five-point
functions and show explicitly how we can determine order by order structures that we
interpret as corrections to the non-Abelian BornInfeld action.

3. The four-point function


Now we consider one-loop contributions with four external vector fields. They
correspond to box-type diagrams as the one shown in Fig. 1. We label the external momenta
p1 , p2 , p3 , p4 and use the notation introduced in (2.9) with n = 4. A diagram like the one
in Fig. 1 gives rise to a momentum integral

1
I4 = d4 k 2
(3.1)
2
k (k + p2 ) (k + p2 + p3 )2 (k + p2 + p3 + p4 )2
and to a background field dependence



 (p3 )W
 (p4 ) .
d4 TrAd W (p1 )W (p2 )W

Fig. 1. Four-point amplitude.

(3.2)

70

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

The complete answer is obtained by summing to the above expression all permutations in
the 2, 3, 4 indices. We are interested in rewriting the superspace result in components and
in particular we want to study the bosonic gauge field contributions. To this end we use the
definitions in (2.2) and the relations in the appendices. First we obtain

 (p3 , a3 )W
 (p4 , a4 )
d4 W (p1 , a1 )W (p2 , a2 )W

f4 ) ,
 (f1 ) (f2 ) (f3 ) (

(3.3)

where fi f (pi , ai ). Then using (A.6), (B.3) and (B.4) we have



1
 2 
b ) ,
F ab + i(Fa ) (F
(fa ) (fb ) =
4

1
 2 

b ) .
F ab i(Fa ) (F
(fa ) (fb ) =
4
In general we use the notation
 n
F a an (Fa1 )1 2 (Fa2 )2 3 (Fan )n 1

(3.4)
(3.5)

(3.6)

with the only exception of


 2
F ab (Fa )1 2 (Fb )1 2

(3.7)

which we keep with indices contracted in standard manner. Making use of (A.10) and
the final result becomes
(A.11) to eliminate F

1
g(a1 , a2 , a3 , a4 )
d4 k 2
k (k + p2 )2 (k + p2 + p3 )2 (k + p2 + p3 + p4 )2
 
 
1  
4 F4 a a a a + 4 F4 a a a a + 4 F4 a a a a
1 2 3 4
1 2 4 3
1 3 2 4
8

 2
 2
 
 2
 2
 2
F a a F a a F a a F a a F a a F2 a a .
(3.8)
1 2

3 4

1 3

2 4

1 4

2 3

Now following the general procedure outlined in the previous section we rewrite (3.8) as an
infinite series of local terms in a low-energy expansion with higher derivatives. Feynmann
combining the propagators in (3.1) we obtain

1
I4 =
d4 k 2
2
k (k + p2 ) (k + p2 + p3 )2 (k + p2 + p3 + p4 )2

1
x1
x2
4
 3! d k dx1 dx2 dx3


k + M + (1 x1 )(p2 + p3 + p4 )2 + (x1 x2 )(p2 + p3 )2


2 4

+ (x2 x3 )(p2 )2 (1 x3 )p2 + (1 x2 )p3 + (1 x1 )p4
 4



H4 (xi , pj )
1
p
4
=
d4 k 2

4!
d
k
dx
+
O
,
i
2
4
2
2
5
(k + M )
(k + M )
M8
2

(3.9)

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

71

where
H4 (xi , pj ) = (1 x1 )(p2 + p3 + p4 )2 + (x1 x2 )(p2 + p3 )2 + (x2 x3 )(p2 )2
2

(1 x3 )p2 + (1 x2 )p3 + (1 x1 )p4 .
(3.10)
First we look at the leading term in (3.9). It does not depend on the external momenta
and therefore in (3.8) we can freely symmetrize on them. We obtain the F 4 non-Abelian
contribution to the BornInfeld action

 
 
4  g(a1 , a2 , a3 , a4 ) F 4 a a a a + 2 F 4 a a a a .
1 2 3 4

1 2 4 3

 
 

 
1  2 

F a a F2 a a + 2 F2 a a F2 a a
1
3
2
4
1
2
3
4
4

= TrAd F F F F + 2F F F F



1 
F F F F + 2F F F F .
4

(3.11)

We notice that if the gauge group is SU(N) we can express the gauge group matrices which
appear in (3.11) using the standard fundamental representation. If we do so we find that
the planar sector (large N limit) reproduces the symmetrized trace structure [5]


  
4  Tr T a1 T a2 T a3 T a4 + permutations F 4 a


1  2 2
4
STr F F
.
4

1 a2 a3 a4

 
1  2
F a a F2 a a
1
2
3 4
4

(3.12)

Next we study the subleading contribution which corresponds to terms with two derivatives. Including the overall factor (1/2)4 (1/2!)223 from the vertices and combinatorics,
the integral containing H4 (xi , pj ) in (3.10) gives



1 2 1 
10(p2 )2 + 8p2 p3 + 2p2 p4 g(a1 , a2 , a3 , a4 )
6
40 12 M
 
 
 
 
1   4 
4 F a a a a + 4 F4 a a a a + 4 F4 a a a a F2 a a F2 a a

1
2
3
4
1
2
4
3
1
3
2
4
1
2
3 4
8
 2
 2


 2
 2
F a a F a a F a a F a a + permutations a2 , a3 , a4 .

4(der) =

1 3

2 4

1 4

2 3

(3.13)
These terms give rise to contributions that are not gauge invariant since from p i
ordinary derivatives are produced. The correct covariantization of the result is obtained
adding terms with one and two background connections from the five- and six-point
functions, respectively. In the next sections we will compute these terms explicitly.

72

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

4. The five-point function


The one-loop contributions to the effective action corresponding to diagrams with five
external vector fields group themselves into two distinct classes:
 vertices (and corresponding hermitian conjugate
(i) graphs with two W and three W
ones) which produce gauge-invariant structures;
 vertices and one vertex which, as we will show,
(ii) graphs with two W, two W
contribute to the covariantization of the four-point function.
We start computing the first class of terms.
 3 five-point terms
4.1. The W 2 W
 vertices as in Fig. 2(a). After
Now we consider diagrams with two W and three W
completion of the D-algebra in the loop we obtain terms with five scalar propagators and
a typical background dependence of the form

 (p3 , a3 )D
 (p5 , a5 ).
 (p4 , a4 )W
 W
d4 W (p1 , a1 )W (p2 , a2 )W
(4.1)

Including the combinatorial factors and the loop-momentum integration we obtain the
corresponding bosonic contributions in terms of the field strengths
5(1)



1 51 1 4

2 (f1 ) (f2 ) (f3 ) (f4 ) (f3 ) g(a1 , a2 , a3 , a4 , a5 )


=
2 2! 3!

1
d4 k 2
2
k (k + p2 ) (k + p2 + p3 )2 (k + p2 + p3 + p4 )2
1

.
(4.2)
(k + p2 + p3 + p4 + p5 )2

(a)

(b)
Fig. 2. Five-point amplitudes.

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

Once again in order to simplify the notation we define


 n
f a a an (fa1 ) (fa2 ) (fan ) ,
1 2
 n

f a a an (fa1 ) (fa2 ) (fan ) .


1 2

73

(4.3)

It is easy to show that (f n )a1 a2 an is symmetric under cyclic permutation of the indices
a1 a2 an for n even, while it is antisymmetric for n odd. Using the definitions in (4.3)
(1)
we denote the contributions corresponding to 5 simply as (f 2 )a1 a2 (f3 )a3 a4 a5 . In the
same way we obtain another graph interchanging the external background W(p2 , a2 ) with
 3 , a3 ), e.g.,
W(p


 
1 5 1 1 4 2
(2)
2 f a a f3 a a a g(a1 , a3 , a2 , a4 , a5 )
5 =
1 2
3 4 5
2 2! 3!

1
d4 k 2
k (k + p3 )2 (k + p2 + p3 )2 (k + p2 + p3 + p4 )2
1

,
(4.4)
(k + p2 + p3 + p4 + p5 )2
In addition to (4.2) and (4.4) we have graphs obtained through a permutation of couples
of indices (pi , ai ), i = 3, 4, 5 and (p1 , a1 )(p2 , a2 ). In this way we end up with a total
of 3! 2! 2 = 4! inequivalent contributions. The final result can be rewritten in terms of
the field strengths F . Indeed using the formulas (B.5) and (B.6) in Appendix B we can
express the product of three fs (or equivalently three f s) as
1   3
c )
2 F abc + i(Fa ) (Fb ) (F
8


a )
+ i(F
(Fb ) (Fc ) .

(fa ) (fb ) (fc ) =

(4.5)

In the same way using (B.3) and (B.4) two fs (or two f s) can be rewritten as in (3.4) and
(3.5). The product of (3.4) and (4.5) gives

1  2 
a2 )
F a a + i(Fa1 ) (F
1
2
32
  

a5 ) + i(F
a3 ) (Fa4 ) (Fa5 ) .
2 F 3 a a a + i(Fa3 ) (Fa4 ) (F
3 4 5

(4.6)

and adding the hermitian conjugate


Finally using (A.10) in order to reexpress the two F
contributions, we obtain

 
 
1 5
2Fa1 a4 a3 a2 a5 2Fa51 a4 a5 a2 a3 + 2Fa51 a5 a4 a2 a3
 f 2 a a f3 a a a =
1 2
3 4 5
32
2Fa51 a3 a4 a2 a5 + 2Fa21 a2 Fa33 a4 a5 Fa21 a5 Fa32 a3 a4

Fa22 a5 Fa31 a3 a4 Fa21 a3 Fa32 a4 a5 Fa22 a3 Fa31 a4 a5 .
(4.7)
Now, as we have done in the previous section, we study the loop momentum integral in the
low-energy approximation

1
d4 k 2
I5 =
2
k (k + p2 ) (k + p2 + p3 )2 (k + p2 + p3 + p4 )2

74

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

1
(k + p2 + p3 + p4 + p5 )2
1
x1
x2
x3

4
 4! d k dx1 dx2 dx3 dx4

k + M + (1 x1 )(p2 + p3 + p4 + p5 )2 + (x1 x2 )(p2 + p3 + p4 )2


2

+ (x2 x3 )(p2 + p3 )2 + (x3 x4 )(p2 )2 (1 x4 )2 (p2 )2 (1 x3 )2 (p3 )2


5
(1 x2 )2 (p4 )2 (1 x1 )2 (p5 )2
 4 



H5 (xi , pj )
1
p
4
4
=
d k 2
5! d k dxi 2
+O
(4.8)
,
2
5
2
6
(k + M )
(k + M )
M 10
where
H5 (xi , pj ) = (1 x1 )(p2 + p3 + p4 + p5 )2 + (x1 x2 )(p2 + p3 + p4 )2
+ (x2 x3 )(p2 + p3 )2 + (x3 x4 )(p2 )2
2

(1 x4 )(p2 ) + (1 x3 )(p3 ) + (1 x2 )(p4 ) + (1 x1 )(p5 ) .
(4.9)
We are interested only in the leading order contributions which again are independent of
the external momenta p1 , . . . , p5 , so that the various terms can be combined as




 2
 3
1
1
(tot)
4

d k 2

f
f
5 =
g(a1 {a2 , a3 }, a4 , a5 )
a1 a2
a3 a4 a5
12
(k + M 2 )5

+ permutations of a1 , a2 and a3 , a4 , a5 .
(4.10)
Taking into account the symmetry properties of (f n )a1 a2 an , we obtain




 3
 2
1
1
(tot)
4

5 =

f
d k 2
f
a1 a2
a3 a4 a5
12
(k + M 2 )5


2 g(a1 {a2 , a3 }[a4, a5 ]) + g(a1 {a2 , a4 }[a5, a3 ]) + g(a1 {a2 , a5 }[a3 , a4 ]) .
(4.11)
Finally we can use the result in (4.7) and perform explicitly the commutator algebra in the
gauge group structures


1 2 1
(tot)
5 =
g(a1 , a2 , a3 , a4 , a5 )
32 12 M 6
  
 
 
2 F 5 a a a a a F 5 a a a a a + 3 F 5 a a a a a
1 2 3 4 5
1 4 2 5 3
1 4 3 2 5
 3
 3

 2
 2
+ 1/2 F a a F a a a 1/2 F a a F a a a .
(4.12)
2 4

1 3 5

3 4

1 2 5

The above expression can be manipulated further so that the F 2 F 3 terms are eliminated in
favor of the F 5 terms, using the identities (C.4) depicted graphically in Fig. 5.


1 2 1
(tot)
5 =
g(a1 , a2 , a3 , a4 , a5 )
32 12 M 6

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486



 
8 
4 
F5 a a a a a F5 a a a a a + 4 F5 a a a a a .
1 2 3 4 5
1 4 2 5 3
1 4 3 2 5
5
5

75

(4.13)

In Section 6 we will come back to (4.13) and assemble all our results. Now we turn to the
computation of the second class of diagrams that contribute to the five-point function.
W
 five-point terms
4.2. The W W W
In this subsection we analyze contributions from the five-point function with a
 W
 . These terms are not gauge invariant,
background dependence of the form WWW
but as we have previously stated, they do contribute to the covariantization of derivative
terms obtained from the four-point function. In order to prove this we start considering a
typical graph of this type as shown in Fig. 2(b). After completion of the D-algebra it gives
rise to

 (p3 , a3 )W
 (p4 , a4 )
d4 (p5 , a5 )W (p1 , a1 )W (p2 , a2 )W

 ( 5 ) (f1 ) (f2 ) (f3 ) (f4 ) .

(4.14)

Including the loop-momentum integration we find



i(k p1 )
1
(cov1)
5
=
d4 k 2
2
8
k (k + p2 ) (k + p2 + p3 )2 (k + p2 + p3 + p4 )2 (k p1 )2


g(a1 , a2 , a3 , a4 , a5 )( 5 ) (f1 ) (f2 ) (f3 ) (f4 ) .
(4.15)
The complete set of diagrams is obtained considering all the permutations of the indices
(p1 , a1 ) (p4 , a4 ). The next step is to evaluate the momentum integral to leading order in
the low-energy expansion


1
(cov1)
 24 d4 k dxi 2
I5
(k + M 2 )5


p2 (x4 x1 ) + p3 (x3 x1 ) + p4 (x2 x1 ) x1 p1 .
(4.16)
The final result is
5(cov1)



i 2 1
g(a1 , a2 , a3 , a4 , a5 )[4p1 + 3p2 + 2p3 + p4 ]
=
40 12 M 6


( 5 ) (f1 ) (f2 ) (f3 ) (f4 )



+ permutations of (p1 , a1 ) (p4 , a4 ) .

(4.17)

Now we want to show that (4.17) gives the correct contribution linear in to covariantize
(3.13). (The part quadratic in is produced at the level of the six-point function and it will
be discussed in the next section.) Thus we go back to (3.13) and there we substitute every
ordinary derivative with a covariant derivative whose expression in momentum space can
be written as

()c (p) = ()c (p) + d4 q ( a )(q)b (p q)f abc .
(4.18)

76

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

In so doing we find that the terms which are linear in exactly amount to the ones in
(4.17). This can be checked in a simple manner as follows: first we write the derivative
contributions to the four-point function as



1 2 1 
(4 )der =
10(p2 )2 + 8p2 p3 + 2p2 p4
6
40 12 M



1
TrAd A(p1 )B(p2 )C(p3 )D(p4 ) + permutations of A, B, C, D ,
4
(4.19)

where A, B, C, D denote either f or f , with all permutations included while keeping


the momenta in the order p1 , p2 , p3 , p4 . Then we can use the ciclicity and inversion
property of the trace so that we always have the two derivatives acting both on the first
field or on the first and the second or on the first and the third. In this way the rules for the
covariantization are simply given by (here we consider the terms linear in only)


TrAd 2 ABCD  2 TrAd ([ , A]BCD)
 2TrAd ( ABCD) 2 TrAd ( BCDA),
TrAd (ABCD)  TrAd (A[ , B]CD) + TrAd ([ , A]BCD)
 TrAd ( BCDA) TrAd ( CDAB)
+ TrAd ( ABCD) TrAd ( BCDA),
TrAd (ABCD)  TrAd (AB[ , C]D) + TrAd ([ , A]BCD)
 TrAd ( CDAB) TrAd ( DABC)
+ TrAd ( ABCD) TrAd ( BCDA).

(4.20)

In order to streamline the notation we have written the above expressions in the gauge
= 0, a choice actually not necessary to prove the result. Since we have to include
a sum on all permutations we can freely rename the fields and write the final result in the
form


i 2 1
(4 )cov1 =
[4p1 + 3p2 + 2p3 + p4 ]
40 12 M 6



TrAd (p5 )A(p1 )B(p2 )C(p3 )D(p4 )

+ permutations of A, B, C, D
(4.21)
which exactly reproduces the formula (4.17).

5. contributions from the six-point function


Now we study contributions from the six-point function containing a background
dependence which will complete the covariantization of the two-derivative part of the fourpoint function. There are two distinct types of diagrams as shown in Fig. 3. We consider
them in the next two subsections.

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

(a)

77

(b)
Fig. 3. Six-point amplitude.

5.1. Diagrams with a vertex


The typical background dependence is given by

 (p3 , a3 )W
 (p4 , a4 )
d4 (p5 , a5 ) (p6 , a6 )W (p1 , a1 )W (p2 , a2 )W

 ( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 ) .

(5.1)

Including the momentum integration one finds


 
 2

1
1
1 4

24
6 =
2
4
2!


g(a1 , a2 , a3 , a4 , a5 , a6 )( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )

1
d4 k 2
2
k (k + p1 ) (k + p1 + p2 )2 (k + p1 + p2 + p3 )2
1

.
(5.2)
(k + p1 + p2 + p3 + p4 )2
To the above contribution we have to sum all the ones obtained permuting the a1 , . . . , a4
indices. Once again we compute the momentum integral at the leading order in the lowenergy expansion. We have


1 2 1
g(a1 , a2 , a3 , a4 , a5 , a6 )
6 
16 12 M 6


( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )



+ permutations of (p1 , a1 ) (p4 , a4 )


1 2 1

=
( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )
16 12 M 6

78

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486



g(a1 , a2 , a3 , a4 , a5 , a6 ) + permutations of a1 a4 .

(5.3)


Notice that the permutations are only on the indices of the Ws and the Ws.

W

5.2. Diagrams ( )( )W W W

Depending on the relative position of the two vertices one obtains three
different diagrams. The first one contains the two adjacent vertices and produces a
contribution of the form
   

1 6 1 3 5


=
2
d4 k( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )
6
2
2!
k (k + p6 )
2
k (k + p6 )2 (k + p6 + p1 )2 (k + p6 + p1 + p2 )2
1

g(a1 , a2 , a3 , a4 , a5 , a6 ).
(5.4)
(k p5 p4 )2 (k p5 )2
The leading order in the low-energy expansion gives
6

5!

16

1

x4
dx1

d k
0

d x5 ( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )

1
2
g(a1 , a2 , a3 , a4 , a5 , a6 )
(k + M 2 )6


k k + p6 (1 x5 ) + p1 (1 x4 )


+ p2 (1 x3 ) + p3 (1 x2 ) + p4 (1 x1 )

p6 (1 x5 ) + p1 (1 x4 ) + p2 (1 x3 )
 
+ p3 (1 x2 ) + p4 (1 x1 ) p6 .

(5.5)

From the above contribution we need extract the terms proportional to k k


1 2
2 k C C which are the ones relevant for the covariatization of the four-point function.
The other two diagrams lead to similar contributions. We end up with the following
result


1 2 1


6 tot =
( 5 ) ( 6 ) (f1 ) (f2 ) (f3 ) (f4 )
32 30 M 6

2g(a1 , a2 , a3 , a4 , a5 , a6 ) + 2g(a1 , a6 , a2 , a3 , a4 , a5 )


+ g(a1 , a2 , a6 , a3 , a4 , a5 ) + permutations of a1 , a2 , a3 , a4 .
(5.6)
Notice that the first two graphs have a multiplicity which is double as compared to the third
one. All together they reconstruct the 5! permutations of the a2 , . . . , a6 indices.
5.3. Covariantization of 4 with two s
The starting point is the result in (4.19). We proceed in complete analogy with what
we have done for the part linear in , so that we obtain simple rules also for the terms

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

79

quadratic in :


TrAd 2 ABCD  TrAd ([ , [ , A]]BCD)
 TrAd ( ABCD) 2TrAd ( A BCD) + TrAd ( BCDA),
TrAd (ABCD)  TrAd ([ , A][ , B]CD)
 TrAd ( A BCD) TrAd ( AB CD)
TrAd ( BCDA) + TrAd ( B CDA),
TrAd (ABCD)  TrAd ([ , A]B[ , C]D)
 TrAd ( AB CD) TrAd ( ABC D)
TrAd ( B CDA) + TrAd ( BC DA).

(5.7)

Once again, using the properties of the trace operation it is possible to set one of the s
in first position, and the other one either in second or in third or in fourth position. Then
renaming the fields and using the formulas (2.18) we obtain




1 2 1 
3TrAd (p5 ) (p6 )A(p1 )B(p2 )C(p3 )D(p4 )
(4 )cov2 =
80 12 M 6

+ 2TrAd (p5 )A(p1 ) (p6 )B(p2 )C(p3 )D(p4 )

+ (p5 )A(p1 )B(p2 ) (p6 )C(p3 )D(p4 )

+ permutations of A, B, C, D
(5.8)
which is just the sum of (5.3) and of (5.6). This completes the proof of the covariantization
of the two-derivative terms for the four-point function.

6. The final result


In this section we collect all the contributions at order M 6 that we have computed
so far, i.e., the two derivative terms from the first subleading contribution of the fourpoint function in (3.13), and the leading terms of the five-point function in (4.13). We start
reexamining the two derivative terms.
Covariantizing the expression in (3.13) and taking into account the cyclic and inversion
properties of the trace operation (cf. (2.9), (2.10)), we have



1 2 1
2
2
Tr
(4 )der =
Ad F F F F 2 F F F F
8 12 M 6
1
1
4
+ 2 F F F F + 2 F F F F F F F F
2
4
5
4
1
2
F F F F F F F F F F F F
5
5
5
4
1
1
F F F F + F F F F + F F F F
5
5
5
1
1
+ F F F F + F F F F
5
10

80

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486


1
F F F F ,
20

(6.1)

where it is understood that the indices on the two are to be contracted. We integrate by
parts in such a way to have the derivatives acting either on the same or on two adjacent
field strengths. We obtain




1 2 1
Tr
F F F F
(4 )der =
Ad
20 12 M 6

+ F F F F + F F F F


1
F F F F + F F F F + F F F F
4



1 2 1
Tr
2 2 F F F F 4 2 F F F F

Ad
20 12 M 6

1
+ 2 F F F F + 2 F F F F .
(6.2)
2
Now we show that quite generally we can rewrite terms which contain 2 and four F s
as F 5 contributions. Indeed, using the equations of motion F = 0 and the Bianchi
identities [ F] = 0, we have the following identities
2 F = F

Bianchi

( F + F )

e.o.m

= [ , ]F [ , ]F
= +i[F , F ] + i[F , F ] = 2i(F F F F ),

(6.3)

where we have used [ , ]F = i[F , F ].


Therefore we can write
 
   
2 F a = 2fabc F b F c .

(6.4)

The above equation leads to


     
2 F b F c F d F e g(b, c, d, e)
       
= 2 F a F b F c F d F e g([a, b], c, d, e).

(6.5)

Thus in order to study corrections to the symmetrized trace prescription one has to
consistently take into account derivative contributions. In particular the above relations
clearly show that for the case of two covariant derivatives one has to consider not only the
antisymmetrized products [ ] , as it was already noticed [7,8], but also the symmetrized
ones ( ) .
Using (6.5), we can rewrite the last two lines of the two derivative result in (6.2) as


 5
 5
1 2 1
 5
4
F
+
4
F
+
8
F aaaaa

a
a
a
a
a
a
a
a
a
a
1 2 3 4 5
1 2 4 3 5
1 4 5 2 3
20 12 M 6

 3
 3
 2
 2
+ 4 F a a F a a a + 2 F a a F a a a g(a1 , a2 , a3 , a4 , a5 ).
(6.6)
3 4

1 2 5

2 5

1 3 4

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

81

These terms can be combined with the F 5 contributions we have obtained from the fivepoint function in (4.13). Making use of the identities (C.4) the sum of (6.6) and of (4.13)
can be written as


1 2 1
 5
7 
3 
F a a a a a F5 a a a a a F5 a a a a a
6
1 2 3 4 5
1 2 4 3 5
1 4 2 5 3
20 12 M
2
2

 5
+ 2 F a a a a a g(a1 , a2 , a3 , a4 , a5 )
1 3 2 5 4



1 2 1
7
Tr
=
Ad F F F F F F F F F F
6
20 12 M
2

3
(6.7)
F F F F F + 2F F F F F .
2
Thus our final result to the M 6 order is:




1 2 1
tot =
Tr
F F F F
Ad
6
20 12 M

+ F F F F + F F F F


1
F F F F + F F F F + F F F F
4



7
1 2 1
Tr
+
Ad F F F F F F F F F F
20 12 M 6
2

3
F F F F F + 2F F F F F .
(6.8)
2
Up to an overall numerical factor, the first two lines in (6.8) reproduce the corresponding
result in Ref. [14], formula (3.3), obtained from an open superstring scattering amplitude
(see also [19]). So, our first result is that, at the level of the four-point function, the
supersymmetric YangMills effective action exactly reproduces the structure of the nonAbelian BornInfeld theory, including the first derivative corrections. The terms F 5 , which
were also computed in [14], are more difficult to compare with ours: this is essentially
due to the fact that the result quoted in [14] is not written in a canonical form and
moreover it requires additional symmetrizations which we are not clear how to interpret
unambiguously.
We would like to conclude summarizing our results. We have considered the N = 4
supersymmetric YangMills theory and computed at one-loop the four- and five-point
functions with external vector fields. From the superfield result we have extracted the part
of the bosonic components which contain the field strengths F . These non-local one-loop
contributions have been expanded in a low energy approximation and expressed as a sum
of an infinite series of local terms. We have argued that these local expressions reproduce
contributions to the non-Abelian BornInfeld action, if supersymmetry has to determine
its structure. We have explicitly computed the leading contributions from the four- and the
five-point functions and the subleading terms of the four-point function. These latter ones
contain two derivatives acting on the four F . From the four-point function calculation
one simply obtains ordinary derivatives, thus the result is not in a gauge invariant form. We

82

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

have checked that the correct covariantization of the result is obtained via contributions
from the five- and the six-point functions with one and two connections, respectively.
We have confirmed that beyond the F 4 order the non-Abelian BornInfeld action
contains terms which are not included in the symmetrized trace prescription. In order
to compute corrections one has to consider terms with derivatives, and in particular
we have shown that one cannot disregard terms with symmetrized derivatives. Both the
antisymmetrized products [ ] and the symmetrized ones ( ) are on equal footing.
The part of the six-point function needed in order to check the gauge invariance of
our result has been computationally rather difficult, but clearly not impossible. Taking
advantage of the superfield approach and of the background field method the determination
of the full six-point function seems at hand.

Acknowledgements
The authors have been partially supported by INFN, MURST, and the European
Commission RTN program HPRN-CT-2000-00113 in which they are associated to the
University of Torino.

Appendix A. Notation and conventions


Our notations and conventions are the ones in [15]. We use a metric
 = (, +, . . . , +)

(A.1)

and raise and lower spinor indices with





0 i
0

C =
,
C =
i 0
i


i
,
0

C = C .

C = C ,

(A.2)

The matrices are defined in terms of the Pauli matrices


( ) (1,  ),

( )
C C ( ) = (1,  ),

1
( ) ( ) ,
4
1

( ) ( ) .
4
Vector indices are transformed into spinor notation by

(A.3)

 
V = V

(A.4)

F = 2f C + 2f C ,

(A.5)

V = ( )
V .
2
For an antisymmetric tensor of rank two we have

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

83

where f and f are symmetric bispinors. The above relations can be inverted
1
f = ( ) F ,
2
1
f = ( ) F .
(A.6)
2
Therefore an antisymmetric tensor is always expressible in terms of two symmetric
bispinors.

F = ( ) f ( ) f .

(A.7)

In terms of the LeviCivita tensor   0123 = 1 we define the Hodge-dual of F


as
1  F .
(3F ) = F
2

(A.8)

Using the relation


  = [ ]

(A.9)

one obtains the general formula


b ) = 1   (Fa ) (Fb )
a ) (F
(F
4
 

1  2 

= F ab (Fa ) (Fb ) (Fa ) (Fb )


2


+ (Fa ) (Fb ) (Fa ) (Fb ) (Fa ) (Fb ) (A.10)
and also
a Fb F
c ) = (Fa Fb Fc ) ,
(F
 
a ) (F
b ) = 1 F 2 + (Fa ) (Fb ) .
(F
ab
2

(A.11)

Appendix B. Further identities involving the matrices


The basic relations we use are
= ( ) i ,

(B.1)

= ( ) + i .

(B.2)

These allow to obtain the following results for the trace of two matrices


1
i
Tr( ) = + + 
4
2
+ completely antisymmetrized in (, )(, )

(B.3)

84

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

and similarly


1
i
Tr( ) = + 
4
2
+ completely antisymmetrized in (, )(, ).

(B.4)

For the trace of three matrices we obtain




1
i
1
Tr(  ) =  + (   +  ) +   
8
16
32
+ completely antisymmetrize in (, )(, )(, )
(B.5)
and similarly

1
i
1

Tr(  ) =  (   +  ) +   
8
16
32
+ completely antisymmetrized in (, )(, )(, ).
(B.6)


Appendix C. Special identities for the F field strengths


In this appendix we study structures with five F field strengths. We can contract the
indices in different ways, moreover we can place the various F s in different positions
with respect to the gauge group trace. For example we can write
 
TrAd (F F F F F ) = F 5 a a a a a g(a1 , a2 , a3 , a4 , a5 ),
1 2 3 4 5
 
 3
TrAd (F F F F F ) = F a a a F 2 a a g(a1 , a2 , a3 , a4 , a5 ).
(C.1)
1 2 3

4 5

Clearly not all the possible structures are independent, since we can freely use the ciclicity
and inversion property of the trace. It is easy to realize that among the F 5 structures only
four are independent: we can list them as follows
 
A: TrAd (F F F F F ) = F 5 a a a a a g(a1 , a2 , a3 , a4 , a5 ),
1 2 3 4 5
 5
B: TrAd (F F F F F ) = F a a a a a g(a1 , a2 , a3 , a4 , a5 ),
1 2 3 5 4
 5
C: TrAd (F F F F F ) = F a a a a a g(a1 , a2 , a3 , a4 , a5 ),
1 3 2 5 4
 5
D: TrAd (F F F F F ) = F a a a a a g(a1 , a2 , a3 , a4 , a5 ).
(C.2)
1 4 2 5 3

In the same way one finds that there exist only two independent structures F 2 F 3
 
 
E: TrAd (F F F F F ) = F 3 a a a F 2 a a g(a1 , a2 , a3 , a4 , a5 ),
1 2 3
4 5
 2
 3
F : TrAd (F F F F F ) = F a a a F a a g(a1 , a2 , a3 , a4 , a5 ).
1 3 4

2 5

(C.3)

The terms in (C.2) and (C.3) can be represented graphically as shown in Fig. 4 and
explained in [8]. Every vertex corresponds to a F field strength and the trace operation

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

85

Fig. 4. F 5 structures.

Fig. 5. Special identities.

has to be taken following the vertices of the pentagon clockwise. The arrows denote how
the indices are contracted.
A quite remarkable result is that the F 2 F 3 structures can be reexpressed in terms of the
F 5 structures as follows:
3
1
F = A B + C + D,
5
5

3
1
E = A + B + C + D.
5
5

(C.4)

We can give a graphical representation of these relations as is shown in Fig. 5. So actually


there are only 4 independent structures containing 5 F .

86

A. Refolli et al. / Nuclear Physics B 613 (2001) 6486

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]

[12]

[13]
[14]
[15]
[16]
[17]
[18]
[19]

M. Born, L. Infeld, Proc. R. Soc. A 144 (1934) 425.


E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 163 (1985) 123.
A. Abouelsaood, C. Callan, C. Nappi, S. Yost, Nucl. Phys. B 280 (1987) 599.
A.A. Tseytlin, BornInfeld action, supersymmetry and string theory, in: M. Shifman (Ed.), Yuri
Golfand Memorial Volume, World Scientific, 2000, hep-th/9908105.
A.A. Tseytlin, Nucl. Phys. B 501 (1997) 41, hep-th/9701125.
A. Hashimoto, W.I. Taylor, Nucl. Phys. B 503 (1997) 193, hep-th/9703217.
E. Bergshoeff, M. de Roo, A. Sevrin, On the supersymmetric non Abelian BornInfeld action,
hep-th/0011264.
A. Sevrin, J. Troost, W. Troost, The non-Abelian BornInfeld action at order F 6 , hepth/0101192.
D. Gross, E. Witten, Nucl. Phys. B 277 (1986) 1.
A.A. Tseytlin, Nucl. Phys. B 276 (1986) 391;
A.A. Tseytlin, Nucl. Phys. B 291 (1986) 876.
S.V. Ketov, Phys. Lett. B 491 (2000) 207, hep-th/0005265;
S.V. Ketov, Class. Quant. Grav. 17 (2000) L91, hep-th/0005126;
S. Gonorazky, F.A. Schaposnik, G. Silva, Phys. Lett. B 449 (1999) 187, hep-th/9812094;
A. De Giovanni, A. Santambrogio, D. Zanon, Phys. Lett. B 472 (2000) 94;
A. De Giovanni, A. Santambrogio, D. Zanon, Phys. Lett. B 478 (2000) 457, hep-th/9907214;
A. Refolli, N. Terzi, D. Zanon, Phys. Lett. B 486 (2000) 337, hep-th/0006067;
M. Cederwall, B.E. Nilsson, D. Tsimpis, The structure of maximally supersymmetric Yang
Mills theory: constraining higher-order corrections, hep-th/0102009;
M. Cederwall, B.E. Nilsson, D. Tsimpis, D = 10 super-YangMills at O( 2 ), hep-th/0104236.
S.J. Gates, S. Vashakidze, Nucl. Phys. B 291 (1987) 172;
E. Bergshoeff, M. Rakowski, E. Sezgin, Phys. Lett. B 185 (1987) 371;
R.R. Metsaev, M.A. Rakhmanov, A.A. Tseytlin, Phys. Lett. B 193 (1987) 207;
I.L. Buchbinder, S.M. Kuzenko, A.A. Tseytlin, Phys. Rev. D 62 (2000) 045001, hepth/9911221;
L. De Fosse, P. Koerber, A. Sevrin, The uniqueness of the Abelian BornInfeld action, hepth/0103015.
M. Pernici, A. Santambrogio, D. Zanon, Phys. Lett. B 504 (2001) 131, hep-th/0011140.
Y. Kitazawa, Nucl. Phys. B 289 (1987) 599.
S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace, BenjaminCummings, Reading,
MA, 1983.
M.T. Grisaru, W. Siegel, Nucl. Phys. B 201 (1982) 292.
M.T. Grisaru, D. Zanon, Nucl. Phys B 252 (1985) 578.
M.T. Grisaru, W. Siegel, Phys. Lett. B 110 (1982) 49.
R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 298 (1988) 109.

Nuclear Physics B 613 (2001) 87104


www.elsevier.com/locate/npe

Closed strings from SO(8) YangMills instantons


Ruben Minasian a , Samson L. Shatashvili b,1 , Pierre Vanhove c
a Centre de Physique Thorique, cole polytechnique, 91128 Palaiseau, France 2
b Department of Physics, Yale University, New Haven, CT 06520-8120, USA
c Service de Physique Thorique, CEA-Saclay, F-91191 Gif-sur-Yvette Cedex, France

Received 2 July 2001; accepted 25 July 2001

Abstract
When eight-dimensional instantons, satisfying F F = 8 (F F ), shrink to zero size, we
find stringy objects in higher order ten-dimensional YangMills (viewed as a low-energy limit of
open string theory). The associated F 4 action is a combination of two independent parts having a
single-trace and a double-trace structure. As a result we get a D-string from the single-trace term
and a fundamental string from the double-trace. The latter has (8, 0) supersymmetry on the worldsheet and couplings to the background gauge fields of a heterotic string. A correlation between the
conformal factor of the instanton and the tachyon field is conjectured. 2001 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.-w; 11.15.-q; 11.10.-z

1. Introduction
Recent developments in string theory have given some hope for answering important
questions which require off-shell formulation of a second quantized string theory. In the
case of open string field theory, several off-shell formulations are currently known [13].
The situation with the closed string field theory is more complicated. One might consider
the possibility that in fact the complete and consistent off-shell string theory can originate
from properly defined open string field theory alone, with closed string being understood
in terms of classical open string field theory (for motivations and references see, [4,5]).
The possibility that a closed string world-sheet action arises from classical solutions in
open string field theory will be the main theme of this paper. Although a complete picture
might require incorporating the whole field theory of open strings, one might hope to get
E-mail addresses: ruben.minasian@cpht.polytechnique.fr (R. Minasian), samson.shatashvili@yale.edu
(S.L. Shatashvili), vanhove@spht.saclay.cea.fr (P. Vanhove).
1 On leave of absence from St. Petersburg Steklov Mathematical Institute, St. Petersburg, Russia.
2 Unit mixte du CNRS et de lEP, UMR 7644.
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 1 ) 0 0 3 6 9 - 8

88

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

important insights by looking at low energy sector only. This is what we will attempt here.
We start from a non-abelian super-YangMills theory including the higher-derivative F 4
terms in ten dimensions in the background of a SO(8) instanton and find stringy objects. 3
The presence of higher-derivative terms in the effective action is crucial for our analysis
since not only the lowest order F 2 term cannot lead to stringy objects with a finite tension
but, as we will argue, does not play any role. In zero size limit of the instanton, a Dstring and a fundamental heterotic string are found. The heterotic string solution will be
supersymmetric only under a linear combination of the standard supersymmetry of the
YangMills theory together with an extra supersymmetry transformation. It is tempting
to interpret this extra supersymmetry transformation as originating from the non-linear
supersymmetry of a non-BPS D-brane sitting in our vacuum.
The consistency of the heterotic string background is related to the original symmetries
of the YangMills background; the chirality of the background will be reflected in the
chirality of the string world-sheet theory. In particular, starting from an anomaly free
open string theory (a correct choice of the gauge group for which the top term in the
anomaly polynomial vanishes is what is meant here) will surely result in a closed heterotic
string theory in a consistent gravitational and YangMills background.
The D-string appears from the presence of tr F 4 in the effective action, and its tension
is quantized according the value of 7 (G) = Z for G allowing non-trivial embedding
of SO(8) groups [6,7]. For obvious reasons, we chose G = SO(32). The novelty of our
study resides in the origin of a fundamental heterotic string from the double-trace structure
(tr F 2 )2 and its supersymmetric completion in the effective action.
In the next section, the precise structure of these terms will be shown from the
supersymmetry analysis but we would like to make some qualitative comments here from
a different point of view. Some relevant ideas come just from looking at the (bosonic part
of) open string action


S = Sopen (, B, T ) d 10 x det(G + F ) f (T )
(1.1)
written here for the constant tachyon field T (the precise analog of this action for nonabelian gauge groups is not known yet but can be conceptually derived in any given string
field theory). According to the well-known conjectures of Sen [8], f (T ) has zero at the
minimum of the tachyon potential where the closed string vacuum is expected. It can
be calculated using the formalism of [2]; for bosonic string field theory it is given by
f (T ) = eT while the tachyon potential is VB = (1 + T )eT and for the superstring,
2
f (T ) coincides with the tachyon potential VF (T ) = eT [9]. Since our analysis relies on
supersymmetry, from now on we will replace f by V .
In a general string field theory the Lagrangian is defined up to field redefinitions and a
correct choice of fields is dictated by the principle that equations of motion are proportional
to world-sheet -function with some Zamolodchikov metric G: i S = Gij j . Here G is
3 The structure and the supersymmetry of F 4 terms will be discussed in details in Section 2. For now we just

remark that due to different ways of contracting the gauge indices they can be of two types single-trace terms
tr F 4 and double-trace (tr F 2 )2 terms.

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

89

the metric on the space of fields and has to be non-degenerate. One can note that the perfect
square part of double-trace terms, (tr F F )2 , in a classical open string field theory
Lagrangian can be generated by a field redefinition T T + tr F 2 . The compatibility
of the supersymmetry and the above principle should fix these terms unambiguously as
well as complete the needed t(8)(tr F 2 )2 .
We conclude this introduction by a comment on the world-sheet topology. We already
mentioned that by expanding around the solution we obtain couplings for collective modes
which turn out to reproduce those of the heterotic string, and this should already indicate
that the string under the consideration is closed. Much of this structure is directly inherited
form the SO(8) instanton construction. It is yet another feature of the construction that the
solution has in fact SO(9) symmetry and the fact that we are using only the SO(8) part
leads to the periodicity of one of the string world-sheet coordinates.
A brief outline of the paper is the following. In Section 2, we present the Yang
Mills action and the supersymmetry transformations. In Section 3, we look for BPS
configurations associated with eight-dimensional instanton configurations having SO(8)
symmetry. We consider the supersymmetric excitations in the Section 4 and show the
emergence of the two-dimensional action with couplings of the heterotic string with
a particular embedding of the gauge group. Finally the Section 5 contains a general
discussion of our results. We have collected various definitions and identities needed in
the main text in two appendices.

2. The effective action


Since we need to consider non-abelian and supersymmetric extensions of the effective
theory (1.1), we start by recalling some facts about supersymmetrisation of open string
effective actions.
2.1. The linear supersymmetry
The effective action can be expanded in power series of  , and at each order
supersymmetric action can be obtained by Noether procedure. This construction is valid for
any gauge group. This is done in the context of linearly realized N10 = 1 supersymmetry
transformations. The Lagrangian for the non-abelian degrees of freedom constructed by
this procedure up to and including (  )2 F 4 terms is [10]:
1
L = F A F A 8 AD
/ A
4




(  )2
MABCD F A F B F C F D 4F A F B F C F D

32



+ (  )2 MABCD 4F A F B C (D )D


+ 2F A F B C (D )D
+ quartic fermions.

(2.1)

90

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

This action (2.1) is invariant under the linear supersymmetry transformations (by a
straightforward generalization of the results of [11])

 
 C D
 C D
(  )2 A
A
& A A
M BCD 2 & B F
=

4
&

F 8 & B F
F

8



(2.2)
& 1 4 B FC1 2 FD3 4 ,
 ()


1
(  )2 A
A
+ 5
7 8 & 1 6 & FB1 2 FC3 4 FD5 6 ,
M BCD t(8)
& A = &F
8
2 4!
(2.3)
are the SO(1, 9) -matrices, as well as all the spinors in this expression; the tendimensional spacetime indices are labeled by Greek letters from the middle of the alphabet
, = 0, 1, . . . , 9.
All the fields are in the adjoint of the gauge group G = SO(32), labeled by a capital
letter from the beginning of Latin alphabet. The quadratic expressions in the field are
contracted with the metric AB = trfund (TA TB ) and quartic expressions with the completely
symmetric four rank tensor MABCD [10,12]. For any group, there are two independent fourrank symmetric tensors given by the symmetrized trace in the fundamental representation
trfund(T(A TB TC TD) ) = Str( ) and the double-trace (AB CD) trfund ( ) trfund ( ) in
the fundamental representation
MABCD = m1 trfund (T(A TB TC TD) ) + m2 (AB CD) .

(2.4)

The analysis of [10,12] was purely in the framework of N10 = 1 super-YangMills


theory without any reference to a particular string vacuum. Linear supersymmetry fixes
the independent tensorials structures, or superinvariants, at each order in the derivative
expansion, up to a number of unfixed coefficients (not affected by field redefinitions). We
emphasize once more that in principle such structure can also be fixed from the requirement
that the equations of motion are proportional to the world-sheet -function.
When specifying the string vacuum of reference the coefficients multiplying the group
tensors become field-dependent functions. In particular these coefficients will have a
dependence
on the dilaton. The open string effective action is normalized as S =

(  )2 d 10 x L/(gs go2 ) with the standard normalisation value for the coupling constant
2
= 216 14 (  )6 [13]. The first term in (2.4) receives contributions
go4 = 210 7 (  )2 (10)
starting from the tree-level (disc) open string diagram: m1 ()/gs = 1/gs + 1/2 +
o(gs ); the second starts from one-loop open string diagrams: m2 ()/gs = 1/16 + o(gs ).
A dependence on the open string tachyon field can appear too. As we will see this
difference will lead eventually to having D-string and a fundamental string tensions for
solitons corresponding to the single- and the double-trace parts, respectively. Looking at
this theory as the low-energy limit of the SO(32) heterotic string, the coefficients m1 and
m2 will have different dilaton dependence. These two different functional dependences
on the dilaton have been shown in [14] to be compatible with the strong coupling
duality between type I and heterotic string. Bearing this in mind, we keep the field
dependence of the coefficients unspecified, relying only on their independence under
supersymmetrisation.

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

91

The presence of the two independent terms in (2.4) will be fundamental for our analysis.
The single-trace term will carry the charge of the instanton introduced in Section 3 and
the double-trace term will be at the origin of the interactions of the fluctuations around the
instanton.
2.2. The non-linear supersymmetry
Looking at the effective action as related to D-brane configurations in a type II vacuum,
it is natural to look for the possibility of non-linear supersymmetry resulting from the
breaking of some of the original supersymmetries of the vacuum. There are various ways
to introduce non-linear supersymmetry depending on which fields are coupled to the
supersymmetry parameter [10,15] in = + O().
For the case of open string actions with tachyon field and abelian gauge field, the
quadratic effective action takes the form (abelian field have tilde)


1
V (T )
/ + 1 T T +
1 F
L=
(2.5)
F + D
gs
4
2
and is invariant under the non-linear supersymmetry transformations


1
= 1 + T ,
2

T = 2 2 T
2 T T ,
A =
.

(2.6)

For later reference we notice that the supersymmetry transformation on abelian gaugino
take the form = g(T ) where g(T ) is function of the tachyon field only, taken its
values in the spinorial representation of SO(1, 9).
In a way, the non-linear supersymmetry involving the tachyon field is forced on us
by the desire of working with pure YangMills without the presence of the supergravity
multiplet. The open string vacuum, where this second symmetry appears, will be important
for us since as we will see in Section 4 a linear combination of the linear and non-linear
supersymmetries guarantees the supersymmetry of the fluctuations around the instanton.

3. Introducing the SO(8) instanton


The main point of this paper is to search for closed strings inside the field theory of
open strings. From the point of view of low-energy effective Lagrangian this means that
we shall look for classical solutions in the sector of theory containing the gauge fields
and tachyon. Because we are looking for a string-like solitonic objects it is natural to
investigate eight-dimensional gauge configurations. It is then not hard to see that due to
the gauge configuration having codimension two, an expansion around such a soliton will
result in a string. Understanding the nature of this string will be our task.
On a more technical level, the presence in the supersymmetry transformations of the
gauginos of terms linear and cubic in the the gauge field strengths is also indicative of

92

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

a special role of eight-dimensional gauge configurations. Luckily one such configuration,


namely the eight-dimensional SO(8) instanton, is well known [16,17], and has been used
for constructing string solitons previously in [6,18]. We start from examining the (  )2
corrections to & and observe that indeed it significantly simplifies for the gauge field
given by [16,17]
2
(3.1)
,
+ 2 )2
where is the size of the instanton, m, n = 1, . . . , 8, and a, b are SO(8) indices of positive
chirality. 4 Hereafter the form factor will be abbreviated as f (, x) = 2 /(x 2 + 2 )2 . We
introduce the SO(8) chirality projector
ab
Fmn
(x) = 4(mn P+ )ab

(x 2

1
P = (1 (8)).
2
The SO(1, 9) fermionic indices in (2.2) and (2.3) are split according to SO(8) indices as
& = (&a , &a ) (note all the ten-dimensional spinors are MajoranaWeyl). Most importantly
for our analysis the antisymmetric product of two SO(1, 9) -matrices decompose under
SO(8) -matrices as
mn = mn P+ + mn P .
As shown in Appendix A, the eight-dimensional solution (3.1) satisfies a very important
identity


1
r1 r8
r1 r8
MABCD t(8)
(3.2)
+ (8)
FrA1 r2 FrB3 r4 FrC5 r6 = 0,
2
with the relative positive sign for our choice of the SO(8) representation for the instanton.
(For the single-trace part this has been noticed in [6].)
For the solution (3.1), using (3.2) and the previous conventions, the (  )2 correction to
the gaugino supersymmetry transformation (2.3) takes the form


(  )2 A
(3.3)
M BCD m1 m6 P & FmB1 m2 FmC3 m4 FmD5 m6 ,
4
2 4!
the appearance of the projector P is related to the choice of the SO(8) representation used
to define the instanton.
Since we have already restricted ourselves to eight-dimensional configurations, it is easy
to see that there may be further relations between &(0) and &(2) by virtue of Hodge
duality. But in order to see these we need some facts about eight-dimensional self-duality
equations and their solutions [16,17], discussed in more details in Appendix A. The gauge
field (3.1) is a solution of the following self-duality equation in flat R8 [16,17]



MABCD F C F D 8 F C F D = 0.
(3.4)
&(2) A =

This equation is conformal, and the self-duality is guaranteed by the properties of the SO(8)
-matrices.
4 We refer to Appendix A for details about SO(8) instantons, in particular the chirality discussion and their
generalisations to multi-instantons solutions.

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

93

One can check than the solution (3.1) also satisfies a relation
2
1 
m1 4f (, x) 8 F A = M A BCD F B F C F D .
(3.5)
2
Note that we are not using the curved (non-conformal) counterpart of (3.4), but stay in the
flat space and simply use another property of the solution. Eq. (3.5) allows to establish a
(0)
connection between the quadratic and quartic term in the gauge field in (2.1) and & and
(2)
& . Notice that in the lhs of (3.5) the dependence on m2 has dropped out by virtue of the
first Pontryagin class p1 (F ) vanishing on the solution. We finally rewrite the full resulting
supersymmetry as



2
1 rs A
4
A
 2
& = Frs 1 + ( ) m1 f (, x) P &.
(3.6)
8
3
Using (B.7) to work out the contraction between the spacetime -matrices and the SO(8)
-matrices used to construct the instanton we see that the supersymmetry transformation
of the SO(8) part of the gaugino ab takes the form



2
4
 2
&
(
& ab = 7f (, x)(P+ )ab
(3.7)
f
(,
x)
)
m
P
,
1
+
1

that may be further reduced to 5


& ab = 7f (, x)(P+ )ab
& ,

(3.8)

by remarking that the operator (P ) projects on the chirality space opposite to the one
used to construct the instanton
(P+ )ab
P = 0.
Therefore the (  )2 contribution to the supersymmetry transformation & vanishes
(2)
identically. Since there is no background for the fermions & A = 0 for the instanton (3.1)
and the (  )2 piece of the Lagrangian (2.1), in the SO(8) bosonic background (3.1), is
invariant under the lowest-order linear supersymmetry transformations

&(0) L(2) for (3.1) = 0.
(3.9)
(2)

From now on we will concentrate on this part of the YangMills action.


The supersymmetry transformation (3.8) is reminiscent of the non-linear supersymmetry
transformations (2.6) = up to the dependence on the profile of the instanton. That
the instanton (3.1) realizes an identification between the spacetime SO(8) Lorentz group
and the SO(8) sub-group of the gauge group, will be central, Section 4, for getting a
supersymmetric sigma-model from the fluctuations around this instanton.
The effects of the instanton are all controlled by the behavior of the form factor
f (, x) = 2 /( 2 + x 2 )2 . Running a little bit ahead of the time one can already note that
for or for 0 and not sitting on the instanton x = 0 the form factor goes to
5 The notation (P )ab makes explicit that this operator projects on the positive SO(8) chirality space labeled
+
by the indices a and b.

94

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

zero, and the limit corresponds to the usual open string vacuum

, for all x,
open string vacuum: lim f (, x) = 0 for
0,
and x = 0.

(3.10)

In this regime the linear supersymmetry (2.3) is not broken and the non-linear supersymmetry (2.6) is present. There is no relation between these two supersymmetries.
When sitting on the instanton x = 0, the zero-size limit behavior is different as the form
factor blows up to infinity: this is the closed string regime
closed string vacuum: lim f (, 0) 2 = .
0

(3.11)

In this regime the instanton dissolves and the non-linear supersymmetry get promoted to
linear supersymmetry and a linear combination of the two supersymmetries (3.8) and (2.6)
is expected to survive.

4. The string soliton


Since for the instanton (3.1), the quadratic L(0) and quartic L(2) pieces of the effective
action (2.1) are decoupled (see (3.9)), and we can concentrate on the quartic piece of the
action showing how the sigma-model for an heterotic string comes out

(  )2
(2)
d 10 x L(2) .
S = 9 7  3
(4.1)
2 ( ) gs
We shall now move to discussion of the string world-sheet and apply the standard technique
of expansion around the soliton. We take x m = x m ( ), where ( = 0, 1) are two
bosonic zero modes to be identified momentarily with the world-sheet coordinates, and at
lowest order in zm = x m x0m ,


m = Am = 4imn zn f (, x0 ) + x n ( z x)f  (, x0 ) + .
F
(4.2)
As the size of the instanton goes to zero, lim0 (f (, x0 ))4 = (8) (x0 ) and this leads to
emergence of two-dimensional string world-sheet action from the corresponding Yang
Mills action. Note that while the instanton solution enjoys a full SO(9) [17], only the
SO(8) subgroup is left explicit by the identification of R8 as the complement of the point
r9 = . Any other point would have been as good. This remaining rotational invariance
reflects itself in the periodicity of the world-sheet coordinate 1 . The construction puts no
restriction on the world-sheet coordinate 0 . As far as the world-sheet action is concerned,
at the leading order only the first term matters and realizes the pull-back from the spacetime
to the worldvolume theory. Using the decomposition of the antisymmetric product of
ab + c ab , where c
two SO(8) gamma-matrices as (mn )ab = mn
mn
mnab is the completely
antisymmetric SO(8) octonionic tensor, and covariantizing the result one gets from the F 4
terms in (4.1) the expected





1
m1
17 m1
1 m2
2
m
S (2) =
(4.3)

2
+
+
z

z
+

.
d

m
2 
gs
1920 gs
60 gs

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

95

We would like to take this a bit further and in particular study the coupling to background
gauge fields.
Before we go on we comment on the coefficients in (4.3). The first constant term is the
energy of the instanton, and thanks to the identities of Section A.2, it receives contributions
from the single-trace term only. It is therefore proportional to m1 . The second term
from the fluctuations around the instanton in the SO(8) part of the gauge group receives
contributions from both the single- and the double-trace part, and depends on m1 and m2 .
We observe here that this leads to emergence of two different string tensions here the
single-trace part m1 /(2  gs ) 1/(2  gs ) has a tension of a D-string while the doubletrace part m2 /(2  gs ) 1/(2  ) has a fundamental string tension.
A more interesting test comes from considering the next order perturbation on the space
(1)
(0)
of connections Am = A(0)
m + Am (m = 1, . . . , 8) around the instantonic solution (3.1) Am .
(1)
The fluctuation Am = am lives in the SO(24) part of the gauge group. Since we choose a
purely bosonic instanton background, the fermions are all of the first order type = (1) .
m and
Fermions, respectively in SO(8) and SO(24), are associated with the fluctuations F
(1)
A . At this point another crucial difference between the single- and double-trace parts
emerges. The former does not allow any mixing of the SO(8) and SO(24) parts and an
action (4.3) with half of the original supersymmetries preserved is all one gets. The doubletrace term however contains the supersymmetric extension

 F
 )
8L(2) 4m2 trSO(8) (
D ) trSO(8)(F

 F
 ) ,
D ) trSO(8)(F
+ 2m2 trSO(8) (
(4.4)
whose supersymmetry variation under (3.8) is



3
& 8L(2) m2 f (, x) (& P+ D).

(4.5)

Contrary to the action evaluated for the classical background, the action for the fluctuations
is not invariant by itself under the supersymmetry generated by &. We have observed
already the similarity between the linear and non-linear supersymmetries, and it looks
natural to try to use the latter in order to compensate for & (8L(2) ). As noticed below (2.6)
the geometry of the non-linear supersymmetry transformation is controlled by the profile
of the tachyon field. Moreover, beyond the group theory identifications of the gauge SO(8)
and the Lorentz SO(8), a tachyon field dependence can occur in the map between the
two gaugini and . We confine this freedom to a function W (T ) whose relation to the
potential is yet unclear. Acting with the non-linear supersymmetry (2.6) on (4.4) we get


D),
8L(2) m2 W (T ) (
(4.6)
The fluctuations around the instanton (3.1) can be made supersymmetric under a linear
combination of the two supersymmetry transformations & and provided the following
identification between the supersymmetry parameters
=

(f (, x))3
P+ &.
W (T )

(4.7)

96

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

It follows that the action (4.4) is supersymmetric granted


P = 0

(4.8)

is satisfied and the positive-chirality component of + survives. Notice that this chirality
is the one of the SO(8) representation used to construct the instanton. The remaining
supersymmetry, generated by + , will give rise to light-cone GreenSchwarz fermions
for an heterotic string.
As we will see in the next subsection, these fermionic zero modes will give rise to
both the fermions of a chiral (8, 0) supersymmetric sigma-model and the gauge degrees
of freedom for the wordsheet theory of an heterotic string obtained from the fluctuations
around the instantons (3.1).
4.1. Gauge coupling
(1)

The coupling between Am = am to the world-sheet fermions , both in SO(24), is


essential for understanding the nature of the string. The effective action (2.1) has only two
terms with ten-dimensional fermion bilinears, of the form [1,3] DF 2 . Once again, the
gauge indices are contracted with the tensor MABCD (2.4) containing a single-trace and
a double-trace structure. Since the single-trace part gives a supersymmetric D-string, for
which the SO(24) is totally decoupled, we concentrate here on the double-trace part only.
We proceed with decomposing the fermions into a SO(8) part, , and SO(1, 1) fermions

(x) S( ),
for A Adj(SO(8)),
A =
(4.9)
(x) A ( ), for A Adj(SO(24)).
S a is a eight-dimensional MajoranaWeyl fermion defined from the SO(8) component of
by ab = S (P+ )ab
. Its SO(8) chirality is the same as the one used to construct the
instanton (3.1). The kinetic terms for the fermions S will come from the SO(8) SO(8)
part of the action. Since the fermions are the fluctuations in the complement of the SO(8),
used to construct the instanton, in SO(32) the kinetic and gauge couplings for the these
come from the SO(8) SO(24) cross terms.
From the fermion bilinear terms in (2.1)

) trSO(8) (Fn Fn )
m2 d 10 x trSO(8) (
we readily derive the kinetic terms for the light-cone fermions

Sh ,
m2 d 2 S

(4.10)

with induced two-dimensional metric




2
n
n
h := z z lim d 8 x f (, x) (x).
0

Notice that (3.8) implies that (x) scales as f (, x) and the previous integral has a finite
non-zero limit when goes to zero. The other fermion bilinear in (2.1) involving
does not contribute to the kinetic term for the light-cone fermions S.

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

97

From the interactions between the SO(8) deformation (4.2) and the SO(24) fermions

m2 d 10 x trSO(24) (
) trSO(8) (Fn Fn )
we derive the kinetic term for the twenty-four fermions

)h .
m2 d 2 trSO(24)(

(4.11)

From the coupling



[am , ]) trSO(8) (Fn Fnm )
m2 d 10 x trSO(24) (
we derive the coupling between the fermions and the background SO(24) gauge field

[am , ])e m .
m2 d 2 trSO(24)(
(4.12)
where the contractions of the indices are performed with the vielbein


2
e m := zm lim d 8 x f (, x) (x).
0

Finally the term in (2.1) does not contribute.


Putting everything together, we arrive at the world-sheet theory for a heterotic string
in the light-cone gauge with the particular embedding of the spin connection into the
SO(8) part of the SO(32) gauge group. 6 In particular, the kinetic term proportional to
m2 in (4.3) and (4.10) gives the (8, 0) supersymmetric sigma-model invariant under the
supersymmetries generated by . The kinetic term for the SO(24) fermions is obtained
in (4.11) which together with the coupling (4.12) to the external field, gives a covariant
derivative acting on the fermions.
It is well-known that the heterotic string can appear as a soliton in type I theory, and it is
natural to inquire about the difference of our mechanism from the S-duality correspondence
between type I and heterotic string of [20]. The latter requires a background metric (with
the Lorentz invariance broken to SO(1, 1)SO(8)) and a non-vanishing vev for the dilaton,
while we only have a YangMills background and a flat metric.

5. Discussion
We would like to underline some of the important points and list some open questions.
Starting from pure ten-dimensional YangMills action we are able to produce a Dstring and more surprisingly/controversially a heterotic string in the background of
eight-dimensional SO(8) instanton. The former has a non-vanishing vacuum energy and
supersymmetric (8, 0) world-sheet. Its very plausible connections to non-commutative
6 Curiously enough, one of very few features of ordinary four-dimensional instantons that is replicated by the

SO(8) instanton is the possibility of extending the solution to a curved space by setting the gauge connection
equal to spin connection [19].

98

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

solitons and to D1D9 (D0D8) systems are not very clear to us at the moment and are
beyond the scope of this paper. There may be some interesting questions to be asked in
this context. As for the heterotic string to which we will turn now, the SO(9) symmetry
of the instanton solution gets reflected in the periodicity of one of the string world-sheet
coordinates.
Our main interest is in the appearance of the fundamental heterotic string. Technically
speaking, its existence is due to the presence of the double-trace structures in the Yang
Mills F 4 terms responsible for mixing of SO(8) and SO(24). These terms were first seen
from the supersymmetry analysis on the ten-dimensional YangMills action, and as we
tried to argue here may also be related to the field redefinition freedom (fixed by the
requirement that the equations of motion are proportional to the world-sheet -function
and by supersymmetry) in the classical open string field theory Lagrangian. As a result
by expanding around the soliton we obtain a string action in the background where the
SO(8) gauge connection is set to the spin connection. It might be interesting to notice that
this double-trace structure also allows to generate kinetic terms for YangMills fields in
the bulk by giving classical values to tr F 2 . Indeed, in the SO(8) instanton background,
t(8) (tr F 2 )2 tr F F , and in a way the SO(8) part is used as the center of the gauge
group.
Finally, a crucial point in the analysis was the mutual cancellation of linear and nonlinear supersymmetry. The latter is known for the abelian case but applicable to the
SO(8) group by the magic of the SO(8) instantons, which realizes an identification of
the Lorentz and gauge SO(8) groups. The resulting connection between linear and nonlinear supersymmetry parameters contains a dependence of the tachyon and the size of the
instanton via (f (, x))3 /W (T ). Of course it is natural to expect that this ratio is a constant
and thus one gets a correlation between the tachyon and the size of the instanton (e.g.,
since f (, x) is singular when the instanton shrinks to the zero size, W (T ) should behave
accordingly). This correlation is not very explicit though, since it is not yet completely
clear how W (T ) is related to the tachyon potential. One could hope to understand better
what are the respective regimes where the D-string and the heterotic string discussed above
dominate. Such a possibility of at least partially recovering the tachyon potential from the
low-energy analysis in our opinion deserves further study.

Acknowledgements

It is a pleasure to thank Carlo Angelantonj, Costas Bachas, Emilian Dudas, Elias Kiritis
and Hong Liu for discussions, as well as Tigran Tchrakian for guidance into the SO(8)
instanton literature. The work of R.M. and P.V. is supported in part by EEC contract HPRNCT-2000-00122; R.M. is also supported by the INTAS contract 55-1-590. The work of S.S.
is supported in part by the OJI award from DOE.

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

99

Appendix A. Instantons in eight dimensions


In this appendix we review some facts about SO(8) instantons, and introduce the
notations used in the main text. We also briefly discuss different variants of the self-duality
equations and their solutions.
A.1. Definitions
 The SO(8) instantons satisfying the self-duality condition in Euclidean flat eightdimensional R8
1
(8)(F F )
(A.1)
4!
have been constructed in [16,17]. For a particular gauge choice, this solution is defined on
the eight-sphere of radius , rm r m + r9 r 9 = 2 in terms of projective coordinates (xm ) on
R8 , identified with the complement of the point r9 =
F F = 8 (F F ) =

rm = 2

2xm
(m = 1, . . . , 8),
2 + x2

r9 =

2 x2
.
2 + x2

Thanks to the SO(9) rotational invariance of the solution, we could have considered the
complement of r9 = +, without any change in the discussion of the main text. The
solution with no singularity at the origin is given by
(A )m = 4imn P

x n 2
( 2 + x 2 )2

(A.2)

and its curvature is


(F )mn = 4mn P

2
,
( 2 + x 2 )2

(A.3)

where mn P+ = mn belong to the 8s representation of SO(8) and correspond to the + sign


in (A.1). The other representation of opposite SO(8) chirality mn P will be abbreviated
as mn . The two representations correspond to the choice of sign in (A.1). We refer to
Appendix B for details on -matrix algebra. These mn matrices are the SO(8) equivalent
of t Hooft eta-symbols and this solution generalizes the JackiwRebbi instanton.
In the main text we make the choice of the positive SO(8) chirality solution F+ , and drop
the + subscript.
 There is another solution [17] related to the one above by a conformal transformation,
with a singular behavior at the origin
(A )m = 4imn P

x n 2
x 2 ( 2 + x 2 )2

(A.4)

and its associated curvature is


(F )mn = 4mn P

6
.
x 4 ( 2 + x 2 )2

(A.5)

100

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

 The solution (A.3) solves


8 F = F F F ,

(A.6)

with a conformal metric


gmn = 4f (, x)mn .

(A.7)

(A.5) has a form factor singular at the origin and therefore does not qualify for such nonconformal generalisation.
 The solution (A.3) satisfies a relation

2
8 f (, x) 8 F = F F F ,
(A.8)
with the flat metric gmn = mn . Some of the discussion in the main text relies heavily on
this relation.
 One can show that in 4d dimensions the following double-duality relation holds for
any Einstein metric
m m

(2d)!2 d R 1 2d 1 2d
n n

= & m1 m2d n1 n2d d R 1 2d 1 2d & 1 2d 1 2d ,
where d R is the completely antisymmetrized product of 4d-dimensional Riemann
curvatures. Similarly to four-dimensional case [21], for d = 2 this double-duality implies
a single duality relation (A.1) for a gauge field A [19], provided
1 mn
1 mn

F
= R
P mn .
A
(A.9)
= P mn ,
2
2
Thus the procedure of setting the gauge connection to spin connection enables one to
extend the construction of the instanton to curved backgrounds.
 This setup is generalized to multiple instanton solutions classified by maps of the
eight-sphere S 8 onto itself [22]:
n wp m wq
pq P ,
(1 + wl wl )2

N
7

wl l = x8 I +
x m m .
(F )mn = 2

(A.10)

m=1

Such instanton carries a topological charge classified by 7 (SO(8)) = Z Z.


 Finally we remark that the SO(8) instantons can be embedded in any SO(n) group
with n  8 (7 (SO(n)) = Z for n > 8) but not in E8 E8 group, since there is no nontrivial embedding of SO(8) in E8 . This fact is easily understood by noticing that E8 has
no independent quartic invariants tr(F F F F ) tr(F F ) tr(F F ), and that
tr(F F ) = 0 for the both the solutions (A.3) and (A.5).
 Since all the F 4 action is expressed in term of the completely symmetric tensor [10,
12] we define the topological charge carried by the instanton as

1
q :=
d 8 x MABCD FA FB FC FD ,
4!

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

101

which can be rewritten using first and second Pontryagin classes





1
q =
d 8 x 4m1p2 (F ) + 2(m1 + m2 )p1 (F ) p1 (F ) .
4!
Since for the instantons (A.3), (A.5) and (A.10) p1 (F ) = 0, the charge carried by the
instanton is

m1
q =
(A.11)
d 8 x p2 (F ).
4!
(A.3) and (A.5) have charge q = 28 4 , and (A.10) has charge q = 28 4 N .
A.2. Identities
We show that the SO(8) instanton (A.3) satisfies the identity


1
B
C
D
MABCD t(8) r1 r8 (8) r1 r8 (F )A
r1 r2 (F )r3 r4 (F )r5 r6 Nr7 r8 = 0
2

(A.12)

for any matrix N in the adjoint of the group, generalising the identity used in [6]. The
completely symmetric tensor MABCD of Eq. (2.4) (defined in [10]) has two parts the
symmetrized single-trace and the double-trace. It is therefore sufficient to check that


tr t(8)r1 r8 (F )r1 r2 (F )r3 r4 (F )r5 r6 Nr7 r8


2 
= 25 (7d 11)(d 4) f (, x) tr (F )r7 r8 Nr7 r8 ,


tr (8)r1 r8 (F )r1 r2 (F )r3 r4 (F )r5 r6 Nr7 r8

2 

= 24 6! f (, x) tr (F )r7 r8 Mr7 r8
therefore for d = 8 (A.12) is satisfied.
t(8) r1 r8 tr[(F )r1 r2 (F )r3 r4 ] tr[(F )r5 r6 Nr7 r8 ]

2 

= 28 (d 1)(d 8) f (, x) tr (F )r7 r8 Mr7 r8 ,


B
tr (8)r1 r8 (F )A
r1 r2 (F )r3 r4 (F )r5 r6 Nr7 r8 = 0
therefore for d = 8 (A.12) is satisfied.
A.3. An action for the instanton
This solution has been shown [6,18] to be of importance for constructing the string as
a soliton for the five-brane in ten dimension. It was there shown that the solution satisfies
the equation-of-motion [23]
E (0) (A) =



1
D f (, x)F = 0,
f (, x)

and
E (2) (A) =

(  )2 e
M A BCD
go2 3! 8gs

(A.13)

102

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104



  C D 1  B C D 
D F B F
F D F F F
4

(A.14)

derived from the effective action (2.1).


Thanks to the equality (A.12), E (2) (A) = 0 and the Bianchi identity, the instanton (A.1)
is an extremum of the (  )2 part of the effective action, L(2) , but not a global extremum.

Appendix B. Spinology, t8 -ology and -gymnastic


 For a non-abelian gauge field
t(8) tr F 4 = 8 tr(F F F F ) + 16 tr(F F F F )
4 tr(F F F F ) 2 tr(F F F F ),
t(8) tr F tr F 2 = 2 tr(F F ) tr(F F ) + 16 tr(F F ) tr(F F )
2

4 tr(F F ) tr(F F ) + 8 tr(F F ) tr(F F ),


which is easily seen to be given by the following symmetrized trace formula


1
tr t(8) F 4 = 4! Str F F F F (F F )2 .
4

(B.1)

(B.2)

 The SO(1, 9) gamma matrices are decomposed into SO(8) gamma matrices as


0
.
=
(B.3)
0
acts from the space of spinor of positive SO(8) chirality into the space of negative SO(8)
chirality: : S+ S . does the opposite. Antisymmetric products of even numbers
of Gamma matrices respects the direct sum S+ S . In particular

0
=
(B.4)
= P+ + P .
0
 The antisymmetric product of two SO(8) matrices can be expressed in terms of the
completely antisymmetric octonionic structure constants
ab
+ cmn ab .
(mn )ab = mn

(B.5)

 We need the following matrices identities (all antisymmetrizations are with unit
weight)
{ , } = 2 ,
pq

rs

= pq

rs

(B.6)
[s

+ 4[p q]

r]

sr
+ 2pq
,

= (d 2) + (d 1) ,


sn
,
tr r1 rn sm s1 = 24 n! rs11r
n
(8)r1 r8 = r1 r8 (8) ,
(8)

r1 rn

s1 s8n

rn r1

(B.7)
(B.8)
(B.9)
(B.10)

= n! s1 s8n (8) .

(B.11)

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

103

References
[1] E. Witten, Interacting field theory of open superstrings, Nucl. Phys. B 276 (86) 291.
[2] E. Witten, On background independent open string field theory, Phys. Rev. D 46 (1992) 5467,
hep-th/9208027;
E. Witten, Some computations in background independent off-shell string theory, Phys. Rev.
D 47 (1993) 3405, hep-th/9210065;
S.L. Shatashvili, Comment on the background independent open string theory, Phys. Lett. B 311
(1993) 83, hep-th/9303143;
S.L. Shatashvili, On the problems with background independence in string theory, Algebra and
Anal. 6 (1994) 215, hep-th/9311177.
[3] N. Berkovits, A new approach to superstring field theory, Fortsch. Phys. 48 (2000) 31, hepth/9912121.
[4] S.L. Shatashvili, On field theory of open strings, tachyon condensation and closed strings, Talk
at Strings 2001, Mumbai, India, hep-th/0105076.
[5] D.J. Gross, W. Taylor, Split string field theory II, hep-th/0106036.
[6] M.J. Duff, J.X. Lu, Strings from five-branes, Phys. Rev. Lett. 66 (1991) 1402.
[7] E. Witten, D-branes and K-theory, JHEP 12 (1998) 019, hep-th/9810188.
[8] A. Sen, Descent relations among bosonic D-branes, Int. J. Mod. Phys. A 14 (1999) 4061, hepth/9902105;
A. Sen, Non-BPS states and branes in string theory, hep-th/9904207;
A. Sen, Universality of the tachyon potential, JHEP 9912 (1999) 027, hep-th/9911116.
[9] A.A. Gerasimov, S.L. Shatashvili, On exact tachyon potential in open string field theory,
JHEP 0010 (2000) 034, hep-th/0009103;
D. Kutasov, M. Mario, G. Moore, Some exact results on tachyon condensation in string field
theory, JHEP 0010 (2000) 045, hep-th/0009148;
D. Kutasov, M. Mario, G. Moore, Remarks on tachyon condensation in superstring field
theory, hep-th/0010108;
A.A. Gerasimov, S.L. Shatashvili, Stringy Higgs mechanism and the fate of open strings,
JHEP 0101 (2001) 019, hep-th/0011009.
[10] M. Cederwall, B.E. Nilsson, D. Tsimpis, D = 10 super YangMills at O( 2 ), hep-th/0104236.
[11] K. Peeters, P. Vanhove, A. Westerberg, Supersymmetric higher-derivative actions in ten and
eleven dimensions, the associated superalgebras and their formulation in superspace, Class.
Quant. Grav. 18 (2001) 843, hep-th/0010167.
[12] M. Cederwall, B.E. Nilsson, D. Tsimpis, The structure of maximally supersymmetric Yang
Mills theory: constraining higher-order corrections, hep-th/0102009.
[13] N. Sakai, M. Abe, Coupling constant relations and effective Lagrangian in the type I superstring,
Prog. Theor. Phys. 80 (1988) 162.
[14] C. Bachas, C. Fabre, E. Kiritsis, N.A. Obers, P. Vanhove, Heterotic/type-I duality and D-brane
instantons, Nucl. Phys. B 509 (1998) 33, hep-th/9707126.
[15] J. Bagger, A. Galperin, A new Goldstone multiplet for partially broken supersymmetry, Phys.
Rev. D 55 (1997) 1091, hep-th/9608177;
T. Hara, T. Yoneya, Nonlinear supersymmetry without the GSO projection and unstable D9brane, Nucl. Phys. B 602 (2001) 499, hep-th/0010173;
E. Dudas, J. Mourad, Consistent gravitino couplings in non-supersymmetric strings, hepth/0012071;
S. Terashima, T. Uesugi, On the supersymmetry of non-BPS D-brane, JHEP 0105 (2001) 054,
hep-th/0104176.
[16] D. Tchrakian, Spherically symmetric gauge field configurations with finite action in 4P dimensions (P = Integer), Phys. Lett. B 150 (1985) 360.

104

R. Minasian et al. / Nuclear Physics B 613 (2001) 87104

[17] B. Grossman, T. Kephart, J. Stasheff, Solutions to YangMills field equations in eight


dimensions and the last Hopf map, Commun. Math. Phys. 96 (1984) 431.
[18] M.J. Duff, J.X. Lu, A duality between strings and five-branes, Class. Quant. Grav. 9 (1992) 1.
[19] G. OBrien, D. Tchrakian, Spin connection generalized YangMills fields on double dual
generalized EinsteinCartan backgrounds, J. Math. Phys. 29 (1988) 1212.
[20] J. Polchinski, E. Witten, Evidence for heterotic Type I String Duality, Nucl. Phys. B 460
(1996) 525, hep-th/9510169.
[21] J.M. Charap, M.J. Duff, Gravitational effects on YangMills topology, Phys. Lett. B 69 (1977)
445.
[22] A. Dundarer, Multi-instantons solutions in eight-dimensional curved space, Mod. Phys. Lett.
A 6 (1991) 409.
[23] B. Grossman, T. Kephart, J. Stasheff, Solutions to gauge field equations in eight dimensions:
conformal invariance and the last Hopf map, Phys. Lett. B 220 (1989) 431.

Nuclear Physics B 613 (2001) 105126


www.elsevier.com/locate/npe

Loop equation and Wilson line correlators


in non-commutative gauge theories
Avinash Dhar 1 , Yoshihisa Kitazawa
Laboratory for Particle and Nuclear Physics, High Energy Accelerator Research Organization (KEK),
Tsukuba, Ibaraki 305-0801, Japan
Received 11 April 2001; accepted 18 July 2001

Abstract
We investigate SchwingerDyson equations for correlators of Wilson line operators in noncommutative gauge theories. We point out that, unlike what happens for closed Wilson loops, the
joining term survives in the planar equations. This fact may be used to relate the correlator of an
arbitrary number of Wilson lines eventually to a set of closed Wilson loops, obtained by joining
the individual Wilson lines together by a series of well-defined cutting and joining manipulations.
For closed loops, we find that the non-planar contributions do not have a smooth limit in the
limit of vanishing non-commutativity and hence the equations do not reduce to their commutative
counterparts. We use the SchwingerDyson equations to derive loop equations for the correlators of
Wilson observables. In the planar limit, this gives us a new loop equation which relates the correlators
of Wilson lines to the expectation values of closed Wilson loops. We discuss perturbative verification
of the loop equation for the 2-point function in some detail. We also suggest a possible connection
between Wilson line based on an arbitrary contour and the string field of closed string. 2001
Published by Elsevier Science B.V.

1. Introduction
Non-commutative gauge theories are realized on branes in the zero slope limit in the
presence of a large NSNS B-field [18]. Recently these theories have attracted a lot of
attention. Various aspects of these theories have been studied in [936].
In ordinary gauge theories, a generic gauge-invariant observable is provided by an
arbitrary closed Wilson loop. Non-commutative gauge theories have more general gaugeinvariant observables, defined on open contours. Such gauge-invariant observables in noncommutative gauge theories were constructed in [16]. Different aspects of these were
* Corresponding author.

E-mail addresses: adhar@post.kek.jp (A. Dhar), kitazawa@post.kek.jp (Y. Kitazawa).


1 On leave from Department of Theoretical Physics, Tata Institute, Mumbai 400005, India.

0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.


PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 4 3 7 - 0

106

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

studied in [23,24,3746]. Roughly speaking, these gauge-invariant observables can be


written as Fourier transforms of open Wilson lines. In the operator formalism they are
given by the following expression



 

 ik.x
,
WC [y] = Tr P exp i d y ( )A x + y( ) e
(1.1)
C

where
[x , x ] = i ,

(1.2)

and the trace in (1.1) is over both the gauge group, taken here to be U (N), as well as the
operator Hilbert space. These open Wilson lines are gauge-invariant in non-commutative
gauge theories, unlike in ordinary gauge theories, provided the momentum k associated
with the Wilson line is fixed in terms of the straight line joining the end points of the
path C, given by y ( ) where 0   1, by the relation
y (1) y (0) = k .

(1.3)

The path C is otherwise completely arbitrary. When k vanishes, the two ends of the path
C must meet and we have a closed Wilson loop.
In an earlier work [41], based on a perturbative analysis of correlation functions of
straight Wilson lines with generic momenta, we had suggested that at large momenta the
Wilson lines are bound into the set of closed Wilson loops that can be formed by joining the
Wilson lines together in all possible different ways. In the present work we will establish
a more general connection between correlators of Wilson lines and expectation values of
Wilson loops in a non-perturbative setting, for arbitrary Wilson lines. In this generic case,
however, the closed Wilson loops to which the Wilson lines are related are not formed
by simply joining the Wilson lines together, but by more complicated cutting and joining
manipulations. Also, the statement is valid for arbitrary momenta, not necessarily large,
carried by the Wilson lines. We will use the framework of SchwingerDyson equations and
the closely related loop equations in the planar limit. In the context of non-commutative
gauge theories similar equations have been studied earlier in [4749].
This paper is organized as follows. In the next section we summarize some aspects of
operator formulation of non-commutative gauge theories, which is used throughout this
paper, and in particular list some useful identities. In Section 3 we derive Schwinger
Dyson equations for the correlators of open and closed Wilson observables and discuss
these at finite N as well as in the planar limit. As in commutative gauge theories, the
splitting term disappears from the planar equations. However, unlike in the case of closed
Wilson loops, the joining term survives in the planar equations for open Wilson lines. This
has the consequence of eventually relating them to closed Wilson loops. We also find that at
finite N , the splitting term does not reduce to the ordinary gauge theory result in the limit in
which the non-commutativity is removed. We trace this result to the UV-IR mixing in noncommutative gauge theories. In Section 4 we use the results of Section 3 to write down loop
equations for the correlators of open and closed Wilson observables. We consider the loop

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

107

equation for the 2-point function of the open Wilson lines in the planar limit and discuss
the verification of this equation in t Hooft perturbation theory in some detail. Section 5
contains a discussion of a possible connection between a Wilson line based on an arbitrary
contour and string field for closed string and the non-perturbative meaning of the new loop
equations derived here. In Appendix A we give details of the perturbative calculations.

2. Non-commutative gauge theories operator formulation


We will be working in 4-dimensional Euclidean space with a generic non-commutativity
parameter in (1.2). The non-commutative gauge theory action that we will consider is

2
1
S = 2 Tr F (x)
(2.1)
+ ,
4g
where


F (x)
= A (x)
A (x)
+ i A (x),
A (x)
.
The dots stand for possible bosonic and fermionic matter coupled to the gauge field and the
trace Tr = trU (N) trH is over the gauge group U (N) as well as the operator Hilbert space
H. To define this latter trace more precisely, let us assume that has the canonical form
01 = 10 = a ,

23 = 32 = b ,

(2.2)

with all other components vanishing, and let us define the operators
x2 + x3
x0 + x1
,
b=
,
a=
2a
2b
which satisfy the standard harmonic oscillator algebra,




a, a = 1,
b, b = 1.
The operator Hilbert space trace is then defined by

x)|n

x)
trH O(

na , nb |O(
a , nb .
= (2)2 a b

(2.3)

(2.4)

(2.5)

na ,nb

Note that with this definition of the operator Hilbert space trace, the coupling constant g
appearing in the action (2.1) is dimensionless.
We use the standard Weyl operator ordering,


x)
O(
= d 4 y O(y) (4)(x y),
(2.6)
where the operator delta-function is defined in terms of the Heisenberg group elements by

d 4 k ik.y ik.x
(4)
e
e .
(x y) =
(2.7)
(2)4
The use of this operator delta-function simplifies many calculation because it shares some
properties of the usual delta-function. For example, (2.6) and
trH (4)(x y) = 1.

(2.8)

108

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

There are, of course, differences as in the following identity which encodes the star product:
i

(4) (x y) (4) (x z) = e 2 + (4) (x y+ ) (4) (y ),

(2.9)

where y+ = (y + z)/2 and y = y z. Below we give two identities involving these


operator delta-functions which will be used in deriving the SchwingerDyson equations
in the next section. The first one joins together two operators which appear inside two
different traces,






1 (x)t

2 (x)t

1 (x)

2 (x)
Tr O
a (4) (x z) Tr O
a (4) (x z) = Tr O
O
,
d 4z
a

(2.10)
and the second one splits two operators which are inside the same trace,



1 (x)t

2 (x)t
d 4z
Tr O
a (4) (x z)O
a (4) (x z)
a

1
(2)4 det





2 (x)

1 (x)
Tr O
.
Tr O

(2.11)

Here the t a s are the generators for the gauge group, which we have taken to be U (N),
with the normalization dictated by the completeness condition

a
(2.12)
tija tkl
= il j k .
a

2.1. Wilson observables and cyclic symmetry


The generic gauge-invariant Wilson observable is given in (1.1). We will also need the
Wilson operator

 

 ik .x

C [y]0s = P exp i d y ( )A x + y( ) y(0) e s .


W
(2.13)
C

Here the subscripts 0s indicate that the path-ordered phase factor runs from = 0 to

C [y]s1 , which
= s along the curve C, and y(s) y(0) = ks . The Wilson operator W
runs from = s to = 1, is defined similarly:

 

 i ks .x

C [y]s1 = P exp i d y ( )A x + y( ) y(s) e


W
(2.14)
,
C

where y(1) y(s) = ks . These operators are related to the Wilson observable as follows:



C [y]01 = WC [y] eik.y(0).


Tr W
(2.15)
The Wilson observable WC [y] possesses a cyclic symmetry because of the trace over
both the gauge group and the operator Hilbert space. To arrive at a mathematical expression

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

109

of this symmetry, we note that

C [y]01 = e 2 ks k W

C [y]0s W

C [y]s1,
W
i

(2.16)

and 2 therefore,


i

C [y]0s W

C [y]s1
WC [y] = eik.y(0)e 2 ks k Tr W



Cs [ys ]
= eik.y(s) Tr W
WCs [ys ],

(2.17)

where in the second step we have used the cyclic property of the trace and recombined the
two operators in the opposite order. The contour Cs is given by
ys ( ) = y( + s),

0   (1 s)

= y( 1 + s) + y(1) y(0),

(1 s)   1.

(2.18)

It is obtained from the curve C by cutting it at a point = s and rejoining the two pieces
in the opposite order, as shown in Fig. 1.
If the original curve is a straight line, then the transformed curve is also a straight line
shifted by an amount s( k). It is easy to see more directly that such shifts are a symmetry
of the straight Wilson line. This symmetry was used very effectively in [41] to simplify
perturbative calculations. More generally, the cyclic symmetry relates Wilson observables
defined on contours that are nontrivially different. As we shall see later, the quantity

(k)
VC [y] = i dy ( ) eik.y( )
(2.19)
C

frequently appears in perturbative calculations of multipoint Wilson line correlation


functions. It is easy to see that in fact this quantity is invariant under the cyclic symmetry
(2.18), and that may be reason for its appearance. It is also noteworthy that the above
quantity is very similar to the vector vertex operator of open string theory. It would be
interesting to have a better understanding of these connections and the implications of the
cyclic symmetry.

Fig. 1. Cyclic symmetry of Wilson line.


2 In the following W

C [y]01 , which goes over the full parametric range, will be denoted by W

C [y] for notational


convenience.

110

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

3. SchwingerDyson equation
In this section we will first derive the SchwingerDyson equation for multipoint
correlators of Wilson observables and then analyse it at finite N as well as in the
planar limit. The SchwingerDyson equation follows from the standard functional integral
identity






S 

d 4z
0=
DAb (x)
e Tr WC1 [y1 ]t a (4) (x z)
a
A (z)
a





Cn [yn ] .

C2 [y2 ] Tr W
Tr W
(3.1)
Using

Aa (z)

C [y] = i
W



i

C [y]0s t a (4) (x z) W

C [y]s1,
dy (s) e 2 ks k W

(3.2)

and the joining and splitting identities, (2.10) and (2.11), we get
 



1  

C2 [y2 ] Tr W

Cn [yn ]

(x)
Tr WC1 [y1 ]D
Tr W
2
g
n 

i
dyl (s) e 2 kls kl
=i
l=2 C



 




Cl [yl ]s1 Tr W

C1 [y1 ]W

Cn [yn ]

C2 [y2 ] Tr W

Cl [yl ]0s W
Tr W

i
i
+
dy1 (s) e 2 k1s k1
4
(2) det
C1

 



 
 

C1 [y1 ]0s Tr W

Cn [yn ] .

C1 [y1 ]s1 Tr W

C2 [y2 ] Tr W
Tr W

(3.3)

In this equation yl (s) yl (0) = kls and each of the contours C1 , C2 , . . . , Cn may be open
or closed.
3.1. Closed Wilson loop
Let us first consider a single closed Wilson loop. In this case Eq. (3.3) reduces to

 


 
1  

(x)

C [y]s1 .

C [y]0s Tr W
Tr
[y]
D

=
dy (s) Tr W
W
C
2
4
g
(2) det
C
(3.4)
Here C is a closed curve. The right-hand side of (3.4) has a disconnected piece.
However, because of momentum conservation, the disconnected piece contributes only

C [y]0s ) and
Tr(W

C [y]s1) are proportional to


when y(s) = y(0). In fact, both
Tr(W
4
(4)
(2) (ks ). One of these gives rise to the total spacetime volume V , while the other
factor can be rewritten as (2)4 det (4) (y(s) y(0)). Thus, we may rewrite (3.4) as

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

111


1  

(x)

F
Tr WC [y]D

2
g


 
 

i

C [y]0s Tr W

C [y]s1
dy (s) (4) y(s) y(0) Tr W
=
V
C

i
(2)4 det

 

 

C [y]s1

C [y]0s Tr W
dy (s) Tr W
.
conn.

(3.5)

The second term on the right-hand side of (3.5) contains the connected part of the 2-point
function of open Wilson lines. At finite N , it is easy to see that this term is down by a
factor of 1/N 2 relative to the other terms in the equation. In the planar limit, therefore, this
term drops out and the planar equation looks formally like the corresponding equation in
commutative gauge theory. This is consistent with the perturbative result that, except for in
an overall phase, the dependence on the non-commutative parameter drops out of planar
diagrams. However, there are new gauge-invariant observables in non-commutative gauge
theory, the open Wilson lines, and so there are new equations. As we shall see shortly,
these new equations have a non-trivial planar limit. One might then say that it is these new
equations that reflect the new physics of non-commutative gauge theory.
At finite N the second term on the right-hand side of (3.5) contributes. One might
wonder whether this term reduces to its commutative counterpart in the limit of small
non-commutative parameter. An argument has been presented in [49] suggesting that this
is the case. However, we find that, in fact, the small limit of this term is not smooth, at
least in perturbation theory, as we will now show.
At the lowest order in perturbation theory, the second term on the right-hand side of
(3.5) evaluates to
ig 2 N
(2)4 det


C

1
dy (s) 2
ks

 s
dy( ) e
0

iks .y( )

 1




dy( ) e

iks .y(  )

(3.6)

For simplicity, let us specialize to a rectangular contour of sides L1 and L2 . We will also
take to be of the form in (2.2) with a = b = 0 . In this case the diagrams that contribute
to (3.6) are shown in Fig. 2.

Fig. 2. Lowest order diagrams contributing to the non-planar term.

112

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

We can easily evaluate (3.6) in this case. The result is







g2 N
2
i 2
1 i

e
f
()

f
i(L

L
)
1

f
()
+
i
e
()
,
1
2
(2)4 02
(3.7)
where = L1 1 L2 is the magnetic flux passing through the rectangular contour and


ds 
1 eis .
f () =
(3.8)
s
In the limit of small non-commutativity, is large, and then f () ln . In this case
the right-hand side of (3.5) is divergent with the leading term going as ln /02 . So we
see that if we take the limit of small non-commutativity first, keeping N finite, we do
not recover the commutative result. It is easy to see in perturbation theory that the origin
of this problem lies in UV-IR mixing. It has been argued in [22] that this phenomenon
renders loop diagrams finite in non-commutative field theory. Now, the diagrams in Fig. 2
that contribute to the right-hand side of (3.5) at order 1/N 2 in the lowest order in t Hooft
perturbation theory actually come from non-planar one-loop diagrams on the left-hand side
of this equation, as shown in Fig. 3.
In fact, the relevant amplitude is






d 4p
d 4 q eip(y1 y3 ) eiq(y2 y4 ) ipq
g 4 N dy1 dy3 dy2 dy4
e
.
(2)4
(2)4
p2
q2
(3.9)
We can estimate the above momentum integral as follows


d 4 q eip(y1y3 ) eiq(y2y4 ) ipq
d 4p
e
4
(2)
(2)4
p2
q2

1
1
1
1

if |y1 y3 ||y2 y4 | > 0 ,

4 2 (y1 y3 )2 4 2 (y2 y4 )2


=
(3.10)
1
1

ln

y
||y

y
|
<

.
if
|y
1
3 2
4
0
2(2)4 2
|y1 y3 ||y2 y4 |
0
In commutative gauge theory these diagrams have short distance singularities which
are linearly divergent. In the non-commutative theory they get regularized at the noncommutativity scale, as can be seen from the above expression. The singularities of
the commutative theory reappear in the limit of small non-commutativity. This is what

Fig. 3. Non-planar one-loop diagrams giving rise to diagrams in Fig. 2.

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

113

is reflected in the singular behaviour of the right-hand side of (3.5) for small noncommutativity.
3.2. Open Wilson lines
The generic equation satisfied by the n-point function of Wilson lines is (3.3). The
second term on the right-hand side of this equation has a disconnected part which is given
by



 
 
 

Cn [yn ]

C1 [y1 ]s1 Tr W

C2 [y2 ] Tr W

C1 [y1 ]0s Tr W
Tr W
 



 
 

C1 [y1 ]s1 Tr W

Cn [yn ] .

C1 [y1 ]0s Tr W

C2 [y2 ] Tr W
+ Tr W
(3.11)
Because of momentum conservation, the first term contributes only for y(s) = y(0), while
the second term contributes only for y(s) = y(1). In either term we get back the original npoint function. This is just like for the closed Wilson loop discussed above. The connected
part of the second term on the right-hand side can be easily seen to be down by a factor of
1/N 2 compared to the other terms in the equation. In the planar limit, therefore, it drops
out, leaving only the joining term (the first term on the right-hand side), apart from the
disconnected term mentioned above. We then have the result that the planar Schwinger
Dyson equation for Wilson lines expresses any n-point function entirely in terms of (n1)point functions. By iterating this procedure (n 1) times we may, in principle, express any
n-point function entirely in terms of closed Wilson loops.
The simplest example of the above phenomenon is provided by the 2-point function. In
this case, the planar SchwingerDyson equation reads
 

1  

(x)

C2 [y2 ]
Tr WC1 [y1 ]D
Tr W
2
g

 

i

C2 [y2 ]0s W

C2 [y2 ]s1

C1 [y1 ]W
= i dy2 (s) e 2 k2s k2 Tr W
C2

dy1 (s) e 2 k1s k1

(2)4 det
C1

 
 
 


C1 [y1 ]s1 Tr W

C2 [y2 ]

C1 [y1 ]0s Tr W
Tr W
 
 

 

C1 [y1 ]0s Tr W

C2 [y2 ] .

C1 [y1 ]s1 Tr W
+ Tr W

(3.12)

We see that the right-hand side involves closed Wilson loops, apart from the the 2-point
function itself. The closed curves involved are obtained by first traversing the curve C1
given by y1 ( ), 0   1 and then the curve given by
y( ) = y2 ( + s) y2 (s) + y1 (1),

0   (1 s),

= y2 ( 1 + s) y2 (s) + y1 (0),

(1 s)   1,

(3.13)

for different values of s, 0  s  1. Note that the closed curves obtained in this way are
continuous because of momentum conservation.

114

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

Similarly, the 3-point function involves two different 2-point functions,


 

 

C3 [y3 ] ,

C2 [y2]s1 Tr W

C2 [y2 ]0s W

C1 [y1 ]W
Tr W
 
 


C3 [y3 ]0s W

C2 [y2 ] ,

C3 [y3]s1 Tr W

C1 [y1 ]W
Tr W

(3.14)

depending on whether C1 and C2 or C1 and C3 combine into a single curve. These 2-point
functions are themselves related to closed loops, as discussed above. Thus, the 3-point
function can eventually be related to closed Wilson loops, there being two distinct sets
of structures for the closed contours involved. These closed contours can be obtained
explicitly, as we have done above for the case of the 2-point function. For n-point function
one eventually gets (n 1)! distinct structures for the closed loops.
The above discussion establishes a general link between the correlators of Wilson lines
and the expectation value of closed Wilson loops. In a previous work [41] we have
presented a perturbative proof for long straight Wilson lines to be bound together into
closed loops. The connection we have found here is more general in that the Wilson lines
are based on arbitrary contours and the momenta need not be large. The closed loops we
now find are also more general. We believe that the planar SchwingerDyson equation
indeed supports our previous claim for the high energy behaviour of the Wilson lines.
This is because, firstly, it can be argued that the planar approximation is always valid in
the high energy limit (or large limit) since the splitting term is suppressed by 1/ det .
Secondly, we need to repeatedly insert the equation of the motion operator into the Wilson
line correlator in order to eventually relate it to Wilson loops. Such an operation picks
up contact terms in the sense that it makes two Wilson lines touch each other. In high
energy limit, we expect to rediscover such contact terms, although the real singularities are
expected to be regulated by the noncommutativity. Finally, the set of relevant closed loops
may collapse to that found in [41], which is the set of extreme configurations of Wilson
loops obtained by simply joining the Wilson lines end-to-end.

4. Loop equation for Wilson lines


In this section we will first derive the loop equation for multipoint correlators of Wilson
lines. We will then consider the case of the 2-point function in detail and verify the planar
loop equation in this case upto second order in t Hooft perturbation theory.
The loop equation is basically the SchwingerDyson equation (3.3) with the insertion of
the equation of motion operator replaced by a geometric variation of the contour. This is
done with the help of the identity

C [y]
2 W
y ( )y (  )





i

(x)

C [y]  i  y (  )F

(x)

C [y]  1
= e 2 k  k W
W
W
C [y]0 i y ( )F


i

C [y]0 i y ( )D

(x)

C [y] 1 ,
(  )e 2 k k W
W
(4.1)

where, as before, k = y( ) y(0) and k = k k . Note that this identity is valid at


interior points of the Wilson line. At the boundaries of the Wilson line one has to be more

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

115

careful. However, if we vary both the ends keeping k fixed, and assume that the tangents to
the contour at the ends are identical, 3 then a very similar identity is valid at the ends also.
We need to separate out the equation of motion piece on the right hand side of (4.1).
This may formally be done by defining the loop laplacian [50]
2
lim
y 2 ( ) 00

0
dt

2
.
y ( + t/2)y ( t/2)

(4.2)

In principle, if the quantum theory is regularized then the first term in (4.1) does not have
any singularities as  and so the loop laplacian picks up only the delta-function term
on the right-hand side of this equation. 4 We then get

C [y]


i
2W

C [y] 1 .

C [y]0 i y ( )D

(x)
= e 2 k k W
W
y 2 ( )

(4.3)

Using this in (3.3) we get the loop equation for Wilson line correlators


1 2 
WC1 [y1 ]WC2 [y2 ] WCn [yn ]
2
2
g y1 ( )
n 



ds i y1 ( ).is yl (s) eik1 .y1 ( )ikl .yl (s)
=
l=2 C






C1 [y1 ]W

Cls [yls ] WCn [yn ]


WC2 [y2 ] Tr W


 ik .y ( )+ i k k
1
2 1s 1
+
ds
i
y
(
).i
y
(s)
e 1 1

1
s
1
(2)4 det
C1

 

 


C1 [y1 ]s1 WC2 [y2 ] WCn [yn ] ,

C1 [y1 ]0s Tr W
Tr W
(4.4)

where the contours C1 and C1s are as defined in (2.18). Also, the operator WC [y] has
been defined in (2.13) and WC [y] is the gauge-invariant observable defined in (1.1).
4.1. Two-point function
We will now discuss the case of the 2-point function in some detail. For the 2-point
function, the loop equation reduces to


1 2 
W
[y
]W
[y
]
C
1
C
2
1
2
g 2 y12 ( )

 




C1 [y1 ]W

C2s [y2s ]
= ds i y1 ( ).is y2 (s) eik1 .y1 ( )ik2 .y2 (s) Tr W
C2

3 Under these conditions the variation of contour at the end points is effectively like at an interior point.
4 In practice the separation of the two terms on the right hand side of (4.1) is a nontrivial issue. For a discussion

in the case of commutative gauge theory, see, for example, [5052].

116

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

1
(2)4 det

C1



i
ds i y1 ( ).is y1 (s) eik1 .y1 ( )+ 2 k1s k1
 

 


C1 [y1 ]0s Tr W

C1 [y1 ]s1 WC2 [y2 ] .


Tr W

(4.5)

We are interested in the planar limit of this equation. In this limit only the disconnected
part of the 3-point function, appearing on the right-hand side of (4.5), survives. This
disconnected part is
 

 


C1 [y1 ]0s Tr W

C1 [y1 ]s1 WC2 [y2 ]


Tr W
 
 



C1 [y1 ]s1 Tr W

C1 [y1 ]0s WC2 [y2 ] .


+ Tr W
Because of momentum conservation, the first term survives only when y1 (s) = y1 (0),
while the second term survives only when y1 (s) = y1 (1). Assuming that the contour C1
has no self-intersection, the first condition is satisfied only for s = 0, while the second
condition is satisfied only for s = 1. Thus the disconnected part of the 3-point function
takes the form
 
  
 



C1 [y1 ]0s + Tr W

C1 [y1 ]s1

C1 [y1 ] WC2 [y2 ] ,


Tr W
Tr W
which, using (2.17), is equivalent to
 

  


C1 [y1 ]0s + Tr W

C1 [y1 ]s1
eik1 .y1 ( ) Tr W
WC1 [y1 ]WC2 [y2 ] .
Using this in (4.5), together with the fact that y ( ) gives tangents at the two ends of the
contour C1 , which are equal by construction, we get


1 2 
WC1 [y1 ]WC2 [y2 ]
2
2
g y1 ( )

 
  

( y1 ( ))2

C1 [y1 ]s1

C1 [y1 ]0s + Tr W
ds
Tr
W
=
(2)4 det
C1



WC1 [y1 ]WC2 [y2 ]




 

C2s [y2s ] .

C1 [y1 ]W
+ ds i y1 ( ).is y2 (s) eik1 .y1 ( )ik2 .y2 (s) Tr W
C2

(4.6)
This is the final form of the planar loop equation for the 2-point function. Notice that the
first term on the right-hand side of this equation is proportional to ( y1 ( ))2 and also
it involves the original 2-point function. Taken together with the left hand side, the two
terms have the form of string hamiltonian acting on the 2-point function. However, it is
not clear that this term is really physically meaningful. In fact, a corresponding term in the
loop equation for the commuting gauge theory is often ignored in the regularized theory.
In the present case also, an evaluation of the coefficient of ( y1 ( ))2 cannot be done
unambiguously. This is because such a calculation involves computation of amplitudes for
splitting off of tiny bits at the two ends of the Wilson line defined on the contour C1 . The

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

117

computation of this is delicate and needs a regulator. So the physical significance of this
term remains unclear.
We should mention here that Eqs. (4.4) and (4.6) are new type of loop equations since
there is no analogue of these in commutative gauge theory. Also, it is clear that the planar
equation (4.6) relates the 2-point function of open Wilson lines to the expectation value of
a closed Wilson loop. The latter may be obtained by solving the planar loop equation for
a closed Wilson loop. Thus one needs equations for both types of Wilson observables to
form a closed system of equations.
4.2. Perturbative verification
Perhaps the most interesting aspect of the loop equation (4.6) is its stringy interpretation.
Investigating this aspect of the loop equation is bound to be inherently non-perturbative. In
fact, recently such an exercise has been successfully carried out in [51,52] for the loop
equation in commutative gauge theory, using the AdS/CFT correspondence. A similar
exercise for the present non-commutative case seems to require a better understanding
of the connection between non-commutative gauge theory and its conjectured gravity dual
[10,11], and is beyond the scope of the present work. Here we will restrict ourselves to a
perturbative verification of (4.6). We will do the computations upto the second order in the
t Hooft coupling.
Separating out the momentum conserving delta-function, we may parametrize the
2-point function as


WC1 [y1 ]WC2 [y2 ] = (2)4 (4) (k1 + k2 )GC1 C2 [y1 , y2 ],
(4.7)
where, in perturbation theory,
(1)

(2)

GC1 C2 [y1 , y2 ] = GC1 C2 [y1 , y2 ] + 2 GC1 C2 [y1 , y2 ] + .

(4.8)

Here = g 2 N is the t Hooft coupling constant. As indicated in (4.8), in perturbation


theory the function GC1 C2 [y1 , y2 ] starts at first order in the t Hooft coupling and is order
one in N . The left-hand side of (4.6) is, therefore, of order N , the same as the right-hand
side.
The lowest order diagram contributing to the 2-point function is shown in Fig. 4.
A simple calculation gives the result
(1)

GC1 C2 [y1 , y2 ] =

1 (k1 )
(k )
V [y1 ].VC22 [y2 ],
k12 C1

(4.9)

(k)

where VC [y] has been defined in (2.19). Operating the loop laplacian on (4.9), we get the
lowest order expression for the left-hand side of (4.6) 5
 (k )

i y1 ( )eik1 .y1 ( ) VC22 [y2].
(4.10)
5 Here and in the following we have omitted the momentum conserving delta-function factor (2 )4 (4) (k +
1
k2 ) which is present on both sides of (4.6).

118

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

Fig. 4. Lowest order diagram contributing to the 2-point function.

Fig. 5. Examples of diagrams contributing to the 2-point function at second order in perturbation
theory.
(k)

In arriving at this expression we have used that k.VC [y] = 0, which is true because of the
identity k.y(1) = k.y(0) which follows from the definition of k, (1.3).
On the right-hand side of (4.6), at the lowest order in , the first term does not contribute.
The relevant contribution comes from the second term by setting the gauge field to zero in
each of the two Wilson line operators involved in making the closed Wilson loop. Omitting
the momentum conserving delta-function, we get precisely the expression in (4.10).
At the next order in , there are several different types of diagrams that contribute to the
2-point function. Fig. 5 shows a representative example from each type.
(2)
We have done a calculation of the quantity GC1 C2 [y1 , y2 ], which gives the second order
contribution to the 2-point function. Some details of this calculation and the result are given
in Appendix A.
At the second order in , both terms on the right-hand side of (4.6) contribute. The
contribution of the first term comes from the lowest order calculation of the 2-point
function, while that of the second term comes from a one gauge boson exchange. In
Appendix A we have discussed in detail how each of these contributions arises as a result
of operating the loop laplacian on G(2)
C1 C2 [y1 , y2 ]. Here we only mention that some of the
terms from different sets of diagrams that appear in the calculation of the left-hand side
of (4.6) have a structure that does not occur on the right-hand side. However, there are
non-trivial cancellations between the contributions of different sets of diagrams. We have
checked that many such terms disappear from the overall result for the left-hand side as
well, but we have not attempted a complete verification of this.
It would be interesting to extend the present perturbative analysis to all orders in .
To verify the new loop equation (4.4) non-perturbatively, one needs to understand

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

119

what the multipoint correlators of Wilson lines based on arbitrary contours map on to
in the string/supergravity dual. A better understanding of the non-commutative gauge
theory/string theory duality than we have at present appears to be necessary for this.

5. Discussion
In this paper we have investigated SchwingerDyson and loop equations in noncommutative gauge theory. A major difference from the commutative case is the existence
of gauge-invariant Wilson line observables based on open contours, in addition to those
on closed contours. The SchwingerDyson and loop equations in non-commutative gauge
theories, therefore, involve both types of gauge-invariant Wilson observables. In the planar
limit, the equations for a closed Wilson loop simplify and, like in their commutative
counterparts, involve only closed loops. There are, however, new equations, those for
open Wilson lines. These involve closed Wilson loops as well, so both types of Wilson
observables are needed for a closed set of equations in the planar limit. In fact, as we have
seen, these latter equations determine correlators of open Wilson lines entirely in terms of
closed Wilson loops.
Recently in several works it has been argued [39,4246] that local operators in noncommutative gauge theory with straight Wilson lines attached to them are dual to bulk
supergravity modes. In this context it is relevant to ask what bulk observables are dual
to Wilson lines based on arbitrary open contours. This question is also important for a
non-perturbative study of the new loop equations derived here. Our proposal is to identify
a Wilson line based on an arbitrary open contour with the operator dual to bulk closed
string. This proposal is based on the following reasoning.
The momentum variable appearing in a Wilson line satisfies the condition (1.3). This is a
constraint on the contour enforced by gauge invariance. The contour is otherwise arbitrary.
This condition may be regarded as a boundary condition on the curves involved. A generic
curve with this boundary condition may be parametrized as y( ) = ( k) + y  ( ), where
0   1 and y  ( ) satisfies periodic boundary conditions. Thus the freedom contained
in a generic Wilson line is exactly the one needed to describe a closed string!
Actually, we can take this line of reasoning further. Let us confine our attention to smooth
curves, with the additional condition that the tangents to the curve at the two ends are equal.
In this case we may parametrize the curves as
y( ) = y0 ( ) + y( ),
y0 ( ) = y0 (0) + ( k),




y( ) =
n e2in + n e2in .

(5.1)

n=1

As the above parametrization suggests, what we are going to do is to assume that deviations
of the given curve from a straight line are small and expand the Wilson line, based on the
given curve, around the corresponding straight Wilson line. This gives

120

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

1
WC [y] = WC0 [y0 ] +

d y ( )

WC [y]
y ( )

1
+
2

1

1
d

y=y0

2 WC [y]
d y ( ) y ( )
y ( )y (  )



+ .

(5.2)

y=y0

Here C0 refers to the straight line contour.


The first term in the above equation is known to be the non-commutative gauge
theory operator dual to the bulk closed string tachyon. The second term vanishes, since
(WC [y]/y( ))y=y0 is independent of , which can be easily verified using the cyclic
symmetry of a straight Wilson line, and since y( ) has no zero mode. The first non-trivial
contribution comes from the third term. Using a generalization of the identity in (4.1), we
may rewrite this term as
1

1
d

d  y ( ) y (  )





C0 (,  )

C0 (0, ) il F

(x + y0 ( )) U
Tr U





C0 (  , 1)eik.x (  ) + (  )

(x + y0 (  )) U
il F
1
+






C0 (0, ) il D

C0 (, 1)eik.x

(x + y0 ( )) U
d y ( ) y ( ) Tr U

1
+




C0 (0, )F

C0 (, 1)eik.x ,

(x + y0 ( ))U
d y ( ) i y ( ) Tr U

(5.3)

C0 (1 , 2 ) is the path-ordered phase factor, running along the straight line contour
where U

C (0, 1)eik.x ).
C0 , from the point 1 to 2 . Note that in this notation WC [y] = Tr(U
0
Substituting for y( ) from (5.1) in this expression and extracting the part of the n = 1
term symmetric in the indices , , we get precisely the operator that has been identified
in [44] as being dual to the bulk graviton (in the bosonic string), polarized along the brane
directions, modulo factors that connect the open string metric with the closed string metric
and terms involving the scalar fields. Note that the last term in (5.3) is purely antisymmetric
in the indices , and hence contributes only to the operator dual to the bulk antisymmetric
tensor field.
It seems quite likely that the above procedure gives us all the gauge theory operators
dual to bulk string modes. It is, therefore, tempting to identify the Wilson line based
on generic curves of the type described by (5.1) as dual to the bulk closed string. An
expansion of the Wilson line around the corresponding straight line contour would then be
like the expansion of the closed string field in terms of the various string modes carrying
a definite momentum. If this is true, then multipoint correlators of Wilson lines should be

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

121

identified with closed string scattering amplitudes. In particular, the 2-point function would
then have the interpretation of closed string propagator and (4.6) would be the equation of
motion satisfied by the propagator. Such a non-perturbative interpretation of (4.6), or more
generally (4.4), should further enhance our understanding of gauge theory/string theory
duality. The above discussion applies to the bosonic string. It would be interesting to extend
these ideas to the case of the superstring.

Appendix A
(2)

In this appendix we will give some details of the calculation of GC1 C2 [y1 , y2 ]. We will
also describe how the action of the loop laplacian on it reproduces the right-hand side of
the loop equation (4.6).
(2)
The diagrams that contribute to GC1 C2 [y1, y2 ] can be collected into four different types
of groups. A representative from each of these has been shown in Fig. 5. There are six
self-energy type of diagrams, Fig. 5(a). Their total contribution to G(2)
C1 C2 [y1 , y2 ] is



1
(y1 (1 ).y2 (2  ))(y2 (2 ).y2 (2  )) ik1 .(y1 (1 )y2 (2  ))
e
4 2 |y2 (2 ) y2 (2  )|2
k12
1 2 >2  > 
2

(y1 (1 ).y2 (2  ))(y2 (2 ).y2 (2  )) ik1 .(y1 (1 )y2 (2  ))


e
4 2 |y2 (2 ) y2 (2  ) + k1 |2

(y1 (1 ).y2 (2 ))(y2 (2  ).y2 (2  )) ik1 .(y1 (1 )y2 (2 ))
+
e
4 2 |y2 (2  ) y2 (2  )|2
+

+ (1 2),

(A.1)

where a dot on y stands for a derivative with respect to the argument and the last
contribution is obtained by the 1 2 interchange of the subscripts on k, y and .
There are two diagrams in the second set represented by Fig. 5(b). Their total
contribution to G(2)
C1 C2 [y1 , y2 ] is


 
(y1 (1 ).y2 (2 ))(y1 (1  ).y2 (2  ))
4 ik1 .z
d ze
4 2 |z + y1 (1 ) y2 (2 )|2 4 2 |z + y1 (1  ) y2 (2  ) + k1 |2
1 >1  2 >2 


(y1 (1  ).y2 (2 ))(y1 (1 ).y2 (2  ))
.
+
4 2 |z + y1 (1 ) y2 (2  )|2 4 2 |z + y1 (1  ) y2 (2 )|2
(A.2)
In the third set, represented by Fig. 5(c), also there are two diagrams. Their total
contribution to G(2)
C1 C2 [y1 , y2 ] is
 

1
eik1 .(z+y2 (2 )y1 (1 ))
2 d 4z
4 2 z2
k1
1 2 >2 




ik1 y1 (1 ) y2 (2  )y2 (2 ) 2y2 (2 )y2 (2  )


+ y1 (1 ) y2 (2  )y2 (2 ) + y2 (2 )y2 (2  ) 2y2 (2 ).y2 (2  ) z

122

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

1
1

4 2 |z + y2 (2 ) y2 (2  ) + k1 |2 4 2 |z + y2 (2 ) y2 (2  )|2
+ (1 2).

(A.3)

Finally, we have the gauge boson self-energy diagrams like Fig. 5(d), including those
with ghosts. Their total contribution to G(2)
C1 C2 [y1 , y2 ] is
 

1
y1 (1 )y2 (2 )

eik1 .(zy1 (1 )+y2 (2 ))


d 4z
2
2
4 2 z2
(k1 )
1 2




 
1
1
2
2

z + 2i k1 .z 5k1 + 8z z
.
4 2 |z + k1 |2 4 2 z2
(A.4)
Let us now evaluate the action of the loop laplacian, 2 /y12 ( ), on the expression for
the second order contribution to the 2-point function given above. In the first term in (A.1),
(k )
the only dependence on y1 is in the form of VC11 [y1 ], which has been defined in (2.19).
Applying the loop laplacian on it gives the result



2
(k )
V 1 [y1 ] = k12 k1 k1 y1 ( )eik1 .y1 ( ) .
y12 ( ) C1

(A.5)

The k12 in the first term above cancels the factor of 1/k12 in front of the first term in (A.1).
The rest of this factor can be seen to precisely reproduce that contribution of the second
term on the right-hand side of the loop equation (4.6) in which a self-energy insertion
is present on the contour C2s . The three terms correspond to the three possibilities that
the marked point s on the contour C2s is entirely above, entirely below or in-between the
points where the self-energy insertion takes place. The second term in (A.5) gives rise to
the following contribution from the first term in (A.1):

k1 .y1 ( ) ik1 .y1 ( )


e
k12


(k1 .y2 (2  ))(y2 (2 ).y2 (2  )) ik1 .y2 (2  )
e

4 2 |y2 (2 ) y2 (2  )|2
2 >2  >2

(k1 .y2 (2  ))(y2 (2 ).y2 (2  )) ik1 .y2 (2  )


e
4 2 |y2 (2 ) y2 (2  ) + k1 |2

(k1 .y2 (2 ))(y2 (2  ).y2 (2  )) ik1 .y2 (2 )
.
+
e
4 2 |y2 (2  ) y2 (2  )|2
+

(A.6)

This can be simplified to


ik1 .y1 ( ) ik1 .y1 ( )
e
k12




 
y2 (2 ).y2 (2  ) eik1 .y2 (2 ) eik1 .y2 (2 )

2 >2 


1
1

.
4 2 |y2 (2 ) y2 (2  ) + k1 |2 4 2 |y2 (2 ) y2 (2  )|2

(A.7)

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

123

Now, it is easy to see that on the right-hand side of (4.6) there are no terms at this order
having the above structure. Therefore, for consistency of the loop equation, (A.7) must get
cancelled by another term in the 2-point function. In fact, this does happen and the required
(k )
term comes from the first term of (A.3). In this term also y1 is in the form of VC11 [y1], so
applying the loop laplacian results in two terms because of (A.5). Let us look at the second
term. It is


eik1 .(z+y2 (2 )) 
k1 .y1 ( ) ik1 .y1 ( )
4
e
z
d
ik1.y2 (2  )k1 .y2 (2 )
4 2 z2
k12
2 >2 



+ k1 y2 (2  )y2 (2 ) + y2 (2 )y2 (2  ) 2y2 (2 ).y2 (2  ) z


1
1

. (A.8)
4 2 |z + y2 (2 ) y2 (2  ) + k1 |2 4 2 |z + y2 (2 ) y2 (2  )|2
In the last term in the square brackets above, let us rewrite 2k1.z as i{(k1 + iz )2 +
z2 k12 } and then use (k1 + iz )2 (eik1 .z /z2 ) = eik1 .z z2 (1/z2 ) and z2 (1/z2 ) = 4 2 (4)(z).
As a result of this simplification, one of the terms we get is precisely (A.7), but with
opposite sign. So this unwanted term cancels in a rather nontrivial way, since the
cancellation involves terms which come from two entirely different diagrams.
Going back to (A.1), let us now look at the second term, which is obtained from the first
by 1 2 interchange:



(y2 (2 ).y1 (1  ))(y1 (1 ).y1 (1  )) ik1 .(y2 (2 )y1 (1  ))
1
e
4 2 |y1 (1 ) y1 (1  )|2
k12
2 1 >1  > 
1

(y2 (2 ).y1 (1  ))(y1 (1 ).y1 (1  )) ik1 .(y2 (2 )y1 (1  ))


e
4 2 |y1 (1 ) y1 (1  ) k1 |2

(y2 (2 ).y1 (1 ))(y1 (1  ).y1 (1  )) ik1 .(y2 (2 )y1 (1 ))
+
e
.
4 2 |y1 (1  ) y1 (1  )|2

(A.9)

The y1 structure of this term is much more complicated than that of the first term in
(A.1). So the result of applying the loop laplacian on it is also more complicated. For
example, let us consider the first term in the above expression. If the loop laplacian acts

on y1 (1  )eik1 .y1 (1 ) , the result is simple and, in fact, just reproduces that contribution
of the second term on the right-hand side of the loop equation (4.6) in which a self-energy
insertion is present on the contour C1 entirely above the marked point . A similar
operation of the loop laplacian on the other two terms in (A.9) reproduces the other
two contributions in which the self-energy insertion is either entirely below the marked
point or across it. On the other hand, if the loop laplacian acts on the propagator
1/4 2 |y1 (1 ) y1(1  )|2 , the result is a delta-function type of contribution. Together with
a similar contribution from the last term in (A.9) (the middle term has no contribution
of this type since the delta-function does not click), this precisely reproduces the entire
contribution of the first term on the right-hand side of the loop equation in this order.
Let us now go to the next term, (A.2). Here, the loop laplacian may act on any of
the four propagators, resulting in a delta-function. This gives four different terms and

124

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

these precisely reproduce that contribution of the second term on the right-hand side of
the loop equation (4.6) in which a gauge boson is exchanged between the two contours
C1 and C2s . For this it is essential to remember that the loop on the right-hand side,

C2s [y2s ]), involves the hatted operators. As defined in (2.17), these differ

C1 [y1 ]W
Tr(W
from the unhatted ones in that the argument of the gauge field is shifted by the starting
point of the contour. For the contours C1 and C2s , the starting points are, respectively,
y1 (0) = y1 ( ) and y2s (0) = y2 (s). The four terms mentioned above correspond to the
four different possibilities of the two ends of the gauge field propagator landing above or
below the marked points on the two contours.
Thus, we see that the action of the loop laplacian on the second order contribution
to the 2-point function reproduces all the terms expected on the right-hand side of the
loop equation (4.6). There are also other terms produced in the process of applying
the loop laplacian on the contribution of individual diagrams to the 2-point function.
We have seen above an explicit example of one such term which, however, eventually
disappeared because of a nontrivial cancellation with another term. We expect that all other
similar terms (which are produced in the process of applying the loop laplacian on the
contributions of different diagrams to the 2-point function, but are not present on the righthand side of the loop equation) will eventually disappear through similar cancellations, but
we have not attempted a complete verification of this.

References
[1] A. Connes, M. Douglas, A. Schwarz, Non-commutative geometry and Matrix theory:
compactification on tori, JHEP 9802 (1998) 002, hep-th/9711162.
[2] M.R. Douglas, C. Hull, D-branes and the non-commutative torus, JHEP 9802 (1998) 008, hepth/9711165.
[3] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Mixed branes and Matrix theory on noncommutative torus, hep-th/9803067;
F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Non-commutative geometry from strings and
branes, JHEP 9902 (1999) 016, hep-th/9810072.
[4] Y. Cheung, M. Krogh, Non-commutative geometry from D0 branes in a background B-field,
Nucl. Phys. B 528 (1999) 185, hep-th/9803031.
[5] H. Garcia-Compean, On the deformation quantization description of matrix compactifications,
Nucl. Phys. B 541 (1999) 651, hep-th/9804188.
[6] C. Chu, P. Ho, Non-commutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hepth/9812219.
[7] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/
9903205.
[8] N. Seiberg, E. Witten, String theory and non-commutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[9] N. Nekrasov, A.S. Schwarz, Instantons on non-commutative R 4 and (2, 0) superconformal sixdimensional theory, Commun. Math. Phys. 198 (1998) 689, hep-th/9802068.
[10] J. Maldacena, J. Russo, Large N limit of non-commutative gauge theories, hep-th/9908134.
[11] A. Hashimoto, N. Itzhaki, Non-commutative YangMills and the AdS/CFT correspondence,
Phys. Lett. B 465 (1999) 142, hep-th/9907166.

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

125

[12] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Non-commutative YangMills in


IIB matrix model, Nucl. Phys. 565 (2000) 176, hep-th/9908141.
[13] M. Li, Strings from IIB matrices, Nucl. Phys. 499 (1997) 149, hep-th/9612222.
[14] L. Cornalba, D-brane physics and noncommutative YangMills theory, hep-th/9909081.
[15] N. Ishibashi, A relation between commutative and non-commutative descriptions D-branes,
hep-th/9909176.
[16] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in non-commutative YangMills,
Nucl. Phys. B 573 (2000) 573, hep-th/9910004.
[17] N. Seiberg, E. Witten, D1/D5 system and singular CFT, JHEP 9904 (1999) 017, hep-th/
9903224.
[18] A. Dhar, G. Mandal, S.R. Wadia, K.P. Yogendran, D1/D5 system with B field, non-commutative
geometry and the CFT of the Higgs branch, Nucl. Phys. B 575 (2000) 177, hep-th/9910194.
[19] S.R. Das, S. Kalyana Rama, S. Trivedi, Supergravity with self-dual B fields and instantons in
non-commutative gauge theory, JHEP 0003 (2000) 004, hep-th/9911137.
[20] I. Bars, D. Minic, Non-commutative geometry on a discrete periodic lattice and gauge theory,
Phys. Rev. D 62 (2000) 1050, hep-th/9910091.
[21] J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Finite N matrix models of non-commutative
gauge theory, JHEP 9911 (1999) 029, hep-th/9911041;
J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Non-perturbative dynamics of noncommutative gauge theory, Phys. Lett. B 480 399, hep-th/0002158;
J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Lattice gauge fields and discrete noncommutative YangMills theory, JHEP 0007 (2000) 013, hep-th/0003208.
[22] S. Minwalla, M.V. Raamsdonk, N. Seiberg, Non-commutative perturbative dynamics,
JHEP 0007 (2000) 013, hep-th/0003160.
[23] O. Andreev, H. Dorn, On open string sigma-model and non-commutative gauge fields, Phys.
Lett. B 476 (2000) 402, hep-th/9912070.
[24] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, String scale in non-commutative YangMills, Nucl.
Phys. B 583 (2000) 159, hep-th/0004038.
[25] S.R. Das, B. Ghosh, A note on supergravity duals of non-commutative YangMills theory,
JHEP 0006 (2000) 043, hep-th/0005007.
[26] L. Alvarez-Gaum, S.R. Wadia, Gauge theory on a quantum phase space, hep-th/0006219.
[27] A.H. Fatollahi, Gauge symmetry as symmetry of matrix coordinates, hep-th/0007023.
[28] D. Jatkar, G. Mandal, S.R. Wadia, NielsenOlesen vortices in noncommutative abelian Higgs
model, JHEP 0009 (2000) 018, hep-th/0007078.
[29] J.A. Harvey, P. Kraus, F. Larsen, E. Martinec, D-branes and strings as non-commutative
solitons, JHEP 0007 (2000) 042, hep-th/0005031.
[30] K. Dasgupta, S. Mukhi, G. Rajesh, Non-commutative tachyons, JHEP 0006 (2000) 022, hep-th/
0005006.
[31] E. Witten, Non-commutative tachyons and string field theory, hep-th/0006071.
[32] R. Gopakumar, S. Minwalla, A. Strominger, Symmetry restoration and tachyon condensation
in open string theory, hep-th/0007226;
R. Gopakumar, S. Minwalla, A. Strominger, Non-commutative solitons, JHEP 0005 (2000)
020, hep-th/0003160.
[33] N. Seiberg, A note on background independence in non-commutative gauge theories, matrix
model and tachyon condensation, hep-th/0008013.
[34] P. Kraus, A. Rajaraman, S. Shenker, Tachyon condensation in non-commutative gauge theory,
hep-th/0010016.
[35] G. Mandal, S.R. Wadia, Matrix model, noncommutative gauge theory and the tachyon potential,
hep-th/0011094.
[36] A. Dhar, Y. Kitazawa, Wilson loops in strongly coupled noncommutative gauge theories, hepth/0010256, Phys. Rev. D, in press.

126

A. Dhar, Y. Kitazawa / Nuclear Physics B 613 (2001) 105126

[37] S.-J. Rey, R. von Unge, S-duality, non-critical open string and non-commutative gauge theory,
hep-th/0007089.
[38] S.R. Das, S.-J. Rey, Open Wilson lines in non-commutative gauge theory and tomography of
holographic dual supergravity, hep-th/0008042.
[39] D.J. Gross, A. Hashimoto, N. Itzhaki, Observables of non-commutative gauge theories, hep-th/
0008075.
[40] A. Dhar, S.R. Wadia, A note on gauge invariant operators in non-commutative gauge theories
and the matrix model, Phys. Lett. B 495 (2000) 413, hep-th/0008144.
[41] A. Dhar, Y. Kitazawa, High energy behaviour of Wilson lines, JHEP 0102 (2001) 004, hep-th/
0012170.
[42] H. Liu, -Trek II, hep-th/0011125.
[43] S.R. Das, S. Trivedi, Supergravity couplings to noncommutative branes, open Wilson line and
generalized star products, hep-th/0011131.
[44] Y. Okawa, H. Ooguri, How noncommutative gauge theories couple to gravity, hep-th/0012218;
Y. Okawa, H. Ooguri, Energy momentum tensors in Matrix theory and in noncommutative
gauge theories, hep-th/0103124.
[45] H. Liu, J. Michelson, Supergravity couplings of noncommutative D-branes, hep-th/0101016.
[46] K. Okuyama, Comments on open Wilson lines and generalized star products, hep-th/0101177.
[47] M. Fukuma, H. Kawai, Y. Kitazawa, A. Tsuchiya, String field theory from IIB matrix model,
Nucl. Phys. B 510 (1998) 158, hep-th/9705128.
[48] A. Dhar, S.R. Wadia, unpublished.
[49] M. Abou-Zeid, H. Dorn, Dynamics of Wilson observables in non-commutative gauge theory,
hep-th/0009231.
[50] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic, 1997.
[51] A.M. Polyakov, V.S. Rychkov, Gauge fields-strings duality and the loop equation, hep-th/
0002106;
A.M. Polyakov, V.S. Rychkov, Loop dynamics and AdS/CFT correspondence, hep-th/0005173.
[52] N. Drukker, D.J. Gross, H. Ooguri, Wilson loops and minimal surfaces, Phys. Rev. D 60 (1999)
125006, hep-th/9904191.

Nuclear Physics B 613 (2001) 127146


www.elsevier.com/locate/npe

Hopf-reductions, fluxes and branes


Ruben Minasian a , Dimitrios Tsimpis b
a Centre de Physique Thorique, Ecole Polytechnique 1 , F-91128 Palaiseau, France
b Department of Theoretical Physics, Gteborg University and Chalmers University of Technology,

SE-412 96 Gteborg, Sweden


Received 4 July 2001; accepted 2 August 2001

Abstract
We use a series of reductions, T -dualities and liftings to construct connections between fractional
brane solutions in IIA, IIB and M-theory. We find a number of phantom branes that are not supported
by the geometry, however materialize upon untwisting and/or Hopf-reduction. 2001 Published by
Elsevier Science B.V.
PACS: 11.25.-w; 04.65.+e

1. Introduction and discussion


The study of M-theory backgrounds with U (1) isometries and their associated Hopfreductions dates back to the eighties (see, e.g., [1,2]). In the presence of branes and
depending on whether the direction of the reduction is parallel or transverse to the brane,
the dimensionality of the brane worldvolume may or may not change under the reduction.
A subsequent T -duality transformation can be applied. In the case of AdSDd S d
backgrounds the fact that both the AdSDd and the S d factors can be thought of as
U (1) fibrations has been used extensively in order to establish new vacua through Hopfreductions/T -dualities. One of the lessons was that these operations can untwist the U (1)
bundles and break supersymmetry at the level of supergravity solutions (although this may
not be true at the level of full string theory) [35]. The recent interest in supergravity
solutions involving branes wrapping cycles in conifold backgrounds motivates us to
investigate some reductions and dualities in this case.
Our starting point is the singular type IIB solution of [6] (the KT solution). The geometry
is of the form (warped) R1,3 C(T 1,1 ) where the transverse space C(T 1,1 ) is the conifold
of [7], which is a cone over T 1,1 . The latter is topologically S 2 S 3 . Since this can
E-mail addresses: ruben@cpht.polytechnique.fr (R. Minasian), tsimpis@fy.chalmers.se (D. Tsimpis).
1 Unit mixte du CNRS et de lEP, UMR 7644.

0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.


PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 8 7 - X

128

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

be thought of as a U (1) fibration over S 2 S 2 , we can Hopf-reduce/T -dualize along


the U (1). The KT solution involves a system of ordinary and fractional D3-branes. The
latter are thought of in this context as D5-branes wrapping the S 2 of T 1,1 . The T -duality
in this case will be along a direction transverse to the branes.
A closely related non-singular system is the KS solution [8]. In this case the transverse
space is instead the deformed conifold of [7]. The latter is obtained from the conifold by
replacing the apex by an S 3 . In the context of KS, the D5 wrapping the S 2 produces flux
through the S 3 , which is thus stabilized (kept finite). The deformed conifold is also a U (1)
fibration, although Hopf-reducing is more complicated technically.
In a related development [9,10] it was realized that a system of D6-branes in IIA theory
wrapping the S 3 of the resolved conifold (the AMV solution) can be lifted in M-theory
to a manifold of G2 holonomy. The resolved conifold is obtained from the conifold by
replacing the apex by an S 2 . In the case of AMV, the D6-branes wrapping the S 3 produce
flux through the S 2 , which is stabilized. 2 Once more, metamorphoses of U (1) bundles in
this and related setups provide interesting connections.
This paper contains a series of Hopf-reductions, T -dualities and liftings involving branes
in the geometries discussed above. There are two points that we find worth emphasizing.
While T -duality can untwist the U (1) bundles, the M-theory lifting (in the presence of
fractional branes) can induce new twistings. This has received much attention lately due to
[10]. In the liftings considered here, orientation issues play a subtle role. Starting from IIB
theories in a background involving a T 1,1 space we may end up in an M-theory background
with a different T 1,1 .
We find a number of examples with twisted geometries (with respect to the U (1) fibre)
where there are fluxes, but where there are no corresponding non-contractible cycles for
the fluxes to be integrated over into charges. We think of these backgrounds as phantom
branes that are not supported by the (twisted) geometry. However, there are two instances
in which real branes can emerge from phantom ones.
(a) By switching-off some fluxes (typically corresponding to fractional branes) the
geometry may get untwisted giving thus rise to the cycles necessary for supporting
the branes. Note that this is not a continuous process since it involves taking the
number of fractional branes (an integer) to zero.
(b) By Hopf-reduction.
In a way this could be considered the discrete analogue of the large-N duality between
fluxes in twisted geometry and branes. A somewhat similar phenomenon which has already
appeared in the literature is the case of winding strings in the background of the Kaluza
Klein monopole [12]. We comment on this in Section 3.
The outline of the paper is as follows. Since the properties of T 1,1 spaces will be
important in the following, we review these in Section 2. Special attention is paid to issues
2 In the terminology of [10], this is the situation where the D6-branes have disappeared and have been replaced
by the F2 flux. Here we use the notion of brane modulo the large N duality of [11], namely both for N
D6-branes wrapping T S 3 and for the physically equivalent case of the resolved conifold with N units of flux
through the S 2 .

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

129

related to orientations of cycles. The T -dual and the M-theory lift of the KT solution and its
S-dual are discussed in Sections 3 and 4, respectively. Here we find a number of phantom
branes that are not supported by the geometry, however materialize upon Hopf-reduction
and/or untwisting. Section 5 is more speculative and concerns fivebranes in the geometry of
AMV. Finally in Section 6 we explore the U (1) isometry of the deformed conifold and find
a new set of variables in which this isometry is explicit. Some useful formulae concerning
different reductions and T -dualities are collected in Appendix A.

2. The geometry of T p,q spaces


T p,q spaces can either be thought of as U (1) fibrations over S 2 S 2 or as SU(2)
SU(2)/U (1) coset spaces. Let 0  i  2 , 0  i  , i = 1, 2 parametrize the two S 2
and let 0   4 be the coordinate of the U (1) fibre. The most general T p,q metric
reads
dsT2 = D(d + p cos 1 d1 + q cos 2 d2 )2




+ A sin2 1 d12 + d12 + C sin2 2 d22 + d22

+2B cos (d1 d2 sin 1 sin 2 d1 d2 )

+ sin (sin 1 d1 d2 + sin 2 d2 d1 ) ,

(2.1)

where A, B, C, D are constants.


It is convenient to introduce the following basis of one-forms, motivated by the geometry
of coset spaces [13]:
 1 

 3 


e
sin 1 d1
e
cos sin
sin 2 d2
=
,
=
,
sin
cos
e2
d1
e4
d2
(p,q)

e5 = d + A1

(p,q)

A1

= p cos 1 d1 + q cos 2 d2 .

(2.2)

(p,q)

A1
is the connection 1-form of the (p, q) U (1) bundle over S 2 S 2 . In terms of the
above base the metric (2.1) becomes
 2
 2  2 
 2  2 


dsT2 = D e5 + A e1 + e2 + C e3 + e4 + 2B e1 e3 + e2 e4 .

(2.3)

Of particular interest to us is the space T 1,1 which is topologically S 3 S 2 . A basis for the
harmonic representatives of the (one-dimensional) spaces H 2 (T 1,1 , Z), H 3 (T 1,1 , Z) was
constructed in [14]:
2 = (sin 1 d1 d1 sin 2 d2 d2 ) = e1 e2 e3 e4 ,
3 = e5 2 .

(2.4)

Both 2 and 3 are closed. Locally we can write


2 = d1 ,

1 = cos 1 d1 cos 2 d2 .

(2.5)

130

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

We denote by S 2 , S 3 the basis of homology cycles for the spaces H2 (T 1,1 , Z), H3 (T 1,1 , Z)
so that


(2.6)
3 = 2 = 1.
S3

S2

Since b2 (T 1,1 ) = 1 (b3 (T 1,1 ) = 1), there is only one S 2 (S 3 ) homologically. Consider the
two different 3-spheres corresponding to the orbits of each of the two SU(2) factors in
the isometries of T 1,1 . Let us denote them by 3(i) , i = 1, 2. They are parametrized by
(i)
(i , i , ). Similarly, let us denote by 2 the two 2-spheres parametrized by (i , i ),
i = 1, 2. The homology basis can be written as [14,15]:
(1)

(2)

S 3 = 3 3 ,

(1)

(2)

S 2 = 2 2 .

(2.7)

Following [14,15] we will interpret a (fractional) brane wrapping (1) as being equivalent
to an antibrane wrapping (2) . Under the transformation 2 2 , 1 becomes the
U (1) connection over a T 1,1 with reversed orientation. We have,
A1 1 ,

2 dA1 ,

(2)
(2)
2,3
2,3
,

(2.8)

where A1 := A1(1,1) . Under this orientation-reversal, a brane wrapping (2) becomes an


antibrane and vice versa. We also note that (2.6) can be written as



= 1.
(2.9)
(1)

(2)

On the other hand,




+
= 0.
(1)

(2.10)

(2)

Taking (2.8) into account, a similar set of relations can be easily derived for the integrals
of dA1 , e5 dA1 over 2,3 .
Finally we can define the T 1,1 space which differs from T 1,1 in the relative sign
between the two factors in the U (1) connection or, equivalently, in the relative orientation
of the two S 2 in the base. An equation similar to (2.8) relates the T 1,1 with its orientationreversal. Note that A1 can be either the U (1) connection of a T 1,1 or of a T 1,1 with
reversed orientation. Similarly, 1 can be either the U (1) connection of a T 1,1 or of
a T 1,1 with reversed orientation.

3. The KlebanovTseytlin solution


We start from the ten-dimensional IIB singular solution for N ordinary and M fractional
D3-branes, presented in [6]. The geometry is of the form (warped) R 1,3 C(T 1,1 ) where
the transverse space C(T 1,1 ) is the conifold of [7] which is a cone over T 1,1 . T 1,1 has a
U (1) isometry and we can apply the techniques of Hopf-reduction/T -duality explained

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

131

in Appendix A. Reducing along the U (1) to nine-dimensional IIB, dualizing to IIA and
lifting back to ten dimensions, we find that the U (1) bundle gets untwisted so that the
transverse space is foliated by 2(1) 2(2) S 1 . We will see how in a further lifting
to eleven dimensions a new twisted direction develops so that the transverse space is
foliated by T 1,1 S 1 .
The KT solution reads
r
B2 = 3gs M ln 2 ,
F3 = M3 ,
e = gs ,
r0


r
F5 = F5 + F5 ,
(3.1)
2 3 ,
F5 = N + 3gs M 2 ln
r0
where 2 , 3 where defined in the previous section. The metric is


2
= h1/2 dx dx + h1/2 dr 2 + r 2 dsT2 ,
ds10
where
h(r) = b0 +



c gs
r
3
2
2
g
M
+
3g
M
ln
N
+
.
s
s
r4
4
r0

(3.2)

(3.3)

The metric dsT2 is the T 1,1 metric defined in the previous section for the special case D =
1/9, A = 1/6, B = 0. With this choice of constants T 1,1 becomes Einstein. The constant
b0 can be fixed by demanding that the asymptotically flat limit of the metric be the standard
Minkowski. Similarly the constant c can be fixed by requiring that for M = 0 the metric
reduce to the standard AdS5 S 5 in the near-horizon limit. We also have, 3


r
F3 = M,
F5 = N + 3gs M 2 ln .
(3.4)
r0
S3

T 1,1

The identification of fractional D3-branes with D5-branes wrapping the S 2 of the conifold,
was justified in [16].
3.1. The IIA versions of KT solution
As already mentioned, reducing the KT solution along the U (1) fibre to ninedimensional IIB, dualizing to IIA and lifting to ten-dimensional IIA, we find that the
U (1) bundle gets untwisted so that the transverse space is foliated by 2(1) 2(2)
S 1 . The interpretation of the T -dualized (IIA) version of KT, depends on our choice of
orientation for (2) . Geometrically speaking, there seems to be no reason for preferring
one orientation over the other, but the brane content is entirely different in each case.
It must be noted that while reversing the orientation of a solution is also a solution to
the supergravity equations, the amount of preserved supersymmetry may change. In fact,
if an Einstein space that is not a round sphere has Killing spinors, its orientation reversal
gives a space with no Killing spinors [17]. So trying to preserve the original amount of
supersymmetry will choose an orientation for us, however we will consider both cases
3 We are dropping numerical constants of proportionality.

132

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

here. Concerning the effect of T -duality/Hopf-reduction we note that T -duality at the full
string theory level always preserves as much supersymmetry as has survived the U (1)
compactification. The U (1) compactification in the case where the compactifying direction
is naturally periodic (as is here) also leaves supersymmetry unbroken at the full string
theory level. At the level of supergravity however, it is known that these operations can
brake the supersymmetries of p-brane solutions (see [5] for a detailed discussion).
3.1.1. T-duality
Let us start with the orientation for 2(2) which is inherited from (2.7). Reducing to nine
dimensions, dualizing and lifting back to ten-dimensional type IIA, we obtain the following
solution
dr
2 + dA1 dz,
F3 = 3gs M
F2 = M2 ,
r


r
2 2 ,
F4 = N + 3gs M 2 ln
r0

(3.5)

while the metric reads





1/4 
2



1
1
2
1/8
(i)
2
d2
= gs
h1/2 dx dx + h1/2 dr 2 + r 2
ds10
h1/2 r 2
9
6
i=1

3/4
7/8 1 1/2 2
h r
+ gs
(3.6)
dz2 ,
9
where (d2(i) )2 := sin2 i di2 + di2 . We see that the dualization has untwisted the U (1)
bundle so that the transverse space is foliated by S 2 S 2 S 1 .
There is a charge Q(6) corresponding to D6-branes wrapping an S 2 S 1 . This gives
fractional D3-branes! Indeed, let us denote by 1,2 the two S 2 factors. We have


Q(6) =
(3.7)
F2
F2 = M,
(1)

(2)

as can be seen from (2.6).


There is however zero net Q(5) charge corresponding to NS5-branes wrapping an S 2


Q(5) =
F3
F3 = 0.
(3.8)
(1)

2 S 1

(2)

2 S 1

Finally, there is a Q(4) charge corresponding to D4-branes wrapping S 1



r
(4)
Q =
F4 = N + 3gs M 2 ln .
r0

(3.9)

(1)
(2)
2 2

The brane content is in agreement with the (naive) expectations from T -duality: the D5
(fractional D3) become D6, the D3 become D4 wrapping the T -duality circle.

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

133

3.1.2. The orientation-reversed version


We now perform an orientation-reversal transformation on the KT solution in IIA. Upon
changing the orientation, the geometrical interpretation of the solution changes. This is
because a brane wrapping a non-trivial cycle becomes an antibrane upon changing the
orientation of the cycle. More specifically there is now zero net charge Q(6) corresponding
to D6-branes wrapping an S 2 S 1


(6)
Q =
F2 +
F2 = 0.
(3.10)
(1)

(2)

In other words, there is an equal number of D6-branes and antibranes.


There is however non-zero net Q(5) charge corresponding to an NS5-brane wrapping S 2


Q(5) =
F3 +
F3 = 1.
(3.11)
(1)

2 S 1

(2)

2 S 1

The Q(4) charge corresponding to D4-branes wrapping S 1 is still given by



r
(4)
F4 = N + 3gs M 2 ln .
Q =
r0

(3.12)

(1)
(2)
2 2

The above brane content should be compared with the system of [15,1820] where there
are two NS5-branes, one of them stretching along the directions x 05 , and the other along
x 03,7,8 . The x 6 direction is a circle. There are also two types of D4-branes along x 03,6 .
N of them going around the circle and M of them stretching between the two NS5-branes.
(1)
In order to compare to the situation at hand, we should identify x 4,5 with 2 , say, and
(2)
x 7,8 with 2 . The circle x 6 should be identified with the S 1 . The constant factor N on
the right-hand side of (3.12) is coming from the D4-branes going around the circle, while
the r-dependent piece should be attributed to the D4-branes stretching between the NS5branes. Having these two types of fourbranes will be important when the solution is lifted
to eleven dimensions.
3.2. The M-theory versions of KT
Lifting to eleven-dimensional supergravity will produce two distinct solutions, corresponding to the two versions of KT in type IIA. In both cases a new twisted U (1) develops.
Let us start with the second case discussed in the last section (solution (3.10)(3.12))
and lift to eleven dimensions. The solution reads



 
2
1 1/2 2 1/3 1/2
1 2  (i) 2
2
1/2
2
ds11 =
h r
d2
dx dx + h
h
dr + r
9
6
i=1



1 1/2 2 2/3  1 2
h r
+
(3.13)
gs dz + gs (d M1 )2 .
9
Note that for M = 0 the bundle is untwisted and the transverse geometry is foliated by
S 2 S 2 S 1 S 1 . However in the presence of fractional branes M = 0 and a twist is

134

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

induced. The transverse space becomes foliated by T 1,1 S 1 . More accurately, this is a
T 1,1 space whose S 3 fibre (when T 1,1 is viewed as S 3 over S 2 ) is replaced by the lens space
S 3 /ZM . Indeed, under the redefinition M, d M1 becomes proportional to
the covariant displacement on the U (1)/ZM fibre of a T 1,1 with reversed orientation. This
is reminiscent of the situation in [5] (see also [21]).
We also have


r
2
2 2
F4 = N + 3gs M ln
r0


dr
2 + dA1 dz (d M1 ).
3gs M
(3.14)
r
Note that dF4 = 0 as it should. To prove that, use the fact that 2 dA1 = 0.
The solution contains M5-brane charge wrapping the S 2 of T 1,1 ,

(5,1)
Q
=
F4 .

(3.15)

S 3 S 1

This will reduce to NS5 wrapping S 2 as in (3.11).


Naively, the solution contains M5-branes wrapping both the untwisted S 1 and the
(1)
(2)
(5,2) is given by
U (1) fibre of the base 2 2 over the T 1,1 . Their charge Q

r
(5,2) =
Q
(3.16)
F4 = N + 3gs M 2 ln .
r0
(1)

(2)

2 2

(1)
(2)
(5,2) charge can be taken
Since there is no non-trivial 2 2 cycle in T 1,1 such a Q
to zero, even though there is non-vanishing F4 flux. However upon reduction to IIA, these
phantom M5-branes come to existence in the form of the D4-branes wrapping the S 1 , as
in (3.12)! Also, as noted below (3.13), in the absence of fractional branes the U (1) bundle
is untwisted and again the phantom M5-branes become ordinary M5-branes wrapping
S 1 S 1 . In the following, all the phantom charges will carry tildes.
This is the first instance in which we encounter a phenomenon whereby there are
fluxes unsupported by the geometry, which however materialize into physical branes upon
untwisting and/or Hopf-reduction. The U (1) (un)twisting is the key to this, and in a way
provides a discrete analogue of the large N duality between the fluxes and branes. We will
meet more examples of this in the sequel.
Let us compare to the situation in [12] where a string with non-zero winding number
in the background of a KaluzaKlein monopole was considered. The string can unwind
since the total space has 1 = 0, but the charge associated to the winding number of the
string is conserved. From this one concludes that there is a zero mode among the collective
coordinates of the KK monopole which couples to the charge associated with the winding
of the string. In our case 1 = ZM is non-zero, but there are still no non-trivial one-cycles
in homology (b1 = 0).
The lifting of the solution (3.7)(3.9) is similar. The resulting geometry is now
T 1,1 S 1 (for M = 0) and the charge Q(5,1) = 0. As before the S 3 fibre of the T 1,1

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

135

is replaced by the lens space S 3 /ZM . Again, there is a flux through the contractible
(1)
(2)
2 2 . Upon reduction this becomes the D4-brane flux.

4. The S-dual KT
We saw in the previous section that upon T -dualizing the KT solution we get
(1)
(2)
a configuration with untwisted 2 2 S 1 geometry for the level surfaces. It
would be interesting to have a situation where by T -dualizing a IIB solution we get level
surfaces with twisted geometry. The reason why this does not work with the KT solution
(3.1) is clear. From the T -duality transformations (A.22) we see that in order to get a
twisted U (1) fibration upon going from IIB to IIA, the original IIB solution would have to
(3)
have ANS
1 = 0. Upon T -dualizing this becomes the connection A1 of the U (1) fibration in
NS
IIA. Since H3 = dANS
2 dA1 dz, this means that the original IIB solution would have
to have H3 with non-zero flux through the Hopf fibre. We can indeed obtain a solution of
IIB with the aforementioned property by performing an S-duality transformation on the
KT solution:
 

 
a + b
H3
d c
H3
,

(4.1)
b a
F3
F3
c + d
with the rest of the fields inert and


a b
SL(2, Z).
:= + ie ,
c d

(4.2)

The KT solution (3.1) transforms to


c2
ac + bdgs2
:= g s ,
= 2
,
gs
c + d 2 gs2
dr
2 + cM3 ,
H3 = 3dgs M
r
dr
2 + aM3 .
F3 = 3bgs M
(4.3)
r
The metric and the five-form field strength are still given by (3.1)(3.3). Reducing to nine
dimensions T -dualizing and lifting to IIA we get two possible situations depending on the
choice of relative orientation of the two S 2 . Namely, a geometry with level surfaces given
either by T 1,1 with reversed orientation or by T 1,1 .



1/4 
2
1 2  (i) 2
1/8 1 1/2 2
2
1/2

1/2
2
ds10 = gs
h
dr + r
h r
d2
dx dx + h
9
6
i=1

3/4
7/8 1 1/2 2
h r
+ gs
(4.4)
(dz + cM1 )2 .
9
e = gs d 2 +

Note that dz + cM1 is proportional to the covariant displacement on the U (1)/ZcM fibre
either of a T 1,1 with reversed orientation or of a T 1,1 . We have

136

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

F2 = (a c)M2 ,
dr
2 + dA1 (dz cM1 ),
F3 = 3dgs M
r


r
dr
F4 = N + 3gs M 2 ln
2 (dz cM1 ).
2 2 + 3(b d)Mgs
r0
r
(4.5)
The brane content is different in each case.
First we discuss the T 1,1 with reversed orientation. The solution has zero units of
D6-branes wrapping S 3

(6)
Q = F2 = 0.
(4.6)
S2

There is however non-zero net Q(5) charge corresponding to NS5-branes wrapping S 2



(5)
Q = F3 = cM.
(4.7)
S3

(4) phantom charge corresponding to D4-branes wrapping the Hopf


Finally, there is a Q
fibre,

r
(4) =
Q
(4.8)
F4 = N + 3gs M 2 ln .
r0
2(1) 2(2)

Again, as in (3.16), note that this makes sense either upon Hopf-reducing or in the absence
of fractional branes in which case the U (1) fibre gets untwisted.
Turning to the T 1,1 case we find Q(6) = (a c)M and Q(5) = 0. What was said
before about (4.8) is true here as well.
4.1. The M-theory lift
Upon lifting to M-theory, yet another twisted U (1) fibre develops. The resulting
transverse geometry is foliated by a non-trivial S 1 /ZaM bundle over T 1,1 ZcM such that
reducing the fibre along any one of the two circles produces a T 1,1 space (or a T (1,1) , by
a reasoning that should be familiar by now). In this sense there is a similarity with the AMV
solution in the next section. 4 The transverse geometry in that case is an S 3 over S 3 bundle.
Viewing the base as a Hopf fibration over S 2 and reducing along the U (1) fibre produces an
S 3 (up to moding out by a discrete group) over S 2 bundle, which is actually a T 1,1 space.
Analytically, the eleven-dimensional metric reads



 
2
1 1/2 2 1/3 1/2
1 2  (i) 2
2
1/2
2
d2
dx dx + h
h
dr + r
h r
ds11 =
9
6
i=1

4 In spite of the similarity it is easy to see that the resulting geometry cannot be that of AMV simply because
the original ten-dimensional one is not that of the resolved conifold.

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

1 1/2 2
+
h r
9

2/3

 1 
2
gs (d aM1 ) (dz cM1 )

+ gs (dz + cM1 )2 ,

while the four-form is given by




4 = F4 F3 (d aM1 ) (dz cM1 ) ,
F

137

(4.9)

(4.10)

with F4 , F3 as in (4.5). Note that the fibre is in fact T 2 /ZaM ZcM .


The solution contains phantom M5-branes wrapping the T 2 fibre over the base 2(1)
(2)
(5,1) is given by
2 . Their charge Q

r
(5,1) =
Q
(4.11)
F4 = N + 3gs M 2 ln .
r0
(1)

(2)

2 2

Upon reduction to IIA these become the D4-branes wrapping the Hopf fibre of T 1,1 , as
in (4.8). Depending on the orientation there can be phantom M5-brane charge wrapping S 2

(5,2) =
Q
F4 ,
(4.12)
S 1 S 3

which reduces to NS5 wrapping S 2 as in (4.7).


If the opposite orientation is chosen, Q(5,2) = 0 which of course reduces to Q(5) = 0
(5,2) charge is now carried by
in IIA. The information about the previously nontrivial Q
ZaM ZcM .

5. Five-branes in the AMV solution


The AMV solution [10] involving M-theory on a manifold of G2 holonomy (locally
an S 3 R4 ) reduced to ten dimensions is given by warped 4D Minkowski times the
resolved conifold of [7,22]. In addition, there is a non-zero F2 flux through the S 2 . As
before, modulo the large N duality (S 3 flop), we think of this as the supergravity solution
corresponding to D6 wrapping the S 3 with a stabilizing flux through S 2 and thus producing
the geometry of the resolved conifold. For some related work see also [2329].
From (A.22) we immediately see that upon T -dualizing this solution to IIB, we get
(1)
(2)
untwisted 2 2 S 1 geometry: in order to get a twisted U (1) fibration upon
going from IIA to IIB, the original IIA solution would have to have A(2)
1 = 0. Upon T dualizing this becomes the connection A1 of the U (1) fibration in IIB. Since, as we see
3 = F (1) dA(2) e5 , this means that the original IIA solution would have
from (A.14), F
3
1
to have an F3 with non-zero flux through the Hopf fibre. This would signal the presence of
NS5-branes wrapping the S 2 , which are absent from the solution of [10].
This brings us to the question of whether such a configuration of NS5s and D6s is
known and if so, what would it lift to in M-theory. Note that the five-brane and sixbranes stabilize the S 2 and S 3 factors, respectively. The S 2 which is wrapped by the NS5

138

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

upon lifting becomes the base of a twisted U (1) fibration and thus is no longer able to
support an M5. This is in agreement with the fact that G2 holonomy manifolds do not have
calibrated two-dimensional submanifolds.
We should therefore look for an eleven-dimensional G4 that has a flux through the
(eleven-dimensional) Hopf fibre. Reducing this will give an F3 flux, which upon integration
over the S 3 will give NS5 charge. Since there are two S 3 s in the original geometry, an
obvious possibility is to consider a flux

(5.1)
G4 ,
S 3 S1

which upon reduction will give an NS5 wrapping S 2 . Although the ten-dimensional
geometry is foliated by T 1,1 , it cannot be that of the resolved conifold, since the flux
of the NS three-form will stabilize the S 3 part as well. At the moment we are not able to
write down explicitly such a solution containing simultaneously D6-branes wrapping S 3
and NS5-branes wrapping S 2 and having four common non-compact directions.
M-theory solutions with a flux as in (5.1) have recently appeared in the literature [26,
28]. It would clearly be interesting to obtain their IIA reductions and their IIB T -duals.
The hope is to provide an explicit realization of the mirror symmetry between D5-branes
on S 2 of the deformed conifold in type IIB and D6 on S 3 of the resolved in type IIA, as a
single T -duality along the Hopf fibre of the T 1,1 . 5 This task is complicated, however, by
the form of the deformed conifold metric, as will be discussed in the next section.

6. The KS solution
Although we have already excluded the possibility that the KS solution [8] is the T -dual
of the AMV solution since T -dualizing the latter gives untwisted transverse geometry
whereas the former has a twisted one we can still try to perform the analysis of the
previous sections to this case. Technically this is complicated by the fact that the metric
of the deformed conifold is written in terms of variables in which the U (1) isometry is
implicit [13]. Here we make partial progress by identifying a set of new variables in which
the U (1) isometry of the metric will be manifest, but we are yet unable to write down
explicitly the metric in terms of these new variables.
The KS solution reads
2
= h1/2 ( ) dx dx + h1/2 ( ) ds62 ,
ds10




F3 = M e5 g 3 g 4 + d F ( ) g 1 g 3 + g 2 g 4 ,


B2 = gs M f ( )g 1 g 2 + k( )g 3 g 4 ,

F5 = gs M 2 l( )g 1 g 2 g 3 g 4 e5 ,

(6.1)

5 Note that as we have already remarked, the existence of NS5-branes is required in addition to the D-branes
for this duality to work.

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

where
ds62


 
 3 2  4 2 
1 4/3
1  2  5 2 
2
d + e
g
+ cosh
= / K( )
+ g
2
3K( )
2
 

 1 2  2 2 
g
+ g
+ sinh2
2

139

(6.2)

and is a radial coordinate. The explicit form of the functions F, f, k, l, K will not be
important in the following. The one-forms g i are defined as
e1 e3
e2 e4
e1 + e3
g2 = ,
g3 = ,
,
2
2
2
2 + e4
e
g4 = ,
(6.3)
g 5 = e5 ,
2
so that the level surfaces of (6.2) are of the general form (2.3).
As we noted in Section 4, for this solution to T -dualize to a twisted transverse geometry,
B2 would have to have a non-zero component along the direction corresponding to the U (1)
isometry.
Trying to Hopf-reduce runs into trouble because the solution is written in terms of
coordinates in which the U (1) isometry is not simply a shift in . In fact from (2.2),
(2.3) we see that the isometry reads





cos c sin c
sin 2 d2
sin 2 d2

.
+ c,
(6.4)
sin c cos c
d2
d2
g1 =

We would like to find a coordinate transformation for 2 , 2 which has the effect of the
second line in (6.4). Clearly this would have to be a special case of the SU(2) group of
isometries of the sphere
az + b
,
+ a
bz

2
, |a|2 + |b|2 = 1.
(6.5)
2
Indeed for a = cos //2, b = sin //2, (6.5) implies (6.4) to order O(/), provided we identify
z

z := ei2 tan

sin 2
.
sin 2
Explicitly the coordinate transformation reads, to O(/) order,
c=/

(6.6)

sin 2
,
sin 2
/ sin 2 cot 2 ,
+/

2 2 + / cos 2 .

(6.7)
e5

It is easy to check that (6.7) leaves


invariant. We have therefore succeeded in
constructing the killing vector k corresponding to the U (1) isometry,
k=

sin 2
sin 2 cot 2
+ cos 2
.
sin 2
2
2

(6.8)

It remains to find a set of new variables  , 2 , 2 in which k locally takes the form
k = /  . Rewriting (6.2) in terms of these new variables would result in a metric

140

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

whose components do not depend on  and Hopf-reduction would proceed as before.


The variables we are looking for would therefore have to satisfy the equation



2

2 2




2 2 2

2 2

  
2 2
One can easily verify that


k2
0

k 2 = 0 .

k
1

sin 2
1

,
L (f1 + f2 ) =
2
sin 2
where


,
L := k k

1

L (f1 f2 ) = 1,
2

(6.9)


Lh(sin 2 sin 2 ) = 0,



sin 2 sin 2
f1,2 (2 , 2 ) := arctan
,
cos 2 cos 2

(6.10)

(6.11)

and h is an arbitrary function of sin 2 sin 2 .


Taking (6.10) into account, we can write down an explicit special solution to (6.9): 6
 = + f1 (2 , 2 ),

1
 = + f1 (2 , 2 ) + f2 (2 , 2 ) ,
2
 = h(sin 2 sin 2 ).

(6.12)

The Jacobian of the above transformation reads


J = sin 2 h (x)|x=sin 2 sin 2 .

(6.13)

Finally one can derive the relation of the differentials of the old coordinates in terms of the
new,

cos 2
cot 2 sin 2 cot 2 sin 2
2

2
2
h (cos 2 + cos 2 sin 2 ) sin 2


d 

d2

cos 2 sin 2
d 
d2 =
cos 2
cos 2
2

 (cos2 + cos2 sin2 )


h
2
2
2
d 

2
sin 2
sin 2
cos 2 cot 2
1

sin 2
sin 2
h (cos2 2 + cos2 2 sin2 2 )
(6.14)
Using the above relation, one can examine whether or not B2 in (6.1) has a non-trivial
component along the U (1) fibre. The answer is that it does and therefore, as already
explained, the T -dual version of KS has twisted transverse geometry.
6 This is not the most general form. One can add arbitrary functions of sin sin to the rhs. Also one could
2
2
take arbitrary linear combinations of 2 , 2 in the second and third lines.

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

141

Note added
After this paper was posted on the hep-th archive, we were informed by I. Klebanov,
A. Tseytlin and L. Pando Zayas of their related independent work which overlaps with
Section 3 of this paper. See the talk by I. Klebanov at the Avatars of M-theory conference
(UCSB) http://online.itp.ucsb.edu/online/mtheory_c01/klebanov/.

Acknowledgements
It is a pleasure to thank C. Bachas, M. Cvetic, M. Douglas and G. Ferretti for discussions.
The work of R.M. is supported in part by EU contract HPRN-CT-2000-00122 and by
INTAS contract 55-1-590; D.T. is supported in part by EU contract HPRN-CT-2000-00122
and by the SNSRC. D.T. would like to thank Ecole Normale Suprieure and CERN for
hospitality.

Appendix A. Review of KK on a circle


This is a collection of useful formulae. There is nothing here that cannot be found in the
literature. However we include this appendix as a self-contained set of conventions.
A.1. Reduction of 11D supergravity to 10D type IIA
The starting point is the eleven-dimensional Lagrangian
2 + C.S.,
1 F
e1 L = R
48 4
where
4 = d A
3 .
F

(A.1)

(A.2)

Note that all eleven- (ten-)dimensional field strengths and potentials are denoted with
(without) a hat. Assuming the geometry has a U (1) isometry, we can put the metric in
the form:
1

2
2
= e 6 ds10
+ e 3 (A1 + dz)2 ,
ds11

(A.3)

3 and the metric in ds 2 , are all assumed independent of the U (1)


where , A1 , A
10
coordinate z. We also reduce the 3-form potential
3 = A3 + A2 dz.
A

(A.4)

The eleven-dimensional Lagrangian becomes


1
1 3
1
1 1
e1 L = R ()2 e 2 F22 e F32 e 2 F42 + C.S.,
2
4
12
48

(A.5)

142

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

where
F4 = dA3 + dA2 A1 ,

F3 = dA2 ,

F2 = dA1 ,

(A.6)

so that
4 = F4 + F3 (dz + A1 ).
F

(A.7)

A.2. Reduction of 10D type IIA to 9D N = 2


We start with the Lagrangian
1 3 2
1 2
1 1 2
1 ( )
2 e 2 F
e1 L = R
e F3 e 2 F
2
4 + C.S.,
2
4
12
48
where
4 = d A
3 + d A
2 A
1 ,
F

3 = d A
2 ,
F

2 = d A
1 .
F

(A.8)

(A.9)

We assume that there is a U (1) isometry so that the ten-dimensional metric can be cast in
the form

1
2
7  (3)

2
= e 2 7 ds92 + e 2 A1 + dz ,
ds10
(A.10)
2
where , A(3)
1 and the nine-dimensional metric ds9 do not depend on the U (1) coordinate z.
We reduce the potentials as follows

1 = A(1) + A0 dz,
A
1

2 = A(1) + A(2) dz,


A
2
1

3 = A3 + A(2) dz. (A.11)


A
2

The Lagrangian reduces to

7
1
1
1 3
1 3  (2) 2
7
e1 L = R ()2 ()2 e 2 2 (F1 )2 e
F2
2
2
2
4
1 3 + 1  (1)2 1 4  (3) 2
1 + 1  (1) 2
7
F3
e 2 2 7 F2
e 7 F2
e
4
4
12
1 1 5  (2) 2
1 1 + 3
e 2 2 7 F3
e 2 2 7 (F4 )2 + C.S.,
12
48
where

(A.12)

= ,
F1 = dA0,
(1)

(1)

(3)

F2 = dA1 + dA0 A1 ,
F2(2) = dA(2)
1 ,
F2(3) = dA(3)
1 ,
(2)
(3)
F3(1) = dA(1)
2 + dA1 A1 ,
(1)
(2)
(1)
F3(2) = dA(2)
2 + A1 dA1 + A0 dA2 ,
(1)
(2)
(3)
(2)
(1)
(3)
(1)
(3)
F4 = dA3 + dA(1)
2 A1 + dA2 A1 dA1 A1 A1 A0 dA2 A1 ,

(A.13)

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

143

so that


4 = F4 + F (2) dz + A(3) ,
F
3
1


3 = F (1) + F (2) dz + A(3) ,
F
3
2
1


2 = F (1) + F1 dz + A(3) .
F
2
1

(A.14)

A.3. Reduction of 10D type IIB to 9D N = 2


We start with the Lagrangian
1
1  2
1 ( )
2 e2 ( )
e1 L = R
2 e F
3
2
2
12
1
1  2
 2
F5 + C.S.,
e H
3
12
4 5!

(A.15)

where
1 NS R 1 R NS
R
5 = d A
F
4 d A2 A2 + d A2 A2 = F5 ,
2
2
R
NS




NS .
F3 = d A2 d A2 ,
H3 = d A
2

(A.16)

We assume that there is a U (1) isometry so that the ten-dimensional metric can be cast in
the form
2
=e
ds10

2 7

ds92 + e

7
2

(A1 + dz)2 ,

(A.17)

where , A1 and the nine-dimensional metric ds92 do not depend on the U (1) coordinate z.
We reduce the potentials as follows
R
R
R
A
2 = A2 + A1 dz,
R
R
R
A
4 = A4 + A3 dz.

NS = ANS + ANS dz,


A
2
2
1
(A.18)

The Lagrangian reduces to


1
1
1
e1 L = R ()2 ()2 e2 ()2
2
2
2
1 3  NS 2 1 3  R 2 1 4
7
7
F2
F2 e 7 (F2 )2
e
e
4
4
4
1 + 1  NS 2
1 + 1  R 2
1 2  2
7
7
e
F3
F3 e 7 F4R + C.S.,
e
12
12
48
(A.19)
where

= ,

= ,

NS
F2R = dAR
F2NS = dANS
F2
1 + dA1 ,
1 ,
NS
NS
NS
F3 = dA2 + A1 dA1 ,
 NS

R
NS
F3R = dAR
2 + dA1 A1 dA2 + dA1 A1 ,

= dA1 ,

144

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

F4R = dAR
3 +


1
R
NS
R
R
NS
R
NS
dANS
2 A1 A2 dA1 + dA2 A1 + A2 dA1 ,
2
(A.20)

so that
n = Fn + Fn1 (dz + A1 ).
F

(A.21)

A.4. 9D T-duality
The two nine-dimensional Lagrangians described above, are in fact related to each other
by the following local field transformations:

3
7
B ,
A B
4
4

3
7
A
B B ,
4
4

R
NS
A(1)
1 A1 + A1 ,

A(2)
1 A1 ,

(1)

A0 ,
NS
A(3)
1 A1 ,

(2)

NS
A2 ANS
2 A1 A1 ,

R
A2 AR
2 + A1 A1 ,

1 NS
1 R
R
NS
NS
R
A3 AR
3 + A1 A2 A1 A2 A1 A1 A1 .
2
2

(A.22)

Or, in terms of field strenghs:


F1 d,
(1)

F2 F2 ,

(2)

(1)

F3 F3R ,

F2 F2R ,

(3)

F2 F2NS ,

(2)

F3 F3NS ,
F4 F4R .

(A.23)

A.5. String vs. Einstein frame


In ten dimensions the Einstein metric is related to the string metric through
1

str
.
g = e 2 g

(A.24)

The reduction of the ten-dimensional metric in the string frame reads


1

str
ds10
= e2

( 1 )
7

ds92 + e 2 (+

7)

(dz + A1 )2 .

(A.25)

In the string frame the ten-dimensional type IIB Lagrangian becomes


 2  1

3 ( )
2 1 H
2
e1 L = e2 R
+ 4( )
12
2
1  2
1  2

F3
F5 + C.S.
12
4 5!

(A.26)

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

145

References
[1] B.E.W. Nilsson, C.N. Pope, Hopf fibration of eleven-dimensional supergravity, Class. Quant.
Grav. 1 (1984) 499.
[2] M.J. Duff, P.S. Howe, T. Inami, K.S. Stelle, Superstrings in D = 10 from supermembranes in
D = 11, Phys. Lett. B 191 (1987) 70.
[3] M.J. Duff, H. Lu, C.N. Pope, Supersymmetry without supersymmetry, Phys. Lett. B 409 (1997)
136, hep-th/9704186.
[4] M.J. Duff, H. Lu, C.N. Pope, AdS(5) S(5) untwisted, Nucl. Phys. B 532 (1998) 181, hepth/9803061.
[5] M.J. Duff, H. Lu, C.N. Pope, AdS(3) S 3 (un)twisted and squashed, and an O(2, 2, Z)
multiplet of dyonic strings, Nucl. Phys. B 544 (1999) 145, hep-th/9807173.
[6] I.R. Klebanov, A.A. Tseytlin, Gravity duals of supersymmetric SU(N) SU(N + M) gauge
theories, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
[7] P. Candelas, X.C. de la Ossa, Comments on conifolds, Nucl. Phys. B 342 (1990) 246.
[8] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and
chiSB-resolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[9] B.S. Acharya, On realising N = 1 super-YangMills in M-theory, hep-th/0011089.
[10] M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large n duality, hep-th/0011256.
[11] C. Vafa, Superstrings and topological strings at large N , hep-th/0008142.
[12] R. Gregory, J.A. Harvey, G. Moore, Unwinding strings and T -duality of KaluzaKlein and
H -monopoles, Adv. Theor. Math. Phys. 1 (1997) 283, hep-th/9708086.
[13] R. Minasian, D. Tsimpis, On the geometry of non-trivially embedded branes, Nucl. Phys. B 572
(2000) 499, hep-th/9911042.
[14] S.S. Gubser, I.R. Klebanov, Baryons and domain walls in an N = 1 superconformal gauge
theory, Phys. Rev. D 58 (1998) 125025, hep-th/9808075.
[15] K. Dasgupta, S. Mukhi, Brane constructions, fractional branes and anti-de Sitter domain walls,
JHEP 9907 (1999) 008, hep-th/9904131.
[16] I.R. Klebanov, N.A. Nekrasov, Gravity duals of fractional branes and logarithmic RG flow,
Nucl. Phys. B 574 (2000) 263, hep-th/9911096.
[17] M.J. Duff, B.E.W. Nilsson, C.N. Pope, KaluzaKlein supergravity, Phys. Rep. 130 (1986) 1.
[18] E. Witten, Solutions of four-dimensional field theories via M-theory, Nucl. Phys. B 500 (1997)
3, hep-th/9703166.
[19] E. Witten, Branes and the dynamics of QCD, Nucl. Phys. B 507 (1997) 658, hep-th/9706109.
[20] K. Dasgupta, S. Mukhi, Brane constructions, conifolds and M-theory, Nucl. Phys. B 551 (1999)
204, hep-th/9811139.
[21] M. Cvetic, H. Lu, C.N. Pope, Decoupling limit, lens spaces and Taub-NUT: D = 4 black hole
microscopics from D = 5 black holes, Nucl. Phys. B 549 (1999) 194, hep-th/9811107.
[22] L.A. Pando Zayas, A.A. Tseytlin, 3-branes on resolved conifold, JHEP 0011 (2000) 028, hepth/0010088.
[23] J.M. Maldacena, C. Nunez, Towards the large N limit of pure N = 1 super-YangMills, Phys.
Rev. Lett. 86 (2001) 588, hep-th/0008001.
[24] J.D. Edelstein, C. Nunez, D6-branes and M-theory geometrical transitions from gauged
supergravity, JHEP 0104 (2001) 028, hep-th/0103167.
[25] K. Dasgupta, K. Oh, R. Tatar, Geometric transition, large N dualities and MQCD dynamics,
hep-th/0105066.
[26] M. Cvetic, H. Lu, C.N. Pope, Massless 3-branes in M-theory, hep-th/0105096.

146

R. Minasian, D. Tsimpis / Nuclear Physics B 613 (2001) 127146

[27] M. Aganagic, C. Vafa, Mirror symmetry and a G(2) flop, hep-th/0105225.


[28] M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, M3-branes, G(2) manifolds and pseudosupersymmetry, hep-th/0106026.
[29] A. Brandhuber, J. Gomis, S.S. Gubser, S. Gukov, Gauge theory at large N and new G(2)
holonomy metrics, hep-th/0106034.

Nuclear Physics B 613 (2001) 147166


www.elsevier.com/locate/npe

GUT breaking on the brane


Yasunori Nomura a,1 , David Smith a , Neal Weiner b
a Department of Physics, and Theoretical Physics Group, Lawrence Berkeley National Laboratory,

University of California, Berkeley, CA 94720, USA


b Department of Physics, University of Washington, Seattle, WA 98195, USA

Received 18 April 2001; accepted 2 August 2001

Abstract
We present a five-dimensional supersymmetric SU(5) theory in which the gauge symmetry is
broken maximally (i.e., at the 5D Planck scale M ) on the same 4D brane where chiral matter is
localized. Masses of the lightest KaluzaKlein modes for the colored Higgs and X and Y gauge fields
are determined by the compactification scale of the fifth dimension, MC 1015 GeV, rather than
by M . These fields wave functions are repelled from the GUT-breaking brane, so that proton decay
rates are suppressed below experimental limits. Above the compactification scale, the differences
between the standard model gauge couplings evolve logarithmically, so that ordinary logarithmic
gauge coupling unification is preserved. The maximal breaking of the grand unified group can also
lead to other effects, such as O(1) deviations from SU(5) predictions of Yukawa couplings, even in
models utilizing the FroggattNielsen mechanism. 2001 Published by Elsevier Science B.V.
PACS: 12.10.D; 04.50

1. Introduction
The standard model of particle physics, described by the gauge group SU(3) SU(2)
U(1) is one of the most successful physical theories ever. Nonetheless, the standard
model is theoretically unsatisfying for a number of reasons. For instance, the instability of
the weak scale against radiative corrections has motivated the study of numerous theories,
including technicolor, supersymmetry and theories with additional dimensions.
While the gauge hierarchy may be the most compelling motivation for new physics, there
are additional reasons to consider theories with further structure. It was long ago realized
that the gauge group of the standard model could be embedded into a simple group [1].

This work was supported in part by the US Department of Energy under Contracts DE-AC03-76SF00098, in
part by the National Science Foundation under grant PHY-95-14797.
1 Research fellow, Miller Institute for Basic Research in Science.
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 8 8 - 1

148

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

The appeal of this idea was substantiated by measurements of electroweak observables


which suggested that the values of the standard model gauge couplings unify at a scale
MGUT 1015 GeV [2].
As data became increasingly precise, it became apparent that the simplest grand unified
theory (GUT), minimal SU(5), predicted a value for sin2 W that was incompatible with
observation. However, the combination of supersymmetry and grand unification [3] has
met with great success in predicting sin2 W [4]. For these supersymmetric theories, the
GUT symmetry must be broken at a somewhat higher scale, MGUT 2 1016 GeV.
One smoking gun signal of grand unification is proton decay, which arises from
dimension-six operators generated by exchange of the X and Y gauge bosons, and, in
supersymmetric theories, from dimension-five operators generated by colored Higgsino
exchange [5]. In supersymmetric GUTs, the dimension-six operators typically do not
lead to observable proton decay, but the dimension-five operators do, and place strong
limits on these theories [6]. Recently Ref. [7] studied the dimension-five proton decay and
found that even large masses for the first-two generation sparticles cannot save minimal
supersymmetric SU(5). They found that successful gauge coupling unification requires
that the Higgs triplet mass fall in the range 1.1 1014 GeV  MHC  9.3 1015 GeV.
However, even with decoupling of the superpartners, the limit from proton decay was
shown to require MHC  9.4 1016 GeV.
While additional field content may generate threshold effects to rectify this, in this
paper we propose an alternative possibility. We will embed a unified gauge theory into
five dimensions in which gauge and Higgs fields propagate in the additional dimension.
There has been much work on gauge fields propagating in extra dimensions, with gauge
symmetries broken by Wilson lines or boundary conditions [810]. We will not employ
orbifold breaking of the grand unified group, but rather, we will show that if the GUT is
broken on a brane at a high scale (i.e., M , the Planck scale of the five-dimensional theory),
the wave functions of the gauge fields and Higgs triplets can develop an approximate
node on the brane, suppressing proton decay operators arising from their exchange. We
will show that the running above the compactification scale is not only consistent with,
but is in fact equivalent to ordinary four-dimensional logarithmic unification with the
compactification scale identified roughly as the mass of the Higgs triplet.
In doing so, we reformulate the question of the MGUT /MPl hierarchy because the
breaking occurs at the fundamental scale of the theory (which, in general, is still larger
than 1016 GeV). In its place we require a somewhat large hierarchy (O(100)) between the
Planck and compactification scales. The traditional GUT scale, 2 1016 GeV, is a derived
scale in this framework: there is no new physics there. Because of the O(1) breaking of
the grand unified group on the brane, we will also easily understand deviations from SU(5)
predicted relationships among Yukawa couplings.
The layout of the paper is as follows. In Section 2, we describe the setup, in which the
grand unified group is broken maximally on the brane. We then calculate the spectrum
of the KaluzaKlein (KK) towers of both the X and Y gauge fields as well as the Higgs
triplets. In Section 3 we study the evolution of gauge couplings in the presence of shifted
KK towers, and demonstrate that while the gauge couplings g1 , g2 and g3 exhibit power

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

149

law running, the majority of it is SU(5) universal. Nonetheless, we can extract a nonuniversal piece, which is proportional to the MSSM running, and logarithmic, but at a rate
slowed compared to the MSSM, such that ordinary, logarithmic unification can occur at
the higher-dimensional Planck scale. In Section 4 we will see how proton decay operators
are naturally suppressed by suppressing the KK wave functions on the brane. In Section 5
we will discuss deviations from SU(5) Yukawa relationships can arise from the maximal
breaking of the GUT on the brane. In Section 6 we discuss extensions to larger gauge
groups. Finally, we conclude in Section 7.

2. Framework
We consider a 5D supersymmetric SU(5) theory with the fifth dimension compactified
on an S1 /Z2 orbifold (in Section 6 we will consider the possibility of larger gauge groups).
The bulk exhibits N = 1 supersymmetry in five dimensions, which translates into an N = 2
supersymmetry in four dimensions. The fields that propagate in the bulk are contained
in a gauge supermultiplet and two Higgs hypermultiplets transforming as 5 and 5 under
SU(5). Under the 4D N = 1 supersymmetry preserved after the orbifold compactification,
the gauge supermultiplet decomposes into a vector superfield V and an associated chiral
adjoint . The two hypermultiplets yield four chiral multiplets H5 , H5c , H5 and H c . (H5
5

and H c transform as 5 under the SU(5), while H5c and H5 as 5.) Under the orbifold Z2 ,
5
which flips the sign of the fifth coordinate (y y), the various superfields transform as
H5,5 H5,5 ,
V V,

c
c
H5,5
H5,5
,

(1)

which leaves zero modes only for the MSSM fields and their GUT counterparts.
The S1 /Z2 orbifold has fixed points at y = 0 and y = R. We take chiral matter to be
localized to the y = R fixed point, along with an adjoint chiral superfield . The SU(5)
symmetry is broken by the vacuum expectation value (vev) of , which we assume to have
the form

2/5 0
0
0
0
0 2/5 0
0
0


 ij =
(2)
0
0
2/5
0
0

.
0
0
0 3/5
0 2
0

3/5

We imagine that this breaking occurs at or near the five-dimensional Planck scale M ,
which we assume to be much larger than the conventional GUT scale of 2 1016 GeV. We
will refer to the y = R fixed point as the GUT-breaking brane.
Bulk fields even under the orbifold Z2 can couple directly to the fields on the GUTbreaking brane. For instance, the Higgs fields can couple to the chiral matter fields T10
and F5 as (u T10 T10 H5 + d T10 F5 H5 )(y R) in the superpotential. We also take there
to be a bare Higgs mass term, H5 H5 (y R), as well as Higgs coupling to the GUT

150

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

breaking, (/M )H5 H5 (y R), in the superpotential. 2 While it would be interesting


to study standard solutions to the doublettriplet splitting problem in this framework,
we will not endeavor to do so here. Rather, we will simply tune these two contributions
against one another so that the SU(2) doublets Hu ( H5 ) and Hd ( H5 ) are nearly
massless, while the colored triplets HC ( H5 ) and HC ( H5 ) end up with a large mass
term HC HC (y R). Note that is a dimensionless parameter since HC and HC are
bulk fields and have dimension 3/2. Here, arises as a sum of terms (with higher order
terms potentially relevant in the strongly coupled theory). In addition to the bare mass
term, the H5 H5 term contributes. The natural size of this operator is restricted only by
perturbativity, with an upper bound 6 2 suggested by naive dimensional analysis in
higher dimensions [11].
2.1. Spectrum: gauge fields
Once the gauge symmetry is broken, the X and Y bosons acquire brane-localized masses
through the interaction

L (y R)g52  2 A,a Aa ,

(3)

where a indexes the broken generators of the group and g52 is the 5D gauge coupling, with
mass dimension 1. If we were to decompose each X and Y bulk vector field
in a naive KK
basis, we would estimate that the zero modes have mass MV = g5  / 2R = g4  .
Such an estimate incorrectly assumes that the gauge boson wave functions do not change
appreciably in the presence of the large localized mass term. To find the correct spectrum
we must solve the differential equation
y2 A + (y R)g52  2 A = m2 A ,

(4)

which leads to KK masses MnG given by [11,12]



 g 2  2
,
MnG tan MnG R = 5
2

(5)

where n = 0, 1, 2, . . . . For g52  2 R  1 this equation gives a spectrum whose low-lying


levels are approximately MnG = (n + 1/2)MC (with MC 1/R), whereas the usual KK
spectrum is mn = nMC . We see that the KK tower has been shifted up one half unit.
This is easy to understand intuitively. The gauge field picks up a mass of g4   1/R
if it does not avoid the brane. If it avoids the brane entirely, it does not see the mass
term, and picks up a smaller mass of 1/(2R) = MC /2 from a nontrivial profile of the wave
function in the extra dimension. The localized mass term acts merely to dynamically assign
the boundary condition that the wave functions should vanish at the GUT-breaking brane.
The boundary condition is not absolute, of course, so that the spectrum is not precisely
2 In this paper we will work within the context of effective field theory and will not concern ourselves with
an explicit string realization. In string theory the boundary terms such as H5 H5 can arise only from higherdimensional operators SH5 H5 , where S is a standard-model singlet brane field which has a vev of the order of
the 5D Planck scale.

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

151

Fig. 1. Mass spectrum for the lowest KK modes of the gauge fields (a), and Higgs fields (b) in our
model. As explained in the text, the transition to degeneracy between the X, Y and 3-2-1 towers
occurs much more gradually than shown in (a). In (b), the limit of very large is taken, so that the
slight non-degeneracy between the colored hypermultiplet pairs at each level is not resolved. In both
(a) and (b), the triplet of numbers below each tower corresponds to the beta function contribution
(b1 , b2 , b3 ) that comes from each level in that tower (for the colored Higgs, the contributions from
the nearly-degenerate hypermultiplet pairs are combined).

spaced according to (n + 1/2)MC . The value of the wave function at the GUT-breaking
brane goes as cos(MnG R) = (2n + 1)MC /(g52  2 ), and there are corrections to the mass
which go like (2n + 1)MC2 /(g52  2 ). Modes with masses near or above the scale
g52  2 become essentially degenerate with the ordinary KK tower, mn = nMC .
We have not broken supersymmetry, and thus the gauginos corresponding to the broken
SU(5) generators have an identical spectrum. The KK towers for the gauge fields are
illustrated in Fig. 1(a). For each unbroken SU(3) SU(2) U(1) (3-2-1) generator there
is a massless vector multiplet and massive vector multiplets at MC , 2MC , 3MC , and
so on. For each broken (X, Y ) generator there is a tower of massive vector multiplets
whose low levels are offset from those of the 3-2-1 tower by MC /2, but whose higher

152

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

levels relax into alignment with the 3-2-1 tower. Although in Fig. 1(a) this transition to
degeneracy has already been achieved at 5MC , for the parameters of interest in our model,
the relaxation will actually occur much more gradually, over the span of g52  2 /MC
O(10100) KK levels. Note that for each X, Y massive vector multiplet but one there
is a corresponding 3-2-1 massive vector multiplet. The remaining one is paired with the
massless 3-2-1 vector multiplet, indicating a larger total number of states in the X, Y
tower relative to the 3-2-1 tower. These excess states are the eaten Goldstone degrees of
freedom from the multiplet. This means that above the scale g52  2 the KK towers of
gauge fields contributing to the renormalization group (RG) running are completely SU(5)
symmetric except that one chiral adjoint for each 3-2-1 vector multiplet is missing. These
missing degrees of freedom are provided by the physical fields which remain after the
Goldstone components of are eaten by the X, Y gauge multiplets. Therefore, above the
scale g52  2 and m (the mass of physical fields), the spectrum contributing to the
running is completely SU(5) symmetric.
2.2. Spectrum: Higgs fields
The spectrum analysis for the Higgs triplet fields HC and HC is somewhat different,
because the brane coupling is dimensionless, and hence the naive mass that the zero
modes would pick up by not avoiding the brane is just MC which is comparable to MC
for order one . For our purposes, we will take to be a large parameter, roughly O(10),
so that the Higgs wave functions will in fact be strongly repelled from the brane. This size
for is quite consistent with the general features of the theory, which must be somewhat
strongly coupled in order to achieve order one top Yukawa and gauge couplings in spite of
the large radius.
The equations that determine the colored Higgsino spectrum are
c (y R)H
= 0,
i HC y H
C
C

c


i H y H (y R)HC = 0,

C
C
c
C = 0,
i HC + y H

c
= 0.
i HC + y H
C

(6)
(7)
(8)
(9)

We can analyze these equations by writing the solutions as


(c)
(c)
HC(c) =
C,n (x)gC,n
(y),

(10)

HC (c) =

(c)
(c)
C,n
(x)gC,n
(y),

(11)

c
c
where gC,n and gC,n
are even under y y, while gC,n and gC,n
are odd. If we integrate
the first two of these equations in a region (R /, R + /) where / 0, we obtain two
constraints,
c
(x) =
C,n

gC,n
(R)
c
2gC,n
(R /)

C,n
(x),

(12)

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166


c
C,n
(x) =

gC,n (R)
c (R /) C,n (x).
2gC,n

153

(13)

Here, two constraints must be satisfied mode by mode. Let us define


(I,J ),n =

gI,n (R)
c (R /) ,
2gJ,n

(14)

and identify both i C,n (x) = MnH (x) and i


where I, J take C, C,
C,n
H
C,n
(x) = Mn C,n (x). This allows us to rewrite Eqs. (6)(9) as (neglecting singular terms)
c
MnH gC,n (y) (C,C),n
y gC,n
(y) = 0,

(15)

c
MnH gC,n
y gC,n
(y) (C,C),n
(y) = 0,

(16)

c
MnH (C,C),n
gC,n (y) + y gC,n (y) = 0,

(17)

c
MnH (C,C),n
gC,n
(y) = 0,

(y) + y gC,n

(18)

which give solutions


 H 
gC,C,n
(y) = Nn cos Mn y ,

 H 
c
gC,
(y) = Nn sin Mn y ,
C,n

(19)

with MnH defined by



 2
tan2 MnH R = .
4
Thus, the KK masses for the colored Higgs are given by


1
arctan(/2)
MnH =
n+
,
R

(20)

(21)

where n runs from negative infinity to positive infinity (the physical masses are given by
the absolute value of MnH ). For large , arctan(/2)  /2, so MnH  (n + 1/2)MC , and
H
). Note that, in contrast to the
the masses fall into nearly degenerate pairs, (MnH , Mn1
case of the gauge fields, even for high values of n, the triplet Higgsinos do not become
degenerate with the doublet Higgsinos.
Supersymmetry fixes the masses and couplings of the Higgs scalars to be the same
as for the Higgsinos. The KK towers for the Higgs fields are shown in Fig. 1(b). The
doublet spectrum consists of a pair of massless N = 1 chiral multiplets, and pairs of exactly
degenerate hypermultiplets at MC , 2MC , 3MC , and so on. For large , the triplet spectrum
consists of nearly degenerate pairs of hypermultiplets at MC /2, 3MC /2, 5MC /2, and so
on. In Fig. 1(b), the non-degeneracy between these pairs is not resolved.

3. Differential running
A crucial question in our model is how the gauge couplings evolve above the
compactification scale. To address this question, it is useful to focus on the differential
running, i.e., the non-uniform evolution of the gauge couplings. Above MC /2 a whole

154

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

tower of modes contributes to the evolution of g1 , g2 and g3 . However, we should


emphasize that this scenario does not employ power law unification. In contrast with
Ref. [13], the overwhelming power-law contribution to the running above MC /2 is SU(5)
universal, so that only a small, logarithmically evolving piece contributes to the unification.
Similarly, while this behavior looks from the low energy observer as a logarithmic
unification, it is distinct from Ref. [14], as there is a genuine power-law behavior above
the compactification scale.
Because the dominant component of the running is SU(5) universal, it is useful to focus
only on the quantities which distinguish the couplings and lead to non-uniform evolution
above MC /2. For instance, the matter fields fall into complete SU(5) multiplets so their
effects only change the coupling value at unification, but do not influence whether and at
what scale the couplings unify. In contrast, the presence of the Higgs doublets leads to nonuniform running. Since the KK towers of the Higgs triplets and X and Y gauge bosons have
been merely shifted, we will see shortly that above MC /2, the differential running arises
from a sum of a large number of threshold effects, retaining logarithmic unification as a
feature of the theory. 3
Let us define the scale-dependent quantities
i () i1 () 11 (),

(22)

with 1 vanishing trivially. The evolution of i () above MC /2 (but below m ) takes the
form
i () = i (MC /2)


1  H
Ri () + RiG () ,
2

(23)

where RiH () represents contributions arising from Higgs loops and RiG () represents
contributions arising from gauge loops. Contributions from matter loops vanish because
they contribute universally to the running. Using the known spectra for the various KK
towers and their beta function contributions (shown in Fig. 1), we can calculate these
quantities. For the Higgs contributions we find

2
4
H
R2 () = log
log
log

+
,
5
MC /2
5
nMC
5 H
|MnH |
0<nMC <
|Mn |<
(24)

3
6
log
log
+
R3H () = log

.
5
MC /2
5
nMC
5 H
|MnH |
0<nMC <
|Mn |<
(25)
Here the MnH are the colored Higgs masses given in Eq. (21). Recall that there are
actually two slightly non-degenerate |MnH | near each explicitly shown colored Higgs level
in Fig. 1(b).
3 The threshold correction from KK towers are also discussed in the context of power-law unification [15] and
orbifold breaking of the unified gauge symmetry [16].

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

For the gauge contributions we have

G
log
log
R2 () = 6 log
+4
,
4
MC /2
nMC
MnG
0<nMC <
MnG <

G
log
log
R3 () = 9 log
+6
,
6
MC /2
nMC
MnG
G
0<nMC <

155

(26)

(27)

Mn <

where MnG are the X, Y vector multiplet masses, given by the solutions to Eq. (5).
3.1. Higgs contributions
The running of i () is given by


di
1 dRiH () dRiG ()
.
=
+
d
2
d
d

(28)

Let us consider the Higgs and gauge contributions to this running separately, starting with
the Higgs loops first. We have

dR2H
1
=
(# of doublets with mass ) (# of triplets with mass) ,
d
5

(29)

and

dR3H
3
=
(# of doublets with mass < ) (# of triplets with mass ) . (30)
d
10
Here we mean the number of doublet and triplet chiral superfields: there are two massless
doublets, four triplets with mass  MC /2, four doublets with mass MC , and so on.
Both of the above expressions have the expected 1/ characteristic of logarithmic
running, but have coefficients which average out to zero. Below MC /2 only the massless
Hu and Hd doublets contribute, so R2H (R3H ) runs in the positive (negative) direction. At
MC /2 we encounter four triplet chiral superfields and R2H (R3H ) now runs in the negative
(positive) direction. The directions reverse again when we gain four doublets at MC , and
again at 3MC /2 when we gain another four triplets. These threshold effects continue, but
become increasingly negligible. Taking the triplet masses to be (n + 1/2 )MC , the
threshold effects from the states between nMc and (n + 1)MC are proportional to

log
+ log
log
nMC
(n + 1)MC
(n + 1/2 + )MC

(n + 1/2 + )(n + 1/2 )


log
(31)
= log
(n + 1/2 )MC
n(n + 1)
which vanishes as (1/4 2 )/n2 for large n. 4
4 Here we use the step-function approximation for RG running. However, had we used a Jacobi function as
in Ref. [13], we would still find the high mass threshold effects to be negligible.

156

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

We conclude that the differential running due to the Higgs fields dies off quickly above
the compactification scale. Thus the total Higgs contribution to the running between MC /2
and a high scale such as m or g52  2 is essentially independent of the high scale. Taking
the pairs of triplet hypermultiplets to be exactly degenerate, we find that for high scales,


2
2MC
MC
3MC /2
H
+ log
R2 =
log
log
5
MC /2
MC
3MC /2



5MC /2
log
+
2MC
 
2
,
= log
(32)
5
2
and similarly,

3
H
R3 = log
(33)
.
5
2
Note that the ratio R2H /R3H = 2/3 is the same as one would have calculated by including
the contributions from the massless doublets alone. This fact will be important when we
compare with the running in 4D minimal SU(5) in Section 3.3.
3.2. Gauge contributions
In contrast to the Higgs threshold effects, those arising from gauge field loops do not
quickly die off. In fact, we will see that these effects add up to an effective logarithmic
running all the way up to m or g52  2 .
The definitions we have used for RiG are very convenient for making connection
to ordinary, four-dimensional running. In particular, 1 receives no gauge contribution
below the GUT scale in four-dimensional theories, and likewise here R1G will not
receive contributions to differential running. R2G and R3G , on the other hand, will receive
corrections above the compactification scale.
Referring to Eqs. (26) and (27), we see that the quantities RiG are proportional to the
quadratic Casimir coefficients C2 (G) of the gauge groups. Thus, we can write the gauge
contributions as
RiG () = C2 (Gi )6(),

(34)

where C2 (G) is 2 for SU(2) and 3 for SU(3). Furthermore, we have




1
d6()
=
3 + 2 # of (V , )321 levels below
d



2 # of(V , )X,Y levels below .

(35)

The notation here is as in Fig. 1(a). Note that in contrast to the case for the Higgs multiplets,
here the coefficient multiplying 1/ does not average to zero. Rather, for low modes (such
that MnG  (n + 1/2)MC ), the coefficient is 3 up to MC /2, 1 between MC /2 and MC , then
3 again until 3MC /2, and so on. For the higher mass modes, such that the X, Y and 3-2-1
multiplets are nearly degenerate, the coefficient becomes fixed at 1.

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

157

Fig. 2. The qualitative picture for gauge coupling unification in our 5D model. We define
i () i1 () 11 (). The conventional unification scale MGUT 2 1016 GeV is a derived
scale rather than a physical one. Here we assume m < g52  2 , so that unification is achieved near
g52  2 .

If the running were coming entirely from the massless 3-2-1 vector multiplet, the
coefficient multiplying 1/ would be 3, so we see that above the compactification scale,
the differential running due to the gauge loops is slowed somewhat relative to the ordinary
logarithmic evolution in 4D. Note that even once the X, Y and 3-2-1 multiplets become
degenerate, the differential running continues. In this regime, the running is due entirely to
the eaten Goldstone states contained in the X, Y KK tower. The differential running stops
completely only once we reach the scale m , when the rest of the degrees of freedom
begin to propagate in the loops. 5
The qualitative picture for gauge coupling unification in our model is depicted in Fig. 2.
Above the compactification scale, the non-uniform evolution of the gauge couplings slows,
so that unification occurs at a larger scale than usual. The unification scale will essentially
be the larger of m and g52  2 . Next we will verify that this picture is in fact correct, and
we will make more precise the connection to unification in ordinary minimal SU(5) in 4D.
3.3. Unification and minimal SU(5)
Up to this point we have discussed the differential evolution of the couplings above the
compactification scale, but have not explicitly demonstrated that the couplings in our model
unify in a manner consistent with electroweak scale measurements of the 3-2-1 couplings.
Here we will demonstrate that, at one loop, the successful prediction of sin2 W of minimal
SU(5) with MHC < MGUT [17] is equivalent to the successful prediction of sin2 W in this
5 Or, if m happens to be below the the scale at which the X, Y and 3-2-1 towers become degenerate

(determined by g52  2 ), the differential running due to the Goldstone degrees of freedom ceases above m ,
but the total differential running does not die off until a scale g52  2 .

158

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

model with MC MHC . That is, the fact that minimal SU(5) with a triplet Higgs below
the GUT scale can give sin2 W correctly guarantees that this model can do as well.
To see this we must reexamine the running of gauge couplings in four-dimensional,
minimal SU(5). Above the Higgs triplet mass, the gauge couplings evolve as


 (4d)

1 

7 3C2 (Gi ) log


.
i1 () = i1 MHC
(36)
(4d)
2
MH
C
To make connection with our discussion of differential running in the five-dimensional
theory, we subtract off a universal piece 11 from each gauge coupling, and we find


 (4d)
3
MGUT
4d
4d
C2 (Gi ) log
i (MGUT ) i MHC =
(37)
,
(4d)
2
MH
C
(4d)
, the
where MGUT is the conventional GUT scale 2 1016 GeV. Below the scale MH
C
running of the gauge couplings in the 5D theory is the same as for the 4D theory. Thus, if
 (4d)
has the same form as Eq. (37), we will have
we can demonstrate that i5d () i5d MH
C
shown that successful unification in ordinary minimal SU(5) implies successful unification
(4d)
in our model. Here the superscript in MH
C is to emphasize that it is the colored Higgs mass
that gives successful unification in the 4D theory that is meant, rather than the lightest KK
mode of the colored Higgs in the 5D theory.
 (4d)
are
The Higgs contributions to i5d () i5d MH
C


1
MC /2

1
,
c
+
log
ci log
(38)
i
(4d)
2
2
2
MHC

where (c2 , c3 ) = (2/5, 3/5). The first term comes from the ordinary running due to Hu
and Hd below MC /2, and the second term is the finite running above MC /2 found in
Eqs. (32) and (33). Since the two terms are both proportional to ci , we see that the Higgs
(4d)
contribution to the differential running in 4D (which takes place entirely below MHC )
is emulated in 5D with the proper choice of compactification scale. Had the differential
running due to the Higgs multiplets turned off right at MC /2, we would simply identify
(4d)
. Because we find a small amount of differential running above MC /2, we
MC /2 = MH
C
(4d)
instead identify MC / = MH
. We expect small additional corrections to this quantity
C
from further finite one-loop effects, but the precise value MC / is not important: what
(4d)
that leads to a successful prediction of sin2 W in
matters is that, given any value of MH
C

(4d) 
the 4D theory, there is a value for MC MH
such that the total differential running
C
due to Higgs loops in the 5D theory will be the same as in the 4D theory.
 (4d)
arises entirely from loops of
Having chosen this value for MC , i5d () i5d MH
C
gauge and physical fields:


 (4d)
3
MC /2
5d
5d
C2 (Gi ) log
i () i MHC =
(4d)
2
MHC



1

C2 (Gi ) 6() ( m ) log


+
(39)
,
2
m

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

159

where the first term comes from the ordinary 4D running below MC /2, and two terms in
the curly bracket represent the contributions from gauge KK towers and physical fields,
respectively (for the parameterization of the gauge contribution, see Eq. (34)). We find that
this has the same form as Eq. (37), with the required identification



MGUT

.
= 3 log
6() log
m
MC /2

(40)

Note that 6() log(/M) under , where M is some mass scale which is a
function of MC and g52  2 , so that the quantity in the left-hand side has a well-defined
value which is a function of MC , g52  2 and m . (g52  2 is the scale at which the
differential running almost ceases in the 5D theory.)
We see that the form of differential running above the Higgs triplet mass in minimal
SU(5) is precisely what arises from threshold effects in the case of the five-dimensional
theory. This holds irrespective of whether the step-function approximation of fields in the
RG running is very good or not. For instance, had we used the full one-loop calculation
as in Ref. [13], all gauge contributions would still be proportional to C2 (G). At one loop,
the structure of unification is equivalent to that of minimal SU(5). That this model can
successfully yield sin2 W is no more nor less remarkable than in four-dimensional unified
theories. The remarkable feature is that the grand unified group is broken maximally, at
a scale well above the masses of the lightest X and Y gauge bosons. This feature changes
the predictions of the theory, as we will see in Sections 4 and 5.
Let us make one final comment regarding the differential running of the couplings. In
a sense, its logarithmic (as opposed to power law) behavior is natural. The power law
evolution is characteristic of the bulk (i.e., five-dimensional) nature of the gauge coupling.
In contrast, the SU(5) breaking occurs strictly on a brane. Simply by Lorentz invariance,
the brane breaking cannot contribute to the bulk coupling, but instead should just generate
a brane contribution to the gauge kinetic term. Consequently, we would not a priori have
expected the effects of this SU(5) breaking to exhibit a power law behavior.
3.4. Scales
Now that we have shown the direct correspondence between the differential running
in our model to that of ordinary minimal SU(5), we can determine the scale at which
successful unification occurs. In other words, given a compactification scale MC , we can
find the values of m and  that give sin2 W correctly.
The allowed compactification scales are already known: these are related to the allowed
values of MHC in minimal SU(5) by MC MHC . Then we need only to find the
values of m and  for which Eq. (40) holds. These values are indicated in Fig. 3
for compactification scales between 6 1014 GeV and 6 1015 GeV (corresponding to
MHC 2 1014 GeV and 2 1015 GeV). For these compactification scales, the 5D Planck
scale M is (1 3) 1017 GeV, and Fig. 3 shows that the GUT-breaking scale can be
at or near this scale and the unification of three gauge couplings can be attained below

160

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

Fig. 3. Values of M and g52  2 which generate the same threshold effect above MC as is
generated in minimal supersymmetric SU(5) between MHC and MGUT . The dashed line corresponds
to MC = 6 1014 GeV, while the solid line corresponds to MC = 6 1015 GeV. To calculate these
lines we have used first-order solutions in (g52  2 R)1 for the X and Y masses.

M depending on the parameters of the model. 6 In the parameter region where the field
theoretic unification works, the ratio of the cut-off to the compactification scales is O(10
100), and no gauge or Yukawa couplings become nonperturbative below the cut-off scale.
Thus, our one-loop treatment of the running is well justified. We also find that  must be
somewhat smaller than M in this parameter region. This is consistent with the observation
that the theory is more or less strongly coupled at the scale M to have O(1) Yukawa and
gauge couplings in 4D; if the theory is strongly coupled at M ,  is naturally  
M /(4), since the superpotential giving the vev of scales like M 2 + 4 3 . Even
then, however, the couples to the other fields in the combination of 4/M , so that
various operators feel order one GUT breaking, since 4 /M 1. Therefore, we treat
as if has a vev of order M in the following discussions, although all the arguments also
apply in the strongly coupled case,   M /(4).
3.5. Brane operators

We finally comment on effects from brane operators like d 2 (/M )m W W (y
R) where m = 1, 2, . . . , and W is the field strength superfield. Although it is possible
that these operators are somehow suppressed at the cut-off scale, from effective field theory
point of view it is generically expected that they are present with order one coefficients.
They give tree-level splitting of three gauge couplings for SU(3), SU(2) and U(1), and
could affect the previous analysis of the gauge coupling unification. In particular, since
6 If m or g 2  2 is above the 5D Planck scale, the gauge unification is not completed in a field theoretic

5
regime. In this case, our field theoretic treatment is not fully trustable above M and would have some

uncertainties coming from the cut-off scale physics.

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

161

we are considering  M , one might think that they give O(1) correction to sin2 W .
However, we can expect that the effect of these operators are actually smaller by making
the following observations.
As an example, let us first consider the extreme case where all the interactions are
strongly coupled at the scale M . Then, the operators involving the field strength superfield
scale as



M
1 4

W
W
+
(y

R)
W
W
+

,
L5 = d 2
(41)

24 3
16 2 M
where 4 /M is an O(1) quantity. (Strictly speaking, 4 /M must be somewhat
smaller than 1 so that all higher-dimensional operators involving (4 /M )m do not
equally contribute and make the theory unpredictable.) On integrating over y, the zeromode 4D gauge couplings g0 are given as
1
M R
1

+
.
2
2
2
12
16
g0

(42)

Here, the first and second terms are SU(5)-preserving and SU(5)-violating contributions,
respectively. Since we know that g0 1, we have to take M R 12 2 in this strongly
coupled case. This shows that the SU(5)-violating contribution coming from brane
operators is suppressed by a factor of 1/(16 2 ) in this case.
In fact, the theory is not truly strongly coupled at the scale M in the realistic case
discussed in previous sections so that the one-loop treatment of gauge coupling evolutions
is reliable. Nevertheless, the above argument applies more generically; the SU(5)-violating
brane contribution is small relative to the SU(5)-preserving bulk contribution due to the
large volume factor 2RM . Thus, the correction to sin2 W is expected to be small. We
will not discuss possible effects of the brane operators further, and assume that they are
negligible in the subsequent discussions.

4. Proton decay
One of the key signals of grand unification is proton decay. X and Y gauge boson
exchange generates dimension-six proton decay operators in the low energy theory, and
Higgsino triplet exchange generates dimension-five operators. One might expect that since
the particles appear at a scale MC , this will be the suppression scale of the proton decay
operators. One interesting possibility offered by this framework is that the rate of proton
decay is not constrained by gauge invariance to be related to the strength of gauge and
Yukawa interactions in the usual way.
We have already noted that the dangerous particles have wave functions at the y = R
fixed point that are small compared with those of their 3-2-1 or doublet counterparts. On the
other hand, there is a tower of states that can mediate proton decay, and we must sum each
contribution. In this section, we investigate proton decay operators generated by exchanges
of X, Y gauge bosons and colored Higgsinos, and show that the present model can satisfy
the constraints coming from experimental lower bounds on the proton lifetime.

162

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

4.1. Dimension-six operators


Dimension-six proton decay operators arise from the exchange of X and Y gauge
bosons. The coupling of these bosons to fields on the brane is suppressed by a factor
cos(MnG R). Comparison with four-dimensional theories can be made by replacing
1
2
MX,Y


cos2 (M G R)
n=0

n
2
G
Mn

(43)

where MnG are the masses of the X and Y gauge boson KK modes. These masses satisfy
Eq. (5), which allows us to approximate the sum as

tanh(g52  2 /2MC )
4
=
.
g 4  4 + MC2 (2n + 1)2
g52  2 MC
n=0 5

(44)
1/2

Thus, the effective mass of the X and Y bosons is g5  MC /(2)1/2 = g4  . For


 at or near the five-dimensional Planck scale, this will typically be larger than MGUT ,
although it is not required to satisfy experimental constraints. In any case, despite the fact
that the lightest X and Y gauge bosons have masses of only 1015 GeV, it is easy to satisfy
proton decay constraints coming from dimension-six operators in this model.
4.2. Dimension-five operators
Dimension-five operators come from integrating out the Higgsino triplets. Again, we
make connection with the four-dimensional theory by finding an effective triplet Higgsino
mass by taking the whole sum over KK modes. Here we make the replacement


cos2 (MnH R)
1

,
MHC
MnH
n=

(45)

where MnH are the masses of the colored Higgs KK modes, given in Eq. (21). Using this
spectrum, we obtain the sum


8
4
1
=
.
(46)
2
2
+ 4 nMC + MC arctan(/2)/
( + 4)MC
n=
Thus, proton decay from dimension-five operators is suppressed by a factor 8/ 3
compared to the compactification scale. In ordinary minimal SU(5), proton decay limits
require MHC > 9.4 1016 GeV, so for a compactification scale of MC = 2 1015 GeV,
we would need a parameter of ten to adequately suppress proton decay operators. This
leaves open the possibility that proton decay could be observed in future experiments.
However, for larger detection becomes increasingly unlikely.
4.3. Derivative operators
In addition to the ordinary Yukawa couplings between the Higgs and matter fields on
the brane, there can be derivative couplings of the conjugate Higgs fields to the matter

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

163

fields as well. 7 These operators such as (u T10 T10 y H c + d T10 F5 y H5c )(y R) can
5
lead to proton decay. Of course, coefficients u and d of these operators are not related
to the usual Yukawa couplings, u and d , by SU(5), so that their actual significance is
unknown. Furthermore, there is an ambiguity in what we mean by the derivative of the
conjugate field, which is not differentiable at the point y = R. However, using Eqs. (6)
and (7), we can rewrite the coupling as




c
H
Mn gC,n (R)
g (R) C,n (x),
(y R)y HC = (y R)
2R m C,n
n
(47)
where the summation over m arises from rewriting the -function as a sum of the KK mode.
While we can compute proton decay diagrams involving these vertices, it is now
apparent that such diagrams have a strong dependence on how we cut off the sum of the
KK mode. If the cut-off is done near the fundamental scale, these diagrams can, at least
in principle, give comparable contribution to those involving the usual Yukawa couplings.
However, it is not entirely clear what the cut-off for these diagrams should be. In the case
where the brane is dynamical, the summation in the above operators are cut off at the scale
of the brane tension [18]. For couplings at the orbifold fixed point, which is not dynamical
and cannot fluctuate, it is unclear whether the cut-off is the fundamental scale or a lower
scale, such as the radion mass.
For our purposes here, we will not address these issues further. Since we cannot a priori
know the size of the couplings of these operators and their flavor structure, estimating
the resulting proton decay rate is already very uncertain. However, it is possible that such
operators may provide an opportunity for detectable dimension-five proton decay in the
near future.

5. Yukawa couplings
So far, we have a framework that looks like ordinary SU(5) except with suppressed
proton decay. While the breaking occurs at a scale near the 5D Planck scale M , the
X, Y gauge bosons are much lighter. Still, in a very real sense, the breaking of SU(5)
is maximal, which can manifest itself in deviations from SU(5) expectations. We here
consider the effects of such maximal GUT breaking on the fermion Yukawa couplings.
In the minimal SU(5), one important prediction is the unification of the Yukawa
couplings. A successful prediction of the theory is the unification of the bottom and
Yukawa couplings at the GUT scale [19]. However, it is well known that the SU(5) relations
fail in the lighter first two generations. For instance, an SU(5) relation me /m = md /ms
fails by a factor of ten. 8
7 We thank M. Graesser for bringing this to our attention.
8 One may be able to correct this prediction by means of contributions from supersymmetry-breaking A terms

to the Yukawa couplings [20].

164

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

In ordinary 4D SU(5) GUT, it has been suggested that the operators involving the field
can correct this discrepancy [21]. However, this mechanism does not work in most theories
where the fermion mass hierarchy is explained by the FroggattNielsen mechanism [22].
In this mechanism, the matter fields carry generation dependent flavor U(1) charges and
the various Yukawa couplings are generated through the U(1) breaking spurion. This is
an attractive mechanism in that it not only suppresses the first-two generation Yukawa
couplings but also suppresses dangerous tree-level dimension-five proton decay operators.
However, having employed this mechanism, the GUT-breaking operators involving 
can modify SU(5) mass relations only by an amount suppressed by a factors of MGUT /MPl ,
which is too small to accommodate order one deviations suggested by the observed quark
and lepton masses. 9
In the present model, on the other hand, there is no suppression of higher-dimensional
operators involving GUT-breaking effects, since  is near the cut-off scale M . In this
sense, fields living on the GUT-breaking brane see the GUT broken maximally, and their
Yukawa couplings need not respect the SU(5) symmetry. Thus, the failure of SU(5) to
describe (me /m )/(md /ms ) seems quite natural. In this framework, however, the success
of the b unification must be viewed as an accident, unless there is a reason why the
coupling of to the third generation is somewhat suppressed.

6. Larger gauge groups


Up to this point, we have restricted our attention to a scenario with SU(5) gauge group,
but it is interesting to consider larger gauge groups. One possible difference comes from
additional matter fields required by larger gauge groups. For instance, in SO(10) there is an
additional state, right-handed neutrino. If this additional field acquires a mass around the
cut-off scale M from SO(10) breaking, then there will be a running effect between MC
and M which is not SO(10) universal.
It is, however, important that this does not affect our previous analyses; the 3-2-1
unification still works as in Section 3 even in the case of SO(10). The additional Higgs
states do not contribute to differential running above MC , and the threshold effects are
proportional to 3-2-1 -functions. This means that we can have larger gauge groups broken
all the way to 3-2-1 around the cut-off scale. In the SO(10) case, this may result in too large
right-handed neutrino masses to give an appropriate mass scale for atmospheric oscillations
[24] through see-saw mechanism [25]. Then, we may need somewhat small coefficient in
front of the operator which gives right-handed neutrino masses or have to resort to other
ways to generate neutrino masses within supersymmetric models, for instance, though Rparity violation [26] or supersymmetry breaking [27].
9 One possible way to evade this conclusion is to assign a non-vanishing FroggattNielsen charge to the
field [23].

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

165

7. Conclusions
The possibility that the standard model gauge group is merely a subgroup of a larger,
simple group is an attractive one. Unfortunately, the simplest version of supersymmetric
SU(5) predicts proton decay at rates incompatible with experiment.
We have demonstrated that the incorporation of just one new ingredientan additional
dimension in which gauge and Higgs fields propagatebrings about crucial changes
relative to ordinary GUTs. The couplings of the lightest X and Y bosons are no longer
directly linked through gauge symmetry to those of the lightest 3-2-1 bosons, and the
couplings of the lightest Higgs triplets are no longer directly linked to those of the Higgs
doublets. The strong breaking of the GUT symmetry pushes away the wave functions
of these states, suppressing the generated proton decay below experimental limits, all
while retaining ordinary, logarithmic unification. Such features seem special to a fivedimensional theory, incapable of reproduction in simple four-dimensional theories.
The presence of extra dimensions of this size has been motivated previously as a means
to resolve the supersymmetric flavor problem [28]. Our GUT-breaking picture can nicely fit
into this framework of supersymmetry-breaking mediation. For instance, if supersymmetry
is broken at y = 0 fixed point by F -term vev of singlet or non-singlet field, it naturally
realizes the scenarios of Refs. [29] and [30], respectively. The incorporation of our GUT
picture within these scenarios may give interesting signatures, since we now have one
more piece of information about parameters of the model; the compactification radius
is determined by the gauge coupling unification. We leave an investigation of detailed
phenomenology for future work.
To summarize, the framework we have described is very simple, but can describe
various observed features such as gauge coupling unification, lack of unification in Yukawa
couplings, the absence of proton decay, and so on. It will be interesting to add non-minimal
structure to the model as a means of deriving experimental signatures.

Acknowledgement
We thank M.L. Graesser, L.J. Hall, D.B. Kaplan, D.E. Kaplan, A.E. Nelson, M.E. Peskin
and L. Randall for useful discussions.

References
[1] H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438.
[2] H. Georgi, H.R. Quinn, S. Weinberg, Phys. Rev. Lett. 33 (1974) 451.
[3] S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150;
N. Sakai, Z. Phys. C 11 (1981) 153.
[4] S. Dimopoulos, S. Raby, F. Wilczek, Phys. Rev. D 24 (1981) 1681;
L.E. Ibanez, G.G. Ross, Phys. Lett. B 105 (1981) 439.
[5] N. Sakai, T. Yanagida, Nucl. Phys. B 197 (1982) 533;
S. Weinberg, Phys. Rev. D 26 (1982) 287.

166

Y. Nomura et al. / Nuclear Physics B 613 (2001) 147166

[6] J. Hisano, H. Murayama, T. Yanagida, Nucl. Phys. B 402 (1993) 46, hep-ph/9207279;
For a recent analysis, see: T. Goto, T. Nihei, Phys. Rev. D 59 (1999) 115009, hep-ph/9808255.
[7] H. Murayama, A. Pierce, in press.
[8] P. Candelas, G.T. Horowitz, A. Strominger, E. Witten, Nucl. Phys. B 258 (1985) 46;
E. Witten, Nucl. Phys. B 258 (1985) 75.
[9] J.D. Breit, B.A. Ovrut, G.C. Segre, Phys. Lett. B 158 (1985) 33.
[10] A. Sen, Phys. Rev. Lett. 55 (1985) 33.
[11] Z. Chacko, M.A. Luty, E. Ponton, JHEP 0007 (2000) 036, hep-ph/9909248.
[12] N. Arkani-Hamed, L. Hall, Y. Nomura, D. Smith, N. Weiner, hep-ph/0102090.
[13] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
[14] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[15] D. Ghilencea, G.G. Ross, Phys. Lett. B 442 (1998) 165, hep-ph/9809217;
D. Ghilencea, G.G. Ross, Nucl. Phys. B 569 (2000) 391, hep-ph/9908369.
[16] L. Hall, Y. Nomura, hep-ph/0103125.
[17] J. Hisano, H. Murayama, T. Yanagida, Phys. Rev. Lett. 69 (1992) 1014.
[18] M. Bando, T. Kugo, T. Noguchi, K. Yoshioka, Phys. Rev. Lett. 83 (1999) 3601, hep-ph/9906549.
[19] M.S. Chanowitz, J. Ellis, M.K. Gaillard, Nucl. Phys. B 128 (1977) 506.
[20] J.L. Diaz-Cruz, H. Murayama, A. Pierce, hep-ph/0012275.
[21] J. Ellis, M.K. Gaillard, Phys. Lett. B 88 (1979) 315.
[22] C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
[23] K.I. Izawa, K. Kurosawa, Y. Nomura, T. Yanagida, Phys. Rev. D 60 (1999) 115016, hepph/9904303.
[24] Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562, hepex/9807003.
[25] T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proc. of the Workshop on the Unified Theory
and Baryon Number in the Universe, KEK report 79-18, 1979, p. 95;
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D.Z. Freedman (Eds.),
Supergravity, North-Holland, Amsterdam, 1979, p. 315.
[26] L.J. Hall, M. Suzuki, Nucl. Phys. B 231 (1984) 419;
I. Lee, Nucl. Phys. B 246 (1984) 120.
[27] N. Arkani-Hamed, L. Hall, H. Murayama, D. Smith, N. Weiner, hep-ph/0006312;
N. Arkani-Hamed, L. Hall, H. Murayama, D. Smith, N. Weiner, hep-ph/0007001;
F. Borzumati, Y. Nomura, hep-ph/0007018.
[28] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155.
[29] D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323.
[30] D.E. Kaplan, G.D. Kribs, JHEP 0009 (2000) 048, hep-ph/0009195.

Nuclear Physics B 613 (2001) 167188


www.elsevier.com/locate/npe

Massless 3-branes in M-theory


M. Cvetic a,1 , H. L b,2 , C.N. Pope c,3
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
b Michigan Center for Theoretical Physics, University of Michigan, Ann Arbor, MI 48109, USA
c Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA

Received 29 June 2001; accepted 2 August 2001

Abstract
We construct supersymmetric M3-brane solutions in D = 11 supergravity. They can be viewed as
deformations of backgrounds taking the form of a direct product of four-dimensional Minkowski
spacetime and a non-compact Ricci-flat manifold of G2 holonomy. Although the 4-form field
strength is turned on it carries no charge, and the 3-branes are correspondingly massless. We also
obtain 3-branes of a different type, arising as M5-branes wrapped over S 2 . 2001 Published by
Elsevier Science B.V.

1. Introduction
The standard D3-brane provides a natural supergravity dual of four-dimensional N = 4
superconformal YangMills theory, via the AdS/CFT correspondence [13]. Branes with
less supersymmetry can in general be constructed by replacing the spheres that form the
level surfaces in the flat transverse space by some other Einstein space that admits a lesser
number of Killing spinors [4]. It was proposed that D3-branes on the six-dimensional
conifold, in which the level surfaces are the T 1,1 space, is dual to an N = 1 superconformal
theory in D = 4 with gauge group SU(N) SU(N) [5]. The conformal symmetry can
then itself be broken, by introducing fractional branes corresponding to the wrapping of
D5-branes on 2-cycles. The corresponding supergravity solutions were obtained in [6,7].
It has been proposed that M-theory compactified on a certain singular seven-dimensional
space with G2 holonomy might be related to a N = 1, D = 4 gauge theory [810], which
has no conformal symmetry to begin with. (See also the recent papers [1115].) This leads
to the question of whether there might exist a 3-brane configuration in M-theory, whose
E-mail address: cvetic@cvetic.hep.upenn.edu (M. Cvetic).
1 Research supported in part by DOE grant DE-FG02-95ER40893 and NATO grant 976951.
2 Research supported in full by DOE grant DE-FG02-95ER40899.
3 Research supported in part by DOE grant DE-FG03-95ER40917.

0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.


PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 8 1 - 9

168

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

transverse space is a deformation of a Ricci-flat space of G2 holonomy, in which the 4-form


field is turned on. In this paper we shall indeed obtain 3-brane solutions of this deformed
type.
So far, three explicit metrics for seven-dimensional manifolds of G2 holonomy are
known [16,17]. They all have cohomogeneity one. The first two have principal orbits that
are CP3 or SU(3)/(U (1) U (1)), written as an S 2 bundle over S 4 or CP2 , respectively.
The associated 7-manifolds have the topology of R3 bundles over S 4 or CP2 . The third
manifold has principal orbits that are topologically S 3 S 3 , written as an S 3 bundle over
S 3 , and the 7-manifold is topologically R4 S 3 . In order to construct a non-trivial 3-brane
configuration on such a background in eleven-dimensional supergravity, it is necessary that
the background G2 manifold itself should admit a well-behaved harmonic 4-form (or dual
3-form). It was shown in [18,19] that such harmonic forms exist in all three of these explicit
examples. In this paper we construct M3-brane configurations describing deformations
away from backgrounds having a G2 manifold as the transverse space, taking the 4-form
field strength of M-theory to be proportional to the appropriately deformed harmonic 4form. We first obtain the second-order equations for the fields in our ansatz, which follow
from those of eleven-dimensional supergravity, and then we show that in a Lagrangian
formulation of these equations the potential can be derived from a superpotential. This
leads to first-order equations which we are able to solve explicitly.
The exact solutions that we obtain by this method describe configurations with a fourdimensional Poincar invariance in the world-volume, and a seven-dimensional transverse
space that is a deformation of the original Ricci-flat metric of G2 holonomy. We may thus
view them as being 3-brane solutions of M-theory. At large distance they approach the
product of 4-dimensional Minkowski spacetime and the Ricci-flat metric of G2 holonomy.
The rate at which the metrics approach this asymptotic form is rapid enough that the ADM
mass vanishes, and so they may be thought of as massless M3-branes. In common with
other examples of massless branes, they have naked singularities at short distance. We
show that the M3-brane solutions are supersymmetric.
It is of interest also to look for 3-brane configurations in M-theory within a more general
framework. Another natural candidate for a 3-brane is to look for an M5-brane wrapped
on a supersymmetric 2-cycle. Wrapped supersymmetric M5-branes have been discussed in
previous papers [30,31], and typically these have been of the form AdSd H7d , where
Hn denotes the n-dimensional hyperbolic space. In Section 5 we shall consider M5-branes
wrapping around a 2-sphere. The solutions can be obtained by starting with SU(2)-gauged
AdS supergravity in D = 7, and looking for 3-branes supported by the YangMills fields.
We obtain the equations of motion for the general non-abelian case, and show that when
only a U (1) subgroup is turned on, we can construct first-order equations derivable from a
superpotential. The general solution can be reduced to Abels equation, and the structure of
the resulting configurations can be analysed. The solutions can be lifted to D = 11, where
they describe 3-branes as M5-branes wrapped on S 2 .

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

169

2. M3-branes in backgrounds of R3 bundles over S 4 or CP2


In this section we shall construct M3-brane solutions that can be viewed as living in
backgrounds where the 7-dimensional transverse space is a manifold of G2 holonomy with
the topology of the R3 bundle over S 4 or CP2 .
2.1. The ansatz
Let us consider the D = 11 ansatz
2
ds11
= H 2 dx dx + d 2 + a 2 Di Di + b2 d42 ,

(1)

j
where i i = 1 and Di = di + ij k A(1) k , and Ai(1) is the SU(2) YangMills instanton
on S 4 , whose unit metric is d42 . The functions H , a and b will be taken to depend only
on the radial coordinate in the transverse space. This describes the case of the R3 bundle

over S 4 . The second possibility is obtained by replacing the S 4 by CP2 . This does not
affect the form of the equations for H , a and b. 4 The constrained i coordinates can be
expressed in terms of two angular coordinates on S 2 in a standard way,
1 = sin sin ,

2 = sin cos ,

3 = cos .

The vielbein components in the S 2 fibre directions are then given by




e1 = a d A1(1) cos + A2(1) sin ,


e2 = a sin d + A1(1) cot sin + A2(1) cot cos A3(1) .

(2)

(3)

There is clearly a vacuum solution of the form (1) that is simply the direct product
of four-dimensional Minkowski spacetime and the associated Ricci-flat seven-dimensional
manifold with G2 holonomy. In the vacuum we shall have H = 1, with a and b being given
by [16,17]. We should now like to turn on the 4-form field strength of eleven-dimensional
supergravity. The 4-form ansatz that respects the symmetry of the metric is given by [17,
19]
F(4) = f1 (4) + f2 X(2) Y(2) + f3 d Y(3) ,
F (4) = H 4 a 2 b4 f1 (4) d X(2) + H 4 a 2 f2 (4) d Y(2)
+ H 4 f3 (4) X(3) ,

(4)

where the fi are functions depending only on , and


1
i
i
Y(2) i F(2)
,
X(3) Di F(2)
,
X(2) ij k i Dj Dk ,
2
k
,
(4) dt dx1 dx2 dx3 .
Y(3) ij k i Dj F(2)

(5)

(Note that F(4) could in principle have had a term of the form d X(3) as well, but this
is ruled out by the field equation dF(4) = 0.) The Bianchi identity dF(4) = 0 implies that
4 In everything that follows, results obtained for the case of the S 2 bundle over S 4 apply equally, mutatis
mutandis, to the case of the S 2 bundle over CP2 .

170

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

f1 = 4f3 and f2 = 2f3 , so we can take f1 = 2f , f2 = f and f3 = 12 f  , giving


 1

F(4) = f 2(4) + X(2) Y(2) + f  d Y(3) .
2

(6)

In fact, F(4) = dA(3) with A(3) = 12 f Y(3) . The field equation dF(4) = 0 implies





1
2a 4 + b4 H 4 f a 2b4 H 4 f  = 0.
(7)
2
(The F(4) F(4) term vanishes here.)
In order to impose the D = 11 Einstein equation it is convenient to perform a Kaluza
Klein reduction on the 4-dimensional world-volume of the 3-brane, so that the problem
can be reformulated from a seven-dimensional point of view. This allows us to make use of
curvature calculations for 7-metrics of this type that were performed in [17]. The relevant
seven-dimensional Lagrangian is given by
1
1
e1 L7 = R ()2 e
2
48

8
5

2
F(4)
,

(8)

with the ansatz (1) now taking the form


ds72 = dt 2 + a 2 Di Di + b 2 d42 ,
 1

F(4) = f 2(4) + X(2) Y(2) + f  d Y(3) .
2
The metric in D = 7 is related to the one in D = 11 by
2
d s11

=e

32

2
5

ds72

+e

1
3

5
2

dx dx .

(9)

(10)

Thus we have
dt = H 4/5d,

a = H 4/5a,

b = H 4/5b.

(11)

The original eleven-dimensional Einstein equation is now recast as the seven-dimensional dilaton equation and Einstein equation. The dilaton equation gives


 2 4 4   2 4  2 2f 2 4f 2 a 2
a b H
(12)
= H f + 2 +
,
3
a
b4
which can be rewritten as
f 2
6H  18H  2 12a  H  24b H 
2f 2
4f 2
+
+
=
+
+
+
.
H
H2
aH
bH
a 2 b4 a 4 b4
b8
From the Einstein equation we get three separate equations, namely,
5
5a 2
5a  28a  H  5a  2 20a b 16b H  4H  12H  2
+
+ 2 +
+
+
+
2 4
2
a
aH
a
ab
bH
H
H
a
b
2

2
2
f
2f
6f
= 2 4 4 4+ 8 ,
4a b
a b
b

(13)

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

5b 36bH  15b2 10a b 8a  H  4H  12H  2 15 5a 2


+
+ 2 +
+
+
+
2+ 4
b
bH
b
ab
aH
H
H2
b
b
f 2
f2
4f 2
= 2 4+ 4 4 8 ,
4a b
2a b
b




8a H
20b
16b H  24H  12H 2
10a
+
+
+
+
+
a
aH
b
bH
H
H2
f 2
3f 2
6f 2
= 2 4 + 4 4 + 8 .
a b
a b
b
This set of equations can be derived from the Lagrangian L = T V , where

171

(14)

40bH
15H 2
5f2
5a 2 20a b 15b 2 20a H

+
,
ab
aH
bH
2a 2
b2
H2
8a 2b4



 

5
V = H 8 2a 2b4 a 4 + b4 6a 2 b2 2a 4 + b4 f 2 ,
(15)
4
together with the constraint T + V = 0. Note that here a dot is a derivative with respect
to the coordinate , defined by d = a 2 (bH )4 d. The kinetic term T can be expressed as
T = 12 gij i j , where a = e1 , b = e2 , H = e3 and f = 4 , with

5 20 20
0
20 30 40
0
.
gij =
(16)
20 40 30
0
T =

5
4a 2 b4

Since gij is field dependent, the system is a non-linear sigma-model. Nevertheless, we


find that the potential V can be expressed in terms of a superpotential W , so that V =
12 g ij i W j W , where

5 
W = H 4 a 2 + b2 4a 2 b4 + f 2 .
(17)
2
2.2. First-order equations and general solution
From the superpotential (17), we can derive the first-order equations i = g ij j W ,
which, in terms of the original radial coordinate of Eq. (1) become
6a 4b4 6a 2 b6 + a 2 f 2 2b2f 2
2(a 2 + b2 )f
,
,
f =
4
K
3ab K
12a 4b4 + 4a 2 f 2 + b2 f 2
(a 2 + b2 )f 2 H
b =
,
H =
,
2
3
6a b K
3a 2b4 K
where a prime denotes d/d, and we have defined

K 4a 2 b4 + f 2 .
a =

(18)

(19)

Solutions of these equations will necessarily satisfy the original second-order equations,
implying that we shall have solutions of the D = 11 supergravity equations. One may note

172

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

from (18) that


ab2H 3 f = ,

(20)

where is a constant of integration.


In order to solve the first-order equations (18) it is helpful to define new hatted variables
as follows:

b = H 1/2b,

a = H 1/2a,

f = H 3/2f.

At the same time, we introduce a new radial variable , defined by d =


metric ansatz (1) now assumes the form


2
= H 2 dx dx + H 1 d 2 + a 2 D2 + b 2 d42 .
ds11

(21)
H 1/2 d.

The
(22)

The first-order equations (18) are considerably simplified, becoming




(a 2 b 2 )K
K
d a
d b
=
=
,
,
d
d
2a b 4
2b 3

(a 2 + b 2 )f2
1 dH
1 d f (a 2 + b 2 )K
=
=
,
(23)
,

H d
2a b 4
3a 2b 4 K
f d

 4a 2 b 4 + f2 .
where K
 d , puts these equations
A further change of radial coordinate to r, defined by dr = K
in the form
(a 2 b 2 )
d a
=
,
dr
2a b 4
(a 2 + b 2 )
1 d f
=
,
2a b 4
f dr

1
d b
=
,
dr
2b 3
(a 2 + b 2 )f2
1 dH
=
.
H dr
3a 2 b 4 (4a 2 b 4 + f2 )

(24)

In particular, the equations for a and b have now decoupled from the rest, and they are in
fact nothing but the first-order equations for the G2 metrics on the R3 bundle over S 4 (see,
for example, [20]).
The system of first-order equations is now completely solvable. After a final change of
radial variable r 12 r 4 , we find that the general solution for the metric in the ansatz (1) is
1
2
= H 2 dx dx + 2H 7 U 1 dr 2 + r 2 H 1 U Di Di + r 2 H 1 d42 ,
ds11
2
where

1/6
c2
/4
H = 1 + 12 2
,
U =1 4,
r
2r U

(25)

(26)

and c is a constant. The function f appearing in the 4-form ansatz (6) is given by
f=

c
r 3 H 3/2U 1/2

Note that the constant appearing in (20) is precisely the constant c.

(27)

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

173

2.3. Properties of the 3-brane solution


The general solution contains two non-trivial integration constants, / and c. The constant
/ measures the scale size of the gravitional instanton of the background G2 manifold.
The constant c, on the other hand, measures the strength of the the 4-form field F(4) .
Asymptotically at large distance, the gravitational instanton contributions, going as /4 /r 4 ,
dominate in comparison to the F(4) contribution, going as c2 /r 12 .
When the constant c is set to zero, the 4-form field is turned off, and consequently the
configuration reduces to the vacuum solution of a direct product M4 M7 of Minkowski 4spacetime and the seven-dimensional smooth manifold with G2 holonomy. When c is nonvanishing, the solution describes a 3-brane in D = 11, with a four-dimensional Poincar
symmetry in its world-volume. Asymptotically, the solution approaches M4 M7 . If the
parameter / is taken to be zero, then the smooth 7-manifold M7 has a singular limit to
the cone over the S 2 bundle over S 4 (or CP2 ). At small distance, near r = /, the metric
becomes singular, with the limiting form


2
1/2

1/2 1
i
i
2
D D + d4 + d 2 ,
dx dx +
/ = 0: ds11 =
2
1
2
/ = 0: ds11
(28)
= 2/5 dx dx + 1/5 Di Di + 4/5 d42 + d 2
2
as the proper distance tends to zero. We can also consider the the possibility of sending
/4 /4 . In this case, the metric behaviour at small proper distance (i.e., near r = 0)
becomes


1
2
1/4

i
i
ds11 =
(29)
dx dx + D D + 1/2 d4 + d 2 .
2
The 4-form flux is given by

Q = 2f (4),

(30)

S4

and from (27) this can be seen to vanish when r is sent to infinity. Thus our M3-brane
configurations do not carry any conserved charge, and might be described as 3-branes
without 3-branes. The solution should really be thought of as a gravitional monopole
involving the supergravity multiplet. The two constants / and c are both continuous
parameters.
What is perhaps the most interesting feature of the solution is that it is massless. The
leading-order r-dependence in the function H at large r is c2 /r 12 , whilst a mass term for
a 7-dimensional transverse space would have a leading-order r-dependence of the form
m/r 5 . The existence of the naked singularity may be related to the fact that the solution
is massless. In fact, all the previously known massless p-brane solutions in supergravity
contain naked singularities [2125]. A repulsion mechanism was proposed in string theory
[26] to resolve such a naked singularity in the massless dyonic string [25]. A further
observation is there appears not to exist a natural and non-trivial decoupling limit. This

174

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

may also be a consequence of the masslessness. For example, a massive dyonic string has
a decoupling limit, but this ceases to exist when the charges are tuned for masslessness.
Unlike the dyonic string, where the masslessness is achieved by making an adjustment of
integration constants, in our new M3-brane solution there is no mass integration constant.
From the point of view of supersymmetry, the masslessness is consistent with the absence
of a non-vanishing conserved 4-form charge. However, we shall defer a more detailed
investigation of the supersymmetry of this solution until Section 4.
3. M3-brane in background of R4 bundle over S 3
In this section we construct an analogous 3-brane solution in the background of the third
manifold of G2 holonomy, whose topology is R4 S 3 . As we shall see, the configuration
is again a no-braner, which carries no conserved brane charge.
3.1. The ansatz
In this case, we consider the eleven-dimensional metric ansatz
2
ds11
= H 2 dx dx + d 2 + a 2 i2 + b2 i2 ,

(31)

where i i
and i and i are two sets of left-invariant on two independent
SU(2) group manifolds. The level surfaces r = constant are therefore an S 3 bundle over
S 3 . Since the bundle is a trivial one, the level surfaces are topologically S 3 S 3 . There is a
vacuum solution which is a complete Ricci-flat manifold, namely the direct product of fourdimensional Minkowski spacetime and the known seven-dimensional manifold R4 S 3
with G2 holonomy [16,17].
Again we should now like to turn on the 4-form field strength, in order to introduce
a 3-brane configuration in this background. The ansatz for the 4-form, respecting the
symmetries of the vacuum, can be written as [18]
1
2 i ,

F(4) = f1 i j i j + f2 d 1 2 3
1
+ f3 ij k d i j k .
2
The Bianchi identity dF(4) = 0 gives
1
1
f1 f2 + f3 = 0,
8
2
and the field equation d F(4) = 0 gives

(32)

(33)



2f1 H 4 
3f1 H 4 
+ f3 ab1 H 4 = 0,
+ 2f2 b3 a 3 H 4 = 0.
(34)

ab
ab
It is again convenient to derive the conditions implied by the eleven-dimensional Einstein
equation in terms of a dimensional reduction to D = 7. The dilaton equation gives
f32
2f12
f22
H  3H  2 3a  H  3b H 
+
+

= 0,
H
aH
bH
H2
a 4 b4 6a 6 2a 2 b4

(35)

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

175

and finally, the seven-dimensional Einstein equation gives


5
5a 2
5b 4H  15a b 10b2 12a H  32bH  12H 2
+
+
+ 2 +
+
+
2+
2
b
H
ab
aH
bH
b
H
2b
16b4
2
2
2
f
2f
3f
+ 4 14 26 + 23 4 = 0,
a b
2a
2a b




4H
15a b
10a 2 32a H  12bH  12H 2
5a
5
5a 2
+
+
+ 2 +
+
+
2
2
a
H
ab
a
aH
bH
H
2a
32b4
2f 2
2f 2
f2
+ 4 14 + 26 2 34 = 0,
a b
a
a b


15b
24H  12a  H  12b H  12H  2
15a
+
+
+
+
+
a
b
H
aH
bH
H2
3f 2
18f 2 f 2
4 14 + 26 + 2 34 = 0.
(36)
a b
a
a b
3.2. First-order equations and general solution
From (34), we can solve for f1 and f2 ,




1
3
f1 = abH 4 f3 ab1 H 4 ,
(37)
f2 = a 3 b3 H 4 + f3 ab1 H 4 ,
2
4
where is a constant of integration. The remaining equations for a, b, H and f3 can then
be obtained from the Lagrangian L = T V , together with the constraint T + V = 0,
where T = 12 gij i j with i = (log a, log b, log H, f3 ), and a dot denotes a derivative
with respect to defined by dt = a 3 b3 H 4 d. We have

15f 2
15f 2
60f 2
3
60 b43
90 + b43
120 b4 3 15f
b4
2

15f 2
60f 2
15f3
90 + 15f43

60 b43
120 + b4 3

b
b4 .
gij =
(38)
60f32
60f32
240f32
60f3
120 4

120
+
120

4
4
4
b
b
b
b
3
15f
b4

15f3
b4

3
60f
b4

15
b4

The potential V is given by



15/32 8 4  4 2
a a b + 16a 2b4 + 16b6 + 3a 4 b2 f32 + 16b2f32
H


45
+ a 6 2ab1H 4 f3 .
(39)
32
As in the previous case, the kinetic term T is of the form of a non-linear sigma model,
with a fairly complicated field-dependent gij . The inverse is relatively simpler, given by

1
1
1
1
90
90
45
90 f3
1
7
1

1
90

90
45
90 f3
g ij =
(40)
.
1
1
1
1

90
90

f
3
72
18
V=

1
90
f3

7
90 f3

1
18
f3

3 4
45
b +

13 2
45 f3

176

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

If the integration constant is taken to be zero, 5 we find that the potential can be
expressed in terms of a superpotential W , i.e., V = 12 g ij i W j W , with


15
W = H 4 a 2 b1 a 2 + 4b2 b4 f32 .
(41)
4
From this we can obtain first-order equations, given by
a 2 b4 4b6 2a 2 f32
2a 2b4 + (a 2 4b2)f32

,
b
,
=
8b4 K
8ab3K
H (a 2 + 4b2 )f32
f3 ((12b2 a 2 )b4 f32 (a 2 + 20b2))

H =
(42)
,
f
,
=
3
8ab4K
8ab4K

where K b4 f32 and a prime denotes a derivative with respect to the original
coordinate appearing in (31). Note that these first-order equations again imply an algebraic
relation among the functions, analogous to (20). This time, we have
a =

a 3 bH 6f3 = .

(43)

The equations can be solved by defining new quantities a H a, b H b and f3


After manipulations analogous to those in Section 2, we arrive at the general
solution
H 2 f3 .

4
2
= H 2 dx dx + 12H 4U 1 dr 2 + r 2 H 2 U i2 + r 2 H 2 i2 ,
ds11
3
where

1/6
c2
/3
H = 1 12 3
.
U 1 3,
r
r U

(44)

(45)

The function f3 is given by


f3 =

c
r 4 H 2 U 3/2

(46)

Note that the constant in (43) is related to c by = 8c/(3 3 ).


Thus we see that the general solution has two non-trivial integration constants, /
and c. The constant / measures the scale size of the the gravitional instanton of the
G2 manifold, whilst the constant c measures the contribution from the 4-form field
strength. Again, the solution is massless and carries no charge, and it can be thought
of as a gravitional monopole involving the supergravity multiplet fields. Asymptotically,
the solution a becomes a product of four-dimensional Minkowski spacetime and the G2
5 If the integration constant were non-vanishing, which would correspond to a configuration including M5branes wrapped on 3-cycles in S 3 S 3 , it is not clear how one would solve the second-order equations. Similar
remarks apply to the previous case in Section 2 also. We chose to omit an analogous constant of integration, in the
discussion above (6), in order to obtain a formulation of the second-order equations in terms of a superpotential
and hence a gradient flow. Had we retained the constant of integration, which would give a non-vanishing flux
(4) for F(4) corresponding to M5-branes wrapping on 2-cycles in CP3 or SU(3)/(U (1) U (1)), it is again
not clear how one would solve the second-order equations. We thank S.S. Gubser for raising this question about
our procedure for obtaining gradient flows.

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

177

manifold with R4 S 3 topology, since the contribution to the metric is dominated at large
r by the instanton contribution /4 /r 4 , in comparison to the F(4) contribution which is of
order c2 /r 12 . At small distance, the solution has a naked singularity.

4. Supersymmetry of the M3-branes


Since the configurations that we have obtained in the previous two sections arise as
the solutions of first-order systems of equations, it is natural to expect that they should
be supersymmetric. In other words, one would expect that the first-order equations would
have the interpretation of being precisely the integrability conditions for supersymmetry.
However, since they were not obtained by explicitly requiring supersymmetry, but rather by
finding a superpotential for the Lagrangian formulation of the original bosonic supergravity
equations of motion, the question of supersymmetry remains to be investigated.
4.1. Solutions in the R4 bundle over S 3 background
First, we shall study the supersymmetry for the solutions obtained in Section 3, where the
background metric in the transverse space is the R4 bundle over S 3 . It is a straightforward
matter to calculate the spin connection for the metric (31) directly in eleven dimensions,
and then to substitute this and the field strength ansatz (32) directly into the gravitino
transformation rule
1
1
FBCDE A BCDE  + FABCD BCD .

288
36
We make a standard 4 + 7 decomposition of the Dirac matrices, as follows:
A = DA 

 = 1,

a = 5 a .

(47)

(48)

Substituting into (47), and examining first the world-volume directions , we find that a
Killing spinor of the form  =  must satisfy 5  = . For the case of 5  = + we
find

1/2
1/2

= g b2 + f3
(49)
1 + g b2 f3
2 ,
where 1 and 2 are constant spinors in the transverse 7-space, satisfying the projection
conditions
(1 + i4 )2 = 0,

23 1 = 042 ,

(26 35 )1 = 2i04 2 .

(50)

Note that these conditions uniquely determine 1 and 2 , up to an overall scale. For the
case where 5  = , the associated spinor is given again by (49), but with f3 replaced
by f3 . The dependence of the overall function g on the coordinates of the transverse
space is undetermined by the components of A = 0.
The components of A = 0 lying in the directions A = a of the transverse space
will now determine the dependence of g on the transverse coordinates. It is easiest
first to examine the radial direction (i.e., the 0 direction), which determines the radial

178

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

dependence of the Killing spinor. Then, by looking at the remaining transverse directions,
we find that the Killing spinor has no dependence on the angular coordinates of the two
3-spheres. The conclusion is that the function g in (49) is given by
g = b1 H 1/2.

(51)

Thus the first-order equations (37) (where f1 and f2 are given by (42) with = 0) are
precisely the integrability conditions for the existence of a spinor  satisfying A = 0 in
(47). Since we have 2 solutions (corresponding to two spinors  in the M3-brane worldvolume) for each of the cases 5  = , the general Killing spinor has four real solutions,
corresponding to N = 1 supersymmetry on the world-volume of the M3-brane. Of course,
if the constant c is set to zero, so that the 4-form is turned off, these Killing spinors reduce
to the usual ones in the product of four-dimensional Minkowski spacetime and the Ricciflat metric of G2 holonomy.
4.2. Solutions in the R3 bundle over S 4 background
Here, we repeat the analysis of the supersymmetry transformations in the case of the R3
bundle over S 4 background, for which the M3-brane solution was constructed in Section 2.
Again we begin by considering the components of A = 0 lying in the world-volume
of the 3-brane. From = 0 we deduce that a Killing spinor of the form  =  will
be given by 5  = , and for the case 5  = + we have


= gP (K f )1/2 1 + (K + f )1/2 2 ,
(52)
where K is given by (19), P is given by


 
sin 12 
1
1
1
1
K
cos

if
sin

+
K
cos

+
if
sin

02
01
2
2
2
2
2ab2


+ cos 12 cos 12 + sin 12 12 ,

(53)

and the constant 8-component spinors 1 and 2 are uniquely specified (up to scale) by the
projections
(0 3456)2 = 0,

(34 + 56 2012)2 = 0,

41 = (135 146 + 236 + 245 )2 .

(54)

Here the explicit indices 1 and 2 on the Dirac matrices refer to the two directions on the S 2
fibres, as in (3), whilst the indices 3, 4, 5 and 6 refer to the directions in the S 4 base. The
index 0 refers to the radial direction. For the case when 5  = , the associated spinor
in the transverse space will be given again by (52), but now with f sent to f in (52) and
in (53).
The dependence of the overall prefactor g on the coordinates of the transverse space
is not determined by = 0 in the world-volume directions. (We have, however, made
a convenient choice of and dependent overall factors, in anticipation of subsequent
results.)

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

179

From the radial component 0 = 0 in the transverse space, we can again determine the
radial dependence of the function g, finding
g=

g
,
H ba 1/2

(55)

where g depends only on the angular coordinates of the transverse space. Examination of
A = 0 in the S 2 directions then implies that g is independent of these two coordinates.
Finally, the components in the S 4 directions determine that g has the dependence associated
with the singlet fermion zero-mode in the YangMills instanton background (as in [28,29]).
We have seen that again, the first-order system of equations for this 3-brane in the
background of the R3 bundle over S 4 have turned out to be precisely the integrability
conditions for the existence of a Killing spinor. There are in total four real solutions,
implying N = 1 supersymmetry on the world-volume of the M3-brane. As in the previous
example, if the field strength in the solution is taken to zero, by setting the constant
c = 0, the Killing spinor reduces to the standard one in the vacuum of four-dimensional
Minkowski spacetime times the Ricci-flat metric on the R3 bundle over S 4 .

5. Dual formulations and phase transitions


In a standard massive BPS p-brane solution, the charge Q arises as a constant prefactor
in the field strength supporting the solution, and Q also appears linearly in the harmonic
function H in the p-brane metric. By contrast, in our massless M3-brane solutions the
analogous constant c that arises as the prefactor in the expressions for the 4-form field
appears quadratically in the metrics (25) and (44). This means that the metrics would
continue to be real if we were to send c ic. The same would also be true of the reduced
metrics in D = 7 that formed the starting-points of our derivations in Sections 2 and 3.
Of course sending c ic would imply that the 4-form field strength would become
imaginary. However, it should be recalled that our original 7-dimensional starting point was
in the Euclidean-signatured theory obtained by dimensional reduction on the world-volume
of the M3-brane. In Euclidean signature, if the 4-form field strength F(4) is dualised to a
3-form F(3) , then its kinetic term in the D = 7 Lagrangian will undergo the replacement
1 8/5 2
1
2
e
(56)
F(4) + e 8/5 F(3)
.
48
12
This change of sign of the kinetic term, which is generic to all dualisations in Euclidean
signature, indicates that we could achieve the same effect as sending
c ic by instead

1 8/5 2
F(3) kinetic term
using a real 3-form field in D = 7, but with the canonical 12 e
instead of the sign-reversed one in (56) that arose by dualising the 4-form.
The upshot of the above discussion is that we can obtain real solutions in D = 7 that are
just like those in Sections 2 and 3, but for the opposite sign of c2 . These will be solutions
of the equations coming from the seven-dimensional Lagrangian

7 = R 1 ()2 1 e 8/5 F 2 .
e1 L
(3)
2
12

(57)

180

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

The expression for F(3) for each solution will be given by F(3) = iF(4) , where F(4) is
the corresponding expression given in Section 2 or 3. The i factor in this relation between
F(3) and F(4) is precisely removed by the i factor that we acquire upon sending c ic.
A difference now arises when we consider the higher-dimensional origin of the sevendimensional Lagrangian. We viewed (8) in Sections 2 and 3 as coming from the Kaluza
Klein reduction of D = 11 supergravity on the world-volume of the M3-brane. Instead, we
should now view (57) as coming from the KaluzaKlein reduction of type IIA, type IIB or
type I supergravity on the world-volume of a 2-brane. In other words, we obtain the 3-form
in D = 7 as the direct world-volume reduction of a 3-form in D = 10. Accordingly, we
can then lift the D = 7 solutions of Sections 2 and 3, after sending c ic, to real solutions
of ten-dimensional supergravity. Thus there is a phase transition from one type of brane to
another, when we change the modulus parameter c of the solution from real to imaginary.
For the case of the S 2 bundle over S 4 in Section 2, we find that the corresponding
massless 2-brane solution in D = 10 is given by
1
2
= H 3/2 dx dx + 2H 9/2U 1 dr 2 + r 2 H 3/2U Di Di
ds10
2
+ r 2 H 3/2 d42 ,

(58)

where
U =1

/4
,
r4


H = 1

c2 1/6
.
2r 12 U 2

(59)

For the S 3 bundle over S 3 of Section 3, the corresponding massless 2-brane solution in
D = 10 is given by
4
2
= H 3/2 dx dx + 12H 13/2U 1 dr 2 + r 2 H 1/2 U i2 + r 2 H 1/2i2 ,
ds10
3
where
1/6

c2
/3
H = 1 + 12 3
.
U 1 3,
r
r U

(60)

(61)

We have written the solutions that come from reducing the NSNS 3-form of the tendimensional supergravity. Of course in the case of type IIB we could instead use the RR
3-form, in which case the lifted solutions in D = 10 would simply be the S-duals of those
we have just presented.
One can also, of course, further lift the above configurations, if viewed as solutions of
type IIA supergravity, to D = 11. For the case corresponding to the S 2 bundle over S 4 we
then find
1
2
ds11
= H 2 dx dx + H 4 dz2 + 2H 5 U 1 dr 2 + r 2 H U Di Di
2
+ r 2 H d42 ,

(62)

where z is the eleventh coordinate, and U and H are again given by (59). For the case
corresponding to the S 3 bundle over S 3 , we find

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

4
2
ds11
= H 2 dx dx + H 4 dz2 + 12H 6U 1 dr 2 + r 2 U i2 + r 2 i2 ,
3
where U and H are given by (61).

181

(63)

6. D = 7 3-brane and S 2 -wrapped M5-brane


It is of interest also to study more general 3-brane configurations in M-theory. Another
natural candidate is an M5-brane wrapped around a supersymmetric 2-cycle. M5-branes
wrapped on supersymmetric cycles have been discussed previously [30,31]. Typically, they
admit solutions of the form AdSd H7d , where Hn denotes the n-dimensional hyperbolic
space. In this section, we shall consider an M5-brane wrapped around a 2-sphere. The
solution can be obtained by looking first at gauged supergravity in D = 7.
6.1. D = 7 AdS7 3-brane
Consider the D = 7, N = 2 gauged supergravity, whose bosonic Lagrangian is
1 4
1
L7 = R1 d d U 1 e 10 F(4) F(4)
2
2
1 2
1 i
1
i
e 10 F(2) F(2) + F(2) F(2)
A(3) gF(4) A(3) .
(64)
2
2
2 2
where F(4) = dA(3) and U is the scalar potential in the D = 7 gauged supergravity,


8
3
2
2 1 10
10
10
e
.
U =g
(65)
2e
2e
4
In addition, the 4-form satisfies the first-order odd-dimensional self-duality equation
e

4
10

1
1
F(4) = gA(3) + (3) .
2
2

(66)
j

i
16 gij k Ai(1) A(1) Ak(1) . Domain wall and AdS7 black
Here, we have (3) Ai(1) F(2)
hole solutions in this theory have been constructed [3234], which can be viewed after
lifting back to M-theory as distributed or rotating M5-branes, respectively.
Here we consider a 3-brane configuration, which is supported by one component of the
SU(2) YangMills gauge fields. We take the ansatz to be


ds72 = e2A dx dx + e2B dr 2 + d22 ,
(67)
3
= 2 ,
F(2)

F(4) = 0.

(68)

The resulting equations of motion can be derived from the Lagrangian L = T V , where
1
T = 12A 2 16A B 2B 2 + 2 ,
2


1 2 2B+ 2
8A+2B
2B
10
V =e
(69)
2e U e
,
2
together with the constraint T + V = 0. Here the dot denotes a derivative with respect to
, defined by d = e8A+2B dr.

182

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

Fig. 1. The flow defined by the 2-vector (b , f  ). The abscissa is b and the ordinate is f .

We find that V can be derived from a superpotential W , provided that g = 1. It is given


by
1 4A+ 1

1
4A+2B 1
4A+2B+ 4
10 +
10 + ge
10 .
2g e
W = 2 2ge
2

(70)

The associated first-order equations, after setting g = 1 without loss of generality, are given
by
4b2
a
= b 2f 4
+ 2f,
a
f

b
4b2
= b2f 4
8f,
b
f

f
4b2
= 4b2 f 4
+ 2f,
f
f
1

(71)

where a = eA , b = eB , f = e 10 , and a prime here denotes a derivative with respect to ,


which is defined by d = 1 eB dr.
10 2
It is not clear how to solve these first-order equations analytically, but the general
behaviour of the solutions to the gradient flow can nevertheless be analysed in terms of
a phase-plane diagram. From (71), we can plot the 2-dimensional vector (b , f  ), and it
shows that the solution flows from (b , f 1) to (b 0, f ). (See Fig. 1.
Note that f is always non-negative.)
It suffices to analyse the solution in the regions (b , f 1) and (b 0, f ).
Consider first the behaviour when f 1. In this case, the approximate form of the solution
is

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

183

6 181
4 11
1
1
+ +
+ ,
b2
+
+ ,
10 5
30
10 5
15


638
+ 40 log 1120 3 log + ,
f 1 2 + 2
(72)
21

up to the first few orders in . It is easy to see that r 0, and so the metric
approaches
a2


1
(73)
dx dx + d22 + dr 2 .
2
r
Since r tends to zero here, this describes the large-distance asymptotic region.
Now consider the behaviour when f , with b approaching zero. In this case, we
have
ds72

1
f (1 )2/7 ,
2
with tending to 1 from below. The metric then has the form


ds72 = (r r0 )2/15 dx dx + d22 + dr 2 .
a 2 (1 )1/7 b2 ,

(74)

(75)

Since r r0 in this case, it clearly corresponds to the region at small proper distance. The
solution at r = 0 would be singular, but it is also a horizon
We can dimensionally reduce the solution on d22 , to obtain a domain wall in D = 5, of
the form


ds52 = e2C dx dx + dr 2 .
(76)
The radial coordinate r runs from r = 0, which is the AdS5 horizon, to r = r0 , which is a
null singularity. The conformal factor in these two regions is given by
r 0:
r r0 :

e2C

1
,
r 10/3

e2C (r r0 )2/9 ,

35
,
4r 2
V

1
.
12(r r0 )2

(77)

Thus the system has a discrete spectrum, indicating confinement.


6.2. Lifting to S 2 -wrapped M5-brane
The consistent S 4 reduction of eleven-dimensional supergravity was obtained in [35
37]. Using the explicit reduction ansatz given in [36], we can lift the above solution to give
an S 2 -wrapped M5-brane in D = 11, with



2
= 1/3 a 2 dx dx + b2 dr 2 + d22 + 2f 1/3 1/3 d 2
ds11


1
22 ,
+ 2/3 f cos2 2 + d
2


1
1
(2) .
(2) (2) + cos2 1 f 4
A(3) = sin
(78)
2
2

184

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

Here
= f 4 sin2 + f 1 cos2 ,

(2).
d = (2) +

In the asymptotic region at large proper distance, the metric becomes





1
1
2
22 ,
2 dx dx + d22 + dr 2 + 2d 2 + cos2 2 + d
ds11
2
r
where r 0. Note that the 4-form field strength F(4) = dA(3) has a term
1
2 + ,
F(4) = 1 f 4 cos3 ( ) d
2 2
implying that this 3-brane configuration has non-vanishing M5-brane charge.

(79)

(80)

(81)

6.3. General solutions


Although we have not obtained the general solution explicitly, we can nevertheless
show that the first-order equations (71) can be reduced to a single non-linear first-order
differential equation. Defining X b/f , Y bf 4 , and dt = 5f d, we have

Y
a 1 
X
= XY 4X2 + 2 ,
= XY 2,
= 3XY 4X2 .
(82)
a
5
X
Y
The first equation gives a, once X and Y have been found using the remaining equations.
The second equation may be solved for Y , and substituted into the third. This gives
XX + X + 4X3 X + 10XX + 8X4 + 12X2 = 0.
2

(83)

Now let v X , so that X = v  = (dv/dX)(dX/d) = v dv/dX, and hence (83) becomes
vX

dv
+ v 2 + 4vX3 + 10vX + 8X4 + 12X2 = 0.
dX

(84)

A further change of variable from v to w, defined by w 12 vX, then gives


dw
+ 2wX2 + 5w + 2X4 + 3X2 = 0.
dX
Finally, we let z X2 + 5/2. This transforms (85) into
X1 w

(85)

1
dw
+ zw + (z 1)(2z 5) = 0.
(86)
dz
2
This is a particular case of Abels equation, but unfortunately it appears to be difficult to
obtain the solution in closed form.
w

6.4. Non-abelian solutions in D = 7 supergravity


So far we have made use only of a U (1) subgroup of the SU(2) gauge fiends. It is
possible also to turn on the full SU(2) gauge fields, with the ansatz
A1(1) = v sin d,

A2(1) = v d,

A3(1) = cos d,

(87)

where v is a function of r, and (, ) are the coordinates on the 2-spheres foliating the
transverse 3-space. (This ansatz was used, for example, in [38].) The Hamiltonian H =

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

185

T + V for this case is given by


1 2
2B+ 2  2
2
2
10 v ,
T = 12A 16A B  2B  +  + e
2


1 2B+ 2 2
10 (v 1)2 ,
V = e8A+2B 2 e2B U e
2

(88)

where U is the scalar potential in D = 7 gauged supergravity, as given in (65).


We have not found a superpotential for this system. The earlier U (1) result corresponds
to v = 0. There is a singular scaling limit in which the first two terms in the scalar potential
U in (65) vanish, and then the theory can be viewed as the S 3 reduction of N = 1, D = 10
supergravity [39]. A supersymmetric 3-brane with SU(2) YangMills fields does then exist,
and it is non-singular [40]. The solution is the lifting to D = 10 of the SU(2) black hole
constructed in [38]. The superpotential for this system was obtained in [27].

7. Conclusions
In the context of four-dimensional field theories, it is of considerable interest to construct
3-brane configurations in M-theory. One class of such solutions has been obtained by
wrapping M5-branes on a certain supersymmetric two cycles, such as Riemannian surfaces
[30]. These solutions are deformations of an AdS5 H 2 S 4 vacuum, where H 2 is the
hyperbolic plane.
In this paper we have constructed two new types of 3-brane configuration. In the first
type, we exploit the fact that the transverse space of the 3-brane in D = 11 is sevendimensional, and that there exist non-trivial seven-dimensional Ricci-flat manifolds with
G2 holonomy. It has been proposed that compactifications of M-theory on G2 are related to
N = 1, D = 4 YangMills theory [810]. Three explicit complete non-compact manifolds
of G2 holonomy are currently known [16,17], and they can be used to smooth out the
singularities of compact G2 orbifolds. Each of them admits a harmonic 4-form [18,
19], which suggests the possibility of turning on the 4-form field strength of D = 11
supergravity in solutions that correspond to deformations of four-dimensional Minkowski
space times a Ricci-flat G2 manifold. We have indeed managed to obtain exact solutions
of this type, which can be viewed as 3-branes in M-theory.
The general solutions contain two continuous parameters; /, which measures the size of
the gravitional instanton, and c, which measures the strength of the 4-form. The solutions
are massless, and carry no 4-form charge. In common with all other known massless brane
solutions, there are naked singularities at short distance. At large distance the solution
approaches the product of four-dimensional Minkowski spacetime and the original Ricciflat G2 manifold. The solution can be viewed as a supergravitional monopole, involving
both the metric and the 4-form in the supergravity multiplet.
We obtained the solutions by deriving first-order equations from a superpotential, and
we showed that these are precisely the integrability conditions for the existence of a
Killing spinor. Thus the two M3-brane solutions that we have constructed in this paper
are supersymmetric.

186

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

It is interesting to observe that although the harmonic 4-forms in the undeformed


manifolds of G2 holonomy, on the R3 bundle over S 4 and the R4 bundle over S 3 , have
quite different properties (the former being L2 normalisable whilst the latter is not), the
corresponding deformed 3-brane solutions in Sections 2 and 3 have very similar qualitative
behaviour. In contrast, the properties of the metrics for the fractional D2-brane [19] and
NSNS 2-brane [18], which make use of these same two Ricci-flat metrics, are significantly
different.
We also obtained M3-brane solutions of a different kind, by lifting 3-brane solutions in
D = 7 gauged supergravity back to D = 11. They carry magnetic 4-form charge, and can
be viewed as M5-branes wrapped on S 2 .

Note added
In an earlier version of this paper it was claimed that the M3-brane solutions were not
supersymmetric, but instead were pseudo-supersymmetric with respect to a modified
D = 11 supersymmetry transformation rule. This incorrect conclusion resulted from a
systematic error in a computer program that we used for calculating the Killing spinors.
We are grateful to Jim Liu for calculations that encouraged us to recheck the computer
programs and discover the error.

Acknowledgements
We are grateful to Mike Duff, Gary Gibbons, James Liu and Jianxin Lu for useful
discussions. C.N.P. is grateful to the Michigan Center for Theoretical Physics for
hospitality during the completion of this work.

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/980215.
[4] M.J. Duff, H. L, C.N. Pope, E. Sezgin, Supermembranes with fewer supersymmetries, Phys.
Lett. B 371 (1996) 206, hep-th/9511162.
[5] I.R. Klebanov, E. Witten, Superconformal field theory on the threebranes at a CalabiYau
singularity, Nucl. Phys. B 536 (1998) 199, hep-th/9807080.
[6] I.R. Klebanov, A.A. Tseytlin, Gravity duals of supersymmetric SU(N) SU(N + m) gauge
theories, Nucl. Phys. B 578 (2000) 123, hep-th/0002159.
[7] I.R. Klebanov, M.J. Strassler, Supergravity and a confining gauge theory: duality cascades and
(chi)SB-resolution of naked singularities, JHEP 0008 (2000) 052, hep-th/0007191.
[8] B.S. Acharya, On realising N = 1 super-YangMills in M-theory, hep-th/0011089.

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

[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]

187

M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large n duality, hep-th/0011256.


E. Witten, Talk presented at the Santa Barbara David Fest.
J. Gomis, D-branes, holonomy and M-theory, hep-th/0103115.
J.D. Edelstein, C. Nunez, D6-branes and M-theory geometrical transitions from gauged
supergravity, JHEP 0104 (2001) 028, hep-th/0103167.
S. Kachru, J. McGreevy, M-theory on manifolds of G2 holonomy and type IIA orientifolds,
hep-th/0103223.
J. Gutowski, G. Papadopoulos, Moduli spaces and brane solitons for M-theory compactifications on holonomy G2 manifolds, hep-th/0104105.
P. Kaste, A. Kehagias, H. Partouche, Phases of supersymmetric gauge theories from M-theory
on G2 manifolds, hep-th/0104124.
R.L. Bryant, S. Salamon, On the construction of some complete metrics with exceptional
holonomy, Duke Math. J. 58 (1989) 829.
G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S 3 , R3 and R4 bundles, Commun.
Math. Phys. 127 (1990) 529.
M. Cvetic, H. L, C.N. Pope, Brane resolution through transgression, Nucl. Phys. B 600 (2001)
103, hep-th/0011023.
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Supersymmetric non-singular fractional D2-branes
and NSNS 2-branes, hep-th/0101096.
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Hyper-Khler Calabi metrics, L2 harmonic forms,
resolved M2-branes and AdS4 /CFT 3 correspondence, hep-th/0102185.
K. Behrndt, About a class of exact string backgrounds, Nucl. Phys. B 455 (1995) 188, hepth/9506106.
R. Kallosh, A. Linde, Exact supersymmetric massive and massless white holes, Phys. Rev. D 52
(1995) 7137, hep-th/9507022.
M. Cvetic, D. Youm, Singular BPS saturated states and enhanced symmetries of fourdimensional N = 4 supersymmetric string vacua, Phys. Lett. B 359 (1995) 87, hep-th/9507160.
M. Cvetic, D. Youm, All the static spherically symmetric black holes of heterotic string on a
six-torus, Nucl. Phys. B 472 (1996) 249, hep-th/9512127.
M.J. Duff, H. L, C.N. Pope, Heterotic phase transitions and singularities of the gauge dyonic
string, Phys. Lett. B 378 (1996) 101, hep-th/9603037.
C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulsion singularities,
Phys. Rev. D 61 (2000) 086001, hep-th/9911161.
G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2sphere, Class. Quantum Grav. 18 (2001) 1333, hep-th/0012034.
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, Ricci-flat metrics, harmonic forms and brane
resolutions, hep-th/0012011.
M. Cvetic, G.W. Gibbons, H. L, C.N. Pope, New complete non-compact Spin(7) manifolds,
hep-th/0103155.
J. Maldacena, C. Nunez, Supergravity description of field theories on curved manifolds and a
no-go theorem, Int. J. Mod. Phys. A 16 (2001) 822, hep-th/0007018.
J.P. Gauntlett, N. Kim, D. Waldram, M-fivebranes wrapped on supersymmetric cycles, hepth/0012195.
M. Cvetic, M.J. Duff, P. Hoxha, J.T. Liu, H. L, J.X. Lu, R. Martinez-Acosta, C.N. Pope,
H. Sati, T.A. Tran, Embedding AdS black holes in ten and eleven dimensions, Nucl. Phys.
B 558 (1999) 96, hep-th/9903214.
J.T. Liu, R. Minasian, Black holes and membranes in AdS7 , Phys. Lett. B 457 (1999) 39, hepth/9903269.
M. Cvetic, S.S. Gubser, H. L, 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.

188

M. Cvetic et al. / Nuclear Physics B 613 (2001) 167188

[35] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11d


supergravity on AdS7 S 4 and self-duality in odd dimensions, Phys. Lett. B 469 (1999) 96,
hep-th/9905075.
[36] H. L, C.N. Pope, Exact embedding of N = 1, D = 7 gauged supergravity in D = 11, Phys.
Lett. B 467 (1999) 67, hep-th/9906168.
[37] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS(7) S(4) reduction and
the origin of self-duality in odd dimensions, Nucl. Phys. B 581 (2000) 179, hep-th/9911238.
[38] A.H. Chamseddine, M.S. Volkov, Non-Abelian BPS monopoles in N = 4 gauged supergravity,
Phys. Rev. Lett. 79 (1997) 3343, hep-th/9707176.
[39] A.H. Chamseddine, W.A. Sabra, D = 7 SU(2) gauged supergravity from D = 10 supergravity,
Phys. Lett. B 476 (2000) 415, hep-th/9911180.
[40] J.M. Maldacena, C. Nunez, Towards the large n limit of pure N = 1 super-YangMills, Phys.
Rev. Lett. 86 (2001) 588, hep-th/0008001.

Nuclear Physics B 613 (2001) 189217


www.elsevier.com/locate/npe

Four-dimensional supergravities from


five-dimensional brane worlds
Adam Falkowski a , Zygmunt Lalak a,b , Stefan Pokorski a
a Institute of Theoretical Physics University of Warsaw, Poland
b Theory Division, CERN CH-1211 Geneva 23, Switzerland

Received 14 March 2001; accepted 17 July 2001

Abstract
We give the explicit form of the four-dimensional effective supergravity action, which describes
low-energy physics of the RandallSundrum model with moduli fields in the bulk and charged chiral
matter living on the branes. The relation between 5d and 4d physics is explicit: the low-energy action
is derived from the compactification of a locally supersymmetric model in five dimensions. The
presence of odd Z2 parity scalars in the bulk gives rise to effective potential for the radion in four
dimensions. We describe the mechanism of supersymmetry breaking mediation, which relies on nontrivial configuration of these Z2 -odd bulk fields. Broken supersymmetry leads to stabilization of the
interbrane distance. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
There is a growing theoretical evidence that brane models with warped geometries
may play a prominent role in understanding the variety of field theoretical incarnations
of the hierarchy problem [13]. Nonetheless, the initial hope that the mere presence of
extra dimensions would be a natural tool to control mass scales in gauge theories coupled
to gravity turned out to be premature. Various hierarchy issues have resisted numerous
proposals, and revealed another, slightly disguised, face of the fine-tuning, well-known
from four dimensions (an example is the cosmological constant problem, see [4] and
references therein). The most stumbling observation is that whenever one finds a flat 4d
foliation as the solution of higher-dimensional Einstein equations, which seems to be
necessary for the existence of a realistic 4d effective theory, it is accompanied by a special
choice of various parameters in the higher-dimensional Lagrangian. The fine-tuning seems
to be even worse in 5d than in 4d, since typically one must correlate parameters living
E-mail address: zygmunt.lalak@cern.ch (Z. Lalak).
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 1 ) 0 0 3 7 6 - 5

190

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

on spatially separated branes. Then there appears immediately the problem of stabilizing
these special relations against quantum corrections.
This situation has prompted the proposal [57], that it is a version of branebulk
supersymmetry that may be able to explain apparent fine-tunings and stabilize hierarchies
against quantum corrections. And indeed, the branebulk supersymmetry turns out
to correlate in the right way the brane tensions and bulk cosmological constant in
the supersymmetric RandallSundrum model. In addition, supergravity is likely to be
necessary to embed brane worlds in string theory.
Hence, there are good reasons to believe that supersymmetry is an important ingredient
of the higher-dimensional unification and the quest for consistent supersymmetric versions
of brane worlds goes on, see [59,1117,1925]. In earlier papers [1215,26] the attention
has been focused on models with solitonic (thick) branes and several no-go theorems were
established (but see [17,19,20,24]). Finally, in papers [5,7,9,11] explicit supersymmetric
models with delta-type (thin) branes were constructed.
The distinguishing feature of the pure supergravity Lagrangians proposed in [7] is
imposing the Z2 symmetry, such that gravitino masses are Z2 -odd. An elegant formulation
of the model is given in Ref. [11], where additional non-propagating fields are introduced
to independently supersymmetrize the branes and the bulk. In the on-shell picture for these
fields the models of Ref. [11] and Refs. [7,9] are the same.
In Ref. [11] an extension of the model to include vector multiplets has been worked out.
On the other hand, in Refs. [7,9] it has been noted that supersymmetric RandallSundrumtype models can be generalized to include the universal hypermultiplet and gauge fields and
matter on the branes. The Lagrangian of such a construction has been given in [7,9]. This
opens up a phenomenological avenue, which we follow in the present paper, and allows us
to study issues such as supersymmetry breaking and its transmission through the bulk. We
want to stress that it is impossible to perform a trustworthy research of these issues without
having a complete, explicit, locally supersymmetric model including matter on the branes
embedded in extra dimensions. This is the main drawback of the phenomenological studies
of supersymmetric brane worlds published so far. Further, it is our opinion that one should
study thoroughly the classic, and in a sense minimal, version of the 5d brane worlds where
charged matter and observable gauge interactions are confined to branes. The point is that
putting matter and gauge fields into the 5d bulk amounts extending in a non-trivial way the
very attractive MSSM. Hence, for the time being we prefer to clarify the situation in the
minimal models with just neutral fields living in the bulk. On the other hand, populating
the bulk with 5d gravitational fields alone does not seem to be realistic. In typical string
compactifications, just on the basis of simple dimensional reduction of higher-dimensional
supergravities, one expects neutral (with respect to the SM group) matter, namely moduli
fields, to coexist in the 5d bulk with supergravity multiplet. To represent such bulk matter,
we choose to work in the present paper with bulk hypermultiplets, which couple also to
the branes. In the explicit calculations which shall lead us to a consistent 4d supergravity
model in the final section of this paper, we shall employ a universal hypermultiplet, which
reduces to the dilatonic chiral multiplet in 4d.

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

191

Still, it should be clear that putting gauge fields (and more) in the bulk is a viable
alternative to our models, and as shown in [23] it may also provide a mechanism to stabilize
the extra dimensions.
The final goal of this paper is to formulate the effective low-energy theory that
describes properly the physics of the warped five-dimensional models with gauge sectors
on the branes. On the way to four-dimensional theory we investigate supersymmetry,
supersymmetry breakdown and moduli stabilization using five-dimensional tools. In
particular, we show that in the class of models that we consider, i.e., models without nontrivial gauge sectors in the bulk, unbroken N = 1 local supersymmetry (classical solutions
with four unbroken supercharges) implies vanishing of the effective cosmological constant.
We demonstrate the link between vanishing of the 4d cosmological constant, minimization
of effective potentials in 5d and 4d, and moduli stabilization in five-dimensional
supersymmetric models presented in this paper. We also discuss supersymmetry breaking
due to a global obstruction against the extension of bulk Killing spinors to the branes,
which is a phenomenon observed earlier in the HoravaWitten model in 11d and 5d.
First steps towards the 4d effective theory were made in [22,23] (where the Khler
function for the radion field was identified). In the set-up considered in this paper,
supersymmetry in 5d is first broken from eight down to four supercharges by the BPS
vacuum wall, and then again broken spontaneously down to N = 0 by a switching on
of expectation values of sources living on the branes. The general strategy follows the one
[2730] that led to the complete and accurate description of the low-energy supersymmetry
breakdown in the HoravaWitten models, see [27,29]. We are able to find maximally
symmetric solutions to the 5d equations of motion within our supersymmetric model,
and deduce the Khler potential, superpotential and gauge kinetic functions describing
physics of corresponding vacua in four dimensions. It turns out that the warped background
modifies in an interesting way the kinetic terms for matter fields and the gauge kinetic
function on the warped wall. There also appears a potential for the radion superfield, its
origin being a modulus-dependent prefactor multiplying the superpotential on the warped
wall in the expression for the 4d effective superpotential. We do not need to introduce
any non-trivial gauge sector in the bulk to generate a potential for the T modulus. It
is interesting to note that the structure of the effective 4d supergravity is completely
different from that of the no-scale models. In these the 4d cosmological constant vanishes,
while FT is undetermined and sets the supersymmetry breaking scale. In our model FT
vanishes, supersymmetry is broken by FS and non-zero cosmological constant is induced.
The complete and phenomenologically relevant 4d N = 1 supergravity model which we
managed to construct in this paper, should finally facilitate a detailed investigation of the
low-energy physics of warped compactifications.

2. Unbroken supersymmetry in the brane-world scenarios


We begin with a brief review of the original RS model. The action is that of 5d gravity
on M4 S1 /Z2 with negative cosmological constant:

192

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217


S = M3

d 5x


 

 



1
g R + 6k 2 + d 5 x gi 1 x 5 2 x 5 .
2

(1)
Three-branes of non-zero tension are located at Z2 fixed points. The ansatz for the vacuum
solution preserving 4d Poincare invariance has the warped product form:
 

2
ds 2 = a 2 x 5 dx dx + R02 dx 5 .
(2)
The breathing mode of the fifth dimension is parametrized by R0 . The solution for the warp
factor a(x 5 ) is:
 
 

a x 5 = exp R0 k x 5  .
(3)
It has an exponential form that can generate a large hierarchy of scales between the branes.
Matching delta functions in the equations of motion requires fine tuning of the brane
tensions:
1 = 2 = 6k.

(4)

With the choice (4) the matching conditions are satisfied for arbitrary R0 , so the fifth
dimension is not stabilized in the original RS model. Thus R0 enters the 4d effective theory
as a massless scalar (radion), which couples to gravity in manner of a BransDicke scalar.
This is at odds with the precision tests of general relativity, so any realistic model should
contain a potential for the radion field.
Relaxing the condition (4) we are still able to find a solution in the maximally symmetric
form, but only if we allow for non-zero 4d curvature (AdS4 or dS4 ) [31]. In such a case
radion is stabilized and its vacuum expectation value is determined by the brane tensions
and the bulk cosmological constant.
The RandallSundrum model can be extended to a locally supersymmetric model [5
7]. The basic set-up consists of 5d N = 2 gauged supergravity [32,33] which includes the
gravity multiplet (em , A , A ), that is the metric (vielbein), a pair of symplectic Majorana
gravitinos, and a vector field called the graviphoton. The U (1) subgroup of the R-symmetry
group is gauged, the gauge charge g being constant between two branes but antisymmetric
in the x 5 coordinate. The form of the gauging is fully characterized by the SU(2) valued
prepotential. Choosing the prepotential along the 3 direction:
 
3
gP = k" x 5 i 3
2 2

(5)

reproduces the bosonic part of the RS bulk action (1). Moreover, because of the
antisymmetric function "(x 5 ), the supersymmetry variation of the action contains terms
proportional to the delta function, that cancel if the brane tensions satisfy the relation (4).
Thus the fine-tuning present in the original RS model can be explained by the requirement
of local supersymmetry [7].
New bosonic and fermionic fields do not affect the vacuum solution so the equations
of motion for the warp factor are the same as in the original, non-supersymmetric RS
model. The RS solution satisfies the BPS conditions and preserves one half of the

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

193

supercharges, which corresponds to unbroken N = 1 supersymmetry in four dimensions. In


the supersymmetric version the brane tensions are fixed. As a consequence, the exponential
solution (3) is the only maximally symmetric solution and the radion is still not stabilized.
We now turn to studying the supersymmetric RS model coupled to matter fields. We
want to investigate how general are the features present in the minimal supersymmetric
RS model. We find that unbroken local supersymmetry implies flat 4d spacetime in a
wider class of 5d supergravities coupled to hyper- or vector multiplets, in which the scalar
potential is generated by gauging a subgroup of R-symmetry.
Let us consider a version of the RS model, which apart from the gravity multiplet
includes an arbitrary number of hypermultiplets. In five dimensions a hypermultiplet
consists of a pair of symplectic Majorana fermions a and of four real scalars q u . The
scalar potential can be generated by gauging a U (1) subgroup of the R-symmetry group
and at the same time, by gauging isometries of the sigma model [28,32,33]. The bosonic
action we consider is:


1

 
3
5
u v
d x e5 2 R huv D q D q V (q) d 4 x e4 1 x 5
S=M



d 4 x e4 2 x 5 ,
(6)


u v
2 1
V (q) = g 2 16
3 P + 2 huv k k .

(7)
i 1 +

The sigma-model metric huv is quaternionic. The SU(2) valued prepotential P = P1


P2 i 2 + P3 i 3 describes gauging of the R-symmetry group while the Killing spinor k u
describes gauging the isometries of the quaternionic manifold; the covariant derivative
acting on scalars is D q u = q u + gk u A . Generically, both k u and P are functions of
the hypermultiplet scalars and satisfy the Killing prepotential equation k u Kuw = w P +
[w , P], where is the spin connection and K is the Khler form of the quaternionic
manifold. As usually, we assume that the gauge charge g is odd: g 1 6k"(x 5). The
2
brane tensions are:
1 = 2 = 24kP3

(8)

as was pointed out in [9]. This relation is necessary to cancel the variation of the action
which arise because of the presence of "(x 5 ) in the gauge coupling and is an equivalent
of (4) in a model with hypermultiplets.
The relevant part of the supersymmetry transformation laws is (we use the normalization
of [28]):

 A
2
1
A
A
ab A
gP3 3 B " B ,
= " + ab " +
4
3
1 u Aa
a
Aa
u 5
= iVu 5 q "A + g k Vu "A .
(9)
2
Our objective is to show that, in the class of warped compactification, unbroken
supersymmetry implies flat 4d spacetime. To achieve this, we will derive the effective
theory of the 4d metric degrees of freedom. It will turn out that supersymmetry requires

194

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

vanishing of the 4d effective potential. We make the ansatz:


 

2
ds 2 = a 2 x 5 g dx dx + R02 dx 5 ,
 
qu = qu x5

(10)

which describes oscillations of the 4d metric g about some vacuum solution. Using the
ansatz (10) the action (6) can be rewritten in the form which reveals the BPS structure:



2


 5 
6 a
1
3
5
4
2

+ 4kR0 " x
S=M
R+ 2
P
d x g R0 a
2a 2
R0 a




+ 9k 2 huv 4 u P 2 v P 2 k u k v

  
  
1
2 huv 5 u 6kR0 " x 5 u P 2 5 v 6kR0 " x 5 v P 2
R0
 

 5 
4 a
2

+ 3k" x
45 a
P
a




  

 

4
1
5
4
2
5
+ a 24k P 2 x + a 24k P 2 x .
(11)
To show that the effective potential vanishes when supersymmetry is preserved requires
some calculations. First, A = 0 conditions can be solved in terms of the warp factor
yielding:
 
a
= 4kR0 " x 5 P 2 .
(12)
a
Furthermore, the a = 0 condition yields:


12 2kR0 u  1a
1a
u
k Vu P3 + Vu2a (P1 iP2 ) ,
Vu 5 q = i
4 P 2


12 2kR0 u 
2a
u
Vu 5 q = i
(13)
k Vu2a P3 + Vu2a (P1 + iP2 ) .
4 P 2
Using the Killing prepotential equation it is possible to eliminate k u from (13). After some
calculations we get:

huw 5 q w = 6kR0 u P 2 .
(14)
Another manipulations of (13) yield the relation huv 5 q u 5 q v = 9R02 k 2 huv k u k v .
Summarizing, BPS conditions imply the following relations between the scalars, the
warp factor and the Killing vector and prepotential:

a
= 4kR0 P 2 ,
a

huw 5 q w = 6kR0 u P 2 ,
huv 5 q u 5 q v = 9k 2 R02 huv k u k v .

(15)

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

195

Plugging the above formulae into the action (11) and integrating over x 5 , we obtain the 4d
effective action for the metric g :



2
Veff ,
S4 = MPL
d 4 x 12 R







2
MPL
Veff = M 3 a 4 (0) 24k P 2 (0) 1 + a 4 () 24k P 2 () 2 ,

2
MPL
(16)
M 3 R0 dx 5 a 2 .
In Section 3 we explain in detail that if the P1 and/or P2 components of the prepotential
are non-zero at the Z2 fixed point then supersymmetry is broken. The reason is that
in such
a case the BPS solution is not global. Thus unbroken supersymmetry requires P 2 (0) =
P3 (0) (the same at x 5 = ) and, in consequence, the 4d effective potential vanishes (recall
that the brane tensions satisfy (8)).
Similar conclusions hold also in the RS model, with vector multiplets constructed in
Ref. [11]. The proof goes the same way as in the hypermultiplet case; the ultimate reason
for the vanishing of the effective potential being the supersymmetric tuning of the brane
tensions.
Summarizing this section, unbroken local supersymmetry in 5d RS-type scenarios
implies flat space solutions. This result is somewhat unexpected, since within the
framework of 4d supergravities we can a priori obtain an anti-de Sitter solution and
preserve supersymmetry at the same time. This means that compactifications of the
supersymmetric RS scenarios yield a very special subclass of 4d supergravities. This
should be kept in mind when phenomenological models are constructed.
Another consequence of the vanishing 4d potential is that RS-type models with
unbroken supersymmetry cannot incorporate a mechanism of radion stabilization. The
BPS conditions together with the supersymmetric tuning of the brane tensions yield the
4d effective potential which is identically zero. Thus the 4d equation of motion for the
M 2
= 0.
= 0, which can be satisfied by any value of R0 , since R
radion field is PL R
R0

3. Supersymmetry breaking and radion stabilization in a model with the universal


hypermultiplet
In the preceding section we argued that RS-type models with unbroken supersymmetry
require vanishing of the 4d cosmological constant. At the same time they cannot
incorporate a mechanism of radion stabilization. But in a realistic model supersymmetry
must be broken. Therefore in the remainder of this paper we will study dynamics of the RS
model with broken supersymmetry.
The RS scenario with spontaneously broken supersymmetry was already discussed in
Ref. [23] in the context of radion stabilization. The effective potential for the radion field
was generated through the interaction of the radion with gaugino condensates in the bulk.
Communication of supersymmetry breaking to the visible brane occurs at the level of fourdimensional physics with the help of the anomaly mediation mechanism proposed in [34].

196

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

In this paper we investigate an alternative mechanism of supersymmetry breaking,


similar to that studied in M-theoretical scenarios [28,29,35]. It is triggered by brane sources
coupled to the scalar fields in the bulk, which are odd with respect to the Z2 parity.
One way to see that supersymmetry is broken is to notice that the Killing spinor cannot
be defined globally. The odd fields are the agents that transmit supersymmetry breaking
between the hidden and visible branes. Below we present a general description of our
mechanism and then apply it to a specific model of 5d gauged supergravity with the
universal hypermultiplet.
Already in the previous section we signalled the possibility of breaking supersymmetry
by inducing non-trivial vev of the P1 and/or P2 component of the prepotential. Recall,
that the Z2 acts on the supersymmetry parameter " A as 5 " 1 (x 5 ) = " 1 (x 5 ), 5 " 2 (x 5 ) =
" 2 (x 5 ). At the Z2 fixed points half of the components are projected out (so that only
"R1 and "L2 are non-zero and they correspond to a single Majorana spinor which generates
N = 1 supersymmetry in four dimensions). At the same time, the BPS conditions impose
restrictions on the Killing spinors. If we insist on preserving N = 1 supersymmetry, the
form of the Killing spinors must be consistent with the orbifold projection; in other words,
the Killing spinors must have only Z2 even components at the fixed points.
It is straightforward to check that as long as P 3 the Killing spinors which generate
unbroken N = 1 supersymmetry has definite, even Z2 parity. But as soon as we switch on
non-zero P1 or P2 the Killing spinors satisfy the relation:

P 2 5 " 1 P3 " 1
2
" =
(17)
.
P1 iP2
The above equality implies that if P1 or P2 are non-zero at the Z2 fixed points, then the
Killing spinors have not definite parity there and supersymmetry is completely broken.
It is still possible to find a Killing spinor locally, but it cannot be defined globally.
Supersymmetry is broken because of the misalignment between the bulk and the brane
supersymmetry.
But for P1 or P2 to be non-zero a non-trivial configuration of the Z2 odd fields is
necessary. Indeed, from the form of the covariant derivative A + gVi (P i )AB A B
it is straightforward to deduce that while P3 is even with respect to Z2 , P1 and P2 must
be odd (recall that g "(x 5 )). Non-trivial configuration of the odd fields can be induced
by sources located on the branes. The mechanism involves coupling the odd scalars to
the branes through their fifth derivatives (which are even and thus well defined on the
boundary). The explicit realization of brane sources are boundary superpotential and
boundary gaugino condensates, studied in [7,10]. In Appendix B we present a general
supersymmetric form of such couplings in 5d supergravity with an arbitrary number of
hypermultiplets.
Studying supersymmetry breaking in 5d supergravity with general hypermultiplet
spectrum is a difficult task. Therefore, from now on we concentrate on a simpler model
with only one, so called universal hypermultiplet which includes two even scalars V , and
two odd scalars , (the basic properties of the universal hypermultiplet are summarized
in Appendix A). One may hope, that the features present in this toy-model persist in

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

197

more general scenarios, in which supersymmetry breaking is transmitted by a non-trivial


configuration of the odd bulk fields.
To generate a scalar potential we gauge the isometry ei of the quaternionic
manifold [9]. Solving the Killing prepotential equation we find:
Re( )
P1 = ,
2 V

Im( )
P2 = ,
2 V

P3 =



| |2
1
1
.
4
V

(18)

We see that P1 and P2 are non-zero in this model. Thus, inducing non-zero vev of the
field will break supersymmetry. To this end, we couple (in a supersymmetric way) the fifth
derivative of to sources located on the boundaries. The rest of the bosonic action we
consider is that given in (6) rewritten for our special choice of the quaternionic manifold
and gauging:


1
1
R 3
( A )2 |D |2
S = M 3 d 5 x e5
( V )2
2
4
V
4V 2


1
1
2
2
4
| |
+ 6k 1 +
| |
2V
2V 2





e5   
| |2
M 3 d 5 x 5 x 5 x 5 6k 1
V
e5




1
 W
 
2
x 5 W1 5 + 2 x 5
d 5 x e5
V g55
M3



2

W
+ (x 5 )W2 5 + 2 x 5
(19)
+
h.c.
.
M3
In the first line we displayed the relevant kinetic terms (we neglected the field which does
not play any role in the following). The
covariant derivative acting on is D = +
ig A with the gauge charge g = 3k 2 "(x 5 ). The boundary terms in the second line are
proportional to P3 and are necessary for local supersymmetry of the 5d action [9]. The third
line contains derivative coupling of the odd field to the boundary sources W ; the presence
of singular 2 terms is commented on in Appendix B. For the sake of concreteness we have
concentrated on the case, where sources of supersymmetry breakdown are represented by
expectation values of brane superpotentials (gaugino condensates will be discussed later in
the paper).
Locally in the bulk, it is possible to find a flat BPS solution which preserves
half of the supersymmetry (see Appendix A for details), but as soon as we switch
on non-zero sources W the BPS solution does not satisfy the matching conditions
at the Z2 fixed points. In such case we search for maximally symmetric, non-BPS
solutions. We take the ansatz that allows for AdS4 foliation of the 4d metric, g =
3
3
3
diag(a 2 (x 5 )e2Lx , a 2 (x 5 )e2Lx , a 2 (x 5 )e2Lx , a 2 (x 5 ), R02 ) (the generalization to the de
Sitter foliation is straightforward); the size of the 5th dimension is parametrized by R0 . Our
vacuum has the form of a constant curvature foliation (otherwise it could lead to violation
of Lorentz invariance after integrating out the extra dimensions [41]). The ansatz for the

198

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

scalar field and the graviphoton is:


 
 
= x 5 , V = V x 5 , = const,

A = 0,

A5 = const.

(20)

We start with the RS solution a = ekR0 |y| , = 0 and treat the boundary sources as a
perturbation. This procedure is justified since we expect that the parameters of the bulk
Lagrangian, which determine the zeroth-order solution, are close to the Planck scale, while
the boundary sources should be of the order of supersymmetry breaking scale. Since the
sources set the boundary value of we expect that will get (W/M 3 ) corrections. On the
other hand, the enters quadratically into the equation for the metric, so the metric will get
the correction only at the order (W/M 3 )2 . We will be able to find a solution valid to the
order (W/M 3 )2 . First we consider the linearized equation of motion for the field which
has the form:




a
+ 4 + 2igA5 + 3k 2 R0 4igkR0 A5 g 2 A25
a
1
2
W
W
= 2 (x 5 ) 3 2 (x 5 ) 3 .
(21)
M
M
The solution is:
= C"(x 5 )ek(R0 3

2i A5 )|y|

(22)

Matching the in the equation of motion yields the boundary conditions: (0+ ) =
1 /M 3 , ( ) = W
2 /M 3 . As a consequence:
W

2
1
W
W
(R0 3 2 i A5 )k
,
Ce
=
.
(23)
M3
M3
In the absence of supersymmetry breaking the moduli R0 and A5 could have arbitrary constant values. When we switch on the sources for the odd fields and break supersymmetry,
the expectation value of the moduli is determined by the boundary sources Wi . In other
words, the moduli are stabilized. Hence, in our model the supersymmetry breaking indeed
leads to stabilization of the fifth dimension. In the 4d effective theory R0 and A5 enter
the so-called T supermultiplet. In the next section we show, that the configuration of the
field which we have found gives rise to the 4d effective potential which is able to stabilize
T modulus when the 4d dilaton is frozen.
The non-zero value of of order W will produce a back-reaction on the metric and the
scalar V of order W 2 . Keeping only terms relevant to the order W 2 , the Einstein equations
read:
 2


a
1
L2
1 2
a
2
2
| |
+ 3 2 + |D5 | 6k 1 +
3 +3
a
a
V
2V
a


  


| |2
,
= x 5 x 5 6k 1
V
 2


a
L2
1 2
1
2
6
(24)
| |2 = 0.
+ 6 2 |D
5 | 6k 1 +
a
a
V
2V

C =

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

199

The terms involving boundary sources W gather nicely into the full square with the kinetic
term of . The boundary conditions (23) ensure that the hatted derivative contains no delta
functions. Thus the only boundary terms we have to consider are those displayed explicitly
in (24). Matching the delta functions in the first equation yields the boundary conditions
for the warp factor:




a
a
(25)
(0) = 6k 1 |(0)|2 ,
() = 6k 1 |()|2 .
a
a
Plugging in our solution for we can satisfy the boundary conditions if the warp factor has
the form:
 
|C|2 kR0 |x 5 |
5
e
.
a x 5 = ekR0 |x | +
2V

(26)

k |C|
Away from the branes the Einstein equations can be satisfied if we choose L2 = 16
6
V .
2
2
k |C|

, hence this solution corresponds to
This means that the 4d curvature is R = 32
2

anti-de Sitter 4d foliation.


Similarly, the scalar V will get the correction of the order of (W/M 3 )2 . The equation of
motion is:
  


4a
5 | 6(kR0 )2 | |2 = 12kR0 | |2 x 5 x 5
V + 2|D
V +
(27)
a
which is solved by V = V0 |C|2 e2kR0 |x | .
Summarizing, we have found a perturbative (to the order |W 2 |) solution to the equation
of motion in the presence of non-zero boundary sources for the odd field :
5

= C"(x 5 )ek(R0 3
a = ekR0 |x | +
5

2 ik A5 )|x 5 |

|C|2 kR0 |x 5 |
e
,
2V0

V = V0 |C|2 e2kR0 |x | ,
5

1
2
W
W
k(R0 3 2 ik A5 )
,
Ce
= 3,
3
M
M
2 |C|2
k
8
L2 =
,
3 V0

C=

= 0 .

(28)

In the above solution V0 and 0 are arbitrary constants. Hence these moduli are not
stabilized. In fact, matching conditions in the equation of motion for V require V0
. This means that V0 exhibits the runaway behaviour. To achieve stabilization of all the
moduli we need to complicate our model by adding boundary sectors that couple to V .
As a cross-check of the above results we can calculate the 4d effective potential, obtained
by integrating out the 5d bosonic action in the background (28). The result (to the order
W 2 /M 3 ) is:


M3 
 1
k 2 |C|2
2kR0

L4 = g
(29)
.
1e
R+8
k
2
V0
We denoted by g the oscillations of the 4d metric around the vacuum solution. Solving the
= 32k 2 |C|2 /V0 which is consistent
Einstein equations in the 4d effective theory yields R
1
2
with the value of L 12 R in (28). We also see that V0 enters the denominator of the
effective potential, which explains its runaway behaviour commented on earlier.

200

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

Before closing the discussion of the 5d classical solutions and supersymmetry breakdown, let us comment on proposals [17,18,36] of solving the cosmological constant problem due to supersymmetry of the bulkbrane system. To put the issue into the perspective,
let us note that the second equation in (24) does not contain second derivatives of fields,
hence it acts as a sort of constraint on the solutions of the remaining equations. This becomes more clear in the Hamiltonian approach towards the flow along the fifth dimension,
where this equation arises as the Hamiltonian constraint H = 0, and is usually used to illustrate the way the conservation of the 4d curvature L2 is achieved through the compensation
between gradient and potential terms along the classical flow. However, this classical conservation hinges upon fulfilling certain consistency conditions between brane sources, or
between boundary conditions induced by them, as illustrated by the model above. When
one perturbs the boundary terms on one wall, then to stay within the family of maximally
symmetric foliations one of two things must happen. Either the distance between branes
must change, or the source at the distant brane must be retuned. In the class of models
which we constructed, if the 4d curvature is present then supersymmetry is broken, and
does not take care of such a retuning. Furthermore, even if retuning takes place, the size
of 4d curvature, i.e., of the effective cosmological constant, does change as well; moreover, the magnitude of the effective cosmological constant has quadratic dependence on
the boundary terms which induce supersymmetry breakdown. Hence, any perturbation of
the boundary, instead of being screened by the bulk physics, contributes quadratically to
the effective cosmological constant. Of course, we are talking about perturbations that can
be considered quasi-classical on the brane. Thus we do not see here any special new effect
of the extra dimension in the cancellation of the cosmological constant. The positive aspect
of supersymmetry is exactly the one we know from 4d physics. Supersymmetry, even the
broken one, limits the size of the brane terms inducing supersymmetry breakdown, thus
limiting the magnitude of the 4d cosmological constant, since the two effects are strictly
related to each other.

4. Four-dimensional effective theory


In this section we give the form of the effective four-dimensional supergravity describing
zero-mode fluctuations in the model presented in the previous section. Since in the
5d set-up supersymmetry was broken spontaneously, it is safe to assume that in 4d
this supersymmetry breakdown can be considered as a spontaneous breakdown in a
4d supergravity Lagrangian described with the help of certain Khler potential K,
superpotential W and gauge kinetic functions H . The goal is to identify reliably these
functions starting from the maximally symmetric approximate solutions (28) we have
found in the previous section. Our procedure is perturbative in the supersymmetry breaking
parameter W1 . The configurations (28) are solving equations of motion and boundary
conditions to the second order in W1 /M 3 . However, it is sufficient to identify the functions
we are looking for from the terms which can be reliably read at the order (W1 /M 3 )1 .
Such terms include the gravitino mass term. In addition, we have at our disposal the

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

201

complete kinetic terms for moduli, gauge and matter fields, which are of order (W1 )0 and
are sufficient to read off the Khler potential for moduli and matter fields. The complete
procedure consists of solving to the given order for all background fields, including the Z2 odd ones, substituting the solutions back to the 5d Lagrangian and integrating over the fifth
dimension. This procedure can be carried out to the full extent, however here, taking the
existence of the effective 4d supergravity for granted, we shall perform the integration only
for certain relevant terms the ones which give direct information about K, W and H .
4.1. Khler function for the radion and 4d dilaton
To the lowest order in field fluctuations the ansatz defining the radion modulus in the
RandallSundrum background, which is also the lowest order solution in W1 -expansion, is
ds 2 = ekR0 (x

)|x 5 |

 
2
g dx dx + R02 x dx 5 .

This ansatz leads to the Khler potential:




3MP2 log f (T + T ) ,
K = MP2 log(S + S)

S = V0 + i0 ,
T = k(R0 + i 2 A5 ),

(30)

(31)

where we defined
MP2 =
with =


M3 
1 e2kR0  ,
k

M3
kMP2



f = 1 e(T +T )

. The form of the Khler potential for the multiplet T was previously derived

in [22,23].
To see in more detail how the argument goes, let us summarize those terms in the 5d
action that are most relevant to the forthcoming discussion:



1 A
1
3
5
5
A B
D A 3ik"(x ) PAB ,
d x e5 R
Sgrav = M
2
2


 1
1 
1 a D a
D
Shyp = M 3 d 5 x e5 2 V V + D
V
2
4V

1
5 q u ,
A B uAB D
(32)
2
where

5 = 5 + 3i 2 k"(x 5 ) A5 + 2(xi )W
i ,
D
the prepotential is given by


1 x2 + y2
x
y
P=

i 1 +
i 2
i 3
1/2
4
4V
2V
2V 1/2
and the SU(2) spin connection is given in Appendix A.

202

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

The goal is to reduce various terms in this action down to the Einstein frame in four
dimensions, where the matter-supergravity action is of the standard form of Cremmer et al.
[37,38]. After inserting the ansatz (30) into the 5d gravitational action, substituting the 5d
1 = a 1/2 ( ) , 2 = a 1/2 ( )
gravitino with the 4d zero-modes R
R 4d
L 4d (see [9,39]),
L
and integrating over x 5 one finds the following graviton and gravitino kinetic terms in 4d:
1 2
(4) 1 MP2 e4 (1 e2kR0 ) D + , (33)
MP e4 (1 e2kR0 )R
2
2
where the R0 dependent bracket is simply the scalar part of the real vector superfield f (in
what follows we shall use the symbol f for this scalar function as well). After performing
the Weyl rescaling of the graviton and gravitino
ea ea f 1/2 ,

f 1/4 ,

(34)

one arrives at the canonical action for graviton and gravitino in four dimensions.
Substituting the ansatz (30) into the 5d
 action results also in (a part of) the kinetic terms for
R0 Skin = 3M 3 k()2 e2kR0 d 4 x e4 R0 R0 ; together with another term with
two derivatives on R0 generated by the Weyl rescaling, this gives exactly the kinetic
energy reproduced by the Khler potential K given in (31). A procedure of rescalings and
integrating over x 5 applied to the kinetic term of V (and ) also produces 4d kinetic terms
which are immediately seen to be exactly given by (31). It is worth noticing at this point
that, contrary to assumptions usually made, the natural 5d frame where we have coupled
bulk gravity and moduli with general gauge sectors on branes does not directly give us the
superspace frame of the 4d supergravity. This is seen from the fact, that the Weyl rescaling
in 4d is not given by the complete Khler function, but only by the part of it that depends
on the radion. Fortunately, radion and hypermultiplet moduli do not have a kinetic mixing
in the canonical 5d frame.
4.2. Effective superpotential
Knowing the Khler function and performing the reduction of the gravitino mass terms
(the ones which in five dimensions couple Z2 even components with even components)
one can identify the effective 4d superpotential


W = 2 2 W1 + e3T W2 .
(35)
One can see that the contribution to the effective superpotential given by the second
brane at x 5 = is suppressed by the factor a03 with respect to the contribution from the
Planck brane. This fits nicely the notion of a universal down-scaling of all the mass scales
on the visible brane.
The derivation of the superpotential goes as follows. There are three contributions to the
4d gravitino mass terms:

1
1
3  5  1 2
1 5 2
D
Lgm = M 3 e5
5 k" x
2
2
V

3
+ iA5 k"(x 5 ) 1 2 + h.c. .
(36)
2 2

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

203

The first contribution comes from the term in (32) involving the part of the spin
connection , the second from the 5d gravitino mass term involving the prepotential P.
Finally the term with the graviphoton A5 comes from the SU(2) covariant derivative in
the gravitino kinetic term: D A = A + gA P A B B . Plugging in the gravitino zero
modes in the solution (28), integrating over x 5 and Weyl rescaling, we get:



1
1 1 ek(2R0 +3 2 i A5 ) R L + h.c.
L4dgm =
e
W
4
2
3/2
2 V0 f MP


1
1 + e3kR0 W
2 R L + h.c.
=
e4 W
2
3/2
2 V0 f MP

(37)

In the last step we used the boundary condition (23). In the standard 4d supergravity
formulation [37] this term has the form e24 eG/2 R L + h.c. with G = K + ln |W |2 .
The form of the Khler potential (31) and the holomorphicity of the superpotential requires
the superpotential of our 4d model to be exactly (35).
The non-trivial test of the consistency of the superpotential (35) comes from minimizing
the 4d effective scalar potential derived from (31) and (35). In the standard formulation it

has the form: V = e4 eG (Gi Gi j Gj 3) or explicitly:


V = e4

4
MP2 V0 3 (1 e2kR0 )3






|W1 |2 3e2kR0 2 + |W2 |2 3e4kR0 2e6kR0


2 e3kR0 i 2A5 + W2 W
1 e3kR0 +i 2A5 .
+ W1 W

(38)

Minimizing the above scalar potential with respect to R0 and A5 yields the relation (23),
consistently with the 5d picture. Also, the value of the prepotential at the minimum (that
is the cosmological constant) is consistent with the solution (28). Note also that, the 4d
effective potential is of the runaway type with respect to V0 .
Just as a remainder one should mention that the usual redefinition of the 4d components
of the gravitino leading to removal of the kinetic mixing terms between A and 5B is
needed A A + 15 5 5A . Inspecting more closely the fermionic mass matrix one
2e5

notices readily the mass terms proportional to 5  and   which mix 4d gravitini with 5
and hyperini . The origin of these mass terms is analogous to that of gravitini masses, and
their presence signals that the superHiggs mechanism is at work (as expected). To see that
the higgsino is a mixture of hyperino and modulino 5 one can inspect the supersymmetry
transformation laws of these fermions:
  2ie5
1
51 = 5 " 2 + i x 5 5 W " 2
V
V



1B Im( )  2 1B
 5
Re( )
5 "B ,
1
+ i2k" x
+
2 V
2 V
  2ie5
1
"1
52 = + 5 " 1 i x 5 5 W
V
V

204

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217



 
Re( )  1B Im( )  2 1B
+ i2k" x 5 1
+
5 "B ,
2 V
2 V

 5 2 V
 
i
1
5
2
" 2 + 3k" x 5 VuA1 k u "A ,
W
= + 5 " i x
a
2V

 


V 1
i
2
2 = + 5 5 " 1 i x 5
W " + 3k" x 5 VuA2 k u "A
a
2V

(39)

just repeating the procedure given in [7] for the HoravaWitten model. In the above VuAb
are SU(2) vielbeins given in Appendix A. Since the theory has a mass gap <m = mKK
Ma(), to find the low-energy goldstino one needs to substitute into above equations the
vacuum solutions for bulk scalars and to project resulting expressions onto their zero-mode
components.
Another way to identify the 4d goldstino is to derive FS  and FT  with the help of the
4d effective Lagrangian and we shall give the result at the end of this chapter.
4.3. Gauge kinetic functions
To arrive at a realistic model, one should introduce gauge and charged matter fields. One
option with gauge fields living in five-dimensional vector multiplets was studied
in [23,39] and yields the gauge kinetic function H T . Here we present an alternative, a
model with gauge and matter fields confined on the boundaries. The action of the gauge
sector is:


e5  
V
V a a 1
a a
F
F
aD
/ a
SYM = d 5 x 5 x 5 F
F
4
4
2
e5
 a
V
a F
+
4


3i V 
1 a a
i
+ 5 a 5 a F5 (
)F a 5 a
4
8
4 2 e5



 
V  a a
a
a
5 L R 5 + R L 5 + (4 fermi)
(40)
2e5
and similarly on the visible brane at . In the above the bulk fermions appear in their
even (and Majorana in the 4d sense) combinations defined as:

 2 


L
i1L
=
(41)
,

=
,
2
V
1
i2R
R
and brane supersymmetry is generated by the Majorana fermion
 2
"
" = L1 .
"R
The tree-level gauge kinetic function turns out to be universal and equal:
H (S) = S

(42)

on either brane. However, we expect corrections to this universality (see the forthcoming
discussion).

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

205

4.4. Khler function for matter fields


Now we introduce the matter fields living on the branes. We allow the superpotentials
W to depend on the scalar . Let us concentrate on the matter living on the visible brane
at x 5 = . The part of the action relevant to our discussion is (more terms and corrections
to fermionic transformations are given in Appendix C):


i

e5  5
4 Wi W

+ .
Lm, brane = x D D
(43)

g55
V
In passing from the canonical 5d frame to the 4d Einstein frame one needs two rescalings
of the metric. First, one factorizes out of the original metric the warp factor and, second,
one performs the Weyl rescaling ea ea f 1/2 . The second, potential term in the brane
Lagrangian (43) does not get corrected through the Weyl rescaling of the curvature scalar R
(all new terms borne this way carry two spacetime derivatives); after rescalings it becomes


1 2 4kR0  W2 2
Lpot, b = 4e4 f e
(44)
  .
V0

The canonical 4d SUGRA expression for such a term is e4 eK g | W |2 , where g


is the inverse Khler metric for the matter fields. Comparing these two expressions, one
obtains information about the matter Khler metric. In addition, one should remember
that in the limit k 0 the matter kinetic terms are reproduced with the Khler function
3 log(T + T ||2 ) with a suitable coefficient . This suggests a
K0 = log(S + S)
trial function of the form
 

3MP2 log f T + T ||2 ,
= MP2 log(S + S)
K(T , T ; , )
(45)

where f = (1 e(T +T || ) ). And indeed, the term (44) is reproduced upon



k
substituting K |=0 into the standard supergravity expression with = 3M
3 . In addition

2
we need to redefine the real part of the T modulus: Re T = R0 2 || .
The additional check comes from the first term in (43) which after rescalings
In canonical supergravity it should equal
takes the form e4 f 1 e2kR0 D D .
2K


e4 D D and indeed it is for our choice (45) with = k/3M 3 (one should
notice that fT |=0 = e2kR0 ).
The above discussion applies to matter living on the warped brane (whenever necessary
we should denote it by 2 ). To include matter living on the Planck brane (1 ) we need
to improve the Khler potential further. The expression which reproduces properly also
matter Lagrangian on the first, unwarped, brane is


 
3MP2 log f T + T |2 |2 |1 |2 , (46)
K(1 , 2 ) = MP2 log(S + S)
2

k
where as before = 3M
3.
To summarize, we have deduced the zeroth-order approximations to the Khler function,
superpotential and gauge kinetic functions of the 4d effective supergravity for 5d warped
RandallSundrum model with general gauge and matter sectors on the branes.

206

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

It turns out that the leading effects of the RandallSundrum brane tensions are encoded
in the exponential dependence of the effective Khler function on radion and matter fields,
and in the exponential suppression of the contributions to the effective superpotential borne
on the warped brane.
4.5. Gaugino condensates
In this context it is interesting to ask the question about the proper immersion of the
gaugino condensates into the effective supergravity picture. Let us discuss this issue at
the level of the 4d model we have just constructed. The basic expression for the effective
potential including the contribution from the gaugino condensates we start with is

2

1

V = eK g S S DS W + eK/2   .
4

(47)

Usually, for canonically normalized gauge and gaugino fields, one replaces the condensate
3
by 3c = MGUT
e3 Re(S)/2b0 . Using holomorphicity and R-symmetry of the gauge sector
one often promotes this contribution to the one generated by an effective superpotential
for the dilaton superfield S. However, the question arises of what we shall substitute for
MGUT in the warped case. To answer this we should carefully recompute the condensation
scale c using the one-loop renormalization group equation for the gauge coupling and
watching the rescalings we make before reaching the 4d canonical frame. The point is
that, to achieve canonical normalization of gauginos, we must perform the rescaling
a 3/2f 3/4 . These rescalings are anomalous and amount to threshold corrections at the 4d
upper scale of running which is MP


1
1
Re(S) +
log f log a 2b0 .
(Re(S) =) 2
(48)
g (MP )
2
Now the renormalization group running looks like




1
1
MP
= Re(S) +
log f log a 2b0 2b0 log
.
g 2 (p)
2
p

(49)

This gives the condensation scale


3c = MP3 a()3f 3/2 e3 Re(S)/2b0 .

(50)

Thus we see that the effective MGUT at the warped brane equals essentially MP a() (but
MGUT = MP on the Planck brane). We can substitute the condensation scale that we have
just derived into the potential (47) to obtain

2


1
2
3 3kR0 3 Re(S)/2b0 

e
V = e (S + S) DS W + (S + S)MP e

4
K

(51)

(where by W we mean a perturbative superpotential plus possible constant contributions


inherited from higher dimensions).

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

207

It is tempting to replace the contribution from the condensates by the effective


superpotential
Wnpert = MP3 e3T e3S/2b0

(52)

(the factor Re(S) multiplying the exponential dependence on S can safely be neglected).
This would fit nicely with the formula for the perturbative 4d superpotential. However, one
should notice that plugging such a superpotential into the standard SUGRA expression
would generate in the potential new terms which are not present in the expression (51),
namely these containing DT Wnpert , which would be non-vanishing. This inconsistency
is alleviated if one notices that together with threshold corrections we have used, there
are suitable one-loop corrections to gauge kinetic function on the warped brane. The
threshold correction on that brane may be split into two pieces. The first is the same as on
the Planck brane and comes from the Weyl rescaling by the power of f . The f became
identified with a scalar component of a real superfield defining the Khler potential, hence
this part of the 1-loop corrections is fully analogous to result of the Weyl rescaling leading
from superspace to canonical frame given by Bagger et al. [42]. The second piece is nonuniversal, and is associated with additional powers of the warp factor a multiplying the
gaugino terms on the warped brane. Since already at the level of the effective perturbative
superpotential we have found it consistent to promote a to a chiral superfield, a eT ,
then also here this part of the correction should be understood as a correction to the gauge
kinetic function on the warped brane, which now becomes
Hwarped (S, T ) = S + 2b0 T .

(53)

The supergravity model defined with (52) and (53) gives effective potential suitable to
study moduli stabilization and supersymmetry breakdown due to gaugino condensation in
the effective four-dimensional theory.
T , T ; , )
=
The four-dimensional supergravity model defined by K(S, S;

k
k
2
2
2
2


MP log(S + S) 3MP log(f (T + T 3M 3 |2 | ) 3M 3 |1 | ), W = 2 2 (W1 +
e3T W2 ), Hwarped (S, T ) = S + 2b0 T and Hplanck(S) = S has been constructed as a small
perturbation around the generalized RandallSundrum background. However, it is very
likely that extrapolating this model away from the original vacuum and using it on its
own makes sense as a tool to explore phenomenology of warped compactifications at TeV
energies.
Let us summarize basic features of our model. The F-terms take at the minimum the
expectation values


2 a 2 ()|W2 |2 1 a 2 () 2 = 0,
|F S |2 = 8eK (S + S)
|F T |2 = 0
which means that supersymmetry is broken along the dilaton direction. The potential
energy at this vacuum is negative:
Vvac =

a 2 ()
8|W2 |2
.
V0 (M 3 /kMP2 )3 MP2 1 a 2 ()

(54)

208

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

The mass of the canonically normalized radion is


m2R =

a 2 ()|W2 |2 1
24
V0 (M 3 /kMP2 )3 (1 a 2 ()) MP4

(55)

and the gravitino mass term is given by the expression


a()
|W2 |
2
.
m3/2 =
2
2
1/2
3
3/2
V0 (M /kMP ) (1 a ())
MP2

(56)

It is interesting to compare these features to those of the no-scale models: there F S = 0,


Vvac = 0, and F T is undetermined at tree-level [43].

5. Summary
The main result of this paper is the four-dimensional effective supergravity action which
describes low-energy physics of the RandallSundrum model with moduli fields in the bulk
and charged chiral matter living on the branes.
The relation between 5d and 4d physics has been made explicit; the low-energy action
has been read off from a compactification of a locally supersymmetric model in five
dimensions. The exponential warp factor has interesting consequences for the form of
the effective 4d supergravity. The asymmetry between the warped and unwarped walls is
visible in the Khler function, gauge kinetic functions and in the superpotential. Roughly
speaking the contributions to these functions which come from the warped wall are
suppressed by an exponential factor containing the radion superfield. This is the way the
warp factor and (and RS brane tensions) are encoded in the low-energy Lagrangian.
We have described the mechanism of supersymmetry breaking mediation which relies
on non-trivial configuration of the Z2 -odd fields in the bulk. We point out that the oddZ2 -parity fields can be an important ingredient of 5d supersymmetric models. They play a
crucial role in communication between spatially separated branes.
Moreover, we have demonstrated that, after freezing the dilaton, it is possible to stabilize
the radion field in the backgrounds with broken supersymmetry and excited odd-parity
fields. To achieve this we do not need to add vector fields and/or exotic charged matter in
the bulk.
We believe that the class of models we have constructed in this paper provides the proper
explicit setup to study low-energy phenomenology of the supersymmetric brane models
with warped vacua.

Acknowledgements
This work has been supported by RTN programs HPRN-CT-2000-00152 and HPRN-CT2000-00148. Z.L. and S.P. are supported by the Polish Committee for Scientific Research
under grant 5 P03B 119 20 (2001-2002).

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

209

Appendix A. The BPS solution in 5d supergravity coupled to the universal


hypermultiplet
First, let us summarize the basic facts about the geometry of the universal hypermultiplet. The 4 real coordinates are denoted q u = {V , , x, y}; when convenient, we also use
= x + iy.
The metric is:
1
,
4V 2
y2
1
hxx = + 2 ,
V
V
y
h x = 2 ,
2V
The inverse metric:
hV V =

hV V = 4V 2 ,
hxx = V ,
h x = 2V y,

1
,
4V 2
x2
1
hyy = + 2 ,
V
V
x
hy =
.
2V 2

h =

h = 4V 2 + | |2 ,
x2
hyy = V + 2 ,
V
hy = 2V x.

h V = 0,
hxy =

xy
,
V2
(A.1)

h V = 0,
hxy = 0,
(A.2)

We can introduce a vielbein one-form V = Vu


a vielbein as:

dq u .

Metric can be expressed in terms of

huv = "AB ab VuAa VvBb ,

(A.3)

where " and are totally antisymmetric and we choose "12 = 21 = 1. This formula
determines the vielbein (up to SU(2) transformation) to be:


1
1
y
1
1
2
3
2
dV +
d + +
dx
V=
2 2V
2 2V
2V
2V


1
x
2 dy
+ iI
(A.4)
2V
2V
or more explicitly




1
0
0
i 1
2 2V
2 2V
VV =
=
,
V
,

1
0
i 1
0
2 2V
2 2V


 1
 1
i
i y
i x
2V
2V
2V
2V
,
V
.
=
Vx =
(A.5)
y
i y
1
i x
i 1
2V

2V

2V

2V

The inverse vierbein V u satisfies the equation:


u
VAa
VvAa = vu .

The explicit form is:



0
2V
V
V =
,
2V
0

(A.6)

V =

i 2V
i 2 V


i 2 V
,
i 2V

210

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217


V =
x

V
2

0


V
2

V =
y

V
2

0



V
2

(A.7)

The SU(2) spin connection :


V = 0,

1
i 3 ,
4V
x
1
i 3 1/2 i 1 ,
y =
2V
V

y
1
i 3 1/2 i 2 ,
2V
V
and the Sp(1) spin connection :
x =

(A.8)

3
i 3 ,
4V
3y 3
3x 3
i ,
i .
x =
(A.9)
y =
2V
2V
The Khler form (which for quaternionic manifolds that can occur in supersymmetric
theories is minus the curvature form) is:
V = 0,

1
1
y
i 3 ,
KV x = 3/2 i 2 +
i 3 ,
8V 2
4V
4V 2
1
x
KV y = 3/2 i 1
i 3
4V
4V 2
1
1
K x =
i 1 ,
Ky = 3/2 i 2 ,
3/2
4V
4V
1
x
y
3
1
Kxy =
(A.10)
i +
i
i 2 .
3/2
2V
2V
2V 3/2
To illustrate our discussion of supersymmetry breaking let us study the explicit
model presented in [9]. It is based on 5d-dimensional supergravity with the universal
hypermultiplet and the U (1) symmetry ei gauged. The Killing vector and
prepotential corresponding to our gauging have components:
KV =

k x = y,

k y = x,

k V = k = 0,
x
y
P2 = ,
P1 = ,
2 V
2 V





1
1
| |2
| |2
P3 =
P 2 =
1
,
1+
.
4
V
4
V

(A.11)

Locally, even with non-zero sources for the odd field it is possible to find a BPS
solution, which preserves half of the supersymmetry. Plugging in the explicit metric and
the prepotential into the BPS conditions (15) we get:
(V )

V = 6kR0 | |2 ,

( )

2(yx xy ) = 0,

y
2(yx xy ) = 3kR0 x,
x
V

x

2(yx xy ) = 3kR0 y,
y +
V

(x)
(y)

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217



| |2
a
= kR0 1 +
.
a
V
It is straightforward to find the solution:
   

x = C1 exp 3kR0 x 5  " x 5 ,
   

y = C2 exp 3kR0 x 5  " x 5 ,




V = V0 C12 + C22 exp 6kR0 |x 5 | ,

211

(A.12)

(A.13)

= 0 .
Integrating equation for the warp factor we find:
 
 



5 1/6
a x 5 = exp kR0 x 5  V0 C12 + C22 e3kR0 |x |
.
The matching condition in the Einstein equation yields:




a
a
| |2
| |2
(0) = kR0 1
(0) ,
() = k 1
() .
a
V
a
V

(A.14)

(A.15)

If C1 = C2 = 0, the above set is satisfied and the flat BPS solution satisfies all boundary
conditions. But as soon as we induce a non-zero value of (by coupling 5 to the
boundary sources) it is impossible to find any solution to (A.15) and supersymmetry is
broken.

Appendix B. Supersymmetric coupling of the odd hypermultiplet fields to the brane


sources
In this appendix we investigate the supersymmetric coupling of the hypermultiplet odd
fields to the brane sources. It is obvious that away from the branes the theory does not
distinguish between the even and odd fields. However, matching conditions at the fixed
points change in a significant way. To see this we factorize the x 5 dependence of an odd
field :
   
 
x 5 = " x 5 x 5  ,
(B.1)
where is smooth near the fixed points; hence:
  



5 = + 2 x 5 (0) x 5 () ,
  


 

5 5 = " x 5 + 2 x 5 (0) x 5 () .

(B.2)

The odd fields that are non-zero at the Z2 fixed points must have a jump there. As
a consequence their first derivatives have delta functions singularities and their second
derivatives have derivatives of the delta. The equations of motion are second order, thus we
must somehow cancel singularities by a suitable choice of the brane action. One way to
achieve this is to couple the odd fields to the sources on the boundary through their fifth
derivative. Such couplings naturally arise in M-theory [40] and its compactification to 5d

212

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

[28] where the field from the universal hypermultiplet couples to gaugino bilinears and
to the superpotential of the boundary scalars.
Local supersymmetry considerably restricts the possible form of such derivative
couplings to the brane sources. We will show that couplings of the fifth derivative of an odd
field to the boundary sources r has to enter the bosonic action (6) in a form of the full
square. That is we have to replace 5 q r 5 q r 5 q r + (x 5 )( r )1 + (x 5 )( r )2 ,
in the kinetic terms of the odd fields (note that it implies that 2 singularities appear in the
Lagrangian).
Let us introduce the following coupling of an odd scalar q r to the source located on,
say, the hidden brane:
  e4
L = x 5 5 5 q r r (qu ).
(B.3)
e5
We allow for the dependence of the source on the even hypermultiplet scalar (it will
turn out crucial for the consistency). Now we want to supersymmetrize (B.3). Consider the
variation of the (odd) scalar q in L :
i r A a
e4 i
r A a
" L = 5 5 (VAa
" )r .
q r = VAa
2
e5 2

(B.4)

First consider the fifth derivative acting on ". Using the fact that 5A = 5 " A this variation
can be cancelled by adding a new term to the boundary Lagrangian:
  e4 i r  a A 
5 r .
L5 = x 5 5 VAa
e5 2

(B.5)

The variation with the 5th derivative acting on can be cancelled by the variation of the
hyperino kinetic term provided we modify the hyperino transformation law in the following
way:
a =

(x 5 ) i ab s 5 B
VbB " s .
e55 2

(B.6)

This modification has an immediate consequence that 2 terms are necessary in the brane
Lagrangian because is already present in the brane Lagrangian. Namely, varying (B.5)
we get:
  2 e4 1 r ab s
V VbB 5 A " B r s
L5 = x 5
5 2 4 Aa
(e5 )
  5 2 e4 1  A 5  rs
= x
5 "A h r s .
(e55 )2 8

(B.7)

Recalling, that e55 = 12 (5 A 5 "A ) the above can be cancelled by adding a singular term
to the brane Lagrangian:
  2 e4 1 rs
h r s .
L = x 5
e55 4

(B.8)

This is exactly what we need to cancel the 2 singularities in the equations of motion.

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

213

Note that (B.3), (B.8) and the scalar kinetic term can be gathered in a full square:
u qw
LF S = e5 g huw 
q 

(B.9)

where we defined:
 5
u
5 1 ur
u

q = q h r x .
2

(B.10)

Thus we can guess that all the terms needed to supersymmetrize the action can be found
by exchanging in the bulk Lagrangian the fifth derivative of the scalars by the hatted one.
This procedure leads to appearance of the following new terms in the brane Lagrangian:

 5
i
1
r
rs
LB = e4 x
a 5 A VAa
+ A 5 B (r )A
B h s
2
4

1
a 5 b (r )ab hrs s
(B.11)
4
and in the SUSY transformation laws:
 5
1
B rs
5A = (r )A
B " h s x ,
2
(x 5 )
i
s
a = 5 " B ab VbB
s 5 .
2
e5

(B.12)

Note that this procedure correctly reproduces the terms we determined before, e.g., the
correction to the hyperino transformation laws.
Nevertheless, exchanging the derivatives with their hatted counterpart does not ensure
that the action is supersymmetric. If we make the supersymmetry transformation of the
modified Lagrangian we get:
 L

L u
L
u
u qw =
u qw
+
L , 
q + 2

5 q , 
5 q 
5
5 q 
5
u 5
w



q
5
5 q u 
5q
L
L
L
w
+
5 q u + 2
5 q u 
5q
u

u 
w

5 q

q
q
5
5


1  ur  5
L
L
w

h r (x )
+2
5 q .
u
u
w
2

5q

5 q 
5q
(B.13)

The first three terms vanish here if they vanish in the unmodified 5d supergravity. But
the last term does not vanish automatically. The easiest way to nullify this contribution
is to assume that the supersymmetry variation (hur r ) is identically zero. This can be
achieved if the function which determines the coupling of odd scalars to the boundary
can be expressed as:
r = hrs As ,
where As are constants (which determine the boundary values of the odd fields).

(B.14)

214

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

Let us turn to the example of the universal hypermultiplet. Eq. (B.14) implies that the
coupling of the odd scalars to the brane has the form:
y 
x 


y 2 xy A
x 2 xy A
Ax
Ay
Ax
Ay
1+
5 x +
1+
5 y.
L =
V
V
V
V

(B.15)

Appendix C. Supersymmetric coupling of gauge sectors on the boundaries to the


bulk
For completeness let us summarize the relevant parts of the bulk bosonic action and the
brane action coupled to it. Derivation of the complete result including four-fermions terms
can be found in [10]. The action is S = Sbulk + SYM + Smatter where



1
1
R
5

Sbulk = d x e5
V V +
2
4V 2
V


1
1
2
2
4
| |
| |
+ 6k 1 +
2V
2V 2



e5 
V
V a a 1
a a
F
F
aD
/ a
SYM i = d 5 x 5 x 5 xi5 F
F
4
4
2
e5
 a
V
a F
+
4

 a

3i V 
1
i
+ 5 a 5 a F5 a F
a 5 a
4
8
4 2 e5





V 
5 a L Ra 5 + a R La 5
2e5

V 3/2  a 2 2

+ (4 fermions)
+ (0)
8






 5
2
| |2
5 e5
55
i
W + W
Smatter i = d x 5 x xi "i 6k 1

g
V
V
e5
i



4 Wi W
+ V 3/2 W
a a
i
D i D
+ (0) 4W W
R L
i

V i




1
W 
L R 2/3 W L L
V
V

i 5

+ 3/2 e5 W (R5 L ) + h.c.


V

(C.1)

where i = 1, 2 labels branes and "1,2 = +1, 1. In the above Z2 -even 4d Majorana
fermions are defined as:
 2 



L
i1L
=
(C.2)
2
V
,

=
,
1
i2R
R

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

215

and brane supersymmetry is generated by the Majorana fermion


 2
"
" = L1 .
"R
The relevant parts of the supersymmetry transformation of bulk fermions, which depend
explicitly on brane operators that are allowed to take an expectation value, are


2ie5
i "R ) x 5 xi5 ,
5 = 5 (Wi "L W
V



i "R ) x 5 x 5 .
= 2 V (Wi "L W
i

(C.3)

Supersymmetry transformations of the charged fermions on branes are = 12 (D p



p
R ) "R + 1 L (" 5 ) 1 W "L . After solving the equations of motion

8V

p
V

for , boundary operators appear also in bulk parts of the modulini supersymmetry
transformations (these are given in the main text).

Appendix D. Conventions and normalizations


The normalizations are mainly those of [28]. The signature of the metric tensor is
( + + + +). The indices , , . . . are five-dimensional (0, . . . , 3, 5), while 4d indices
are denoted by , , . . . . We define


1 0
.
5 =
0 1
The Z2 symmetry acts as reflection x 5 x 5 and is represented in such a way that
bosonic fields (em , e55 , A5 ) are even, and (e5m , e5 , A ) are odd. The action of Z2 on the
gravitino is 5 A (x 5 ) = ( 3 )A B B (x 5 ) 5 5A (x 5 ) = ( 3 )5 B 5B (x 5 ) (so that, for
1 is even and 1 is odd). The action of Z on the SUSY parameter " A is the
instance, R
2
5R
same as the action on A .
In the spinor basis we use, the SU(2) R-symmetry is manifest. The index A of the
gravitino transforms in the fundamental representation of SU(2). The SU(2) indices are
raised with an antisymmetric tensor " AB . We choose " 12 = "12 = 1. Explicitly: 1 = 2
2 = 1 . Note that A A and A A , so one has to be careful about the position
of the bar. The Majorana condition is:
 T
A = A C,

(D.90)

where C is the 5d charge conjugation matrix satisfying C C 1 = ( )T . In particular


1 = C5 (2 )T , 2 = C5 (1 )T . In the chiral basis C = i 2 0 5 .
The rule for dealing with symplectic spinors is A 1 ...n B = B n ...1 A . From
the above formula one can deduce: 1 " 1 = 2 " 1 = 2 " 1 = " 1 2 =
"1 2 = " 2 2 .

216

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

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]

L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.


L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690.
J. Lykken, L. Randall, JHEP 0006 (2000) 014.
S. Forste, Z. Lalak, S. Lavignac, H. Nilles, JHEP 0009 (2000) 034;
S. Forste, Z. Lalak, S. Lavignac, H. Nilles, Phys. Lett. B 481 (2000) 360.
R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario, hepth/0003117.
T. Gherghetta, A. Pomarol, Bulk fields and supersymmetry in a slice of AdS, hep-th/0003129.
A. Falkowski, Z. Lalak, S. Pokorski, Phys. Lett. B 491 (2000) 172.
N. Alonso-Alberca, P. Meesen, T. Ortin, Supersymmetric brane-worlds, hep-th/0003248.
A. Falkowski, Z. Lalak, S. Pokorski, Five-dimensional supergravities with universal hypermultiplet and warped brane worlds, hep-th/0009167, Phys. Lett. B in press.
A. Falkowski, M.Sc. Thesis, info.fuw.edu.pl/afalkows.
E. Bergshoeff, R. Kallosh, A. Van Proeyen, Supersymmetry in singular spaces, hep-th/0007044.
K. Behrndt, M. Cvetic, Phys. Rev. D 61 (2000) 101901.
K. Behrndt, M. Cvetic, Phys. Lett. B 475 (2000) 253.
R. Kallosh, A. Linde, JHEP 0002 (2000) 0005 and references therein.
R. Kallosh, A. Linde, M. Shmakova, JHEP 9911 (1999) 010.
K. Behrndt, Nonsingular infrared flow from D = 5 gauged supergravity, hep-th/0005185.
P. Mayr, Stringy world branes and exponential hierarchies, hep-th/0006204.
H. Verlinde, Supersymmetry at large distance scales, hep-th/0004003.
M.J. Duff, J.T. Liu, K.S. Stelle, A supersymmetric type IIB RandallSundrum realization, hepth/0007120.
K. Behrndt, C. Herrmann, J. Louis, S. Thomas, Domain walls in five-dimensional supergravity
with non-trivial hypermultiplets, hep-th/0008112.
M. Zucker, Supersymmetric brane worlds scenarios from off-shell supergravity, hepth/0009083.
J. Bagger, D. Nemeschansky, R. Zhang, Supersymmetric radion in the RandallSundrum
scenario, hep-th/0012163.
M. Luty, R. Sundrum, Hierarchy stabilization in warped supersymmetry, hep-th/0012158.
M. Cvetic, H. Lu, C. Pope, Localized gravity in the singular domain wall backgrounds?, hepth/0002054.
S. Nojiri, S.D. Odintsov, Supersymmetric new brane world, hep-th/0102032.
J. Maldacena, C. Nunez, Supergravity description of field theories on curved manifolds and a
no go theorem, hep-th/0007018.
E. Mirabelli, M. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Nucl. Phys. B 552 (1999) 246, hep-th/9806051.
J. Ellis, Z. Lalak, S. Pokorski, W. Pokorski, Nucl. Phys. B 540 (1999) 149, hep-th/9805377.
J. Ellis, Z. Lalak, W. Pokorski, Nucl. Phys. B 559 (1999) 71.
A. Karch, L. Randall, Locally localized gravity, hep-th/0011156.
M. Gunaydin, G. Sierra, P.K. Townsend, Nucl. Phys. B 253 (1985) 573;
M. Gunaydin, M. Zagermann, Nucl. Phys. B 572 (2000) 131.
A. Ceresole, G. DallAgata, General matter coupled N = 2, D = 5 gauged supergravity, hepth/0004111.
L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79.
P. Horava, Phys. Rev. D 54 (1996) 7561.
C. Schmidhuber, Nucl. Phys. B 585 (2000) 385.
E. Cremmer, S. Ferrara, L. Girardello, A. Van Proyen, Nucl. Phys. B 212 (1983) 413.
J. Wess, J. Bagger, Supersymmetry and Supergravity, 2nd ed., Princeton Univ. Press, 1992.

A. Falkowski et al. / Nuclear Physics B 613 (2001) 189217

217

[39] T. Gherghetta, A. Pomarol, A warped supersymmetric standard model, hep-ph/0012378.


[40] P. Horava, E. Witten, Nucl.Phys. B 475 (1996) 94.
[41] C. Csaki, J. Erlich, C. Grojean, Gravitational Lorentz violations and adjustment of the
cosmological constant in asymmetrically warped spacetimes, hep-th/0012143.
[42] J.A. Bagger, T. Moroi, E. Poppitz, Nucl. Phys. B 594 (2001) 354.
[43] E. Cremmer, S. Ferrara, C. Kounnas, D. Nanopoulos, Phys. Lett. B 133 (1983) 61;
J. Ellis, A.B. Lahanas, D.V. Nanopoulos, K. Tamvakis, Phys. Lett. B 134 (1984) 429.

Nuclear Physics B 613 (2001) 218236


www.elsevier.com/locate/npe

A note on the supergravity description


of dielectric branes
D. Brecher, P.M. Saffin
Centre for Particle Theory, Department of Mathematical Sciences, University of Durham, South Road,
Durham DH1 3LE, UK
Received 5 July 2001; accepted 2 August 2001

Abstract
We comment on the recent papers by Costa et al. and Emparan, which show how one might
generate supergravity solutions describing certain dielectric branes in ten dimensions. The basic
such solutions describe either N fundamental strings or N D4-branes expanding into a D6-brane,
with topology M2 S 5 or M5 S 2 , respectively. Treating these solutions in a unified way, we note
that they allow for precisely two values of the radius of the relevant sphere, and that the solution with
the smaller value of the radius has the lower energy. Moreover, the possible radii in both cases agree
up to numerical factors with the corresponding solutions of the D6-brane worldvolume theory. We
thus argue that these supergravity solutions are the correct gravitational description of the dielectric
branes of Emparan and Myers. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.65.+e

1. Introduction
The higher-dimensional counterparts of the Melvin universe [1], which describe
gravitating magnetic fluxbranes, have appeared in various string- and M-theoretic guises
over the last few years (see, e.g., [215]). Within the KaluzaKlein context, the
most natural way to generate such fluxbranes is via a twisted compactification of flat
space [1618]. Moreover, starting instead with the Euclidean Schwarzschild solution
cross a trivial time direction, one can generate spherical branes via just such a twisted
compactification [18]. In slightly different language, one can think of this as follows:
take a D-dimensional uncharged black string; analytically continue the worldvolume
coordinates; and perform a double-dimensional reduction along a twisted direction. The
E-mail addresses: dominic.brecher@durham.ac.uk (D. Brecher), p.m.saffin@durham.ac.uk (P.M. Saffin).
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 1 ) 0 0 3 8 2 - 0

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

219

resulting (D 1)-dimensional solution describes a spherical (D 5)-brane in the core of


a magnetic fluxbrane. It has local, but zero net, magnetic charge. 1
If one now adds N units of 2-form charge to the D-dimensional black string, the
(D 1)-dimensional solution will exhibit N units of 1-form charge, so is more rightly
interpreted as a collection of N particles expanding into a sphere under the influence
of the background (D 3)-form magnetic field. Of course, these techniques need not
only be applied to black strings and, in recent work, Costa et al. [19] and Emparan [20]
considered their application in an M-theoretic context. Starting with the basic branes of
eleven-dimensional supergravity, then, these authors showed how to generate solutions
of type IIA supergravity which describe various ten-dimensional branes expanding into
a D6-brane under the influence of a background 8-form magnetic field. In particular,
the reduction of certain M2- or M5-brane solutions generates ten-dimensional solutions
describing F -strings or D4-branes expanding into a D6-brane with topology M2 S 5 or
M5 S 2 , respectively, where Mn denotes an n-dimensional Minkowski space. Applying
T -duality generates more general configurations [20] Dp-branes expanding into a
D(p + 2)-brane, and F -strings expanding into a Dp-brane, for arbitrary values of p
but these solutions will necessarily be smeared along certain directions. 2
One of the motivations behind this work was to try to get a handle on the possible
supergravity description of the dielectric branes of Emparan [21] and Myers [22]. Imposing
T -duality invariance on the non-Abelian action relevant to the description of multiple
Dp-branes, gives rise to a whole host of worldvolume couplings which are not present
in the Abelian theory [22,23]. In particular, the ChernSimons piece of the action must
include couplings to RamondRamond (RR) potentials of degree greater than (p + 1),
giving rise to possible couplings to higher-dimensional branes. The presence of such terms
allows for the dielectric effect, the simplest example of which is that of N Dp-branes
expanding into a D(p + 2)-brane under the influence of the (p + 3)-form RR potential
associated with the latter [22,24]. The resulting dielectric brane, which is a minimum
2 , where S 2 is the
of the worldvolume energy functional, has worldvolume Mp+1 SNC
NC
noncommutative or fuzzy two-sphere.
Although this process is most properly described from within the non-Abelian
worldvolume theory of the Dp-branes, one can alternatively consider the dual description
from within the Abelian theory of the D(p + 2)-brane. In the large N limit, the
noncommutative nature of the sphere is lost, and these two descriptions agree [22].
Since it is unlikely that such noncommutative structures will appear within a supergravity
perspective, any supergravity solution purporting to describe this dielectric effect should be
compared to solutions of the D(p + 2)-brane worldvolume theory. From this perspective,
the standard coupling of a (p + 1)-form RR potential to the D(p + 2)-brane gives rise
to dissolved Dp-branes within the worldvolume of the D(p + 2)-brane [25]. And it is
1 More generally, spherical (D 3 2n)-branes in the core of n intersecting fluxbranes can be generated by
applying n twists [18].
2 Further solutions, generated from certain intersecting configurations of M-branes, were considered by
Emparan in [20] but we will say no more about these here.

220

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

the presence of this dissolved Dp-brane charge which allows for spherical solutions of the
D(p + 2)-brane theory.
The existence of such spherical D(p + 2)-branes with dissolved Dp-brane charge was
anticipated by Emparan in [21]. The main point of this work, however, was to analyse
spherical D(p + 2)-brane solutions with dissolved F -string charge. The interpretation of
such solutions was that of N BornInfeld strings [26,27] expanding into a sphere under
the influence of an external RR potential. Various arguments can then be used to identify
the string-like solution of BornInfeld theory as a fundamental string, but it is clear that
one lacks a proper description of this process from the point of view of the F -strings
themselves.
Many features of the supergravity solutions discovered by Costa et al. [19] and
Emparan [20] suggest that they should, indeed, be thought of as the proper gravitational
description of these dielectric branes for p = 4. 3 We will provide new evidence which
significantly strengthens the case for such a connection, and clarifies the relationship
between these geometries and the unstable solutions of [18].
In Sections 2 and 3 we review possible spherical solutions of the D6-brane worldvolume theory due to the presence of dissolved D4-brane and F -string charge, respectively.
We comment briefly on the more general case of the spherical Dp-brane with dissolved
F -string charge, for arbitrary p, the p = 2 case being qualitatively different to the other
examples. In Section 4, we turn to the supergravity solutions of [19,20]. To comment on
them, a brief review of their construction is necessary. We treat both the F -string and
D4-brane solutions in a unified way, since they have the same structure, and then compare
directly with the worldvolume analysis of Sections 2 and 3. In both cases, the solution is
consistent for precisely two values of the radius of the five- or two-dimensional sphere,
a point missed in the treatment of [19] because the limiting procedure used therein sends
one of the radii to infinity. Moreover, the functional form of these radii as a function of N ,
the number of D4-branes or F -strings, precisely matches that of the world volume calculation. That this should be the case is rather unexpected. Moreover, a brief consideration
of the smeared solutions of Emparan [20] shows that the atypical structure of the D2-brane
with dissolved F -string charge is also captured by the corresponding supergravity solution.
In Section 4.3, we comment on the energetics of these two consistent solutions, and
show that that with the smaller value of the radius has the lower energy, in agreement with
the worldvolume picture. We then turn to a consideration of off-shell configurations,
those with arbitrary values of the radius of the sphere. In general, such solutions exhibit
conical deficits which can be viewed as deficit branes providing the tension necessary
to hold the dielectric brane in equilibrium. We then argue that the tension of these deficit
branes gives us a handle on the stability of the supergravity solutions, with the results that
the solution at the smaller radius is in fact stable at least to radial perturbations and
that the second solution is unstable. 4 These additional pieces of evidence exhibit the close
3 Corresponding, unsmeared supergravity solutions for different values of p should exist, but they cannot be

generated using the techniques of [19,20].


4 We are grateful to Roberto Emparan for pointing out these arguments to us.

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

221

relationship between the worldvolume and supergravity descriptions. We believe further


study of this connection will greatly enhance our understanding of the dielectric effect.

2. Spherical D6-branes from D4-branes the worldvolume theory


Consider, then, N flat D4-branes polarized by a 7-form RR potential into a D6-brane.
From the point of view of the D6-brane, such a configuration is described by a solution with
topology M5 S 2 and N units of dissolved D4-brane flux. The flat space approximation 5
analogous to that considered by Myers [22] a flat background geometry and a constant
8-form field strength does indeed allow for such solutions. We will see that the
supergravity solutions of Costa et al. [19] and Emparan [20] correspond to this simple
model with a surprising degree of accuracy.
The flat ten-dimensional metric is written as
 
 
ds 2 = ds 2 M5 + dr 2 + r 2 d22 + ds 2 E2 ,
(2.1)
where dn2 denotes the round metric on a unit n-sphere, and the worldvolume of the
D6-brane is taken to be the obvious M5 S 2 . We are interested in static solutions of
the worldvolume theory for r = R = const. The constant background 8-form field strength
and corresponding 7-form potential is chosen to be
 
 
F[8] = 2f r 2  M5 dr  S 2 ,
(2.2)
 2
2 3  5
C[7] = f r  M  S ,
(2.3)
3
where (M) denotes the volume form on the space M, and where a convenient choice of
gauge has been made.
The D6-brane action is a sum of DiracBornInfeld [28] and ChernSimons [25,29,30]
terms:
S = SBI + SCS ,


T6
SBI =
e det(P [G]ab + Fab ),
gs
D6

 
T6
SCS =
P
C[n] eF ,
gs

(2.4)
(2.5)

(2.6)

D6

where we have set the KalbRamond 2-form to zero, F is the Abelian BornInfeld 2-form
field strength, P [ ] denotes a pull-back to the worldvolume of the spacetime fields,
T6 = 6 = ((2)6 7/2 )1 and = 2 . The ChernSimons action is a sum of terms
containing all the RR potentials, C[n] , for 1  n  7:





T6  
SCS =
(2.7)
P C[7] + P C[5] F + ,
gs
D6
5 Approximation in the sense that it is not a consistent supergravity background.

222

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

which are the relevant couplings of these potentials to Dp-branes for 0  p  6. The Born
Infeld field strength is then chosen such that the second term in (2.7) mimics the coupling
of N D4-branes to C[5] . With
F=

N  2
 S ,
2

we have

(2.8)


P [C[5] ] F = NT4

T6
M5 S 2

P [C[5] ],

(2.9)

M5

where we have used the relation 2Tp+2 = Tp .


Substituting for this BornInfeld field strength, and for the radius r = R, in the
background metric (2.1), we find that the action (2.4) gives the potential [22]


4T6 V4

R 4 + 14 2 N 2 23 f R 3 ,
V(R) =
(2.10)
gs
where we have taken the dilaton to vanish, and where Vn denotes the volume of En . If we
introduce the dimensionless worldvolume quantities
D = f R,
1
ND = f 2 N,
2

gs
4 + N 2 2 3 ,
VD =
f 2 V = D
D
4T6 V4
3 D

(2.11)
(2.12)
(2.13)

then the dimensionless potential, VD , has extrema when the following condition holds:
2
2
4
ND
= D
D
.

For ND <

(2.14)

1
2

there are two nontrivial extrema, given by




1

2
2 ,
D
(2.15)
= 1 1 4ND
2
as shown in Fig. 1. There are thus two solutions describing spherical D6-branes, one a
maximum of the potential, the other a local minimum. In Fig. 2(a) we show how the
stationary points of the potential change as the number of D4-branes change. Fig. 2(b)
shows the value of the potential at these extrema.
2 in which case
Myers [22] considers the case where ND < 1/2 in the limit ND D
4

3
one may expand the potential (2.13) as VD ND + 2NDD 23 D
+ . The single nonzero
extremum is then at D ND so, in this limit, we lose the information about the maximum.
A similar limit was taken in the supergravity solution of [19] which again obscured the fact
that there is a second consistent solution. We shall see below that, prior to any limit being
taken, the allowed radii of the supergravity solution have precisely the same functional
form as (2.14). We believe this to be strong evidence that the worldvolume and supergravity
pictures are describing the same phenomenon.

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

223

Fig. 1. This plot shows the potential energy, (2.13), of the spherical D6-brane for different values of
dissolved D4-brane charge, ND (2.12). The top curve is for ND = 0 and the lower one for ND = 12 .
The potential has been shifted by a constant such that the energy at D = 0 is zero.

Fig. 2. Here we show how the charge, ND (2.14), and energy, VD (2.13), at the stationary points vary
with the radius of the spherical D6-brane with dissolved D4-brane charge.

3. Spherical D6-branes from F -strings the worldvolume theory


We now turn to the case considered by Emparan [21], in which N F -strings expand
into a D6-brane with topology M2 S 5 . Taking the same action as in (2.4), we place the
D6-brane at constant r = R in the flat spacetime given by
 
 
ds 2 = ds 2 M2 + dr 2 + r 2 d52 + ds 2 E2 .
(3.16)

224

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

The BornInfeld field strength is taken to be


 
F = E M2 ,

(3.17)

where E is a constant electric field, and we take the constant RR 8-form field strength and
corresponding 7-form potential to be
 
F[8] = 5h M2 dr r 5 (5 ),

(3.18)

5  
C[7] = h M2 r 6 (5 ).
6

(3.19)

The action (2.4) then reduces to


S =

T6 5 l
gs


 
5h
dt R 5 1 2 E 2 R 6 dt L,
6

(3.20)

where l is the length of the string and n denotes the volume of a unit n-sphere.
Introducing the conjugate momentum, D = L/E, we find that the Hamiltonian,
(ED L), is given by [21]
T6 5 l
H=
gs



R 10

gs
+
T6 5

2
D2


5h 6
R .
6

(3.21)

To relate D to N , the number of dissolved F -strings, we take R to zero to obtain a stringlike state with energy per unit length D/ = DTF (TF
being the F -string tension), so we
simply associate D = N [21]. Now we use R11 = gs , 5 = 3 , and introduce the
following dimensionless quantities
F = hR,

(3.22)

NF = 8h5 R11 2 N,

H
5
VF = 83 R11 h5
= F10 + NF2 F6 .
V1
6

(3.23)
(3.24)

The extrema of the potential (3.24) are given by solutions of


NF2 = F8 F10 .

(3.25)

Again there is a maximum charge, above which there are no stable solutions. To calculate
it, we use the fact that at this maximum charge, the two extrema
 merge into a point
of inflection, so VF = VF = 0 at the extrema. This gives NF  (4/5)4 (4/5)5 . In
Fig. 3 we show how the potential changes with the number of F -strings dissolved in the
D6-brane. How the location of the extrema changes with NF is shown in Fig. 4(a), there
clearly being two extrema for NF less than the maximum value. This system is seen to be
qualitatively the same as the one considered in Section 2, the D4-brane blowing up into a
D6-brane.

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

225

Fig. 3. This plot shows the potential energy, (3.24), of the spherical D6-brane for different values
of dissolved F -string charge,
 NF (3.23). The top curve is for NF = 0 and the lower one for the
maximum charge NF = (4/5)4 (4/5)5 . The potential has again been shifted by a constant to
ensure that the zero of energy is at F = 0.

Fig. 4. Here we show how the charge, NF (3.25), and energy, VF (3.24), at the stationary points vary
with the radius of the spherical D6-brane with dissolved F -string charge.

3.1. Spherical Dp-branes from F -strings


The above analysis is not peculiar to D6-branes, in that spherical Dp-branes with
dissolved F -string charge exist for arbitrary values of 2  p  8 [21]. It is easy to show
that the dimensionless potential in the general case has the form [21]

226

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236


VF =

2(p1)

+ NF2

p1 p
F ,
p

(3.26)

in terms of dimensionless quantities analogous to those defined in (3.22) and (3.25) above.
The extrema of this potential are given by solutions of
2(p2)

NF2 = F

2(p1)

(3.27)

so for p > 2 there are always two extrema. The p = 2 case is qualitatively different,
however. There is no minimum in this case, the single nontrivial extremum being at

F = 1 NF2 ,
(3.28)
and this is a maximum of the potential. We will see that the relevant (smeared) supergravity
solution describing this configuration has a similar structure.

4. The supergravity description


The eleven-dimensional starting point of the discussion of [19,20] is the double analytic
continuation of the rotating black M2- and M5-branes: 6

2 

 d1 
kl cosh 2
2
2/d
2

sin d
+ f d +
ds = H
ds M
r d f
 2


f 2
2/d dr
2
2
2
2
2

sin d + cos dd1


+H
(4.29)
,
+ r d +

f
f
= (3, 6) and (6, 3) for the M2- and M5-branes, respectively, and where
where (d, d)
H =1+

k sinh2
r d

f =1

k
r d

l 2 cos2
r d l 2 r d2 k
,

=
1

.
f =
r2
r d
For the M2-brane, the 3-form potential is given by

k sinh  2  
 M cosh d l sin2 d ,

r d H
whereas for the M5-brane it is



 
k sinh
l2
l
cos3 1 2 cosh d 2 d  S 2 .
A[3] =

r
r
A[3] =

(4.30)

(4.31)

(4.32)

The location, rH , of the Euclidean horizon or bolt is given by the zero of f, so that

 2
d2
rH l 2 ,
k = rH
(4.33)
6 The standard versions of which are to be found in [31,32].

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

227

and the Euclidean angular velocity, , is


=

l
.
2 l2)
cosh (rH

(4.34)

To avoid a conical singularity at r = rH , must be periodic with period 2R11 , where

2k cosh
R11 =
(4.35)
= gs .

d1 (d 2)l 2 r d3
dr
H
H
Finally, we must also consider the quantization of M2- and M5-brane charge which gives

k cosh sinh = cd N(R11 )d/3 ,

(4.36)

where wehave used the fact that the eleven-dimensional Planck length, lP , is given by
1/3
lP = gs
. N denotes the number of M2- or M5-branes, and the constant cd is

cd =

(2)2d/3
,

d

(4.37)

d+1

so that c6 = 8 and c3 = 1/2.


As explained in [19,20], the ten-dimensional geometry is found by reducing along orbits
of the Killing vector
K=

+B ,

(4.38)

the fixed point set of the corresponding isometry being {r = rH , = 0}. To dimensionally
reduce, one identifies points separated by a distance of 2R11 along integral curves of K.
Introducing a new angular coordinate = B , which has standard 2 periodicity and
is constant along orbits of K, gives K = / . Then the ten-dimensional string frame
metric is

 2




dr
2
ds 2 = 1/2 H 3/d ds 2 Md1 + 1/2 H 13/d
+ r 2 d 2 + cos2 dd1

f
+ 1/2 H 13/d fr 2 sin2 d 2 ,

(4.39)

where



f
Bkl cosh sin2 2
(Br sin )2 ,
+H
=f 1+

f
r d f

(4.40)

and the dilaton is given by 7


e2 = 3/2 H 3/d .

(4.41)

In general, the metric (4.39) has a conical singularity in the r plane at r = rH . An


analysis of the rrH limit shows that the deficit angle is given by 2(1 a), where
a=

H (d 2)l 2 rH 1
1 dr
.
2 l2)
2 l + B cosh (rH

(4.42)

7 We will not write down the dimensional reduction of the 3-form potentials (4.31) and (4.32) since they are
superfluous to our discussion.

228

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

Then by imposing
B=

H + (d 2)l
1
dr
=
,
R11
2rH (rH + l) cosh

(4.43)

we have a = 1, and so no conical singularity. As described in [19,20], the ten-dimensional


metric has the correct form to describe an F -string or D4-brane expanding into a D6-brane.
The fixed point set {r = rH , = 0} of K leads to a naked singularity in the metric (4.39),
the singular surface being identified with the spherical part of the worldvolume of the
F -string or D4-brane: M2 S 5 or M5 S 2 , respectively. Moreover, an analysis of the
near-core, {r = rH , = 0}, spacetime shows that, in both cases, the radius of the sphere
into which the F -string or D4-brane expands is precisely equal to rH [19,20].
4.1. Smeared solutions: F -strings expanding into a Dp-brane
We have seen in Section 3.1, from the worldvolume point of view, that the case of
F -strings expanding into a D2-brane is qualitatively different to that of expansion into
any other Dp-brane: there is only one static solution and it is unstable. Here we obtain
a supergravity description, albeit a smeared one [20], describing N F -strings expanding
into a Dp-brane, for arbitrary 2  p < 6. We will see that the p = 2 is a special case in
supergravity as well. The starting point in this case is a smeared version of the M2-brane
metric (4.29):

2 

 
kl cosh 2

sin

ds 2 = H 2/3 ds 2 M2 + f d +
r p f



 6p  dr 2
f 2
1/3
2
2
2
2
2
2

sin d + cos dp1


ds E
+
+ r d +
+H
f
f
(4.44)
where the functions H, f, f and are given by (4.30), but with d replaced by p. The
solution is thus smeared over the obvious 6 p transverse directions. The relations (4.33)
and (4.35) are unchanged up to the replacement of d with p, as is the 3-form potential
(4.31). The charge quantization condition is then
k cosh sinh = cp N

(R11 )2
,
V6p

(4.45)

where
cp =

(2)4
.
p1+p

(4.46)

Dimensional reduction proceeds as above, and gives rise to a D6-brane with topology
M2 E6p S p1 , so that T -duality along the flat transverse directions gives the desired
solution describing N F -strings expanding into a Dp-brane with topology M2 S p1 [20].
Again, the potential conical singularity is avoided if B is given by (4.43), up to the
replacement of d with p.

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

229

4.2. The radius of the sphere


The question with which we are concerned here is, for what values of rH does the
solution (4.39) make sense? That is, what values of rH are allowed and, furthermore, do the
allowed values of rH match the radii of the polarized branes discussed in Sections 2 and 3?
To address this question, we need to think about which quantities to hold fixed as we vary
the values of N and B. At first sight, one might consider holding the rotation parameter, l,
fixed, but this is incorrect; we should rather be considering the set of possible solutions for
2 = (2)4 R 2 3 . In other words, we should fix R , in order that our family
constant 210
11
11
of ten-dimensional solutions have the same Newtons constant.
We thus use the four relations (4.33), (4.35), (4.36) and (4.43) to eliminate l, and k,
and to give a relation between rH , N , B and R11 . First rewrite (4.43) as
d
(4.47)
k = 0,
2
Now take (4.33) to eliminate l from (4.47) and (4.35). Then eliminate by using (4.35),
giving

d1
d2
lrH
l 2 rH
+ kB cosh rH

1
(d 2)2 (1 BR11 )2 k 2
4

d

2d
+ (d 2)(1 BR11 )2 + 1 rH
k + (1 BR11 )2 1 rH
= 0.

(4.48)

We may now solve this quadratic for k and substitute into (4.36). We note that the
requirement k  0 gives the restriction
0  BR11  1,

(4.49)

so that B = 1/R11 is the maximum magnetic field [18]. However, for the ten-dimensional
KaluzaKlein picture to be valid, we need all ten-dimensional length scales to be larger
than the compactification scale, R11 . The magnetic field gives just such a length scale, 1/B,
so the solution should have a reasonable ten-dimensional description only for BR11  1.
A more detailed examination of the region in which (the near-core limit of) these solutions
is ten-dimensional can be found in [19]. At any rate, we now find

d
k = rH
,

(4.50)

where we have defined




2
d 2)(1 BR11 )2 (d 2)(1 BR11 )2 1 .
1 + d(
=
(d 2)2 (1 BR11 )2
(4.51)
With



d 2
1
1+
,
A=

we have
2 cd2

1
A2d

2
R11

2d/3

(4.52)

rH
N +
AR11
2

2d2

rH

AR11

2d4

= 0.

(4.53)

230

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

Fig. 5. In order to find rH for a given set of parameters, R11 , B and N we need the dimensionless
functions A(BR11 ), (BR11 ). (We plot A1 (BR11 ) for clarity.) The scales on the left of the graphs,
and the solid lines, correspond to d = 3 (D4 D6) and the right-hand scales, and dashed lines,
correspond to d = 6 (F1 D6).

The functions A(BR11 ) and (BR11 ) which control this equation, are plotted in Fig. 5. By
introducing the dimensionless quantities
rH
,
AR11


cd
d/3
NS =
N,
2
Ad R11
S =

(4.54)
(4.55)

relevant for the supergravity solutions, we finally find

2(d2)

NS2 = S

2(d1)

(4.56)

which is of precisely the same form as (2.14) for d = 3, and (3.25) for d = 6. Although
we cannot reproduce the correct numerical factors for any value of BR11 , it is surprising
enough that the form of the equations are the same when one considers that the
worldvolume calculation did not make use of a consistent supergravity background.
In Fig. 6 we show how the radius, S , of the (d 1)-sphere depends on NS , the
dimensionless quantity defining the (d 2)-brane charge of the solution. There are various
noteworthy points. For NS less than some critical value, there are always two values of rH ,
which we denote by r+ and r (where r+ > r ). As long as the F -string or D4-brane
charge is not too large, then, there are two distinct supergravity solutions of the form
(4.39), for which the spacetime has no conical singularities. In the following subsection,
we will see that the solution with rH = r has the lower energy of the two. As we decrease
the charge, r+ increases, and r decreases. For zero charge essentially the situation
considered by Dowker et al. [18] there is only a single nontrivial value of rH , this

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

231

Fig. 6. Analogous plots to Figs. 2 and 4 for the supergravity case without conical singularities. They
show how the charge, NS (4.56), and energy, ES (4.64), vary as the radii of the static solutions
change. The solid line is for d = 3 (D4 D6) and the dashed line is for d = 6 (F1 D6).

solution being unstable [18]. It corresponds to = 0, whereas the trivial solution rH = 0


corresponds to k = 0, which we shall use as a background to calculate the energy of the
charged solutions.
We now have all the information we need to consider the B 0 limit for fixed N
and R11 (we consider the case d = 3). For BR11  1, BR11 and A1 BR11 (see
Fig. 5), so from (4.55) we have NS A2 0. Fig. 6 shows that there are two radii for
which NS 0, one at S 0 and the other at S 1. The former has NS S (see
Fig. 6) as S 0 so, from (4.54), we have rH /R11 A1 0. The latter has S 1 so
rH /R11 A . The solution with rH = r was identified as the stable solution of
the worldvolume theory, so as the magnetic field is reduced this radius goes to zero; the
effect disappears, as expected. The other solution, rH = r+ , is a charged version of the one
found by Dowker et al. [18] where it was shown that the small B limit gave rH , an
effectively planar brane.
It is not difficult to generalize the above analysis to include the smeared solution
describing N F -strings expanding into a Dp-brane for arbitrary 2  p < 6, as discussed in
Section 4.1. For p = 2, we need only everywhere replace d with p. The ten-dimensional
solutions thus obey
2(p2)

NS2 = S

2(p1)

(4.57)

which is of the same form as (3.27). The p = 2 case is simpler since the equation analogous
to (4.48) is then only linear in k. We have
2

,
k = 1 (1 BR11 )2 rH
(4.58)
so that, in terms of the relevant dimensionless quantities, we find

232

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

S =


1 NS2 ,

(4.59)

which is of the same form as (3.28). We note that this single consistent solution corresponds
to the larger of the two radii, rH = r+ , in the case of general p. As we will see below, it is
only the smaller radius which is stable, so it would seem that F -strings cannot expand into
a stable spherical D2-brane. At any rate, the atypical properties of the spherical D2-brane
with dissolved F -string charge are precisely captured by the corresponding (smeared)
supergravity solution. We expect that a localized solution should exist, and that it should
also exhibit the same behaviour.
4.3. Energetics and stability
Following [19], to compute the energy of the ten-dimensional solutions we make use
of the background subtraction method [33], where the relevant background is the k = 0
solution: the standard Melvin-like flux 7-brane solution [9] written in spherical oblate
coordinates. That is to say we define the zero of energy to be the k = 0 solution. We
also need to transform the metric (4.39) to the Einstein frame for this calculation, using
dsE2 = e/2 dsS2 . Specifically [33]





1
8
8
E = 2
(4.60)
d x h N K d x h0 N0 K0 ,
210

where hij is the metric induced on a constant r slice of the constant time slice, h is
its determinant and the integration is performed as r . The lapse function, N , and
extrinsic curvature, K, of the constant r slice, are given by
N=

gt t ,

1
K = hij r hij ,
grr

(4.61)

and a 0 subscript denotes the corresponding quantities for the reference k = 0


background.
We find 8
E=

d+1
Vd2 
2 3
(2)4 R11


d sinh2 + d + 2 k.

(4.62)

Defining the dimensionless energy, ES , as

ES =

cd (2)2(2d/3) 3

d2
Ad Vd2 R11

gives

ES = Sd2

E,



2 2
1 + S ,
d

(4.63)

(4.64)

8 The energy of these dielectric branes is thus slightly higher than the corresponding energy of the standard
black F -string and D4-brane solutions, which have a factor of (d + 1) rather than (d + 2) [34].

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

233

In both cases, the solution with rH = r


which is plotted in Fig. 6 for the two values of d.
has lower energy than that with rH = r+ , which agrees with the pictures presented in
Sections 2 and 3.
Although the solution with rH = r has lower energy than that with rH = r+ , this does
not imply stability. We can argue, however, that it is in fact stable to radial perturbations by
looking at off-shell configurations, that is by holding B and N constant and varying rH .
This still yields a supergravity solution but in general there will be a conical singularity
on the surface of, and inside, the dielectric sphere. The interpretation of this singularity is
that it provides the tension necessary to hold the dielectric sphere at that radius. One can
associate a deficit brane with this conical singularity and, just as for a cosmic string [35],
its tension is proportional to the deficit angle, 2(1 a), with a as in (4.42). So if the
deficit angle is positive, then the deficit brane has positive tension, and is having to pull
on the dielectric sphere to keep it static; the dielectric sphere wants to expand. Similarly,
when the tension is negative, the sphere wants to contract.
To compute the tension (mass per unit volume) of the deficit brane, we take the rrH
limit of the metric (4.39) and again make use of the background subtraction method [33],
as applied to the resulting spacetime. As explained in [33], the relevant background in this
case is a spacetime identical in all respects except that it is free of the conical singularity.
Denoting the energy in which we are interested by T , and applying the formulae (4.60)
and (4.61) to the case at hand, we find
T =
=

 2

d1
rH
l + B cosh rH
2
(2)4 R11 3

d2
d+1
Vd2 R11
TS
(2)4 3
d+1
Vd2


l 2 (1 a)
(4.65)

which, as promised, is proportional to the deficit angle. TS is the mass of the deficit brane
in terms of the dimensionless quantities. It is interesting to note that we can derive this
expression using different techniques. In particular, we have considered methods analogous
to those used in [36] to compute the effective energymomentum tensor of the conical
singularity between two collinear black holes in four dimensions; and these methods give
exactly the same result for the mass T .
To analyse the form of (4.65) further, we fix B and N , and substitute for l, and
k using the expressions (4.33), (4.35) and (4.36). Of course, we no longer have access
to the expression (4.43) for B, since this was derived to ensure the absence of conical
singularities. Indeed, we can no longer solve for or k analytically, and must proceed
using numerical methods. The plot of the mass, TS , as a function of S is shown in Fig. 7,
for both the F -string and D4-brane solutions.
As argued previously, a deficit brane with positive tension indicates that the dielectric
sphere wants to expand, and negative tension implies that it wants to contract. We see from
Fig. 7, then, that at rH = r+ the sphere is unstable to both expansion and contraction,
whereas at rH = r the sphere is stable. The sphere at rH = r+ would decay either into
the sphere at rH = r , or into a sphere which expands forever. We should note that these
arguments apply only to radial perturbations.

234

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

Fig. 7. This plot shows the dimensionless mass (4.65) of the deficit brane for d = 3 (solid line,
left-hand scale), d = 6 (dashed line, right-hand scale), and for some specific values of B and N . The
inset is a magnification of the region around rH = r . The arrows show whether the sphere wants to
expand or contract as one moves away from the zeros of TS .

5. Discussion
We have provided strong evidence that the supergravity solutions of Costa et al. [19]
and Emparan [20] are describing the dielectric effect whereby F -strings or D4-branes
expand into spherical D6-branes. We have presented the two supergravity solutions in a
unified manner showing that, in both cases, there are two distinct static solutions which
are free of conical singularities. Moreover, the radius of these solutions has precisely the
same functional form as that implied by the worldvolume analyses of Emparan [21] and
Myers [22]. Of course, since the solutions in question do not preserve any supersymmetry,
one should ask whether they would be modified by higher derivative and/or string loop
corrections. It seems likely that they would be so modified, and presumably the spherical
worldvolume solutions discussed above would also be affected by such corrections.
Certainly one cannot trust the solutions in regions of very large curvature.
On the supergravity side we have seen that the energy of the rH = r solution was
lower than that with rH = r+ , although this alone does not guarantee stability. To address
the latter issue, we considered the tension of the deficit brane formed by static solutions
away from rH = r , which indicated that the rH = r solution was stable and that with
rH = r+ was unstable. Thus we see that the supergravity solutions capture the worldvolume
properties with remarkable accuracy.
To further strengthen the interpretation of the supergravity solutions, we considered the
special case of N F -strings expanding into a D2-brane, for which the worldvolume analysis
indicates that no stable dielectric solution should exist. Although the supergravity solution
for this configuration was smeared in certain transverse directions, we have seen that indeed
there was no stable solution.

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

235

A potentially interesting area of research would be to look at the possible tunnelling of


these configurations. From the worldvolume perspective, it is clear that tunnelling should
occur from the classically stable configuration at rH = r to some expanding solution. On
the supergravity side, in the case with no dissolved charge, instantons describing such
a process exist, and have been described in some detail by Dowker et al. [18]. They
correspond to the nucleation of spherical branes within a fluxbrane. It would be interesting
to look for analogous instantons in the case with dissolved charge.

Acknowledgements
We would like to thank Miguel Costa, Bert Janssen and Simon Ross for useful
discussions and comments on a draft of this paper. We are also indebted to Roberto
Emparan for explaining the stability arguments and urging us to look at the smeared
solutions. D.B. is supported in part by the EPSRC grant GR/N34840/01 and PMS by
PPARC.

References
[1] M.A. Melvin, Pure magnetic and electric geons, Phys. Lett. 8 (1964) 65.
[2] G.W. Gibbons, K. Maeda, Black holes and membranes in higher-dimensional theories with
Dilaton fields, Nucl. Phys. B 448 (1988) 741.
[3] J.G. Russo, A.A. Tseytlin, Constant magnetic field in closed string theory: an exactly solvable
model, Nucl. Phys. B 448 (1995) 293, hep-th/9411099.
[4] J.G. Russo, A.A. Tseytlin, Magnetic flux tube models in superstring theory, Nucl. Phys. B 461
(1996) 131, hep-th/9508068.
[5] A.A. Tseytlin, Closed superstrings in magnetic flux background, Nucl. Phys. Proc. Suppl. 49
(1996) 338, hep-th/9510041.
[6] J.G. Russo, A.A. Tseytlin, GreenSchwarz superstring action in a curved magnetic Ramond
Ramond background, JHEP 9804 (1998) 014, hep-th/9804076.
[7] R. Emparan, Composite black holes in external fields, Nucl. Phys. B 490 (1997) 365, hepth/9610170.
[8] C. Chen, D.V. Galtsov, S.A. Sharakin, Intersecting M-fluxbranes, Grav. Cosmol. 5 (1999) 45,
hep-th/9908132.
[9] M.S. Costa, M. Gutperle, The KaluzaKlein Melvin solution in M-theory, JHEP 0103 (2001)
027, hep-th/0012072.
[10] G.W. Gibbons, C.A.R. Herdeiro, The Melvin universe in BornInfeld theory and other theories
of non-linear electrodynamics, Class. Quantum Grav. 18 (2001) 1677, hep-th/0101229.
[11] P.M. Saffin, Gravitating fluxbranes, gr-qc/0104014.
[12] M. Gutperle, A. Strominger, Fluxbranes in string theory, hep-th/0104136.
[13] J.G. Russo, A.A. Tseytlin, Magnetic backgrounds and tachyonic instabilities in closed
superstring theory and M-theory, hep-th/0104238.
[14] J.F. Sparks, KaluzaKlein branes, hep-th/0105209.
[15] P.M. Saffin, Fluxbranes from p-branes, hep-th/0105220.
[16] F. Dowker, J.P. Gauntlett, D.A. Kastor, J. Traschen, Pair creation of dilaton black holes, Phys.
Rev. D 49 (1994) 2909, hep-th/9309075.

236

D. Brecher, P.M. Saffin / Nuclear Physics B 613 (2001) 218236

[17] F. Dowker, J.P. Gauntlett, G.W. Gibbons, G.T. Horowitz, The decay of magnetic fields in
KaluzaKlein theory, Phys. Rev. D 52 (1995) 6929, hep-th/9507143.
[18] F. Dowker, J.P. Gauntlett, G.W. Gibbons, G.T. Horowitz, Nucleation of p-branes and
fundamental strings, Phys. Rev. D 53 (1996) 7115, hep-th/9512154.
[19] M.S. Costa, C.A.R. Herdeiro, L. Cornalba, Flux-branes and the dielectric effect in string theory,
hep-th/0105023.
[20] R. Emparan, Tubular branes in fluxbranes, hep-th/0105062.
[21] R. Emparan, BornInfeld strings tunneling to D-branes, Phys. Lett. B 423 (1998) 71, hepth/9711106.
[22] R.C. Myers, Dielectric branes, JHEP 12 (1999) 022, hep-th/9910053.
[23] W. Taylor, M. Van Raamsdonk, Multiple Dp-branes in weak background fields, Nucl. Phys.
B 573 (2000) 703, hep-th/9910052.
[24] S.P. Trivedi, S. Vaidya, Fuzzy cosets and their gravity duals, JHEP 0009 (2000) 041, hepth/0007011.
[25] M.R. Douglas, Branes within branes, hep-th/9512077.
[26] C.G. Callan, J.M. Maldacena, Brane dynamics from the BornInfeld action, Nucl. Phys. B 513
(1998) 198, hep-th/9708147.
[27] G.W. Gibbons, BornInfeld particles and Dirichlet p-branes, Nucl. Phys. B 514 (1998) 603,
hep-th/9709027.
[28] R.G. Leigh, DiracBornInfeld action from Dirichlet sigma model, Mod. Phys. Lett. A 4 (1989)
2767.
[29] M. Li, Boundary states of D-branes and Dy-strings, Nucl. Phys. B 460 (1996) 351, hepth/9510161.
[30] M. Green, J.A. Harvey, G. Moore, I-brane inflow and anomalous couplings on D-branes, Class.
Qauntum Grav. 14 (1997) 47, hep-th/9605033.
[31] M. Cvetic, D. Youm, Rotating intersecting M-branes, Nucl. Phys. B 499 (1997) 253, hepth/9612229.
[32] C. Cski, J. Russo, K. Sfetsos, J. Terning, Supergravity models for (3 + 1)-dimensional QCD,
Phys. Rev. D 60 (1999) 044001, hep-th/9902067.
[33] S.W. Hawking, G.T. Horowitz, The gravitational Hamiltonian, action, entropy, and surface
terms, Class. Quantum. Grav. 13 (1996) 1487, gr-qc/9501014.
[34] M.J. Duff, H. Lu, C.N. Pope, The black branes of M-theory, Phys. Lett. B 382 (1996) 73,
hep-th/9604052.
[35] A. Vilenkin, E.P.S. Shellard, Cosmic Strings and other Topological Defects, CUP, 1994.
[36] M.S. Costa, M.J. Perry, Interacting black holes, Nucl. Phys. B 591 (2000) 469, hep-th/0008106.

Nuclear Physics B 613 (2001) 237259


www.elsevier.com/locate/npe

The d = 6 trace anomaly from quantum field theory


four-loop graphs in one dimension
Agapitos Hatzinikitas a,b , Renato Portugal c
a University of Crete, Department of Applied Mathematics, L. Knosou-Ambelokipi, 71409 Iraklio Crete, Greece
b University of Athens, Nuclear and Particle Physics Division, Panepistimioupoli GR-15771 Athens, Greece
c Laboratrio Nacional de Computao Cientfica, Av. Getulio Vargas, 333,

Petrpolis, RJ, Brazil, Cep 25651-070


Received 22 December 2000; accepted 27 February 2001

Abstract
We calculate the integrated trace anomaly for a real spin-0 scalar field in six dimensions in a
torsionless curved space without a boundary. We use a path-integral approach for a corresponding
supersymmetric quantum mechanical model. Weyl ordering the corresponding Hamiltonian in phase
space, an extra two-loop counterterm 18 (R + g ij kil ljk ) is produced in the action. Applying a
recursive method we evaluate the components of the metric tensor in Riemann normal coordinates in
six dimensions and construct the interaction lagrangian density by employing the background field
method. The calculation of the anomaly is based on the end-point scalar propagator and not on the
string inspired center-of-mass propagator which gives incorrect results for the local trace anomaly.
The manipulation of the Feynman diagrams is partly relied on the factorization of four-dimensional
subdiagrams and partly on a brute force computer algebra program developed to serve this specific
purpose. The computer program enables one to perform index contractions of twelve quantum fields
(10 395 in the present case) a task which cannot be accomplished otherwise. We observe that the
contribution of the disconnected diagrams is no longer proportional to the two-dimensional trace
anomaly (which vanishes in four dimensions). The integrated trace anomaly is finally expressed in
terms of the 17 linearly independent scalar monomials constructed out of covariant derivatives and
Riemann tensors. 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.62.+v; 31.15.K; 02.40.K; 02.70
Keywords: Weyl anomaly in six dimensions; Riemannian geometry; Computational techniques; Path integral
methods

1. Introduction
The trace anomaly, namely the breaking of classical conformal invariance of gravity
actions under Weyl rescaling of the metric:
E-mail addresses: ahatzini@tem.uoc.gr (A. Hatzinikitas), portugal@lncc.br (R. Portugal).
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 1 ) 0 0 0 9 8 - 0

238

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

gij (x) (x)gij (x),

(1)

has a long history with numerous applications and implications to high-energy physics,
general relativity and statistical mechanics. The literature on this subject is vast and in
this brief review we will only concentrate on those aspects which are associated with the
present problem. For a historical review the reader is advised to consult [1].
Interesting different methods have been developed to investigate and calculate Weyl
anomaly in four and higher dimensions. The authors in [2] were able to express the
integrated trace anomaly in four dimensions as a linear combination of two invariants,
the square of Weyl tensor
Cij kl C ij kl = Rij2 kl 2Rij2 + 13 R 2

(2)

and the only parity-even candidate


E4 = Rij kl R ij kl = Rij2 kl 4Rij2 + R 2

(3)

which is proportional to the well-known GaussBonnet topological density and denotes


the dual. To be more concrete, the gravitational contribution to the anomaly depends on
only two constants (call them and ) and is expressed as


g ij Tij  = C ij kl Cij kl + 23 R + E4 .
(4)
The numerical values of the constants are
1 1 1
1 1 1
,
= 2
,
2 30 64
90 64
and can also be found using the Feynman diagram scheme in [4].
It was soon realised [3] that these invariants were manifested in the t-independent
b2 coefficient of the SchwingerDeWitt asymptotic expansion of the heat kernel of the
appropriate differential operators. The four- (and partially the six-) dimensional anomaly
was later rederived by Bonora et al. [5] who established the connection between Weyl
anomalies and cocycles by relying on a cohomological method and using the Wess
Zumino consistency condition. Although these authors explicitly specified the invariants
(see Appendix A.2 for their classification) they did not express the six-dimensional trace
anomaly in terms of these invariants. From the representation theory point of view, Fulling
et al. [6] were also able to determine the number of independent scalar monomials of
each order and degree up to twelve in derivatives of the metric. The first explicit result
was given in the literature by Gilkey [7] and later by Avramidi [8], who by making some
modifications and innovations to the heat kernel or proper time method, presented
a new covariant nonrecursive procedure and found the one-loop effective action in the
presence of arbitrary background fields in six and eight curved spacetime dimensions.
Later Deser and Schwimmer [9] reexamined the different origin of the topological (type-A)
versus local conformal scalar polynomials involving powers of the Weyl tensor and its
derivatives (type-B) contributions to the anomaly in general dimensions.
In the present work we follow the supersymmetric quantum mechanical approach first
pioneered by Alvarez-Gaum and Witten [10] and used to compute chiral anomalies.
=

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

239

According to this method the operators 5 , , x , are represented by operators of


a corresponding quantum mechanical model, and by turning these operator expressions
into path integrals, one finds that anomalies of quantum field theories can be written in
terms of Feynman diagrams for certain sigma models on the worldline. Bastianelli and
van Nieuwenhuizen applied this method to trace anomalies [11]. These authors used mode
regularization, a scheme widely used at the time. Subsequent work by de Boer et al. [12]
showed how to use time-slicing and gave a completely and unambiguous derivation of trace
anomalies in terms of path integrals, using as input Einstein Hamiltonians. In this article
we apply the regularization method of [12] to the calculation of trace anomalies, following
the set up of [11].
The outline of the paper is as follows. In the next section we present explicitly one of
the basic ingredients of the background field method, namely the expansion of the metric
tensor components in six dimensions in Riemann normal coordinates. The method we use
is recursive and enables one with the application of (9) and (11) to determine the expansion
of the metric tensor up to the desired order.
Section 3 begins with a very rapid introduction to trace anomalies from the onedimensional path integral point of view. We write down the interaction Lagrangian density
and the propagators of the fields involved.
Section 4 is devoted to the calculation of the perturbative expansion depicting at the
same time the Feynman diagrams associated to each vertex.
A vital tool in our journey is a computer algebra algorithm which proved to be very
efficient especially in finding the contribution of I6 vertex with twelve quantum fields. Thus
Section 5 is devoted to a brief description of this program. We illustrate its capabilities by
applying it to the I7 interaction with eight quantum fields.
Our conclusions are given in Section 6. Several appendices follow to assist to a deeper
understanding of the technical obstructions the reader might face.

2. The recursive expansion of the metric components in RNC in six spacetime


dimensions
Before embarking on the background field method we discuss the expansion of the
metric tensor in Riemann normal coordinates (RNC).
RNC have the appealing feature that the geodesics passing through the origin have the
same form as the equations of straight lines passing through the origin of a cartesian system
of coordinates in euclidean geometry [13]. Locally no two geodesics through a point P
intersect at another point, and the power series solution of the geodesic equation is
y l = i1 s +



1 l
i1 i2 ik P i1 i2 ik s k ,
k!

(5)

k=2

where (il1 i2 ik )P are the generalized Christoffel symbols at the point P and the geodesics
through P which are straight lines are defined in terms of the arc length s by
y l = l s.

(6)

240

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

By induction, one can easily prove that


l
(i1 i2 ik2 ik1
ik ) = 0.

(7)

Paraphrasing Eq. (7), one can state that all symmetric derivatives of the affine connection
vanish at the origin in RNC.
In general a covariant second-rank tensor field on a manifold can be expanded according
to




1

Tk1 k2 ()i1 i2 in .
) = Tk1 k2 () +

Tk1 k2 (
(8)
n! i1 i2
in
n=1

The coefficients of the Taylor expansion are tensors and can be expressed in terms of the
l
components Rmnp
of the Riemann curvature tensor and the covariant derivatives Dk Tlm
l
and Dk Rmnp . Without much effort one can prove that
(i1 i2 in1 iln )k




n1
l
l

D(i1 Di2 Din2 R
=
in1 kin ) + (i1 i2 in2 in1 k in )
n+1
 l


(i1 i2 in3 in2
Rin1 kin ) l , k
 l


(i1 i2 in4 in3
Din2 Rin1 kin ) l , k
..
.





(i1 i2l Di3 Din2 R
in1 kin ) l , k ,

(9)

where the interchange of covariant and contravariant indices act independently and
symmetrization acts only on i indices. Expression (9) reproduces for various values of
n 1 the following results:
l
(i1 i2l )k = 13 R
(i1 ki2 ) ,
l
(i1 i2 i3l )k = 12 D(i1 R
i2 ki3 ) ,

2 l
l

(i1 i2 i3 i4l )k = 35 D(i1 Di2 R
i3 ki4 ) + 9 R(i1 i2 Ri3 i4 )k ,

l
l
(i1 i2 i3 i4 i5l )k = 23 D(i1 Di2 Di3 R
i4 ki5 ) D(i1 Ri2 ki3 Ri4 i5 ) ,

l
(i1 i2 i3 i4 i5 i6l )k = 57 D(i1 Di4 R
i5 ki6 )


1
l R

l
5 7D(i1 Di2 R
i3 i4 i5 i6 )k + D(i1 Di2 Ri3 ki4 Ri5 i6 )

16 l
iki Di4 R
l

+ 32 D(i1 R
i5 i6 ) 45 R(i1 i2 Ri3 i4 Ri5 i6 )k ,
2 3
..
.
(10)
The coefficients of (8) can be rewritten as:

(i1 i2 in ) T k1 k2 = D(i1 Di2 Din ) T k1 k2 + (i1 i2 in1 ) ink1 ) T k2 + k1 k2


1 In [14] there is a misprint for the n = 4 case. A minus sign is needed in front of the 2 -term.
9

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

241


+ (i1 i2 in2 ) in1
k1 Din ) Tk2 + k1 k2
..
.

+ (i1 i2 i3k1 Di4 Din ) T k2 + k1 k2


+ (i1 i2k1 Di3 Din ) T k2 + k1 k2 .

(11)

Expressions (9) and (11) compose the building blocks of the current recursive method
which produces the following results for different values of n:



(i1 i2 ) T k1 k2 = D(i1 Di2 ) T k1 k2 13 R
(i1 k1 i2 ) Tk2 + k1 k2 ,
(i1 i2 i3 ) T k1 k2 = D(i1 Di2 Di3 ) T k1 k2

+ (i1 i2 i3 )k1 T k2 + 2(i1 i2 k1 Di3 ) T k2 + k1 k2






13 R
(i1 k1 i2 Di3 ) Tk2 + k1 k2 ,
(i1 i2 i3 i4 )T k1 k2 = D(i1 Di2 Di3 Di4 ) T k1 k2

+ (i1 i2 i3 i4 )k1 T k2 + 3(i1 i2 i3 k1 Di4 ) T k2







+ 3(i1 i2 k1 Di3 Di4 ) T k2 13 R
i3 i4 ) T k2 + k2

+ k1 k2 ,
(i1 i2 i3 i4 i5 ) T k1 k2 = D(i1 Di2 Di3 Di4 Di5 ) T k1 k2





10
3 R(i1 k1 i2 Di3 Di5 ) Tk2 + 5D(i1 Ri2 k1 i3 Di4 Di5 ) Tk2
ik i Di5 ) T k2 + 2 D(i1 Di2 Di3 R


+ 3D(i1 Di2 R
i4 k1 i5 ) Tk2
3
3 1 4
R

23 D(i1 R
i2 k1 i3 i4 i5 ) Tk2



ik i R

+ D(i1 R
i4 i5 ) Tk2 + k2
2 1 3




T k2 + k2 R
+ 23 D(i1 R
i4 i5 )k1 k1 k2 ,
i2 i3
(i1 i2 i3 i4 i5 i6 ) T k1 k2
= D(i1 Di6 ) T k1 k2

+ (i1 i5 i6)k1 T k2 + 6(i1 i4 i5k1 Di6 ) T k2
+ 15(i1 i3 i4k1 Di5 Di6 ) T k2 + 20(i1 i2 i3k1 Di4 Di6 ) T k2



+ 14(i1 i2k1 Di3 Di6 ) T k2 + 10(i1 i3 i4k1 i5 i6 ) T k2 + k2





+ 10(i1 i2 i3k1 i4 i5 i6 ) T k2 + k2



+ 36(i1 i2 i3k1 i4 i5 Di6 ) T k2 + k2





+ 5(i1 i2k1 i3 i5 i6 ) T k2 + k2



+ 24(i1 i2k1 i3 i4 i5 Di6 ) T k2 + k2





+ 45(i1 i2k1 i3 i4 Di5 Di6 ) T k2 + k2

242

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259


+ 15 (i1 i2k1 i3 i4 i5 i6) T k2 + k2 + k2

+ (i1 i2k1 Di3 Di6 ) T k2 + k1 k2 .

(12)

If the second-rank tensor with components T k1 k2 is replaced by the metric components


gk1 k2 then the related covariant derivatives (provided we deal with a torsion free affine
connection) vanish and the above expressions are simplified. One could derive for n = 5
the result:




k1 i4 i5 k2 + 2 Di1 R
k1 i2 i3 R

(i1 i5 ) gk1 k2 = 43 Di1 Di3 R
(13)
i4 i5 k2 + k1 k2 .
On the other hand, for n = 6 one gets:
(i1 i2 i3 i4 i5 i6 ) gk1 k2 =

k1 i5 i6 )k2
Di4 R




+ 34
7 D(i1 Di2 Rk1 i3 i4 Ri5 i6 )k2 + k1 k2

10
7 D(i1

55


7 D(i1 Rk1 i2 i3 Di4 Ri5 i6 )k2

16
l
7 Rk1 (i1 i2 Ri3 i4 l Ri5 i6 )k2 .

(14)
Thus, plugging into (8) expressions (13) and (14) we end up with the following expansion
of the metric tensor in RNC:
k1 i1 i2 k2 i1 i2 + 1 Di1 R
k1 i2 i3 k2 i1 i3
gk1 k2 = gk1 k2 + 2!1 23 R
3!

i
i4
1
k1 i3 i4 k2 + 8 R

m
+ 4!1 65 Di1 Di2 R
9 k1 i1 i2 m Ri3 i4 k2


i

i5
1
k1 i4 i5 k2 + 2 Di1 R
k1 i2 i3 R

+ 5!1 43 Di1 Di3 R
i4 i5 k2 + k1 k2



17



+ 6!1 10
7 Di1 Di4 Rk1 i5 i6 k2 + 5 Di1 Di2 Rk1 i3 i4 Ri5 i6 k2 + k1 k2

i
8
i6
1


l
+ 11
2 Di1 Rk1 i2 i3 Di4 Ri5 i6 k2 + 5 Rk1 i1 i2 Ri3 i4 l Ri5 i6 k2


+ O i1 i7 .
(15)
The expression (15) will play a crucial role in calculating the contribution steming from
the quadratic Christoffel symbol term in the interaction Lagrangian. It is also in perfect
agreement with the closed formula for these coefficients which are encoded in the integral
representation of [15].

3. The integrated trace anomaly


Anomalies in (even) n-dimensional quantum field theories 2 are expressed in the
Fujikawa [16] approach as: 3


AnW = lim Tr J e R/h ,
(16)
0

2 For a manifold having odd dimensionality, one cannot form a scalar out of an odd number of derivatives.
3 We perform the usual redefinition of the scalar fields 
= g 1/4 which leaves the Jacobian invariant under
the similarity transformation g 1/4 .

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

243

for
(x)/ 
(y) and the regulator R
where J is the Jacobian
(y) = f (x)
of the fields
consistent anomalies is uniquely determined [17]. For local Weyl anomalies and for real
scalar fields one finds that
=H
1 h 2 R(x)
R
= 12 g 1/4 (x)
p i g ij (x)g
1/2 (x)
pj g 1/4 (x)
12 h 2 R(x),

g(x)
= det gij (x).

(17)

The second term in (17) is the well-known improvement potential term and the
dimensionless coefficient
(n 2)
=
4(n 1)
takes the value (n = 6) = 15 in the present case. The Hamiltonian is Einstein invariant
commutes with the generator
which means that H


) = 1 p i i (x)
G(
+ i (x)
p i
2 ih

of the infinitesimal target space diffeomorphisms xi xi + i (x).


Consider a spin-0 field which lives on an n-dimensional compact riemannian manifold
equipped with its standard (metric-compatible, torsion-free) connection and having no
boundary. In addition decompose the paths x i ( ) into a constant part x0i satisfying the free
field equations and a quantum fluctuating q i ( ) one vanishing at the time boundaries. The
Weyl anomaly for a real spin-0 field may represented by the euclidean quantum mechanical
path integral [4]:


AnW (s = 0, n) = lim Tr f (x) e H
0

1
= lim
0 (2 h
)n/2


dx0i


n 

 i   i  S int /h 
g x0 f x0 e
,

(18)

i=1

where f (x) is an arbitrary function, h1 S int = S1int + S2int and 4


S1int

1
=
2 h
1
=
h


gij (x0 + q) gij (x0 ) q i q j + bi cj + a i a j d

6 Rii1 i2 j q

+
+
+

q +

i1 i2

1
i1 i2 i3

12 Di1 Rii2 i3 j q q q

i1
i4
3
8

k
5! Di1 Di2 Rii3 i4 j + 9 Rii1 i2 k R i3 i4 j q q

 i

i5
4
1


k
6! Di1 Di3 Rii4 i5 j + 2 Di1 Rii2 i3 k R i4 i5 j + i j q q



5
17


k
7! Di1 Di4 Rii5 i6 j + 5 Di1 Di2 Rii3 i4 k R i5 i6 j + i j
 i
i6
8
1

k
k
l
+ 11
2 Di1 Rii2 i3 k Di4 R i5 i6 j + 5 Rii1 i2 k R i3 i4 l R i5 i6 j q q

4 The expectation value   means that all quantum fields must be contracted using the appropriate
propagators. A bar over the various geometrical quantities indicates that they depend exclusively on x0i .

244

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259





+ O q i1 q i7 q i q j + bi cj + a i a j d,
S2int

1
= h
8
1
= h
8

(1 4 )R + g ij l ki k lj (x0 + q) d


+ q i1 Di1 R
+
(1 4 ) R

(19)

0
1

1 i1
3! q

1 i1 i2

2! q q Di1 Di2 R

+
q i3 Di1 Di3 R

1 i1
4! q



q i4 Di1 Di4 R

+ g ij i1 l ik i2 k lj q i1 q i2


+ 2!1 g ij i1 i2 l ik i3 k lj + i1 l ik i2 i3 k lj q i1 q i3
 i j
i1 i2 i3 l ik i4 k lj 1 g ij i1 i2 l ik i3 i4 k lj
13 R
4


ij 1
l
g 3! i1 i3 ik i4 k lj + i1 l ik i2 i4 k lj



q i1 q i4 + O q i1 q i5 d.
(20)
As emphasized in [12], the above expression is the continuum limit of a rigorous discrete
result.
In the action (19), (bi , cj ) and i is a set of anticommuting and commuting LeeYang
ghosts, respectively [11]. Their existence is imposed by the integration over the momenta

pi ( ) in the transition from phase space to configuration space. A measure factor g is


then produced at each point of the path and by exponentiating it, introducing the LeeYang
ghosts, we are led to a perfectly regular term in the action. The existence of ghost fields
also removes ultraviolet divergences at higher loops and as a consequence all integrals are
finite.
The noncovariant h 2 terms, R and , which are essential for the general coordinate
invariance of the transition element are created [12] by Weyl ordering the Hamiltonian
of (17). The contribution of these terms to the trace anomaly is found by first Taylor
H
expanding them in RNC and then substituting the partial derivatives of the Christoffel
symbols by the polynomials of R and DR:
l (i k)i ,
i1 l (i2 k) = 23 R
2
1
l

l i2 ki3 ) ,
(i1 i2 i3 )k = 12 D(i1 R

l i ki )
(i1 i2 i3 l i4 )k = 35 D(i1 Di2 R
3 4

(21)

l (i i R
i i )k .
+ 29 R
1 2
3 4

(22)
(23)

The symmetrization of the various indices in the above identities is understood with the
inclusion of a 1/n! factor. All other extra terms in (19) and (20) are produced by expanding
the metric in Riemann normal coordinates with the help of (15).
Propagators are derived in closed form through the discretised configuration space path
integrals via a midpoint rule. In this way ambiguities arising from products of distributions
are resolved. Taking the continuum limits of the propagators one can read off the Feynman
rules. One then finds that ( ) is to be considered as a Kronecker delta ij and the

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

245

propagators depend on the discrete Heaviside function ii = 1/2. The continuous twopoint Green function may also be determined by the equation
2
( ) = ( ),
2
subjected to the boundary conditions on the interval [1, 0] (end-point approach):
(0, ) = (1, ) = (, 0) = (, 1) = 0.

(24)

(25)

The Feynman propagator is then formally found to be:


(, ) = ( + 1) + ( ) ( ).

(26)

The propagators of the various fields are proportional to h and given by:
q i ( )q j ( ) = h g ij (, ),

q i ( )q j ( ) = h g ij + ( ) = h g ij (, ),

q i ( )q j ( ) = h g ij + ( ) = h g ij (, ),

q i ( )q j ( ) = h g ij 1 ( ) = h g ij (, ),

(27)
(28)
(29)
(30)

b ( )c ( ) = 2 h g ( ),

(31)

a ( )a ( ) = + h g ( ).

(32)

ij

ij

All other Wick contractions of the fields vanish. Regarding the vertices they may be read
directly from the continuum S int given by (19) and (20).

4. The Feynman diagrams and the associated contributions


Perturbative expansion of eS /h , keeping only terms that cancel the ( h )3 factor
in the measure of the trace anomaly, provides us with the following distinct interactions as
well as connected and disconnected diagrams having at most four loops: 5
int

1.

1 5
I1 =
h 7!

 0

3 1
5
) 120
7! ( h

ii5 i6 j +
Di1 Di4 R

17
5



ii3 i4 k R
k i5 i6 j + i j
Di1 Di2 R

ii1 i2 k R
k i3 i4 l R
l i5 i6 j
+ 85 R




q i1 q i6 q i q j + bi cj + a i a j ( ) d
+

11

k
2 Di1 Rii2 i3 k Di4 R i5 i6 j

 2

+ 2Da D a R

R

5 The recipe the diagrams have been drawn is: a line closing on a single vertex is represented by a Green

function G(, ) with one integration variable while a line connecting two vertices stands for G(, ) and
indicates two integration variables.

246

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

1
60


ab + Da Db D a Dc R
ab + Da Db R
bc
Da Db R

1
120

1
120

1
120

ab + Da Db Dc D b R
ac + Da Db Dc D a R
bc
+ Da Db R

34 ab
3 abcd
dacb
ab D c D d R
Rabcd + 2R
5 R Rab + 2 R

abcd Dd Db R
abcd D(e Dd) R
ac + 12R
e abc
+ 2R

2
11
3
2
d

2 (Da Rbc ) + 4Da Rbc D Rdcab + 2 (Da Rbcde )

abcd D e R
ebcd + 3D a R
bcde De R
bcda
+ 3Da R

8 b ac
a edc R
bcde
ab R
+ 9R
5 Rab Rc R


e a f c R
bedf + 7 R

ab ef cd
abcd R
R
2 abcd Ref R


1
= ( h )3 7!5 120
12K4 + 12K5 + 4K6 2K7 + 8K8

+ 6K10 + 2K11 16K12 + 14K13 20K14 5K16 + 9K17

34
5 (2K4 + 2K5 + 4K6 2K7 + 8K8 + 3K10 + 5K11 K12 )

11
2
2 (11K13 10K14 + 3K15 ) + 5 (2K4 + 9K6 + 7K7 + 2K8 ) ;

(33)
2.
1
I2 = h
8

 0

4! (1 4 )Di1

= ( h )3 18

1 R
i i1 i2 j i3 l ik i4 k lj
Di4 R
3

+ 14 g ij i1 i2 l ik i3 i4 k lj



+ 3!1 g ij i1 i3 l ik i4 k lj + i1 l ik i2 i4 k lj


i1
i4
q q ( ) d

1
720 (1 4 )


 2
2Da R
ab Db R
2R
ab Da Db R

3 R

1 1 ab cde
Rbcde
540 2 R Ra

abcd R
a e c f R
bedf
+ 13 R

abcd R
ab ef R
cdef
+ 13
12 R


1
bc Dd R
ab d c + 1 D a R
bc 2
Da R
1728
17280

5
abcd De R
e bcd
2304 Da R

1
480

1259 abcd ef
Rab Rcdef
518400 R

1 abcd
ac
Dd Db R
7200 R

7 abcd
Rabcd
3200 R

7 abcd
e abc
De Dd R
1600 R

209 abcd e f
Ra c Rbedf
129600 R

bcde
Da R

2

a bc
1

5760 D R Dc Rab

a bcde
1
bcda
De R
360 D R

47 ab cde
Rbcde
86400 R Ra

23 abcd
e abc
Dd De R
4800 R

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259


= ( h )3 18

163
43200 K6
1
1800 K12

91
43
259200 K7 8100 K8
7
11
1440 K13 + 2160 K14

13
2880 K11
1
3600 K16

1
1200 K17

(34)


3.

247

   0 0 
 1 2



1 1 2
ii2 i3 j q i1 q i3 q i q j + bi cj + a i a j ( )
I3 =
Di1 R
12
2 h
1 1




ki5 i6 l q i4 q i6 q k q l + bk cl + a k a l ( )
Di4 R




ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
+ 5!1 R
 i

i6
3
ki5 i6 l + 8 R

m
Di3 Di4 R
9 ki3 i4 m R i5 i6 l q q




q k q l + bk cl + a k a l ( ) d d
= 12

3 1
1
bcde D e R
bcda
) 10 Da R
144 ( h

2

abcd
15 Da Rbc Dd R

1
ab c
48 Da R D Rcb

5!


1 108

1
abcd D e R
ebcd
10 Da R

1
2
30 (Da Rbc )

1
ab
48 Da R Db R

1
2
192 (Da R)

15

abcd +
abcd R
R

72
ab
15 Rab R

b acde
432
Rbcde
30 Da D R

144
bcad R
cd
15 Da Db R

64
b d eaf c
15 Rabcd R e f R

64 b ca
15 Rab Rc R

R

32 R

432 a
bcde R
acde
30 D Db R

144
adbc
15 Da Db Rcd R

a Db R
ab + 4 R(
R
ab )2
3RD
3

192
bcde R
ea
5 Rabcd R

224
abef R
cd ef
15 Rabcd R

R
abcd )2
+ 2R(
 1 11

1
1
1
= ( h )3 12 144
30 K13 3 K14 + 10 K15 48 K16
1
1
1
1
23
+ 5!1 108
K2 + 72
K3 + 27
K4 + 15
K5 135
K7 +


1
1
+ 10
K10 + 16 K11 30
K12 ;

4.

1
I4 =
8

 0 0


+

1
12 (1 4 )

2
1

20 (Da Rbcde )

32
135 K8




ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
R

1 1

i3 q i4 ( )
Di3 Di4 Rq

1
48 K9

(35)

248

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259




ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
R

g mn i3 l mk i4 k ln q i3 q i4 ( )




1
ii2 i3 j q i1 q i3 q i q j + bi cj + a i a j ( )
(1 4 ) Di1 R
+ 12


i4 ( )
Di4 Rq


mi3 i4 n + 8 R

k
+ 5!3 (1 4 ) Di1 Di2 R
9 mi1 i2 k R i3 i4 n

1
6




d d
q i1 q i4 q m q n + bm cn + a m a n ( )R
= ( h )3 18

2
1 1
1 1
60 24 RR 6 144 R(Rabcd )

1 1
ab 1
2
60 24 Da R Db R + 48 (Da R)

1
1
1
ab 1 R(
R
ab )2
200 24 RR + 12 RDa Db R
27


1
31
1
1
= ( h )3 18 5400
K2 21600
K3 + 900
K9 + 1440
K16 ;

5.

2
1
18 R(Rabcd )


(36)

   0 0

1 h 2
i1 ( ) Di2 Rq
i2 ( )
I5 =
(1 4 )2 Di1 Rq
2 8
1 1

( )R

+ q i1 q i2 Di1 Di2 R

g ij i1 l ik i2 k lj q i1 q i2 ( ) d d
+ 2(1 4 )R
 1 2  1

 2 1
2
1
a 1

= ( h )3 128
5
12 Da RD R + 6 RR 5 36 R(Rabcd )
1
 7

1
= ( h )3 8!1 63
5 6 K9 + 12 K16 2 K3 ;
3



+
+
+

6.




+
+

(37)

 3  0 0 0




1
1
ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
I6 =
R
1296 h
111




ki3 i4 l q i3 q i4 q k q l + bk cl + a k a l ( )
R


 m n


m
n
m
n
mi5 i6 n q q q q + b c + a a () d d d
R

1
= ( h )3 1296

1 3
64 R

37 ab cd
360 R R Racbd

R
ab )2
18 R(

i5 i6

2
3
40751 ab cde
a Rbcde
16 (Rabcd ) R 10080 R R

20911 abcd ef
cdef
R ab R
10080 R

137 ab c
2016 R Rb Rca

409 abcd e f
R a c Rbedf
280 R

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259


1
= ( h )3 1296



2

7.

1
1
64 K1 8 K2

409
280 K8 ;

3
16 K3

137
2016 K4

37
360 K5

40751
20160 K6

20911
10080 K7

(38)

1 1
(1 4 )
I7 =
576 h

249

 0 0 0
111




ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
R





ki3 i4 l q i3 q i4 q k q l + bk cl + a k a l ( )
R

d d d
R


2
2
1 1 1 2
1
1
= ( h )3 576
5 R 16 R 6 (Rab ) 4 (Rabcd )
1

1
1
= ( h )3 14
8! 16 K1 6 K2 4 K3 ;

(39)

8. [ ]2
 0 0 0
 




3 1 21
2
ii1 i2 j q i1 q i2 q i q j + bi cj + a i a j ( )
(1 4 )
R
I8 = h
3! 8 6
1 1 1

R
d d d
R
= ( h )3 8!1 21
10 K1 ;

(40)

9. [ ]3
 0 0 0



1 h 3
3
R
R
d d d
I9 =
(1 4 )
R
3!
8
1 1 1

21
= 18 200

K1 .

(41)

Some comments are in order.


There are Feynman diagrams that satisfy the factorization property according to which
a diagram breaks down into simpler subdiagrams. The four-dimensional building
block diagrams are depicted with their corresponding contributions in what follows:
2 ,
= 14 R
= 16 (Rab )2 ,
= 14 (Rabcd )2 ,
=

2
1
16 R(Rabcd )

1 2
24 R(Rab )

1

+ 12
R.

250

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

n/2 terms. In n = 2 dimensions the trace


Let us examine now the contribution of R
anomaly is
2 =

1
R.
24

(42)

2 terms, steming from the


The situation changes in four dimensions in which the R
interactions

 0
 i j

1
i
i
i
j
i
j

IR 2 =
q 1 q 2 q q + b c + a a d
Rii i j R
144 1 2
1

1
ki1 i2 l R
mi3 i4 n
R
+
72( h )2

q q

i3 i4

 0 0




q i1 q i2 q k q l + bk cl + a k a l ( )

11




( h )2 2
R ,
q q + b c + a a ( ) d d +
1152
m n

m n

m n

(43)

and represented by the following diagrams:


 1 2 2
,
R
= ( h )2 24




1 2 2
1 2
R 432
= ( h )2 12 24
(Rab ) ,
 1 2 2
,
= ( h )2 12 24
R
produce a vanishing result.
terms created by the interaction
The same observation holds for the R
 0



1 1
1
i1 i2 i j
i j
i j
IR =
q q q q + b c + a a d h
Rii1 i2 j
R
h 6
24

(44)

and associated with the Feynman diagrams


1 1 .
+ = h R
24
24
One making use of the identities (9)(24) of Appendix A and the 17 linearly independent
terms listed below:
K1 = R 3 ,
K4 = Rab Rcb R ac ,
K7 = Rabcd R abef R cd ef ,
K10 = Rab R ab ,
K13 = (Da Rbc )2 ,
K16 = (Da R)2 ,

K2 = R(Rab )2 ,
K5 = R ab R cd Rcabd ,
K8 = R abcd Rb ef c Reaf d ,
K11 = Rabcd R abcd ,
K14 = D a R bc Db Rac ,
K17 = 2 R,

K3 = R(Rabcd )2 ,
K6 = Rab R acde R b cde ,
K9 = RR,
K12 = R ab Da Db R,
2

K15 = D a R bcde ,
(45)

can deduce for the integrated trace anomaly:





n 

 i  i  i
1
i
AnW (s = 0, n) = lim
g x0 f x0 I x0 ,
dx0
0 (2 h
)3
i=1

(46)

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

251

where
I (x0i )
=
( h )3

1 7
8! 225 K1

98
45 K2

14
15 K3

661
108
9 K8 5 K10

1
6
5 K16 5 K17 .

2701
324 K4

191
12 K11

6047
405 K5

16
3 K12

18413
648 K7

98
3 K14

2K15

159421
3240 K6

127
3 K13

(47)

5. The computer algebra program


Vertices I1 I9 were calculated using the Riegeom package [18], which is a Maple
package for manipulating generic symbolic tensor expressions in the context of riemannian
geometry. In this section we show the process of calculating the vertex I7 . The lines
beginning with > in typewrite font are the input in a Maple worksheet. A colon at the
end of the command hides the result. We start loading the package
> with(Riegeom);
defining Christoffel(Gamma), Riemann(R), Weyl(C), Ricci(R), LeviCivita(eta),
TraceFreeRicci(S) for Dimension = 4, CoordinateName = X, MetricName = g

[absorbg, changedumind, cleartensor, codiff , coordinate, definetensor, dimension,


expandcodiff , lptensor, ltensor, metric, normalform, off , on, printtensor,
replace, simpLC, simptensor, sreplace, switches, symmetrize, symmetry, tdiff ]
Next command reads file Vertex.mpl that contains procedures written specifically for the
calculation of vertices I1 I9 using Maple programming language and Riegeom commands.
> read(Vertex.mpl):
We setup spacetime dimension using Riegeom command dimension.
> dimension(6):
Next command enters the tensor coefficient of vertex I7 .
> tensor_coeff := printtensor(R[-mu1,-i1,-i2,-nu1]
>
*R[-mu2,-i3,-i4,-nu2]);
tensor_coeff := R1 i1 i2 1 R2 i3 i4 2
Printtensor is the Riegeom interface command. Indices with minus sign are covariant
and with plus sign are contravariant. In the next command we enter the field terms. We
use a Maple list (which preserves order) instead of an expression written in terms of the
commuting product operator. In this form we have full control over the order of the fields.
> L := printtensor([[q[i1],q[i2]](sigma),
> [[qdot[mu1],qdot [nu1]],
> [b[mu1],c[nu1]],[a[mu1],a[nu1]]](sigma),
> [q[i3],q[i4]](tau),
> [[qdot[mu2],qdot[nu2]],[b[mu2],c[nu2]],
> [a[mu2],a[nu2]]](tau)]);


L := q i1 ( ), q i2 ( ) ,


qdot1 ( ), qdot1 ( ) , b1 ( ), c1 ( ) , a 1 ( ), a 1( ) ,

252

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

q i3 ( ), q i4 ( ) ,

qdot2 ( ), qdot2 ( ) , b2 ( ), c2( ) , a 2 ( ), a 2( )

Next command finds all independent 6 index configurations that contribute to the final
result.
> Lic := ind_config([i1,i2,mu1,nu1,i3,i4,mu2,nu2]):
> N := nops(Lic);
N := 105
In the case of 8 field indices, there are 105 independent index configurations. We show the
first 3 ones.
> for i to 3 do evaln(Lic[i])=Lic[i] od;
Lic1 = [i1, i2, 1, 1, i3, i4, 2, 2]
Lic2 = [i1, i2, 1, 1, i3, 2, i4, 2]
Lic3 = [i1, i2, 1, 1, i3, 2, 2, i4]
Next command is a loop over the 105 index configurations. The results are stored in a table
of results called res.
> for i to N do
> expr := WickContractions(L, Lic[i]):
> res[i] := Vertex(tensor_coeff*expr):
> od:
Command WickContractions performs the Wick contractions given by Eqs. (27)
(32). We show an example with the fourth index configuration.
> expr := factor(WickContractions(L, Lic[4]));

2
expr := 4 hbar4 g i1 i2 g 2 2 g 1 i3 g 1 i4 (1 + ) + Heaviside( )
Command Vertex simply multiplies the result of Wick contractions to the tensor
coefficient, simplifies the tensor expression, and isolates the terms to be integrated, since
we noticed that, for the most complicate vertices, Maple spends more time integrating
Dirac delta and step functions than simplifying the tensorial terms.
> Vertex(tensor_coeff*expr);

2

4
hbar4 R 2 1 R2 1 , (1 + ) + Heaviside( )
Next command adds the results of all index configurations after integrating over variables
and . Functions int_tau and int_sigma replace the usual Maple integrator, since
Maple int command fails to return the correct value for complicated vertices.
> final_value := simptensor(add(res[i][1]*
6 The number of independent index configurations is given by n!/((n/2)!2n/2 ) where (n/2)! represents the
possible permutations of index pairs and 2n/2 the permutation of indices within the pairs.

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

253

> int_tau(int_sigma(res[i][2])),i=1..N));
final_value :=

4
4 1 2
1 2 4
1 4
R1 2
16 R hbar 6 hbar R
1 4
4 2 i1 _b0 _c0
4 hbar R
R2 i1 _b0 _c0

Collecting h 4 4 , we obtain the final form.


> collect(final_value, [beta,hbar]);
 1 2 1 1 2
R R
R
1 R 2 i1 _b0 _c0 R
16

1 2

2 i1 _b0 _c0

hbar4 4

For vertices I2 and I6 , the method described above has some extra complications. Vertex
I6 has a large number of index configurations (N := 10395) and vertex I2 uses huge side
identities obtained in Riemann normal coordinates as described in Appendix A.3.
This computer algorithm can easily be extended to higher dimensions to predict the
trace anomalies in 8 and 10 dimensions. Of course there are strong limitations implied by
the exponential growth of the problem which can be relaxed by increasing the available
computing power. The tensorial upper bound of the program is reached by the product of
eight Riemann curvature tensors but such a case is beyond the scope of the present work.
6. Conclusions
In this article we calculate the integrated trace anomaly for a real spin-0 scalar field
living on a curved six-dimensional manifold. This is achieved by relying on a recursive
computation of the metric tensor components in Riemann normal coordinates. One can use
the general formulae (9) and (11) to reach the desired order of metric expansion induced
by the dimensionality of the manifold. Adopting the path integral formalism of quantun
mechanical nonlinear sigma models, we evaluate all the vertices and the corresponding
Feynman diagrams that contribute to the present anomaly. A computer based program
is used to perform the otherwise inevitable task of integration over distributions and
contractions of the various tensors involved. The final result derived in this process involves
17 scalar monomials consisting of covariant derivatives and/or Riemann tensors.
Acknowledgements
We are grateful to F. Bastianelli for his critical comments, S. Frolov for bringing
to our attention their work [19] and K. Lake for advising A.H. to contact R.P. for the
computational manipulation of the problem.
Appendix A
A.1. Useful identities
We consider a riemannian (or pseudo-riemannian) manifol equipped with its standard
(metric compatible, torsion free) connection. The Riemann curvature is defined as:
R a bcd = c a bd + e bd a ec c d,

(A.1)

254

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

possessing the familiar symmetries:


antisymmety:
Rabcd = Rbacd = Rabdc ;

(A.2)

pair symmetry:
Rabcd = Rbadc ;

(A.3)

cyclic symmetry:
Rabcd + Radbc + Racdb = 0;

(A.4)

Bianchi symmetry and the related identities:


Rabcd;e + Rabec;d + Rabde;c = 0,

(A.5)

Rbc;a Rab;c + R d bca;d = 0,


 a 1 a 
Rb 2 b R ;a = 0.

(A.6)
(A.7)

The Ricci curvature and scalar are:


Rab = R q aqb ,

R = Rq q .

(A.8)

In this paper the following identities have been exploited to simplify our expressions:
2
Rabcd R acbd = 12 Rabcd
,

(A.9)

= 12 Rabcd R ebcd ,
Rabcd R
(A.10)
e c b f 7 abef

e(af )b (d c
cd
1
R e) = 8 Rabcd R af R ed + 2 R
Ref ,
Ra(bc)d R
(A.11)
ab
1 2
Da Db R = 2 R,
(A.12)
Da Db R ab = 12 2 R 12 (Da R)2 2Rab D a D b R + 2Rab Rcb R ac 2R ab R cd Racbd
4Da Rbc D b R ac + 3(Da Rbc )2 + 12 Rabcd R abcd
2Rabcd R e af c R be df + 2R ab R cde a Rbcde
ecbd

12 R abcd Rabef Rcd ef + Rab R ab ,


Da Db D a Dc R bc = 12 2 R 14 (Da R)2 12 Rab D a D b R,
b

ac

bc

Da Db Dc D R

Da Db Dc D R

12 (Da R)2 2Rab D a D b R + R abcd Dd Db Rac


+ 2Rab Rcb R ac 2R ab R cd Racbd 3Da Rbc D b R ac
+ Rab R ab + 2(Da Rbc )2 ,

(A.13)
(A.14)

1 2
2 R

2
a b
b ac
1 2
1
2 R 2 (Da R) 2Rab D D R + 2Rab Rc R
2R ab R cd Racbd 3Da Rbc D b R ac + Rab R ab + 2(Da Rbc )2

(A.15)

+ 14 Rabcd R abcd R abcd R e af c R be df + R ab R cde a Rbcde


14 R abcd Rabef Rcd ef ,
D a R bc Dd Rab d c = (Da Rbc )2 Da Rbc D b R ac ,

(A.16)

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

255

R abcd Dd Db Rac = 14 R abcd Rabcd + 12 Rab R acde R b cde R abcd R e a f c Rbedf


14 R abcd R ef ab Rcdef ,
ab

R D Dd Rca

b =R

ab

Rab

(A.17)

ab c
1 ab
2 R Da Db R + R Ra Rcb

R R Racbd , (A.18)
ab

cd

R abcd Dd De R e abc = R abcd Dd Db Rac ,


Da R abcd De R e bcd = 2 (Da Rbc )2 Da Rbc D b R ac ,


De Dd R abc =
ebcd
D R
De Rabcd =
a
Da RD R = 12 R 2 RR,
R ab Db Da R = 2R ab D c Db Rac
R

abcd

(A.19)
(A.20)

1 abcd
Rabcd ,
4R
1
2
2 (Da Rbcde ) ,

(A.21)
(A.22)
(A.23)

2R R Racbd + 2R
ab

cd

ab

Rac Rbc .

Proof of (A.9). By cyclic symmetry,




2
.
Rabcd R acbd = Rabcd R adcb + R abdc = Rabcd R adcb + Rabcd

(A.24)

(A.25)

Rename the indices in the first term:


Rabcd R adcb = Rabdc R acdb = Rabcd R acbd .

(A.26)

Solve for the desired object. In the same vein we can prove (A.10) as well as
Da R abcd D e Recbd = 12 (Da R abcd )2 and R abcd Racbd = 12 R abcd Rabcd .
Proof of (A.11). Expanding out the products, the resulting sum can be expressed in terms
of the invariants:
L1 = Rabcd R abef R c e c f ,

L2 = Rabcd R abef Ref cd ,

L3 = Rabcd R e f a c R f de b ,

L4 = Rabcd R e af c R b ed f ,

(A.27)

and with the assistance of the identities:


L1 = 12 L2 ,

L3 = 14 L2 ,

L4 = L3 + Rabcd R e af c R b f d e ,

(A.28)

one can recover the proposed formula.


Proof of (A.12)(A.24). With the assistance of (A.7)(A.9) it is a straightforward exercise
to show their validity.
A.2. Six-dimensional invariants
In d dimensions the polynomials C abcd Ccdef C ef ab and Cabcd C ebcf C a ef d (type-B
anomaly according to the geometric classification of [9]) are written as:
Inv1 = C abcd Ccdef C ef ab
12
6
2
Rabcd R abce Red +
RRabcd
d 2
(d 1)(d 2)
8(2d 3)
24(2d 3)
2
+
R3
RRab
2
3
(d 1) (d 2)
(d 1)(d 2)3

= R abcd Rcdef R ef ab

256

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

16(d 1) ab
24
R Rbc Rac +
R abcd Rac Rbd ,
3
(d 2)
(d 2)2

(A.29)

Inv2 = Cabcd C ebcf C a ef d


=

(d 2 + d 4)
3(d 2 + d 4)
3
3
2
2
RRabcd
R

RRab
+
(d 1)2 (d 2)3
(d 1)(d 2)3
2(d 1)(d 2)
2(3d 4)
3d
3
Rabcd R abce Red
+
Rab R bc Rca +
Rabcd R ac R bd
d 2
(d 2)3
(d 2)2
+ Rabcd R ebcf R a ef d .

(A.30)

Expressions (A.29) and (A.30) reproduce in d = 6 the unique already known Weyl
invariant polynomials of [5] constructed only out of Weyl tensors. In higher dimensions
the independent ways to contract indices among a number of Weyl tensors increases.
Another invariant made out of covariant derivatives of the Weyl tensor is
Inv3 = R 3 + 8RR ab Rab + 2RR abcd Rabcd 10R ab Rbc Rca 10Rab R acdb Rcd
+ 12 RR 5R ab Rab + 5R abcd Rabcd .

(A.31)

The final nontrivial 7 invariant (type-A) we would like to consider stems from the Euler
form which exists in any even dimension d = 2n:

1
:a1 a2 a2n R a1 a2 R a2n1 a2n
E2n =
(4)n n!
M

1
1
=
:a1 a2 a2n : b1 b2 b2n R a1 a2 b1 b2 R a2n1 a2n b2n1 b2n dV
(4)n n! 2n
M

1
=
(A.32)
E2n dV ,
(4)n n!
M

where
1
:a a a : b1 b2 b2n R a1 a2 b1 b2 R a2n1 a2n b2n1 b2n
2n 1 2 2n
 
1
= n det abii R a1 a2 b1 b2 R a2n1 a2n b2n1 b2n , i = 1, . . . , 2n,
(A.33)
2
is the Euler number which is a total divergence, vanishes in all lower integer dimensions

and dV = e1 e2n = g d2n x is the volume form. In six dimensions the Euler
number becomes:
E2n =

Inv4 = E6 = R 3 12RR ab Rab + 16R ab Rbc Rac + 24R ab R cd Racbd + 3RR abcd Rabcd
+ 4R abcd Rab ef Rcdef 24R ab Ra cde Rbcde 8R abcd Ra e c f Rbedf .
(A.34)
7 The terminology trivial invariants is adopted to justify the existence of covariant total derivatives of

polynomials over Riemann tensor and its covariant derivatives. These invariants coincide with variations of all
independent local counterterms to effective action.

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

257

A.3. The vertex I2


The key idea for calculating the contribution of this vertex is to express each interaction
term in terms of polynomials of the Riemann curvature and its covariant derivatives. This
can be achieved by making use of the identities (23) and the symmetries implied by the
structure of each term separately.
d
d one may rewrite the second
= 23 R
With the help of the identity a bcd = a (bc)
(bc)a
term of I2 as: 8

i i1 i2 j i3 ikl i4 ljk = 4 R
i i1 i2 j R
l
k
13 R
(ik)i3 R(lj )i4 .
27

(A.35)

e
e
= 12 D(a R
Expanding out the identity (a b c)d
b|d|c) we get:


l + Di2 R
l + Dk R
l
i1 i2 kil + i1 k i2l i + i2 k i1l i = 14 Di1 R
i2 ik
i1 ik
i1 ii2


l + Di1 R
l + Di2 R
l .
+ Dk R
i2 ii1
kii2
kii1
(A.36)

Notice that the third term of I2 is symmetric under the interchange of the lower indices
of the Christoffel symbols namely k and l . So in (A.36) we interchange
these indices and add the two expressions together. We then obtain:


2 i1 i2 kil + i1 (k i)il 2 + i2 (k i)il 1

l + Di2 R
l + Di1 R
l + Di2 R
l + Dk R
l + Dk R
l
= 14 Di1 R
kii2
kii1
iki2
iki1
i2 ii1
i1 ii2

(A.37)
l + Di R
l
.
+ Di R
i2 ki1

i2 ki1

We would like now to express the second and third terms of the left-hand side
l
of (A.37) in terms of i1 i2 k
and covariant derivatives of the Riemann curvature.
Starting from
l = i1 i2 l i1 i l
Di1 R
ii2 k
ik
ki2

l
l
l
i1 (i k)i
= i1 i2 (ik)
Di1 R
(i|i2 |k) ,
2
(A.38)

one has
l
l
Di2 R(i|i
.
i1 (k i)il 2 + i2 (k i)il 1 = 2i1 i2 kil Di1 R(i|i
2 |k)
1 |k)

(A.39)

Substituting (A.39) into (A.37) one gets:



l
l
l
l
kii
kii
iki
iki
il ii
3i1 i2 kil = 18 Di1 R
+ Di2 R
+ Di1 R
+ Di2 R
+ Dk R
2
1
2
1
2 1

l
l
l



+ Dk Ri1 ii2 + Di Ri2 ki1 + Di Ri2 ki1


1
l
l
+ 2 Di1 Riil 2 k + Di1 Rki
(A.40)
.
+ Di2 Riil 1 k + Di2 Rki
2i
1i
1 d
d
8 Identical results arise if one makes use of the relation
(a b)c = 3 R
(bca) which communicates with the
d
one employed in the text by the relation (a bc) = 0.

258

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

Finally the expression we are going to use for the contribution of the third term in the
action I2 is:

1
l + Di2 R
l + Di1 R
l + Di2 R
l + Dk R
l
i1 i2 kil = 24
Di1 R
kii2
kii1
iki2
iki1
i2 ii1

l
l
l



+ Dk Ri1 ii2 + Di Ri2 ki1 + Di Ri2 ki1


l
l
1
+ Di2 Riil 1 k + Di2 Rki
+ 6 Di1 Riil 2 k + Di1 Rki
(A.41)
.
2i
1i
and a similar relation holds for the term i3 i4 lik .
References
[1] M.J. Duff, Twenty years of the Weyl anomaly, Class. Quantum Grav. 11 (1994) 1387.
[2] S. Deser, M.J. Duff, C.J. Isham, Non-local conformal anomalies, Nucl. Phys. B 111 (1976) 45;
M.J. Duff, Comment on quantum gravity and world topology, QMC-7629, 1976;
M.J. Duff, Observations on conformal anomalies, Nucl. Phys. B 125 (1977) 334.
[3] B.S. DeWitt, Dynamical Theory of Groups and Fields, Gordon and Breach, New York, 1965,
first published, in: Relativity, Groups and Topology, Les Houghes Summer School, 1964;
B.S. DeWitt, Phys. Rep. C 19 (1975) 297;
L.S. Brown, Stress tensor trace anomaly in a gravitational metric: scalar field, Phys. Rev. D 15
(1977) 1469;
L.S. Brown, C.J. Cassidy, Stress tensor trace anomaly in a gravitational metric: general theory,
Maxwell field, Phys. Rev. D 15 (1977) 2810;
A. Duncan, Conformal anomalies in curved spacetime, Phys. Lett. B 66 (1977) 170;
J.S. Dowker, R. Critchley, The stress-tensor conformal anomaly for scalar and spinor fields,
Phys. Rev. D 16 (1977) 3390;
A.E.M. van de Ven, Index-free heat kernel coefficients, hep-th/9708152.
[4] A. Hatzinikitas, K. Schalm, P. van Nieuwenhuizen, Trace and chiral anomalies in string and
ordinary field theory from Feynman diagrams for non-linear sigma models, Nucl. Phys. B 518
(1998) 424.
[5] L. Bonora, P. Cotta-Ramusino, C. Reina, Conformal anomaly and cohomology, Phys. Lett.
B 126 (1983) 305;
L. Bonora, P. Pasti, M. Bregola, Weyl cocycles, Class. Quantum Grav. 3 (1986) 635.
[6] S.A. Fulling, R.C. King, B.G. Wybourne, C.J. Cummins, Normal forms for tensor polynomials:
the Riemann tensor, Class. Quantum Grav. 9 (1992) 1151.
[7] P.B. Gilkey, The spectral geometry of a Riemann manifold, J. Differential Geom. 10 (1975)
601.
[8] I.G. Avramidi, A covariant technique for the calculation of the one-loop effective action, Nucl.
Phys. B 355 (1991) 712.
[9] S. Deser, A. Schwimmer, Geometric classification of conformal anomalies in arbitrary
dimensions, Phys. Lett. B 309 (1993) 279.
[10] L. Alvarez-Gaum, E. Witten, Gravitational anomalies, Nucl. Phys. B 234 (1983) 269.
[11] F. Bastianelli, P. van Nieuwenhuizen, Trace anomalies from quantum mechanics, Nucl. Phys.
B 389 (1993) 53.
[12] J. de Boer, B. Peeters, K. Skenderis, P. van Nieuwenhuizen, Loop calculations in quantum
mechanical non-linear sigma models with fermions and applications to anomalies, Nucl. Phys.
B 446 (1995) 211, Nucl. Phys. B 459 (1996) 631.
[13] L. Eisenhart, Riemannian Geometry, Princeton Univ. Press, Princeton, NJ, 1965;
V.I. Petrov, Einstein Spaces, Pergamon, Oxford, 1969;

A. Hatzinikitas, R. Portugal / Nuclear Physics B 613 (2001) 237259

[14]
[15]
[16]

[17]
[18]
[19]

259

A.D. Dolgrov, L.B. Khriplovich, Normal coordinates along a geodesic, J. Gen. Rel. Grav. 15
(1983) 1033.
L. Alvarez-Gaum, D.Z. Freedman, S. Mukhi, The background field method and the ultraviolet
structure of the supersymmetric nonlinear -model, Ann. Phys. 134 (1981) 85.
U. Mller, C. Schubert, A.M.E. van de Ven, A closed formula for the Riemann normal
coordinate expansion, J. Gen. Rel. Grav. 31 (1999) 1759.
K. Fujikawa, Path-integral measure for gauge-invariant fermion theories, Phys. Rev. Lett. 42
(1979) 1195;
K. Fujikawa, Comment on chiral and conformal anomalies, Phys. Rev. Lett. 44 (1980) 1733;
K. Fujikawa, Energy momentum tensor in quantum field theory, Phys. Rev. D 23 (1981) 2262;
K. Fujikawa, Phys. Rev. Lett. 42 (1979) 1195.
A. Diaz, W. Troost, A. Van Proeyen, P. van Nieuwenhuizen, Int. J. Mod. Phys. A 4 (1989) 3959.
R. Portugal, The Riegeom package: abstract tensor calculations, Comput. Phys. Commun. 126
(2000) 261268.
F. Bastianelli, S. Frolov, A.A. Tseytlin, Conformal anomaly of (2, 0) tensor multiplet in six
dimensions and AdS/CFT correspondence, JHEP 013 (2000) 0002;
F. Bastianelli, O. Corradini, 6D trace anomalies from quantum mechanical path integrals, Phys.
Rev. D 63 (2001) 065005.

Nuclear Physics B 613 (2001) 260284


www.elsevier.com/locate/npe

Power corrections in flavour-singlet deep inelastic


scattering
G.E. Smye 1
Dipartimento di Fisica, Universit di Milano-Bicocca, and INFN Sezione di Milano, Italy
Received 30 May 2001; accepted 9 August 2001

Abstract
We investigate the 1/Q2 power-suppressed corrections to structure functions in the flavour-singlet
channel of deep inelastic lepton scattering arising from renormalon insertions into an initial-state
gluon, as obtained using the dispersive approach. The pinch-technique is used as a convenient tool in
the separation of contributions. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.38.-t; 12.38.Cy; 13.60.Hb

1. Introduction
We have learned much about the strong interaction of particle physics from deep
inelastic scattering (DIS) experiments over many years. This process is continuing, with
present and future experiments generating more data over larger regions of phase space.
Alongside this ongoing experimental work, theoretical developments continue to be
made. QCD, the established theory of strong interactions, continues to pose challenges:
perturbation theory has been relatively successful at high energies, but the series expansion
even here is at best asymptotic. In the non-perturbative rgime lattice techniques are
making advances. Yet a clean distinction between perturbative and non-perturbative
cannot be made: all QCD observables involve some interplay between them.
Although the structure functions (for example) cannot be calculated using perturbative
QCD, the general shape of their asymptotic Q2 behaviour is well known: the observed
Bjorken scaling at high Q2 is violated by additional smaller terms. The dominant scaling
violation is a logarithmic Q2 dependence, originating in the scale dependence of the parton
E-mail address: graham.smye@mib.infn.it (G.E. Smye).
1 Research supported by the EU Fourth Framework Programme, Training and Mobility of Researchers,

Network Quantum Chromodynamics and the Deep Structure of Elementary Particles, contract FMRX-CT980194 (DG12 - MIHT).
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 1 ) 0 0 3 9 3 - 5

G.E. Smye / Nuclear Physics B 613 (2001) 260284

261

density functions of the incident hadron. This can be calculated using perturbative QCD
and used to measure the strong coupling s .
In addition to the logarithmic scaling violations, there are known to be contributions
behaving as inverse powers of the hard scale, i.e., as 1/Qn . These include both corrections
due to the non-zero hadron mass M, which are suppressed by a factor M 2 /Q2 ,
and non-perturbative power-suppressed terms arising from higher-twist operators in the
operator product expansion. Such contributions are not included in fixed-order perturbative
calculations, yet they are known to be important over the wide Q2 range of available data.
Over a number of years the higher-twist power-suppressed terms of a wide variety of
observables have been estimated using two related approaches, the renormalon (see [1]
for a review) and dispersive [2] models, although it is the dispersive approach that is
used here. If we consider graphs with an arbitrary number of loop insertions in a gluon
propagator, we assume that we can reconstruct a well-defined effective strong coupling
at the scale of the gluon virtuality. The difference between this true coupling and
that reconstructed using fixed order perturbation theory gives rise to non-perturbative
corrections, which are typically power-behaved. For a given observable, the shape of the
leading correction can be found; an estimation of its magnitude requires the additional
assumption of universality. Thus, starting from perturbative QCD, we aim to investigate
the transition to the non-perturbative region.
These approaches have been applied to various QCD observables, with the assumption
of universality approximately holding [3]. In flavour non-singlet DIS there are results for
structure functions [4,5], fragmentation functions [6] and event shape variables [7], while
studies have been made of power corrections to structure functions [8,9] and fragmentation
functions [10] in the flavour singlet channel.
The flavour singlet contribution to DIS is that involving the interaction of the gluons
within the hadron. (We do not study the fermionic singlet, arising from the total sum of
quark distributions, since these behave just like the non-singlet contribution [4,5].) Power
corrections are calculated using the renormalon or dispersive models by making insertions
into the gluon propagators as shown in Fig. 1. Here the lower part of the diagram represents
the finding of a virtual gluon within the proton, and the upper part represents photongluon
fusion.
In the normal perturbative treatment of DIS, the asymptotic freedom of QCD enables
us to treat the initial-state partons as free particles confined within the nucleon; so in
a singlet calculation we would start with a free gluon and convolute the perturbative result
with the gluon distribution function g(x). We cannot however do this in a calculation
of power corrections, since the models we use consider modifications to the gluon
propagator (loop insertions in the renormalon model, or, equivalently, a mass in the
dispersive approach). We therefore consider the initial-state gluon to be generated by some
perturbative mechanism, the simplest of which is by radiation from an on-shell parton.
We immediately encounter a very serious problem: if the gluon is not on shell, both
the upper and lower parts of this diagram are gauge-dependent. Gauge independence is
achieved only when we include all the elements of some subset of diagrams of a given order
in s , and this includes diagrams that cannot be separated into the two halves of Fig. 1:

262

G.E. Smye / Nuclear Physics B 613 (2001) 260284

Fig. 1. Power corrections in flavour singlet DIS.

Fig. 2. Radiation of gluon from quark line.

diagrams which do not have analogous gluon propagators in which to make insertions or
from which to take a scale for the running coupling.
This is precisely the same problem encountered in the two-loop calculations of power
corrections to event shapes in e+ e annihilation, as discussed in [11], except that the
gluon is now in a different channel. We might thus expect a similar solution: namely,
the use of the pinch technique [12] to generate the running coupling at the scale of the
gluon virtuality [13] in squared diagrams with two exchanged gluons, and the remainder
of diagrams contributing to the power correction via some as-yet unknown mechanism.
The first studies [810] of these singlet quantities considered the incoming virtual gluon
to be radiated from a quark line, as in Fig. 2. This is the simplest case, since the lower half
of the diagram is gauge independent, as is the upper half to leading order, O(s ).
There is however another problem to be overcome, concerning the interpretation of the
lower half of the diagram. We may try to recover the singlet contribution to the power
corrections by deconvoluting the full result with the quark-to-gluon splitting function (this
is the approach taken in [8]), or we may leave the result as it is and interpret it as a genuine
second-order contribution. These two interpretations give very different predictions for the
magnitudes of the power-suppressed corrections. I argue below (in Section 6) that the latter
approach should be adopted.
In Section 2 we discuss the flavour singlet contribution to deep inelastic scattering. The
dispersive approach to power corrections is then briefly reviewed in Section 3. Section 4
examines in detail the virtual gluon production represented by the bottom half of Fig. 1,
and the application of the pinch technique to this, while in Section 5 we calculate the
contribution to the structure functions from photongluon fusion. Section 6 investigates
the distribution of virtual gluons in the proton, paying particular attention to the 1/Q2
power corrections. Finally a summary is given in Section 7.

G.E. Smye / Nuclear Physics B 613 (2001) 260284

263

2. Flavour singlet DIS


Consider the deep inelastic scattering of a lepton with 4-momentum l from a hadron with
4-momentum P , as shown in Fig. 3. If the momentum transfer is q, the usual kinematic
variables are Q2 = q 2 , the Bjorken variable x = Q2 /2P q, and y = P q/P l  Q2 /xs,
where s is the square of the energy in the c.m. frame.
Then the differential cross section is


2 2 
d 2
1 + (1 y)2 FT (x) + 2(1 y)FL (x) ,
=
(2.1)
dx dQ2
Q4
where FT (x) = 2F1 (x) and FL (x) = F2 (x)/x 2F1 (x) are the transverse and longitudinal
structure functions, which also have a weak Q2 dependence which we do not show
explicitly. (For simplicity we are neglecting any contribution from weak interactions, i.e.,
Z0 or W exchange.)
We can consider the photon to interact with an asymptotically free parton, moving
collinearly with the hadron with momentum s = xP / (x   1). Then in the parton
model, to order s0 , we have
 

FT (x) =
(2.2)
eq2 q(x) + q(x)

,
FL (x) = 0,
q

where q(x) and q(x)

are the quark and antiquark distributions in the target hadron. Thus
at this level there is no contribution arising from the gluon distribution in the hadron.
The O(s ) contributions are most easily given as the distribution in the final-state
variable = P r/P q, (0   1):



d
s  2 d 
Fi (x) =
eq

)
CF Ci,q (, ) q(x/ ) + q(x/
d
2 q

x

+ TR Ci,g (, )g(x/ ) ,
1

Fig. 3. Flavour singlet contribution to deep inelastic scattering.

(2.3)

264

G.E. Smye / Nuclear Physics B 613 (2001) 260284

where g(x) is the gluon distribution in the target hadron, CF = 4/3, TR = 1/2 and the
coefficient functions Ci,j (, ) [14] are
2 + 2
+ 2 + 2,
(1 )(1 )
CL,q (, ) = 4 ,

 2 + (1 )2
,
CT ,g (, ) = 2 + (1 )2
(1 )
CL,g (, ) = 8(1 ).
CT ,q (, ) =

(2.4)
(2.5)
(2.6)
(2.7)

An integration over the entire range of from 0 to 1 requires the implementation of


a factorisation scheme to regulate the collinear divergences. In our calculations this will be
effected by the introduction of a small gluon mass. In addition the coefficient functions
Ci,q (, ) acquire contributions at = 1 from virtual gluon emission.
We are interested in the singlet contributions Ci,g (, ), corresponding to photongluon
fusion; the 1/Q2 power corrections from the non-singlet components Ci,q (, ) were
successfully analysed in [4,5].

3. The dispersive approach to power corrections


We assume that the QCD running coupling s (k 2 ) can be defined for all positive k 2 , and
that apart from a branch cut along the negative real axis there are no singularities in the
complex plane. It follows that we may write the formal dispersion relation:
 
s k 2 =

 
d2
s 2 ,
2
+k

2
0

(3.1)

where the spectral function s represents the discontinuity across the cut:
 




1   2 i 
1
s e s 2 ei =
Disc s 2 .
s 2 =
2i
2i

(3.2)

0
To lowest order in perturbation theory we have s (2 ) = 4
|s (2 )|2 .
Non-perturbative effects at long distances are expected to give rise to a non-perturbative
modification to the perturbatively-calculated strong coupling at low scales, s (2 ) =
s (2 ) sPT (2 ), where sPT (2 ) is the perturbatively-calculated running coupling [2].
Note that here sPT (2 ) refers only to the contribution to the running coupling from a fixed
(next-to-leading) order perturbative calculation, and so is itself well-behaved down to low
scales, without any divergences: the Landau pole appears when we include an arbitrary
number of loop insertions in the propagator. Hence s (2 ) is assumed to be well-defined
for all positive 2 .
We now consider the calculation of some observable F in an improved approximation
which takes into account fixed-order contributions plus those higher-order terms that lead
to the running of s . As is well documented, for processes involving a single gluon, it
is required to calculate the relevant contributions as though the gluon had a small mass

G.E. Smye / Nuclear Physics B 613 (2001) 260284

265

2 = $Q2 . Calculations of power corrections in e+ e annihilation, in non-singlet DIS,


and in the DrellYan process, use a single such gluon. In these cases the 1/Qn corrections
are found to be proportional to
CF
An
2

d2 n  2 
s .
2

(3.3)

Numerical values for these parameters must be obtained from data: fits of the 1/Q2
corrections to DIS structure functions suggest that A2  0.2 GeV2 [4].
However, DIS in the singlet channel involves two such gluons. Both gluons have an
associated dispersive variable, so we obtain a characteristic function F ($1 , $2 ), where
$i = 2i /Q2 , which is simply the observable F calculated as though the gluons had masses
1 and 2 , and without any factors of s .
Since both gluons are constrained to have the same 4-momentum s, we can simplify
this to require only one dispersive variable. By defining = s 2 /Q2 , we see that the
dependence of the characteristic function F on $1 and $2 is given by

d f ()
,
F ($1 , $2 ) =
(3.4)
( + $1 )( + $2 )
where the integration limits and the function f depend on the particular calculation. This
may be re-expressed in the form
F ($1 , $2 ) =
where

($1 ) $2 F
($2 )
$1 F
,
$1 $2

d
f ().
( + $)
So it is sufficient to perform the calculation with one mass set equal to zero, the
giving us the form of the characteristic function.
The alternative expression


1
1
d f ()

F ($) =
d f ()
$
$
+$
found in [9,10]. In
shows how this relates to the slightly different definition of F
($) =
F

(3.5)

(3.6)
other

(3.7)

these
references it is shown that a characteristic function of the form (3.5) leads to a nonperturbative correction to the observable F given by

F =

d2
 2   2    2 2  2 2 
2s s s
G /Q ,
2

(3.8)

where, using Cauchys theorem,




  i 
= 1 Disc F
($) = 1 F
$ei = f ($).
$e F
G($)
(3.9)
2i
2i
(Another method of arriving at (3.8) in this particular case is simply to insert the running
coupling s (s 2 ) into the matrix element before integrating.)

266

G.E. Smye / Nuclear Physics B 613 (2001) 260284

Since s (2 ) is small in the perturbative rgime, and vanishes as 2 , the


at small $. So we perform an expansion of
correction (3.8) depends on the behaviour of F
this function about $ = 0.
($) is subtracted off. These are the terms causing the running
Any divergent term in F
of parton distributions, which gives rise to logarithmic scaling violations: the gluon mass
here behaves as a regulator.
The remaining terms, which vanish as $ 0, give power corrections. We will find the
dominant terms to be:
a1 $ log $
F

F = a1

D1
,
Q2

(3.10)

and
1 a2 $ log2 $
F
2

F = a2

D1
D2
log 2 ,
2
Q
Q

(3.11)

where the non-perturbative parameters D1 and D2 are defined by:



D1

d2 2
 2   2    2 2
,
2s s s
2

(3.12)

1
log D2
D1

      
d2 2
2
2
2s 2 s 2 s 2
.

log

(3.13)

While we expect the form of s (2 ), and hence D1 and D2 , to be universal, we have as


yet no numerical values for them. It will be necessary therefore to extract values for D1
and D2 , either from experimental results or from some model of the form of s (2 ).

4. Production of the gluon


4.1. Gluon kinematics and polarisations
Let us consider the radiation of a virtual gluon with 4-momentum s from a massless
parton with 4-momentum p, as shown for the case of a quark in Fig. 2. The parton model
assumption is p = xP / , where P is the 4-momentum of the incoming hadron. Since
the gluon is virtual, it need not be collinear with the proton, nor need it be transversely
polarised.
We describe the kinematics by three additional variables, and  defined by
Q2

=
,
1 + 2s q

s2
=

= ,
2
Q2

(4.1)

and the azimuthal angle in the Breit frame. Then the relevant 4-momenta are, in the Breit

G.E. Smye / Nuclear Physics B 613 (2001) 260284

267

frame,
1
q = Q(0, 0, 0, 2),
2
1
p = Q(1/, 0, 0, 1/ ),
2
1
s = Q(s0 , s cos , s sin , s3 ),
2
where

(4.2)
(4.3)
(4.4)

1 + (1 2/  )
,

4(1 /  )(1 /  )
2
s
=
,
2
1+
s3 =  .

s0 =

(4.5)
(4.6)
(4.7)

The gluon is therefore produced with three degrees of freedom:  determines the
longitudinal momentum fraction x/  of the gluon, is proportional to the gluons
virtuality, and gives its azimuthal angle. If we are concerned only with the produced
gluon, and not with the parton that emitted it, then is also free. Note that in the
limit 0, the gluon 4-momentum s becomes xP /  , the usual parton model result,
independent of and . This limit gives the standard photongluon fusion contribution
to the observable under consideration. However in Section 3 we saw that it is terms of
higher order in that generate power corrections: the transverse momentum of the gluon
is crucially important. Consequently we might like the gluon to have a 4-dimensional
4-momentum distribution within the proton: this will be discussed in Section 6.
With these definitions, the integration measure for the splitting becomes


1
d 3 p
=
(2)3 2p 0 (4)2

1

d 


2
0

d
2



d
(1 ) s 2  .

(4.8)

Let us also introduce polarisation vectors $i , (i = 1, 2, 3), for the radiated gluon. We

must choose the $i to satisfy s $i = 0 and $i $j = ij . Then we have the identity


3

i=1

$i $i = g +

ss
.
s2

(4.9)

For those contributions to Fig. 1 where both the top and bottom halves are gaugeindependent (such as the lowest-order contribution to Fig. 2), we can use (4.9) to replace
the gluon propagators with sums over polarisation vectors, thereby detaching the top and
bottom halves of the diagram. Where it is not gauge independent, this gives the contribution
in Landau gauge.
For the purpose of later discussion, let us introduce two such sets of polarisation vectors.
Firstly, let us define the natural polarisation vectors $i by imposing q $2 = q $3 =

268

G.E. Smye / Nuclear Physics B 613 (2001) 260284

p $3 = 0. Then the identity (4.9) gives us

$1 =

$2 =

(1 + )  s + 2q
,

iQ(1 )
[(1 + )  2 ]s + [2 (1 + )/  ]q (1 )2 p

Q(1 ) (1 /  )(1 /  )

(4.10)

and $3 is orthogonal to the vector space spanned by s, q and p.


The vectors $2 and $3 are both orthogonal to q, and so are the transverse polarisation
vectors in frames in which s and q are (anti-)parallel, and thus in all physical frames. Also,
since s is spacelike, the longitudinal polarisation vector $1 is timelike. (The factor i in the
expression for $1 then causes the identity (4.9) to hold as written; we could alternatively
have introduced a factor 1 in the sum over polarisation vectors.)
These are the natural polarisation vectors because $1 is longitudinal and $2 and $3 are
transverse: they are thus the polarisation vectors that should be used when considering
parton splitting, and the splitting functions arise naturally from their use.
Secondly, let us define the diagonal basis of polarisation vectors $i by imposing p $2 =
p $3 = q $3 = 0. It will be seen below that this is the basis that diagonalises the matrix

describing gluon production, and is formed by mixing $1 and $2 to give the new basis
vectors

$1 =

$2 =

2  p  s

,
iQ
[2  (1 + ) ]p  s (/  )q

,
Q (1 /  )(1 /  )

$3 = $3 .

(4.11)

The disadvantage of this basis is that $1 and $2 are neither transverse nor longitudinal, and
therefore are not suitable for a discussion of parton splitting or of splitting functions. The
use of this basis will be denoted by a tilde on appropriate symbols.
4.2. Radiation from quark line
The matrix element for the process shown in Fig. 2, integrated using the measure (4.8),
gives the following parton-level contribution to some observable F :




1
2

2(p $i )(p $j )
1
d
d
s CF  d 
q
(1 )  ij
F =
2

2

2
s2
i,j


Fij (, , , ),

(4.12)

where Fij is the contribution due to photongluon fusion with the appropriate gluon
polarisations.

G.E. Smye / Nuclear Physics B 613 (2001) 260284

269

Using the natural basis (4.10), we find




1
2

d
d q
s CF  d 
q
M (/  , )Fij (,  , , ),
F =
2

2
ij
i,j

where

q
Mij (z, ) =

(4.13)

2(1z)(1z)
(1)z

(2zz) (1z)(1z)
i(1)z

(2zz) (1z)(1z)
i(1)z
(2zz)2
2(1)z

(1)z
2

ij

(4.14)

Several comments may be made about this matrix:


(i) The elements on the diagonal correspond to the exchange of a gluon with definite
polarisation; the two off-diagonal elements give rise to an interference term between the
longitudinal and one of the transverse polarisations.
(ii) The usual q g splitting function can be recovered from the total transverse piece
q
q
M22 + M33 in the limit 0:
 q
1 + (1 z)2
q 
.
lim M22 + M33 = Pqg (z) =
0
z

(4.15)

The standard factorised expression for the leading


(logarithmic) divergence arises from

the fact that as 0, s xP /  and $1 s / s 2 , so F11 , F12 and F21 all become
subleading and (for an unpolarised observable) F22 and F33 become equal and functions
of  only. Thus the leading logarithmic divergence in F is
s CF
F leading( ) =
2

1

d 


d
Pqg (/  )Cg (  ),

(4.16)

where Cg represents the contribution from a transversely-polarised initial-state gluon.


This however applies only to the leading divergence: other pieces, in particular those nondivergent pieces generating power corrections, do not factorise in this way. In general, the
functions Fij have dependence on and as well as and  , i.e., for the purposes of
higher twist contributions the gluon remembers the momentum fraction of the particle
that emitted it.
q
(iii) The matrix Mij has zero determinant, and hence has a zero eigenvalue. The other
two eigenvalues are both equal to (1 )z/2. It is diagonalised if we use the diagonal
basis (4.11), in which case the decomposition (4.12) simplifies considerably to:
s CF
F =
2

1

d 


2
0

d
2


d (1 )  
33 .
F22 + F

2 

(4.17)

In other words, only two gluon polarisations, $2 and $3 , are permitted, but they are not
the transverse polarisations assumed in the parton model. Since this basis fails to make

270

G.E. Smye / Nuclear Physics B 613 (2001) 260284

apparent the form of the leading divergence and does not naturally give rise to the splitting
function, it is clear that any attempt to remove the initial quark line from the power
corrections calculated in [9] and [10] cannot be simply a matter, as with the logarithmic
divergences, of deconvoluting with the quark to gluon splitting function. Also, since all the
elements of (4.14) contribute to the power correction we would need to be able to include
the unphysical polarisations in the gluon distribution.
4.3. Radiation from gluon line
The radiation of the virtual gluon from a gluon line is shown in Fig. 4. This diagram is
gauge-dependent, and while its logarithmically divergent piece is gauge-independent and
given by the g g splitting function, the remainder of the diagram, and in particular the
pieces generating power corrections, are not.
The matrix element for the process shown in Fig. 4, integrated using (4.8) and using
the natural basis of polarisation vectors (4.10), gives the following contribution to the
observable F :


1
2

d
d g
s CA  d 
F =
M (/  , )Fij (,  , , ),
2

2
ij
g

i,j

(4.18)

where

(2z)2
2z

2)
(2z)(22z+z

2iz 1z

2iz 1z

24z+4z2 2z3 +z4


z(1z)

(2z)(22z+z2)
g

Mij (z, ) =

z(22z+z2 )
1z

Fig. 4. Radiation of gluon from gluon line.

+ O().

ij

(4.19)

G.E. Smye / Nuclear Physics B 613 (2001) 260284

271

To compare with the quark case, several comments should be made:


(i) The terms of O() are gauge-dependent, and therefore of not much use until we can
isolate pieces of other diagrams to give a gauge-invariant total. The pinch technique
(see next section) succeeds in giving a gauge-invariant result, but at the expense of
changing this leading order matrix.
(ii) Again there are on-diagonal elements corresponding to exchange of a gluon of
definite polarisation, and off-diagonal elements giving interferences. In fact all six
off-diagonal elements are non-zero at O() level, unlike the quark case. In addition

the use of the basis $i does not diagonalise the matrix, even to leading order.
g
g
(iii) The transverse elements M22 and M33 yield polarised splitting functions, and their
sum gives
 g
2(1 z + z2 )2
g 
= Pgg (z).
lim M22 + M33 =
0
z(1 z)
Thus the leading divergence is the well-known result
s CA
F leading( ) =
2

1

d 


(4.20)

d
Pgg (/  )Cg (  ).

(4.21)

(iv) The matrix


diverges at z = 1, i.e., when  = . This is due to the gluon
with momentum p becoming soft, and the divergence is cancelled by virtual
corrections.
g
Mij

4.4. Use of the pinch technique


When we considered the diagram in Fig. 4, we found that the terms giving rise to power
corrections (i.e., the terms in the integrand of (4.18) that do not diverge as 0) are
gauge-dependent. We must consequently include pieces of other diagrams to restore gauge

invariance. We could add in all O(s2 ) diagrams that contribute to the process g q qg,
as in Milan factor calculations for event shape variables (see [11]), but that would force
us to include diagrams unlike that of Fig. 1: it is not clear what relation (if any) these
diagrams have to the scale chosen for the running coupling.
First we notice that Fig. 4 is related by crossing to the e+ e annihilation diagram shown
in Fig. 5. So, just as in e+ e annihilation in [11], we can apply the pinch technique to

Fig. 5. Diagram from e+ e annihilation related by crossing to the present case.

272

G.E. Smye / Nuclear Physics B 613 (2001) 260284

the diagram to define a gauge-invariant contribution. But while the pinch technique has
previously been applied to internal gluon loops [13] and cut loops containing outgoing
partons [11], it is now being applied to a cut loop where one of the particles is incoming
and the other outgoing. This difference is irrelevant since the algebra is identical.
The matrix element for Fig. 5, on applying the pinch technique, is given in Eq. (A.9)
of [11]: in our notation it may be written


1 
|M|2 = g 2 CA 4 8 s 2 g s s + 2(p + p )(p + p )
$i $j Fij (s),
s
i,j

(4.22)
where Fij is the contribution from the shaded blob. Using the crossing relation and
integrating using (4.8), we obtain the following form for the observable F :




1
2

2(p $i )(p $j )

d
d
s CA  d 
g
(1 )  2ij
F =
2

2

s2
i,j


Fij (, , , ).

(4.23)

Note that this differs from the incoming quark case only by having a different constant
multiplying ij . In fact the only linearly independent terms allowed inside the square
bracket, consistent with gauge invariance and having the correct dimensions, are the two
shown. This was also seen in the e+ e annihilation calculation of [11].
Using the natural basis (4.10), we find


1
2

d
d g
s CA  d 
M (/  , )Fij (,  , , ),
F =
2

2
ij
g

i,j

where

g
Mij (z, ) =

(4.24)

(23z+z)(2+z3z)
2(1)z

(2zz) (1z)(1z)
i(1)z

(2zz) (1z)(1z)
i(1)z
(2zz)2
3(1)z
2(1)z +
2

2(1 )z

ij

(4.25)
As previously, there are several comments to be made:
(i) Although this matrix is gauge invariant, it does not reduce to (4.19) in the limit
0. One consequence of this is that the g g splitting function is no longer trivially
g
g
recovered from the total transverse piece M22 + M33 in the limit 0:
 g
2(1 z + 2z2 )
2z3
g 
= Pgg (z)
,
lim M22 + M33 =
0
z
1z

(4.26)

where the usual splitting function is


Pgg (z) =

2(1 z + z2 )2
.
z(1 z)

(4.27)

G.E. Smye / Nuclear Physics B 613 (2001) 260284

273

The additional piece arises from the fact that the pinch part of the diagram was removed (or
alternatively that the pinch parts of other diagrams were added) in order to secure gauge
invariance. This can be understood as follows: the factor 1/(1 z) in the splitting function
can only arise from the collinear limit of the gauge-dependent piece (n p)/(n p ) in the
sum over polarisations of the p gluon in Fig. 4. In order to achieve gauge invariance we
must add or remove terms so as to cancel all n-dependence, not just in the collinear limit
but identically for all p and p , and so terms with the factor 1/(1 z) must necessarily
disappear.
This happens even though the remaining diagrams themselves have no logarithmic
divergence associated with the gluon splitting (although of course they may have
divergences arising elsewhere): applying the pinch technique separates the diagrams into
a pinch part and a remainder, which have equal and opposite logarithmically divergent
pieces. Thus this is not a sensible way to study splitting functions. Also, since the only two
linearly-independent gauge-invariant terms with the correct dimension are those in (4.23),
and no linear combination of these can reproduce the splitting function, we see that we
cannot have a fully gauge invariant expression for the splitting g g + g (with some
suitable definition for what is meant by g , e.g., use of pinch technique) while retaining
the full splitting function.
We also notice that the contribution (4.26) is no longer invariant under z 1 z: this is
because the two daughters of the splitting are no longer identical, one being a real on-shell
gluon and the other a virtual gluon with a modified propagator.
(ii) The difference between the production matrix employing the pinch technique and
that using standard perturbation theory is

2z

(2z)z
g
Mij (z, ) =

2i 1z

(2z)z

2i 1z
z3
1z

0
+ O()
z3

1z

(4.28)

ij

(where O() terms are gauge-dependent). This is the contribution that gives the additions
to the splitting function. It becomes zero as z 0, which is as the interacting virtual gluon
becomes soft.
(iii) As in the quark and conventional gluon cases, there are both on- and off-diagonal
g
elements. Now, however, the matrix Mij is diagonalised using the diagonal basis (4.11),
just as in the case where the emission is from a quark line. (Indeed one can trivially see
that any linear combination of the two terms in (4.23) is diagonalised using this basis.) One
eigenvalue is 3(1 )z/2, and the other two are both 2(1 )z. Then the decomposition
(4.23) simplifies considerably to:

s CA
Fg =
2

1

d 


2
0

d
2


0



d 2(1 ) 3 


F11 + F22 + F33 .


4

(4.29)

274

G.E. Smye / Nuclear Physics B 613 (2001) 260284

5. Structure functions
Consider now the part of the process involving the interaction of the virtual gluon
within the proton with the virtual photon. This is the process shown in diagram 3, which
corresponds to the quantities Fij above.
Working in the Breit frame, the 4-momentum of the incoming photon and gluon are
given by (4.2) and (4.4); in this part of the process the variables  , and are considered
fixed. Let us introduce the variables = P r/P q, = P k/P q, the azimuthal angle
between r and s, and the azimuthal angle between k and s. The kinematics are then
given by:


r = 12 Q z0 , z cos( + ), z sin( + ), z3 ,
(5.1)


1
k = 2 Q z 0 , z cos( + ), z sin( + ), z 3 .
(5.2)
The definitions of and along with the on-shell conditions for the outgoing particles
require that
2
z
,
4
z 2
z 0 = + ,
4

z0 = +

2
z
,
4
z 2
z 3 = ,
4

z3 =

(5.3)
(5.4)

whence conservation of the 0th and 3rd components of 4-momentum give the conditions
+ = 1 /  2 ,
2
z 2
z
(1  ) + (1 /  )
+ =
,
4 4


(5.5)
(5.6)

while conservation of transverse momentum requires that s , z and z satisfy the triangle
inequalities
|z s |  z ,

(5.7)

|z s |  z .

(5.8)

An additional variable, , is required to parametrise the permitted values of z and z .


Let us choose to write:



2
z
(5.9)
= 4 a + b + 2 cos ab ,



2
z
(5.10)
= 4 a + b 2 cos ab ,
where
a=

(1  )(  )
,
 2 (1 /  2 )2

b=

(1 /  )(1 /  )
.
 2 (1 /  2 )2

(5.11)

Given s , z and z , the angles and are determined up to a sign. We may then
choose and as the independent variables, with phase space
0   1 /  2 ,

0   .

(5.12)

G.E. Smye / Nuclear Physics B 613 (2001) 260284

275

To integrate the matrix elements we apply the operator



d 3r
d 3k
(2)4 4 (q + s r k).
(5.13)
3
0
(2) 2r (2)3 2k 0
Integrating with respect to k and making substitutions for r gives


1
z dz d d (A B 2s z cos )
(5.14)
,

4(2)2
B 2s z cos
where A and B do not depend on .
Next we integrate over . There are two values satisfying the integration condition, and
they differ by a sign. This gives

2
dz
1
d
.
(5.15)
2
4(2)
s z | sin |
Applying the parametrisation in terms of , we find that
 2

 z 


(5.16)

  = 8 sin ab,




s z | sin | = 4 1 /  2 sin ab,


(5.17)
and therefore the integral operator for the photongluon fusion is
1
8 2 (1 /  2 )

2
1/


d
0

d.

(5.18)

We calculate the quantities C ij (,  , ) as defined by




2
d
s TR  2
ij

Fij (,  , , ),
eq  C (, , ) =
2
2

q

(5.19)

q

where the sum over represents the outgoing quarkantiquark flavours. We first perform
the integral: the contributions to FL have no -dependence, while the contributions to FT
have terms proportional to cos and cos2 . Then the integration over can be performed
by writing t = tan(/2), and that over by carefully expanding in powers of =  2 as
described in [9]. We obtain, for the longitudinal structure function,


 
CL11 = 32(1  ) (3 + 4  ) + (1  ) log + O 2 ,
(5.20)



CL12 = CL21 = 32i 1 /  (1  )(1 3   log ) + O 2 ,
(5.21)


22






2
CL = 8 (1 ) + 8 1/ + 2/ 10 8 + 23 + 9 17


 

+ /  3 2 + 8  + 2  6  2 log + O 2 ,
(5.22)


33






2
CL = 8 (1 ) + 8 1/ + 2/ 10 8 + 23 + 7 15


 

+ /  3 2 + 8  + 2  6  2 log + O 2 .
(5.23)
Hence in the limit 0 we recover the well-known results CL11 = CL12 = CL21 = 0 and
CL22 = CL33 = 8  (1  ).

276

G.E. Smye / Nuclear Physics B 613 (2001) 260284

For the transverse structure function we obtain




CT11 = 8  (1  ) + 8 2/3  + 25/3 + 2 15  2  + 8  2

 
+ 2(1  )2 log + O 2 ,
(5.24)
4i 
/3  2 2/3  11/3  + 10/3 + 18 20 
CT12 = CT21 =
1 / 

 

14  + 16  2 2(1  )(4  3 ) log + O 2 ,
(5.25)


22

2
CT = 2 1 2 + 2 (2 + log )

2

2 /3 3 + 8/3  2 8/3  + 13 2 /3  2 88/3  + 76/3


1 / 

14 2 /  + 60 44  + 16 2 48  + 30  2

+ 2 /  2 1/  + 2 /  2 7/  + 6 2 /  + 9 7 


 
+ 2 6  + 4  2 log + O 2 ,
(5.26)


CT33 = 2 1 2  + 2  2 (2 + log )

2
2 /3 3 + 4/  2 4/  + 11 2 /3  2 28/  + 24

1 / 

18 2 /  + 68 52  + 16 2 48  + 34  2

+ 2 /  2 1/  + 2 /  2 7/  + 6 3 2 /  + 13 11 


 
+ 3 2 10  + 8  2 log + O 2 .
(5.27)
As 0 the pieces CT22 and CT33 diverge logarithmically: this is the piece associated with
the collinear splitting of the gluon into a quarkantiquark pair.

6. Power corrections
We noted above that the interaction of the produced virtual gluon is given independently
of the production mechanism by the function Fij (,  , , ), where i and j represent
polarisations and ,  , and parametrise the gluons 4-momentum. In general,
calculations of power corrections will depend on all four of these quantities.
Thus a natural and intuitive way to combine this with a gluon distribution function might
be to write


1
2
1
  d  d   d  d
gij (x/  , , /  )Fij (,  , , ),
F (x) =


2

i,j x

(6.1)

where gij (x/  , , /  ) is the relevant gluon distribution function which by symmetry
does not depend on . Of course this needs to be treated with some care, since both halves
of the diagram in Fig. 1 are gauge-dependent and so therefore are all these distributions
when the gluon is off-shell, i.e., = 0. Nevertheless, we can still make progress: we

G.E. Smye / Nuclear Physics B 613 (2001) 260284

277

present an intuitive argument that, although mathematically not totally rigorous, is helpful
in understanding the underlying physics.
For the diagonal elements we can view this probabilistically: the expected number of
gluons of polarisation i in an element of parameter space du d dz at (u, , z), where
u = x/  is the longitudinal momentum fraction of the proton carried by the gluon, and
z = /  , is:
gii (u, , z) du

d dz
.
z

(6.2)

This is intended to be schematic only 2 clearly any gluon not collinear with the proton,
i.e., with = 0, will experience the confining effect of the QCD potential, and so cannot
exist for more than a short time (where short in this context means 1/QCD ). Such
particles are not asymptotically free: they can only be resolved at high momentum scales,
and it is precisely this resolution of gluons with non-zero transverse momentum that gives
rise to the running of the parton distributions. Power corrections are also known to arise
from the interactions of gluons with non-zero transverse momentum (see, e.g., [15]), and
so are also generated by the resolution of short-lived virtual gluons within the proton. Thus
we expect power corrections in singlet DIS to be intimately related to the running of parton
distributions.
Consider first the transverse polarisations: these are the physical polarisations. Speaking
schematically (i.e., not being too precise about our choice of factorisation scheme), we
may view the polarised gluon distribution function gi (u, M 2 ) as representing the gluons
with transverse momentum less than M. So


gi u, M 2 =
for i = 2, 3, where
Therefore,
gi (u, M
M2
M 2

1

dz
z

(M
 2 )

d
gii (u, , z),

(6.3)

2)
(M

is the value of corresponding to a transverse momentum M 2 .

2)

1
=
u

dz M 2
gii (u, ,
z),
z M 2

(6.4)

and, using Eq. (4.6), we obtain


M 2
1 z

= 1 + O().

=
2
M
1 2z

Further, we know from the DGLAP equation that


gi (u, M
M2
M 2

2)

s
=
2

1


dz 
i
i
CF qtot(u/z)Pqg
(z) + CA g(u/z)Pgg
(z) ,
z

(6.5)

(6.6)

u
2 It could of course be more rigorously formulated in terms of well-defined unintegrated parton distributions,
but that is unnecessary for the purposes of the argument given here.

278

G.E. Smye / Nuclear Physics B 613 (2001) 260284

so for the sake of illustration let us make the simplest consistent assignment for the
differential gluon distribution gii (u, ,
z), which is

s 
i
i
CF qtot (u/z)Pqg
z) =
(z) + CA g(u/z)Pgg
(z) + O().

gii (u, ,
(6.7)
2

Here qtot(u) = q [q(u) + q(u)]

is the total quark and antiquark content of the proton.


So let us return to (6.1) and define the coefficient function
Cg (  ) = Fii (,  , 0, )

(6.8)

which is independent of , and polarisation i. Let us separate off the piece that generates
the collinear divergence by writing
2

Fii (,  , , )


  
d
= Cg (  ) M 2 + Fii (,  , ),
2

(6.9)

we then obtain the contribution to the observable F from incoming transversely-polarised


gluons as:


1  
2
3 1

d
d
d
d
gii (x/  , , /  )Fii (,  , , )
Ftrans (x) =

2
i=2 x

1
=

(6.10)


d  
g x/  , M 2 Cg (  )


x


1  
3 1

d
d
d
gii (x/  , , /  )Fii (,  , ).
+


i=2 x

(6.11)

The first term on the right-hand side of (6.11) is simply the standard contribution to
photongluon fusion, evaluated with an on-shell initial gluon and convoluted with the

gluon distribution g(x) = 3i=2 gi (x). This contains logarithmic scaling violations given
by the running of the gluon distribution, but contains no power corrections associated with
the virtuality of this gluon.
The second term gives us the required power corrections, since this is the term with an
integral over the gluons virtuality. Let us now substitute the differential gluon distribution
with expression (6.7), thus obtaining


1  
3 1

d
d
d
gii (x/  , , /  )Fii (,  , )
Ftrans (x) =


i=2 x

0


1  
3 1
d
d 
s  d
i
(/  )
=
CF qtot (x/ )Pqg

2

i=2 x

i
(/  ) + O()
+ CA g(x/ )Pgg

Fii .

(6.12)

G.E. Smye / Nuclear Physics B 613 (2001) 260284

279

Although the above argument has been restricted to the contributions from transverse
polarisations, and even these have been treated only schematically, we are now in a position
to be able to see what is happening physically. We have gone round in a big circle: starting
with Figs. 2 and 4, we have detached the lower parts of the diagrams corresponding to gluon
production, to give the piece corresponding to photongluon fusion. But then in order to
convolute with the correct differential gluon distribution within the proton we had to make
use of the DGLAP equation, which reintroduced those lower legs of the diagrams. Thus for
the purposes of power corrections we find that the lower parts of those diagrams genuinely
are important and cannot be removed. This also explains the presence of the longitudinal
gluon polarisation: we are not dealing with a real incoming gluon and so we need not be
restricted to physical polarisations. Eq. (6.12) is intended to be illustrative only it was
not rigorously derived. Yet it is clear that while the leading perturbative contribution to
singlet DIS, given by the first term in Eq. (6.11), is O(s ), the 1/Q2 power corrections
as well as the logarithmic scaling violations are O(s2 ), the additional factor of s arising
from the DGLAP equation. (Contrast this with flavour non-singlet DIS, whose lowest order
is O(s0 ) but which has leading power corrections at O(s ).)
Thus the leading singlet power corrections are from the diagrams 2 and 4 as they appear:
we have two independent renormalon chains, and the magnitude of the power corrections
is given by the quantities D1 and D2 of Eqs. (3.12) and (3.13).
($) generating power corrections to structure functions is
The characteristic function F
thus given by

  1

d TR CF
TR CA
2
q
g

eq 
qtot(x/ )C (, $) +
g(x/ )C (, $) ,
F ($) =
(2)2
(2)2

q

(6.13)
where

1

C q/g (, $) =

d 


1/
0

 

d  q/g 
Mij /  ,  2 C ij ,  ,  2 ,
+$

(6.14)

i,j

and the C ij were defined in Eq. (5.19) and calculated in (5.20) to (5.27). The sum over
q  represents outgoing quarkantiquark flavours in the photongluon fusion. We retain
($) up to O($) that are non-analytic as $ 0. In order to define a gaugeterms in F
invariant quantity with a natural scale for the running coupling, we use the matrix (4.25)
evaluated using the pinch technique the remaining diagrams are also expected to give
power corrections, but as in [11] they are expected to have a different structure.
Using (4.13) and (4.24) we find

q
CT = 29 2 63 + 63 2 2 3 + 12 log 27 log

27 2 log + 12 3 log log $



+ 23 4 + 3 3 2 4 3 + 6 log + 6 2 log log 1 + 12 log $ log $


+ 25 2 + 25 2 25 3 2 5 + 15 2 log + 15 3 log $ log $



2 5 2 5 3 + 2 2 log + 2 3 log log 1 + 12 log $ $ log $, (6.15)

280

G.E. Smye / Nuclear Physics B 613 (2001) 260284



q
CL = 83 1 3 + 2 3 3 2 log log $

8
17 + 75 2 125 3 + 33 5 + 30 log
225

+ 75 3 log 45 5 log $ log $



8
2 15 2 + 10 3 + 3 5 15 3 log log 1 + 12 log $ $ log $,
+ 15
(6.16)
which is the result given in [9], and

g
CT = 49 1 99 + 99 2 3 + 6 log 27 log

54 2 log + 33 3 log log $



+ 43 2 + 6 + 3 2 11 3 + 3 log + 12 2 log log 1 + 12 log $ log $


+ 45 1 + 40 2 35 3 6 5 + 15 2 log + 30 3 log $ log $



8 5 2 5 3 + 2 2 log + 2 3 log log 1 + 12 log $ $ log $, (6.17)


g
CL = 83 1 3 9 2 + 11 3 12 2 log log $

8
17 + 1275 2 1475 3 + 183 5 + 30 log + 450 2 log
225

+ 750 3 log 270 5 log $ log $



2
3
5
3
1
+ 16
15 1 15 + 5 + 9 30 log log 1 + 2 log $ $ log $.
(6.18)
Thus we obtain the
quark distribution are
q
Fi (x) =

1/Q2

power corrections to FT (x) and FL (x): those arising from the


 1
TR CF  2
d
q
qtot(x/ )Ci ( ),
eq 
(2)2


q

(6.19)

where

D1 2(2 + 25 2 25 3 2 5 + 15 2 log + 15 3 log )
Q2
5



D2
2
2
2 5 5 + 2 log + 2 log log
(6.20)
,
eQ2

D1
8(17 + 75 2 125 3 + 33 5 + 30 log + 75 3 log 45 5 log )
q
CL ( ) = 2
Q
225

2
3
5
3
D2
8(2 15 + 10 + 3 15 log )
log
+
(6.21)
15
eQ2

q
CT ( ) =

and those from the gluon distribution are


g
Fi (x) =


 1
d
TR CA  2
g
g(x/ )Ci ( ),
eq 

(2)2

q

(6.22)

G.E. Smye / Nuclear Physics B 613 (2001) 260284

281

where


D1 4(1 + 40 2 35 3 6 5 + 15 2 log + 30 3 log )
Q2
5



D2
,
8 5 5 2 + 2 log + 2 2 log log
eQ2

8 
D1
g
CL ( ) = 2
17 + 1275 2 1475 3 + 183 5 + 30 log
225
Q

+ 450 2 log + 750 3 log 270 5 log

D2
16(1 15 2 + 5 3 + 9 5 30 3 log )
log
+
.
15
eQ2
g

CT ( ) =

(6.23)

(6.24)

7. Results and conclusions


Fig. 6 shows the magnitudes K of the 1/Q2 power corrections to structure functions,
given by
q/g

FT /L (x)
(0)
FT (x)

D1 q/g
K (x),
Q2 T /L

(7.1)

(0)

where FT (x) is the Born-level result for the transverse structure function given in
Eq. (2.2). These were calculated at Q2 = 500 GeV2 , using the corresponding MRST

Fig. 6. Graph showing KT (x) and KL (x). The dotted lines represent the contributions from the
quark distribution; the dashed lines represent the gluon contributions. The totals are represented by
the solid lines.

282

G.E. Smye / Nuclear Physics B 613 (2001) 260284

Fig. 7. Graph showing K2 (x)/x. The dotted line represents the contribution from the quark
distribution; the dashed line represents the gluon contribution. The total is represented by the solid
line.

(central gluon) parton distributions [16], and assuming four flavours of outgoing quark
antiquark pairs. The unknown value of D2 /e was set to 0.06 GeV2 , i.e., approximately 2 ,
although the qualitative behaviour of these power corrections does not change provided we
keep D2  Q2 .
The dominant contribution to both of these comes from the gluon distribution, as may be
expected at these relatively low values of x. All the contributions tend to zero at large x, and
become large at small x. The contributions to KT are positive and those to KL negative,
but they have similar magnitudes.
The corresponding quantity related to F2 /x, which is K2 /x = KT + KL , is shown in
Fig. 7. The positive and negative contributions partially cancel, giving a power correction
smaller by a factor 3 or 4 than that for FT or FL .
The power corrections are multiplied by the unknown factor D1 , defined in (3.12). If
D1 is positive, we might expect FT (x) to show a positive 1/Q2 power correction and
FL (x) a negative one. These results are all qualitatively the same as those in [9], which
took into account only the quark contribution, but the inclusion of the gluon contribution
increases considerably the size of the correction, in the case of FT by more than an order
of magnitude.

8. Discussion
The calculation of power corrections using the renormalon model or the dispersive
approach in flavour singlet DIS is a non-trivial problem, due to the fact that Fig. 1 as it

G.E. Smye / Nuclear Physics B 613 (2001) 260284

283

stands is gauge dependent, both in the upper and lower halves of the diagram. These halves
may both be considered to be cut insertions into a single gluon propagator, albeit in another
channel, and so the pinch technique can be used to define a natural gauge-invariant quantity.
The diagram 1 evaluated using the pinch technique gives only one gauge-independent part
of the full O(s2 ) amplitude, and the expectation is that the other parts will also give power
corrections, although not taking the form of two renormalon chains with equal 4-momenta.
Power corrections thus arise when a virtual gluon is radiated from a quark or gluon, and
this virtual gluon then interacts with the photon. We found that, as well as the physical
transverse polarisations, we also have contributions to power corrections from longitudinal
polarisations and interference terms. In addition, since the gluons 4-momentum is not
simply some multiple of that of the proton, the contribution from the photongluon fusion
is a function of all four momentum components. In particular the kinematics of the virtual
gluon production are important: where it is radiated from an on-shell parton, the gluon
remembers the momentum fraction of the parton that emitted it. This, along with the
discussion of the virtual gluon distribution in Section 6, indicates that we cannot detach
the two halves of Fig. 1 but rather that the power corrections arise from the full diagram
and are thus an O(s2 ) effect.
Calculations of power corrections to structure functions and fragmentation functions due
to gluon radiation by a quark are already published in the literature [810]. However the
contributions from radiation by a gluon dominate, because of the relative behaviours of
the quark and gluon parton densities. This larger correction was evaluated above for the
structure functions. Above x = 0.05 the power corrections to the structure functions are
small, but we predict that for x below 0.05 the corrections grow. So the correction to F2
at x = 0.01 and Q2 = 4 GeV2 might be around 2% (assuming a reasonable value of
D1 0.1 GeV2 ). The corrections to FT and FL are expected to be slightly larger. In any
case, we do not expect to see the large ( 50%) corrections predicted by [8].
It was seen that the use of the pinch technique lead to a failure to recover the usual g g
splitting function. This is not a problem the remaining pieces come from the remaining
diagrams. All the pinch technique has done is move certain terms from one diagram to
another. But the point of using this particular separation of terms is that one can define
a gauge-invariant QCD effective charge, so the virtuality of the gluon enters as the natural
scale for the coupling; it is not clear that this is the case for the other terms.
Finally, there is clearly some similarity between the situation here and that of a
decaying outgoing virtual gluon in e+ e annihilation (Fig. 5), where we also have two
renormalon chains with equal 4-momenta k, for some k. Again we have an integral over
the virtuality k 2 , in which the coupling appears as |s (k 2 )|2 . But in that case we use
Eq. (3.2) to convert s2 into the spectral function s (k 2 ), and the two chains became one,
with a cut bubble insertion. There are two reasons why we cannot do the same here. Firstly,
in the e+ e case, the virtual gluon is timelike, k 2 is positive and thus we have s (k 2 )
integrated over positive k 2 . But here the gluon is spacelike, so k 2 is negative, and while
it is quite possible to convert the two factors of s into a single s (k 2 ), it is integrated
over negative values of its argument. It is therefore not naturally manipulated into standard
single-chain form. Secondly, although the algebra of the Feynman diagrams is identical

284

G.E. Smye / Nuclear Physics B 613 (2001) 260284

(i.e., related by crossing), in the e+ e case we can simply integrate over the cut bubble
but in DIS we must include parton density functions for the incoming quark or gluon. This
makes it quite impossible to integrate out the crossed cut bubble that is the mechanism for
production of the virtual gluon. So, while there are interesting and useful parallels between
power corrections to singlet DIS and e+ e annihilation with outgoing gluon splitting, there
are also significant differences and the relationship between them is not as simple as may
naively have been supposed.

Acknowledgements
The author wishes to thank Yuri Dokshitzer, Mark Smith, David Summers, Jay Watson
and Bryan Webber for helpful discussions and comments.

References
[1] M. Beneke, Phys. Rep. 317 (1999) 1, hep-ph/9807443.
[2] Yu.L. Dokshitzer, G. Marchesini, B.R. Webber, Nucl. Phys. B 469 (1996) 93, hep-ph/9512336.
[3] O. Biebel, Contribution to the proceedings of the DIS 2000 conference, April 2000, hepex/0006020.
[4] M. Dasgupta, B.R. Webber, Phys. Lett. B 382 (1996) 273, hep-ph/9604388.
[5] E. Stein, M. Meyer-Hermann, L. Mankiewicz, A. Schfer, Phys. Lett. B 376 (1996) 177, hepph/9601356;
M. Meyer-Hermann, M. Maul, L. Mankiewicz, E. Stein, A. Schfer, Phys. Lett. B 383 (1996)
463, hep-ph/9605229;
M. Meyer-Hermann, M. Maul, L. Mankiewicz, E. Stein, A. Schfer, Phys. Lett. B 393 (1997)
487, Erratum;
M. Maul, E. Stein, A. Schfer, L. Mankiewicz, Phys. Lett. B 401 (1997) 100, hep-ph/9612300;
M. Maul, E. Stein, L. Mankiewicz, M. Meyer-Hermann, A. Schfer, hep-ph/9710392;
M. Meyer-Hermann, A. Schfer, hep-ph/9709349.
[6] M. Dasgupta, G.E. Smye, B.R. Webber, J. High Energy Phys. 04 (1998) 017, hep-ph/9803382.
[7] M. Dasgupta, B.R. Webber, Eur. Phys. J. C 1 (1998) 539, hep-ph/9704297;
M. Dasgupta, B.R. Webber, J. High Energy Phys. 10 (1998) 001, hep-ph/9809247.
[8] E. Stein, M. Maul, L. Mankiewicz, A. Schfer, Nucl. Phys. B 536 (1998) 318, hep-ph/9803342.
[9] G.E. Smye, Nucl. Phys. B 546 (1999) 315, hep-ph/9810292.
[10] G.E. Smye, Nucl. Phys. B 549 (1999) 263, hep-ph/9812251.
[11] G.E. Smye, JHEP 0105 (2001) 005, hep-ph/0101323.
[12] J.M. Cornwall, Phys. Rev. D 26 (1982) 1453;
J.M. Cornwall, J. Papavassiliou, Phys. Rev. D 40 (1989) 3474;
N.J. Watson, Nucl. Phys. B 552 (1999) 461, hep-ph/9812202;
J. Papavassiliou, Phys. Rev. Lett. 84 (2000) 2782, hep-ph/9912336.
[13] N.J. Watson, Nucl. Phys. B 494 (1997) 388, hep-ph/9606381.
[14] R.D. Peccei, R. Rckl, Nucl. Phys. B 162 (1980) 125.
[15] R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 207 (1982) 1;
R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 212 (1983) 29.
[16] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 1 (1998) 463, hepph/9803445.

Nuclear Physics B 613 (2001) 285305


www.elsevier.com/locate/npe

Constraining models with vector-like fermions


from FCNC in K and B physics
G. Barenboim a,b , F.J. Botella c , O. Vives c
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Fermilab Theory Group, MS 106 P.O. Box 500, Batavia, IL 60510, USA
c Departament de Fsica Terica and IFIC, Universitat de Valncia, CSIC, E-46100, Burjassot, Spain

Received 12 June 2001; accepted 6 August 2001

Abstract
In this work we update the constraints on tree level FCNC couplings in the framework of a theory
with n isosinglet vector-like down quarks. In this context, we emphasize the sensitivity of the B
J /KS CP asymmetry to the presence of new vector-like down quarks. This CP asymmetry, together
with the rare decays B Xs,d l l and K are the best options to further constrain the FCNC
tree level couplings or even to point out, in the near future, the possible presence of vector-like quarks
in the low energy spectrum, as suggested by GUT theories or models of large extra dimensions at the
TeV scale. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.15.Ff; 12.60.-i; 13.25.Hw; 13.25.Es; 12.15.Mm

1. Introduction
The study of flavor changing neutral currents (FCNC) in particle physics phenomenology has played a key role in the advance of high energy physics in the past decades. Already
in the early seventies the non-observation of FCNC was used to predict the existence of the
charm quark through the GIM mechanism [1] well before its direct experimental discovery.
Subsequently, a precise prediction of its mass was made from FCNC kaon processes [2].
 mass difference
Later on, the discovery of the bottom quark and the measure of the BB
indicated the presence of a heavy top quark. Due to this GIM mechanism, FCNC in the
SM arise only at higher loop level and are suppressed by powers of light quark masses and
small mixing angles. This strong suppression make FCNC phenomena a privileged ground
to search for signs of new physics beyond the SM. In many extensions of the SM there is no
analog of the GIM mechanism to protect FCNC processes and hence potentially large new
E-mail address: vives@ific.uv.es (O. Vives).
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 1 ) 0 0 3 9 0 - X

286

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

physics contributions can be expected. For instance, a minimal supersymmetric extension


of the SM with generic soft breaking terms gives rise to dangerously large contributions to
FCNC and CP violating observables through loop contributions with the supersymmetric
particles running in the loop [3]. Even more challenging (or perhaps more dangerous) than
these loop induced FCNCs is the inclusion of tree level FCNC couplings.
Indeed, the presence of tree level FCNC is phenomenologically, a very interesting
possibility. A minimal extension of the SM with the only addition of an extra isosinglet
down quark in a vector like representation of the SM gauge group induces FCNC couplings
in the Z and neutral Higgs boson couplings. Some, more realistic models, include at least
one vector-like quark per generation. These models naturally arise, for instance, as the lowenergy limit of an E6 grand unified theory [4]. Moreover, vector-like quarks can appear in
models of large extra dimensions. The existence of extra spatial dimensions of the TeV
size implies the existence of towers of KaluzaKlein modes for the particles that propagate
in the new (extra) dimensions. Therefore, if some SM fermion propagates in the extra
dimensions, for each of these chiral quarks and leptons there is a tower of vector-like
fermions whose separation is of order 1/R, being R the radius of the extra dimensions [5,
6]. From a more phenomenological point of view, models with isosinglet quarks provide
the simplest self-consistent framework to study deviations of 3 3 unitarity of the CKM
matrix as well as flavor changing neutral currents at tree level. Of course, these FCNC
couplings are severely restricted by the low energy results on the different FCNC processes.
Nevertheless, it is well known that even fulfilling these strong constraints these couplings
can have large effects on B factory experiments on CP violation [8,10,11]. In a recent
paper [11] we showed that the possible mismatch between the SM expectations and the
measured value for the CP asymmetry in the B J /Ks decay can be easily explained in
a model with an additional vector like down quark. In this paper we complete the previous
analysis and generalize it to the case where a tower of vector-like down quarks (VLdQ) is
present. In first place, we update the strong low-energy constraints on the tree-level FCNC
couplings and then we concentrate on the CP violation observables in the B factories as
the most efficient observables to find or at least constrain these tree level FCNC couplings.
Finally, we compare the simplest case with a single VLdQ with models with several vectorlike down quarks.

2. FCNC in the presence of isosinglet quarks


As explained in the introduction, vector-like fermions appear in different extensions
of the SM, as for instance E6 GUT models or, remarkably enough, theories with large
extra dimensions on the TeV scale. All these models have several new vector like fermions
that mix with the SM fermions and can have important effects on the low energy FCNC
phenomenology. In this paper we will study FCNC processes with external down quarks,
and so we will be mainly interested in models with isosinglet vector-like down quarks
(VLdQ). Note that the presence of extra vector-like down (up) quarks generates FCNC
in the down (up) couplings to the Z and H . In view of these new possibilities of having

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

287

VLdQs in the low energy spectrum, and especially, to reach a rough understanding of
FCNC effects in models with large extra dimensions [5,6], we present our bounds in a
general framework with n additional VLdQs. A general analysis with both up and down
isosinglets and isodoublets will be presented elsewhere [7].
From the low energy point of view, the model we analyze here has been thoroughly
described in Ref. [8,9]. Its main feature is that the additional VLdQs mix with the three
ordinary quarks and consequently the mass matrix in the down sector is now (3 + n)
(3 + n). This matrix is diagonalized by two (3 + n) (3 + n) mixing matrices and only
the left-handed rotation is observable giving rise to the CKM matrix. As we have already
announced in the introduction, the fingerprint of this model is that it allows tree level FCNC
in the Z and H vertices. To see how these FCNC couplings appear, we can work in the
basis where the up quark mass matrix is diagonal. Here the down quark mass matrix is
diagonalized by a (3 + n) (3 + n) unitary matrix, V . In this model charged current
couplings are unchanged except that VCKM is now the 3 (3 + n) upper submatrix of
V , and at low energies a non-unitary 3 3 mixing matrix appears at tree level. However,
the mixing of doublet and singlet weak eigenstates modifies the neutral current couplings
that in terms of flavor eigenstates will be V Diag(1, 1, 1, 0)V = I44 . So, neutral current
interactions are given by


g

u Li uLi dL U dL 2 sin2 W Jem


Z ,
LZ =
2 cos W
LH =


g 
u Li mui uLi dL U md dL ,
2MW

U =

3


Vi
Vi ,

(1)

i=1

where Uds , Ubs or Ubd = 0 would signal new physics and the presence of FCNC at tree
level.
In the following we analyze the effects of these new couplings in FCNC processes and
we update the phenomenological bounds on them. More specifically, in processes changing
flavor in one unit, %F = 1, that in the SM go through electroweak penguin diagrams
of order GF Vti Vtj , we simply consider the dominant tree level FCNC contributions,
order GF Uij . Similarly, in %F = 2 processes mediated by boxes in the SM of order
GF (Vti Vtj )2 we include two different additional contributions, a pure tree level diagram
with two FCNC vertices, order GF Uij2 and a mixed SM vertex contribution with a tree
level vertex, roughly order GF Vti Vtj Uij [12]. Other new physics contributions in this
framework will always be suppressed by additional loop factors or higher powers of the
FCNC couplings with respect to the contributions considered here [13]. Therefore, the
effective low energy theory we consider is then identical to the SM with a non-unitary
CKM matrix and the presence of tree level FCNC as shown in Eq. (1).

3. Experimental constraints on the extended mixing matrix


As we have seen in the previous section, all the new physics effects in our model are
encoded in the extended mixing matrix that gives rise to tree level FCNC in the Z and H

288

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

vertices. The minimal extension from the SM would consist in the addition of a single
vector-like down quark. In this situation the mixing matrix can be parametrized as [14],
SM
V = R34 (34, 0)R24 (24 , 3 )R14 (14 , 2 )VCKM
(12 , 13 , 23, 1 ),

(2)

where
is 4 4 block diagonal matrix composed of the standard
CKM [15,16] and a 1 1 identity in the (4, 4) element, and Rij (ij , k ) is a complex
rotation between the i and j families. Note that, in the limit of small new angles, we
follow the usual phase conventions. In fact, this 4 4 mixing matrix can represent a good
approximation to more complete models with several vector-like generations if the mixings
are hierarchical, similarly to the standard CKM matrix.
On the other hand, given that the deviations from unitarity of the CKM mixing matrix
are experimentally known to be small, it is possible to use an approximate parametrization
of the mixing matrix valid for an arbitrary number of vector-like quarks [9]. This is an
extension of the usual Wolfenstein parametrization of the CKM [17] in terms of


Udd =
|Vid |2 = 1 Dd2 ,
Usd =
Vis Vid ,
Vus = ,
SM ( , , , )
VCKM
12 13 23 1

i=u,c,t

Vcb = A ,
2

Uss =

i=u,c,t
2

= 1 Ds2 ,

|Vis |

|Vib |

= 1 Db2 ,

Ubs =

i=u,c,t

Vub = A e ,
3 i

Ubb =

Vib Vis ,

i=u,c,t
2

Ubd =

i=u,c,t

Vib Vid ,

(3)

i=u,c,t

n+4

with (Usd , Ubd , Ubs ) general complex numbers and


= i=4 |Vij |2 positive real
numbers. It is possible to obtain the remaining elements of the 3 3 submatrix
corresponding to the SM mixing matrix as a function of (, A, , , Dd2 , Ds2 , Db2 , Usd ,
Ubd , Ubs ). In fact, taking into account that (Dd2 , Ds2 , Db2 ) can be at most of order 3 and
(Usd , Ubd , Ubs ) are experimentally constrained to be  O(4 ), we can keep quadratic or
linear terms in (Dd2 , Ds2 , Db2 ) and (Usd , Ubd , Ubs ), respectively, and we obtain to O(6 ),
Dj2

2 4 1 + 8A2 2 6 Dd2 Dd2 2Ds2 2

2
8
16
2
4
 
D4
d + Re{Usd } + O 7 ,
8
2 1 + 4A2 4 1 4A2 + 16A2 cos 6

Vcs =1
2
8
16
 
D2 D2
D4
s s 2 + A2 Re{Ubs } s + O 7 ,
2
4
8
 
A2 4 A2 2 6 Db2 Ds4

+ O 7 ,
Vt b =1
2
2
2
8
 2

2 D2
D
A
s
A2 ei 5 + d
+ Usd
Vcd = +
2
2


 
U
+ Ubd A sd 2 + O 7 ,
2
Vud =1

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

289



A
A
Vt d =A 1 ei 3 + ei 5 + Ds2 3 + Ubd
2
2


 
A 2
2
i 3
Db Dd 1 e
A2 Usd + O 7 ,
+
2



A
A 2
2
i
Vt s =A +
Ds Db2 2
4 + Ubs +
Ae
2
2
 7
A 6
+ +O .
(4)
8
This is a completely general parametrization of the 3 3 submatrix of a complete (3 + n)
(3 + n) unitary matrix. 1 Hence, it includes also the simplest case of a single vector-like
quark. In this last case, there are several relations among the 13 parameters of this general
matrix and only the nine independent parameters of a unitary 4 4 mixing matrix remain.
In fact, we have
Dd2 = |V4d |2 ,

V4d ,
Usd = V4s

Ds2 = |V4s |2 ,

Db2 = |V4b |2 ,

Ubs = V4b
V4s ,

Ubd = V4b
V4d .

(5)

Hence, we have 4 relations among the parameters in Eq. (3),


|Usd |2 = Ds2 Dd2 ,

|Ubs |2 = Db2 Ds2 ,

|Ubd |2 = Db2 Dd2 ,

Usd Ubs Ubd


= Db2 Ds2 Dd2 .

(6)

In the general case these equalities are replaced by inequalities,


|Usd |2  Ds2 Dd2 ,

|Ubs |2  Db2 Ds2 ,

|Ubd |2  Db2 Dd2 .

(7)

Naturally, the elements of the extended mixing matrix corresponding to CKM elements
are experimentally measured in low energy experiments. Some of these measurements
are obtained from tree-level processes, therefore not affected by new physics to a high
degree of approximation. Hence, they constrain directly the corresponding elements of our
extended mixing matrix. Specifically, we have [16,18], at 95% CL,

0.2150 < |Vus | = < 0.2242,


0.0364 < |Vcb | = A2 < 0.0440,
0.074 < |Vub /Vcb | = < 0.106,
0.9719 < |Vud | < 0.9751,
0.192 < |Vcd | < 0.256,
0.948 < |Vcs | < 1.0.

These constraints restrict the different parameters in Eq. (4), and so from the bounds on
|Vud | and |Vcs |, we obtain Dd2  5.5 103 and Ds2  5.6 102 , respectively. Another
constraint [8,9,19] comes from the SU(2)L coupling of the Z 0 to bb. In the SM, this

VCKM )bb = 1, but in this model it is modified to Ubb ; hence, we have


coupling is (VCKM
2
[19] Db  9 103 .
1 Notice that this parametrization is approximate to O(6 ) but an exact solution can be obtained numerically [9].

290

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

Notice that, in the general case, the Di2 parameters are completely independent from
the FCNC couplings U . However, in a more definite model, as for instance the single
vector-like quark model, due to the unitarity of the 4 4 matrix, these constraints have
a strong impact on all other elements of the extended mixing matrix and consequently
also on the tree level FCNC couplings, as shown in Eq. (6). In any case, even with these
restrictions, the FCNC couplings can have large effects on rare processes, where the SM
contributes at the 1 loop level and new physics is allowed to compete on equal footing. In
fact, most of the experimental results are well accommodated within a pure 3 generations
SM and hence these measurements provide additional constraints on the FCNC couplings.
In other observables, like B 0 CP asymmetries or B rare decays, the experimental results
may still differ from the SM prediction once the experimental accuracy is increased in the
near future. Therefore, we analyze these observables from a slightly different point of view,
and in the last section we show what are the possible deviations from the SM consistent
with the constraints discussed here.
In the first place we analyze the constraints associated with kaon physics and then the
constraints we get from the B system.
3.1. Kaon physics constraints
Rare kaon decays and CP-violating observables in the kaon sector can receive important
contributions from the FCNC coupling Usd . In fact, this new coupling is constrained
mainly by the decay (KL + )SD and the experimental value of  /. Other
observables constraining this coupling are K + + and K [20].
The decay KL + is CP-conserving and in addition to its short distance part,
given by Z penguins and box diagrams, receives important contributions from the
two photon intermediate state which are difficult to calculate reliably. Unfortunately,
the separation of the short-distance part (similar to K + + , free of hadronic
uncertainties) from the long-distance piece in the measured rate is rather difficult.
Therefore, the full branching ratio is generally written as a sum of a dispersive and
absorptive contributions, of which the latter can be calculated using the data for KL
. The dispersive contribution can be decomposed as a long distance and a short distance
part. Following [20,21], we can write down

 2
Br(KL )SD = 6.32 103 CU 2Z Re(Usd ) + c + Y0 (xt ) Re sd
t
 2.8 109 ,
(8)

2 / 2 )1  92.7, ab = V V ,
c = 6.54 105 is
where CU 2Z = ( 2GF MW
ia ib
i
the charm quark contribution and Y0 is the InamiLim [22] function as defined in [23].
Experimentally, we must require this branching ratio to be BR(KL + )SD  2.8
109 [21].
A second observable constraining Usd is  /. Within the SM there is a cancellation
between QCD and electroweak penguin contributions (dominated by Z penguins) which
suppresses this ratio. When new physics enters into the game and if, as it is expected on
general grounds, it does not affect considerably the QCD contributions but does so with the

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

291

Z penguins, the abovementioned cancellation does not take place and significant deviations
from the SM results (or strong bounds on the new parameters) can be expected. We now
decompose  / as follows
 

= U CU 2Z Im(Usd ) + t Im sd
t ,


(3/2) 
,
U = 1.2 Rs rZ B8

(1/2)
(3/2) 
t = U C0 2.3 + Rs 1.1 rZ B6
+ (1.0 + 0.12 rZ )B8
,

(9)

where the first term comes from the Z piece and the other one contains all the remaining
ones [20]. It is worth noting that unlike previous observables, here the theoretical errors
overwhelm the experimental precision. The main sources of uncertainties lie in the
(1/2)
(3/2)
and B8 . The importance of these uncertainties is somehow increased
parameters, B6
because of the cancellation we have mentioned.
(3/2)
= 0.8 0.2 [20]. The value
Here, we take Rs = 1.5 0.5, rZ = 7.5 1 and B8
(1/2)
has caused some controversy in the literature because of the different values
of B6
obtained in different schemes. We take two different values in order to illustrate the
situation where new physics in the sd sector is needed or not needed to reproduce the
(1/2)
= 1.0 0.2 with the other parameters
experimental value of  /. For set I, we use B6
as given above, as in Refs. [20,24,25] which comes from large Nc calculations [26] and
lattice analysis, and this tends to favor the presence of new physics in Uds . In set II we use
(1/2)
= 1.3 0.5 in order to incorporate the predictions of Refs. [27,28], where inclusion
B6
of the correction from final-state interactions in a chiral perturbation theory analysis tends
(1/2)
and
to favor the SM range. Still, there are other schemes where different values for B6
(3/2)
B8
are obtained [2931]. In fact, we have explicitly checked that with the values in
(1/2)
(3/2)
= 2.5 0.4 and B8
= 1.35 0.20, the constraints on the FCNC couplings
[29], B6
U are still consistent with the results presented below.
Numerically, we have for the central values

 

6.2
= 7.8 CU 2Z Im(Usd ) +
Im sd
t

9.9

(10)

with the parameters in set I and set II, respectively, and we calculate its errors with a
Gaussian method. This observable has to reproduce the experimentally obtained value of
( /)exp = (2.12 0.46) 103 .
A theoretically cleaner constraint [32], although less restrictive than the previous two
constraints is provided by BR(K + + ),

2
BR(K + + ) = 1.55 104 CU 2Z Im{Usd } + X0 (xt ) Im sd
t
2 

,
+ CU 2Z Re{Usd } 2.11 104 + X0 (xt ) Re sd
t

(11)

with X0 (xt ) = C0 (xt ) 4B0 (xt ) a gauge invariant loop function combination of boxes and
vertices [22,23]. This decay has already been measured in the experiment E787 at BNL
[33], however, so far, a single event has been found and this is not enough to provide a

292

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

definite value for the branching ratio. Hence we take here only the upper limit at 95% CL
BR(K + + )  8.3 1010 . 2
Finally, we include also K , whose leading-order expression is [12,23]:

ei/4 GF BK FK2 mK

K =
Im t t (Usd )2 +
6%mK
4 sin2 W

t
t


sd
sd sd
,
8
t i Y0 (xi )i Usd
ij S0 (xi , xj )i j
(12)
i,j =c

i=c

where S0 is another InamiLim function [23] and the QCD correction factors (which take
into account short distance QCD effects) are given as follows,
cc = 1.38 0.20,

t t = 0.57 0.01,

t c = 0.47 0.04.

(13)

Here, contrary to Ref. [20], the coefficient Y0 (x) of the linear term in Uds is characteristic
of the present model, therefore, in principle, the irrelevance of K to constraint Uds is not
fully guaranteed.
At this point, it is important to emphasize that Eqs. (8), (9), (11) and (12) are completely
valid in the general model with n additional VLdQs.
In Fig. 1 we present the effects of the different constraints in the Usd coupling. As
we can see in this figure, the most efficient constraints are provided by KL + ,
that constrains Re(Usd ), and  /, that bounds Im(Usd ) to the rectangular box in the
center. However, these constraints will be difficult to improve due to the large hadronic
uncertainties. Similarly, the constraints from K are very precise on the experimental side
and the precision of this limit on Usd is determined by the hadronic parameter BK =
0.85 0.15 [23,25]. Hence, it is not expected to improve largely in the near future. On
the other hand, the decay K + + is much cleaner from the theoretical point of view
[32]. The experimental value for the branching ratio, based in the single event found so
far is
10
,
BR(K + + ) = 1.5+3.4
1.2 10

(14)

which gives an upper bound of


BR(K + + )  8.3 1010 at 95% CL.

(15)

Unfortunately, it is clear from the errors that we cannot obtain a lower bound on this BR
at 80% CL, still we have included in the figure a dashed circle showing the effect of a
future lower limit which would correspond to a hypothetical value of 8 1011 and would
exclude the region to the left of the small rectangle. Moreover, it is interesting to notice
that improving the upper bound to the level of 2 1010 would provide a bound on the
same level as the combined bounds from KL + and  /. In fact, these values could
be reached after the analysis of the stored data from E-787 and the sensitivity of the new
2 In Ref. [11], the old value of this BR at 1 was used, therefore both a lower and an upper limit were
considered.

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

293

Fig. 1. Effect of the constraints from K , K + + ,


KL + and  / on the Usd FCNC
coupling.

experiment E-949 [34] (already approved) will reach 1011 in the next few years. Hence,
this decay will provide the most stringent and clean bounds on the Usd coupling in the near
future, although at present the bounds are still obtained from KL + and  /.
In Figs. 2 and 3 we present a scatter plot of the allowed values of Usd in the general
model with n additional vector-like down quarks, although we must emphasize that this
allowed region does not change at all even if we go to the more restricted case of a single
VLdQ. We impose all the constraints described above, with the  / parameters from set I
and set II in Figs. 2 and 3, respectively. In both approaches the bound for the real part is,
1.3 105 < Re{Usd } < 4.0 106 .

(16)

This constraint is directly obtained from Eq. (8) with the limit values of Re(sd
t ) which
4 and Re(sd )
4 from Eq. (4) to O(5 ) for
)
=
4.9

10
=
2.1

10
are Re(sd
max
min
t
t
= , 0 respectively. Therefore, this implies that the additional correlations among the
parameters in the 1-VLdQ model are irrelevant for the Usd bound.
The constraints on the imaginary part of this coupling depend slightly on the adopted
(1/2)
(1/2)
= 1.0 0.2, we can see that Usd is necessarily
value for B6 . For set I, with B6
positive and does not reach the origin, indicating marginally the need of new physics for
 /. The allowed range is 1.0 107 < Im{Uds } < 6.5 106 . On the other hand, in
set II, we get an allowed area, 1.5 106 < Im{Uds } < 6.0 106 , including the SM
as one of the possible points in agreement with the experimental results. This bounds are
4 and Im(sd )
5
directly obtained from Eq. (10) with Im(sd
t )max = 1.4 10
t min = 6 10

294

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

Fig. 2. Phenomenological bounds on the FCNC coupling Usd with the  / hadronic parameters in
set I.

Fig. 3. Phenomenological bounds on the FCNC coupling Usd with the  / hadronic parameters
from set II.

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

295

assuming a 50% error, which corresponds to the Gaussian errors that we use in this
constraint. In view of these two different options to calculate  /, we take a conservative
bound on Im(Usd ) including both possibilities,
1.5 106 < Im{Usd } < 6.5 106 .

(17)

It is interesting to notice, that these bounds turn out to be much more stringent than
the direct bounds on FCNC couplings usually quoted in the literature for VLdQ models
[35]. This improvement is mainly due to the inclusion of the SM contributions that
were completely neglected in the calculation of K in the bounds presented in [35], the
improvement of the experimental results in (KL + )SD , K + + , and to the
inclusion of the bound from  /. In this way, our bounds basically agree with the general
bounds in [20].
3.2. B physics constraints
In this section, we study the constraints on the Ubd and Ubs couplings from B physics
FCNC. In fact, the experimental information in the B system is improving rapidly with the
new results from B factories and so, it is important to update the bounds on these FCNC
couplings. In all the following processes, the choice of set I or set II to calculate  / has no
relevant effects on the Ubd and Ubs couplings. Hence, we analyze the constraints making no
distinction on the hadronic parameters used to calculate  /. In first place, we concentrate
on the constraints from CP conserving processes and then we add the information from
the B 0 CP asymmetries. The main CP conserving processes constraining Ubd and Ubs are
BR(B Xd,s l + l ) and %MBd,s .
From the upper bound on BR(B Xs l + l )  4.2 105 [36] and assuming BR(B
Xd l + l )  BR(B Xs l + l ) we have [23,37],
|Y0 (xt )t + CU 2Z Ubq |2
(B Xq l + l )
2
=
(B Xc e+ e ) 2 sin4 W
|Vcb |2 f (mc /mb )


1
 4 104
2 sin4 W sin2 W +
4
bq

(18)

with f (z) = (1 8z2 + 8z6 z8 24z4 log z) a phase space factor due to the mass of the
charm quark. From here we get



Y0 (xt )bq
(19)
t + CU 2Z Ubq < 0.15.
3
i ) + O(5 )  8 103 (1 0.4 e i ) with Y  1,
Replacing here bd
0
t  A (1 e
it is clear that the bd
t contribution can be safely neglected and we get a bound |Ubd | 
2
1.7 103 . However, this is not true in the case of bs
t  A  0.040 that shifts the
3
|Ubs | constraint to a circle of radius 1.7 10 centered in (0.040/92.7, 0).
 mixing we have [1012]
Additionally, from BB
B
M12q

2 B f2 m
G2F MW
Bq Bq Bq Bq

12 2

bq 2

S0 (xt )t

bq ,

296

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

d mixing in the n-VLdQ model (left) and in the


Fig. 4. Constraints on the Ubd coupling from Bd B
1-VLdQ model (right). The circle shows the bound obtained from the B Xd l + l .

bq = 1 3.3

Ubq
bq

+ 165

Ubq
bq

2

(20)

where the new parameters are defined in Ref. [23], and the mass difference %MBq =
 B 
2M q . The experimental values for these observables are %MB = (0.472 0.017)
12

1012 s1 and %MBs > 14.9 1012 s1 at 95% CL [38,39].


At this point, it is important to compare the bounds that can be obtained from the mass
difference and the semileptonic decay. From BR(B Xd,s l + l ), we get a bound on
bq
|t Y0 + CU 2Z Ubq |2 in Eq. (19), while from the mass difference, neglecting the linear
bq
2 |. Hence, it is
term, we obtain a constraint on the combination |(t )2 S0 + 4CU 2Z Ubq
clear that the relative size of the tree level FCNC with respect to the SM contribution is
always bigger in the semileptonic decay due to the presence of an additional factor CU 2Z 
92.7. Nevertheless, the experimental constraints from BR(B Xd,s l + l ) are still much
larger than the typical SM prediction while the %MBd measurement is already saturated
by the SM contribution. This implies that the %MBd constraint is more effective in the
case of Ubd and both experiments give rise to constraints of the same order of magnitude,
dominating at the end the %MBd bound. This can be seen in Fig. 4, where we show the
allowed region of the parameter space both in the general model with an arbitrary number
of VLdQs and in the minimal model with a single VLdQ. In these figures, the constraint
from BR(B Xd l + l ) is shown as a circle slightly shifted from the origin and a radius of
1.7 103 . However, as we can see here, the bounds on Ubd are | Re{Ubd }|  0.7 103
and | Im{Ubd }|  1.2 103, which are directly obtained from the %MBd constraint as the
2
2
bd
envelope of the curves |(bd
t ) S0 + 4CU 2Z Ubd | with all the allowed values of t . Notice
that already in this most constrained model, i.e., the model with a single VLdQ, the Ubd
coupling is only bounded by these processes and therefore we obtain the same bounds both
in this case and in the general case with n VLdQ.
In the case of the Ubs coupling the situation is completely different. In first place %MBs
has not been measured yet, and so only a lower bound on the mass difference is available

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

297

Fig. 5. Constraints on the Ubs coupling from B Xs l + l in the n-VLdQ model (left) and in the
1-VLdQ model (right).

which is not useful to set a constraint on Ubs . Moreover, it is easy to see from Eq. (20)
that for similar values of Ubd and Ubs , the FCNC effects on %MBs will be suppressed
bs 2
2
by a factor (bd
t /t )  when compared with the effects on %MBd . This implies that
s mixing. 3 Hence,
in this model we cannot expect to observe the FCNC effects in Bs B
this coupling is only constrained by the upper bound on BR(B Xs l + l ). In Fig. 5 we
show the constraint from BR(B Xs l + l ) as a circle slightly shifted from the origin,
that implies an upper bound of |Ubs |  2 103 [37,41] both in the general model with
n VLdQs and in the model with a single VLdQ. Similarly, the b s branching ratio
provides constraints of the same order of magnitude [42].
Despite these strong constraints from CP conserving processes, the CP asymmetries in
B decays are still very effective to constrain the Ubd coupling [43,44]. Recently, the arrival
of the first measurements of the B J /KS CP asymmetry, aJ / , from the B factories
has caused a great excitement in the high energy physics community.

0.34 0.20 0.05 (BaBar [45]),

+0.32+0.09
(Belle [46]),
aJ / = 0.580.340.10
(21)

+0.41
0.790.44
(CDF [47]).
These values correspond to a world average of aJ / = 0.51 0.18, that can be
compared with the SM expectations of 0.59  aJSM
/ = sin(2)  0.82 with =
/(V V )). The errors are still too large to draw any firm conclusion. In fact,
arg(Vcd Vcb
td tb
if we take the world average at 95% CL we do not get any improvement over the constraints
from CP conserving processes. However, anticipating the improvement of the experimental
errors from B factories, we take the world average at 1 level to see the effects on the Ubd
coupling. From Eq. (20), it is straightforward to obtain aJ / as,


aJ / = sin 2 arg(bd ) .
(22)
3 This is not always true in the presence of an up vector-like quark, as can be seen in Ref. [40].

298

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

Fig. 6. Allowed region for the Ubd coupling requiring a value of aJ / asymmetry to reproduce the
world average at 1 in the n-VLdQ model (left) and 1-VLdQ model (right).

Using the world average at 1 as the experimentally allowed range, we show in Fig. 6
the resulting region for the Ubd coupling. As we can see here, the Ubd allowed region is
sizeably modified from this constraint. The outer regions in the second and fourth quadrants
are reduced while the central region corresponding to the SM remains filled; this situation
represents an improvement over the analysis presented in Ref. [44]. Note that the results
are essentially similar for the n-VLdQ and the 1-VLdQ cases. As we can see here, with
the world average for the aJ / CP asymmetry there is no need of new physics in the Ubd
sector.

4. Discovery potential in B factories


As we have explained above, the present results in the aJ / asymmetry in Eq. (21) are
not precise enough to make a definitive statement about the presence/absence of large
new physics contributions in B decays. Still, these measurements, and especially the
BaBar value which is the most precise one, leave room for an asymmetry considerably
smaller than the standard model expectations. The SM range is certainly outside the
1 BaBar range but not outside the world average. This potential discrepancy is at the
origin of several papers [48,49] studying the implications of a small aJ / in the search
of new physics. The papers in Ref. [48] provide general parametrizations of new physics
contributions to B mixing and decays. They essentially show that this possible mismatch
among the measured value and the SM expectations may be due to the presence of new
physics either directly in B physics or in K physics modifying indirectly the unitarity
triangle fit. However, these papers do not provide a definite new physics example fulfilling
this task. On the other hand, the two papers in Ref. [49] analyze SUSY models, where no
sizeable effects in B mixing and decays are generally expected and the unitarity triangle fit
can be modified only through new SUSY contributions to K physics. In the following, we
show that a model with tree level FCNC from the mixing with vector-like quarks would

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

299

Fig. 7. Distribution of the values of aJ / for 95 000 events in the n-VLdQ (left) and in the 1-VLdQ
(right).

be a natural candidate for a model with clear deviations from the SM expectations in CP
asymmetries, satisfying simultaneously all other experimental constraints.
3
2
From the expression of aJ / in Eqs. (20) and (22), with |bd
t | [5 10 , 1.2 10 ]
3
and |Ubd |  10 , it is clear that arg(bd ) can be very large. Indeed, in the general model
with an arbitrary number of VLdQ, there is no further influence on Ubd from constraints
on other elements of the mixing matrix. In fact, only the inequalities in Eq. (7) can restrict
this coupling, however, given the bounds on Di2 , they are almost inoperative in this case.
Hence, all the points inside the allowed contour in Fig. 4 are equally probable and any value
of the asymmetry is possible with a comparable probability. Still, the especial geometry
of this allowed region and the preferred orientation of tbd slightly favors the SM range
over other values. Furthermore, even in a more constrained model, as in the model with
a single VLdQ large departures from the SM range are possible. The expected values
for the asymmetry both in the n VLdQs and in the 1-VLdQ model are shown in Fig. 7
as an histogram of the distribution of 95 000 events satisfying all the constraints in the
previous sections. As expected from the above discussion, in the general case all possible
values of the asymmetry have similar probability. On the other hand, in the model with a
single VLdQ, this distribution is clearly peaked on the SM range. Moderate deviations are
rather frequent and larger departures are more rare but still possible for any value of the
asymmetry.
In Figs. 8 and 9 we show, in the 1-VLdQ model, the correlation of the possible values
of the asymmetry with |Ubd | and |Ubs |, respectively. Here, we see that the range 0.55 
aJ /  0.85 corresponding to the SM expected range concentrates most of the events and
can be reproduced with |Ubd | = 0 and any allowed value of |Ubs |. However, for different
values of the asymmetry there is a clear correlation between aJ / and the minimum value
of |Ubd | required to obtain this asymmetry. For instance, to obtain an asymmetry below
0.5, a |Ubd |  2 104 is needed. This required minimum is also true in the general
model with an arbitrary number of VLdQs. Similarly, we see in Fig. 9 that these large
values of Ubd correspond to low values of Ubs [11], however it is important to emphasize

300

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

Fig. 8. Correlation between the values of aJ / and |Ubd | in the model with a single VLdQ.

Fig. 9. Correlation between the values of aJ / and |Ubs | in the model with a single VLdQ.

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

301

Fig. 10. Allowed region for the Ubd coupling consistent with the BaBar result for the aJ /
asymmetry at 1 in the n-VLdQ (left) and in the 1-VLdQ (right).

Fig. 11. Allowed region for the Ubs coupling consistent with the BaBar result for the aJ /
asymmetry at 1 in the n-VLdQ (left) and in the 1-VLdQ (right).

that this correlation is only true in the minimal model, but not in a model with an arbitrary
number of VLdQs. To see this we show in Figs. 10 and 11 both in the n-VLdQ and
in the 1-VLdQ models an scatter plot of Re(Ubd ) versus Im(Ubd ) and Re(Ubs ) versus
Im(Ubs ) for an asymmetry corresponding to the BaBar result at 1 , 0.14  aJ /  0.54
[45]. In this Fig. 10 we see that the great majority of the allowed points are in the range
1 104 (2 104 )  |Ubd |  1.2 103 , in the n-VLdQ (1-VLdQ) model, i.e., a large,
non-vanishing Ubd coupling is required to reproduce the BaBar asymmetry. In particular,
this means that, within this model, a low CP asymmetry implies the presence of new
physics in bd transitions. This conclusion would be unchanged in the general model.
On the other hand, we see that, for these points, in the 1-VLdQ model, the coupling Ubs
is always restricted to the range |Ubs |  3 104 ; hence, all the allowed points have
simultaneously high |Ubd | and low |Ubs |. Indeed, it is easy to obtain in the 1-VLdQ model,

302

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

= U D 2 . The region in the U plane does not change


from Eq. (5), the relation Ubd Ubs
sd b
ds
with the inclusion of the aJ / constraint, and then we still have, |Usd |  6 106 and
Db2  0.009. Taking into account that a low aJ / requires |Ubd |  2 104 , this clearly
implies an absolute upper bound, |Ubs |  3 104 . Therefore, for this set of points, we
can not expect a new-physics contribution in the b s transition. However, this relation
is not valid in the n-VLdQ model and therefore this correlation is lost as can be seen in
Fig. 11.
Also in Fig. 10, we find a few points ( 0.1% of the points) which have simultaneously
|Ubs |  1 103 and |Ubd |  3 105 . This second class of points is only possible in
the vicinity of the SM and they disappear if the value of the asymmetry is reduced to
aJ /  0.52. 4
Therefore, we can conclude that in a model with a single VLdQ a low CP asymmetry
aJ /  0.52 implies |Ubd |  2 104 and simultaneously |Ubs |  3 104 . On the
contrary, in a general model with an arbitrary number of VLdQs, we still have |Ubd | 
2 104 , but this has no influence on the Ubs coupling, and it is only constrained by the
B Xs l l branching ratio.
At this point, it is very interesting to examine the predicted branching ratios of the
decays B Xd,s l l for this set of points. From Fig. 10, where we have included the circle
corresponding to the experimental bounds in these decays, it is clear that we can also expect
In this case, the branching ratios for the Xd decays
a very large contribution to B Xd l l.
are strongly enhanced from the SM prediction, reaching values of 1.0 106  BR(B
Xd l + l )  1.8 105 and 6.0 105  BR(B Xd )  1.0 104 . While, on the
other hand, the low values of Ubs imply that the Xs decays remain roughly at the SM value.
Conversely, in the points with small Ubd and large Ubs , there is no sizeable departure from
the SM expectations in B Xd l l and the Xs decays are now close to the experimental
upper range. Namely, we obtain, for the points to the right of Fig. 11, with Re(Ubs ) 
1 104 , BR(B Xs l + l )  2.7 105 to be compared with the experimental upper
bounds of BR(B Xs l + l )  4.2 105 . However, this possibility is only marginal
in the 1 BaBar range for the minimal model with a single VLdQ. In any case in the
general model, the possibility of large Ubs couplings is again open. For analysis of the
s mixing and
phenomenological effects of this coupling in radiative b s decays, Bs B
rare decays in a slightly different context, see Refs. [4042].

5. Conclusions
In this work we have updated the constraints on tree level FCNC couplings in the
framework of a theory with n isosinglet vector-like down-quarks. The inclusion of the
constraints from (KL + )SD ,  /, K and K has allowed to improve
sizeably the bound on the Usd coupling. The precise range of allowed values for this
coupling depends strongly on the hadronic input in the calculation of  /. Our summary
4 Still, it is important to emphasize that these points also require the presence of new physics in B decays.

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

303

Table 1
Constraints on the tree level FCNC couplings from rare
processes in the K and B systems
Re{Usd }
Im{Usd }
| Re{Ubd }|
| Im{Ubd }|
|Ubs |

 1.3 105
 4.0 106
 1.5 106
 6.5 106
4
 7 10
 1.2 103
 2 103

in Table 1 includes the main theoretical approaches [20,2431]. In the near future, K +
+ will be quite useful to further constrain Usd without large theoretical uncertainties.
We have calculated the constraints on Ubd and Ubs from %MBd , B Xs,d l l and the
B J /KS CP asymmetry. In Table 1, we summarize our results on these bounds
without the additional constraint from the B J /KS CP. Note that all these bounds
on Ubq are independent of the hadronic inputs in  /. In addition, we have shown that
the B J /KS CP asymmetry is especially sensitive to the presence of a Ubd FCNC
coupling. In this regard, we have shown that the 1 value of the world average is already
able to exclude half of the allowed region for the Ubd coupling. However, due to the still
large experimental errors this constraint has a small effect at 95% CL. Assuming a low
value of the aJ / asymmetry below 0.52 (following BaBar range at 1 ) we have shown
that, in models with n VLdQs, tree-level FCNC in the bd sector are mandatory. The
FCNC coupling Ubd is required to be greater than 1 104 , rising to 2 104 in the
1-VLdQ model. The Ubs coupling in the general model is bounded by 2 103 , although
for the 1-VLdQ model the bound goes down to 3 104 .
Therefore, a clear favorable scenario for these models would be to find simultaneously
at least one order of magnitude
a low aJ / (below 0.5) and a large BR(Bd Xd l l),
close to the present
bigger than the SM expectations. Values of BR(Bd Xs l l)
experimental bounds are still possible in the n-VLdQ model. Nevertheless, in the 1-VLdQ
model these BR are expected to be similar to the SM values. In summary, this aJ / CP
asymmetry, together with the rare decays B Xs,d l l and K are the best options to
further constraint the FCNC tree level couplings or to discover the presence of vector-like
quarks in the low energy spectrum as suggested by GUT theories or models of large extra
dimensions at the TeV scale.

Acknowledgements
The work of F.B. and O.V. was partially supported by the Spanish CICYT AEN-99/0692.
O.V. acknowledges financial support from a Marie Curie EC grant (HPMF-CT-200000457).

304

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

References
[1] S.L. Glashow, J. Iliopoulos, L. Maiani, Phys. Rev. D 2 (1970) 1285.
[2] M.K. Gaillard, B.W. Lee, Phys. Rev. D 10 (1974) 897.
[3] J. Ellis, D.V. Nanopoulos, Phys. Lett. B 110 (1982) 44;
R. Barbieri, R. Gatto, Phys. Lett. B 110 (1982) 211;
S. Bertolini, F. Borzumati, A. Masiero, G. Ridolfi, Nucl. Phys. B 353 (1991) 591;
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hepph/9604387;
A. Masiero, O. Vives, hep-ph/0104027, and references therein.
[4] M. Bando, T. Kugo, Prog. Theor. Phys. 101 (1999) 1313, hep-ph/9902204;
M. Bando, T. Kugo, K. Yoshioka, Prog. Theor. Phys. 104 (2000) 211, hep-ph/0003220;
M. Obara, G. Cho, M. Nagashima, A. Sugamoto, hep-ph/0105287.
[5] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[6] For phenomenology of FCNC in models with large extra dimension, see:
A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030, hep-ph/9911252;
F. del Aguila, J. Santiago, Phys. Lett. B 493 (2000) 175, hep-ph/0008143;
S.J. Huber, Q. Shafi, Phys. Lett. B 498 (2001) 256, hep-ph/0010195;
N. Rius, V. Sanz, hep-ph/0103086;
H. Davoudiasl, T.G. Rizzo, hep-ph/0104199;
S.J. Huber, Q. Shafi, hep-ph/0104293.
[7] G. Barenboim, F. J. Botella, O. Vives, work in progress.
[8] F. del Aguila, J. Cortes, Phys. Lett. B 156 (1985) 243;
G.C. Branco, L. Lavoura, Nucl. Phys. B 278 (1986) 738;
F. del Aguila, M.K. Chase, J. Cortes, Nucl. Phys. B 271 (1986) 61;
Y. Nir, D.J. Silverman, Phys. Rev. D 42 (1990) 1477;
D. Silverman, Phys. Rev. D 45 (1992) 1800;
G.C. Branco, T. Morozumi, P.A. Parada, M.N. Rebelo, Phys. Rev. D 48 (1993) 1167;
W. Choong, D. Silverman, Phys. Rev. D 49 (1994) 2322;
V. Barger, M.S. Berger, R.J. Phillips, Phys. Rev. D 52 (1995) 1663, hep-ph/9503204;
D. Silverman, Int. J. Mod. Phys. A 11 (1996) 2253, hep-ph/9504387;
M. Gronau, D. London, Phys. Rev. D 55 (1997) 2845, hep-ph/9608430;
F. del Aguila, J.A. Aguilar-Saavedra, G.C. Branco, Nucl. Phys. B 510 (1998) 39, hepph/9703410.
[9] L. Lavoura, J.P. Silva, Phys. Rev. D 47 (1993) 1117.
[10] G. Barenboim, F.J. Botella, G.C. Branco, O. Vives, Phys. Lett. B 422 (1998) 277, hepph/9709369.
[11] G. Barenboim, F.J. Botella, O. Vives, Phys. Rev. D 64 (2001) 015007, hep-ph/0012197.
[12] G. Barenboim, F.J. Botella, Phys. Lett. B 433 (1998) 385, hep-ph/9708209.
[13] J. Roldan, F.J. Botella, J. Vidal, Phys. Lett. B 283 (1992) 389.
[14] F.J. Botella, L. Chau, Phys. Lett. B 168 (1986) 97.
[15] L. Chau, W. Keung, Phys. Rev. Lett. 53 (1984) 1802.
[16] D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
[17] L. Wolfenstein, Phys. Rev. Lett. 51 (1983) 1945.
[18] R. Barate et al., ALEPH Collaboration, Phys. Lett. B 465 (1999) 349;
P. Abreu et al., DELPHI Collaboration, Phys. Lett. B 439 (1998) 209;
M. Bargiotti et al., Riv. Nuovo Cimento 23 (3) (2000) 1, hep-ph/0001293.
[19] F. del Aguila, J.A. Aguilar-Saavedra, R. Miquel, Phys. Rev. Lett. 82 (1999) 1628, hepph/9808400.
[20] A.J. Buras, L. Silvestrini, Nucl. Phys. B 546 (1999) 299, hep-ph/9811471.

G. Barenboim et al. / Nuclear Physics B 613 (2001) 285305

305

[21] D. Gomez Dumm, A. Pich, Phys. Rev. Lett. 80 (1998) 4633, hep-ph/9801298;
G. DAmbrosio, G. Isidori, J. Portoles, Phys. Lett. B 423 (1998) 385, hep-ph/9708326.
[22] T. Inami, C.S. Lim, Prog. Theor. Phys. 65 (1981) 297.
[23] A.J. Buras, R. Fleischer, Report No. TUM-HEP-275-97, hep-ph/9704376.
[24] M. Ciuchini, E. Franco, L. Giusti, V. Lubicz, G. Martinelli, Report No. ROME-99-1268, hepph/9910237;
M. Ciuchini, G. Martinelli, Nucl. Phys. Proc. Suppl. 99 (2001) 27, hep-ph/0006056;
S. Bosch, A.J. Buras, M. Gorbahn, S. Jager, M. Jamin, M.E. Lautenbacher, L. Silvestrini, Nucl.
Phys. B 565 (2000) 3, hep-ph/9904408.
[25] A.J. Buras, hep-ph/0101336.
[26] T. Hambye, G.O. Kohler, E.A. Paschos, P.H. Soldan, W.A. Bardeen, Phys. Rev. D 58 (1998)
014017, hep-ph/9802300;
T. Hambye, G.O. Kohler, P.H. Soldan, Eur. Phys. J. C 10 (1999) 271, hep-ph/9902334.
[27] S. Bertolini, J.O. Eeg, M. Fabbrichesi, E.I. Lashin, Nucl. Phys. B 514 (1998) 93, hepph/9706260;
S. Bertolini, M. Fabbrichesi, J.O. Eeg, Rev. Mod. Phys. 72 (2000) 65, hep-ph/9802405.
[28] E. Pallante, A. Pich, Phys. Rev. Lett. 84 (2000) 2568, hep-ph/9911233;
E. Pallante, A. Pich, I. Scimemi, hep-ph/0105011.
[29] J. Bijnens, J. Prades, JHEP 0006 (2000) 035, hep-ph/0005189;
J. Bijnens, J. Prades, JHEP 0001 (2000) 002, hep-ph/9911392;
J. Bijnens, J. Prades, Nucl. Phys. Proc. Suppl. 96 (2001) 354, hep-ph/0010008.
[30] S. Narison, Nucl. Phys. B 593 (2001) 3, hep-ph/0004247.
[31] J.F. Donoghue, E. Golowich, Phys. Lett. B 478 (2000) 172, hep-ph/9911309.
[32] G. Buchalla, A.J. Buras, Phys. Rev. D 54 (1996) 6782, hep-ph/9607447.
[33] S. Adler et al., E787 Collaboration, Phys. Rev. Lett. 79 (1997) 2204, hep-ex/9708031;
S. Adler et al., E787 Collaboration, Phys. Rev. Lett. 84 (2000) 3768, hep-ex/0002015.
[34] T.K. Komatsubara, E787 Collaboration, hep-ex/9905014.
[35] Y. Nir, D.J. Silverman, Phys. Rev. D 42 (1990) 1477;
Y. Grossman, Y. Nir, R. Rattazzi, hep-ph/9701231.
[36] S. Glenn et al., CLEO Collaboration, Phys. Rev. Lett. 80 (1998) 2289, hep-ex/9710003.
[37] G. Buchalla, G. Hiller, G. Isidori, Phys. Rev. D 63 (2001) 014015, hep-ph/0006136.
[38] A. Stocchi, hep-ph/0010222.
[39] A. Ali, D. London, Eur. Phys. J. C 18 (2001) 665, hep-ph/0012155.
[40] M. Aoki, M. Nagashima, N. Oshimo, hep-ph/0104063;
M. Aoki, G. Cho, M. Nagashima, N. Oshimo, hep-ph/0102165.
[41] M.R. Ahmady, M. Nagashima, A. Sugamoto, hep-ph/0105049.
[42] C.V. Chang, D. Chang, W. Keung, Phys. Rev. D 61 (2000) 053007;
M. Aoki, E. Asakawa, M. Nagashima, N. Oshimo, A. Sugamoto, Phys. Lett. B 487 (2000) 321,
hep-ph/0005133.
[43] G. Barenboim, G. Eyal, Y. Nir, Phys. Rev. Lett. 83 (1999) 4486, hep-ph/9905397.
[44] G. Eyal, Y. Nir, JHEP 9909 (1999) 013, hep-ph/9908296.
[45] B. Aubert et al., BaBar Collaboration, Report No. SLAC-PUB-8540, hep-ex/0008048.
[46] H. Aihara, Belle Collaboration, Belle note 353, hep-ex/0010008.
[47] T. Affolder et al., CDF Collaboration, Phys. Rev. D 61 (2000) 072005, hep-ex/9909003.
[48] J.P. Silva, L. Wolfenstein, Phys. Rev. D 63 (2001) 056001, hep-ph/0008004;
G. Eyal, Y. Nir, G. Perez, JHEP 0008 (2000) 028, hep-ph/0008009;
Z. Xing, hep-ph/0008018;
A.J. Buras, R. Buras, Phys. Lett. B 501 (2001) 223, hep-ph/0008273.
[49] A. Masiero, O. Vives, Phys. Rev. Lett. 86 (2001) 26, hep-ph/0007320;
A. Masiero, M. Piai, O. Vives, Report No. FTUV-12-08, hep-ph/0012096.

Nuclear Physics B 613 (2001) 306329


www.elsevier.com/locate/npe

Self-tuning solution of the cosmological constant


problem with antisymmetric tensor field
Jihn E. Kim, Bumseok Kyae, Hyun Min Lee
Department of Physics and Center for Theoretical Physics, Seoul National University,
Seoul 151-747, South Korea
Received 5 January 2001; accepted 13 August 2001

Abstract
We present a self-tuning solution of the cosmological constant problem with one extra dimension
which is curved with a warp factor. To separate out the extra dimension and to have a self-tuning
solution, a three index antisymmetric tensor field is introduced with the 1/H 2 term in the Lagrangian.
The standard model fields are located at the y = 0 brane. The existence [1] of the self-tuning solution
(which results without any fine tuning among parameters in the Lagrangian) is crucial to obtain
a vanishing cosmological constant in a 4D effective theory. The de-Sitter and anti-de-Sitter space
solutions are possible. The de-Sitter space solutions have horizons. Restricting to the spaces which
contain the y = 0 brane, the vanishing cosmological constant is chosen in the most probable universe.
For this interpretation to be valid, the existence of the self-tuning solution is crucial in view of the
phase transitions. In this paper, we show explicitly a solution in case the brane tension shifts from one
to another value. We also discuss the case with the H 2 term which leads to one-fine-tuning solutions
at most. 2001 Published by Elsevier Science B.V.
PACS: 04.50.+H; 11.25.Mj; 98.80.Es
Keywords: Cosmological constant; Self-tuning; Four-form field; Brane

1. Introduction
The cosmological constant problem [2] is probably the most important clue to the
physics at the Planck scale. Most attempts toward solutions of the cosmological constant
problem introduce additional ingredients [38], except the wormhole and anthropic
solutions of the problem [6,9].
The wormhole solution is based on the probabilistic interpretation that the probability
to have a universe with a vanishing cosmological constant is the largest [6]. But in
the evolving universe where the true vacuum is chosen as the universe cools down,
E-mail address: jekim@phyp.snu.ac.kr (J.E. Kim).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 9 7 - 2

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

307

the probabilistic interpretation in the early universe is in question since at a later


epoch additional constant may be generated by spontaneous symmetry breaking. For this
interpretation to make sense, there must exists a self-tuning solution. (The self-tuning
solution is defined as the flat space solution without any fine-tuning of parameters in the
action.) The anthropic interpretation [9] for the small cosmological constant is a working
proposal, but does not answer the fundamental question, can we explain the vanishingly
small constant from the fundamental parameters in the theory?
On the other hand, Hawkings probably vanishing cosmological constant [5] relies on
an undetermined integration constant. He showed that the wave function for a flat universe
is infinitely large compared to nonflat universes. Thus, if there exists an undetermined
integration constant, the phase transition will end probably to a flat universe. However,
his original proposal with a three index antisymmetric tensor field does not introduce any
dynamics in the 4D spacetime and the integration constant is just another cosmological
constant [5,8]. Witten [3] also argued that the existence of an undetermined integration
constant may be a clue to the understanding of the flat universe, since a many-body theory
may provide an explanation from a one constant universe to another constant universe [10].
Thus, in the absence of a self-tuning solution of the cosmological constant, we regard that
the cosmological constant problem in Hawkings scenario still remains as an unsolved
problem.
Therefore, it is worthwhile to search for any new solution of the cosmological constant
problem. Since the gravitational law beyond the Planck scale is not a settled issue at
present, we can look for solutions even in models with drastically different gravity beyond
the Planck scale. If the cosmological constant problem is solved in models with a different
gravity, then the new interaction can be studied extensively whether it leads to another
inconsistency in theory or in phenomenology. In this spirit, there have been attempts
to understand the cosmological constant in the 5-dimensional (5D) world with the fifth
dimension y compactified [11] (RSI model) or uncompactified [12] (RSII model). It seems
that there can be a way to understand the cosmological constant problem [13] in these
RS type models since, RSI model for example, the nonvanishing brane tensions k1 and
k2 and bulk cosmological constant k can lead to a flat space solution if these parameters
satisfy two fine-tuning conditions k1 = k = k2 . Therefore, the first step to understand the
cosmological constant is to find the flat space solutions without any fine-tuning between
parameters in the action.
With the fifth dimension compactified (RSI), the attempts to solve the cosmological
constant has failed so far. The original try to find a flat space solution without any finetuning [14] had a singularity, and taking the singularity into account by putting a brane
there reproduced a fine-tuning condition [15] even though the two conditions have been
reduced to just one. Therefore, the first step toward a solution of the cosmological constant
in the RS type models is to have flat space solutions without any fine-tuning between
parameters in the Lagrangian. Restricting just to an exponentially small cosmological
constant, it has been pointed out that it is possible with two or more branes [1618].
Recently, the needed flat space solutions in the RSII type background have been
found [1]. With a smeared-out brane a similar attempt led to a self-tuning solution [19]. In

308

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

the RS type models, the brane(s) has a special meaning in the sense that the matter fields
can reside at the brane only. However, the effective gravitational interaction of the matter
fields is obtained after integrating out the fifth dimension y. For this effective theory to
make sense, we must require that:
(i) the metric is well-behaved in the whole region of the bulk, and
(ii) the resulting 4D effective Planck mass is finite.
The condition (i) is to find a solution without a singularity in the region defined. In this
regard, if the warp factor vanishes at say y = ym , then ym becomes the horizon and the
universe connected to the matter brane is up to y = ym . In the RSI models, it is a disaster
if 0 < ym < 1/2 is between the two branes located at y = 0 and y = 1/2. This is because
one needs both branes for the consistency in the RSI models. But in the RSII models,
if there exists ym then one can consider the space only for ym < y < ym if ym is not
a naked singularity. Indeed, it can be shown that it is consistent to consider the space
up to this point only by calculating the effective cosmological constant by integrating
y to ym . The localized gravity condition (ii) restricts the solutions severely, since the y
integration in some solutions would give a divergent quantity for MP or MP get more
important contributions as y .
The EinsteinHilbert action with a bulk cosmological constant and a brane tension in the
RSII model does not allow a self-tuning solution. Addition of the GaussBonnet term [21]
does not improve the situation, and also it does not help to regularize a naked singularity in
the self-tuning model with a bulk scalar [14]. But addition of the three index antisymmetric
tensor field AMNP (and the field strength HMNP Q and H 2 HMNP Q H MNP Q ) allows a
self-tuning solution [1].
In this paper, we discuss the RSII model with antisymmetric tensor field added. The
case with 1/H 2 term allowing the self-tuning solution is the main motivation for this
extensive study. HMNP Q has been considered before in connection with the cosmological
constant problem [3] and the possible compactification of the seven internal space in
the 11D supergravity [22]. Even in the 5D RandallSundrum model HMNP Q is useful
to separate the 4-dimensional space. The 1/H 2 term looks strange, but the consideration
of the energymomentum tensor would require a nonvanishing HMNP Q , triggering the
separation of the extra dimension from the 4-dimensional space. Below the Planck scale,
we consider that the action is an effective theory. Above the Planck scale, we consider
that the quantum gravity effects may be very important, but at present the final form for
quantum gravity is not known yet.
In this paper, the self-tuning solution means that it does not need a fine-tuning between
the parameters in the action, which is a progress toward understanding the cosmological
constant problem. The self-tuning solutions are found from time independent Einstein
equations. However, the existence of the self-tuning solution alone does not solve the
cosmological constant problem completely. It is because if the ansatz for a time dependent
metric allows, for example, the de-Sitter space solutions with the antisymmetric tensor field
added then choosing the flat space is simply choosing a boundary condition. However,
the existence of the self-tuning solution and the probabilistic interpretation for the wave

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

309

function of the universe can provide a logical understanding of the vanishing cosmological
constant even in this case [5]. Suppose that we start from a flat universe from the
beginning a la the wormhole interpretation of the vanishing cosmological constant. But
this interpretation alone may encounter a difficulty when the phase transitions such as the
electroweak phase transition or the QCD phase transition add a nonvanishing cosmological
constant at a later epoch. However, the existence of the self-tuning solution chooses the flat
space solution out of numerous possibilities when the universe goes through these phase
transitions. If there exists a self-tuning solution, then the wormhole interpretation chooses
the vanishing cosmological constant even after these phase transitions.
We find that the 1/H 2 term does always allow de-Sitter space solutions with localized
gravity. In the RS model, however, a negative brane tension (1 < 0) does not allow
a localized gravity in the de-Sitter space [23], but allows only a nonlocalized gravity.
There are arguments excluding these nonlocalized gravity [24], and in the RSII model
for a negative brane tension one may exclude the de-Sitter space solutions. However, the
nonlocalized gravity cannot become a strong argument for the vanishing cosmological
constant. It simply means that the de-Sitter space solution with nonlocalized gravity
cannot materialize to our universe. At some point, we may invoke an anthropic principle
or turn to a probabilistic interpretation. Namely, as long as there exist solutions for
nonzero cosmological constants whether there results a localized gravity or not, we need
a probabilistic interpretation. For a probabilistic interpretation, the existence of the selftuning solution is crucial in choosing the flat universe.
In Section 2, we present the flat space solutions with 1/H 2 term and with H 2 term.
It is shown that 1/H 2 term allows the flat space self-tuning solution but H 2 term allows
at most one-fine-tuning solutions. De-Sitter space and anti-de-Sitter space solutions are
also commented. In Section 3, it is shown that the 1/H 2 term allows the anti-de-Sitter
and de-Sitter space solutions. We discuss the horizons appearing in our solutions. We also
discuss how the universe chooses the vanishing cosmological constant. In Section 4, we
present a time-dependent solution such that the 4D spacetime remains flat when the brane
tension shift instantaneously to another value. Section 5 is a conclusion.

2. The static solutions


The five-dimensional space is composed of the bulk and a 3-brane located at y = 0
where y is the fifth coordinate. We assume that matter fields live in the brane. For studying
the gravity sector, we include the three index antisymmetric tensor field AMNP whose
field strength is denoted as HMNP Q , where M, N, . . . = 0, 1, 2, 3, 5( y). We find that
there exist solutions for different bulk cosmological constants at y < 0 and y > 0. But for
simplicity of the discussion, we will introduce a Z2 symmetry so that the bulk cosmological
constant is universal. In this section, we summarize the self-tuning solution [1] with the
time independent metric. But for comparison we briefly comment the time dependent
metric, i.e., the de-Sitter space and anti-de-Sitter space solutions with the 1/H 2 term.

310

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

We also present one-fine-tuning solutions for H 2 term with time independent metric, and
compare with the other known tuning solutions [14,19].
2.1. A self-tuning solution of the cosmological constant with 1/H 2
A self-tuning solution exists for the following action,





2 4!
1
4
b + Lm (y) ,
S = d x dy g R +
2
HMNP Q H MNP Q

(1)

which will be called the KKL model [1]. Here we set the fundamental mass parameter M
as 1 and we will recover the mass M wherever it is explicitly needed. We assume a Z2
symmetry of the warp factor solution, (y) = (y). The sign of the 1/H 2 term is chosen
such that at the vacuum the propagating field AMNP has a standard kinetic energy term.
The action contains the 1/H 2 term which does not make sense if H 2 does not develop
a vacuum expectation value. Since the cosmological constant problem is at the bottom
of most cosmological application of particle dynamics, it is worthwhile to study any
solution to the cosmological constant problem. We note that this problem has led to so
many interesting but unfamiliar ideas [35,7]. Therefore, any new idea in the possible
interpretation of the cosmological constant problem is acceptable at this stage. In fact,
we found a very nice solution with the above action and hence we propose the action
(1) as the fundamental one in gravity. Being a part of gravity, we do not worry about the
renormalizability at this stage.
Flat space solution. The ansatz for the metric is taken as
ds 2 = 2 (y) dx dx + dy 2,

(2)

where ( ) = diag(1, +1, +1, +1). Then Einstein tensors are


 
  2

+3
G = g 3
,


G55 = 6

2
,

(3)

where prime denotes differentiation with respect to y. With the brane tension 1 at the
y = 0 brane and the bulk cosmological constant b , the energy momentum tensors are

TMN = gMN b g M N
1 (y)


4
1
1
P QR
+ 4 4!
g
.
H
H
+
MP QR N
MN
2
H4
H2

(4)

The specific form for H 2 HMNP Q H MNP Q in Eq. (1) makes sense only if H 2
develops a vacuum expectation value at the order of the fundamental mass scale. Because
of the gauge invariant four index HMNP Q , four spacetime is singled out from the five
dimensions [22]. The four form field is denoted as H ,
H =

$
,
g
n(y)

(5)

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

311

where , . . . run over the Minkowski indices 0, 1, 2, and 3. With the above ansatz, the field
equation for the four-form field is satisfied,



H MNP Q
M
(6)
g
= 0.
H4
There exists a solution for b < 0. The two relevant Einstein equations are the (55) and
() components,
 2

8
,
6
(7)
= b

A
 2
 

8
3
(8)
= b 1 (y) 3 ,
+3

A
where A is a positive constant in view of Eq. (5). It is easy to check that Eq. (8) in the
bulk is obtained from Eq. (7) for any b , 1 , and A. This property is of the specialty
of the HMNP Q field. Near B1 (the y = 0 brane), the function must be generated by
the second derivative of . The Z2 symmetry, (y) = (y), implies (d/dy)(y)|0+ =
(d/dy)(y)|0 . Thus,


d2
d2
d
(9)
(|y|) = 2 (|y|)
+ 2(y)
(|y|).
d|y|
dy 2
dy
y=0
This -function condition at B1 leads to a boundary condition


k1 ,
y=0+

(10)

where we define ks in terms of the bulk cosmological constant and the brane tension,

b
1
k ,
(11)
k1
.
6
6
Let us find a solution for the bulk equation Eq. (7) with the boundary condition Eq. (10).
We define a in terms of A,

1
.
a=
(12)
6A
The solution of Eq. (7) consistent with the Z2 symmetry is
 1/4


1/4
k
cosh 4k|y| + c
,
(|y|) =
(13)
a
where c is an integration constant to be determined by the boundary condition Eq. (10).
This solution, consistent with condition (i), is possible for any value of the brane
tension 1 . Note that c can take any sign. This solution gives a localized gravity consistent
with the above condition (ii). The boundary condition (10) determines c in terms of b
and 1 ,
 


1
1 k1
1
= tanh
.
c = tanh
(14)

k
6b
A schematic shape of (y) is shown in Fig. 1.

312

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

Fig. 1. (y) as a function of y for the flat space ansatz. It is plotted for k = 1 and a = 1.

The effective 4D Planck mass is finite


 1/2 


1
M3
1
2
3 k
d
MP,eff = 2M
=
F ,
a
cosh(4ky + c)
2ka
2 0
0

M3
=
2ka

1

1(cosh(c))1

dx

.

(1 x 2 ) 1 12 x 2

Here F (, r) is the elliptic integral of the first kind and



= sin1 (cosh(4ky + c) 1)/(cosh(4ky + c)) .

(15)

(16)

Note that the Planck mass is given in terms of the integration constant a, or the
integration constant is expressed in terms of the fundamental mass M and the 4D Planck
mass MP,eff ,

 2

M3
1
a=
(17)
.
F ,
2
MP,eff
2k
2 0
Curved space solution. The curved space solution is a time-dependent solution which
will be discussed in the KKL model of Section 3.
Localization of gravity and no tachyon. The perturbed metric near the above background
is

ds 2 = 2 + h dx dx + 2h 5 dx dy + (1 + h 55 ) dy 2

1 
2
( + h ) dx dx + 2h5 dx dz + (1 + h55 ) dz2 ,
K (z)

(18)
(19)

is interpreted as 5-dimensional graviton. Here we make a change of


where hMN (h MN
 y ) dy
variable by z =
(y) .

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

313

Then the linearized Einstein and energymomentum tensors near the above background
solution are given by

1
hMN + (M P h N)P MN P Q h P Q
2
2
3K
(M hN5 + N hM5 5 hMN )

2K





K P
K 2
K 2
K
K
3MN
+ 2 2 h55
hP 5 3
2 2 hMN ,

K
K
K
K
K

GMN =


K 2
K
3k1
18a 2
+ 6 2 h + (z)
h55 10 h55 ,
= 3
K
K
K
K

(20)
(21)

K 2
6a 2
(22)
h

h55 ,
55
K2
K 10
K 2
T5 = 6 2 h5 ,
(23)
K
where a is given in Eq. (12) and denotes the derivative with respect to z and M
MN N , + z2 . h MN is defined as h MN hMN MN h/2 and h is the trace
for hMN . When deriving the linearized expression for the energymomentum tensor near
the background, we used Eqs. (7) and (8).
The 5-dimensional graviton has 15 degrees of freedom maximally due to its symmetric
property in 5-dimensional spacetime. However, the 5-dimensional graviton has at most 9
physical degrees of freedom, because it is decomposed into the tensor, vector and scalar
modes in 4 dimension spacetime, which have maximally 5, 3, 1 degrees of freedom,
respectively. Hence, we should take at least 6 gauge conditions to analyze the spectrums
including the 4-dimensional massive modes from the 5-dimensional graviton. After solving
the equations of motion we could, of course, require additional gauge conditions and
reduce the degrees of freedom for the massless modes.
As K(z), which is a part of metric, specifies a direction z (or y) the equation of motion
for the graviton gives a relation between h55 and other components of metric fluctuations
as shown in Eqs. (20)(23). Thus we should carefully choose the gauge conditions such that
they are consistent with the relation given by the equations for the 5-dimensional graviton.
We find that the following gauge conditions are appropriate ones,
T55 = 6

M h MN = 0,

(24)

h55 = 0.

(25)

Due to the above gauge condition, the remaining degrees of freedom are 9. Under the
gauges, linearizations of the Einstein equations in the conformal coordinate Eq. (19) are

1
3K
h + h
( h5 + h5 5 h ) = 0,
2
4
2K
h = 0,


K
+6
h5 = 0,
K

(26)
(27)
(28)

314

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

where Eqs. (26)(28) are (), (55), (5) components of the linearized Einstein equations,
respectively.
The trace part of Eqs. (26) and (27) describe the scalar mode in the 5-dimensional
graviton. Here we can confirm the consistency by checking that the trace of Eq. (26) is
the same as Eq. (27). Eq. (26) is also the equation for the tensor mode (4-dimensional
graviton), and Eq. (28) is the equation for the 4-dimensional vector mode. From Eq. (27),
we see the scalar mode h does not have any tachyonic state.
To solve the Eq. (28), let us separate the variables of the vector mode such as h5 (x, z) =
h 5 (x)V (z). Then Eq. (28) is



4 m2 h 5 (z) = 0,


24a 2
12k1
z2
(z) 10 V (z) = m2 V (z),
K
K

(29)
(30)

where m2 is interpreted as the 4-dimensional effective mass for the KK mode. Eq. (30) is
rewritten in the y coordinate, which is defined in Eqs. (18) and (19), as
 2 2

y + y + 12k1 2 (y) + 24a 2 10 + m2 V (y) = 0,
(31)
where we used Eqs. (7) and (8), and in this coordinate h 5 (x, y) h 5 (x) V (y).
Note that under the Z2 symmetry h and h 55 have even parity but h5 has odd parity
h (y) = h (y),

(32)

h 55 (y) = h 55 (y),

(33)

h 5 (y) = h 5 (y).

(34)

Hence, V (y) should satisfy the boundary conditions


V (y = 0) = y2 V (y = 0) = 0.

(35)

We see that if the solution has odd parity, all terms except the second one in Eq. (31), are
explicitly zero at y = 0. Therefore, any stable odd parity solution of h5 is not allowed
due to the boundary conditions, even though there could exist bulk solutions. Note that
the above result was possible in our nonflat background case, because (y = 0) = 0 and
y V (y = 0) = 0 generally in the linear derivative term in Eq. (31). Thus, the Gaussian
normal condition h 55 = h 5 = 0 [20] is a natural gauge in our case.
With Eqs. (25)(27) and h 5 = 0, we derive () equation as


K
 3 5 h = 0,
(36)
K
or

z2 + V (z) (z) = m2 (z),

where h = K 3/2 (z)(z)$ eipx (p2 = m2 ). The potential V (z) is given by

(37)

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

15K 2 3K

2K
4K 2
15 2 3k1
39a 2
k

=
(z)

4K 2
K
4K 10

2
 2 
9 C
3 C
=
+
,
4 z
2 z2

315

V (z) =

(38)

where K(z) = 1/(y(z)) = exp(C). Therefore, we can see that m2  0, i.e., there is no
Tachyon state near the background, by regarding the above equation as a supersymmetric
quantum mechanics [19],



3 C
3 C

Q Q (z) z +
(39)
z +
(z) = m2 (z).
2 z
2 z
Moreover, there appears the localization of gravity on the brane as expected from the finite
4D Planck mass, due to one bound state of massless graviton, 0 (z) e3C/2 = K 3/2 , that
is, h0 = c$ eipx (c is a constant). Note that the zero mode solution 0 (z) automatically


k1

z=0+ = 0 from Eq. (37). Since V (z) 0 as
satisfies the boundary condition z + 2K
z , there is no mass gap for KK massive modes, which start with m2 = 0. The KK
massive modes just gives rise to small corrections to the Newton potential at the length
scale larger than the curvature scale 1/k as in the RS case [12].
More general case with 1/H 2n . So far we presented a model for a simple Lagrangian
with 1/H 2 term only. But we can show that more general Lagrangians containing only
negative powers of H 2 can have the self-tuning solutions. Suppose that Lagrangian


contains n an /(H 2 )n with a1 > 0 and large for the 1/H 2  dominance, the last terms
of the (55) and () Einstein equations, (7) and (8), are changed to

 Cn ()
n

An

 (2n + 1)Cn ()
n

An

(40)

respectively, with A1 > 0. Then checking the two equations, we obtain Cn () = 8n for
the consistency. Then, the (55) equation gives

 8n+2
b

.
| | = 2
(41)
6
6An
n
Eq. (41) gives 0 as 0 if n  0, which guarantees the existence of the solution.
But for n < 0, there exists a naked singularity and there is no solution.
This observation is very useful when we consider the radiative corrections on the
background of our self-tuning solution. Suppose that the 1/H 2  is given with possibly
small additional terms of other negative powers of H 2 . This means that our solution is not
expressible in terms of the cosh1/4 , but there is a solution. Then the radiative corrections

which are powers of H 2 can be brought approximately into the form n0 an /(H 2  +
H 2 )n where the n = 1 term is dominant. Thus, if the corrections contain only the (H 2 )n
type terms, the existence of the self-tuning solution is intact.

316

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

2.2. One-fine-tuning solutions of the cosmological constant with H 2 term


In this section, we obtain just the flat space solution for the H 2 term in the action,
 3



M
M
S = d 4 x dy g
R
HMNP Q H MNP Q
2
2 4!


b +
L(i)
(y

y
)
.
(42)
i
m
i

The ansatze for the metric and the four form field are taken also as given in Eqs. (2) and
(5), respectively. Thus the field equation for AMNP is trivially satisfied again,


M g H MNP Q = 0.
(43)
The two relevant Einstein equations are the (55) and () components,
 2

A
6
= b + 8 ,

 2
 

A
3
+3
= b 1 (y) 2 (y yc ) 8 ,

where A/ 8 1/2n2 expressed in terms of a positive constant A.


The solutions of Eqs. (44) and (45) are
 1/4


1/4
a
sinh |4k|y| + c|
,
for b < 0: (|y|) =
k
 1/4


1/4
a
,
for b > 0: (|y|) =
sin |4k|y| + c|
k

1/4
,
for b = 0: (|y|) = |4a|y| + c|

(44)
(45)

(46)
(47)
(48)

where the a is defined in terms of A,



A
a
(49)
.
6
We note that for a positive c in Eqs. (48)(50) s do not give localized graviton solutions
near the brane B1, and for a negative c s have naked singularities at |y| = c/4k or
c/4a. Therefore, to get the effective four-dimensional gravity or to avoid the sigularities
in the bulk, it is indispensable to cut the extra dimension such that it has a finite length
size by introducing another brane, say B2. Thus, we need at least two branes and the
situation is similar to that of the RSI except for the 4 form field contributions.Since the
extra dimension is finite, the effective four-dimensional Planck mass MP M 3 dy 2 is
also finite. If we introduce two branes, we should satisfy the boundary conditions at the
two branes, consistently with the S 1 /Z2 orbifold symmetry,



d
d2
d2

(|y|).
(50)
(|y|)
=
(|y|)
+ 2 (y) (y yc )

dy 2
dy 2
d|y|
y=0

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

Then the boundary conditions for the above three cases are
 
 
1 k1
1 k2
for b < 0: c = coth
= 4kyc coth
,
k
k
 
 
k1
k2
= 4kyc cot1
,
for b > 0: c = cot1
k
k


a
1
for b = 0: c =
= a 4yc
,
k1
k2

317

(51)
(52)
(53)

where k2 is defined in terms of the brane tension 2 at B2,


k2

2
.
6

(54)

We note that in the case of the H 2 term (not 1/H 2 ) in the action, the one-fine-tuning
relations between k1 and k2 appear always, e.g., the relations (53)(55), while in the case
of RSI model, the two-fine-tuning relations k = k1 = k2 were inevitable.
If we introduce both 1/H 2 and H 2 in the action, there does not exist a flat space selftuning solution. In this case, the derivative of the warp factor satisfies

8
A
b 2

+ 8.
= +
(55)
6
6A 6
A necessary condition for the existence of a flat space solution is 0 and 0 as
y . Certainly, Eq. (55) does not satisfy this necessary condition. On the other hand,
a necessary condition for the de-Sitter space regular horizon is that = finite as 0.
Also, this condition is not met in Eq. (55) and hence there does not exist de-Sitter space
regular horizon.
2.3. Comparison with other tuning solutions
There appeared several self-tuning solutions since early eighties [3,5,14,19]. The early
stage scenario [3,5] used field strength of three index antisymmetric tensor field H to
introduce an integration constant. The vacuum value of H  = $ c introduces an
integration constant c which contributes to the cosmological constant c2 . Therefore,
there exists a value of c such that the effective cosmological constant vanishes for a
range of bare cosmological constant. Once c is determined to give the zero effective
cosmological constant, c cannot change since it is a constant. In this sense, the fourdimensional example is not a working model. Thus, the self-tuning solution needed a
dynamical field to propagate. The three index antisymmetric field is not a dynamical field
in 4D spacetime.
Introduction of the extra dimension opened a new game in the self-tuning solutions of
the cosmological constants [1,14,1719,26]. Here, we briefly comments on the key points
of these solutions.
The California self-tuning solution [14,15] must use the RSI model set up, i.e., introduce
two branes B1 and B2, and introduce a specific form for the potential of a bulk scalar field

318

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

coupling to the brane matter with a desired form. But the model has a naked singularity and
to cure the problem, one has to introduce a brane at this singular point. Then, one needs
a condition at the new brane and leads to one fine-tuning there [15]. If one satisfies this
one-fine-tuning condition between parameters in the Lagrangian, there always exists a flat
space solution. Therefore, the solution is not a self-tuning solution originally anticipated.
One interesting or (disastrous to some) point of this model is that it does not allow the
de-Sitter or anti-de-Sitter space solutions. In this sense, it is an improvement over the
original proposal [3,5] for the self-tuning solutions. However, it may be difficult to obtain
a period of inflation since the de-Sitter space exponential expansion is not possible. One
should see whether the bulk energymomentum tensor satisfies = p = p5 /2 to have
the exponential expansion of the effective 4D space [25]. However, at present it is not
known what matter satisfies this kind of equation of state.
There appeared another interesting self-tuning solution [19,27], which does not
introduce any brane. This model assumes a specific form of the 5D scalar potential
multiplied to the matter Lagrangian in 5D. Even though the metric gives a localized gravity,
the matter fields propagate in the full 5D space since the potential tends to a constant value
as y . It is not certain how we are forbidden to realize the extra dimension in this
model.
Then, there are proposals that the de-Sitter space solution is phenomenologically
acceptable as far as the curvature at present is sufficiently small [1618,26]. In a sense,
it also tries to accommodate the current small vacuum energy [28] with the de-Sitter space
solutions. (On the other hand, note that the quintessence idea is based on the solution of
the cosmological constant problem [29].) The way the proposals make the vacuum energy
small is to separate the distance between branes sufficiently large since the vacuum energy
is exponentially decreasing with the separating distance. The one fine-tuning at B1 is met
but the second fine tuning at B2 is not satisfied and allows the de-Sitter space solutions.
To have the vacuum energy decreased down to the current energy density, one needs that
the separation distance increases as t increases. Namely, the horizon point at the y axis
is required to increase. Then it is possible to relate the current vacuum energy with the
current mass energy. In Ref. [26], the relation = (5/2)m is obtained for total = 1.
The model presented in [1] allows a self-tuning flat space solution, self-tuning de-Sitter
and anti-de-Sitter space solutions. In general it is possible to introduce a period of inflation.
The transition from one flat space to another flat space can arise following the solutions of
the Einstein and field equations, as we discuss in the subsequent sections.

3. De-Sitter space solutions


We find that there exist de-Sitter space solutions in our model with a simple time
dependence. First, let us briefly discuss the de-Sitter space solution [23] in the RSII model
and present de-Sitter space solutions in our model.

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

319

3.1. The RSII model


Let us consider the following metric ansatz for the dS4 brane in the RSII model:

ds 2 = 2 (y) dt 2 + e2 t ij dx i dx j + dy 2
= 2 (y)g dx dx + dy 2

(56)

from which the 4D Ricci tensor is given as


= 3 g . Then, the warp factor (y) of
the dS4 brane solution is given by

 


1

1 k1
sinh k(ym y) , ym = coth
(y) =
(57)
,
k
k
k
(4)
R

where the integration constant ym is determined from the boundary condition at the brane
and

b
1
k1
.
k ,
(58)
6
6
For a positive tension brane, we get a positive value of ym , which gives rise to an event
horizon at the finite proper distance away from the brane. Therefore, the region beyond
the horizon cannot be causally connected to the region where the brane resides. This
bulk horizon resembles the dS4 horizon on the brane. And, the dS4 solution is regular
along the bulk because the curvature tensors become finite at the bulk horizon. To say
about whether the dS4 solution could describe a dynamical compactification of the extra
dimension with the region beyond the horizon cut off, we should check its consistency
from the 4D effective cosmological constant by integrating out the fifth dimension. The
4D effective action is given by


ym

1
d x g (4)
dy 4 R (5) b 1 (y) ,
2


S4,eff 

ym

 2


MP,eff (4)
R 3eff ,
d 4 x g (4)
2




(59)

where g (4) and R (4) are given from g = g + h in the zero mode expansion and the
4D effective Planck mass and the 4D effective cosmological constant are determined as
follows,



1
2
MP,eff = 3 kym + sinh(2kym ) > 0
(60)
2
k
and
1
eff =
3
=

ym

 2



+6
dy 4
+ b + 1 (y)

ym

2
k3

kym +


1
sinh(2kym ) > 0.
2

(61)

320

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

Then, when we compare the 4D Einstein equations of motion derived from the above
4D effective action with those satisfied by the dS4 brane solution, we can show that the
2
ratio eff /MP,eff
is exactly equal to the 4D cosmological constant of the brane solution.
Therefore, the dS4 brane solution can be regarded to reproduce the 4D effective de-Sitter
spacetime with the extra dimension dynamically compactified up to the horizon distance
in the bulk.
It is easy to observe that 1 < 0 does not allow a horizon, which implies a nonlocalized
gravity.
3.2. The KKL model
The background metric is
ds 2 = 2 (y)g dx dx + dy 2,

(62)

where the metric corresponds to one of the 4D maximally symmetric = 0 spaces:

(dS4 background, > 0),


g = diag 1, e2 t , e2 t , e2 t

for de-Sitter space, and

g = diag e2 x3 , e2 x3 , e2 x3 , 1 (AdS4 background, < 0),


for anti-de-Sitter space.
 = 3 g . And, components of the five-dimensional
The 4D Ricci tensor is given as R
Einstein tensor are
  

 2

G = 3
(63)
3
2 g ,
+3

 2

2 ,
G55 = 6
(64)
6

where > 0 for dS4 and < 0 for AdS4 background. The term arises from the time
derivatives of the metric. Then, the (55) and () components of the Einsteins equations
with 1/H 2 follows,

6

3

2

2 = b
6

2


+3

8
,
A

2 = b 1 (y) 3
3

(65)
8
.
A

(66)

We can obtain bulk solutions by solving the (55) component.


We can consider two cases: = 0 at y = ym = (finite) or 0 as y . Otherwise,
the 4D Planck mass diverges and localizable gravity does not follow. The case = 0 at a
finite y arises in the de-Sitter space solution.

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

321

For the anti-de-Sitter space, let us consider the asymptotic behavior of the warp factor
(y) as y . With z = 1/y, we can rewrite the (55) component as
 2
d

k 2 2 a 2 10
(67)
= 4+ 4
,
dz
z
z
z4
where
k2

b
,
6

a2

1
.
6A

(68)

To get the finite 4D Planck mass from the noncompact extra dimension or have the
localized gravity, should be zero as z 0. However, the 4D cosmological constant
term should be divergent as z 0 even if the last two terms are set to be zero. Therefore,
there does not appear the localized gravity for the AdS background with nonzero effective
4D cosmological constants.
In the KKL model [1] the de-Sitter space metric ansatz gives as

= k 2 + k 2 2 a 2 10,
(69)
where
k =

,
6


k=

b
.
6

(70)

The boundary condition at y = 0 relates the brane tension (k1 ) and the bulk cosmological
constant,


,
k1 = k 2 a 2 8 (0) +
(71)
(0)2
where (0) is a function of an integration constant c. The above relation (71) is valid for
both the de-Sitter and anti-de-Sitter space solutions.
Consider the half space y > 0. If we choose the minus sign in front of the square root of
Eq. (69), there always exist ym as in the RSII model.
Even if we choose the plus sign in front of the square root of Eq. (69) near y = 0+ , there
exist horizons. This is because there can exist a point where vanishes for a sufficiently
large in the region where increases. For this not to be realized, should tend to an
asymptotic limit as y . If it goes to an asymptotic limit, the Planck mass becomes
infinite and there does not result a localized gravity. Therefore, we restrict to the case for
turning around and coming down, in which case there results a horizon which is determined
by the point where = 0 at y = ym . Between y = 0 and y = ym , there can (1 < 0) or

cannot (1 > 0) exist a point
= 0. Even if there exists such a point as in the
 where
above example, the integral d/ k 2 + k 2 2 a 2 10 always converges, and we obtain
a localized gravity with the horizon at y = ym . Therefore, there always exist de-Sitter
space solutions, which have the periodic form as shown in Fig. 3. Similarly, there exist
anti-de-Sitter space solutions, which have the periodic form as shown in Fig. 2. But the

322

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

Fig. 2. (y) as a function of y for the anti-de-Sitter space solution. It is plotted for k 2 = 0.01, k = 1

and a = 1, where k 2 = .

Fig. 3. (y) as a function of y for the de-Sitter space solution. It is plotted for k 2 = 0.01, k = 1 and

a = 1, where k 2 = .

anti-de-Sitter space solution does not lead to a localized gravity. Even if we allow the antide-Sitter space solutions, neglecting the unphysical unlocalized gravity, the probability to
choose the anti-de-Sitter space is exponentially small [5].
The curved space solutions in the KKL model have the effective bulk cosmological
constant more negative compared to the RSII case. Therefore, the bound from the weaker
energy condition on the background matter supporting those spacetimes is completely
satisfied. On the other hand the weaker energy condition is saturated in the RSII
case [23]. The weaker energy condition reads TMN M N  0 for a null vector M , which
2 for the background metric Eq. (62). In the KKL case,
corresponds to ( /) 

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

323

the weaker energy condition is always satisfied without the possibility of saturation,
 

2 4a 2 8 <
2 .
=
(72)

Existence of de-Sitter space solutions in the KKL model allows possible cosmological
constants in 4D space as nonnegative. Hawkings probabilistic interpretation of the
cosmological constant chooses the flat space [5]. However, we need a qualification in this
statement. Existence of nonlocalizable de-Sitter space solution can give a large probability.
The Euclidean action can be written schematically as



R4
+ ,
SE d 4 x MP2
(73)
2
after the y integration. The de-Sitter space solutions have horizons as shown in Fig. 3.
If we integrate over the whole region of the y space MP diverges. But the Planck mass
must be from the integration only up to the horizon connected to y = 0 brane which we
call the first universe. Then MP is finite in the first universe. However, if we neglect the
effect of the brane, the probability to go the second universe (the universe between the
next two horizons) is of the same order as the probability to go to the first de-Sitter space
universe, because the horizon points appear periodically. Summing up the probabilities to
go to either of these de-Sitter universes, we would obtain an infinity. Nevertheless, the
second universe does not contain the brane, and we have to exclude this possibility of
transition to the universes not containing the brane. Therefore, MP is considered to be
finite, and the probability to stay in the flat universe is maximum [5].

4. Time-dependent solution of the self-tuning model


The existence of the self-tuning solution will be a great leap toward the understanding
of the vanishing cosmological constant [3,5]. As shown in the previous section, the action
given in Ref. [1] also allows de-Sitter and anti-de-Sitter space solutions.
Note that if there does not exist any de-Sitter space and anti-de-Sitter space solutions
but there exist flat space self-tuning solutions then the cosmological constant may become
automatically zero for a finite range of parameters in the Lagrangian. In this sense, our selftuning solution does not lead to a vanishing cosmological constant automatically but the
cosmological constant is probably zero a la Hawking [5]. However, the automatic solutions
have a difficulty in implementing inflationary period which is needed for our sufficiently
homogeneous and isotropic universe [30].
It is anticipated that the spontaneous symmetry breaking at the brane proceeds at the
electroweak phase transition and the QCD phase transition. Then the effective potential
energy will have a time dependence of the form, 1 (t)(y). According to the shift of the
potential energy at the brane, the solution of the field and Einstein equations will have
a time-dependence. Certainly, there will exist time-dependent solutions, and we will be
interested in solutions a la Hawking [5], changing from one flat space solution with an
integration constant c1 corresponding to the brane cosmological constant old to another

324

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

integration constant c2 corresponding to the brane cosmological constant new . This situation corresponds to a time-dependent energymomentum tensor. This transition from a
flat space solution to another flat space solution is through satisfying the field equations
and is different from Wittens sudden choice of a flat space solution [3]. There can be timedependent solutions from a flat space to de-Sitter or anti-de-Sitter spaces, but the probability for these transitions is exponentially small compared to a flat to flat transition [5].
To study the time-dependence the metric is taken as
ds 2 = n2 (t, y) dt 2 + a 2 (t, y)ij dx i dx j + b2 (t, y) dy 2.
The Einstein tensors are






3 a a
b
3 a a a b
+
+ 2
+

,
G00 = g00 2
a
a a
b
n a a b
b



 

a
1
a b a
n a
b n
Gii = gii 2 2 +
2
2

n
a b a n a
b n
a





a a n a
a
1 n
b n
+2 +
+2
,
+ 2
2 +

b n
a
a
n
a
b n
a






3 a a n a
3 a n a
G55 = g55 2
+ 2

+
,
a
n a a n a
b a n


b a a
a n
,
+

G05 = 3
a n
ba
a

(74)

(75)

where dot and prime denote differentiations with respect to t and y, respectively.
With the brane tension 1 at the y = 0 brane and the bulk cosmological constant b ,
the energymomentum tensor is

1 (y)
TMN = gMN b g M N


4
1
1
P QR
+ 4 4!
HMP QR HN
+ gMN 2 .
H4
2
H

(76)

Considering the homogeneous 3D space, nonzero components of H MNP Q are


1
H = $ 5 5 ,
g

1
H 5ij k = $ 5ij k0 0 .
g

(77)

Then, we have

H 2 = 4! f 2 h2 ,

H0NP Q H0 NP Q = 3!g00 f 2 ,

H5NP Q H5 NP Q = 3!g55 h2 ,
HiNP Q Hj NP Q = 3!gij f 2 h2 ,
H5NP Q H0 NP Q = 3!bnf h,

H5NP Q Hi NP Q = 0,

(78)

where f 2 = 2 /b2 and h2 = 2 /n2 . The TMN appearing in the Einstein equations
GMN = TMN are


6
1
4h2
(y) + 2
+
T00 = g00 b +
,
b
f h2 (f 2 h2 )2

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329



6
1
(y) + 2
Tij = gij b +
,
b
f h2


2
4h2
T55 = g55 b + 2

,
f h2 (f 2 h2 )2

T05 =

4bnf h
.
(f 2 h2 )2

The field equation of the four-form field is




 0ij k5 
 5ij k0 
$
$
g H MNOP

= 0
+ 5
M
H4
H4
H4
 



$ 0ij k5

=
0
5
= 0.
(4!)2
(( /b)2 ( /n)2 )2
(( /b)2 ( /n)2 )2
Note that is given, in the static homogeneous 4D case considered above, by

ab
(|y|) = 2 2A sinh 4kb|y| + c ,
4k

325

(79)

(80)

(81)

with 2/f 2 = 8 /A.


It is very difficult to find a general solution of the above Einstein equations. However, we
are interested in the existence of the interpolating solution between two flat space solutions.

We anticipate that a flat space solution, viz. Eq. (13), with ci = tanh1 (i / 6b ) is
given, probably by the initial condition or Hawkings choice of the cosmological constant.
During a phase transition when old changes to new , we try to show the existence of an
interpolating solution connecting two flat space solutions with c1 and c2 .
Let us consider the case that the brane tension 1 = old changes to 1 = new
instantaneously due to a phase transition by brane matter at the brane. Then the boundary
condition requires a time dependent b,
b(t) = (bnew bold ) (t t0 ) + bold,

(82)

which gives
 1/4


1/4
k
(|y|, t) =
cosh 4kb(t)(|y| + y1 )
,
a

for t = t0 ,

(83)

where bold , bnew and y1 are constants. c in Eq. (13) is endowed with a time-dependence as
given above. The t dependence of generates nonzero time-derivatives of our metric in
the Einstein equation. We suppose that the metric dynamics gives rise to the bulk matter
fluctuation near the vacuum, T (m)M N diag(, p, p, p, p5 ),


3 b
= 2
(84)
+
,
b
 2 


b
b
1
+
p= 2 2 +
(85)
,

b
 
3
.
p5 = 2
(86)

Note that the bulk matter contribution vanishes when t = t0 , because the matter would be
t0 ).
proportional to (t t0 ), 2 (t t0 ) and (t

326

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

If the following relation is satisfied,


 

b
(87)
,
=

b
which leads to G05 = 0, then the 5D continuity equations are automatically satisfied [25],
b

+ 3 ( + p) + ( + p5 ) = 0,
(88)

b


p5 + 3 (p5 p) + ( + p5 ) = 0.
(89)

We will see that Eq. (87) is satisfied for our solution.


The t dependence of b in Eq. (82) gives also the t dependence of the four form field,

H MNP Q $ MNP QR R / g, where is


a
(|y|, t) = 2 2A b(t) sinh 4kb(t)(|y| + y1 ) ,
(90)
4k
where A is the constant. The t dependence of generates also nonzero H ij k5 component
of the four form field in Eq. (77), that would be proportional also to the delta function,
h (t t0 ). Since the h term on the RHS of the Einstein equations appear as
hm /(f 2 h2 )n/2 with 0  m < n, the (t)-function of h gives vanishing contribution at
t = 0. Thus, it turns out that T05 = 0 for any t. For t = t0 , Eq. (83) satisfies the Einstein
equations with the f 2 dependence on the RHS [1]. Near t = t0 , TMN are reduced to those
of the RS [12] due to the divergence of or h (which kills the f 2 dependence), and so the
solution becomes the RS solution [12].
We have shown above three solutions: the flat space solution with c1 = tanh1 (old /

6b ) for t < t0 , the RS solution [12] with the fine-tuning k = k1 , and the flat space

solution with c2 = tanh1 (new / 6b ) for t > t0 . The fine-tuning solution at t = 0


means simply that our solution goes through such an intermediate stage. Namely, to satisfy

the field equation and the Einstein equations, old suddenly jumps to 6b and again
suddenly jumps to new . If the transition from old to new is not abrupt, there would
exist a solution smoothly connecting the two flat solutions with c1 and c2 . But the above
example shows that at the intermediate stage it would go through the RS solution [12].
The consistency of the solution is achieved by showing Eq. (87) using the divergence
(m)
of h. As the time derivatives of the metric in G55 = T55 + T55
have been identified already
(m)
with p5 of T55 , the remnant reads as

1
h2

b 1
2
= b(t)

(91)
+
,
2
2
2

6
3f h
3 (f h2 )2
from which we can see immediately that Eq. (87) is satisfied,

 
1
h2

b
b 1
2

= b
+
2
2
2

b
6
3f h
3 (f h2 )2

1
h2

b 1
2

+
=
t
6
3 f 2 h2 3 (f 2 h2 )2

 
b
,
b

(92)

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

where we used


1

= 0.
t 2 (t t0 )

327

(93)

Note that Eq. (93) can be shown for a specific representation of the delta function.
We should check whether the equation of motion for the four form field Eq. (80) remains
satisfied even when the constant b is changed to b(t) like Eq. (82). But we can also see
easily that even in the case, it is satisfied always regardless of t = t0 or not. When t = t0 , it
is the static case, and so the equation of motion is trivially satisfied. On the other hand, for
t t0 the equation of motion is also satisfied due to the (t t0 ) divergence of h, which is
checked easily if we use Eq. (93).
We have seen that there exists a time-dependent solution when the brane tension changes
from old to new instantaneously. In this case, the universe starting from a flat universe
can go to a flat universe instantaneously. Instead of the sudden shift of the brane tension,
the transition from old to new can be smooth. We argue that in this case also, there exists
a solution since we can replace the function with a smooth function D(t; $) of t defined
in a short interval $, such that lim$0 D(t; $) = (t). The error by introducing $ is at most
$ and hence our solution is approximate, but this shows that there can exists a smooth
solution very close to our approximate solution with an error of order $.
There may exist solutions connecting de-Sitter space and flat space, de-Sitter space and
de-Sitter space, etc. [31]. But the final universe may be chosen probabilistically to be the
flat universe a la Hawking [5].
5. Conclusion
We considered the cosmological constant in the (4 + 1) dimension with a warp factor.
A brane, containing the matter fields, is located at the origin y = 0 of the extra dimension.
To separate the 4D space we introduced a three index antisymmetric tensor field with the
field strength H . The simple form for the H action is considered: the Lagrangian with
1/H 2 and H 2 . We found a self-tuning solution, Eq. (13), with 1/H 2 . The H 2 term does
not allow a self-tuning solution, but allows nonlocalizable gravity with one brane or onefine-tuning solutions with two branes.
We concentrated the discussion related to the self-tuning solution, Eq. (13). We also
found that there exist de-Sitter space and anti-de-Sitter space solutions with 1/H 2 term.
For a finite range of parameters of the bulk cosmological constant b and the brane
tension 1 , there always exists the flat space solution but it is not unique. Therefore,
when the boundary condition is changed at the electroweak or QCD phase transitions,
the cosmological constant after the phase transitions can be anything allowed by the
Einstein equations. However, the probability to choose the vanishing cosmological constant
is infinitely large compared to the others [5]. This argument is applicable in our case since
there exists the self-tuning solution, Eq. (13).
This may not sound so attractive as any model not allowing de-Sitter or anti-de-Sitter
space solutions. However, if there exists a such model, then it may not allow inflation

328

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

which is probably needed in cosmology. In the self-tuning model the transition from
one integration constant to another integration constant is possible through satisfying the
equations of motion.
For the self-tuning solution we presented to work in cosmology, there must be a
natural explanation of the current acceleration of the expansion [28]. There appeared some
proposals [29], but these must be shown to work with the self-tuning solution.

Acknowledgements
This work is supported in part by the BK21 program of Ministry of Education, Korea
Research Foundation Grant No. KRF-2000-015-DP0072, CTP Research Fund of Seoul
National University, and by the Center for High Energy Physics(CHEP) of Kyungpook
National University.

References
[1] J.E. Kim, B. Kyae, H.M. Lee, hep-th/0011118.
[2] M. Veltman, Phys. Rev. Lett. 34 (1975) 777;
For a review, see, S. Weinberg, Rev. Mod. Phys. 61 (1989) 1.
[3] E. Witten, in: R. Jackiw, H. Khuri, S. Weinberg, E. Witten (Eds.), Proc. Shelter Island II
Proceedings, Shelter Island, 1983, MIT Press, 1985, p. 369.
[4] E. Baum, Phys. Lett. B 138 (1983) 185.
[5] S. Hawking, Phys. Lett. B 134 (1984) 403.
[6] S. Coleman, Nucl. Phys. B 310 (1988) 643.
[7] A. Linde, Phys. Lett. B 200 (1988) 272.
[8] M. Duff, Phys. Lett. B 226 (1989) 36.
[9] S. Weinberg, Phys. Rev. Lett. 59 (1987) 2607;
S.M. Barr, J.F. Donoghue, D. Seckel, Phys. Rev. D 57 (1998) 5480;
T. Banks, M. Dine, L. Motl, hep-th/0007206.
[10] G. Dee, J.S. Langer, Phys. Rev. Lett. 50 (1983) 383;
R.C. Brower, D.A. Kessler, J. Koplik, H. Levine, Phys. Rev. Lett. 51 (1983) 1111.
[11] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-th/9905221.
[12] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
[13] V.A. Rubakov, M.E. Shaposhinikov, Phys. Lett. B 125 (1983) 139.
[14] N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, Phys. Lett. B 480 (2000) 193,
hep-th/0001197;
S. Kachru, M. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021, hep-th/0001206;
C. Csaki, J. Erlich, C. Grojean, T. Hollowood, Nucl. Phys. B 584 (2000) 359, hep-th/0004133.
[15] S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, Phys. Lett. B 481 (2000) 360, hep-th/0002164;
S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, JHEP 0009 (2000) 034, hep-th/0006139.
[16] H.B. Kim, Phys. Lett. B 478 (2000) 285.
[17] S.-H.H. Tye, I. Wasserman, hep-th/0006068;
E. Flanagan, N. Jones, H. Stoica, S.-H.H. Tye, I. Wasserman, hep-th/0012129.
[18] J.M. Cline, H. Firouzjahi, hep-ph/0012090;
N. Tetradis, hep-th/0012106.
[19] A. Kehagias, K. Tamvakis, hep-th/0011006.

J.E. Kim et al. / Nuclear Physics B 613 (2001) 306329

[20] J. Garriga, T. Tanaka, Phys. Rev. Lett. 84 (2000) 2778, hep-th/9911055;


S. Giddings, E. Katz, L. Randall, JHEP 0003 (2000) 023, hep-th/0002091;
J.E. Kim, H.M. Lee, hep-th/0010093.
[21] J.E. Kim, B. Kyae, H.M. Lee, Phys. Rev. D 62 (2000) 045013, hep-ph/9912344;
J.E. Kim, B. Kyae, H.M. Lee, Nucl. Phys. B 582 (2000) 296, hep-th/0004005;
I. Low, A. Zee, Nucl. Phys. B 585 (2000) 395, hep-th/0004124.
[22] P.G.O. Freund, M.A. Rubin, Phys. Lett. B 97 (1980) 233.
[23] A. Karch, L. Randall, hep-th/0011156;
See, earlier papers: T. Nihei, Phys. Lett. B 465 (1999) 81, hep-ph/9905487;
H.B. Kim, H.D. Kim, Phys. Rev. D 61 (2000) 064003, hep-th/9909053.
[24] T. Padmanabhan, S. Shankaranarayanan, hep-th/0011159.
[25] J.E. Kim, B. Kyae, Phys. Lett. B 486 (2000) 165, hep-th/0005139.
[26] Ph. Brax, A. Davis, hep-th/0011045.
[27] Z. Kakushadze, Phys. Lett. B 489 (2000) 207, hep-th/0006215.
[28] The Supernova Cosmology Project, Bull. Am. Astron. Soc. 5 (1998) 1351.
[29] P. Binetruy, Phys. Rev. D 60 (1999) 063502;
C. Kolda, D.H. Lyth, Phys. Lett. B 458 (1999) 197;
J.E. Kim, JHEP 9905 (1999) 022;
J.E. Kim, JHEP 0006 (2000) 016;
T. Chiba, Phys. Rev. D 60 (1999) 083508;
P. Brax, J. Martin, Phys. Lett. B 468 (1999) 40;
A. Masiero, M. Pietroni, F. Rosati, Phys. Rev. D 61 (2000) 023504.
[30] A. Guth, Phys. Rev. D 23 (1981) 347.
[31] G.T. Horowitz, I. Low, A. Zee, Phys. Rev. D 62 (2000) 086005, hep-th/0004206;
P. Binetruy, J.M. Cline, C. Grojean, Phys. Lett. B 489 (2000) 403, hep-th/0007029.

329

Nuclear Physics B 613 (2001) 330352


www.elsevier.com/locate/npe

Cosmological time in (2 + 1)-gravity


Riccardo Benedetti a , Enore Guadagnini b
a Dipartimento di Matematica, Universit di Pisa, Via F. Buonarroti 2, I-56127 Pisa, Italy
b Dipartimento di Fisica, Universit di Pisa, Via F. Buonarroti 2, I-56127 Pisa, Italy

Received 16 March 2001; accepted 2 August 2001

Abstract
We consider maximal globally hyperbolic flat (2 + 1)-spacetimes with compact space S of genus
g > 1. For any spacetime M of this type, the length of time that the events have been in existence is M
defines a global time, called the cosmological time CT of M, which reveals deep intrinsic properties
of spacetime. In particular, the past/future asymptotic states of the cosmological time recover and
decouple the linear and the translational parts of the ISO(2, 1)-valued holonomy of the flat spacetime.
The initial singularity can be interpreted as an isometric action of the fundamental group of S on a
suitable real tree. The initial singularity faithfully manifests itself as a lack of smoothness of the
embedding of the CT level surfaces into the spacetime M. The cosmological time determines a real
analytic curve in the Teichmller space of Riemann surfaces of genus g, which connects an interior
point (associated to the linear part of the holonomy) with a point on Thurstons natural boundary
(associated to the initial singularity). 2001 Elsevier Science B.V. All rights reserved.
PACS: 04.60.Kz
Keywords: Cosmological time; Asymptotic states; Real trees; Marked spectra

1. Introduction
We shall be mainly concerned with maximal globally hyperbolic, matter-free spacetimes
M of topological type S R, where S is a compact closed oriented surface of genus g > 1.
The (2 + 1)-dimensional Einstein equation with vanishing cosmological constant actually
implies that M is (Riemann) flat.
After [9] and [26], a large amount of literature has grown up about this (2 + 1)-gravity
topic, regarded as a useful toy-model for the higher-dimensional case. Two main kinds
of description have been experimented. A cosmological approach points to characterize
the spacetimes in terms of some distinguished global time; for instance the constant mean
curvature CMC time has been widely studied [3,17]. A geometric time-free approach
E-mail addresses: benedett@dm.unipi.it (R. Benedetti), guada@df.unipi.it (E. Guadagnini).
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 1 ) 0 0 3 8 6 - 8

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

331

eventually identifies each flat spacetime by means of its ISO(2, 1)-valued holonomy
[16,26]. With the exception of the case with toric space (g = 1), there is not a clear
correspondence between the results obtained in these two approaches.
The aim of this paper is to show that this gap can be filled by using the canonical
Cosmological Time CT, that is the length of time that the events of M have been in
existence (see [2]). It turns out that this is a global time which reveals the fundamental
properties of spacetime. It is canonically defined by means of the very basic spacetimes
structures: its casual structure and the Lorentz distance. The cosmological time is
invariant under diffeomorphisms, therefore, the = a level surfaces Sa provide a gaugeinvariant description of space evolution in M. Both the intrinsic and extrinsic geometry
of the surfaces Sa , as well as their past/future asymptotic states, are intrinsic features of
spacetime. The asymptotic states are defined by the evolution of the observables associated
to the length of closed geodesic curves on the surfaces Sa . Remarkably, they recover and
decouple the linear and the translational parts of the holonomy. The study of the asymptotic
states also leads to understand the initial singularity (we will always assume that the space
is future expanding) and the way how the classical geometry degenerates, but does not
completely disappear. The initial singularity can be interpreted as the isometric action of
the fundamental group of S on a suitable real tree. Differently to the case of the CMC
time (for instance), the level surfaces Sa of the CT are in general only C 1 -embedded into
the spacetime M. This lack of smoothness takes place on a geodesic lamination on Sa and
is a observable large scale manifestation of the intrinsic geometry of the initial singularity.
Thus the initial singularity admits two complementary descriptions: one, in terms of real
trees and, the second, in terms of geodesic laminations. The existence of a duality relation
between real trees and laminations was already known in the context of Thurston theory
of the boundary of the Teichmller space. It is remarkable that Einstein theory of (2 + 1)gravity sheds new light on this subject and puts duality in concrete form.
In [6] we have also used the cosmological time in order to study certain interesting
families of (2 + 1)-spacetimes coupled to particles.
Our main purpose consists of elucidating the central role of the cosmological time and
its asymptotic states in the description of spacetimes. The cosmological time perspective
provides a new interpretation of several facts spread in the literature which are related to
Thurston work. More precisely, the present article is based on, and could be considered a
complement of, Messs fundamental paper [16].
The importance of (2 + 1)-gravity has been pointed out by several authors, see for instance [8,9] and [26]; here we simply add a few comments. In general, the models which
have dynamical degrees of freedom associated with the spacetime geometry are of particular interest in physics. Indeed, gravitational interactions are supposed to be described
by general relativity in which the geometry of spacetime admits a nontrivial dynamical
evolution. A satisfactory knowledge of all the classical and quantum aspects of general relativity is still lacking; so, toy models which provide conceptual hints in this directions are
welcome. The matter-free (2 + 1)-gravity model with compact space is a remarkable example of general relativistic theory because the complete classical solution [16] has been
produced. In this context, one can then find explicit answers to some open problems. In

332

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

our article we have explored a few general topics which are related the problem of time in
classical gravity. The resulting conceptual hints that we have obtained are:
(i) a consistent correspondence between the dynamical and the static pictures of
spacetime exists, and has been explicitly produced in the model;
(ii) one can introduce a global canonical time which corresponds to the age of the
universe, this time codifies intrinsic features of spacetime by means of the associated
asymptotic states;
(iii) the asymptotic states are characterized by spectra of observables;
(iv) the geometric structure of the initial singularity gives rise to observable effects
which can be detected at later times;
(v) the space slices of any global time, which plays the role of the age of the universe,
are not necessarily smoothly embedded into spacetime.

2. The cosmological time function


For the basic notions of Lorentzian geometry and causality we refer, for instance,
to [4,13]. Let N be any time oriented Lorentzian manifold of dimension n + 1. The
cosmological time function, : N (0, ], is defined as follows. Let C (q) be the set of
past-directed causal curves in N that start at q N , then


(q) = sup L(c) : c C (q) ,
where L(c) denotes the Lorentzian length of the curve c:

L(c) = (proper time).
c

(q) can be interpreted as the length of time the event q has been in existence in N . For
example, if N is the standard flat Minkowski space Mn+1 , is the constant -valued function, so in this case it is not very interesting. In [2] (see also [25]) one studies the properties
of a manifold N with regular cosmological time function. Recall that is regular if:
(1) (q) is finite valued for every q N ;
(2) 0 along every past-directed inextensible causal curve.
The existence of a regular cosmological time function has strong consequences on the
structure of N and of the constant- surfaces [2]. In particular when is regular, : N
(0, [ is a continuous function, which is twice differentiable almost everywhere, giving a
global time on N denoted by CT. Each level surface is a future Cauchy surface, so that
N is globally hyperbolic. For each q N there exists a future-directed time-like unit speed
geodesic ray q : (0, (q)] N such that:




q (q) = q,
q (t) = t.
The union of the past asymptotic end-points of these rays can be regarded as the initial
singularity of N .
The cosmological time function is not related to any specific choice of coordinates in N ;
it is gauge-invariant and so it represents an intrinsic feature of spacetime. Thus, when

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

333

the cosmological time is regular, the -constant level surfaces and their properties have a
direct physical meaning as they are observables.
We present now two basic examples of spacetime with regular cosmological time, which
shall be important throughout all the paper. To fix the notations, the standard Minkowski
space M2+1 is endowed with coordinates x = (x 1 , x 2 , x 0 ), so that the metric is given by
ds 2 = (dx 1 )2 + (dx 2)2 (dx 0 )2 . M2+1 is oriented and time-oriented in the usual way.
Example 1. Consider the chronological future of the origin 0 M2+1

 2  2  2

I + (0) = x M2+1 : x 1 + x 2 x 0 < 0, x 0 > 0 .
Its cosmological time, : I + (0) (0, ), is a smooth submersion; the constant-time
{ = a} surfaces are the (upper) hyperboloids
 2  2  2


I(a) = x M2+1 : x 1 + x 2 x 0 = a 2 , x 0 > 0 .
Hence I(a) is a complete space of constant Gaussian curvature equal to 1/a 2 , and of
constant extrinsic mean curvature 1/a. The Lorentzian length of the time-like geodesic
arc connecting any p I + (0) with 0 equals (p); 0 is the initial singularity. Note that
I(a) can be obtained from I(1) by means of a dilatation in M2+1 with constant factor a;
shortly we write I(a) = aI(1). We shall denote by SO(2, 1) the group of oriented Lorentz
transformations acting on M2+1 and by ISO(2, 1) the Poincar group. SO+ (2, 1) denotes
the subgroup of SO(2, 1) transformations which keep I + (0) and each I(a) invariant.
ISO+ (2, 1) is the corresponding subgroup of ISO(2, 1).
Example 2. Let us denote by I + (1, 3) the chronological future in M2+1 of the line {x 1 =
x 0 = 0}

 2  2

I + (1, 3) = x M2+1 : x 1 x 0 < 0, x 0 > 0 .
The Lorentzian length of the time-like geodesic arc connecting any p = (x 1 , x 2 , x 0 )
I + (1, 3) with q = (0, x 2 , 0) equals the cosmological time (p). The level surfaces are
 2  2


I(1, 3, a) = x M2+1 : x 1 x 0 = a 2, x 0 > 0
and have constant extrinsic mean curvature equal to (1/2a). Each surface I(1, 3, a) is
isometric to the flat plane R2 . To make this manifest, it is useful to consider the following
change of coordinates. Let 2+1 = {(u, y, ) R2+1 : > 0} be endowed with the metric
ds 2 = 2 du2 + dy 2 d 2 . Then, x 1 = sh(u), x 2 = y, x 0 = ch(u), is an isometry
between 2+1 and I + (1, 3). The level set { = a} of 2+1 goes isometrically onto
I(1, 3, a), so this is intrinsically flat. Note that the group of oriented isometries of 2+1 is
generated by the translations parallel to the planes { = a}, and the rotation of angle of
the (u, y) coordinates.
We are going to show that any maximal globally hyperbolic, matter-free (2 + 1)spacetime M, with compact space S, actually has regular cosmological time, and its initial
singularity can be accurately described.

334

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

3. Flat (2 + 1)-spacetimes
A flat spacetime is, by definition, locally isometric to the Minkowski space M2+1 .
We assume that our maximal hyperbolic flat spacetimes are time-oriented and future
expanding, and that these orientations locally agree with the usual ones on M2+1 . The
spacetime structures on S R are regarded up to oriented isometry homotopic to the
identity.
3.1. Minkowskian suspensions
We introduce here the simplest (2 + 1)-spacetimes with compact space S of genus g > 1.
Recall that the upper hyperboloid I(1) M2+1 , mentioned in the previous section, is a
classical model for the hyperbolic plane H2 (see [7] for this and other models); the Poincar
disk is another model which can be obtained from I(1) by means of the stereographic
projection shown in Fig. 1. We shall use the Poincar model in Section 4.
Take any hyperbolic surface F = H2 / homeomorphic to S. is a subgroup of
SO+ (2, 1) which acts freely and properly discontinuously on H2
= I(1). I(1) can be
identified with the universal covering of S and with the fundamental group 1 (S). can
be thought also as a group of isometries of the spacetime I + (0) and M(F ) = I + (0)/ is
the required spacetime with compact space homeomorphic to S. We call it the Minkowskian
suspension of F . This construction is well-known; sometimes M(F ) is also called the
Lorentzian cone over F or the Lbell spacetime based on F . I + (0) can be regarded as the
universal covering of M(F ). Let us now consider the cosmological time of M(F ). The CT
of I + (0) naturally induces the CT of M(F ). Indeed, each level surface Sa of M(F ) has
I(a) as universal covering; moreover, S1 = F and Sa = aF . In this case, the CT coincides
with the CMC time and each level surface Sa smoothly embeds into M(F ). The initial
singularity trivially consists of one point.
 its suspension in I + (0)
Notation. Let Y be any subset of I(1), we shall denote by Y


which is defined by Y = a(0,[ aY .

Fig. 1. The hyperbolic plane in the hyperboloid and disk models.

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

335

3.2. (2 + 1)-spacetimes as deformed Minkowskian suspensions


It has been shown in [16] that any maximal globally hyperbolic, future expanding flat
spacetime M with compact space homeomorphic to S, as above, can be regarded as a
deformation of some Minkowskian suspension (see also [26]). In fact M is of the form
M = U (M)/  , where:
(1) The domain U (M) of M2+1 is a convex set


U (M) = x M2+1 ; x 0 > f (x) ,
where f : {x 0 = 0} [0, [ is a convex function.
(2)  is a subgroup of ISO+ (2, 1) (also called the holonomy group of M) acting freely
and properly discontinuously on U (M). Hence U (M) is the universal covering of M and
 is isomorphic to 1 (M)
= 1 (S).
(3) The linear part of  is a subgroup of SO+ (2, 1) which is isomorphic to 1 (S)
and acts freely and properly discontinuously on I(1)
= H2 . This is a nontrivial fact which
follows from a result of Goldman [12]. Each element   is of the form  = + t ( ),
where and t ( ) R3 is a translation. t : R3 is a cocycle representing an element of H 1 (, R3 ). If t = t, R , then U (M ) differs from U (M) by: U (M) =
1 U (M ). When is small, U (M ) is close to I + (0) (M is close to M(F ), F =
H2 / ).
Note that  , whence U (M) and t, are defined up to inner automorphism of ISO+ (2, 1).
3.3. Spacetimes of simplicial type
In this section, we shall consider the flat spacetimes that can be obtained from
Minkowskian suspensions by means of particular deformations. These spacetimes will be
called of simplicial type, the origin of this name is related to the material presented in
Section 4. Spacetimes of simplicial type are important because they are dense in the
set of all spacetimes; the shape of any spacetime and of its CT can be arbitrarily well
approximated by some spacetime of simplicial type (see Proposition 4.23). So, it is enough
to understand these examples in order to have a rather complete qualitative picture of our
general presentation. Moreover, all the statements of this paper can easily be checked in a
spacetime of simplicial type.
Start with a Minkowskian suspension M(F ). Assume that a weighted multi-curve L on
F is given. L is the union of a finite number of disjoint simple closed geodesics on F ,
each one endowed with a strictly positive real weight. L governs a specific deformation of
M(F ) producing a required flat spacetime denoted by M(F, L). A particular spacetime
deformation is associated to each component of L and can be obtained by means of
an appropriate surgery operation in Minkowski space. As the deformations associated to
the components of L act locally and independently from each other, we may assume for
simplicity that L just consists of one component c, with weight r and length s.

336

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

3.3.1. Elementary deformation


In order to illustrate the deformation associated with one geodesic c with weight r,
we shall now introduce a simple hyperbolic surface F0 which can be understood as a
local model for the general surface S. Let 0 be an infinite cyclic subgroup of SO+ (2, 1)
generated by an element g0 acting on I(1) as an isometry of hyperbolic type (see, for
instance, [7] for the classification of the isometries of H2 ). We can assume that g0 is
a Lorentz transformation corresponding to a boost along the x 1 -direction, so that the
g0 -invariant geodesic line on I(1) is the line 0 = I(1) {x 2 = 0}. The hyperbolic surface
F0 = I(1)/0 is homeomorphic to the noncompact annulus S 1 R and its area is not finite.
The image in F0 of the axis of g0 is a simple closed geodesic c of a certain length s; give it
the positive weight r. So we dispose of a one-component weighted multi-curve L0 on F0 ,
as illustrated in Fig. 2.
The suspension M(F0 ) = I + (0)/0 is a flat spacetime. Let us now construct M0 =
M(F0 , L0 ) which represents the deformation of M(F0 ) associated to the weighted multicurve L0 . We shall use the spacetimes I + (0), I + (1, 3) and 2+1 that we have introduced
in Section 2. The universal covering U (M0 ) of M0 will be the union of three domains of
M2+1 : U (M0 ) = A B C, where A = I + (0) {x 2  0}, B = I + (1, 3) {0  x 2  r},
C = C  + r(0, 1, 0) and C  = I + (0) {x 2  0}. In our notations, C  + r(0, 1, 0) denotes the
set of points in M2+1 which can be obtained from C  by means of a translation of length
r along the unit vector (0, 1, 0). It is important to note that the cosmological times of the
different pieces A, B and C fit well together; in fact, the CT level surfaces 
Sa of U (M0 ) are


 



Sa = aI(1) x 2 < 0 I(1, 3, a) 0  x 2  r




aI(1) x 2 > 0 + r(0, 1, 0) .

(1)

As shown in Fig. 3, each surface 


Sa M2+1 can be obtained by cutting the hyperboloid
I(a) along a0 (which is the intersection of I(a) with the {x 2 = 0}-plane) and then by
inserting a band of I(1, 3, a) of depth r.
The surfaces 
Sa are only C 1 -embedded into U (M0 ). The initial singularity of U (M0 ) is
the segment J0 = {x 1 = x 0 = 0 0  x 2  r}.

Fig. 2. The surface F0 with the closed geodesic c.

Fig. 3. Level surfaces 


Sa of U (M0 ).

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

337

Fig. 4. Level surface of M0 .

Remark 3.1. We have the following characterization of J0 . The interior points of this
segment make the subset of U (M0 ) (boundary of the convex set U (M0 )) of the points
with exactly two null supporting planes; the end-points make the subset of U (M) with
more than two null supporting planes. Recall that a supporting plane at x U (M0 ) is a
plane P such that x P and U (M0 ) P = . P is null if it contains some null-lines.
The covering U (M0 ) M2+1 is flat. To get M0 , we only need to specify the action of
1 = 1 (F0 )
= Z on U (M0 ).
3.3.2. Action of the fundamental group
1 acts on A by the restriction of the action of 0 on I(1). The domain B corresponds
(via the isometry established in Section 2) to B  = {(u, y, ) 2+1 ; 0  y  r}, so
that the action of 1 on B transported on B  is just given by the translation (u, y, )
(u + s, y, ). Finally, if is the translation (x 1 , x 2 , x 0 ) (x 1 , x 2 + r, x 0 ) on M2+1 , then
the action of 1 on C is just the conjugation of 0 by .
The CT of the covering U (M0 ) passes to the quotient M0 = U (M0 )/1 ; each level
surface Sa is only C 1 -embedded into M0 , so that it is endowed with an induced
C 1 -Riemannian metric. This allows anyway to define the length of curves traced on the
surface Sa and the derived length-space distance. Let A = A/1 , B = B/1 and C = C/1 .
Then, Sa B is a flat annulus of depth r and parallel geodesic boundary components
of length as; Sa (A C) can be isometrically embedded into aF0 , and has geodesic
boundary curves of length as. As shown in Fig. 4, Sa can be obtained by cutting F0 along
c and by inserting a annulus of depth r.
Remark 3.2. If g is an element of ISO+ (2, 1) acting on X M2+1 as g(X) = QX + w,
the transformed domain g(U (M0 )) = Q(A B C) + w is, of course, an isometric copy
of the universal covering in M2+1 . The curve = Q(0 ) is a geodesic line of I(1); is the
intersection of I(1) with a suitable hyperplane passing at the origin of M2+1 . Let us denote
by the suspension of ; then

Q(B) =
{ + v},
[0,r]

where v is the unitary (in the Minkowski norm) vector tangent to I(1), normal to , and
pointing towards Q(C  ). We also denote Q(B) = B(, v, r). The shape of the CT level
surfaces in g(U (M0 )) is shown in Fig. 5. The initial singularity of g(U (M0 )) is given by
the space-like segment J = Q(J0 ) + w.

338

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

Fig. 5. Level surfaces of g(U (M0 )).

3.3.3. Simplicial type deformation


M0 represents a local model of the deformation M = M(F, L) we are interested in.
In fact, there exists a neighborhood W of B in M0 which embeds isometrically into M,
respecting the cosmological time. Let us denote by W  the image of W in M. Then
M\W  embeds isometrically into the Minkowskian suspension M(F ), respecting again
the cosmological time.
We describe now the universal covering U (M) M2+1 and a cocycle t : R3 which
leads to  I SO + (2, 1) such that M = U (M)/  . The inverse image of c F =
 of disjoint complete
I(1)/ into the covering I(1) is an infinite and locally finite set L
 then L
 = { = (0 ); }. Let L
 I + (0)
geodesic lines. Given any geodesic 0 L,
 The set I(1)\L
 is the union of an infinite
be the suspension of the geodesic lines of L.
 its
number of connected components. Denote by R any such a component, and by R
+

suspension, which is a component of I (0)\L. Every R covers a component FR of F \L;
more precisely, if R is the subgroup of which keeps R invariant, then FR = R/R .
Now, fix one base component R0 and take in it one base point x0 . For each , let
(x0 ) be the point in I(1) which is defined by the action of on x0 . The geodesic arc in
 At each
I(1) connecting x0 with (x0 ) crosses a finite number of lines {i } belonging to L.
2+1
crossing consider the unitary (in the norm of M ) vector vi tangent to I(1) and normal
to i , pointing far from x0 . Then, the required cocycle t ( ) R3 is given by

t ( ) =
rvi .
i

Note that if 1 (x0 ) and 2 (x0 ) belong to the same component R, then t (1 ) = t (2 ),
whence also t (R) = t ( ) for any such that (x0 ) R, is well defined. U (M) is tiled by
 + w, (ii) B(, v, r) + w, for some translation vector w R3 .
tiles of two types: (i) R
More precisely, the tiles of the first type make the open subset of U (M)


 + t (R) .
R=
R
R

 is in the boundary of two regions R , R and we assume that R is


Each line L
closer to x0 than R . Set v the unitary (in the norm of M2+1 ) vector tangent to I(1)
 + t (R ) and R
 + t (R ) are
and normal to , pointing towards R . The two regions R
connected by the tile of the second type B(, v , r) + t (R ), so that


U (M)\R =
B(, v , r) + t (R ) .

L

Note that each tile has its own CT; all the cosmological times fit well together and define
the CT of U (M) which passes to the quotient M.

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

339

Remark 3.3. The construction of M(F ) and of M(F, L) can be performed for any
hyperbolic surface F , not necessarily compact nor of finite area. Similarly, by starting from
any locally finite family of weighted geodesic lines in I(1), the simplicial deformation that
we have just described produces a globally hyperbolic spacetime structure on R2 R with
a regular cosmological time.
Remark 3.4. When F is compact, every region R (defined above) is bounded by infinitely
 In fact, as F is compact, every with = 1 is of hyperbolic type [7].
many lines of L.
 and for every with ( ) = , and ( ) are ultraConsequently, for any L
parallel. This means that the hyperbolic distance satisfies d(, ( )) > 0; moreover,
and ( ) have a common orthogonal geodesic line. Suppose now that a region R is
 In this case, R contains a band E of infinite diameter,
bounded by finitely many lines in L.
 As F is compact, H2 is tiled by tiles of the form
bounded by two half-lines contained in L.
(D), where D is a fundamental domain for of finite diameter. So, one (at least) tile
 = and this contradicts the fact that
(D) must be contained in E. But clearly (D) L

R is a region of H2 \L.
The same conclusion holds if F is of finite area. If F is of infinite area, we can eventually
0 just consists of
have different behaviours. For instance, in the example F0 above, L
2
one component which divides H into two regions. Other examples will be presented in
Subsection 4.1.
4. The cosmological time of (2 + 1)-spacetimes
In this section we describe the main properties of the CT for an arbitrary spacetime M.
We adopt the notations of the previous sections; in particular, M is assumed to be an
expanding matter-free spacetime of topological type S R with compact surface S of
genus g > 1. The validity of the following statements can be quite easily checked for
spacetimes of simplicial type. We shall try to point out the main ideas; we postpone a
commentary on the proofs, with references to the existing literature.
Proposition 4.1. The cosmological time function, : M (0, [, is surjective and
regular, so that it defines a global time on M. It lifts to a regular cosmological time on
the covering, : U (M) (0, [. Each level surface 
Sa of U (M) maps onto Sa of M and
is its universal covering. In other words, the action of 1 (S) on U (M) restricts to a free,
properly discontinuous isometric action on 
Sa such that Sa = 
Sa /1 (S). Each 
Sa (Sa ) is a
future Cauchy surface.
4.1. Initial singularityexternal view
Let us give a description of the initial singularity of M as it appears from the exterior
point of view, that is, from the Minkowski space in which the universal covering U (M) is
placed. In Subsection 4.5 we shall show how the initial singularity can also be characterized
in terms of the observables associated with the CT asymptotic states.

340

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

Let us first give a definition.


Definition 4.2. An R-tree (also called a real tree) is a metric space (T , d) such that for
each couple of points p = q T there exists a unique arc in T with p and q as end-points
and this arc is isometric to the interval [0, d(p, q)] R. This arc is called a segment of T
and is denoted [p, q].
Remark 4.3. The so-called simplicial trees are the simplest examples of real trees.
A simplicial tree is a real tree covered by a countable family of elementary segments,
called the edges of the tree, in such a way that: (a) whenever two edges intersect, then
they just have one common endpoint; (b) the edge-lengths take values in a finite set of
strictly positive numbers. Any endpoint of any edge is called a vertex of the tree. The
distance is the natural length-space distance. Note that a simplicial tree is not necessarily
locally finite; in other words, vertices of infinite valence may occur. In general, a real tree
is more complicated than a simplicial tree because one might find, for instance, a segment
containing a Cantor set made by the endpoints of other segments.
Proposition 4.4. For any p U (M) M2+1 there is a unique time-like geodesic arc
a(p) contained in U (M), which starts at p and is directed in the past of p, such that the
Lorentzian length of a(p) equals (p). The other end-point of a(p), denoted by i(p), belongs to the boundary U (M) of U (M) in M2+1 . If p and q are identified in M by the action of 1 (S), so are a(p) and a(q). The union of the initial points T = {i(p); p U (M)}
is an R-tree. More precisely, each segment of T is a rectifiable space-like curve in U (M)
with its own length. There is a natural isometric action of the fundamental group 1 (S)
on T . The quotient space i(M) = T /1 (S) can be thought as the initial singularity of M.
Remark 4.5. We have already encountered several examples of spaces of the form X =
 1 (S) for some action of 1 on X:
 for instance, F = H2 / , M = U (M)/  , Sa =
X/

Sa /  . Now, the initial singularity of spacetime also is a quotient i(M) = T /1 (S). Instead
 endowed with
of the bare topological quotient space, it is more interesting to consider X
the action of 1 .
Remark 4.6. When M is of simplicial type, the corresponding real tree T is actually a
simplicial real tree. This justifies the name we have given to these special spacetimes. In
this case, the set of edges of T consists of the union of the space-like segments which form
the initial singularity of the different tiles of the form B(, v , r) + t (R ) (see Section 3).
The points of T can also be characterized by the properties discussed in Remark 3.1.
A homeomorphic (not isometric) copy of T can easily be embedded into I(1). Select one
 and consider the set made by the union of all these points.
point in each region of I(1)\L
Connect two points of this set by a geodesic segment of I(1) if and only if they belong to

adjacent regions. In this way we get the required tree. This tree is manifestly dual of L;


in fact, the regions of I(1)\L correspond to the vertices of T and the lines of L correspond
to the edges of T . We shall return on this duality in Section 4.3. Note that, as demonstrated
in Remark 3.4, all the vertices of T are of infinite valence.

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

341

4.1.1. Examples of real trees


A hyperbolic surface F = H2 / is represented in Fig. 6. F is of infinite area and is
homeomorphic to a torus with one puncture; two simple closed geodesics c and a on F are
depicted. The geodesic c cuts open F into a compact surface and an infinite area annulus.
By using the Poincar disk model, a fundamental domain of in H2 is also shown in
Fig. 6. This domain is delimited by four pair-wise ultra-parallel geodesic lines. The inverse
images of c and a are represented on this domain. The first two terms of a sequence of
partial tilings of H2 , made by a finite number of copies of the fundamental domain, are
shown in Fig. 7. The first partial tiling just contains one fundamental domain. The second
is made by the union of 1 + 4 = 5 copies of the fundamental domain. The next partial
tiling of this sequence, which is not shown in the figure, contains 1 + 4 + 12 = 17 tiles,
and so on. For each partial tiling of H2 one can determine a corresponding partial lifting
of the curves c and a. Fig. 8 shows the first two partial liftings c of c and the structure
of the associated partial dual trees. In the limit of the complete infinite tiling of H2 , the
complete lifting of c contains an infinite number of geodesics and the associated real tree
is infinite. In this case, H2 \c has exactly one component with infinitely many boundary
lines (the associated vertex of the dual tree has infinite valence), whereas all the remaining
components have one boundary line. The first three partial liftings a of a are shown in
Fig. 9; the corresponding partial dual trees are also represented. Note that these figures are
just evocative, as they are not geometrically exact.
Remark 4.7. The R-trees and the associated 1 (S)-actions which occur in Proposition 4.4
are not arbitrary (see [20] p. 32). In fact, one can prove that the 1 (S)-action is minimal
with small edge-stabilizers. This means that there is no nonempty strictly subtree which
is invariant for this action, and that, for each segment in the tree, the subgroup of 1 (S)
which keeps the segment invariant is virtually Abelian. We shortly say that a real tree which
admits such a kind of 1 (S)-action, is geometric.

Fig. 6. Two simple curves on F = H2 / .

 = H2 .
Fig. 7. Partial tilings of F

342

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

Fig. 8. Partial c and dual trees.

Fig. 9. Partial a and dual trees.

4.2. Intrinsic and extrinsic geometry of the constant CT surfaces


In order to describe the geometric properties of the surfaces of constant cosmological
time, it is natural to introduce the notion of geodesic lamination.
Definition 4.8. Let G be a surface endowed with a C 1 -Riemannian metric. As usual, this
induces a length-space distance on G and the notion of geodesic arc (line) makes sense.
A geodesic lamination of G is a closed subset K of G, also called the support of the
lamination, which is the disjoint union of complete and simple geodesics, also called the
leaves of the lamination. Complete means that we dispose of arc-length parametrization
defined on the whole real line R; simple means that the geodesic has no self-normal
crossing in G. In other words, each leaf is either a simple closed geodesic or a simple
geodesic which is an isometric copy of R embedded in G. When G is compact, such a
noncompact leaf is not a closed subset of G.
Remark 4.9. A finite union of disjoint simple closed geodesics is called a multi-curve and
is the simplest example of geodesic lamination. We have already introduced multi-curves
in Section 3.3. A generic geodesic lamination K can be more complicated than a multicurve; in fact, if is an arc embedded in G which is transverse to the leaves of K, typically
\K is a Cantor set.
Proposition 4.10. For every a (0, [:
(1) 
Sa is the graph of a positive convex function defined on the plane {x 0 = 0} in M2+1 .
(2) 
Sa is only C 1 -embedded into U (M), so that it carries an induced C 1 -Riemannian
Sa , there is a unique geodesic
metric. 
Sa is geodesically complete and for each p = q 
arc connecting p and q.
a at which the embedding of 
Sa into U (M) is no longer C 2 is a geodesic
(3) The locus L
La is in fact the pull-back of a geodesic lamination La of Sa .
lamination of 
Sa . 

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

343

Remark 4.11. If M is the spacetime of simplicial type which corresponds to the weighted
multi-curve L on the surface F , then La is just made by the boundary components of the
flat annular components embedded into Sa , which are associated to the components of L.
The content of the last remark generalizes as follows. Recall that 1 (S) acts as on
each I(a). For every a (0, [, let us consider the map
Sa I(a)
pa : 
defined as follows: pa (x) is the unique point of I(a) such that the tangent plane to 
Sa
at x is parallel to the tangent plane to I(a) at pa (x). This map is well-defined, surjective
and 1 (S)-equivariant. By taking the union of the pa s we get a 1 (S)-equivariant map
p : U (M) I + (0), respecting the CT. This induces a map p : M M(F ) respecting the
CT.
Proposition 4.12.
(1) There exists a geodesic lamination F on F = I(1)/ , which lifts to a geodesic
 and any leaf of 
 on I(1), such that, for every a, one has pa (
La ) = a F
La
lamination F


is isometrically mapped onto a leaf of a F . That is, the union of pa (La )s covers the
 of F
.
suspension F
(2) F is the disjoint union of two sublaminations
F = L F ,
where L is the maximal multi-curve sublamination of F . Note that either L or F  may be
empty. Then
) isometrically into I + (0) respecting the CT;
(a) p embeds U (M)\p1 (F
1

(b) p embeds U (M)\p (L) continuously into I + (0) respecting the CT.
 is the union of components of the type B(, v , r) + w, so that
(3) The set p1 (L)

1
(p ) (L) Sa is the disjoint union of flat annular components of Sa , like in the case of a
spacetime of simplicial type.
We have an immediate corollary concerning the intrinsic and extrinsic geometry of the
constant CT surfaces.
a = 
a is either
Corollary 4.13. W
Sa \
La is an open dense set of 
Sa . Each component of W
isometric to an open set of I(a) or is a flat band which embeds into I(1, 3, a), and projects
onto an annulus of Sa . Flat annuli do occur only if L is nonempty.
4.3. CT duality
To sum up, two geometric structures are naturally associated to the spacetime M: the
real tree T (the initial singularity) and the geodesic lamination F on F = I(1)/ which
reflects the lack of smoothness of the embedding of Sa into M. We have already noted that
for a spacetime of simplicial type these two objects are dual to each other. Here we want
to strengthen and generalize this point.

344

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

If L is nonempty, we extend the lamination La on Sa to a lamination La , by foliating


the flat annular regions of Sa by closed geodesics parallel to the boundary components. As
.
a denotes the lifted lamination to 
Sa . The above map pa sends 
La onto a F
usually L

We have a natural continuous surjective map ia : Sa T which associates to x the initial
Sa by closed subsets. 1 (S)
point on the arc a(x). So T  = {ia1 (x)}xT is a partition of 
acts also on T  and, clearly, ia induces an 1 -equivariant identification between T  and T .
Sa is:
Proposition 4.14. For every a, each closed set E of the partition T  of 



(1) either the closure of a component of Sa \La ;
(2) or a leaf of the foliation of some band component of 
La which projects onto a flat
annular region of Sa .
We describe how the distance d on the real tree T can be encoded, in dual terms, by
, F , with suitable transverse invariant measures.
equipping the geodesic laminations F
Definition 4.15. A measured geodesic lamination on F is a couple (F , ), where F is
a geodesic lamination and is a transverse invariant measure which consists of a Borel
measure J on each embedded interval J
= [0, 1] in F , transverse to the leaves of F such
that
(1) the support of J coincides with F J ;
(2) if J, J  are arcs, homotopic through arcs which are transverse to the leaves of F ,
keeping the endpoints either on the same leaf or in the same connected components of
, )
which is 1 -equivariant.
F \F , then J (J ) = J  (J  ). (F , ) lifts to (F
Remark 4.16. The simplest measured geodesic laminations of F are the weighted multicurves.
. The map pa lifts J to an arc J 
Let J be an arc in I(1) transverse to the leaves of F

in 
Sa transverse to the leaves of 
La . On the other hand, by means of the map ia , the distance
d on T lifts to a measure J  on J  which finally gives us the required (1 -equivariant)
.
transverse measure on F
One can invert the above construction and associate to each measured geodesic
lamination (F , ) of the hyperbolic surface F a suitable geometric R-tree.
4.3.1. From geodesic laminations to real trees
Take the measured lamination (F , ) of the surface F . F is in general the disjoint union
of two sublaminations
F = L F  ,
where L is the maximal weighted multi-curve sublamination of F . Note that either L or
F  may be empty. F \F consists of a finite number of connected components, the metric
completion of any such a component is isometric to a compact hyperbolic surface with
geodesic polygonal boundary. If L is nonempty, let us consider the spacetime of simplicial

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

345

type associated to L, and let F  be the = 1 level surface of this spacetime. Let us denote
by F  the lamination on F  which coincides with F outside the flat annuli of F  and is
defined as L1 above on these annuli. If L is empty, set F  = F . The measured lamination
(F , ) extends to a measured lamination (F  ,  ) on F  . The flat annular components
are foliated by closed geodesics parallel to the boundary components. These annuli are
endowed with a plain transverse measure of total mass equal to the corresponding annulus
 of F  with the lifted (1 -equivariant) measured
depth. Take the universal covering F



 by closed subsets in the very
geodesic lamination (F , ). Now define a partition of F

S1 , with respect to the lamination 
L1 .
same way we have defined above the partition T of 
Call again this partition T  . We can give it a distance d which makes it an R-tree. If E and
E  are the closure of two components of the complement of the lamination, take two points
 is transverse to
x and x  in these closed sets such that the geodesic segment [x, x  ] of F
the leaves of the lamination. By integration, the transverse measure induces a distance on
the subset of T  made by the closed sets intersecting [x, x  ]. In fact, by the invariance of
the measure, this distance does not depend on the segment [x, x  ]. Finally one verifies that
in this way one can actually define a distance between any two points of T  and that the
resulting (T  , d) is a geometric real tree.
Remark 4.17. Clearly, weighted multi-curves on the surface F dually correspond to
geometric simplicial real trees; the spacetimes of simplicial type do materialize this duality.
4.4. Reconstruction of M = U (M)/ 
Starting from (F = I(1)/, T ) or, equivalently, from (F = I(1)/, (F , )), one can
reconstruct a cocycle t, whence M = U (M)/  . This generalizes what we have done for
a spacetime of simplicial type in Subsection 3.3.
 ,  )
With the notations introduced at the end of the previous subsection, consider (F





on F . To recover a cocycle t do as follows: fix one base point p0 on F out from the
support of the lamination. Let p0 be its image on F  . If is a loop in F  based on p0 , which
 which starts at
represents an element [ ] of (F  , p0 ), lift it to the oriented arc in F
p0 ; up to homotopy we can assume that is transverse to the leaves of the lamination.
Let f be any continuous R3 -valued function on which coincides with the unit normal
 , and oriented in agreement with . Now
to the leaves of the lamination, tangent to F

we can integrate f along by using the transverse measure getting a vector t ([ ]). By
varying [ ], one gets such a cocycle t.
4.5. CT asymptotic states
The above discussion tells us that any spacetime M = U (M)/  is completely
determined by the linear part of its holonomy  (or equivalently by the surface F =
I(1)/ ) and by its initial singularity i(M) = T /1 (S). The aim of this subsection is to
recover these geometric objects from the internal point of view by working inside the
spacetime. More precisely, we would like to show that F and T can be interpreted as

346

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

the future and past asymptotic states for the geometry of the CT level surfaces. To this
aim we shall consider the observables defined by the lengths of the curves on the CT level
surfaces. It is convenient to introduce the concept of Marked Spectrum associated with a
 d) which is endowed with an action of the surfaces fundamental group
metric space (X,
 1 (S). Whenever we shall refer to X, we shall actually refer to the
1 (S), so that X = X/
 d, ) (see Remark 4.5).
triple (X,
Let us denote by C the set of conjugation classes of 1 (S)\{1} which coincide with the
homotopy classes of noncontractible continuous maps f : S 1 S. Each marked spectrum
is a point of the functional space (R0 )C , endowed with the natural product topology. The
Marked Spectrum sX of X (denoted also by sX
 ), is defined as follows: for any c = [ ] C,
1 (S), = 1,


sX (c) = inf d p, (p) .

pX

The spectrum is marked because one takes track of the map in addition to its image.
When X = F or Sa , sX (c) is just the length of a closed geodesic curve (not necessarily
simple; that is, self-crossings could possibly occur in c) which minimizes the length among
the curves in that homotopy class. For this reason, in such a case, sX is called the Marked
 = T , sT can be expressed, in dual terms,
Length Spectrum and is denoted by lX . When X
as the Marked Measure Spectrum of the corresponding measured geodesic lamination F
on F ; usually this is denoted IF . IF (c) is just the minimal transverse measure realized
by the curves in that homotopy class. When T is simplicial, that is when F is a weighted
multi-curve L of F , IL (c) is easily expressed in terms of the geometric intersection number
(this also justifies the notation): assume that all the weights are equal to 1 (that is the length
of all edges of T is equal to 1); then it is easy to see that IL (c) is just the minimum
number of intersection points between L and any curve belonging to c and transverse to
the components of the lamination. For arbitrary weights one just takes multiples of the
contribution of each component of L.
Remark 4.18. Instead of the whole C, one could prefer to use the subset S C of isotopy
classes of simple closed curve in S, and take the corresponding (restricted) marked spectra.
The discussion should proceed without any substantial modification.
4.5.1. On the boundary of the Teichmller space
It is convenient, at this stage, to recall the fundamental facts about the role that the
marked spectra play in the study of the Teichmller space and in Thurstons theory of its
natural boundary. Let us denote by Tg the Teichmller space of the hyperbolic structures
on S up to isometry isotopic to the identity. It is well-known (see [7,11,23]) that the map
l : Tg (R0 )C
defined by l(F = H2 / ) = lF , realizes a meaningful embedding of Tg onto a subset of
(R0 )C homeomorphic to the finite-dimensional open ball B 6g6 . We shall identify Tg
with l(Tg ). In fact Tg is in a natural way a real analytic submanifold of (R0 )C .

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

347

Fix any such a hyperbolic structure F Tg on S. Let us denote by MGL(F ) the


set of measured geodesic laminations on F . Let us denote by GT (S) the set of all
1 (S)-geometric R-trees (Remark 4.7). At the end of Subsection 4.3, we have outlined
a construction which associates to each F MGL(F ) a dual R-tree say (F ) GT (S).
Note that this construction did not use the fact that F was associated to a spacetime M.
Proposition 4.19. : MGL(F ) GT (S) is a bijection, that is it can be naturally
inverted. For each r > 0, (rF ) = r(F ); here we take either the r-multiple of the
measure or the r-multiple of the distance. We can shortly say that respects the positive
rays.
Proposition 4.20. Consider the maps, I : MGL(F ) (R0 )C and s : GT (S) (R0 )C ,
obtained by taking the corresponding marked spectra. Then I = s and is injective. The
image in (R0 )C is a positive cone based on the origin and positive rays go onto positive
rays, in a obvious sense. Moreover, Tg and the image Im(I ) are disjoint subsets of (R0 )C .
Remark 4.21. These spectra represent the actual physical observables in our discussion.
The last two propositions specify the meaning of the duality between laminations and real
trees. As the spectra coincide, they reveal the same physical content.
Similarly to Tg , we identify MGL(F ) and GT (S) with the image Im(I ) (R0 )C ,
endowed with the subspace topology.
Set P + (MGL(F )) = P + (GT (S)) = P + (Im(I )) the projective quotient space, obtained
by identifying to one point each positive ray in Im(I )\{0}. Similarly Tg P + (Im(I )) has
a natural quotient topology.
Proposition 4.22. The pair (Tg , Tg ) = (Tg P + (Im(I )), P + (Im(I ))) is homeomorphic
6g6 , S 6g7 ), where B
6g6 is the closed ball and S 6g7 is its boundary
to the pair (B
sphere. The natural action on Tg of the mapping class group Modg of S extends to an
action on the compactification Tg . This is called the Thurstons natural compactification
and Tg is the natural boundary of the Teichmller space.
We can state precisely how the simplicial trees are dense, as we claimed in Section 3.
Let us denote ST (S) the subset of GT (S) made by the simplicial real trees.
Proposition 4.23. ST (S) is dense in GT (S) in the induced topology by (R0 )C .
Remark 4.24. In this remark we collect a few technical complements concerning the
marked spectra and the geometric meaning of spectra convergence.
(1) The natural compactification of Tg is formally similar to the natural compactification
1 is obtained by adding to I(1) the endpoints
of H2 in the hyperboloid model I(1) where S
of the rays of the future light cone.

348

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

(2) Let Fn be a sequence in Tg considered as a sequence of actions of 1 (S) on H2 .


The meaning of the compactification is the following; up to passing to a subsequence (still
denoted by Fn ), one of the following situations occur: for every c C,
(a) lFn (c) lF0 (c), for some F0 Tg .
(b) There exist a geometric real tree T GT (S) and a positive sequence >n 0, such
that >n lFn (c) sT (c). This is also called the MorganShalen convergence of the sequence
of actions. This can be reformulated in a similar, equivalent, dual way as the convergence
(up to positive multiples) of a sequence of marked length spectra of hyperbolic structures
on S to the marked measure spectrum of a measured geodesic lamination on a fixed base
F0 .
(3) The convergence of marked spectra has a deep geometric content. This can be
expressed in terms of the Gromov convergence. Given two metric spaces (Y, d) and (Y  , d  )
and > > 0, an >-relation is a set R Y Y  (i.e., a relation between the two spaces) such
that:
(a) the two projections of R to Y and Y  are both surjective;
(b) if (y, y  ), (z, z ) R then |d(y, z) d(y  , z )| < >.
Let G be a group, and {G Yn Yn }n1 be a sequence of isometric actions of G on the
metric spaces Yn . We say that (G Yn Yn ) (G Y0 Y0 ) in the sense of Gromov,
if for every compact subset K0 Y0 , for every > > 0 and for every finite subset P of G,
if n is big enough, there are compact subsets Kn Yn and >-relations Rn between Kn and
K0 which are P -equivariant; this means that: if x K0 , g P , g(x) K0 , xn , yn Kn
and (x, xn ), (g(x), yn ) Rn , then dn (g(xn ), yn )  >.
It turns out that in case (a) above we actually have the convergence in the Gromov sense
of the sequence of actions on H2 to an interior point of Tg . In case (b), the MorganShalen
convergence is equivalent to the Gromov convergence for the sequence of actions on >n H2 .
(4) Note that GT (S) is defined by using only the topology of S (its fundamental group
indeed) while in order to adopt the dual view point we have to fix (in an arbitrary way) a
base hyperbolic surface F0 Tg . In fact, the dual view point can be developed by using
the marked measure spectra of the measured (singular) foliations on S (instead of the
measured geodesic laminations on F0 ), which only depend on the differential structure of
S (see [11]). On the other hand, let us consider Tg as a space of complex holomorphic
structures on S (thanks to the classical Uniformization Theorem). By fixing any such
structure F0 , one can realize such a spectrum as the measure spectrum of the horizontal
measured foliation of a unique quadratic differential on F0 . These rigidifications (via
geodesic laminations or quadratic differentials) of softer objects (the measured foliations)
is reminiscent of the role of Hodge theory with respect to De Rham Cohomology.
4.5.2. CT asymptotic states as limit spectra
After this somewhat long but necessary digression, let us come back to the CT
asymptotic states.
Proposition 4.25. (a) lima0 lSa = sT ; (b) lima lSa /a = lF .

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

349

Remark 4.26. This means, in particular, that in a far CT future the spacetime looks like the
Minkowskian suspension M(F ). In order to detect the dual effect of the initial singularity
on the embedding of Sa into M, for large value of the cosmological time one needs to
increase the accuracy in the measurement of geometric quantities. Nevertheless, this effect
is, in principle, observable for any finite value a of the CT.
Proposition 4.27. For every a (0, [, lSa /a belongs to Tg (R0 )C . Hence, the
cosmological time determines a curve M : (0, [ Tg . This is a real analytic curve which
connects F Tg with the point on the natural boundary [T ] Tg (here [.] denotes the
projective class).
Remark 4.28. Consider a spacetime of simplicial type. To prove Proposition 4.25 in
this case, one has to note that the depth of the annular regions is constant on each Sa .
When a 0, the contribution (to the length of any curve on Sa ) of the part contained in
the nonannular components becomes negligible, the length of the annuli boundaries goes
linearly to zero, so that only the transverse crossing of the annuli becomes dominant. When
a , the annuli depth goes to zero because of the rescaling by 1/a, and the length
spectrum converge to the spectrum of F . The general case follows by using the density
stated in Proposition 4.23. Concerning Proposition 4.27, in the special case of a spacetime
of simplicial type, the curve in Tg is just given by the FenchelNielsen flow obtained by
twisting the hyperbolic surface F along the closed geodesic of the multi-curve (see [23]
and also [7]).
4.6. A commentary on the proofs
The identification between cocycles of a spacetime M with measured geodesic
laminations on F = I(1)/ is due to Mess [16]. In fact one can find other examples of
such a construction of cocycles from measured laminations in the contest of Thurstons
theory of bending or earthquakes (see for instance [10]).
Measured geodesic laminations emerged in the original Thurstons approach to the
natural compactification of Tg [2123]. See also [11] for the alternative approach by
using the measured foliations (see Remark 4.24 (4)). For the claim about the quadratic
differentials in Remark 4.24 (4)) see [14]. The dual approach via real trees is due to
MorganShalen [18,19]. This approach does apply to more general, higher-dimensional
situations. The monography [20] contains a rather exhaustive introduction to this matter
and we mostly refer to it (and to its bibliography) for all the details. In particular one can
find in [20] a complete proof of the duality (see Proposition 4.19 and Proposition 4.20). The
delicate point is just the inversion of the map we have described above. The geometric
interpretation of the MorganShalen convergence (see Remark 4.24 (3)) is due to Paulin
and Bestvina (cf. the bibliography of [20]).
It is an amazing fact that the spacetimes materialize this subtle duality in the way we
have seen. Note also that, in the spacetime setting, the choice of the base hyperbolic surface

350

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

F (see Remark 4.24 (4)) is fixed by the linear part of the holonomy of M, that is by its
future asymptotic state.
Concerning Proposition 4.27, the FenchelNielsen flow generalizes to the earthquake
flow (one uses again the density 4.23) with initial data (F, F ) which has real analytic
orbits [15,24].

5. Complements
In this section we add a few comments about the flat spacetimes with compact space
of genus g = 1, and about the spacetimes with negative cosmological constant. Finally we
discuss a conjecture relating the CT and the CMC time.
5.1. Toric space (g = 1)
The case in which the surface S is a torus is a particular example of the socalled Teichmller spacetimes which we have analysed in [5]. So we simply remind
the main points. Each nonstatic spacetime determines a curve : (0, [ T1 , (a) =
(w(a), (a)), where T1 is the cotangent bundle of the Teichmller space T1 of conformal
structures on the torus up to isomorphism isotopic to the identity. Let us recall that T1 is
isometric with the Poincar disk. The cotangent vectors (a) at the point w(a) T1 is
a quadratic differential on a Riemann surface representing w(a). It is not hard to verify
that is just the complete orbit of the Teichmller flow with initial data (w(1), (1))
(see [1,5]). In particular, the projection of onto T1 is a complete geodesic connecting
two boundary points. These points can also be interpreted in terms of marked spectra. Let
us denote by H and by V the horizontal and vertical measured foliations of w(1). Then:
lima lSa /a = IH and lima0 lSa = IV .
5.2. Spacetime with negative cosmological constant
The above discussion on CT for flat spacetimes (i.e., with cosmological constant = 0)
can be adapted to the case of negative which we normalize to be = 1. We denote
by X2+1 the Universal anti-de-Sitter spacetime of dimension 2 + 1. Each spacetime is now
locally isometric to X2+1 . The role played by I + (0) in the flat case, is played now by
the diamond-shaped domain D(2) (see [13] p. 132) isometric to B 2 (/2, /2) with
metric, in coordinates (y 1 , y 2 , t), ds 2 = (cos2 t)h2 dt 2 , where h2 is the usual Poincar
hyperbolic metric on the open disk B 2 .
5.2.1. Anti-de-Sitter suspensions
If F = H2 / is a hyperbolic surface of genus g > 1, then isometrically acts also
on D(2) and D(F ) = D(2)/ is the anti-de-Sitter suspension of F . Up to a translation,
the function t gives the CT and it has many qualitative properties similar to the CT of the
Minkowskian suspensions, but we have now both an initial and a final singularity, both
reduced to one point. In a sense, D(F ) can be obtained by the Minkowskian suspension

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

351

M(F ) by a procedure of warping and doubling; D(F ) and M(F ) have the same initial
singularity; the future asymptotic state of M(F ) becomes the level surface of the CT
on D(F ) where the expansion ends and the collapsing begins. Also the anti-de-Sitter
analogous of I + (1, 3) is easy to figure out.
5.2.2. Deforming anti-de-Sitter suspensions
We want to generalize the above warping and doubling construction. Let M =
U (M)/  as in the former flat-spacetime discussion,  = + t ( ). F = I(1)/ as
usually. For t (/2, 0), (0, ), set t = (/2)e . Denote h(a) the spatial metric
on Sa . On the manifold F (/2, 0) consider the metric ds 2 = cos2 (t)h( )/ 2
dt 2 , getting a spacetime D (M). Similarly, take M and D (M ), where M =
U (M )/  ,  = t ( ), D (M ) is obtained from D (M ) by reversing the
time and the orientation. Finally, set D(M) = D (M) D (M ), by gluing the two pieces
at t = 0. D(M) is locally anti-de-Sitter; up to a translation, t gives the CT. The asymptotic
state for t /2 (i.e., the initial singularity) is equal to the initial singularity of M.
The final singularity (t /2) coincides with the initial singularity of M . The future
asymptotic states of M and M glue at the level surface {t = 0} of the CT where the
expansion ends and the collapse begins. The orbit of D(M) in Tg is given by the union of
two earthquake rays associated to M (pointing to the future) and to M (towards the past);
note that the qualitative behaviour is similar to what we have remarked for g = 1. D(M) is
the quotient of a domain D(2)M X2+1 , which is a deformation of the diamond-shaped
domain D(2). Also in this case the spacetimes with simplicial asymptotic singularities are
significant and particularly simple to be understood.
5.3. CT versus CMC
Assume again that the space S is of genus g > 1, and that the spacetimes are flat. Given
any global time on a spacetime M = U (M)/  , the asymptotic behaviour of the geometry
of the corresponding level surfaces reflects in general a property of the specific time and
not of the spacetime. On the other hand, we have seen that the asymptotic states of the
cosmological time are intrinsic features of the spacetime. In this sense, we can say that a
global time is good when it has the same asymptotic states of the CT. The CMC time,
say, is a widely studied global time. A natural question is whether is a good global
time. We conjecture that this is the case. Let us denote by Wa the { = a} level surfaces of
the CMC time.
Conjecture 5.1. (a) lima lWa = sT ; (b) lima0 lWa /a = lF .
There are some strong evidences supporting the conjecture; in particular by [3] we know
that:
(1) is a global time function with image the interval (0, +).
(2) If : (0, ) Tg is any -orbit in Tg (here Tg is intended as a space of conformal
structures) then:

352

R. Benedetti, E. Guadagnini / Nuclear Physics B 613 (2001) 330352

(i) The lim0 exists in Tg .


(ii) is proper, that is it goes out from any compact set of Tg , roughly it goes to .
An idea to prove the conjecture, should be to confine each Wa between two barriers made
by CT-level surfaces Sa  , Sa  , in such a way that a  and a  depend nicely on a and, when
a or a 0, Sa  and Sa  become more and more geometrically close to each other.
In a recent conversation, L. Andersson confirmed that this should actually work at least for
a spacetime with simplicial initial singularity. A similar conjecture can be formulated in
the anti-de-Sitter context.

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]

W. Abikoff, The Real Analytic Theory of Teichmller Space, Springer-Verlag, Berlin, 1980.
L. Andersson, G.J. Galloway, R. Howard, Class. Quantum Grav. 15 (1998) 309322.
L. Andersson, V. Moncrief, A. Tromba, J. Geom. Phys. 23 (1997) 191205.
J.T. Beem, P.E. Ehrlich, K.L. Easley, Global Lorentzian Geometry, 2nd edn., Pure and Applied
Mathematics, Vol. 202, Dekker.
R. Benedetti, E. Guadagnini, Phys. Lett. B 441 (1998) 6068.
R. Benedetti, E. Guadagnini, Nucl. Phys. B 588 (2000) 436450.
R. Benedetti, C. Petronio, Lectures on Hyperbolic Geometry, Springer-Verlag, 1992.
S. Carlip, Phys. Rev. D 42 (1990) 2647.
S. Deser, R. Jackiw, G. t Hooft, Ann. Phys. (NY) 152 (1984) 220.
D.B.A. Epstein, A. Marden, in: D.B.A. Epstein (Ed.), Analytical and Geometric Aspects of
Hyperbolic Space, London Math. Soc. Lecture Notes Series, Vol. 111, Cambridge Univ. Press,
1987.
A. Fathi, F. Laudenbach, V. Poenaru, Travaux de Thurston sur les surfaces, Astrisque (1979)
6667.
W.M. Goldman, Inventiones Math. 93 (1988) 557607.
S. Hawking, G.F.W. Ellis, The Large Scale Structure of SpaceTime, Cambridge Univ. Press,
Cambridge.
S.P. Kerckhoff, Topology 19 (1) (1980) 23.
S.P. Kerckhoff, Comment. Math. Helvetici 60 (1985) 1730.
G. Mess, Preprint IHES/M/90/28 (1990).
V. Moncrief, J. Math. Phys. 30 (1989) 29072914.
J.W. Morgan, P. Shalen, Ann. Math. 122 (1985) 398476.
J.W. Morgan, P. Shalen, Topology 30 (1991) 143154.
J.-P. Otal, Le thorme dhyperbolisation pour les varits fibres de dimension trois,
Astrisque 235 (1996).
W.P. Thurston, Bull. Amer. Math. Soc. 6 (1982) 357381.
W.P. Thurston, Bull. Am. Math. Soc. 19 (1988) 417432.
W.P. Thurston, Geometry and Topology of 3-manifolds, Princeton Univ. Press, Princeton, 1982.
W.P. Thurston, in: D.B.A. Epstein (Ed.), London Math. Soc. Lecture Notes, Vol. 111,
Cambridge Univ. Press, 1987.
R. Wald, P. Yip, J. Math. Phys. 22 (1981) 26592665.
E. Witten, Nucl. Phys. B 311 (1988) 46.

Nuclear Physics B 613 (2001) 353365


www.elsevier.com/locate/npe

Quantum diffusion of magnetic fields in


a numerical worldline approach
Holger Gies a,b,1 , Kurt Langfeld b
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Institut fr Theoretische Physik, Universitt Tbingen, D-72076 Tbingen, Germany

Received 15 February 2001; accepted 27 July 2001

Abstract
We propose a numerical technique for calculating effective actions of electromagnetic backgrounds based on the worldline formalism. As a conceptually simple example, we consider scalar
electrodynamics in three dimensions to one-loop order. Beyond the constant-magnetic-field case,
serving as a benchmark test, we analyze the effective action of a step-function-like magnetic field
a configuration that is inaccessible to derivative expansions. We observe magnetic-field diffusion,
i.e., nonvanishing magnetic action density at space points near the magnetic step where the classical
field vanishes. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.20.-m; 11.15.Ha
Keywords: Worldline; Effective action; Monte Carlo simulation

Introduction
The worldline formalism was invented by Feynman [1] simultaneously with modern
relativistic second-quantized QED, but for a long time it was used only occasionally for
actual calculations. Eventually, the observation of a close relation between the worldline
formalism and the infinite-string-tension limit of string path integrals triggered further
developments of the worldline formalism for QCD [2] and QED [3], particularly for gauge
particle amplitudes. Certain computational advantages of this formalism were subsequently
recognized and led to numerous applications. Among them, the progress achieved for
QED amplitudes with all-order couplings to an external background field is particularly
remarkable [46]. The formalism works most elegantly for constant electromagnetic
background fields and can be extended to a derivative expansion in the electromagnetic
E-mail address: holder.gies@cern.ch (H. Gies).
1 Emmy Noether fellow.

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 1 ) 0 0 3 7 7 - 7

354

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

background [7,8]. For a comprehensive review of the worldline formalism in QED and
beyond, see [9].
In the present work, we intend to demonstrate that the worldline formalism is moreover
perfectly suited for numerical computations. This is because the path integral over closed
loops in spacetime can be approximated by a finite ensemble of loops, which allows for a
simple and fast evaluation of expectation values. The latter include observables depending
on an arbitrary background field.
We illustrate this proposal by means of a simple example: the one-loop effective action
of Euclidean scalar QED in three dimensions. The computational task boils down to the
calculation of the Wilson loop expectation value using a loop ensemble.
As a first step, we verify the method in Section 2 by considering the constant-magneticfield case, which can also be solved analytically. This serves as a benchmark test for the
concrete procedure that we propose. In a second step, the numerical method is applied
to a magnetic field resembling a step function in one spatial direction (Section 3). This
idealized field configuration represents the simplest configuration, which is inaccessible
to the standard analytical method for inhomogeneous fields: the derivative expansion. As
our main result, we observe diffusion of the magnetic field: the field takes influence on the
region of vanishing background field by inducing a nonzero action density therein.
Our conclusions are summarized in Section 4, where possible generalization and the
road to further application of worldline numerics are sketched.

1. The worldline approach to functional determinants


1.1. Setup
Our starting point is the unrenormalized Euclidean one-loop effective action of scalar
QED in D dimensions in worldline representation [9], corresponding to the determinant of
the gauge-covariant KleinGordon operator

[A] =
1

1/2

dT m2 T
e
N
T

 T 



x 2
+ ie x A x( )
,
Dx( ) exp d
4
0

x(T )=x(0)

(1)
level; 2

where the superscript 1 indicates the one-loop


a gauge-invariant UV regularization at a scale has been performed at the lower bound of the T integral for the sake
of definiteness. In Eq. (1), we encounter a path integral over closed loops in spacetime.
Note that there are no other constraints to the loops except differentiability and closeness;
in particular, they can be arbitrarily self-intersecting and knotty. The normalization can be
2 In an abuse of nomenclature, we call the one-loop contribution to the total effective action also effective

action in the following. The reader should always keep in mind that the Maxwell term (and all higher-order
terms) must be added.

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

determined from the zero-field limit,


 T



x 2 !
d D p p2 T
1
2T
N Dx( ) exp d
= Tr e
=
e
=
.
D
4
(2)
(4T )D/2

355

(2)

Solving Eq. (2) for N and inserting this into Eq. (1) leads us to the compact formula
1
[A] =
(4)D/2


D

d x0

dT
2
em T W [A]
x .
T (D/2)+1

(3)

1/2


Here we have split off the integral over the zero-modes of the path integral, d D x0 , where
center of mass, corresponds to the average position of the loop:
x0 , the so-called
 T loop

x0 := (1/T ) 0 d x ( ). In Eq. (3), we introduced the Wilson loop




T

W [A(x)] = exp ie



d x(
) A x( ) exp ie dx A(x) ,

(4)

and (. . .)
x denotes the expectation value of (. . . ) evaluated over an ensemble of x loops;
the loops are centered upon a common average position
x0 (center of mass) and are
T
distributed according to the Gaussian weight exp[ 0 d x 2 /4].
The following -integral substitution, =: T t, is of crucial importance for the numerical
realization of the path integral; it suggests introducing unit loops y,
1
y(t) := x(T t), t [0, 1],
(5)
T
which are parameterized with a unit propertime t. The remaining integrals can be rewritten
accordingly:
T

x 2 ( )
=
d
4

1

T

y 2 (t)
,
dt
4

1
d x(
)A(x( )) =

dt y(t)




T A T y(t) .

(6)
The important advantage is constituted by the fact that the expectation value of W [A] can
now be evaluated over the unit-loop ensemble y,

W [A(x)]
x W [ T A( T y)]
y ,
(7)
where the exterior T -propertime dependence occurs only as a scaling factor of the gauge
field and its argument. In other words, while approximating the loop path integral by a
finite ensemble of loops, it suffices to have one single unit-loop ensemble at our disposal;
we do not have to generate a new loop ensemble whenever we go over to a new value of T .
Inserting Eq. (7) into Eq. (3), we arrive at the final formula (dropping the subscript
of x0 ):
1 [A] =

1
(4)D/2


dDx
1/2

m2 T
e
W
[
T
A(
T y)]
y + c.t.,
(D/2)+1
dT

(8)

356

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

where we have formally added counter-terms (c.t.) which correspond to a renormalization


of the physical parameters. The details of the c.t.s depend on the dimension, but not on the
type of the background field, and will be discussed in the following.
1.2. Renormalization
For an accurate evaluation of the effective action of Eq. (8), an analytic calculation of the
counter-terms is inevitable in order to obtain accurate results by the numerical procedure.
In the following, we will confine ourselves to the important cases of D = 3 and D = 4
dimensions at the one-loop level.
Let us first discuss the analytical treatment: while for D = 3 the effective action is
rendered finite by dropping a field-independent constant, the complete effective action in
the 4-dimensional case is renormalized in such a way that the quadratic term in the field
strength is given by

1
d 4 x F (x)F (x),
quadr. = 2
(9)
4gR
where F (x) is the field strength tensor, and gR represents the renormalized coupling.
In practice, this is done by subtracting the (infinite) quadratic term of the one-loop
contribution 1 and absorbing it into the bare Maxwell term.
These UV divergencies of the effective action emerge from the lower bound of the
propertime integral in Eq. (8), i.e., T 0. For small values of the propertime, the Wilson
loop expectation value can be calculated exactly (e.g., using heat-kernel techniques):

 
1
W [ T A( T y)]
y = 1 T 2 F [A](x)F [A](x) + O T 4 ,
(10)
12
where x is the loop center of mass. The key observation is that, even for D = 4, the terms
of order T 4 are UV-finite upon the propertime integration in Eq. (8). Hence, the completely
renormalized effective action, which is suitable for numerical simulations, is
1
[A] =
(4)D/2
1

dT

d x

em
(D/2)+1

2T

W [ T A( T y)] 1
y


1 2
T F [A](x)F [A](x)
12

+ cD d D x F [A](x)F [A](x),

where
1 1
+ O(1/),
c3 =
96 m

(11)



  
1 1
2
c4 =
ln 2 C + O m2 2 . (12)
2
12 16
m

Here denotes the UV cutoff, and C is Eulers number. In the D = 4 case, the term c4
will be absorbed into the bare Maxwell term, defining the running of the coupling, and the
cutoff can subsequently be sent to infinity. In D = 3, the theory is super-renormalizable

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

357

and the term c3 represents only a finite shift of the coupling, introducing no running.
In the first two lines of Eq. (11), we have already sent the cutoff two infinity, leaving
2
us with a perfectly finite expression. Note
that in the case D = 3, the term of order T in
Eq. (10) corresponds to a singularity 1/ T , which is integrable. Hence, the subtraction
of this term from the propertime integral is not mandatory. However, aiming at worldline
numerics, the perfect control over the behavior of the integrand at small propertime T is
required in order to augment the accuracy of the numerical evaluation of the propertime
integral.
Now, the numerical renormalization is similar in spirit to the analytical one, but is
complicated by a further problem: evaluating W
with the aid of the loop ensemble
does not produce the small-T behavior of Eq. (10) exactly, but, of course, only within
the numerical accuracy. Unfortunately, even the smallest deviation from Eq. (10) will lead
to huge errors, if we naively plug such a result into Eq. (12); this is because of the factors
of T in the denominator of the integrand for T 0.
Our solution to this problem is to fit the numerical result for W
to a polynomial in T in
the vicinity of T = 0, employing Eq. (10) as a constraint for the first coefficients. Of course,
such a fit is completely justified because of our exact knowledge of W
for small T . This
fit not only represents the renormalization procedure, solving the UV problems, but also
facilitates a more precise estimate of the error bars (see below). Finally, employing this
fitting procedure only close to T = 0, the infrared behavior (T ) of the integrand
remains untouched, and our approach is immediately applicable, also in the case m = 0.
Both renormalization procedures, the analytical as well as the numerical, generalize
straightforwardly to higher dimensions; here, either additional subtractions from the
integrand (for the analytical case), or polynomial fits with higher-order constraints (for
the numerical case) are required: for example, increasing the dimension by 2 requires one
more subtraction/constraint in the small-T behavior of the integrand.
Finally, it should not be concealed that these procedures become increasingly difficult
to higher order in perturbation theory for massive theories in D  4. Then, a mass
renormalization is also necessary, requiring careful analyses of the UV behavior of double
propertime integrals [12].
1.3. Numerical simulation
Now, the route to the effective action is clear:
(1) generate
 1 a unit loop ensemble distributed according to the weight
exp[ 0 dt y 2 /4], e.g., employing the technique of normal (Gaussian) deviates;
(2) compute the integrand for arbitrary values of T (and x); this involves the evaluation
of the Wilson loop expectation value for a given background gauge field;
(3) perform the renormalization procedure and add the c.t.s to the integrand;
(4) integrate over the propertime T in order to obtain the Lagrangian, and also over x
for the action.
From a general viewpoint, there are two sources of systematical error which are
introduced by reducing the degrees of freedom from an infinite to a finite amount: first,

358

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

the loop path integral has to be approximated by a finite number of loops; second, the
propertime t of each loop has to be discretized. Contrary to this, the spacetime does not
require discretization, i.e., the loop ensemble is generated in the continuum.
Of course, the various steps can be carried out using different numerical methods; let us
outline our choice of tools in detail. For this purpose, let Nl denote the number of loops
which are used to estimate the Wilson loop expectation value in Eq. (7), and nl the number
of space points which are employed to specify a particular loop. The points of a particular
loop yi , i = 1, . . . , nl , are generated by a standard heat bath algorithm, where the boundary
conditions y1 = 0, ynl = 0 are enforced. After a proper thermalization, all coordinates yi
are shifted equally in order to implement the center of mass condition yi
= x0 . This
procedure is then repeated Nl times to generate the loop ensemble.
Approximating W
of Eq. (7) by an average over a finite number of loops, the standard
deviation provides an estimate of the statistical error. Approximating the loops by the finite
number nl of space points results in a systematic error that can be estimated by repeating
the calculation for several values nl . The number nl should be chosen large enough to
reduce this systematic error to well below the statistical one. It will turn out that the choice
NL = 1000 and nl = 100 yields results, for the applications below, which are accurate at
the per cent level. We stress, however, that 20000 (dummy) heat bath steps are required
to properly thermalize the loop ensemble.

2. Benchmark test: constant magnetic background field


Since analytic results are available for the case of a constant magnetic background
field B, we will investigate in this section the efficiency of our numerical loop approach to
the scalar functional determinant. 3 For simplicity, we consider three Euclidean spacetime
dimensions, D = 3. Up to a field-independent constant, the one-loop effective-action
density (Lagrangian) L1 for scalar QED in D = 3 is given by [9,11]
 2
B 3/2
m
1
1
,
g
L (B)/V3 =
(13)
3/2
(4)
B
where V3 denotes the volume and

g(z) =
0

dT
I (T , z),
T

I (T , z) =



T
1

1
exp{zT }.
T 2 sinh T

(14)

The exact integrand I (T , z) in Eq. (14) is compared with our loop estimate in Fig. 1
for the case m2 = 0, i.e., z = 0. The numerical estimate for g(m2 /B) is also shown. The
agreement between the two curves is satisfactory and the exact results for I (T , z) as well
as for g(m2 /B) lie well within the error bars produced by the numerics.
3 Incidentally, another exactly solvable system is known [10] with a background field resembling a solitonic
profile in one dimension.

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

359

Fig. 1. Propertime integrand I (T , z = 0) (left panel) and integral g(m2 /B) (right panel) of the
one-loop effective action for the case of a constant magnetic background field. The analytically
known exact results (solid lines) are compared with the numerical findings (circles with error bars).

As mentioned above, the error bars correspond to the statistical error of the ensemble
average. Regarding the quantity W
with its dependence on T , the original error bars
are rather independent of T ; but multiplying W
by a T -dependent
function (cf. Eq. (8))
causes a modulation of the error bars. In particular for T 0, the 1/ T singularity leads
to an unbounded enhancement of the error bars for small T in the function I (T , x) (see
left panel of Fig. 1). Fortunately, the integrand in this regime is known exactly as given
in Eq. (10) in the form of an asymptotic series. As mentioned above for the numerical
renormalization procedure, this information can be used for a constraint fit of the numerical
result to a polynomial in T around T = 0, keeping the first terms corresponding to Eq. (10)
fixed. Then, the error bars of the ensemble average translate into errors for the higher
coefficients of the polynomial. This polynomial is then matched to the pure numerical
result at that value of T where the error bars of the two results are comparable. The final
result is visible on the left panel of Fig. 1.
It should be noted, that the error bars in Fig. 1 (and the following figures) are highly
correlated from point to point, since the same loop ensemble has been used for the
evaluation of each point. This correlation can, of course, easily be reduced by updating
the loop ensemble with a few heat bath steps in-between at the expense of computational
time.

3. Magnetic-field diffusion
In this section, we study the one-loop effective action of Eq. (11) for the case of a
magnetic background field, resembling a step function in space. In particular, we consider
a time-like constant background field B, i.e., a field which is independent of the third
coordinate called Euclidean time. We choose the B field, being a (pseudo-)scalar over the
spatial xy plane, as
B(x, y) = (x)B0 ,

 y) = (x) 1 (y, x)B0 ,


A(x,
2

(15)

360

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

Table 1
Simulation parameters

Set A
Set B
Set C

nl

nT

100
75
50

150000
50000
50000

where (x) is the step function, i.e.,


(x) = 1,

x  0,

(x) = 0,

x < 0.

Note that because of the sudden variation of the background field at the step, the
effective action cannot be obtained within a derivative expansion. By contrast, such a
discontinuity represents no obstacle for the numerical worldline approach proposed in this
work. Discontinuities do not induce (artificial) singularities, but are smoothly controlled
by the properties of the loop ensemble. This ensemble resembles a cloud, exhibiting finite
extension and slowly varying density, and being centered at some particular spacetime
point x0 which is the center of mass of each loop in the cloud. While running with this
x0 towards and across the step, that part of the volume of the loop cloud which feels
the magnetic field increases smoothly; therefore, the effective-action density (effective
Lagrangian) L1 , being the propertime integrated information of what the loop cloud
measures, will be smooth, too.
In order to numerically evaluate the effective action for the case of Eq. (15), we employ
the loop approach outlined in the previous section. This approach provides for statistical
error bars which allow to estimate whether the number of loop ensembles is large enough.
In order to study the systematic errors, such as the limited number of spacetime points nl
specifying a loop and the limited number nT of thermalization sweeps, we shall employ
three sets of loop ensembles (see Table 1). Defining
 2


m
(4)3/2 1 
1/2
g
(16)
:=
,
xB
L x = (x, 0, 0)
0
2
3/2
B0
B0
in analogy to Eq. (13), the final numerical result for the effective action is shown in Fig. 2,
left panel. As expected, the effective-action density is nonzero even in the region x < 0
where the background field B(
x ) vanishes. For increasing mass, the contributions from
large propertime T are exponentially suppressed in the integrand. Since the propertime
controls the size of the loop cloud, the large loop clouds contribute less to the action density
when the mass is large. Hence, the effective-action density for an increasing mass becomes
reduced. (From an alternative viewpoint, the limit of large mass and small field are identical
for dimensional reasons.)
We find that the effective action decreases exponentially in the forbidden region,
which is depicted in Fig. 2, right panel; here we defined the quantity
 2

 1/2 
m
1/2
,
, xB0
g
(17)
=: exp S xB0
B0

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

361

Fig. 2. Effective-action density in the vicinity of the magnetic step (x = 0). On the left panel, the
diffusion profile is visualized. The logarithmic plot on the right panel reveals the exponential nature
of the diffusion depth.
1/2

and plotted S(xB0 ) for several values of the scalar mass m. This phenomenon of
obtaining a nonzero effective-action density even in a region where the background field
vanishes may be called quantum diffusion of the magnetic field.
1/2
Large values of S imply that the effective-action density g(m2 /B0 , xB0 ) is small
due to cancellations. In this case, a sufficiently large number of loop points nl and of
1/2
thermalization sweeps is requested. The results indicate that the function S(xB0 ) can be

fitted for x B0 by the ansatz


S(y) = + y,

1/2

y = xB0 ,

(18)

where the quantities and depend on m2 /B0 . The diffusion depth l of the magnetic field

can be defined as the inverse of , i.e., l = 1/( B0 ). An estimate of this parameter has
been plotted versus the mass m in Fig. 3. As expected, the diffusion coefficient rises with
increasing mass scale m. The function
 2 1/2
  
m
m2 B0 = 0.7627 + 3.255
(19)
B0
nicely fits the high quality numerical data set A (see Fig. 3).
In order to get an understanding of these numbers, let us perform the following heuristic
consideration: at a first glance, one might expect that the magnetic diffusion obeys the
simple law econstmx , since the mass seems to be the only scale in the field-free region
of space; this would correspond to = const (m2 /B0 )0.5 . However, if this simple law
were true, the massless limit m = 0 would be obscure, since then the magnetic field would
diffuse into the field-free region without any damping. Hence, there must be an additional
dependence of the exponent on the magnetic field in order to account for a reasonable
massless limit. Moreover, the mass is indeed not the only scale in the field-free region,
because the loop cloud is a nonlocal object that always feels the strength of the magnetic
field. In fact, it is the constant first term in in Eq. (19) that exactly accounts for this

362

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

Fig. 3. Inverse diffusion depth versus mass-to-magnetic-field ratio. The plot symbols depict the
numerical results, whereas the solid line represents the fit of Eq. (19) adjusted to the optimized loop
ensemble A.

dependence of the magnetic diffusion on the strength of the magnetic field. Only for larger
masses (or weaker fields), m2  B0 /4, the intuitively expected diffusion law of the form
econstmx begins to dominate; in this regime, we find
l 0.31/m.

(20)

In the present work, the precision of the numerics for the action density in the field-free
region restricts the investigation to mass values of m2 < 1.5B0 ; beyond this, the strong
exponential decrease beyond the step prohibits a reliable analysis of the diffusion depth.

4. Conclusion and outlook


Beyond any particular result of the present work, we would like to remark in the first
place that our approach to the worldline formula (see Eq. (8)) for the one-loop effective
action offers a vivid picture of the quantum world. Consider a spacetime point x at a
propertime T ; then, the loop ensemble is centered upon this point x and resembles a
loop cloud with Gaussian density and spread. Increasing or lowering the propertime
T corresponds to bloating or scaling down the loop cloud or, alternatively, zooming out of
or into the microscopic world. The effective-action density at each point x finally receives
contributions from every point of the loop cloud according to its Gaussian weight (times
the mass term and other factors) and averaged over the propertime. This gives rise to
the inherent nonlocality and nonlinearity of the effective action, because every point x
is influenced by the field of any other point in spacetime experienced by the loop cloud.
In particular, we considered the one-loop contribution to the effective action in scalar
electrodynamics in three spacetime dimensions; in the beginning, we were able to
reproduce the analytically well-known case of a constant magnetic background field,
serving as a testing ground for our numerical procedures including renormalization to one-

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

363

loop order. Incidentally, the zero-mass limit (or, alternatively, the ultra-strong magnetic
field limit) is also covered by our approach without additional difficulties.
We furthermore tackled the problem of a step-function-like magnetic background field,
illustrating the stability of our approach also for discontinuous field configurations. For
this case, we observed a diffusion of the magnetic field, i.e., nonzero action density even
at a distance from the magnetic field. This phenomenon is obviously nonlocal, since an
expansion around a point of zero background field gives a zero result to any finite order.
But the diffusion phenomenon appears to be also nonperturbative in the same sense as the
Schwinger mechanism of pair production [11], at least for not too large values of the mass;
1/2
this is suggested by the functional form of the diffusion for small mass: exp(S(xB0 ))

exp(0.7627 B0 x); this result cannot be expanded in terms of the coupling constant,
being rescaled in the field (only an expansion in terms of the square root of the coupling
constant is possible).
Perhaps the main advantage of the numerical worldline approach to functional
determinants is that no a priori information, such as certain symmetries of the background
field or a suitably chosen set of base functions, has to be exploited; if the loop ensemble
has been thermalized properly, any background field can immediately be plugged into the
algorithm.
One drawback of the numerical approach lies in the fact that it applies only to Euclidean
quantum field theory; this is because the action governing the distribution of the loops must
be positive.
The present paper paves the way to further generalizations such as fermionic functional
determinants; our approach could facilitate a systematic numerical investigation of these
determinants. Results could be compared with several results known from analytical
considerations (see, e.g., [13]). Although the treatment of fermions in the worldline
formalism can be most elegantly formulated via Grassmann representations (see, e.g.,
[14]), a numerical approach has to rely on a bosonic representation of the path integrals
[1,15]. In practice, this means that the Dirac algebraic elements in the worldline action are
accompanied by a path-ordering prescription. Similar complications occur for nonabelian
gauge fields and color-charged fermions or scalars. Whereas such path ordering is difficult
to deal with in analytical approaches, a numerical evaluation can immediately take care of
such a prescription. This is because the loops are discretized in the propertime parameter
anyway, consisting of nl links; the path ordering then is nothing but a simple (matrix-)
multiplication of these links.
Recently [16], a new approach was designed to improve the still unpleasant situation [17]
when lattice QCD is studied at finite baryon densities: the basic idea to circumvent the
problems [18] with lattice fermions is a calculation of the continuum fermion determinant
for arbitrary entries of the gluon field, which subsequently is the subject of a lattice
discretization. This program was successfully applied to the case of heavy quarks [16]. The
formulation of the present paper stirs the hope that the concept of [16] might be extended
to the realistic case of light quarks.
Further possible and straightforward applications can be found in the context of
estimating quantum energies to solitonic fields (see [19]), or in the context of thermal field

364

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

theory or Casimir vacua (for analytical worldline approaches, see, e.g., [20]). The latter
cases are connected with a compactification of the Euclidean spacetime manifold; these
topological modifications can be imposed directly on the loop ensemble in our approach.
For example, compactifying the Euclidean time direction at finite temperature will result in
closed loops that wind several times around the time-like cylinder. Work in these directions
is in progress.
Finally, it is obvious that the step-like magnetic field is not physical at all in a strict sense,
because such a sharp drop-off cannot be produced by real laboratory magnets; nevertheless,
the step-like field can be regarded as a limiting case of a real physical situation, being
useful for an estimate of possible effects caused by rapid variations of a magnetic field.
For instance, as a matter of principle, magnetic diffusion can be included in the discussion
of the AharonovBohm effect: if the propagating particle is no longer considered as a
quantum mechanical point particle, but as a quantum field, it will be affected by the
quantum diffusion of the solenoids magnetic field; but a measurable effect can only arise,
if the particles distance from the solenoid is of the order of a few Compton wavelengths.
Beyond that, we would like to mention that our results for the step-like magnetic
field are of some significance for the experiments being currently performed at PVLAS
(Legnaro) and BMV (Toulouse) [21], in which is measured the optical birefringence of the
quantum vacuum exposed to a magnetic field. For example in the PVLAS experiment, the
polarization axis of a laser beam is affected by a (69 T)-magnetic field with a diameter
of 1 m; the magnetic field drops off over a distance of 10 cm, and the question arises as to
whether this drop-off will influence the rotation of the polarization axis as predicted by a
constant-field QED calculation.
The refractive indices of the modified vacuum are proportional to the energy density
induced by the magnetic field (for a review, see [22]). For weak magnetic fields, the energy
density is directly related to the (Euclidean) effective-action density. Provided that our
present results also hold for D = 4 spinor QED at least qualitatively, we can exclude any
further influence of the drop-off region, since the magnetic diffusion in this case (m  B)
occurs at the order of a Compton wavelength 1/m. This zero-result is supported by
considerations within a derivative expansion, where the natural expansion parameter is
given by the Compton wavelength over the length of the field variation (e.g.,  4 1012
for the PVLAS). Only the constant-field result integrated over the size of the magnet
including the drop-off region need be taken into account.

Acknowledgement
We would like to thank W. Dittrich for helpful discussions and for carefully reading the
manuscript. H.G. gratefully acknowledges insightful discussions with C. Schubert. K.L. is
indebted to H. Reinhardt for encouragement and support. This work has been supported in
part by the Deutsche Forschungsgemeinschaft under contract Gi 328/1-1.

H. Gies, K. Langfeld / Nuclear Physics B 613 (2001) 353365

365

References
[1] R.P. Feynman, Phys. Rev. 80 (1950) 440;
R.P. Feynman, Phys. Rev. 84 (1951) 108.
[2] Z. Bern, D.A. Kosower, Nucl. Phys. B 362 (1991) 389;
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
[3] M.J. Strassler, Nucl. Phys. B 385 (1992) 145.
[4] M.G. Schmidt, C. Schubert, Phys. Lett. B 318 (1993) 438, hep-th/9309055.
[5] R. Shaisultanov, Phys. Lett. B 378 (1996) 354, hep-th/9512142.
[6] M. Reuter, M.G. Schmidt, C. Schubert, Ann. Phys. 259 (1997) 313, hep-th/9610191.
[7] D. Cangemi, E. DHoker, G. Dunne, Phys. Rev. D 51 (1995) 2513, hep-th/9409113.
[8] V.P. Gusynin, I.A. Shovkovy, J. Math. Phys. 40 (1999) 5406, hep-th/9804143.
[9] C. Schubert, hep-th/0101036, to appear in Phys. Rep.
[10] D. Cangemi, E. DHoker, G. Dunne, Phys. Rev. D 52 (1995) 3163, hep-th/9506085;
G. Dunne, T. Hall, Phys. Rev. D 58 (1998) 105022, hep-th/9807031;
G. Dunne, T. Hall, Phys. Lett. B 419 (1998) 322, hep-th/9710062.
[11] J. Schwinger, Phys. Rev. 82 (1951) 664.
[12] V.I. Ritus, Sov. Phys. JETP 42 (1975) 774;
W. Dittrich, M. Reuter, Springer Lect. Notes Phys. 220 (1985) 1;
D. Fliegner, M. Reuter, M.G. Schmidt, C. Schubert, Theor. Math. Phys. 113 (1997) 1442,
hep-th/9704194.
[13] M.P. Fry, Int. J. Mod. Phys. A 12 (1997) 1153;
M.P. Fry, Phys. Rev. D 62 (2000) 125007, hep-th/0010008.
[14] E. DHoker, D.G. Gagne, Nucl. Phys. B 467 (1996) 272, hep-th/9508131;
J.W. van Holten, Nucl. Phys. B 457 (1995) 375, hep-th/9508136.
[15] A.O. Barut, I.H. Duru, Phys. Rep. 172 (1989) 1.
[16] K. Langfeld, G. Shin, Nucl. Phys. B 572 (2000) 266.
[17] I.M. Barbour, Talk presented at Workshop on QCD at Finite Baryon Density: A Complex
System with a Complex Action, Bielefeld, Germany, 1998.
[18] M.A. Stephanov, Phys. Rev. Lett. 76 (1996) 4472.
[19] N. Graham, R.L. Jaffe, M. Quandt, H. Weigel, Quantum energies of interfaces, hep-th/0103010;
E. Farhi, N. Graham, R.L. Jaffe, H. Weigel, Nucl. Phys. B 585 (2000) 443, hep-th/0003144;
E. Farhi, N. Graham, R.L. Jaffe, H. Weigel, Phys. Lett. B 475 (2000) 335, hep-th/9912283.
[20] D.G. McKeon, A. Rebhan, Phys. Rev. D 47 (1993) 5487, hep-th/9211076;
I.A. Shovkovy, Phys. Lett. B 441 (1998) 313, hep-th/9806156;
R. Venugopalan, J. Wirstam, hep-th/0102029.
[21] E. Zavattini, in: D. Amati, G. Cantatore, E. Zavattini, R.S. Hayano (Eds.), Proc. of the
QED2000, 2nd Workshop on Frontier Tests of Quantum Electrodynamics and Physics of the
Vacuum, 2001;
C. Rizzo, in: D. Amati, G. Cantatore, E. Zavattini, R.S. Hayano (Eds.), Proc. of the QED2000,
2nd Workshop on Frontier Tests of Quantum Electrodynamics and Physics of the Vacuum,
2001;
See also the URL of the PVLAS experiment: http://sunlnl.lnl.infn.it/ pvlas/home.htm.
[22] W. Dittrich, H. Gies, Springer Tracts Mod. Phys. 166 (2000) 1.

Nuclear Physics B 613 (2001) 366381


www.elsevier.com/locate/npe

Theoretical expectations for the muons electric


dipole moment
Jonathan L. Feng a,b , Konstantin T. Matchev c , Yael Shadmi d
a Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
b Department of Physics and Astronomy, University of California, Irvine, CA 92697, USA
c Theory Division, CERN, CH-1211, Geneva 23, Switzerland
d Department of Particle Physics, Weizmann Institute of Science, Rehovot 76100, Israel

Received 30 July 2001; accepted 2 August 2001

Abstract
We examine the muons electric dipole moment d from a variety of theoretical perspectives. We
point out that the reported deviation in the muons g 2 can be due partially or even entirely to a new
physics contribution to the muons electric dipole moment. In fact, the recent g 2 measurement
provides the most stringent bound on d to date. This ambiguity could be definitively resolved
by the dedicated search for d recently proposed. We then consider both model-independent and
supersymmetric frameworks. Under the assumptions of scalar degeneracy, proportionality, and flavor
conservation, the theoretical expectations for d in supersymmetry fall just below the proposed
sensitivity. However, nondegeneracy can give an order of magnitude enhancement, and lepton flavor
violation can lead to d 1022 e cm, two orders of magnitude above the sensitivity of the d
experiment. We present compact expressions for leptonic dipole moments and lepton flavor violating
amplitudes. We also derive new limits on the amount of flavor violation allowed and demonstrate
that approximations previously used to obtain such limits are highly inaccurate in much of parameter
space. 2001 Published by Elsevier Science B.V.
PACS: 12.60.Jv; 13.40.Em; 14.60.Ef; 14.80.Ly

1. Introduction
Electric dipole moments (EDMs) of elementary particles are predicted to be far below
foreseeable experimental sensitivity in the standard model. In extensions of the standard
model, however, much larger EDMs are possible. Current EDM bounds are already some of
the most stringent constraints on new physics, and they are highly complementary to many
other low energy constraints, since they require CP violation, but not flavor violation.
E-mail address: jlf@mit.edu (J.L. Feng).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 8 3 - 2

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

367

The field of precision muon physics will be transformed in the next few years [1]. The
EDM of the muon is, therefore, of special interest. A new experiment [2] has been proposed
to measure the muons EDM at the level of
d 1024 e cm,

(1)

more than five orders of magnitude below the current bound [3]
d = (3.7 3.4) 1019 e cm.

(2)

The interest in the muons EDM is further heightened by the recent measurement of
the muons anomalous magnetic dipole moment (MDM) a = (g 2)/2, where g is
exp
the muons gyromagnetic ratio. The current measurement a = 11 659 202 (14) (6)
1010 [4] from the muon (g 2) experiment at Brookhaven differs from the standard
model prediction aSM [5,6] by 2.6 :
a a aSM = (43 16) 1010 .
exp

(3)

The muons EDM and a arise from similar operators, and this tentative evidence for a
nonstandard model contribution to a also motivates the search for the muons EDM.
In this study, we examine the prospects for detecting a nonvanishing muon EDM from a
variety of theoretical perspectives. We first note that the reported deviation in the muons
g 2 can be due partially or even entirely to a new physics contribution to the muons
electric dipole moment. In fact, at present the result from the muon (g 2) experiment
provides the most stringent bound on d . We derive this bound and comment on the
conclusions that may be drawn about d from the a measurement alone.
We then move to more concrete frameworks, where additional correlations constrain
our expectations. In particular, we consider supersymmetry and examine quantitatively the
implications of the electron EDM and lepton flavor violating processes [7,8]. Our aim
is to impose as few theoretical prejudices as possible and draw correspondingly general
conclusions. For studies of the muon EDM in specific supersymmetric models, see, e.g.,
Refs. [9,10].
Finally, although we use exact expressions for all flavor-conserving amplitudes in this
study, we also provide compact expressions in the mass insertion approximation for
branching ratios of radiative lepton decays and for leptonic EDMs and MDMs both with
and without lepton flavor violation. These include all leading supersymmetric effects and
are well-suited to numerical calculations.
2. Model-independent bounds from the muon (g 2) experiment
Modern measurements of the muons MDM exploit the equivalence of cyclotron and
spin precession frequencies for g = 2 fermions circulating in a perpendicular and uniform
magnetic field. Measurements of the anomalous spin precession frequency are, therefore,
interpreted as measurements of a .
The spin precession frequency also receives contributions from the muons EDM,
however. For a muon traveling with velocity perpendicular to both a magnetic field B

368

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

 NP NP 
Fig. 1. Regions in the a
, d plane that are consistent with the observed |a | at the 1 and 2
levels. The current 1 and 2 bounds on dNP [3] are also shown.

and an electric field E, the anomalous spin precession vector is




1
e
2c
2
e
a = a
B d B

a
E d E.
m
m c 2 1
h
h

(4)

In recent experiments, the third term of Eq. (4) is removed by running at the magic
29.3, and the last term is negligible. For highly relativistic muons with || 1, then, the
anomalous precession frequency is
 2


1/2

2c
e 2  SM 2
SM NP
NP 2
a + 2a a +
d
,
|a | |B|
(5)
h
m
where NP denotes new physics contributions, and we have assumed aNP aSM and
dNP dSM .
The observed deviation from the standard model prediction for |a | has been assumed
to arise entirely from a MDM and has been attributed to a new physics contribution
of size a . However, from Eq. (5), we see that, more generally, it may be due to
some combination of magnetic and electric dipole moments from new physics. More
quantitatively, the effect can also be due to an EDM contribution


 NP 
aNP


h
e
1

19
d 
aSM a aNP 3.0 10
e cm 1
,
(6)

m c 2
43 1010
to the current central value given in Eq. (3). In Fig. 1 we
where aNP has been normalized

show the regions in the aNP , dNP plane that are consistent with the observed deviation in
|a |.
In fact, the observed anomaly may, in principle, be due entirely to the muons EDM!
This is evident from Eqs. (2) and (6), or from Fig. 1, where the current 1 and 2 upper
bounds on dNP [3] are also shown. Alternatively, in the absence of fine-tuned cancellations

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

369

between aNP and dNP , the results of the muon (g 2) experiment also provide the most
stringent bound on d to date, with 1 and 2 upper limits


a < 59 (75) 1010 dNP  < 3.5 (3.9) 1019 e cm.
(7)
This discussion is completely model-independent. In specific models, however, it may be
difficult to achieve values of d large enough to saturate the bound of Eq. (7). For example,
in supersymmetry, assuming flavor conservation and taking extreme values of superparticle
masses ( 100 GeV) and tan (tan 60) to maximize the effect, the largest possible
value of a is amax 107 [11]. Very roughly, one therefore expects a maximal muon
EDM of order (eh /2m c)amax 1020 e cm in supersymmetry.
Of course, the effects of d and a are physically distinguishable: while a causes
precession around the magnetic fields axis, d leads to oscillation of the muons spin
above and below the plane of motion. This oscillation is detectable in the distribution of
positrons from muon decay, and further analysis of the recent a data should tighten the
current bounds on d . Such analysis is currently in progress [12]. The proposed dedicated
d search will provide a definitive answer, however, by either measuring a nonzero d or
constraining the contribution of d to |a | to be insignificant.
3. Theoretical expectations from the muons MDM
The muons EDM and anomalous MDM are defined through 1
i
LEDM = dNP
mn 5 Fmn ,
2
e

mn Fmn ,
LMDM = aNP
4m

(8)
(9)

where mn = 2i [ m , n ] and F is the electromagnetic field strength. These operators


are closely related. In the absence of all other considerations, one might expect their
coefficients to be of the same order. Parameterizing them as dNP /2 = Im A and
aNP e/(4m ) = Re A with A |A|eiCP , one finds
dNP = 4.0 1022 e cm

aNP
43 1010

tan CP .

(10)

Themeasureddiscrepancy in |a | then constrains CP and dNP . The preferred regions of


the CP , dNP plane are shown in Fig. 2. For natural values of CP 1, dNP is of order
1022 e cm. With the proposed dNP sensitivity of Eq. (1), all of the 2 allowed region with
CP > 102 yields an observable signal.
At the same time, while this model-independent analysis indicates that natural values of
CP prefer dNP well within reach of the proposed muon EDM experiment, very large values
of dNP also require highly fine-tuned CP . For example, the contributions of dNP and aNP
to the observed discrepancy in a are roughly equal only if |/2 CP | 103 . This is
1 Here and below we set h = c = 1.

370

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381



Fig. 2. Regions of the CP , dNP plane allowed by the measured central value of |a | (solid) and its
1 and 2 preferred values (shaded). The horizontal line marks the proposed experimental sensitivity
to dNP .

a consequence of the fact that EDMs are CP -odd and dSM 0, and so dNP appears only
quadratically in |a |. Without a strong motivation for CP /2, it is, therefore, natural
to expect the EDM contribution to |a | to be negligible, and we assume in the following
that the |a | measurement is indeed a measurement of a .
4. The electron EDM and naive scaling
The EDM operator of Eq. (8) couples left- and right-handed muons, and so requires a
mass insertion to flip the chirality. The natural choice for this mass is the lepton mass. On
dimensional grounds, one, therefore, expects
m
dNP 2 ,
(11)
m

where m
is the mass scale of the new physics. If the new physics is flavor blind, df mf
for all fermions f , which we refer to as naive scaling. In particular,
m
de .
d
(12)
me
The current bound on the electron EDM is de = 1.8 (1.2) (1.0) 1027 e cm [13].
Combining the statistical and systematic errors in quadrature, this bound and Eq. (12)
imply
d  9.1 1025 e cm,

(13)

at the 90% CL, which is barely below the sensitivity of Eq. (1). Naive scaling must be
violated if a nonvanishing d is to be observable at the proposed experiment. On the
other hand, the proximity of the limit of Eq. (13) to the projected experimental sensitivity

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

371

of Eq. (1) implies that even relatively small departures from naive scaling may yield an
observable signal.

5. Violations of naive scaling in supersymmetry


Is naive scaling violation well-motivated, and can the violation be large enough to
produce an observable EDM for the muon? To investigate these questions quantitatively,
we consider supersymmetry. (For violations of naive scaling in other models, see, for
example, Ref. [14].) Many additional mass parameters are introduced in supersymmetric
extensions of the standard model. These are in general complex and so are new sources
of CP violation. These parameters may be correlated by a fundamental theory of
supersymmetry breaking that includes a specific mechanism for mediating the breaking. In
fact, all viable mechanisms of mediating supersymmetry breaking are designed to suppress
flavor violation, and so CP -violating observables that also involve flavor violation, such
as K , are also suppressed. However, EDMs do not require flavor violation, and constraints
on quark and electron EDMs are some of the main challenges for supersymmetric models.
For a recent discussion of the supersymmetric CP problem in various supersymmetry
breaking schemes, see Ref. [15].
In full generality, the relevant dimensionful supersymmetry parameters for leptonic
EDMs are the slepton mass matrices m2 , trilinear scalar couplings A, gaugino masses
M1 and M2 , the Higgsino mass , and the dimension two Higgs scalar coupling B.
Schematically, these enter the Lagrangian through the terms

i E

j , Aij Hd L i E

j ,
L m2LL ij L i L j , m2RR ij E

B,

M2 W

, H

u H

d , BHu Hd ,
M1 B

(14)

where i, j are generational indices, L and E denote SU(2) doublet and singlet leptons,

and W

are
respectively, Hu and Hd are the up- and down-type Higgs multiplets, and B
0
0
gauginos, the U(1) Bino and SU(2) Winos. The parameter tan = Hu /Hd , the ratio of
Higgs boson vacuum expectation values, will also enter below.
The U(1)R and U(1)P Q symmetries allow us to remove two phases we choose M2 and
B real. Throughout this study, we assume that the gaugino masses have a common phase, as
is true in many well-motivated theories where the gaugino masses are either unified at some
scale or otherwise have a common origin. We also begin by assuming supersymmetric
flavor conservation, that is, that the sfermion masses and trilinear couplings are diagonal
in the fermion mass basis. (We will consider flavor violation in Section 5.3.) With these
assumptions, only the and A) parameters are complex. We define Arg() and
A) Arg(A) ).
Leptonic electromagnetic dipole moments arise at one-loop from charginosneutrino
and neutralinocharged slepton diagrams. In all figures and results presented here, we
evaluate flavor-conserving amplitudes exactly. However, for purposes of exposition, it
is convenient to consider the fermion mass basis and to adopt the mass insertion
approximation for sleptons, neutralinos, and charginos. (We have checked that the mass

372

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

insertion approximation is accurate to about 5% in almost all of parameter space, justifying


the intuition derived from this simplification.) In the mass insertion approximation, for
large and moderate tan and neglecting subdominant terms, there are five contributions
with the following Feynman diagrams and amplitudes [16]:


 
Aa) = g  2 M1 A) Hd0 m) tan KN M12 , m2) , m2) ,
L



Ab) = g  2 M1 m) tan KN m2) , ||2 , M12 ,
R



1
Ac) = g  2 M1 m) tan KN m2) , ||2 , M12 ,
L
2



1
Ad) = g22 M2 m) tan KN m2) , ||2 , M22 ,
L
2


Ae) = g22 M2 m) tan KC m2 ) , ||2 , M22 .

(15)

In these diagrams, an external photon connected to any charged internal line is implicit,
LR (A H 0  m tan )/m2 . The functions K and K are mass dimension
and ))
)
)
N
C
d
)
4 functions entering the neutralino and chargino diagrams, respectively, and are given in
Appendix A.
Defining
a
b
c
d
e
Atot
) A) + A) + A) + A) + A) ,

(16)

the EDM and anomalous MDM of a lepton ) are simply


1
d) = e Im Atot
a) = m) Re Atot
) ,
) .
2
From the amplitudes of Eq. (15), naive scaling is seen to require

(17)

Degeneracy: Generation-independent m)R , m)L , and m ) .


Proportionality: The A terms must satisfy Im(A) ) m) .
Flavor conservation: Vanishing off-diagonal elements of m2LL , m2RR , and A.
The last requirement, flavor conservation, has been assumed in all of our discussion so
far. As we will see, relaxing this assumption also leads to naive scaling violation. We now
consider violations of each of these properties in turn.

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

373

5.1. Nondegeneracy
Scalar degeneracy is the most obvious way to reduce flavor changing effects to allowable
levels. Therefore, many schemes for mediating supersymmetry breaking try to achieve
degeneracy. However, in many of these, with the exception of simple gauge mediation
models, there may be non-negligible contributions to scalar masses that are generationdependent. Furthermore, there are classes of models that do not require scalar degeneracy
at all. For example, scalar nondegeneracy is typical in alignment models [17], where flavorchanging effects are suppressed by the alignment of scalar and fermion mass matrices
rather than by scalar degeneracy. Scalar nondegeneracy is also typical in models with
anomalous U(1) contributions to the sfermion masses. In fact, in models where the
anomalous U(1) symmetry determines both sfermion and fermion masses, the sfermion
hierarchy is often inverted relative to the fermion mass hierarchy [1820], and so smuons
are lighter than selectrons, as required for an observable d . In summary, there is a wide
variety of models in which deviations from scalar degeneracy exist.
We now consider a simple model-independent parameterization to explore the impact of
nondegenerate selectron and smuon masses. We set meR = meL = me and m R = m L =
m and assume vanishing A parameters. For fixed values of M1 , M2 , ||, and large tan ,
then, to a good approximation both de and d are proportional to sin tan , and we
assume that sin tan saturates the de bound.
Contours of d are given in Fig. 3. The contributions to d have been evaluated exactly
(without the mass insertion approximation). Observable values of d are possible even for
small violations of nondegeneracy; for example, for m /me  0.9, muon EDMs greater

Fig. 3. Contours of d in units of 1024 e cm for varying meR = meL = me and m R = m L = m




for vanishing A terms, fixed || = 500 GeV and M2 = 300 GeV, and M1 = g12 /g22 M2 determined
from gaugino mass unification. The parameter combination sin tan is assumed to saturate the
bound de < 4.4 1027 e cm. The shaded regions are preferred by a at 1 and 2 for tan = 50.

374

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

than 1024 e cm are possible. The current value of a also favors light smuons and large
EDMs. The smuon mass regions preferred by the current a anomaly are given in Fig. 3
for tan = 50. Within the 1 preferred region, d may be as large as 4 (10) 1024 e cm
for me < 1 (2) TeV. Our assumed value of tan is conservative; for smaller tan , the
preferred smuon masses are lower and the possible d values larger.
5.2. Nonproportionality
Naive scaling is also broken if the Im A) are not proportional to Yukawa couplings y) .
Just as in the case of nondegeneracy, deviations from proportionality are found in many
models. Even in models constructed to give proportionality, there are often corrections to
the A terms, so that
Aij = yij A0 + aij ,

(18)

where the second term is smaller than the first term, but violates proportionality. In flavor
models, the A terms do not obey proportionality at all. Rather they are of the form
Aij = cij yij A0 ,

(19)

where cij are order one coefficients. Clearly, violations of proportionality may affect d not
only by changing the magnitude of the A terms, but also through the possible appearance of
new phases, either in aij or in cij . Note that these possibilities also lead to flavor violation,
the subject of the following section.
We will not study the possibility of nonproportionality in detail. For large tan , the A
term contribution to the EDM is suppressed relative to the typically dominant chargino
contribution by roughly a factor of (g22 M2 /g  2 M1 )(y Im tan / Im A ), where we have
used the amplitudes of Eq. (15). However, there are many possibilities that may yield large
effects. In Ref. [8], for example, it was noted that Ae may be such that the chargino and
neutralino contributions to de cancel, while, since Ae = A , there is no cancellation in d ,
and observable values are possible.
5.3. Flavor violation
In all of the discussion so far, we have neglected the possibility of supersymmetric lepton
flavor violation. However, such flavor violation is present, at least at subleading order,
in most models of high-scale supersymmetry breaking [15]. Moreover, large smuonstau
mixing, of particular importance here, is well-motivated by the evidence for large
mixing observed in atmospheric neutrinos [21]. Explicit examples of this connection in
models with leftright gauge symmetry are given in Refs. [9,10]. The relation between
neutrino and slepton mixing is also a general feature of Abelian flavor models [22]. In the
simplest of these models, highly mixed states have similar masses, contradicting the most
straightforward interpretation of the neutrino data. This difficulty may be circumvented in
less minimal models by generating hierarchical neutrino masses from the neutrino mass
matrix and large mixing by arranging for the gauge and mass eigenstates of the SU(2)

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

375

lepton doublets to be related by large rotations. However, because lepton doublets contain
charged leptons in addition to neutrinos, these rotations also generate large misalignments
between the charged leptons and sleptons, producing large slepton flavor mixing.
Smuonstau mixing leads to a potentially significant enhancement in d , because it
breaks naive scaling by introducing contributions with a tau mass insertion so that d
m /m
2 . However, to evaluate the significance of this enhancement, we must first determine
how large the flavor violation may be. This effect may be isolated by assuming that all
charged sleptons are roughly degenerate with characteristic mass m). In the basis with
lepton mass eigenstates and flavor-diagonal gauge interactions, slepton flavor violation
enters through off-diagonal masses in the slepton mass matrix. As usual, we parameterize
LL m2 /m2 and RR m2 /m2 .
the chirality-preserving off-diagonal masses by 23
L 23
E 23
23
)
)
There may also be flavor violation in the leftright couplings; we parameterize these by
LR and RL . We begin by assuming real s; however, very large effects are possible for
23
23
imaginary s, and we consider this possibility at the end of this section.
The off-diagonal masses induce transitions. Eight contributions are parametLR,RL
m ). Their Feynman
rically enhanced by m /m (retaining the possibility that 23
diagrams and amplitudes are
m Aa RR
,
m ln m2 23

m Aa LL
,
m ln m2 23

m Ab RR
,
m ln m2 23

m Ac LL
,
m ln m2 23

m Ad LL
,
m ln m2 23

m Ae LL
,
m ln m2 23

Aa

1 RL
,
LR 23
22

Aa

1 LR
, (20)
LR 23
22

where the amplitudes Ai) are given in Eq. (15). The branching ratio may then be written as
B( ) =


12 3
|ML |2 + |MR |2 B( e e ),
2
2
GF m

where


Atot
Atot
m
1 LR

LL
ML =
+
+ Aa LR 23
,
23
2
2
m ln m
22
ln m
L

(21)

376

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

LL max (left) and RR max / RR max (right). The values max and
Fig. 4. The ratios LLmax /23
23
23 B only
23 B only 23
max
are the upper bounds allowed by determined with all leading contributions and
23 B only

with only the Bino-mediated contribution, respectively. We fix m) 300 GeV and tan = 50. The
shaded regions are excluded by LEP, the requirement of a neutral LSP, and the condition that the
vacuum not be charge-breaking.

MR =

1 RL
m Atot

RR
23
+ Aa LR 23
.
2
m ln m
22

(22)

Eqs. (15), (16), (21) and (22) provide a compact form for the branching ratio for radiative
lepton decays in the mass insertion approximation and are well-suited to numerical
evaluation. (Note that terms subleading in m /m and linear in the s are also easily
2
RR
2
LL
tot
computed as ML = (Atot
/ ln m R )23 and MR = (A / ln m L )23 , but we neglect
them in the analysis to follow.)
The flavor-violating mass insertions are bounded by the current constraint B( )
< 1.1 106 [23]. It is important to note that the Higgsino-mediated decays give the
dominant contribution unless M2 , M1 [24]. The often-used bounds of Gabbiani,
Gabrielli, Masiero and Silvestrini [25] assume a photino neutralino, and so effectively
include only the Bino-mediated contribution. In Fig. 4 we show contours of the ratio of
LL,RR
the upper bound on 23
determined with only the Bino-mediated contribution included
to the upper bound determined with all of the leading diagrams included. We see that
the Bino-only bounds are reasonably accurate only for ||  1 TeV, a region that is
forbidden by the requirement that electromagnetic charge be an unbroken symmetry. (We
have assumed that the diagonal entries of the stau and smuon mass matrices are equal;
in the forbidden region, m2 < 0.) Analyses based solely on the Bino contribution are
1
highly misleading in most regions of parameter space, especially for moderate and large
LL , the constraint from is always far more stringent than
tan . In particular, for 23
one would conclude from a Bino-only analysis. Similar conclusions apply to constraints
from e , e , and e conversion and will be presented elsewhere [26].

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

377

We now determine what effect flavor violation may have on d . Flavor violation induces
four m /m enhanced contributions to the muons EDM:
2 Aa
m
LL RR ,
m ln m2 ln m2 23 23
L

RL
23

Aa

LR ln m2
22
L

LL
23
,

2 Aa
m
LR RL ,
m ln m2 ln m2 23 23
L

LR
23

Aa

LR ln m2
22
R

RR
23
.

The additional flavor-violating contribution to the muons EDM is then simply dFV =
1
FV
2 e Im A , where
AFV
=

2 Aa
LL RR

m
LR RL
23
23 23 + 23
2
2
m ln m ln m
L

RL
23

Aa

LR ln m2
22
L

LL
23
+

LR
23

Aa

LR ln m2
22
R

RR
23
.

(23)

Contours of the maximal possible EDM dmax in the presence of LL and RR flavor
violation are given in Fig. 5. To obtain dmax , is taken to saturate the constraint from de ,
and the sign of the flavor-violating contribution is chosen to add constructively to the
LL,RR
are also taken to maximize d given
flavor-conserving piece. The parameters 23
the constraint from ; we make use of the fact that, subject to the constraint
ax 2 + by 2  c with a, b, c > 0, the product xy is maximized for ax 2 = by 2 = c/2. We
LL,RR
 1/2 so that the mass insertion approximation is valid. We find that
also require 23
flavor violation may enhance d . While the enhancement is not enormous, it does bring
the maximal possible value of d into the range of the proposed experimental sensitivity
in parts of the parameter space.
We have performed a similar analysis for chirality-violating flavor violation. In the case
LR,RL
of nonvanishing 23
, enhancements above the proposed sensitivity are not found.
Note, however, that, to investigate various effects independently, we have assumed real
off-diagonal masses. In fact, however, this division of new physics effects is rather artificial,
as off-diagonal masses need not be real and generically have O(1) phases.
 LL RR 
For concreteness, we consider two cases: in the first, we take Arg 23
23 = and
LR = RL = 0, while in the second, we let LL = RR = 0 and Arg LR RL = . In
23

23
23
23
23 23

378

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

Fig. 5. Contours of dmax in units of 1024 e cm in the presence of flavor violation. Regions consistent
with the observed a at the 1 and 2 levels are also shown. On the left, m) 300 GeV and
tan = 50, and on the right, M2 = 300 GeV and || = 500 GeV. M1 is fixed by gaugino mass
unification. The excluded regions are as in Fig. 4.

Fig. 6. Contours of d (solid) and a (dashed) as functions of , the CP -violating phase present in
 LL RR 
23 =
flavor-violating slepton mass terms. In each pair of contours, the upper is for Arg 23


LR = RL = 0, while the lower is for LL = RR = 0 and Arg LR RL = . We fix
and 23

23
23
23
23 23
m) = 300 GeV, tan = 30, || = 500 GeV, and M2 = 300 GeV, and M1 is fixed by gaugino mass
unification.

both cases, the phase is irrelevant for B( ), as only the magnitudes of the s
enter in Eq. (21). This phase is also not constrained by de , as it has no direct couplings to
the first generation. However, it contributes directly to d , as is clear in Eq. (23).
In Fig. 6 we show the dependence of d on the phase for both cases. We see
that in the LL/RR case, it is easy to achieve values of d 1022 e cm, two orders
of magnitude above the proposed sensitivity. For the LR/RL case, d well above the
proposed sensitivity is also possible. Such values are consistent with all present constraints.

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

379

In particular, note that we have also given values of a including both the flavor-conserving
contribution and the flavor-violating amplitude of Eq. (23). As may be seen in Fig. 6, the
values of are also perfectly consistent with the currently preferred a .

6. Conclusions
The proposal to measure the muon EDM at the level of 1024 e cm potentially improves
existing sensitivities by five orders of magnitude. Such a leap in sensitivity is rare in studies
of basic properties of fundamental particles and merits attention.
In this study we have considered the muon EDM from a number of theoretical
perspectives. We noted that the recent results from the muon (g 2) experiment,
although widely interpreted as evidence for a nonstandard model contribution to a , may
alternatively be ascribed entirely to a nonstandard model contribution to d . In fact, these
results provide the most stringent constraints on the muon EDM at present. Theoretical
prejudices aside, this ambiguity will be definitively resolved only by improved bounds on
(or measurements of) d , such as will be possible in the proposed d experiment.
Considering only the indications from a nonstandard model contribution to a ,
naturalness implies muon EDMs far above the proposed sensitivity. In more concrete
scenarios, however, additional constraints, notably from the electrons EDM and lepton
flavor violation, impose important restrictions. Nevertheless, we have noted a number
of well-motivated possibilities in supersymmetry: nondegeneracy, nonproportionality, and
slepton flavor violation. Each of these may produce a muon EDM above the proposed
sensitivity.
For simplicity, we have focused for the most part on one effect at a time. From a modelbuilding point of view, however, this is rather unnatural. For example, nondegeneracy
of the diagonal elements of the scalar mass matrices is typically accompanied by flavor
violation. At the very least, if the soft masses are diagonal but nondegenerate in a particular
interaction basis, off-diagonal elements will be generated upon rotating to the fermion mass
basis. As noted above, nonproportionality will also typically be accompanied by flavor
violation. In both of these cases, then, the lepton flavor violation leads to new contributions
to d , as well as to new constraints from lepton flavor violating observables. The amount
of flavor violation present is highly model-dependent, and so we have not considered this
in detail. In Section 5.3, however, we have noted that d may be greatly enhanced by two
or more simultaneous effects, leading to values of d 1022 e cm, far above the proposed
sensitivity.
If a nonvanishing d is discovered, it will be unambiguous evidence for physics beyond
the standard model. At the currently envisioned sensitivity, it will also imply naive
scaling violation, with important implications for many new physics models. In addition,
the measurement of nonstandard model contributions to both d and a will provide a
measurement of new CP -violating phases, with little dependence on the overall scale of
the new physics. Such information is difficult to obtain otherwise. Low energy precision
experiments may not only uncover evidence for new physics before high energy collider

380

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

experiments, but may also provide information about the new physics that will be highly
complementary to the information ultimately provided by colliders.

Acknowledgements
We thank R. Carey, G. Kribs, J. Miller, W.M. Morse, Y. Nir, and B.L. Roberts for useful
discussions. This work was supported in part by the US Department of Energy under
cooperative research agreement DF-FC02-94ER40818. KTM thanks the Fermilab Theory
Group (summer visitor program), the Argonne National Lab. Theory Group (Theory
Institute 2001), the Aspen Center for Physics, and D. Rainwater and M. Schmaltz for
hospitality during the completion of this work.

Appendix A
The functions KN and KC of Eq. (15) are






KN x 2 , y 2 , z2 = J5 x 2 , x 2 , y 2 , z2 , z2 + J5 x 2 , x 2 , y 2 , y 2 , z2 ,
(A.1)
 2 2 2
 2 2 2 2
 2 2 2 2
 2 2 2
KC x , y , z = 2I4 x , y , z , z + 2I4 x , y , y , z KN x , y , z , (A.2)
where the loop functions Jn and In are defined iteratively [16] through






1 
2
In x12 , . . . , xn2 = 2
In1 x12 , . . . , xn1
In1 x22 , . . . , xn2 ,
x1 xn2
 2





2
+ xn2 In x12 , . . . , xn2 ,
Jn x1 , . . . , xn2 = In1 x12 , . . . , xn1
with


I2 x12 , x22



x12
x12
x22
x22
1
.
=
ln
+
ln
16 2 x12 x22 2 x22 x12 2

(A.3)
(A.4)

(A.5)

Their arguments are the gaugino masses M1 and M2 , the Higgsino mass parameter ,
and the scalar soft supersymmetry breaking masses




m2) = m2LL )) + 12 + sin2 W m2Z cos 2,
L


m2) = m2RR )) sin2 W m2Z cos 2,
R


m2 ) = m2LL )) + 12 m2Z cos 2.
(A.6)
References
[1] See, e.g., in: Y. Kuno, W.R. Molzon (Eds.), Proceedings of the Joint US/Japan Workshop on
New Initiatives in Muon Lepton Flavor Violation and Neutrino Oscillation with High Intense
Muon and Neutrino Sources, Honolulu, Hawaii, October 26, World Scientific, 2000.
[2] Y.K. Semertzidis et al., hep-ph/0012087.
[3] J. Bailey et al., CERNMainzDaresbury Collaboration, Nucl. Phys. B 150 (1979) 1.

J.L. Feng et al. / Nuclear Physics B 613 (2001) 366381

381

[4] H.N. Brown et al., Muon g 2 Collaboration, Phys. Rev. Lett. 86 (2001) 2227, hep-ex/0102017.
[5] M. Davier, A. Hocker, Phys. Lett. B 435 (1998) 427, hep-ph/9805470.
[6] For recent reviews of standard model contributions to a , see, e.g., W.J. Marciano, B.L. Roberts,
hep-ph/0105056;
K. Melnikov, hep-ph/0105267.
[7] For recent work, see: T. Ibrahim, U. Chattopadhyay, P. Nath, Phys. Rev. D 64 (2001) 016010,
hep-ph/0102324;
A. Bartl, T. Gajdosik, E. Lunghi, A. Masiero, W. Porod, H. Stremnitzer, O. Vives, hepph/0103324;
M. Graesser, S. Thomas, hep-ph/0104254;
Z. Chacko, G.D. Kribs, hep-ph/0104317.
[8] T. Ibrahim, P. Nath, hep-ph/0105025.
[9] K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. Lett. 85 (2000) 5064, hep-ph/0006329.
[10] T. Blazek, S.F. King, hep-ph/0105005.
[11] J.L. Feng, K.T. Matchev, Phys. Rev. Lett. 86 (2001) 3480, hep-ph/0102146.
[12] B.L. Roberts, private communication.
[13] E.D. Commins, S.B. Ross, D. DeMille, B.C. Regan, Phys. Rev. A 50 (1994) 2960.
[14] K.S. Babu, S.M. Barr, I. Dorsner, hep-ph/0012303.
[15] M. Dine, E. Kramer, Y. Nir, Y. Shadmi, Phys. Rev. D 63 (2001) 116005, hep-ph/0101092.
[16] T. Moroi, Phys. Rev. D 53 (1996) 6565;
T. Moroi, Phys. Rev. D 56 (1996) 4424, Erratum, hep-ph/9512396.
[17] Y. Nir, N. Seiberg, Phys. Lett. B 309 (1993) 337, hep-ph/9304307.
[18] E. Dudas, S. Pokorski, C.A. Savoy, Phys. Lett. B 369 (1996) 255, hep-ph/9509410.
[19] E. Dudas, C. Grojean, S. Pokorski, C.A. Savoy, Nucl. Phys. B 481 (1996) 85, hep-ph/9606383.
[20] P. Brax, C.A. Savoy, JHEP 0007 (2000) 048, hep-ph/0004133.
[21] Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562, hepex/9807003;
S. Fukuda et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 85 (2000) 3999, hepex/0009001.
[22] J.L. Feng, Y. Nir, Y. Shadmi, Phys. Rev. D 61 (2000) 113005, hep-ph/9911370.
[23] S. Ahmed et al., CLEO Collaboration, Phys. Rev. D 61 (2000) 071101, hep-ex/9910060.
[24] J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Lett. B 391 (1997) 341;
J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Lett. B 397 (1997) 357, Erratum, hepph/9605296.
[25] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hepph/9604387.
[26] J.L. Feng, K.T. Matchev, Y. Shadmi, in preparation.

Nuclear Physics B 613 (2001) 382406


www.elsevier.com/locate/npe

Enhanced electroweak corrections to inclusive


boson fusion processes at the TeV scale
Marcello Ciafaloni a,b , Paolo Ciafaloni c , Denis Comelli d
a Dipartimento di Fisica, Universit di Firenze, INFN, Sezione di Firenze, Italy
b CERN, Geneva, Switzerland
c INFN, Sezione di Lecce, Via per Arnesano, I-73100 Lecce, Italy
d INFN, Sezione di Ferrara, Via Paradiso 12, I-35131 Ferrara, Italy

Received 10 April 2001; accepted 27 July 2001

Abstract
Electroweak radiative corrections with double-log enhancements occur in inclusive observables
at the TeV scale because of a lack of compensation of virtual corrections with real emission due
to the non-abelian (weak isospin) charges of the accelerator beams. Here we evaluate such Bloch
Nordsieck violating corrections in the case of initial longitudinal bosons, which is experimentally
provided by boson fusion processes, and is related to the GoldstoneHiggs sector. All four states
of this sector are involved in the group structure of the corrections, and cause in particular a novel
double log effect due to hypercharge mixing in the longitudinal states. We study both the light- and
the heavy-Higgs cases, and we analyze the symmetry breaking pattern of the corrections. The latter
turn out to be pretty large, in the 510% range, and show an interesting Higgs mass dependence, even
for processes without Higgs boson in the final state. 2001 Published by Elsevier Science B.V.
PACS: 12.15.Lk

1. Introduction
Recent developments [19] in the treatment of EW radiative corrections at high energies
have brought attention to the fact [7,8] that double-log enhancements of infrared/collinear
origin are present even in inclusive observables associated to hard processes at the
2 ], signal
TeV scale. Such enhancements, involving the effective coupling W log2 [s/MW
the existence of a lack of compensation of the mass singularities associated with the
2 /s 0 limit, due to the non-abelian weak-isospin charge of the initial states (that
MW

Work supported in part by EU QCDNET contract FMRX-CT98-0194 and by MURST (Italy).


E-mail addresses: ciafaloni@fi.infn.it (M. Ciafaloni), paolo.ciafaloni@le.infn.it (P. Ciafaloni),
comelli@fe.infn.it (D. Comelli).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 7 5 - 3

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

383

is, electrons or protons) provided by the accelerator. In other words, the BlochNorsieck
(BN) theorem, valid for abelian theories, is here violated.
For instance, if we take the example of ee+ annihilation into hadron jets in a next linear
collider (NLC [4]), we found [7] that EW corrections to the total cross section at the TeV
threshold are about 5% compared to the strong corrections of 3% and may even be larger
if polarization and angular dependence are considered. The simple fact which underlies
this surprising result is that the emission of a W boson by the incoming electron changes
it into a neutrino whose hadron production cross section off positrons is different from the
electronpositron one, and actually happens to be larger than the latter. This causes a lack
of compensation with virtual corrections, and thus the double-logarithmic enhancement
which turns out to be large in the TeV region. Evaluation of such effects for the next
generation colliders is thus needed.
It is amusing to note that the electroweak sector of the standard model is the first instance
where such non- abelian effect is likely to be seen. In fact, since QCD is in a confined phase,
no colored asymptotic states are allowed: as a consequence, colour averaging eliminates the
double-log singularities and reduces the violation of collinear factorization to higher twist
level [10]. In the EW case, on the other hand, symmetry breaking allows the (non-abelian)
initial states to be prepared by electromagnetic forces, with ensuing violation of the BN
theorem and of collinear factorization theorems at leading level. We remark elsewhere [12]
that a similar effect can occur in broken abelian theories too, due to initial state mixing
allowed by symmetry breaking.
The enhancements of EW radiative corrections due to mass singularities were actually
first pointed out in exclusive processes at double [1] and single [2] logarithmic level. In
view of their size, resummations to all orders are available in the literature [3,5,6,9] but we
feel that no well-established treatment of higher order Sudakov logarithms has emerged yet
in the broken symmetry case, especially for the subleading ones. Here we stress again the
point that the inclusive case is special because it does not depend on the photon effective
mass, whose abelian effects cancel out by the BN theorem. Therefore, exponentiation of
leading logarithms can be established in a clearcut way [7].
In a previous paper [8] we have generalized the evaluation of the BN violating logarithms
to the case of initial gauge bosons, by limiting ourselves to transverse polarizations.
Gauge bosons close to the mass shell are provided indirectly by the accelerator in the
form of boson-fusion processes, whose importance increases with the available energy
in comparison with annihilation type processes. We have thus provided a classification
of the non-canceling logs, which are roughly determined by the Sudakov form factor in
the t-channel weak isospin representations occurring in the overlap matrix of the process.
The latter are found from the ClebschGordan couplings of the t-channel isospin to the
weak isospin states of the initial particles, whether doublets (for fermionic initial states) or
triplets (for bosonic ones). This classification is here summarized in Section 2.
In the present paper, we address the problem of longitudinal initial bosons which is
special because the longitudinal projection (Section 3) singles out, by the equivalence
theorem [16], the HiggsGoldstone boson sector, which is sensitive to symmetry breaking
and in particular to the Higgs boson mass MH .

384

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Since we work in the limit where the hard process invariants are in fixed ratios, and
2 , we are allowed in the fermionic and transverse boson cases to assume symmetry
s  MW
restoration for the squared matrix elements. In the longitudinal case we need more care: if
MH  MW (as hinted at by LEP data [13]), then symmetry restoration is safely assumed
for energies above the weak scale (Section 4). On the other hand, if instead MH  MW ,
in the energy region MH   MW the SU(2)L U (1)Y symmetry is broken by the
mass terms, and in general no easy treatment is allowed. For this reason, we shall consider
in detail the case of very small tan W = g /g, because in the W 0 limit the custodial
SU(2)L SU(2)R symmetry plays a key role [17], and allows in fact a relatively simple
treatment of the BN violating double logs (Section 5) in the energy region MH   MW
also.
Finally, we discuss in Section 6 the example of jet production by boson fusion, by
providing both hard cross section and enhanced radiative corrections for the cases of a
light or a heavy Higgs boson. A few details on the collinear cutoffs and on the symmetry
breaking pattern are derived in the appendices.

2. Electroweak double logs in inclusive processes


EW radiative corrections with leading (double-log) enhancements occur in inclusive
observables because of a lack of compensation of infrared logarithms between virtual
corrections and real emission. This is due in turn to the fact that initial states in a (nonabelian) multiplet may change flavour during emission, and thus interaction probability.
Such enhancements, therefore, have been classified [8] according to the weak isospin
properties of the initial states (fermions or bosons) and the resulting picture is summarized
here.
We consider the structure of soft (infrared/collinear) interactions accompanying a hard
standard model process {I pI } {F pF } where the s and ps denote collectively the
flavours/colours and momenta of initial and final states. The corresponding S-matrix can
be written as an operator in the soft Hilbert space H, that collects the states which are
almost degenerate with the hard ones, and as a matrix in the hard labels, with form [7]




S = UFF F as , as SHF I (pF , pI ) UII I as , as ,
(1)
where U F and U I are operator functionals of the soft emission operators as , as .
An inclusive measurement on the hard process at hand involves squaring the matrix
elements in (1) and summing over the degenerate set (pF ) of soft final states. In this
procedure, the final state interaction operators U F cancel out by unitarity, so that the
relevant quantity is the overlap matrix (Fig. 1)


OI I = S 0|UI S H S H UI I |0S ,
(2)
I I

I I

where |0S denotes the state without soft quanta, as is appropriate for the initial state.
It is also convenient to define the hard (tree-level) overlap matrix OH = S H S H , which
corresponds to Eq. (2) if soft radiative corrections are turned off.

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

385

Fig. 1. Overlap matrix.

Note that physical cross sections of initial particles which are eigenstates of flavours
(weak isospin, charge, etc.) involve diagonal overlap matrix elements only. However, off
diagonal elements (I = I ) are also relevant in the case of mixing (as for the and Z
mass eigenstates) and in order to describe radiative corrections. For this reason we consider
a general label assignment in Eq. (2).
The factorization of soft operators assumed in Eqs. (1) and (2) is typical of a double log
treatment, in which collinear singularities are considered in the soft limit, and makes it clear
that the BN violating double logs are due to the initial state interaction. Generalization to
subleading EW logarithms requires instead a critical revision of the collinear factorization
statements which are valid for QCD.
In order to explicitly evaluate the inclusive double logs, we have used [7] the method of
asymptotic dynamics [11] which builds up the U -operators in (1) in terms of an effective
large-time Hamiltonian, that is constructed in turn by an iterative method to all orders in
2 
the gauge couplings. At leading double log level, and in the high energy limit s  MW
MZ2 M 2 , the key ingredient of the asymptotic Hamiltonian is the eikonal soft boson
emission current [11]
J

(k) = g


i

T ai

pi
,
pi k

(k) = g


i

p
yi i ,
pi k

(3)

which describes the emission of the gauge bosons eigenstates Aa , A0 (a = 1, 2, 3) of

momentum k off the fast charges pi of energy Ei s in the soft boson energy region

s  k  M.

386

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Since we work in the high energy fixed angle limit for the hard process, we can
assume that the SU(2)L U (1)Y symmetry is restored in the squared matrix elements

for s  M. Furthermore, since the inclusive double-logs induced by the e.m. charge Q
cancel out by the BN theorem for abelian theories, there is no net effect from the energy
region k < M and we can work with only one cutoff


W = 2 p k M 2
(4)
and with unmixed indices (a, 0) in Eq. (3).
The isospin T a and hypercharge y operators in Eq. (3) act on the flavour indices I
(1 2 ) of the (two) initial particles, which are doubled, because of the square, in the
overlap matrix O1 2 ,1 2 . Because of the complex conjugation in the square, we shall
normally use the representations T (T T ) for doublets on the -indices and antidoublets
on the -indices (antidoublets on the -indices and doublets on the -indices), with the
identifications
1 2 |O(s)|1 2  = 2 2 |O(t ) |1 1  = O|1 2 1 2 ,

(5)

where the bars denote the conjugate representation and the overlap matrix in the first and
the second case is decomposed along the s- and t-channel flavour structure.
Restoration of gauge symmetry means that the gauge charges are conserved, i.e.,








T ai = O
yi = 0 s  M 2
O
(6)
i

and, as a consequence, the eikonal current (3) is conserved too


OJ A (k)k = 0

(A = (a, 0)).

(7)

Residual effects of symmetry breaking come, of course, from the weak scale cutoff M, and
from the fact that initial particles may be mixed, being a superposition of different isospin
or hypercharge eigenstates. By taking into account soft emission off the legs 1 , 1 (with
momentum p1 ) and 2 , 2 (with momentum p2 ) the eikonal current for the overlap matrix
becomes, by the gauge charge conservation (6):




p
p
p1
p1
2
T aL ,
2
Y,
J a (k) = g
(8)
J 0 (k) = g
p1 k p2 k
p1 k p2 k
where T aL = T a1 + T a1 (Y = y1 + y1 ) denotes the total t-channel isospin (hypercharge).
For initial states with definite hypercharge (y1 = y1 and y2 = y2 ) we can set Y = 0.
(But see Section 4 for a relevant example with Y = 0.)
Radiative corrections to the hard overlap matrix (Fig. 2) come from virtual contributions
(connecting two -indices or two -indices) and from real emission (connecting an
-index with a -index). A simple counting shows that a soft boson loop contributes the
eikonal radiation factor
p1 p2
1  a 2
T 2 W
(9)
J (k) = g 2
2
(p1 k)(p2 k) L
integrated over the phase space of the (on shell) soft boson.

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

387

Fig. 2. One loop dressing of the overlap matrix.

The leading asymptotic Hamiltonian is then found by defining an evolution variable


which is conjugated to the asymptotic time and by slicing the k phase space accordingly [11]:
 3

p1 p2
d k 
g2
W ( ).
(k)
T 2 W T 2L L
H( ) =
(10)
3
(2)
2
(p1 k)(p2 k) L
In general, depending on the choice of whether energy [11] or emission angle [14] or kT
variable [15], higher order terms in the effective Hamiltonian may be not negligible. Here
we stress the point that the present problem is commutative, H( ) being proportional to
T 2L at all values of , and thus any definition of will provide the same result, which is
path ordering independent. By integration we find in fact the Sudakov factor



2 2
2
= P e d H( ) = eL(s,M )T L ,
S s, MW
(11)
where we have set LW L(s, M 2 ) 4W log2 Ms 2 , which depends on the Casimir T 2L of
the total t-channel weak isospin of the problem. Because of gauge symmetry, the hard
overlap matrix OH has the form of a sum of projectors of definite (t-channel) SU(2)U (1)
quantum numbers. It is then straightforward to find the resummed expression for O:


2 2
2
OH =
CtL PtL ,
O = eL(s,M )T L OH =
CtL eL(s,M )tL (tL +1) PtL . (12)
tL

tL

The inclusive double logs are thus found for a generic initial state by the following steps:
First, choose an inclusive hard process which is flavour blind (for instance e e+
hadronic jets, or flavoured jets, summed over flavour charges).

388

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Then, calculate the hard overlap matrix and its couplings to t-channel isospin.
Finally, get the radiative corrections by the weights (11) on each isospin state.
We found [8] by this procedure the following classification.
2.1. Fermionic initial states
This is the most interesting case, which includes electron, muon, and proton colliders
above the TeV scale.
Since left-handed fermions have tL = 1/2, we should consider doublet scattering for
both the ff (which is equal to ff) and f f (equal to ff ) cases. Isospin and CP
conservation provides the constraints
11 = 22 = 1 1 = 2 2 ,

11 = 22 ,

12 = 12

(13)

so that there are only four independent cross sections, e.g., 12 , 22 in the ff case, and
12 , 22 in the f f one.
On the other hand, the t-channel states |1 1  occurring in Eq. (5) give rise to T L = 0, 1
for both the ff and f f cases, thus providing again four independent coefficients [8]:
Off = C0 P0 + C1 P1 ,

0 P0 + C
1 P1 .
Of f = C

(14)

Expressions for the resummed cross sections can be directly obtained by projecting
Eq. (12) on the subset of the states appearing in the projectors that, having tL3 = 0,
correspond to physical cross sections. 1 There are two such states in the case at hand,
providing the equations:





ii O tL = 0, tL3 = 0 = ii OH tL = 0, tL3 = 0 ,




ii O tL = 1, t 3 = 0 = ii OH tL = 1, t 3 = 0 e2LW ,
(15)
L

2 label the initial states on leg 1 and 1 in Fig. 1. Namely, we find:


where i = 1, 2, 1,
The singlet T L = 0 state
 
1
+ |21)
= 1 11 + 22
(|12
2
2
corresponding to the average cross sections
1
1
(12 + 22 ),
( + 22 ).
2
2 12
The triplet T L = 1 states whose T 3L = 0 component is
 
1
+ |21)
= 1 11 22
(|12
2
2
corresponding to the cross-section differences
12 22 ,

12 22 .

(16)

(17)

(18)

(19)

1 The elements of the overlap matrix corresponding to physical cross section always involve a given particle

on leg 2 and its own antiparticle on leg 2 in Fig. 1, and have thus all t-channel additive quantum numbers set to
0.

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

389

By Eq. (15), the average cross section has no radiative corrections, while the difference
is suppressed by the form factor in the adjoint representation, e.g.,
 H
 2L
H
W.
e+ e e+ = e
(20)
+ e e+ e
This leads to corrections 1e e+ /eH e+ of order 5% for e e+ annihilation into hadrons at
the TeV threshold.
2.2. Transverse bosons
This case is needed in order to treat boson-fusion processes, in which nearly on-shell
gauge bosons are exchanged. It must be supplemented by the longitudinal case to the
studied below, which is favoured at very high energies.
Since transverse bosons behave as genuine isospin triplets, it is straightforward to derive
the isospin conservation constraints:
++ = ,

3+ = 3 ,

33 = ++ + + 3+

(21)

so that only three independent cross sections are left


= +

( = +, 3, ).

(22)

Correspondingly, the t-channel isospin can take the values T L = 0, 1, 2, leading to three
projectors and three independent coefficients.
The |1 1  states diagonalizing such projectors are:
The singlet state
1
(| +  + | + + |33)
2

(23)

1
(| +  | +)
2

(24)


corresponding to the cross section average 13 .
Three T L = 1 states, whose T 3L = 0 component is

corresponding to the difference + .


Five T L = 2 states, whose T 3L = 0 component is
1
(| +  + | + 2 |33)
6

(25)

corresponding to the cross section combination


+ + 2 3 .

(26)

Due to the weights (11), the average cross section has no radiative corrections, the
difference + is damped by the form factor in the adjoint representation e2LW ,
much as in Eq. (20), and the combination (26) is damped even more strongly, by the form
factor in the T L = 2 representation e6LW .

390

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Fig. 3. Leading eikonal vertices in the longitudinal (a), (b) and scalar (c) sector.

3. Soft emission in the longitudinal sector and equivalence theorem


It is well-known that, in the high energy limit longitudinal polarizations are of type

2L  p /M + O(M/E), and therefore longitudinal boson amplitudes Aa are related to


the Goldstone bosons ones by the equivalence theorem [16]


p a
(27)
p2  M 2 ,
M (p; . . .)  iM a (p); . . . ,
M
where a = 1, 2, 3 labels the SU(2)L vector boson indices and the remaining momentum
and flavour indices of the amplitudes have been dropped. 2 Therefore, the soft emission
properties of the longitudinal bosons are related to the ones of Goldstone bosons, which are
coupled to the Higgs boson in the scalar sector. This shows that the dynamics of soft boson
emission involves actually four real states ( a , h) or the Higgs doublet and its conjugate.
We shall see that depending on the mass parameters of the problem, the doublet/antidoublet
or triplet/scalar description may be best suited, both in general being needed.
In order to better understand this point, let us consider the emission of a soft boson with
isospin index a, Lorentz indices , and momentum k off an energetic vector boson (2 (p))
via the three boson coupling. The amplitude (Fig. 3a) is proportional to


2 (p) (p, k) = 2 (p) g (2 p + k) g (p + 2 k) g (p k)
(28)
 2p 2 (p) 2 (p)p + O(k),

(29)

where use has been made of the condition 2(p) p = 0. By taking into account the
denominator and isospin factors, we then obtain a generalized eikonal emission amplitude
M(b, 2(p); . . .)|a, , k
2 Strictly speaking, we should have written Eq. (27) for the Z-boson instead of its A3 component
ZL
p
3
MZ M (p; . . .)  M( (p); . . .). But using the relation MZ = MW /cW and electromagnetic current

conservation provides Eq. (27) close to the vector boson mass shell.

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406



M
p

2 (p)M(c, 2L (p + k); . . .)
 ig 2bac M(c, 2(p + k); . . .)
p k 2p k

391

(30)

which differs from the customary one of the unbroken theories by the second term in square
brackets, which is purely longitudinal. If 2(p) is a transverse polarization, the second term
is of order M/E with respect to the first one, and is thus subleading. On the other hand, if
2(p) p/M + O(M/E) is the longitudinal polarization, then the latter term cancels one
half the leading term, and leads to a factorized eikonal emission current off longitudinal
bosons:


 
g
p

1+O
.
M(b, 2L (p); . . .)|a, , k  i 2bac M(c, 2L (p + k); . . .)
2
pk
E
(31)
Note the factor of 1/2, relative to the normal eikonal factorization theorem for transverse
polarizations.
The amplitude (31) is not however, the only possible contribution to soft boson
emission, which can occur via a Higgs boson also (Fig. 3b). Taking into account the weak
scale dependence properly, one finds again the factorized eikonal current, with a Higgs
contribution on the rhs:
M(b, 2L (p); . . .)|a, , k

g p 

i2bac M(c, 2L (p + k); . . .) iab M(h(p + k); . . .) .
2 pk

(32)

Since the Higgs boson can occur as a final state as well (Fig. 3c) it is convenient to
generalize Eq. (32) in matrix form on the boson indices B = (b, 0) (a , h), as follows:






p  a 
M B (p); . . . a, , k  g
T L BC M C (p + k); . . .
pk

by thus introducing the weak isospin matrix in the ( a , h) basis





1
1 i2bac iab
T aL =
TVa + Tha ,
0
2 iac
2

(33)

(34)

where (TVa )bc = i2bac . Similarly one can define the eikonal emission current of the B
A0 boson in the form:






p
YBC M C (p + k); . . . ,
M B (p); . . . 0, , k  g
pk

(35)

where

1 3
TV Th3
(36)
2
is the hypercharge matrix in this sector. Eqs. (34) and (36) are the basis for our treatment
of radiative corrections in the following.
It is important to realize that the gauge generators (34) and (36) have a simple structure
within the custodial symmetry group SU(2)L SU(2)R [17]. In fact it is straightforward
Y=

392

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

to check that TV and Th in Eq. (34) satisfy the algebra of the O(4) group
 a b
 a b
 a b
TV , TV = i2 abc TVc ,
TV , Th = i2 abc Thc ,
Th , Th = i2 abc TVc ,

(37)

As a consequence, the SU(2)L (SU(2)R ) generators T aL (T aR ) are provided by


T aL =


1 a
TV + Tha ,
2

T aR =


1 a
TV Tha .
2

(38)

So that [T aL , T bR ] = 0 and Y = T 3R . In this language the basis ( a , h) is just the (1/2, 1/2)
representation of the group O(4) SU(2)L SU(2)R , characterized by T 2L = T 2R = 3/4,
and the matrices (34) and (36) refer to the 4-vector representation.
Of course, the same type of eikonal current (33) arises by direct study of the Goldstone
Higgs sector and of their couplings to soft bosons, as expected from the equivalence theorem (Fig. 3c). The longitudinal projection just provided helps clarifying some aspects of
the problem, in particular the factor 1/2 occurring in (34), and the occurrence of four states
associated with the longitudinal bosons. It is sometimes useful to arrange such four states
of the (1/2, 1/2) representation in the complex form of the doublet/antidoublet matrix

 



1
1
1 1
h + i3
0 i+
2 + i1
=
=
= (h + ia a ) =
2 2
i 0
2
2 2 + i1 h i3
(39)
UL UR

under the SU(2)L SU(2)R group. The indices


which transforms as
2 here refer to the transformation properties under SU(2)L .
1, 2, 1,

4. Radiative corrections in the light Higgs case


For Higgs bosons and vector bosons nearly degenerate, the gauge symmetry of the
squared matrix elements is restored in the full high-energy region  M = MW  MH
(the common weak scale), while the photon scale plays no role because of the BN theorem
for QED. Therefore, we can assume SU(2)L U (1)Y symmetry for both the hard overlap
matrix and for enhanced radiative corrections, as already discussed in Section 2.
Since the longitudinal bosons are related to the Higgs doublets by the equivalence
theorem (Section 3), and the latter have tL = 1/2, it would seem that we are in
the same case as that of fermion and antifermion doublets. However, a peculiarity
of the longitudinal/Higgs sector is the fact that the physical Higgs boson is by itself
a manifestation of (low-energy) symmetry breaking, being a coherent superposition
(see (39))
+ |2
|0  + |0  |1
=

(40)
2
2
of doublet states of opposite hypercharge. Since the state (40) has no definite hypercharge,
it happens that overlap matrix elements having non-vanishing t-channel hypercharge Y =
1 occur in physical cross sections initiated by neutral states, besides the normal ones with
Y = 0.
|h =

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

393

The fact above has no counterpart in the light fermion case 3 because it is not possible to
prepare a coherent superposition of a fermion and of its antifermion, due to exact fermion
number conservation. On the other hand, a similar phenomenon occurs in the customary
case of B W3 mixing yielding and Z mass eigenstates, as already discussed in
Ref. [8]. In the latter case the mixing is non-abelian, involving tL = 1 and tL = 0 states,
while in the Higgs case the mixing is abelian, involving y = 1/2 hypercharge states.
A peculiar consequence of this fact is that violation of BN theorem can occur in broken
abelian theories too [12].
By using the definitions (39), the gauge symmetry constraints (13) in the longitudinal
/Higgs sector can be written as follows:
= ++ ,

+3 = 3 = +h = h .

(41)

In the purely neutral sector we find, on the other hand, the mixing phenomenon, in the
form:
1
1
33 = hh = (++ + + ) + Re(O),
2
2
1
1
3h = (++ + + ) Re(O),
(42)
2
2
where O O(0 0 0 0 ) and its conjugate denote the Y = 1 and Y = 1 overlap
matrix elements mentioned before.
It is then convenient to refer to a fixed target (the + state, say) in order to enumerate the
independent cross sections. Due to Eqs. (41) we find at once three of them
++ , + , 3+

(43)

similarly to the transverse polarization case and in addition to the neutral overlap Re O just
defined. Finally, we also need another off diagonal matrix element
1
I Im Oh3 Im(O3+h+ ) = (0+ 0 + )
(44)
2
in order to express all possible initial state dependences. All these five quantities, except
Re O, have Y = 0 in the t-channel.
We shall now compute the enhanced (BN violating) radiative corrections to the
observables (42) and (43). Following our method (Section 2), we have to (i) define
the hard process and the inclusive (flavour blind) observable we are interested in (ii)
derive the lowest order (tree level) hard overlap matrix OH1 2 ,1 2 and (iii) derive the
effective Hamiltonian describing the enhanced radiative corrections, for the given initial
state configuration.
Symmetry considerations restrict the hard overlap matrix, according to the possible
t-channel quantum numbers whose eigenstates are constructed out of the |1 1  states
defined before. In the longitudinal/Higgs sector 1 and 1 span the four states of the
3 Note however that mixing occurs in the case of transversely polarized fermions, which are indeed a
superposition of states of opposite chirality, and of different gauge quantum numbers [12].

394

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

(1/2, 1/2) representation, so that the |1 1  states can be classified as follows (see
Appendix B for the explicit expressions):
Eight (Y = 0) states, which can be grouped according to left isospin representation,
and CP = 1 eigenvalues. We find thus the projectors:
T CP (m) = 0 (1),

1 (3),

0+ (1),

1+ (3),

(45)

corresponding to four independent coefficients, each with the multiplicity m of states


in parenthesis. Such four projectors are in close correspondence with the ones for the
fermionic doublets discussed in Section 2. Note that the CP eigenstates +1 (1) are
just symmetric (antisymmetric) under doubletantidoublet interchange.
Four (Y = 1) and four (Y = 1) states, which are interchanged by CP and must thus
be degenerate, which can be classified according to left isospin representation:
T (m) = 0(2),

1(6)

(46)

corresponding to two more independent coefficients.


Actually only five out of the overall six independent coefficients are coupled to the
observables enumerated in Eqs. (43) and (44) (the one corresponding to T L = 0 and Y =
Q = 1 being decoupled from physical states by the e.m. charge superselection rule).
The effective Hamiltonian describing radiative corrections is found by exactly the same
method as in Section 2, with the important difference that now we can have a non-vanishing
t-channel hypercharge, and we thus obtain


W ( ) T 2L + tg 2 W Y 2
H( ) = L
(47)
from which the t-channel Sudakov factors

 2


2
S(T L , Y ) = exp LW s, MW
T L + tg 2 W Y 2

(48)

are found for each one of the states (45) and (46).
In order to assign the Sudakov weights to the observables (41) and (44) we need to
relate them to the explicit form of the states which diagonalize the projectors (45) and (46).
A direct construction of them is obtained from their classification according the SU(2)L
SU(2)R group; this construction is explicitly worked out in Appendix B. From the direct
product of two (1/2, 1/2) representations we obtain the following representations
 


1 1
1 1
,

,
= (0, 0) (0, 1) (1, 0) (1, 1)
(49)
2 2
2 2
which are related to those in (45) and (46) as follows:
The state (0, 0) is the 0 state and corresponds to the average cross section
+ + + 0 + 0 = + + + 3 + h = + + + 23 + 23 , (50)
where we introduce the notation + in which the + target is understood. The
identification (50) is obtained by the explicit construction
1
1
+ |21
+ |12
|21)

+ |00)

= (| +  + | + + |00
|0; 0 = (|12
2
2
(51)

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

395

1
= (| +  + | + + |33 + |hh)
(52)
2
which has, by inspection, tL = tR = 0, and is also antisymmetric into the doublet
Eq. (50) corresponds to the matrix element + |O|0; 0.
indices interchange i i.
The three states (0, 1) are split according to the hypercharge Y = T 3R into (i) the two
degenerate tL = 0, Y = 1 states in (46) and (ii) the symmetric tL = Y = 0 state
in (45). The latter corresponds to the combination
+ + 2 Im Oh3 1 + 2I,

Oh3 = h + |O|3+,

0+

(53)

and to the state 0+ , provided by


|0; tR = 1, Y = 0
1
1
+ |21
|12
+ |21)

|00).

= (|12
(54)
= (| +  | + + |00
2
2
The three states (1, 0) are the antisymmetric 1 states, whose tL3 = 0 term is the only
one related to cross-sections, and corresponds to the combination
+ 2 Im Oh3 1 2I,

(55)

and to the state


3

1, t = 0; 0
L
1
1
|21
+ |12
+ |21)

+ |00).

= (|12
(56)
= (| +  | + |00
2
2
The nine (1, 1) states are split, according to the hypercharge quantum number into (i)
six degenerate tL = 1, Y = 1 states occurring in (46) and (ii) three Y = 0, 1+ states
occurring in (45), whose tL3 = 0 term corresponds to the cross section
+ + 0 0 = + + 3 h
= + + 23 23 ,

1+

(57)

and to the state


3

1, t = 0; 1, t 3 = 0 = 1 (|12
|21
|12
|21)

L
R
2
1
|00).

= (| +  + | + |00
2

(58)

On the other hand, the state corresponding to the overlap O(0 0 0 0 ) in the
in (45).
neutral sector has tL3 + tR3 = Q = 0 and corresponds to the states |00 or |0 0
Other states with Y = 0 give rise to Q = 0 and therefore cannot be coupled to physical
cross sections.
The cross-section assignments just provided are summarized in Table 1(a, b).
We are now in a position to provide the Sudakov evolution in Eq. (47) for each one of
the cross section combinations just introduced. By using the relevant (tL , Y ) assignments,

396

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Table 1
t-Channel quantum number assignments of various cross section combinations. (a) Invariants, (b)
SU(2)L U (1)Y states (MH  MW ) and (c) SU(2)V states (MH  MW ). The unphysical Q = 0
states have been omitted

a
a
b, c
b
c
b
c

(tL , tR )

tLCP

Y2

TV

Cross section

(0,0)
(0,1)
(1,0)
(1,1)
(1,1)
(1,1)
(1,1)

0
0+
1
1+

0
0
0
0

0
1
1

+ + + 3 + h
+ + 2 Im Oh3
+ 2 Im Oh3
+ + 3 h
+ + 2 3
2 Re O
+ + + 3 3h

2
0

we find
+ + + 23 = H + 23H ,
h = 3

 H
H 2LW
+ + 23 = 23 e
,
+ + 2I = 1H + 2I H ,


+ 2I = 1H 2I H e2LW ,
Re O = Re OH eLW (2+tg W ) ,
1
1
3h = (+ + ) Re O,
2
2
1
1
33 = hh = (+ + ) + Re O
(59)
2
2
from which all radiative corrections can be derived once the hard (tree level) cross sections
are known.
2

5. The heavy Higgs case and symmetry breaking pattern


If the Higgs mass happens to be much larger than the vector boson mass, then in the
energy region MH   MW gauge symmetry is broken by the mass spectrum even if the
2 . Therefore, if we want to keep track of the leading
overall energy of the process is s  MH
2 /M 2 ] or the customary ones log[s/M 2 ], we
powers of all logarithms, whether log[MH
W
W
end up with a process with two mass scales, in which symmetry breaking causes a lack of
conservation of the gauge current and therefore requires a more refined treatment of the
BN violation logs and of their scales.
Since we feel that despite recent efforts [9] the direct diagrammatic calculation of
higher order Sudakov logarithms in the many scale, broken symmetry case, is not yet fully
understood, we prefer to treat here the particular case g /g 0, in which the custodial
symmetry SU(2)L SU(2)R of the longitudinal/Higgs sector plays a key role, by providing
a conserved current which considerably simplifies the problem. Note in fact that, at tree

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

397

2  M 2 , the gauge symmetry is still valid and


level and in the high energy limit s  MH
W
is actually extended to the full SU(2)L SU(2)R group in the W 0 limit. On the
other hand, radiative corrections in the energy region MH   MW break the gauge
symmetry, but keep the global custodial symmetry SU(2)V SU(2)L SU(2)R which
then provides the conserved current we are interested in.
In order to understand this point, recall that the infrared/collinear behavior is regulated
by the boson emission current (33) where the weak isospin charge is written as the sum of
its vector component (1/2TVa ) and of a Higgs contribution. In order to take into account
the mass cutoffs in the denominators, we write the current in Eq. (33) in the more precise
form



p
p 1  a
1 a
a
a
a
T W + Th H = g
J (k) = g
T L H + TV W (1 H ) ,
pk 2 V
pk
2

(60)
where



i = 2p k Mi2 ,

i = W, H

(61)

denote the mass cutoffs.


It is apparent from Eq. (60) that the soft boson insertion current is the full gauge current
2 , while it becomes the SU(2) current (with coupling g/2) in the region
for 2p k > MH
V
2
2 . Since each current is conserved in the corresponding validity region,
MH > 2p k > MW
the evolution of radiative corrections follows the same pattern as that outlined in Section 2,
and the evolution Hamiltonian derived from the current (60) becomes

TV2 
W ( ) ,
H ( ) L
(62)
L
4
H ( ), (L
H ( ) L
W ( )) contribution is defined with the phase space slicing
where the L
in Eq. (10), following the cutoff structure H , (H W ) in Eq. (60). After -integration,
2 /M , the two exponents turn out to be (Appendix A):
for E > MH
W

2 
W
W
2 s
2 s
2 MH
log
LW =
,
LH =
2 log
LW 2l h . (63)
log
2
2
2
4
4
MW
MW
MW
H ( )T 2L +
H( ) = L

Note that, since TVa = T aL + T aR , T aL transforms as a vector under TV rotations, and


therefore T 2L transforms as a scalar, i.e., it is invariant. It follows that the two terms in (62)
commute between each other, and therefore [H( ), H( )] = 0 thus making the Sudakov
exponent independent of the type of phase space slicing (whether in or in kT ). By
integration we then obtain




TV2
2
2
2
(LW LH ) .
S s, MH , MW = exp T L LH
(64)
4
Note finally that the Hamiltonian (62) commutes with TVa (but not with T aL ), so that the
overlap matrix radiative corrections are diagonalized by SU(2)V , as anticipated before.
In practice, explicit projectors and cross section evolution are constructed as follows.
The hard overlap matrix possesses full SU(2)L SU(2)R symmetry, and therefore is

398

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

classified directly according to the representation occurring in the direct product (49),
corresponding to four independent coefficients. If radiative corrections are turned on, the
symmetry is broken to SU(2)V and the corresponding states are classified according to the
Casimir TV2 . Eigenstates of TV2 are trivially constructed for the cases
(a) T 2L = TV2 = 0, CP = 1,
(b) T 2L = 0, TV2 = T 2R = 1, CP = +1,
(c) T 2L = TV2 = 1, CP = 1 discussed before, because they are directly provided by
Eqs. (51), (54), (56), respectively.
In the case
(d) T L = T R = 1 the degeneracy is resolved by the vector sum TVa = T aL + T aR , so that
we can have TV = 0, 1, 2. By straightforward algebra (Appendix B) we find
(d.0) The T L = 1, TV = 0 state


1
1

|11 |22 (|12 + |21 + |12 + |21)


2
3
1
+ | +  + | + |00
|00)

= (2|00 2|0 0
2 3
corresponding to the cross section combination
+ + + 3 3h .

(65)

(66)

|22) =
= 0 term is
(|1 1
(d.1) Three T L = TV = 1 states whose Q =
i 2
1
0
|00) corresponding to an overlap of the type Im O(0 0 )
(|0
0
0
i 2
which does not occur in physical cross sections.
(d.2) Five T L = 1, TV = 2 states, whose TV3 = 0 term is
TV3

1
+ |22 |12
|21
|12
|21)

(|1 1
6
1
+ | +  + | + |00
|00)

= (|00 + |0 0
6
corresponding to the cross section combination
+ + 23 .

(67)

(68)

The cross-section assignments just provided are summarized in Table 1(a, c).
Correspondingly, the independent cross sections are + , , 3 , h , while the neutral
sector cross sections can be expressed in terms of the four preceding ones by the
relationships
h3 = h+ = h ,
33 = + + 3 ,
hh = + + + 3 2h .

(69)

The relations on the first line of Eq. (69) are the simple SU(2)V constraints already used in
Section 2. The last one is obtained by straightforward algebra by using the eigenstate (65)
on the T L = 1, TV = 0 projectors (Appendix B).

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

399

The BN violating corrections are finally found by assigning the Sudakov form factors
(64) to each of the cross section combinations found before. We obtain
+ + + 3 + h = H + 23H ,


+ + + 3 3h = H 23H e2LH ,
 1

+ + 2I = 1H + 2I H e 2 (LW LH ) ,


3
+ 2I = 1H 2I H e2LW + 2 (LW LH ) ,


1
+ + 23 = H 2 3H e2LW + 2 (LW LH )

(70)

from which each cross section can be derived.

6. Phenomenological applications
Before applying the radiative corrections just found to some explicit phenomenological
cases it is interesting to solve the systems (59) and (70), and to investigate their asymptotic
limit LW . In such a limit the values of the evolved cross sections with radiative
corrections will depend only on some invariant combinations that do not evolve, i.e.,
are free of double-logs.
For the case with sin W = 0 and MW  MH we have the two invariants
+ + 2 I
+ + + h + 3
0 ,
1
(71)
4
4
with the constraint h = 3 . The first one corresponds to the cross-section average over
the four Goldstone/Higgs states, and the second is the difference between the doublet and
antidoublet averages. As a consequence, all cross sections converge to a combination of
such two invariants:


0H 1H ,
3 , 33 0H ,
I 1H ,
Re O 0,

(72)

where the superscript H means that the invariants are evaluated at Born level.
MH we also have the same two invariants,
For the case with sin W = 0 and MW
but now we find an extra MH dependence in the limit LW : 4


0H el h 1H ,
3 , 33 0H ,
I el h 1H .

(73)

4 We stress that the limit that we are taking is s with M , M fixed. In such a case the combination
H
W
W
2 /M 2 ] 2l is finite.
log2 [MH
LH LW = 2 4
h
W

400

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

Therefore, in the extra limit MH  MW all the physical cross sections would converge to
the invariant 0H , i.e., the average over all states.
The evaluation of the ratios ij /iH give very easily an idea of the strength of the
evolution of the cross section ij under investigation. For instance, the cross section 33 is
of interest, because it can be identified in an e+ e collider by a double tagged experiment
where the two final electrons both have transverse momenta |k |2  M 2 . By Eq. (59) it
has the expression
33 =




1 H
2
H
H
+ 23+
+ 2 Re OH eLW (2+tan W )
+ e2LW H 23+
4

(74)

in the case MH M, and an analogous one in the case MH  M. The limit 33 0H


is obvious at large energies, while the full behavior depends on the explicit values of the
Born cross sections.
Now let us discuss the example of WL WL hadronic jets. This cross section is relevant
to describe hadron jet production at NLC by boson fusion, a process which becomes
competitive with the annihilation process at high energies.
The Born cross sections for WL WL q q are well-known [18]:
H
H
d33
d++
=
= 0,
d cos
d cos
H
2 N N
2
d3+
W
dI H
c f sin
=
=
,
d cos
d cos
8s
8
H
2 N N
2
2
4
d+
W
c f sin (9 18 sin W + 20 sin W )
=
.
d cos
8s
72 cos4 W

(75)

The corresponding evolution, e.g., for 33 in the case MH M becomes, by Eq. (74)
 H


1 H
H
H LW (2+tan2 W )
2+
23+
e
+ + 23+ + e2LW +
4

1 H
H
2LW .
+ 23+ + tan2 W +
+
4

33 =

(76)

In this case 33 becomes non-vanishing for s > M 2 because of radiative corrections,


and 33 /0H 2LW is roughly of the order 10% at the TeV threshold. This provides
an estimate of the double logs, which are therefore of the same order as for fermions and
transverse bosons initial states.
In Fig. 4 we plot the ratios ij /0H (sin W = 0, MH  MW , cos = 0) over a huge
energy range to show the convergence speed of the evolved cross sections.
In Fig. 5 to the left we plot the same ratios as before but with three different MH =
100, 500, 1000 GeV cases (sin W = 0, MH > MW , cos = 0). To the right we plot the
ratio between a cross section calculated with MH = 100 GeV over the same cross section
calculated with MH = 1000 GeV to show the Higgs mass dependence as a function of
the energy. This figure appears interesting from the Higgs physics point of view: while
the hard W W ff cross section is independent of the Higgs mass at high energies, the
evolved ones instead show a pronounced effect coming from the exponential el H which
tends to depress the corrections. This means that also completely inclusive cross sections

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

401

Fig. 4. WL WL f f inclusive cross section for MW  MH .

Fig. 5. WL WL f f inclusive cross sections for MW < MH and sin W = 0. On the left side
the energy evolution of the cross section is plotted for MH = 100, 500, 1000 GeV. For each set of
three curves, the bottom curve corresponds to MH = 1000 GeV in the case of ++ , 33 and to
MH = 100 GeV in the case of + (3+ is MH -independent). On the right we plot the energy
dependence of the ratio between the cross sections calculated with MH = 100 GeV and the ones
calculated with MH = 1000 GeV (3+ coincides with the x-axis).

402

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

not involving explicit Higgs production show enhanced Higgs mass dependence in their
radiative corrections.
A final comment is in order concerning physical cross sections: these can be obtained
through a convolution of the cross sections we study here with the longitudinal bosons
luminosities in the initial beam (electron, proton). The final outcome is strongly dependent
on the type of beam (whether leptonic or hadronic) and on the particular process under
study; a detailed analysis of this kind is outside the scope of the present paper. In any
case we think that the example discussed here shows the importance of this kind of
new standard model effects related to the violation of the BlochNordsieck theorem
in inclusive cross-sections at very high energies.

7. Conclusions
Our study of the longitudinal initial states occurring in boson fusion processes has
shown that, in this case, enhanced radiative corrections have a number of interesting and
sometimes surprising features.
First of all, soft interactions involve all four states of the Higgs sector, and this makes
the group-theoretical classification of the corrections somewhat more involved than in
previously analyzed cases (Section 2). We have provided here (Table 1) the combinations
of cross sections with well-defined form factors, for both MH MW and MH  MW , and
we have studied their radiative evolution.
Secondly, we find important abelian double logs [12], due to the breaking of weak
hypercharge, besides the non-abelian ones. This is surprising at first sight, and is due to
the occurrence of off-diagonal overlap matrix elements, corresponding to a non-trivial
hypercharge in the t-channel (Eq. (59)). In turn, this happens because the longitudinal and
Higgs states are mixed, i.e., are not hypercharge eigenstates.
Finally, the quantitative effects are sizeable, in the 510% range at the TeV threshold as
for the fermionic and transverse bosons initial states investigated previously. Furthermore,
because of the enhanced radiative corrections, the Higgs mass dependence is not negligible
(Fig. 5) even for processes without Higgs bosons in the final states, that are therefore
insensitive to the Higgs mass at tree level.
On the whole, we have now provided a fairly complete description of the enhanced EW
corrections for fermionic and bosonic initial states, which can now be used for detailed
predictions of hard inclusive observables, at the double log level, at future accelerators
energies. Of course, this is not enough: we actually need a generalization of collinear
factorization to broken gauge theories, in order to account for single logarithms as well.

Appendix A. Soft radiation coefficients


H and of their
W, L
Here we derive the expansions of the evolution coefficients L
-integrals for the explicit choice = and = kT of the evolution variable.

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

We start considering the effect of the collinear cutoff




2
H = 2pk MH

403

(A.1)

on one of the incoming momenta (either p = p1 or p = p2 ), and the corresponding


calculation of the evolution Hamiltonian in Eq. (10).
By taking either or kT fixed the constraint (A.1) is written as
 


1
Max M,
MH < < 1 ( = ),
(A.2)

M
 2

kT s
kT < < Min
(A.3)
, s
( = kT ).
2
MH
By then integrating the eikonal radiator factor in Eq. (10) over the region (A.2) we obtain

(E s)




E 
2
2
LH = 2 log E M 2 MH
+ 2 log 2 MH
M 2 E
W
M
MH

(A.4)

for M < = < E, and


E
kT E

LH = 2 log (kT MH ) + 2 log 2 (MH kT )
W
kT
MH

(A.5)

2 /M, while L
W is given by the same expressions in the
for M < = kT < E, E > MH
MH = M limit.
Finally, by -integration, we obtain in both cases


M2
W
2 E
2 MH
2 log
LH =
(A.6)
log
, E> H

M
M
M

which corresponds to Eq. (62) in the text. In the intermediate energy region MH < E <
2 /M we get instead
MH
LH = 2

E
W
log2
,

MH

MH < E <

2
MH
M

(A.7)

as expected for a smooth E = MH limit.

Appendix B. Projectors

The overlap matrix can always be written as a sum of projectors O = Ci Pi . While the
coefficients Ci depend on the dynamics and therefore on the physical process considered,
the projectors Pi can be classified and explicitly constructed according to the relevant
symmetry. We choose to classify the states according to their SU(2)L SU(2)R symmetry
properties. This symmetry acts on the representation (1/2, 1/2) as follows:








 = 1 1 exp iLa a c exp iRb b
()
(B.1)
2 2

404

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

so that using the basis |tL , tL3 ; tR , tR3  we can do the following identification:




1 1 1 1
1 1 1 1


|1 = , ; , ,
|2 = , ; , ,
2 2 2 2
2 2 2 2




1 1 1 1
1 1 1 1

|1 = , ; , ,
|2 = , ; , .
2 2 2 2
2 2 2 2

(B.2)

Then the states coming from the direct sum TL = tL1 + tL1 and TR = tR1 + tR1 generate the
representations (1/2, 1/2) (1/2, 1/2) = (0, 0) + (1, 0) + (0, 1) + (1, 1) that we label as
|TL , TL3 ; TR , TR3 . Explicitly, the states with (TL , TR ) = (1, 1) are given by:

|1, 1; 1, 1 = |2 2,

|1, 1; 1, 0 =

|1, 1; 1, 1 = |22,

|1, 1; 1, 1 = |1 1,
|1, 1; 1, 1 = |11,
+ |2 1

|1 2
|1, 0; 1, 1 =
,

2
|12 + |21
|1, 0; 1, 1 =

.
2

+ |22

|22
,

|1, 1; 1, 0 =

+ |11

|11
,

|1, 0; 1, 0 =

+ |12
+ |21
+ |21

|12
,
2
(B.3)

The states (TL , TR ) = (1, 0) are


|22

|22
,

2
|11

|11
|1, 1; 0, 0 =

|1, 1; 0, 0 =

|1, 0; 0, 0 =

|12
+ |21
|21

|12
,
2
(B.4)

while those (TL , TR ) = (0, 1) are


|21

|12
,

2
|12 |21
|0, 0; 1, 1 =

.
2

|0, 0; 1, 1 =

|0, 0; 1, 0 =

+ |12
|21
|21

|12
,
2
(B.5)

Finally, the singlet (TL , TR ) = (0, 0) is


|12
|21
+ |21

|12
.
(B.6)
2
In the MH  MW case, the gauge symmetry SU(2) U (1) is the relevant one, and we
can classify the states according to the (TL , TR3 = Y ) possible values; however this does
not provide a complete classification of the states since U (1)Y is an abelian symmetry. For
instance, we have two states corresponding to TL = 0, Y = 0; one of them (|0, 0; 0, 0 in
Eq. (B.6)) is CP-odd while the other (|0, 0; 1, 0 in (B.5)) is CP-even. We can then use CP
invariance to further classify the states, obtaining:
|0, 0; 0, 0 =

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

405

Eight (Y = 0) states, whose multiplicity m is written in parenthesis:




TL , Y 2 CP (m) = |0, 0 (1),

|0, 0+ (1),

|1, 0 (3),

|1, 0+ (3).

(B.7)

Four (Y = 1) and four (Y = 1) states, which are interchanged by CP and must thus be
degenerate:


TL , Y 2 (m) = |0, 1(2),

|1, 1(6).

(B.8)

We have then 6 independent projectors, classified by their SU(2) U (1) and CP


properties; 4 of them correspond to Y 2 = 0 and are given in (B.7) and the remaining
2 with Y 2 = 1 are given in (B.8). The subset of states corresponding to physical cross
sections (with TL3 = Y = 0) can be written in terms of the ones appearing in (B.3)(B.6) as
follows:


TL = 0, T 3 = T 3 = 0 + = |0, 0; 1, 0,
L
R


TL = 1, T 3 = T 3 = 0 + = |1, 0; 1, 0,
L
R



TL = 0, T 3 = T 3 = 0 = |0, 0; 0, 0,
L
R


TL = 1, T 3 = T 3 = 0 = |1, 0; 0, 0.
L
R
(B.9)

Because of the mixing phenomenon in the neutral sector (see text), we also need the subset
of (B.8) with Q = TL3 + TR3 = 0 but Y = TR3 = 0:


TL = 1, T 3 = 1, T 3 = 1 = |1, 1; 1, 1,
L
R


TL = 1, T 3 1, = T 3 = 1 = |1, 1; 1, 1.
L
R

(B.10)

In the MH  MW case we need classify the states according to their SU(2)L SU(2)R
properties and their vectorial isospin T V = T L + T R . Writing them as |TL , TR ; TV  we
have then the possibilities:
|0, 0; 0,

|0, 1; 1,

|1, 0; 1,

|1, 1; 0,

|1, 1; 1,

|1, 1; 2.

(B.11)

The first three of them are given directly by Eqs. (B.4)(B.6), the states with TL3 = TR3 = 0
corresponding to physical cross sections. The remaining three can be found by using
the usual rules of isospin composition. It is relatively easy to find the states with Q =
TL3 + TR3 = 0:
1
|1, 1; 0 (|1, 0; 1, 0 + |1, 1; 1, 1 + |1, 1; 1, 1),
3
i
|1, 1; 1 (|1, 1; 1, 1 |1, 1; 1, 1),
2
1
|1, 1; 2 (2|1, 0; 1, 0 + |1, 1; 1, 1 + |1, 1; 1, 1).
6

(B.12)

406

M. Ciafaloni et al. / Nuclear Physics B 613 (2001) 382406

References
[1] P. Ciafaloni, D. Comelli, Phys. Lett. B 446 (1999) 278.
[2] M. Beccaria, P. Ciafaloni, D. Comelli, F.M. Renard, C. Verzegnassi, Phys. Rev. D 61 (2000)
011301;
M. Beccaria, P. Ciafaloni, D. Comelli, F.M. Renard, C. Verzegnassi, Phys. Rev. D 61 (2000)
073005.
[3] J.H. Kuhn, A.A. Penin, hep-ph/9906545.
[4] E. Accomando et al., Phys. Rep. 299 (1998) 1;
Y.Sugimoto, KEK Proceedings-97-2.
[5] P. Ciafaloni, D. Comelli, Phys. Lett. B 476 (2000) 49.
[6] V.S. Fadin, L.N. Lipatov, A.D. Martin, M. Melles, Phys. Rev. D 61 (2000) 094002.
[7] M. Ciafaloni, P. Ciafaloni, D. Comelli, Phys. Rev. Lett. 84 (2000) 410;
M. Ciafaloni, P. Ciafaloni, D. Comelli, Nucl. Phys. B 589 (2000) 359.
[8] M. Ciafaloni, P. Ciafaloni, D. Comelli, Phys. Lett. B 501 (2001) 216.
[9] J.H. Kuhn, A.A. Penin, V.A. Smirnov, Nucl. Phys. Proc. Suppl. 89 (2000) 94;
M. Melles, hep-ph/0004056;
M. Melles, hep-ph/0012157;
W. Beenakker, A. Werthenbach, Phys. Lett. B 489 (2000) 148;
M. Hori, H. Kawamura, J. Kodaira, Phys. Lett. B 491 (2000) 275.
[10] R. Doria, J. Frenkel, J.C. Taylor, Nucl. Phys. B 168 (1980) 93;
G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. Lett. 47 (1981) 1799.
[11] M. Ciafaloni, Phys. Lett. B 150 (1985) 379;
S. Catani, M. Ciafaloni, G. Marchesini, Nucl. Phys. B 264 (1986) 588;
M. Ciafaloni, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics, 1989, p. 491.
[12] M. Ciafaloni, P. Ciafaloni, D. Comelli, hep-ph/0103315.
[13] M. Kopal, hep-ex/0102027;
ALEPH Collaboration, Phys. Lett. B 495 (2000) 1.
[14] B.I. Ermolaev, S. Fadin, JETP Lett. 23 (1981) 269;
S. Fadin, Yad. Fiz. 36 (8) (1982);
A.H. Mueller, Phys. Lett. B 104 (1981) 161;
A.H. Mueller, Nucl. Phys. B 213 (1983) 85;
A.H. Mueller, Nucl. Phys. B 228 (1983) 251;
Yu.L. Dokshitzer, S.I. Troyan, in: Proc. XIXth Winter School, LNPI, 1984.
[15] V.N. Gribov, Sov. J. Nucl. Phys. 5 (1967) 399.
[16] J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. D 10 (1974) 1145;
J.M. Cornwall, D.N. Levin, G. Tiktopoulos, Phys. Rev. D 11 (1975) 972;
B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. D 16 (1977) 1519;
M.S. Chanowitz, M.K. Gaillard, Nucl. Phys. B 261 (1985) 379.
[17] Cf., e.g., A. Dobado, M.J. Herrero, Phys. Lett. B 228 (1989) 495.
[18] J. Layssac, F.M. Renard, G. Gounaris, Z. Phys. C 62 (1994) 139.

Nuclear Physics B 613 [FS] (2001) 409444


www.elsevier.com/locate/npe

Exact spectra of conformal supersymmetric


nonlinear sigma models in two dimensions
N. Read a , H. Saleur b,c
a Department of Physics, Yale University, P.O. Box 208120, New Haven, CT 06520-8120, USA
b Department of Physics, University of Southern California, Los Angeles, CA 90089-0484, USA
c CIT-USC Center for Theoretical Physics, USC, Los Angeles, CA 90089-2535, USA

Received 18 June 2001; accepted 14 August 2001

Abstract
We study two-dimensional nonlinear sigma models in which the target spaces are the coset
supermanifolds U(n + m|n)/[U(1) U(n + m 1|n)]
= CP n+m1|n (projective superspaces) and
2n+m1|2n
(superspheres), n, m integers, 2  m  2;
OSp(2n + m|2n)/OSp(2n + m 1|2n)
S
=
these quantum field theories live in Hilbert spaces with indefinite inner products. These theories
possess non-trivial conformally-invariant renormalization-group fixed points, or in some cases, lines
of fixed points. Some of the conformal fixed-point theories can also be obtained within Landau
Ginzburg theories. We obtain the complete spectra (with multiplicities) of exact conformal weights
of states (or corresponding local operators) in the isolated fixed-point conformal field theories, and
at one special point on each of the lines of fixed points. Although the conformal weights are rational,
the conformal field theories are not, and (with one exception) do not contain the affine versions
of their superalgebras in their chiral algebras. The method involves lattice models that represent
the strong-coupling region, which can be mapped to loop models, and then to a Coulomb gas with
modified boundary conditions. The results apply to percolation, dilute and dense polymers, and other
statistical mechanics models, and also to the spin quantum Hall transition in non-interacting fermions
with quenched disorder. 2001 Published by Elsevier Science B.V.
PACS: 05.50

1. Introduction
In this paper we consider two-dimensional (2D) nonlinear sigma models, and also
other related quantum field theories, that have Z2 -graded Lie algebras (superalgebras) as
internal symmetries. Such models have arisen in various contexts: the integer quantum
Hall (IQH) effect [18], and recent generalizations [922], as well as other problems
of fermions with quenched disorder [2328]; percolation, polymers, and other statistical
E-mail address: saleur@physics1.usc.edu (H. Saleur).
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 9 5 - 9

410

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

mechanics problems (Refs. [13,29] and this paper); and strings in anti-de-Sitter space
(reviewed in Ref. [30], and see Refs. [31,32]). In the models on which we concentrate
here, the target spaces are particular Riemannian supersymmetric spaces [9], in cases
in which the underlying manifolds are compact. These spaces are U(n + m|n)/[U(1)
U(n + m 1|n)]
= CP n+m1|n (projective superspaces) and OSp(2n + m|2n)/OSp(2n +
2n+m1|2n
m 1|2n)
(superspheres). In most cases (in a suitable range for m, and for
=S
n sufficiently large), the beta function for the coupling in the nonlinear sigma model is
nonzero, and there is a single non-trivial renormalization-group (RG) fixed-point theory
for each model. We make use of lattice models that we argue have transitions in the
same respective universality classes to obtain the exact complete spectra of energy and
momentum eigenvalues (i.e., conformal weights) of states and the corresponding local
operators, with their multiplicities, in the conformal field theories (CFTs) that describe
the RG fixed points. In two cases, the beta function in the sigma model is identically zero
nonperturbatively, and we find the exact spectrum at one special point on each of these
two lines of fixed-point theories. In some cases, there are also supersymmetric Landau
Ginzburg theories [29] that have a transition in the same universality class.
In more detail, the field theories and their critical points that we discuss are as follows.
For the case of CP n+m1|n , the fields can be represented by complex components za
(a = 1, . . . , n + m), ( = 1, . . . , n), where za is commuting, is anticommuting.
In these coordinates, at each point in spacetime, the solutions to the constraint za za +
= 1 (we use the conjugation that obeys ( ) = for any , ), modulo U(1)
phase transformations za eiB za , eiB , parametrize CP n+m1|n . The Lagrangian
density in two-dimensional Euclidean spacetime is

1 
( ia )za ( + ia )za + ( ia ) ( + ia )
2
2g
i
(x ay y ax ),
+
2
where a ( = 1, 2) stands for the combination
L=

(1.1)

 

 
i
(1.2)
za za + za za ,
2
and we work on a rectangle with periodic boundary conditions on za , . The fields are
subject to the constraint, and under the U(1) gauge invariance, a transforms as a gauge
potential; a gauge must be fixed in any calculation. This set-up is similar to the nonsupersymmetric CP m1 model in Refs. [33,34]. The coupling constants are g2 , the usual
sigma model coupling (there is only one such coupling, because the target supermanifold is
a supersymmetric space, and hence the metric on the target space is unique up to a constant
factor), and , the coefficient of the topological term, so is defined modulo 2 .
First we note a well-known important point about the supersymmetric models: the
physics is the same for all n, in the following sense. For example, in the present model,
correlation functions of operators that are local functions (possibly including derivatives)
of components a  n1 + m,  n1 , exist for n = n1 , say, and are equal to those of the
same operators for any other value n > n1 , due to cancellation of the unused even and
a =

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

411

odd index values. This can be seen in perturbation theory because the unused index
values appear only in summations in closed loops, and their contributions cancel, but is also
true nonperturbatively (it can be shown in the lattice constructions we discuss below). In
particular, the renormalization group (RG) flow of the coupling g2 is the same as for n = 0,
a non-supersymmetric sigma model. For the case of CP n+m1|n , the perturbative beta
function is the same as for CP m1 , namely (we will not be precise about the normalization
of g2 ),
 
 
dg2
= g2 = mg4 + O g6 ,
dl

(1.3)

where l = ln L, the logarithm of the length scale at which the coupling is defined [see,
e.g., Ref. [35], Eq. (3.4)]. (The beta function for is zero in perturbation theory, and that
for g2 is independent of .) For m > 0, if the coupling is weak at short length scales,
then it flows to larger values at larger length scales. For = (mod 2 ), the coupling
becomes large, the U(n + m|n) symmetry is restored, and the theory is massive. However,
a transition is expected at = ( mod 2). For m > 2, this transition is believed to be first
order, while it is second order for m  2 [36]. In the latter case, the system with =
flows to a conformally-invariant fixed-point theory. At the fixed point, a change in is
a relevant perturbation that makes the theory massive. Our exact results describe these
critical theories for m  2.
For m = 2, the RG flow is the same as for the familiar O(3) sigma model. For n = 0,
the = critical theory is the SU(2) level 1 WessZuminoWitten (WZW) model [37].
For n > 0, a supersymmetric extension of this familiar system is obtained. For m = 1, we
argue that the critical theory is related to the critical CFT of percolation.
For m = 0, the beta function is exactly zero (see Section 2.1), and the sigma model
exhibits a line of fixed points, with continuously-varying scaling dimensions for general
n > 1. We argue that these dimensions do not depend on . We find an exact description of
the spectrum in one theory with the same symmetry, and argue that it represents one special
value of the coupling g2 . This theory describes dense polymers (there are also variant
dense polymer theories with OSp(2n|2n) supersymmetry). These points will be discussed
further in Section 2.1; an important role is played by the fact that for n = 1, this theory is
just non-interacting massless scalar fermions (with central charge c = 2). This case has
similarities with the sigma model with a supergroup manifold SL(n|n)/GL(1) as target
space in Ref. [31,32] (for a careful discussion of the target manifold, see Ref. [27]).
The other nonlinear sigma model has target space S 2n+m1|2n ; it is a supersymmetric
extension of the usual O(m)-vector model. As coordinates, we use a real scalar field,
(1 , . . . , 2n+m , 1 , . . . , 2n ),
(the 2n + m components a are commuting, the remaining 2n components are
anticommuting), and


 =
(1.4)
a a +
J 
a

412

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

is the osp(2n + m|2n)-invariant bilinear form. J is a nondegenerate symplectic form,




for which we can take the standard form, consisting of diagonal blocks 10 10 . In the
nonlinear sigma model, the fields are restricted to obey the constraint
= 1,

(1.5)

which defines the unit supersphere, S 2n+m1|2n . The Lagrangian density of the sigma
model is
1
L = 2 .
(1.6)
2g
The perturbative beta function for the coupling g2 in this model is (again, it is independent
of n to all orders)
 
 
g2 = (m 2)g4 + O g6 .
(1.7)
For m > 2, the RG flow is towards strong coupling at large length scales, and it is expected
that the symmetry is restored, and no transition occurs (for n = 0, m  2, a broken
symmetry phase is not allowed in 2D, and no phase with power-law correlations is known
in this model for n = 0, m > 2). For m < 2, zero coupling is an attractive fixed point, and
one then expects a transition to separate this regime from strong coupling. Our exact results
describe this critical point. In the case m = 1, n = 0, it is best to use a lattice version of
the model, as will be described in Section 2, and then this is simply the 2D Ising phase
transition. For m = 0, this critical point is used to describe dilute polymers (self-avoiding
walks).
For m = 2, the theory is more interesting, as the perturbative beta function for g2
vanishes to all orders, similar to the case of the CP n1|n model. For the action as given,
we again expect that the beta function vanishes exactly, and there is a line of fixed point
theories with continuously-varying scaling dimensions. However, the lattice version of
the model, discussed in Section 2, is the usual XY model for n = 0, and a transition
occurs due to vortex excitations [38], which are not accounted for in the action (1.6).
The supersymmetric extension should have excitations that correspond to vortices, that
can appear as perturbations in the action, and which become relevant above a critical value
of the (renormalized) coupling g2 . Then, as in the KosterlitzThouless (KT) theory of
the 2D XY model [38], there will be a low temperature (small g2 ) phase with power-law
correlations with continuously-varying exponents, and a high-temperature (large g2 ) phase
that is massive, separated from each other by a critical point. It is for this critical theory
that we obtain the exact spectrum.
The critical CFTs in the S 2n+m1|2n models can also be reached starting from Landau
Ginzburg or 4 theories, in which the field , unconstrained this time, has Lagrangian
density
1
1
1
L = + r + ( )2 .
(1.8)
2
2
8
Such a supersymmetric approach to dilute polymers (the m = 0 case) was proposed by
Parisi and Sourlas [29] (this works in any spacetime dimension). In these theories, a

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

413

transition can occur when the renormalized mass-squared r changes sign, again only if
m  2. At the critical point, flows away from zero in any dimension less than four, and
reaches a nontrivial fixed point. These critical points should be the same CFTs as in the
nonlinear sigma models of the same symmetry. The low temperature (r < 0) regime of
these models should correspond to the weakly-coupled regime of the S 2n+m1|2n sigma
models.
We now turn to the lattice models we use, which originated as descriptions of
percolation, polymers, and some other problems in statistical mechanics. The standard
approach to percolation, polymers (self-avoiding walks) and related 2D problems leads
to a CFT description of the critical phenomena involving a single scalar field with a
background charge, and has been well-studied. It is less widely appreciated that there are
representations of the original lattice models using supersymmetry [29]. The cancellation
between fermion and boson states in loops (percolation cluster boundaries, or polymer
loops) in graphical expansions can be used to weight these loops by 1 (for percolation) or
0 (for polymers), in place of the usual Q 1 limit in percolation (based on the Q-state
Potts model) or m 0 limit in polymers (based on the O(m)-vector model). In the cases of
percolation [13] and of dense polymers, we will show that the supersymmetric model is a
vertex model (or in an anisotropic limit, a quantum spin chain) and there is a corresponding
nonlinear sigma model, the CP n+m1|n model at m = 1 (percolation), 0 (dense polymers).
For dilute polymers, the corresponding nonlinear sigma model is the S 2n+m1|2n model at
m = 0. The consequence of the existence of supersymmetric models for these problems
is that there should be versions of the CFT descriptions that exhibit the same global
supersymmetries.
Our results for the percolation case apply directly to the critical theory of the spin
quantum Hall (SQH) transition in non-interacting fermions with quenched disorder, for
which some exact exponents have been found previously by a mapping of a lattice model
onto percolation [13]. While much has been learned about percolation and polymers
without the use of the supersymmetric models, in the random fermion problems such as
the SQH and IQH transitions the supersymmetry is virtually all we have to guide us in
seeking the critical theories. Therefore, a fuller understanding of CFTs with the appropriate
type of supersymmetry should aid our understanding of the latter problems. Our results
constitute another step towards the understanding of previously unknown CFTs that have
superalgebras as symmetries, and central charge 0, 2, or other values. An important part
of our results is that these theories are not rational CFTs, even though all the conformal
weights are rational numbers, because the spectra do not consist of a finite number of
representations of a chiral algebra of integer spin fields. In particular, the chiral algebras
do not contain affine Lie superalgebras, nor do the theories appear to be related to such
superalgebras in any simple way. While the presence of symmetries acting locally on the
degrees of freedom does imply the existence of local currents, these do not generate an
affine Lie superalgebra because they are not purely right- or left-moving (i.e., not chiral
a fact that will manifest itself in logarithms in their correlators), and so cannot be part
of any chiral algebra. We anticipate that such features may be common in the presently
unknown CFTs for the random fermion problems.

414

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

We note that there is other previous work on applications of sigma models with
supergroup manifolds as the target space, in which the CFTs are also not current algebras,
though closely related to them. These theories are related to other random fermion
problems, the so-called chiral ensembles [2328]. There is a proposal to apply one of these
theories, using the work in Refs. [31,32], to the IQH transition [27]. Certain other random
fermion problems can be tackled fruitfully by current-algebra methods [3946].
We now describe our method, in brief overview. We first discuss vertex models with
supersymmetry, and their relation to nonlinear sigma models and LandauGinzburg
theories. Special cases of these models describe some properties of percolation and
polymers. The models can be defined with periodic boundary conditions on, say,
a rectangle with sides l1 , l2 . They can be viewed as 1D quantum systems in the usual
way, say by constructing a transfer matrix T = eH , and then Z = STr eH l2 is the
partition function, which (by construction) is Z = 1 for percolation and dilute polymers,
and Z = 0 for dense polymers. The supertrace, denoted STr, is defined as STr( ) =
Tr[(1)F ], where F = 0, 1 is the grading of states; in many models, F is the fermion
number. The partition function Z is 1 or 0 in these special cases because of cancellations
between states associated with the same energy eigenvalue (i.e., eigenvalue of H ). To
obtain the energy eigenvalues, we will calculate Zmod = Tr eH l2 . This means modifying
the boundary condition in the 2 (imaginary time) direction so as to insert an additional
factor (1)F . For the original periodic boundary conditions, all fields are periodic around
either fundamental cycle of the torus. In the modified partition function, the boundary
condition is modified (twisted) such that odd (fermionic) fields are antiperiodic along
the 2 (time) direction, periodic along the 1 (space) direction, while even (bosonic) fields
remain periodic. By varying the aspect ratio of the system, the energies and momenta,
and their multiplicities, can be read off from the modified partition function. We note that
to incorporate supersymmetry as an internal symmetry, it is convenient to use a space
of states that includes some that have negative norm [6,13] (more pedantically, it is the
norm-squared that is negative). This is connected with the violation of the spinstatistics
theorem in the continuum field theories. This allows H to be Hermitian with respect to
the indefinite inner product, but nonetheless because of the negative norms, viewed as a
matrix H is not diagonalizable [8]. It can be brought to Jordan normal form, and then the
diagonal elements of eH l2 are of the form eEl2 with multiplicities given by the sum of
the sizes of the Jordan blocks that contain E. These Jordan blocks are related to the fact
that these CFTs are logarithmic [47]. We will not obtain any information about the Jordan
block structure, and will simply think of energy eigenvalues E with multiplicities.
By evaluating the transfer matrix products, the vertex models become loop models,
that is sums over configurations of non-intersecting loops, with various weights. In the
supersymmetry point of view, each loop has states in the defining representation of the
superalgebra running around the loop, and the state on a loop is chosen independently of
those on other loops. In particular, for periodic boundary conditions each loop, whether or
not it winds the cycles of the torus, is given a weight which is the supertrace (denoted str)
of 1 in the defining representation of the superalgebra, which is set up to be str 1 = m in
general, and thus 1 (for percolation), 0 (for polymers). The effect of the modified boundary

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

415

condition is that a loop that winds along the 2-direction an odd number of times loses
the cancellation of even and odd states, and is given a factor which is the trace (not the
supertrace) of 1 in the defining representation, while loops that wrap an even number of
times are given unchanged weights.
The loop models can in turn be represented using a scalar field with a background charge,
with certain boundary condition effects (see Ref. [48] and references therein). We will
show that these can be modified to include the differing factor for each loop that winds
the 2 direction an odd number of times. After performing a Poisson resummation, we
obtain the modified partition function in the continuum limit, which gives the spectrum of
each CFT. For low-lying levels, the multiplicities are dimensions of representations of the
superalgebra, and this should continue to hold for higher levels, where the multiplicities
are positive integers.
In Section 2 we describe explicitly the supersymmetric vertex models we use, show
how they can be mapped onto loop models (special cases of which describe percolation
and polymers), and discuss general aspects of the critical phenomena further. In Section 3
we give the construction of the modified partition function for a general loop model. In
Section 4 this is applied to the sigma (or LandauGinzburg) models that correspond to
percolation, dilute and dense polymers, and also some other models. Appendix A gives
details of the derivation of the modified partition function. Appendix B contains results
on finite size systems with free boundary condition in the space direction. These give the
spectrum of the boundary CFT for percolation, whereas the rest of the paper is about the
bulk CFTs. For this case, we show that the states enumerated by the modified partition
function are complete. Some partial results on completeness of states in finite size with
periodic boundary conditions in the space direction are also given.

2. Supersymmetric models
In this section we describe the supersymmetric lattice models for percolation and
polymers. We also explain the relation to the field theories introduced in Section 1, and
give some further field-theoretic arguments.
2.1. Supersymmetric vertex models and CP n+m1|n sigma models
A supersymmetric representation of the standard model of bond percolation in 2D was
constructed in Ref. [13]. Some salient points are reviewed here (some further details can
be found in Appendix B). We also describe the mapping to the corresponding CP n+m1|n
sigma models, and their use in understanding the physics of the lattice models.
The standard model of 2D percolation is bond percolation on the square lattice. Clusters
can be represented alternatively by their boundaries, which are configurations of spacefilling, non-intersecting (and non-selfintersecting) loops on edges of the square lattice (the
loops are allowed to touch, without intersecting, at the vertices). The latter square lattice
is the medial graph of that on which the bonds are placed in percolation. This mapping
has been pictured, and the weights given, many times (see, e.g., [13,4850]), so we will

416

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

omit the details, except for the important point that there is a global factor of 1 for each
loop, in addition to local weights that are independent of the global properties. The model
can be formulated on a simply-connected portion of the plane, or on a cylinder, or on
a torus, that is with periodic boundary conditions in two directions. The global factor per
loop applies to loops that wind around the nontrivial cycles of the cylinder or torus, as
well as to those that are homotopic to a point. Because percolation (or Potts) clusters admit
a two-coloring with opposite colors on the two sides of any segment of a loop, the loops
that wind around the system must do so in pairs, and because they do not intersect, all
the loops that wind the system must be homotopic to one another. The partition function
is the sum over all distinct configurations of such loops, with equal weight, such that
Z = 1.
A popular representation for the percolating clusters uses a Q-state Potts model on
the underlying graph. This yields a loop model with the same local weights but a global

factor Q per loop [49], so percolation corresponds to the limit Q 1 (when the model
is formulated on the torus, there is a slight difference between the loop model and the
Potts model [48]). Affleck [51] showed that the loop model can be reproduced using a
representation of the transfer matrix in a space with m states per site, with Q = m2 . The
transfer matrix has SU(m) symmetry.
We now describe the supersymmetric vertex model, which is closely related to that in
Ref. [51], and a special case of which represents percolation [13]. The transfer matrix
acts in a space that is a Z2 -graded tensor product (denoted ) of Z2 -graded vector spaces
at each site; these spaces have graded dimension n + m|n, with n + m, n  0 (we will
also refer to the total dimension of spaces, which here would be 2n + m). This space
is represented using boson and fermion oscillators, with constraints. We consider sites
labeled i = 0, . . . , 2l1 1, with periodic boundary condition. For i even we have boson

operators bia , bia


, [bia , bjb ] = ij ba (a, b = 1, . . . , n + m), and fermion operators fi , fi
,

{f , f } = ij (, = 1, . . . , n). For i odd, we have similarly boson operators bia , b ,


i

[bia , bjb ] = ij ab (a, b = 1, . . . , n + m), and fermion operators fi , fi , {fi , fj } =

ij (, = 1, . . . , n). Notice the minus sign in the last anticommutator; since our
convention is that the stands for the adjoint, this minus sign implies that the norms of
any two states that are mapped onto each other by the action of a single fi or fi have
opposite signs, and the Hilbert space has an indefinite inner product.
b


b
The supersymmetry generators are the bilinear forms bia
bi , fi
fi , bia
fi , fi
bi for i

b
b

even, and correspondingly bi bia , fi fi , fi bia , bi fi for i odd, which for each i
have the same (anti-)commutators as those for i even. Under the transformations generated

, fi
(i even) transform as the fundamental (defining) representation
by these operators, bia
a

V of gl(n + m|n), bi , fi (i odd) as the dual fundamental V (which differs from the
conjugate of the fundamental, due to the negative norms). We always work in the subspace

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

417

of states that obey the constraints


a

bi + fi
fi =1 (i even),
bia

(2.1)

biabia

(2.2)

fi fi =1

(i odd)

(we use the summation convention for repeated indices of types a or ). These specify
that there is just one particle, either a boson or a fermion, at each site, and so we have the
graded tensor product of alternating irreducible representations V , V as desired. In the
spaces V on the odd sites, the odd states (those with fermion number fi fi equal to
one) have negative norm.
A note on the superalgebras is in order (these remarks are well known, and can
also be found in, e.g., Ref. [32]). The above bilinears generate the superalgebra gl(n +
m|n) in the defining representation V ; they span the Hermitian matrices on V . One
a

bi + fi
fi (or the identity matrix) in the
generator, denoted E, and given by E = bia
defining representation, commutes with the others. The generators which correspond to
the Hermitian matrices with vanishing supertrace (in V ), form a sub-superalgebra, denoted
sl(n + m|n). For m = 0, E has nonzero supertrace, and is eliminated by restricting to
sl(n + m|n). In these cases, sl(n + m|n) is a simple superalgebra. But for m = 0, E has
supertrace 0, and generates an ideal in sl(n|n). The quotient superalgebra, denoted psl(n|n),
can be obtained by taking the quotient of sl(n|n) by the ideal. For n > 1, psl(n|n) is
simple. The spaces V , V are, strictly speaking, not representations of psl(n|n) because E is
nonzero in these spaces (E = +1 in V , 1 in V ), but the tensor products of equal numbers
of factors V and V , as we use in our models, are. We also note that we use lowercase
letters for the superalgebras (which are always over C), while uppercase letters such as
U(n + m|n) denote a corresponding supergroup, with the real form of the underlying Lie
group specifiedin this example, it is U(n + m) U(n). The real form is determined by
the unitary transformations with respect to the inner products in the spaces in our models.
The invariant transfer matrix is constructed as follows. First we note that for any two
sites i (even), j (odd), the combinations
bia bj a + fi fj ,

bjabia
+ fj fi

(2.3)

are invariant under gl(n + m|n), thanks to our use of the dual V of V . Then the transfer
matrix acting on sites i, i + 1 with i even is



Ti = (1 pA ) + pA b a b + f f bia bi+1,a + fi fi+1, ,
(2.4)
i+1 ia

i+1 i

while for i odd



 a



Ti = pB + (1 pB ) b iabi+1,a
+ fi fi+1,
bi+1 bia + fi+1
fi .

(2.5)

The complete transfer matrix is


T T1 T3 T2l1 1 T0 T2 T2l1 2 ,

(2.6)

and i = 2l1 is interpreted as i = 0. Then Z = STr T l2 . We usually assume 0  pS  1,


where S = A, B. By taking either of the two terms in Ti for each vertex in the graph, the
expansion in space-filling loops mentioned above is obtained [13], with factors pS , 1 pS

418

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

for each vertex, and a factor (n + m) m = str 1 = m for each loop (in all the models
considered in this paper, we will consider only m, n integer, with n + m, n  0). The latter
is equal to the supertrace in the fundamental representation (denoted str), of 1, because the
evaluation of contributions for each loop can be viewed in terms of states in V flowing
around the loop. This holds true for topologically nontrivial loops as well as for loops
homotopic to a point. The transfer matrix T on a chain with periodic boundary conditions,
as considered here, is translationally invariant (under translation by two sites), as well as
supersymmetric. We note that these constructions give representations of the Temperley
Lieb algebra (at certain parameter values determined only by m) in the various graded
vector spaces, as discussed further Appendix B. The special case m = 1 gives percolation
(Z = 1 for m = 1). This version has the advantage of being mathematically well-defined,
because there is a mathematically-meaningful, nontrivial transfer matrix at m = 1, unlike
the case of the SU(m) model [51] which requires that a poorly-defined limit m 1 be
taken in order to obtain any quantity of interest. For m = 1, the parameters pA and pB are
the probabilities of links of the square lattice being occupied by bonds in the percolation
problem, with A as the horizontal, and B as the vertical links. For all m, the system is
isotropic when pA = pB , while the transition occurs when pA = 1 pB .
In the anisotropic (time- or 2-continuum) limit pA 0 with pA /(1 pB ) fixed, we can
1/2
1/2
write T  epA (1pB ) H , where the Hamiltonian H is
  a

 a
bi+1 bia + fi+1
fi bi bi+1,a + fi fi+1,
H = 5
i even

5 1



 a



biabi+1,a
bi+1 b ia + fi+1
+ fi fi+1,
fi

i odd

+ (5 + 5 1 )l1 ,

(2.7)

and 5 = pA /(1 pB ). This Hamiltonian, like the transfer matrix above, is invariant
under sl(n + m|n) (or psl(n|n) for m = 0) for all parameter values pA , pB . This
Hamiltonian can be viewed as that of an antiferromagnetic superspin chain, built from the
generators listed above times those on the neighboring site, contracted using the invariant
bilinear form on the superalgebra, up to constant terms. When 5 = 1, the two types of
terms in H have different coefficients, and this perturbation (staggering the coupling) is
a relevant perturbation that takes the system off the transition point (see, e.g., Ref. [13]).
For m = 1, the ground state energy of H is zero, which follows from an argument similar
to that in Ref. [8], or from the facts that the partition function is 1, and that there are no
energies less than zero (which is less obvious). The preceding family of supersymmetric
vertex models, or the corresponding superspin chain Hamiltonians, are the sl(n + m|n)invariant models for which we find exact results in the following.
For such vertex models or spin chains, there is a corresponding continuum quantum
field theory, which is a nonlinear sigma model with target space the symmetry supergroup
[here U(n + m|n)], modulo the isotropy supergroup of the highest weight state (see,
e.g., Refs. [36,52] for related, but non-supersymmetric, examples, and Refs. [7,19] for
supersymmetric examples in random fermion problems). For the present cases we obtain

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

419

U(n + m|n)/U(1) U(n + m 1|n)


= CP n+m1|n , a supersymmetric version of complex
projective space. This mapping can be controlled by generalizing the spin chain model by
replacing the fundamental by a higher spin (larger) irreducible representation (irrep),
represented by the size of the highest weight S, 2S integer (with S = 1/2 for the
fundamental), and its dual on alternate sites. For an appropriate family of such irreps,
such that the isotropy supergroup is the same, the sigma model can be obtained with a bare
coupling constant (at the scale of the lattice spacing) g2 1/S, and the mapping is exact at
large S. In the construction above, this can be done by replacing 1 on the right-hand side of
Eqs. (2.1), (2.2) by the integer 2S, and using the same Hamiltonian (2.7), up to an additive
constant (the transfer matrix of the vertex model would need to be modified). The action of
the nonlinear sigma model obtained this way is as in Eq. (1.1), where the bare value of
is = 2S + O(5 1); note that can be varied by staggering the couplings [36]. In fact,
as 5 varies from 0 to , varies from 0 to 4S, and so passes through (mod 2) 2S
times; for 2S odd, one such point occurs at 5 = 1 for symmetry reasons.
A great deal of insight into the vertex models or spin chains of interest, with S = 1/2,
can now be gained by analyzing the sigma models, and vice versa. When the beta function
Eq. (1.3) for g2 is positive at weak coupling, it is expected that the same fixed point CFT
is reached in the universal large length-scale behavior of the spin chain models for all odd
values of 2S, when 5 = 1 ( = ) (more generally, also at all the 2S critical values of 5,
for all S). In particular, for m = 1, this CFT, to which the CP n1|n sigma model at =
flows, is related to the critical CFT of percolation. We should point out that problems
like percolation are defined geometrically, and that it is possible to define many scaling
dimensions, or correlation functions, for geometric or topological properties of clusters.
On the other hand, we are dealing here with concrete models, and for these the spectra
of states and corresponding operators in CFT are precisely defined, and may not include
all possible operators that yield geometric properties of clusters. The model we are using
emphasizes the properties of the loops, which are the boundaries (hulls) of percolation
clusters [13].
As argued in Section 1, for m = 2, the RG flow in the sigma model is the same as for
the familiar O(3) sigma model. For n = 0, the spin chains are the familiar Heisenberg
spin chains, and for 2S odd, the critical theory is the SU(2) level 1 WessZuminoWitten
(WZW) model [37]. For n > 0, a supersymmetric extension of this familiar system is
obtained. We will return to results for this model in Section 4.4 below.
Also, the case m = 0 (and S = 1/2) can serve as a model for dense polymers. The
general set-up for polymers is discussed more fully in Section 2.2. Here we just specify
that dense polymers are obtained when one or more long polymer loops are forced into
contact. Models of dense polymers involve space-filling non-intersecting loops, with a
global factor m = 0 for each loop. One such model, for loops on the square lattice, is
obtained from the psl(n|n) invariant vertex model. The structure of this model implies that
it contains information on exponents for only even numbers of polymers winding around
the torus. The partition function of this model vanishes, as long as the boundary conditions
are psl(n|n)-invariant, as they are here. The case n = 1 is a free-fermion model [53]. Define
fermion operators Fi = fi1 bi1 (i even), Fi = bi1fi1 (i odd); the space of states is just a

420

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

Fock space, without constraints, in these variables. These operators flip the superspins,
and are related directly to the odd generators of the superalgebra. The Hamiltonian in
terms of F , F is non-interacting, and can easily be solved exactly; indeed, for m = 0,
n = 1, this can be done for all S and 5. The result is free massless scalar complex, or
symplectic, fermions [29,47,53], with Lagrangian density in the continuum limit (,
are anticommuting complex fields):
L=

1
,
2g2

(2.8)

where (for each value of 5) g2 is precisely proportional to 1/2S. By construction,


the free fermions are the Goldstone modes (spin waves) of the spontaneously-broken
supersymmetry of the antiferromagnetic chain. The odd elements of the superalgebra act
on as + , a complex constant, so no other terms with zero or two derivatives
are possible in the action. The spin-wave theory of the antiferromagnet is exact in this case.
In the case of general n, the naive continuum (or large S) limit in this m = 0 case is
the sigma model with target space CP n1|n . In this model, the perturbative beta function
vanishes identically. This can be seen either from direct calculations, which have been
done to at least four-loop order [35], or from an argument similar to that in Ref. [31]: for
n = 1, the sigma model reduces to the massless free fermion theory (2.8) (the sigma model
target space for n = 1 has dimension 0|1 over the complex numbers), and further the theta
term becomes trivial in this case. Thus, for all sigma model couplings g2 > 0, the theory
for n = 1 is non-interacting. The free-fermion theory is conformal, with Virasoro central
charge c = 2; is clearly a redundant perturbation in this case, as it does not appear in
the action (a similar argument appeared in Ref. [18]). By the same argument as before, the
conformal invariance with c = 2 should hold for all n, and also for all g2 and , though
the action is no longer non-interacting in general. Thus the beta function is also identically
zero non-perturbatively. In general, the scaling dimensions will vary with the coupling g2 ,
so changing g2 is an exactly marginal perturbation, though for n = 1 the coupling can
be scaled away, so there is no dependence on the coupling in the exponents related to
those multiplets of operators that survive at n = 1. Hence for n = 1, the exactly-marginal
perturbation that changes g2 is redundant. The dense-polymer problem (the S = 1/2 case)
apparently corresponds to only one special value of g2 . We also expect that, at the special
value of g2 , remains redundant for all n, because the vertex model remains critical when
staggering is present, as we know from the n = 1 case, and in the modified Coulomb gas
derived below, this constrains all the parameters, so staggering does not change any scaling
dimensions even for n > 1 (whereas we cannot change g2 within the vertex model at fixed
S = 1/2, and this leaves open the possibility that scaling dimensions, other than those of
operators that occur for n = 1, change with this coupling). It is likely that is redundant
for all g2 and all n.
Finally, the cases m < 0 of the sl(n + m|n) (or percolation-type) models may also be of
physical interest. For sufficiently weak coupling, the flow of g2 is towards zero at large
length scales, for all . There could be a transition at some finite g2 . When n + m = 1,
the action consists only of interacting scalar fermions, and there can be no theta term.

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

421

Then using similar arguments as in the dense polymer case, the critical theory will be
independent of for all n  1 m. The exact results presented below may describe these
transition points, at least for m = 2, 1.
2.2. Polymers and S2n+m1|2n sigma models
Polymers are modeled as self-avoiding and non-intersecting random walks. There
is a standard method in which this is represented using a scalar field theory with m
component scalar fields, and the limit m 0 in which closed loops disappear yields the
statistics of polymers [54]. Such a model is O(m)-invariant. Parisi and Sourlas introduced
a supersymmetric version, for which we gave the Lagrangian already in Eq. (1.8). The
critical point of that theory for m = 0 represents dilute polymers.
The lattice model usually used in obtaining exact results for 2D polymers is a stronglycoupled version of the field theory, obtained by truncating the high-temperature expansion
of a lattice model [55]. A supersymmetric version of this model can be constructed with
OSp(2n + m|2n) supersymmetry, as follows: We use the honeycomb graph, which can be
constructed by tiling the torus with regular hexagons. The partition function is
 


di
e i i i /2 (1 + Ki j ).
Z=
(2.9)
n+m/2
(2)
i

ij 

Here the 2n + m|2n component field i at each site (vertex) i of the honeycomb graph has
components as in Section 1, and transforms in the fundamental (or vector) representation
V  of OSp(2n + m|2n) [where the real form of the underlying Lie group is O(2n + m)
Sp(2n, R)]. di is the standard OSp-invariant measure on i . Expanding the product over
distinct links ij  of the graph, and performing the integrations over the fields, the partition
function is given by

K L m#loops,
Z=
(2.10)
loops

where the sum is over all configurations of any number of non-space-filling nonintersecting closed loops on the honeycomb graph, L is the total number of edges of
the graph occupied (at most once) by part of a loop, and #loops is the number of loops.
Notice the global factor str 1 = 2n + m 2n = m per closed loop, whether or not the loop

is topologically nontrivial [the supertrace str here is in the
vector
representation V ]. In
general, the critical point of the model occurs at Kc = 1/ 2 + 2 m for 2  m  2
[56]. For m = 0, Z = 1, and at the critical value Kc of K already known from the n = 0
case, this yields the theory of dilute polymers.
The same loop model can be obtained from a similar model in which the field is restricted
to obey i i = 1 for all i, making this a lattice version of the S 2n+m1|2n sigma model.
The partition function of such a model would be
 

Z=
(2.11)
d[i ] eJ ij i j /T ,
i

422

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

where d[i ] is the normalized invariant measure on the supersphere, J is a constant,


and T is temperature. This partition function can be shown to be independent of n,
with no approximation. The high-temperature (strong-coupling) expansion of this model
is obtained by expanding the exponential in powers of J /T . It can be shown to be
independent of n to all orders. In general the expansion involves diagrams with any number
of lines (with associated factor J /T per line) on each link of the graph, with a condition that
an even number of lines end at each vertex. Each line represents the vector representation
propagating along the links. We truncate the expansion of the exponential to the first two
terms [55], and then the graphical expansion is the same as in the model above, with
K J /T . Either of these constructions can be used as the supersymmetric version of
the O(m) model. (For m = 1, n = 0, this model accurately represents the (untruncated)
Ising model on the honeycomb graph, with K = tanh J /T .) Finally, either model can
be represented by a transfer matrix on the tensor product of complex (2n + m + 1|2n)dimensional spaces C V  (where V  has been complexified), one for each vertical link in
a horizontal row of the graph; the states represent either the empty (singlet) or the occupied
(vector) state. The partition function is then the supertrace of this transfer matrix to the
power l2 . One then expects the critical theory of any of these models to be the same
as in those discussed in the introduction. We remark that the Lie superalgebra-invariant
approach to polymers is different from the N = 2 superconformal approach proposed
in [57], although relations between the two are not excluded.
Apart from the m = 0 polymer case, the OSp(2n + m|2n)-invariant models (2.9) with
m = 0 may also be of interest. The cases m > 2 show no critical phenomena, but the cases
m = 1, 2 exhibit the standard Ising and KT transitions for n = 0. Again, for n > 0 we obtain
supersymmetric extensions which have transitions at which the CFTs include the exponents
and correlators of the usual theories. In spite of having the same supersymmetry, these
models do not have the same representation structure as the random-bond Ising model [19].
In Section 4.4 below, we also discuss some results for the case m = 2.
In the O(m) models (2.9), there is also a large K, low T , regime. For all K in the range
Kc < K < , the system flows to the same fixed point CFT (which varies with m). For
m = 2, this coincides with the critical branch, discussed above, but is a distinct CFT for
m < 2 [56]. For m = 0, this yields a model of dense polymers, which in this case has no
restriction on the numbers of loops that can wind the two directions on the torus [58]. We
will consider this briefly below. Naively, or based on the LandauGinzburg theory (1.8), the
low T region of the O(m) or OSp(2n + m|2n)-invariant models would be expected to be a
broken symmetry regime of the model (1.8), described by the weakly coupled S 2n+m1|2n
sigma model, see Eq. (1.6). The perturbative beta function (1.7) for the coupling g2 in this
model is such that for m < 2 zero coupling is an attractive fixed point at large distances
(which tends to confirms the existence of a transition in this range of m). However, this noninteracting behavior is at odds with the conformally-invariant behavior of the O(m) models,
which map to loop models, that we study here, so apparently the latter do not describe this
regime of the S 2n+m1|2n sigma models. Instead, the low-T O(m)/loop models are more
closely related to the sl(n + m|n)-invariant models, as we will see. This distinction is most
likely due to the truncation of the high-T expansions, as discussed above, which does not

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

423

change the universality class in the high-T and critical regions, which are the same as in the
continuum theory (1.8), but could change the low-T behavior. In terms of the physics of
polymers, what we call dense polymers is the m = 0 model in which dense loops are strictly
non-intersecting, and the resulting CFT has c = 2. If we adopt instead de Gennes model
of weak repulsions, as in Lagrangian (1.8), then intersections are suppressed, but allowed,
and it seems that this dense phase is attracted to the non-interacting fixed point at g2 = 0
in the S 2n1|2n sigma model, a theory of free scalar bosons and fermions with central
charge c = 1. (This distinction of the two models relies on the topology of loops in 2D,
so may not hold in higher dimensions.) We expect that the introduction of a small amplitude
for allowing crossings in the former dense polymer model produces a crossover to the latter
fixed pointthe apparent violation of Zamolodchikovs c-theorem being possible, because
the theories in question are not unitary.
In the Introduction we have already briefly discussed the model with partition function
(2.11) for m = 2. For n = 0, it is the XY model, and the transition is understood in terms
of vortices [38], which are characterized by integers, since the fundamental homotopy
group 1 (S 1 ) = Z. On the other hand, for n > 0, 1 (S 2n+1|2n ) = 1 (S 2n+1 ) = 0, and there
are no topologically stable vortices in the continuum S 2n+1|2n sigma model. But we still
believe that the partition function of the lattice model on the torus with periodic boundary
conditions is the same for all n, and similarly for the continuum model (1.6) in which the
fields are assumed continuous, so there are no vortices. At weak coupling, we can attempt a
semiclassical calculation within the continuum sigma model (1.6). For n = 0, one expands
around the saddle point where is constant, and a set of saddles where winds in one
or both directions on the torus. (In fact, such partition functions for winding numbers M  ,
M appear in a different context in Eq. (3.3) below.) For n > 0, the natural saddle points
that correspond to winding configurations for n = 0 are those where the field winds (as a
function of position going around a cycle on the torus) around a great circle of the sphere,
which break the symmetry down to osp(2n|2n). These saddle points are unstable, but the
multiplicity of the unstable modes is 2n|2n, and so the negative eigenvalue cancels in the
formal one-loop fluctuation determinant. Summing over such saddle points and working
to one-loop order, we can reproduce the partition function of the continuum S 1|0 model.
Thus it is very plausible that these configurations are associated with the winding modes
for n = 0, which correspond in CFT to operators that insert vortices.
2.3. Modifying boundary conditions
We outlined in the Introduction how we would modify the boundary condition in the
2 direction so as to evaluate Zmod = Tr eH l2 instead of Z = STr eH l2 . Note that this
modified boundary condition breaks the supersymmetry. More generally, we will evaluate
Zmod = Tr ei P R l1 H l2 , where P is the momentum, and R is a free real parameter. The
insertion of the translation by R l1 in the 1 direction makes the torus non-rectangular.
It is not difficult to see that the effect of the modified boundary condition is that, for
each loop that winds in the 2 direction (crosses the 1 axis), which would previously have
picked up the factor str 1, now picks up str(1)n2 f , where n2 is the number of times it

424

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

cuts the 1-axis, and f = 0 or 1 represents the grading of states in the fundamental: for the
sl(n + m|n)- (respectively, osp(2n + m|2n)-) invariant models, f = 0 for the first n + m
(resp., 2n + m) basis states in V (resp., V  ), f = 1 for the remaining n (resp., 2n). Hence,
loops that wind the 2 direction an even number of times are given the factor m as usual,
while those that wind an odd number of times are given 2n + m (resp., 4n + m).
In discussing the spectrum, we will often refer to operators as well as to states. The
models we are studying are all compact, that is there is a finite number of states per
site in the chain on which the transfer matrix acts (in the naive continuum field theories,
the target spaces are all compact, or effectively so as excursions to large field values are
suppressed). In such systems, one expects that all operators that are strictly local in terms of
these variables (i.e., operators whose correlators with local functions of these variables are
single-valued) correspond to states in the CFT, so the spectrum of one is the spectrum
of the other. There may also be meaningful operators that are not local, such as twist
fields or disorder operators in the Ising model. These are not expected to correspond to
states. Examples in the models we study include the Potts spin operator, for which there
is a nonlocal construction in the supersymmetric vertex model [13], operators studied by
Duplantier [59] and by Cardy [14], and probably the 0-hull (0-leg) operator for percolation
(polymers) [57,60,61].
The n = 1 case of percolation is related to the n = 1 case of the model for the spin QH
transition [13]. In general, the naive continuum theory for the spin QH effect is the sigma
model with target space OSp(2n|2n)/U(n|n), where the real form of the even part of osp
this time is SO*(2n)Sp(2n) (class C in Ref. [9]). This target space is noncompact for
n > 1, but compact and the same as CP 1|1 for n = 1. In the noncompact cases, states lie in
infinite-dimensional representations, and the spectrum is expected to be continuous. On the
other hand, the natural local operators (at least, those that are local functions of the sigma
model fields) transform in finite-dimensional representations, and the general argument
above implies that the operators for n = 1 are a subset of those for n > 1. Thus our results
give us information about the operators in general, whose spectrum will be discrete. Similar
statements about the spectra of states and operators apply to the other noncompact sigma
models, as typically arise in the random fermion problems including the IQH case [9].

3. Modified Coulomb gas partition functions of loop models


In Section 2 we saw that various supersymmetric models can be mapped onto loop
models, where the global weight factor associated with each loop is a supertrace in the
fundamental representation of the appropriate superalgebra. In this section we show how
the modified partition functions of general loop models can be evaluated at the critical
point, in the continuum limit, with periodic boundary conditions (modified, for the 2
direction). We first set up a mapping of loop models onto a Coulomb gas partition function
following Ref. [48], with modifications. It is convenient to begin with the polymer-type
situation (the osp(2n + m|2n)-invariant models of Section 2.2), in which any number

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

425

of loops can wind either direction. Afterwards, we indicate the modification for the
percolation-type models (those in Section 2.1), in which these numbers must both be even.
The starting point for mapping loop models on a lattice onto scalar field theories is to
represent the configurations of loops as configurations of a dual height model that assigns
a real height variable to each plaquette of the lattice. The loops are here viewed as having
two possible orientations (represented by an arrow on the loop) assigned to them, which
will be summed over. The heights associated to a given configuration of directed loops are
defined, up to addition of a constant, so that on crossing a loop from left to right (looking
along the direction of the arrow on the loop), the height changes by 0 , a constant.
Locally, the heights are well-defined, because of the absence of any sources or sinks for the
arrows on the loops. On the torus, loops that wind around the two cycles can lead to heights
that are not single valued, but are single-valued modulo 0 . Suppose the height changes by
1 = M0 in the 1 direction, and by 2 = M  0 in the 2 direction (where 1 and 2 refer
to a choice of fundamental cycles on the torus, which is not necessarily rectangular now).
Then, because the loops are non-intersecting, it can be shown [48] that this corresponds to
at least |M M  | distinct loops, each of which wind the 2 direction |M/(M M  )| times,
and the 1 direction |M  /(M M  )| times (disregarding the orientation of each loop), plus
any number of loops that are homotopic to a point. Here m n denotes the highest common
factor of two integers, with m 0 = m for all m  0.
Now each loop that is homotopic to a point is given a global factor m = 2 cos 0 e0
by the sum over the two possible orientations, and the same factor is desired for each
homotopically non-trivial loop in the unmodified partition function. For our modified
partition function, this can be simply modified to 2 cosh for each loop that winds the 2
direction an odd number of times. The arguments of Ref. [48] can now be easily modified
to show that, in the continuum limit taken at appropriate values for the local weight factors
(discussed in Section 2), the modified partition function can be written as



ZM  ,M gf 2 w(M, d; f, e0 , ),
Zmod [g, 0 , e0 , ] =
(3.1)
M  ,MZ

w(M, d; f, e0 , ) = cos(2df e0 )M/d0 + cosh(d)M/d1 ,

(3.2)

where d = M  M, f = 0 /(2), and in the Kronecker s the congruences are modulo 2.


The notation is somewhat redundant, since the parameters g, f , e0 enter only in the
combinations gf 2 and f e0 , but it is conventional and we will keep it. The continuum
scalar field partition function is given by





ZM  ,M (g) = Z0 (g) exp g M  2 + M 2 R2 + I2 2MM  R /I ,
(3.3)
and
Z0 (g) =

g
1/2

I (q)(q)

(3.4)

= R + iI = 2 /1 is the complex aspect ratio (modular parameter) of the torus (I =


l2 / l1 > 0), q = e2i , and

426

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

(q) = q 1/24




1 qn

(3.5)

n=1

is the Dedekind function. We note that the Virasoro central charge of the continuum CFT
is
c = 1 6e02/g,

(3.6)

and the coupling g for the scalar field depends on the details of the local weights in general.
The expression (3.1) is clearly not modular invariant. However, the underlying unmodified
partition function, in which the weight is always the cosine factor, is modular invariant by
construction, because of the sum over all winding numbers that it contains. Zmod does not
have to be modular invariant, as it is simply a generating function for the spectrum.
Next we perform a Poisson resummation on the M  variable. For each M = 0, this has
to be done by grouping the terms into sums of ZM  ,M (gf 2 ) over M  a multiple of some
positive integer (possibly 1). As in Ref. [48], the general structure of the result is
Zmod [g, 2f, e0 , ]

1

=
q e0 +P /f,0 (g) q e0 +P /f,0 (g)
(q)(q)



mod (M, N; f, e0 , )q P /(Nf ),Mf (g) q P /(Nf ),Mf (g) .

(3.7)

M>0,N>0,
P :N|M,P N=1

The summations are over the integers, except for restrictions shown explicitly; note that,
for any integers N , M, N|M means M is divisible by N . Also

2


1 e
2
e,m (g) = 1 e mg ,
e,m (g) =
(3.8)

+m g ,
4
g
4
g
for any values of the electric charge e and the magnetic charge m. The coefficients
mod (M, N; f, e0 , ) are derived in Appendix A, with the result (more compact than that
in Ref. [48])


 N (M/d)
Nd
mod (M, N; f, e0 , ) = 2
(3.9)
 N  w(M, d; f, e0 , ).
M Nd
d>0:d|M
In this expression, and are the Mbius and Euler functions, respectively, the definitions
of which are recalled in Appendix A. The unmodified partition function Z is given by the
same expressions but with in place of mod ; differs from mod by having cos 2df e0
in place of w(M, d; f, e0 , ), irrespective of whether M/d is even or odd.
In terms of CFT, the right- and left-moving conformal weights of the states are found by
and comparing with the form
expanding Zmod in powers of q and q,


Zmod = (q q)
c/24 Tr q L0 q L0 ;

(3.10)

is the eigenvalue of L0 (L
0 ).
the right- (resp., left-) moving conformal weight h (h)
0 )/ l1 and momentum
Alternatively, in terms of the Hamiltonian H = 2(L0 + L

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

427

0 )/ l1 , the energy and momentum eigenvalues (measured from the ground


P = 2(L0 L
l1 , P = 2(h h)/
l1 . The momentum is related to the
state values) are E = 2(h + h)/

conformal spin s = h h, which is always an integer.


For the percolation-type, sl(n + m|n)-invariant models, M  and M are both even, hence
so is M  M. Then we can move a factor of two into 0 and , so now f = 0 / , and the
modified partition function is given by Zmod [g, 20 = 2f, e0 , 2], as given in Eqs. (3.7)
(3.9). We are now ready to find the spectra of the various models discussed in Section 2.

4. Results
In this section we apply the formulas of the preceding section to the models in Section 2,
by recalling the values of the parameters that enter in the continuum limit as in Ref. [48].
We begin with the percolation (CP n|n ) case as it has direct application to the spin QH
transition of recent interest. We will refer to states or to operators interchangeably,
where the operators are those that correspond to the states. This does not rule out the
possible interest of operators that do not correspond to states, but they would be nonlocal
with respect to the variables used in the construction of our models.
4.1. CP n|n model at = , and percolation
To obtain the model of percolation hulls, or the CP n|n sigma model at = , we have
m = 2 cos(e0 /2) = 1 (we use e0 = 2/3), f = 1/2, 2 cosh = 2n + 1, and we use the
result that in the continuum theory, g = 8/3 [48], and hence c = 0. By the remarks at
the end of Section 3, we must calculate Zmod [8/3, , 2/3, 2]. First we point out some
conformal weights of interest that appear in the spectrum. If we ignore the contributions
from expanding the products in the Dedekind functions, which we view loosely as giving
descendants, then the scalar (momentum or conformal spin zero) states are either of
purely electric type, with dimensions
(3P + 1)2 1
1
=
,
h = h = e0 +2P ,0
24
24
or of purely magnetic type, with dimensions

(4.1)

4M 2 1
1
h = h = 0,M/2
(4.2)
=
.
24
24
The latter correspond to the M-hull operators (M > 0) [60], while the former are
thermal [62], and the sequences overlap infinitely often. There are many other values at
both zero and nonzero conformal spin. For M > 0, the general right-moving conformal
weight is given by
(3P /N + 2M)2 1
(4.3)
24
(plus positive integers in the case of descendants), and the conformal spin is s = P M/N
(plus integers). This h corresponds to position (P /N, M) in the Kac table of c = 0
h=

428

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

Virasoro representations. Since N can be arbitrarily large for some M (as long as N|M
and N P = 1), many non-degenerate Virasoro representations occur in the spectrum.
These include some with h in the range 1/24 < h < 0, however, even when h or h is
negative, h + h (or the excitation energy) is positive. The multiplicities of these states are
nonzero, since the lowest weight h at which a given N occurs is when N = M and P is
close to 2M 2 /3 (such a state has conformal spin of order 2N 2 /3 for large N ). Then
the multiplicity is just mod (N, N), which goes as (2 cosh 2N)/N for large (or n).
Thus there are infinitely many equivalence classes of conformal weights under congruence
modulo 1. This implies that, even though the conformal weights are all rational numbers,
the theory is not a rational CFT, since there can be no chiral algebra (containing only chiral
fields of integer spin) under which the states fall into a finite number of representations.
Next, we give some results for the total multiplicities (including descendants) of lowlying levels. The ground state h = h = 0 is non-degenerate. The next-lowest levels are the
one-hull operators, with h = h = 1/8, which occur in the two series mentioned above. The
total multiplicity with which these occur in Zmod is
D1 = 1 + mod (1, 1) = 1 + 2 cosh 2 = (2n + 1)2 1 = Dadj ,

(4.4)

the dimension of the adjoint representation of sl(n + 1|n), as expected [13]. It is significant
that this number is positive, even, and can be identified as (in general, a sum of) dimensions
of representations of sl(n + 1|n) (it must be even, since except for the ground state all
states must cancel in pairs when Z is calculated). If the CFT underlying this system
contained right- and left-moving affine Lie superalgebra of sl(n + 1|n) at some level,
then the multiplicities of primary fields would have to be products, or in the simplest case
squares, of dimensions of representations of sl(n + 1|n). Taking into account the leftright
symmetry of the system, that is obviously not the case.
The next levels include the two-hull operators, with h = h = 5/8. These occur for
M = 0, P = 1, and M = 2, P = 0, N = 1. The total multiplicity is
D2 = 1 + mod (2, 1) = 1 + cos e0 + cosh 4
= (2n + 1)4 /2 2(2n + 1)2 + 3/2.

(4.5)

For n = 1, this is D2 = 24. This may possibly be a sum of adjoints (each of dimension 8)
and an indecomposable representation of sl(2|1) of dimension 8 (discussed further in
Appendix B). The singlet operator in the indecomposable can appear as a relevant
thermal (non-symmetry-breaking) perturbation of the critical theory.
Among non-scalar operators, the most interesting are the spin-1 currents h = 1, h = 0.
They are obtained from (i) M = 0, as a descendant of the identity, (ii) M = 1, P = 1,
N = 1. The total multiplicity is 1 + mod (1, 1) = D1 = Dadj . Thus they are precisely the
currents of the global sl(n + 1|n) supersymmetry, which must exist by Noethers theorem,
but which are apparently not chiral since if they were they would generate an affine Lie
superalgebra. Hence, they must be logarithmic operators [47]: even though h = 0, these
operators have nontrivial action on the left-moving sector.
There is also a large multiplet of fields (one of them being the stress energy tensor) with
h = 2, h = 0: the total multiplicity is D1 + D2 , which may be interpreted as descendants of

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

429

the currents, together with a primary multiplet of dimension D2 . The latter is presumably
again an indecomposable representation, since the stress tensor, which is invariant under
sl(n + 1|n), must be paired with other fields so that Z can be 1 [63].
= (1, 1). The total
Finally, we mention the marginal operators of dimension (h, h)
multiplicity turns out to be 2D1 . It is likely that these actually form two adjoints, and
that there is no invariant marginal operator. For n > 0, this is much less than would exist if
there were right- and left-moving chiral currents of weight 1, that would generate an affine
Lie superalgebra; in that case there would be at least (Dadj )2 weight (1, 1) states. Since
there are not, we again conclude that the currents are not chiral.
We should comment here that the weights (1/3, 1/3) do not appear in the spectrum.
These were proposed [13] as the weights of a local spin-zero symmetry-breaking
perturbation that had been studied numerically [11]. Problems with this proposal were
pointed out previously [14,63]. The present results show that no such local operator exists
in the model in question, and the operator in the superspin chain found in Ref. [13]
must have some other conformal weights in the long-distance limit. The only spin-zero
operators close by are the one- and two-hull operators identified above. It was argued that
the requisite operator appears as the leading term in the operator product of the one-hull
operator with itself. Since the one-hull operator has even parity [13], it can appear in the
product and now seems to be the best candidate. This is consistent with the fact that the
operator in Ref. [13] is not a singlet, but part of a multiplet, which however does not
contain a singletjust like the one-hull operators in the adjoint representation. This would
=
predict that the exponent of Ref. [11] (see also Ref. [13]) should be = 2/(2 h h)
8/7  1.14, within  25% of the numerical value, 1.45.
4.2. Critical S2n1|2n model and polymers
A similar analysis can be done for the dilute polymer model, that is the critical point
in the S 2n1|2n sigma model. In this case, the parameters are m = 2 cos(e0 ) = 0 (we
use e0 = 1/2), f = 1/2, 2 cosh = 4n, g = 3/2. The modified partition function is
Zmod [3/2, , 1/2, ]. Many aspects of the results are similar to the CP n|n case, already
discussed: the theory is irrational, since it contains infinitely many conformal weights that
do not differ by integers. The weights are
h=

(8P /N + 3M)2 4
96

(4.6)

plus non-negative integers for M > 0, and h = [(4P + 1)2 1]/24 plus non-negative
integers for M = 0. The spin zero operators include some with the former weights with
P = 0 (the M-leg operators, M > 0 [61]), and the latter weights which are those of thermal
operators [56,64].
Some of the low-lying states are as follows. The lowest excited states are at h =
h = 5/96, the 1-leg operators. Their multiplicity is D1 = mod (1, 1) = 4n. This is the
dimension of the defining (vector) representation of osp(2n|2n), as expected (and see
a similar result in Ref. [48]). As in the discussion of percolation, this speaks against

430

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

identifying the CFT as containing an affine osp(2n|2n) Lie superalgebra. The next lowest
spin-zero states are the thermal, 2-leg operators, with h = h = 1/3, and multiplicity D2 =
1 + mod (2, 1) = 8n2 . It makes sense to identify this representation of osp(2n|2n) with
the supersymmetric second-rank tensors built from the graded tensor product of the vector
representation with itself, which has the correct dimension (supersymmetric here means
they are symmetric in the eveneven, and antisymmetric in the oddodd components). It is
indecomposable and contains a singlet. (For n = 1, we can use the isomorphism osp(2|2)

= sl(2|1), and this indecomposable representation is the same as the 8-dimensional one
mentioned in the last subsection.) In a LandauGinzburg view of the critical theory as
an RG fixed point in the theory (1.8), this singlet operator would be , which at
= 0 clearly lies in a multiplet of similar supersymmetric tensors, and this should be
maintained at finite (and hence in the critical theory), because the degeneracy of the
multiplet is needed to maintain the partition function Z = 1 throughout the RG flow.
(Similar arguments regarding the multiplicities of energy levels at each momentum in a
finite size system can be given for the sl(n + m|n)-invariant models, using the sigma model
at weak coupling. Of course, such arguments cannot predict coincidences of energies that
occur in the critical theory, or at special values of the coupling.) There are also other
relevant spin-zero (h = h < 1) operators.
At nonzero spin, there are 8n2 currents with h = 1, h = 0. The currents have multiplicity D2 , which equals the dimension Dadj of the adjoint representation of osp(2n|2n), as
appropriate for the currents. The adjoint representation consists of the super-antisymmetric
second-rank tensors, that are antisymmetric in the eveneven part, and so on. At weights
= (2, 0), there is a total of D1 + D2 states. Even though the subset of D2 states look
(h, h)
formally like descendants of the currents, we prefer to interpret them as transforming in the
supersymmetric representation, with the singlet as the stress tensor [63]. This assignment
would again be expected from the LandauGinzburg point of view, where the stress tensor
has the form + , and so is part of the supersymmetric representation. This is clear
at = 0, and should be maintained at finite , and hence in the critical theory. However, the
currents might be expected to have Virasoro descendants at level 1, so our interpretation
may have some problems in CFT, if not in LG theory; this issue is presumably connected
with the logarithmic operators that are part of the currents.
Finally, there are marginal operators at (1, 1), with multiplicity 2D2 . For n > 0, this
is again fewer than in current algebra. This multiplet is most likely two adjoints, as in
percolation, so that there are no invariant marginal perturbations.
4.3. CP n1|n model and dense polymers
For dense polymers, we first discuss the model with psl(n|n) supersymmetry, then the
osp(2n|2n)-invariant model, which contains additional scaling dimensions. For the former,
the modified partition function we require is Zmod [g, 2f, e0 , 2], with g = 2, f = 1/2,
e0 = 1, 2 cosh = 2n. The central charge is c = 2.
For n = 1, the expression for Zmod simplifies drastically. This is the case in which the
lattice model is essentially a psl(1|1) spin chain, which is a free fermion system that can

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

431

be solved directly, so we expect to obtain the modified partition function of symplectic


fermions [53]. The expression Eq. (3.1) for Zmod becomes for the percolation-type models



Zmod [g, 20 = 2f, e0 , 2] =
(4.7)
ZM  ,M gf 2 w(M, d; f, e0 , 2),
M  ,MZ

w(M, d; f, e0 , 2) = cos(2df e0 )M/d0 + cosh(2d)M/d1 ,

(4.8)

where again d = M  M. Inserting the above values for dense polymers and n = 1, the
Poisson resummation and use of the Jacobi identity give
Zmod [2, , 1, 0] =

1
(q)(q)

q (2e+m)

2 /8

q (2em)

2 /8

m2Z,
eZ+1/2

2



 
 1/12  
n 2
= 4q
1 + q  = det(DAP ),



(4.9)

n=1

where DAP is the Laplacian on the torus, with periodic boundary condition along the 1
cycle, and antiperiodic along the 2 cycle. Thus this is indeed the free fermion partition
function. This conformal field theory contains the psl(1|1) affine Lie superalgebra in its
chiral algebra (as remarked in a similar context in Ref. [57]); the right-moving currents are
, .
For n > 1, there are in general no cancellations, and the situation is similar to the
percolation and dilute polymer cases. The values for the conformal weights, neglecting
descendants, are
h=

(2P /N + M)2 1
,
8

(4.10)

for M > 0, and h = [(2P + 1)2 1]/8 for M = 0. In particular, for spin zero, these give
only the values h = h = (M 2 1)/8 (M > 0) which are the dimensions of the 2M-leg
operators [57]. These are expected as our model contains only even numbers of polymer
loops winding on the torus, and in our model local operators always conserve the number
of polymer lines mod 2. All conformal weights are non-negative.
The ground state is degenerate, and its degeneracy is D1 = 2 + mod (1, 1) = 4n2 . This
was expected, as a natural generalization of the four-fold degeneracy of the symplectic
fermion ground state for n = 1 [65], as in both cases the states form the indecomposable
adjoint representation of gl(n|n) (isomorphic to V V ). The singlet is the identity
operator, and the multiplet represents the 2-leg operators. Notice that this multiplet cannot
be decomposed into representations of right- and left-moving psl(n|n) algebras, even for
n = 1. The next spin-zero states occur at h = h = 3/8, the 4-leg operators. The multiplicity
for general n is D2 = mod (2, 1) = 8n2 (n2 1), so they are absent for n = 1. At nonzero
= (1, 0) a total of 2D1 + D2 = 8n4 states. These are more than
spin, we find at (h, h)
2
simply the 4n 2 currents associated with psl(n|n) symmetry. At (2, 0) we find a
multiplicity 3D1 + D2 = 4n2 (2n2 + 1). The marginal operators, h = 1, h = 1, have total
multiplicity 16n2 (4n4 3n2 + 2)/3 (which is integer for all n). (All these multiplicities

432

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

agree with those for free fermions at n = 1.) For n = 1, the only invariant marginal operator
is redundant, corresponding to the fact that in the free fermion theory, a change in coupling
g2 can be scaled away and has no effect. As explained in Section 2.1, if our theory
describes a special point on the CP n1|n sigma model line of CFTs, an exactly marginal
(but not redundant) operator would be expected for n > 1. There is nothing to rule out the
possibility that the invariant operator of the n = 1 case becomes such an operator for n > 1,
and the identification is presumably correct.
An alternative model of dense polymers is the low T regime of the osp(2n|2n)invariant polymer-type model (2.9). The continuum limit of this model is the same as
for the psl(n|n) model, except that the restriction in the latter to even numbers of loops
M  , M winding the torus is dropped, which means that we use f = 1/4, (not 2),
and 2 cosh = 4n for osp(2n|2n) symmetry. Then the modified partition function we
need is Zmod [2, /2, 1, ]. As expected, in this theory, M-leg operators [57] with all
M > 0 occur, in particular M = 1, with conformal weight h = h = 3/32 [56]. The
latter has multiplicity 4n, the dimension of the vector representation. The identity and
2-leg operators at h = h = 0 have multiplicity 8n2 in this case, which we presume is
the indecomposable supersymmetric representation, with the identity as the singlet. The
structure of higher levels is generally similar to the psl(n|n) dense polymer model: one can
check that all conformal weights in the spectrum of the latter still occur, but there are many
additional weights, and the multiplicities appear to reflect the osp(2n|2n) symmetry, as in
the examples just given.
4.4. Other supersymmetric models
Here we briefly consider some other models that arise by taking m = 0, 1 in the sl(n +
m|n)-invariant models, or m = 0 in the osp(2n + m|2n)-invariant models.
One such model is obtained at m = 2 in the supersymmetric vertex model, or the
CP n+1|n sigma model at = . For n = 0, this is simply the SU(2)-invariant point of the
six-vertex model, or spin-1/2 Heisenberg spin-chain, without staggering of the couplings.
The critical theory is the SU(2) WZW model at level 1. For n > 0, operators with the
same conformal weights as those in the latter theory will appear, and possibly others. We
study this using our modified partition function, with parameter values m = 2, so e0 = 0,
f = 1/2, 2 cosh = 2n + 2, and g takes the self-dual value g = 4. The central charge
c = 1, and the conformal weights are
h = = (P /N + M)2 /4,

(4.11)

plus positive integers. For n = 0, one checks (using Mbius inversion again on G in
Eq. (A.10)) that our formulas reduce to the standard Coulomb gas partition function; in
this case mod (M, N) = 2N,1 . However, for n > 0, all values N > 1 show up, and the
structure is quite similar to the CP n|n case: the theory is not rational for any choice of
chiral algebra.
Some low-lying conformal weights are (1/4, 1/4), which has multiplicity 2+mod (1, 1)
= (2n+2)2 . This is correct for n = 0, and for general n is the square of the dimension of the

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

433

fundamental representation. However, since the fundamental of sl(n + 2|n) is not self-dual,
the WZW model for this symmetry algebra should have twice this multiplicity for these
states (thus this WZW model does not go continuously to the limit n = 0). This is larger
(by 1) than the dimension of the adjoint, which would be the answer expected from the
weak-coupling sigma model analysis alluded to in Section 4.2 (since the partition function
Z > 1 in the present case, such isolated singlets are possible above the ground state), but
it is the minimal result that agrees with the n = 0 case. The currents, with weights (1, 0),
have multiplicity 1 + mod (1, 1) = (2n + 2)2 1 = Dadj again in this case. The next lowest spin-zero states are at (1, 1) and have total multiplicity 3 + mod (2, 1) + mod (1, 1) =
(2n + 2)4 /2 + 1. For n = 0, this is 9 = (Dadj )2 , but for n > 0 is again less than (Dadj )2 .
Even though for n = 0 this theory is precisely a WZW model, for n > 0 it is not, and the
operators in it that correspond to those for n = 0 must have operator products that involve
those not present at n = 0, and that violate the current algebra structure.
It may be useful here to say a few words more formally about the relation of the theories
at different n, which we have mentioned informally or by ad hoc arguments at various
points in this paper. The relevant set-up has been described already in Ref. [32], so we

can be brief. The point is that a suitable element Q = i Qi , a sum of positive odd roots
n
2+
bi
(i even and the corresponding sl(n + 2|n) generators for i odd) in the
Qi = =1 fi
superalgebra, obeys Q2 = 0, and generates a correspondence of the even and odd states that
we would like to cancel. In the space of states, the cohomology space Ker Q/ Im Q (where
Ker is the kernel, Im is the image, of a map) can be shown explicitly to be isomorphic
to the n = 0 Hilbert space. More generally, one can relate n to 0  n < n in this way,
and similar relations hold for other m and for osp algebras. The algebras of operators on
these spaces are similarly related by taking the cohomology of the algebra in the larger
n theory (where Q acts on operators by graded commutation), both in finite size and in
the continuum CFTs. In particular, note that the chiral algebra of the cohomology can be
larger than the cohomology (which is chiral) of the chiral algebra. In the CP n+1|n case just
discussed, the chiral algebra of the cohomology is the SU(2)1 affine Lie algebra, but this is
not contained in the chiral algebra of the sl(n + 2|n) invariant theory. The chiral algebra of
the latter may be more like a W-algebra, as in the theories in Ref. [32]. Similar arguments
apply to the cancellation of integrations over even and odd variables in the integrals that
we have also used earlier.
In the osp(2n+m|2n)-invariant models, the cases m = 1 and m = 2 give supersymmetric
extensions of the Ising and KT transitions, respectively. Similar phenomena as for the
SU(2) model just discussed are expected here; they are not be described by osp(2n + m|2n)
affine Lie superalgebras. In particular for m = 1, n = 0, the Ising transition is related to a
free Majorana fermion field. The loop gas description reproduces the correct modularinvariant partition function on the torus for the Ising model [48]. For the critical point in
the S 2n|2n sigma model with n > 0, one might imagine that the theory becomes 2n + 1
free Majoranas, and n beta-gamma (bosonic ghost) pairs, which is a representation of
osp(2n + 1|2n) current algebra at level 1 (using the normalization for the level that is
standard for O(m) current algebra) [66]. However, that is not what occurs in our model,
and the theory is interacting, not free.

434

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

The other case m = 2 in the osp(2n + m|2n)-invariant, polymer-type models is


interesting. The partition function we find is of similar structure as the others. This S 2n+1|2n
sigma model possesses a line of fixed points with continuously varying scaling dimensions
(like the CP n1|n model, though here c = 1), and the KT point, at which we find the
spectrum, is a special point where all the dimensions become rational. The multiplicities
are quite similar to other cases. For example, the spin-zero states include the vector
representation of multiplicity 4n + 2 at h = h = 1/16, and a traceless supersymmetric
tensor, of multiplicity (4n + 2)2 /2, at h = h = 1/4. The states at h = 1, h = 0 have
multiplicity (4n + 2)2 /2 1, the dimension of the adjoint, so they can be identified with
the currents. There are (4n + 2)4 /4 marginal operators at h = h = 1. This coincides with
(Dadj + 1)2 in this case, so is actually larger then (Dadj )2 . We expect that this set includes
fourth-rank supersymmetric tensors, and others, and there should be just two invariant
marginal operators. One of the perturbations represents a change in the coupling g2 in the
S 2n+1|2n sigma model, which in the absence of the other perturbation is exactly marginal.
By perturbing the action by both these operators, it should be possible to reproduce the
Kosterlitz RG flow diagram near the KT fixed point [38].
The case m = 2 in the critical O(m) (S 2n3|2n ) models is also amusing. Here
the LandauGinzburg theory (1.8) predicts that, for n = 1, the theory is free massless
symplectic fermions [29], and c = 2. For n = 1, we do indeed find the same critical
modified partition function as in Eq. (4.9). For n > 1, we find that the set of conformal
is the same as in the psl(n|n) (CP n1|n ) dense polymer model, but the
weights (h, h)
multiplicities are different, and reflect the dimensions of representations of osp(2n 2|2n).
For example, at h = h = 0, there are 4n operators, which we interpret as two singlets plus
the vector. One invariant operator is the identity, the other is the relevant perturbation that
corresponds to in the LandauGinzburg picture, which for n = 1 is just the fermion
mass term. At (1, 0), we find 8n2 4n + 2 operators, which we interpret as the 8n2 8n + 3
currents in the adjoint, plus the derivatives of the (0, 0) fields other than the identity. At
the next-lowest spin-zero weights, h = h = 3/8, we find 8n(n 1) states, the dimension
of the traceless supersymmetric second-rank tensor of osp(2n 2|2n). In the Ginzburg
Landau picture, the trace of the multiplet of supersymmetric second-rank tensors in is
the invariant identified at h = h = 0; this is allowed by symmetry requirements. It is
clear that the theory for n > 1 is not free scalar fields.

5. Conclusion
To conclude, we have found a rich structure in the conformal field theories of a number
of critical points that arise in nonlinear sigma models with target spaces CP n+m1|n and
S 2n+m1|2n , and are related to models in statistical mechanics such as percolation and
polymers (critical and low T ). For m an integer in the range 2  m  2, these give 14
different series of theories, of which we explicitly mentioned results for 8 (two further cases
with m = 2 lie at c = , so should be excluded). All the theories contain only rational
conformal weights, yet except in some exceptional cases at particular n, for n > 0 there are

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

435

infinitely many fractional conformal weights, and there can be no chiral algebra that will
organize them into a finite number of representations, so the theories are not rational. None
of the theories with n > 0, except the psl(1|1) model, contains the affine Lie superalgebra
associated with its global supersymmetry in its chiral algebra, as shown by the absence of
commuting left and right actions of the global symmetry on the multiplets of states, and
other reasons. On the other hand, the large multiplicities of the states point to an enlarged
global symmetry algebra of nonlocal charges, which is at least as large as the commutant
of the TemperleyLieb algebra in the lattice models (or some analogue in the osp-invariant
cases).
There are many interesting open problems. It would be of great interest to find the full
symmetry structure of the operators, the chiral algebra, and the operator product expansions
of the fields. With this information, it may be possible to extend the results to the n > 1
spin quantum Hall transition, and even to the integer quantum Hall transition and other
noncompact supersymmetric nonlinear sigma models in general.

Acknowledgements
We are grateful to E. Witten for discussions of the CP n1|n nonlinear sigma model, and
to I.A. Gruzberg and A. Ludwig for discussions and help with the references. This work
was supported by the NSF under grant no. DMR-98-18259 (N.R.), and by the DOE (H.S.).

Appendix A. Resummation of modified partition function


In this appendix we give the derivation of the result (3.7), (3.8), (3.9) from Eq. (3.3).
First, in the M = 0 terms, M/M  M is always even, and as in Ref. [48] these terms
can be resummed to give

1

q e0 +P /f,0 (g) q e0 +P /f,0 (g) .
(A.1)
(q)(q)

In the M = 0 terms, the terms |M| give the same result, and we write them in terms of
M > 0, with a factor of 2. Then for each M > 0, the sum can be written




ZM = 2
(A.2)
w(M, d; f, e0 , )
ZM  ,M gf 2 .
d>0:d|M

M  :M  M=d

We need to find the sums





ZM  ,M gf 2
f (d) =

(A.3)

M  :M  M=d

(in the following, d, d  , etc., will always stand for positive divisors of M, i.e., d|M). On
the other hand, sums over M  of the following form can be Poisson-resummed:




1

g(d)
(A.4)
ZM  ,M gf 2 =
q P /(df ),Mf (g) q P /(df ),Mf (g).
d(q)
(
q)



M :d|M

436

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

Clearly, we have

f (d  ).
g(d) =

(A.5)

d  :d|d 

Then one of the Mbius inversion formulas (pp. 234239 in Ref. [67]) can be used to
obtain

(d  /d)g(d  ).
f (d) =
(A.6)
d  :d|d 

The Mbius function is defined by (n) = (1)r , if n is an integer that is a product n =


r
i=1 pi of r distinct primes, (1) = 1, and (x) = 0 otherwise or if x is not an integer.
The two central properties of (x) that lead to the Mbius inversion formulas are [68]

(A.7)
(d  /d) = d,d  ,
d  :d|d  ,d  |d 

which we use here, and



(d  /d  ) = d,d  .

(A.8)

d  :d|d  ,d  |d 

These can be proved by noticing that for d = d  , there are finitely many terms, and they
cancel in pairs. Hence we have


1

ZM =
(A.9)
G(M, d  ; f, e0 , )
q P  /(d  f ),Mf (g) q P  /(d  f ),Mf (g) ,
(q)(q)



d

where
G(M, d  ; f, e0 , ) = 2

w(M, d; f, e0, )(d  /d)/d  .

(A.10)

We now wish to group together the terms in which P  and d  have common factors. For
each term in the sum, let d  = P  d  , and P  = P d  , d  = Nd  , and we can use N > 0
as a summation variable in place of d  . Then


1

ZM =
G(M, d  ; f, e0 , )
q P /(Nf ),Mf (g)q P /(Nf ),Mf (g)
(q)(q)


N>0,P :
N|d  ,P N=1

mod (M, N; f, e0 , )

N>0,P :
N|M,P N=1

where
mod (M, N; f, e0 , ) = 2

1

q P /(Nf ),Mf (g) q P /(Nf ),Mf (g) ,
(q)(q)

(A.11)

w(M, d; f, e0, )(d  /d)/d  .

(A.12)

d,d :N|d  ,d|d 

Finally, the sum over d  can be performed:






(d  /d)/d  = M 1
t M/(dt) = cM/d (M/N)/M,
d  :N|d 

t >0:t |(M/N)

(A.13)

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

437

where t is an integer, and cn (m) is Ramanujans sum (Ref. [67], Theorem 271). By
Theorem 272 of Ref. [67], and some elementary manipulation, we obtain


 (d  /d) N (M/d)
Nd
(A.14)
=
 N  ,
d
M Nd


d :N|d

and Eq. (3.9) follows. Here (n) is Eulers totient function, which is defined for positive
integers n as the number of integers m such that 1  m  n and n m = 1, and is given
explicitly by

(n) = n
(A.15)
(1 1/p),
p:p|n

where the product is over primes p. This completes the derivation of Eq. (3.7).

Appendix B. Finite size and completeness of states


In this appendix we supplement the main text by presenting results for one case,
the percolation model, in a finite size system with free, rather than periodic, boundary
conditions. In this model, we can verify that our set of states is complete, that is, the number
of states of the supersymmetric chain coincides with the number of states counted by the
modified partition function. The exponents found are the boundary hull exponents, and
their descendants. The continuum limit for free boundary conditions in other cases can be
found similarly. We also sketch a partial proof for completeness of the states for periodic
boundary conditions.
We consider the system with free boundary conditions. We write the operators Ti defined
in Section 2.1 as
Ti = (1 pA ) + pA ei

(i even),

Ti = pB + (1 pB )ei

(i odd).

(B.1)

Then ei , i = 0, . . . , 2l1 2, satisfy the TemperleyLieb algebra relations


ei2 = mei ,

ei ei1 ei = ei ,

[ei , ej ] = 0,

j = i, i 1

(B.2)

(in fact, these are satisfied by ei for all i = 0, . . . , 2l1 1). One can also introduce
a trace Trm , which in the supersymmetric construction is Trm = STr, and which
obeys m Trm ei W = Trm W (i = 0, . . . , 2l1 2), where W is any string of ej s with
j = 0, . . . , i 1. The relations (B.2), together with the normalization for the trace Trm 1 =
m2l1 , are sufficient to evaluate the trace Trm of any product of ei , with i = 0, . . . , 2l1 2,
independent of the vector space (or graded vector space) on which the TemperleyLieb
algebra is represented. Here we will consider only m = 1.
We are mainly interested in the spectrum of the transfer matrix obtained by multiplying
the elementary vertex matrices along a given row:
T = T1 T3 T2l1 3 T0 T2 T2l1 2 .

(B.3)

438

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

To obtain this spectrum, we consider the usual trace Tr of the transfer matrix T raised
to the power l2 , that is, the modified partition function of the model on an annulus. This
modified partition function can be expanded graphically. The result is a gas of dense loops
with weights pS or 1 pS (S = AB) for every point where loops come together, and a
global weight factor tr(1)F = 1 for contractible loops. The weight of non-contractible
loops (necessarily in the time direction, since the system has free boundary conditions in
the space direction) is 2n + 1. It would be n + 1 n = 1 if the partition function were
defined as the supertrace STr (or Trm ), leading to the partition function Z = 1. Note that
for free boundary conditions, a loop can wind the time direction at most once.
The modified partition function with weight 2n + 1 per non-contractible loop can be
computed using a mapping onto the 6-vertex model. To do so, we observe that the loop
model partition function is the same as the well-known reformulation of the Q state
Potts model (with Q = 1) partition function using the polygon decompositions of the
surrounding (or medial) lattice [49]. The mapping onto the 6-vertex model then follows
from orienting the loops, and gluing them at the interaction vertices. The resulting 6-vertex
model has anisotropy parameter = 1/2. The loop orientation can be thought of as an
z
isospin S z = 1/2, and the insertion of a term k 2S in the trace finally produces the right
weight for non-contractible loops provided
(2)k = 2n + 1,

(B.4)

where as usual we have set (x)k = (k x k x )/(k k 1 ). In the following we also use the
parameter = ln k, such that (2)k = 2 cosh .
The 6-vertex model has a Hilbert space H6-vertex which is entirely unrelated to the one of
the supersymmetric model; in particular, its dimension is 22l1 ! The point however is that the
6-vertex model is solvable, so all of the transfer matrix eigenvalues can be obtained: these
eigenvalues will also be the ones of the initial supersymmetric problem, up to multiplicities
which we now elucidate.
To proceed, we recall that the 6-vertex model transfer matrix enjoys a crucial property: it
has a quantum group sl(2)t isospin symmetry with t = ei/3 , [T6-vertex , sl(2)t ] = 0. Since
the theory of representations of sl(2)t for t a root of unity is a bit intricate, let us for a
while assume the vertex model is at some generic coupling, with t not a root of unity. In
that case, the quantum group symmetry results in a splitting of H6-vertex into irreducible
representations of sl(2)t according to
 

l1 

2l1
2l1
H6-vertex =
(B.5)

j ,
l1 j
l1 j 1
j =0

where j is the sl(2)t isospin, the j are sl(2)t representations, and the coefficients are
binomial coefficients. Then there are

 

2l1
2l1
(B.6)

l1 j
l1 j 1
representations of isospin j , each coming with its own eigenvalue. The number (B.6) is
nothing but the dimension of the irreducible representation j of the commutant algebra
of sl(2)t , the TemperleyLieb algebra.

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

439

The generating function of the eigenvalues of the transfer matrix in this representation

Zj = j l2 (that is, the generating function of the eigenvalues at isospin j ) can
be obtained by taking the generating function in the sector S z = j and subtracting
the generating function in the sector of isospin S z = j + 1: Zj = ZS z =j ZS z =j +1 .
Commutation with the sl(2)t algebra ensures coincidence of the eigenvalues that have to
be subtracted, even in finite size.
The generating function of the 6-vertex model eigenvalues reads then
Z6-vertex =

l1


(2j + 1)Zj .

(B.7)

j =0

To obtain what will be the generating function of the sl(n + 1|n) model eigenvalues after
the 6-vertex model parameters are adjusted so t = ei/3 , we need to give a weight 2n + 1
z
per non-contractible loop, that is introduce the term k 2S in the trace. Since the 6-vertex
transfer matrix commutes with sl(2)t , this simply gives a multiplicity factor which is the
k-deformation of the one in (B.7):
Zmod = Tr T l2 =

l1


(2j + 1)k Zj .

(B.8)

j =0

The decomposition of the sl(n + 1|n) Hilbert space in terms of irreducible representations
of the TemperleyLieb algebra also follows:
Hsusy =

l1

(2j + 1)k j .

(B.9)

j =0

[For n = 1, the first few multiplicities are (1)k = 1, (3)k = 8, (5)k = 55, . . . , and all are
Fibonacci numbers.] The total cardinality reads then
Dim Hsusy =

l1



(2j + 1)k

j =0

2l1
l1 j

2l1
l1 j 1


.

(B.10)

For future purposes, we note that the cardinality can also be written
Dim Hsusy =




l1 



2l1
2l1
(2j + 1)k (2j 1)k +
l1 j
l1
j =1

=2




l1 

2l1
2l1
cosh 2j +
l1 j
l1
j =1


2l
= e + e 1 = (2n + 1)2l1 .

(B.11)

We now let t = ei/3 . None of the previous results change, except that H6-vertex
splits into indecomposable representations of sl(2)t instead of irreducible ones. Similarly,
the representations of the TemperleyLieb algebra have to be handled more carefully.

440

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

However, the arguments using the counting of states still hold by continuity, and so do
those that lead to the partition functions in the continuum limit.
The conformally-invariant limit for the sl(n + 1/n) model is obtained in the isotropic
case when pS = 1/2. Up to a non-universal bulk free energy term (which is trivial here,
because the unmodified partition function is Z = 1), the expression of Zj in the conformal
limit follows from the computations in [69]. In the limit where t ei/3 , one finds
q j (2j 1)/3 q (j +1)(2j +3)/3
.
(B.12)
P (q)

n
l2 / l1 . The generating function for the spectrum
Here, P (q) =
n=1 (1 q ), and q = e
of the sl(n + 1|n) model with free boundary conditions finally follows:
Zj

Zmod =

(2j + 1)k

j =0

q j (2j 1)/3 q (j +1)(2j +3)/3


.
P (q)

(B.13)

The unmodified partition function is reproduced meanwhile by taking


Z=

l1


(2j + 1)t Zj

(B.14)

j =0

as t ei/3 . That it is equal to 1 in that limit is not obvious from this formula alone:
to prove it, one needs to invoke the additional cancellations that arise from representation
theory of sl(2)t when t = ei/3 . These cancellations completely mimic the ones of the
continuum limit, which give here
Z=


1 
1 q q 2 + q 5 + q 7 q 12 + ,
P (q)

(B.15)

which is equal to one by Eulers pentagonal number theorem.


Of course, the fact that Z = 1 follows much more directly from the fact that it coincides
with the partition function of the supersymmetric model with periodic boundary conditions
in the time direction.
In conclusion, we see that the eigenvalues of the transfer matrix based on sl(n + 1|n) are
the same as the eigenvalues of the 6-vertex transfer matrix. The only things that differ are
multiplicities, none of which are zero, i.e., any eigenvalue of one transfer matrix is also an
eigenvalue of the other.
The operator content of the sl(n + 1|n) model with free boundary conditions is therefore
made only of j hull operators with surface dimensions
j (2j 1)
3
and multiplicities (neglecting the effect of descendants this time)
hj =

(B.16)

Dj = (2j + 1)k .

(B.17)

These multiplicities are much larger than dimensions of irreducible representations


of sl(n + 1|n); the commutant of the TemperleyLieb algebra in the supersymmetric

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

441

representation used here is much larger than (the products of generators of) the global
supersymmetry sl(n + 1|n) [70]. The values of Dj can however be understood by a simple
argument: calling D1 = Dadj = (2n + 1)2 1 the dimension of the adjoint, one finds that
Dj = Dadj Dj 1 Dj 1 Dj 2 .
This is easily interpreted graphically: with a pair of non-contracted lines we associate the
adjoint. Multiplying 2(j 1) non-contracted lines by the adjoint gives either 2j noncontracted lines, or 2(j 1) if one of the added lines gets contracted with the ones existing
before, or 2(j 2) if the two added lines get contracted.
Here, it may be useful to comment briefly on the decomposition of products of the type
(V V )l1 in the case n = 1. Following notations of [71], one has for instance
V V = (0, 0) (0, 1),
(V V )2 = (0, 0) (0, 2) 4(0, 1) (1/2, 3/2)
(1/2, 3/2) [Indecomp.],

(B.18)

where in particular (0, 1) is the adjoint, and [Indecomp.] stands for an eight-dimensional
indecomposable block. Recall that the representations (b, j ) are generically typical,
with dimension 8j and vanishing supertrace, while (j, j ) are atypical, with dimension
4j + 1, and supertrace equal to 1. We observe in the previous examples that the true
singlet appears only once: it can be proved that this is a general feature (invariant vectors,
that are annihilated by all the generators, appear in the indecomposable blocks in general,
but they result from the action of some generators on other vectors inside the block).
This is compatible (though not necessarily equivalent) with having Z = 1. With our
normalizations for the transfer matrix, the eigenvalue associated with the true singlet is
= 1.
A similar approach could presumably be used to study periodic boundary conditions.
Commutation with the quantum group sl(2)t would then be replaced by the existence of
commutative diagrams as discussed in [72]. We are not going to discuss this in detail here,
but would like to indicate how the counting of states works in that case.
The modified partition function of the sl(n + 1|n) model with periodic boundary
condition in the space direction, Zmod , will be obtained by combining partition functions
of the 6-vertex model with a given value of the isospin S z = M, and a twist angle (we use
the notations of [72]). This corresponds in the Coulomb gas at coupling gf 2 = g/4 = 2/3,
to m = S Z = M, e = /2 + P , where P is an arbitrary integer. The 6-vertex partition
function in the continuum limit is

1

/2
q /2 +P ,S z (g/4)q /2 +P ,S z (g/4).
ZS Z =
(B.19)
(q)(q)

The partition function of the supersymmetric model for a given value of M reads
meanwhile, for M = 0,


1

G(M, d  ; 1/2, e0, 2)
q P  /d  ,M (g/4)q P  /d  ,M (g/4). (B.20)
ZM =
(q)(q)



d

442

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

The sum over P  can be decomposed into a sum of d  terms for all the congruences modulo
d  ; it follows that
ZM =

G(M, d ; 1/2, e0, 2)

d

 1
d

k/d 

ZM .

(B.21)

k=0

It is very likely that the same expression holds in the finite case as well. If we assume
this is true, we can check that the right number of states for the supersymmetric model is
/2
obtained. Indeed, each ZS z being a partition function of the 6-vertex model at isospin


S z accounts for l1 2l1S z states, independently of the value of the twist angle . Hence the
number of states contributing to ZM is


2l1
d  G(M, d  ; 1/2, e0, 2).
l1 M

d

Now,

d  G(M, d  ; 1/2, e0, 2) = 2

d

w(M, d; 1/2, e0, 2)(d  /d).

(B.22)

d  ,d


Recall that d|M, d  |M. Then d  (d  /d) = d,M from Eq. (A.7), and the right-hand
side reduces to 2 cosh
 2M. It follows that the number of states contributing to ZM is
2 cosh 2M l1 2l1M , for M = 0. The case M = 0 is trivial: the corresponding partition
function of the supersymmetric model has the form (A.1), and is reproduced in the 6-vertex
z
model
 2l1  by taking S = 0 and /2 = e0 , with a corresponding number of states equal to
l1 . For a system of size 2l1 , the allowed values of the isospin run between 0 and l1 , so
using (B.11) we find the right size for Dim Hsusy .
Notice that in finite size, there will be numerous coincidences of levels ensured by
the commutative diagrams of [72]. These should guarantee that subtractions of partition
functions are possible in the finite size case as well, like in the case of free boundary
conditions.
We should point out that the arguments given in this appendix generalize to other m and
n values, including the range m > 2 where the models are gapped, not conformal. This
includes the cases with n = 0, the SU(m) spin chains with alternating m (or V ) and m
(V )
representations. For 5 = 1, the eigenvalues of these models can be found from the Bethe
ansatz on the corresponding 6-vertex model [72], and the multiplicities can now be found
as here.

References
[1] A.M.M. Pruisken, in: R.E. Prange, S.M. Girvin (Eds.), The Quantum Hall Effect, SpringerVerlag, New York, 1990, Chapter 5.
[2] K. Efetov, Supersymmetry in Disorder and Chaos, Cambridge Univ. Press, Cambridge, 1997.
[3] A.M.M. Pruisken, Nucl. Phys. B 235 (1984) 277.
[4] H.A. Weidenmller, M.R. Zirnbauer, Nucl. Phys. B 305 [FS25] (1988) 339.
[5] J.T. Chalker, P.D. Coddington, J. Phys. C 21 (1988) 2665.

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

[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]
[41]
[42]
[43]
[44]
[45]
[46]
[47]

443

N. Read, unpublished.
M.R. Zirnbauer, Ann. Phys. (Berlin) 3 (1994) 513.
J. Kondev, J.B. Marston, Nucl. Phys. B 497 (1997) 639.
A. Altland, M.R. Zirnbauer, Phys. Rev. B 55 (1997) 1142;
M.R. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
T. Senthil, M.P.A. Fisher, L. Balents, C. Nayak, Phys. Rev. Lett. 81 (1998) 4704.
V. Kagalovsky, B. Horovitz, Y. Avishai, J.T. Chalker, Phys. Rev. Lett. 82 (1999) 3516.
T. Senthil, M.P.A. Fisher, Phys. Rev. B 60 (1999) 6893;
T. Senthil, J.B. Marston, M.P.A. Fisher, Phys. Rev. B 60 (1999) 4245.
I.A. Gruzberg, A.W.W. Ludwig, N. Read, Phys. Rev. Lett. 82 (1999) 4524;
Further details will be given in: I.A. Gruzberg, N. Read, A.W.W. Ludwig, in preparation.
J. Cardy, Phys. Rev. Lett. 84 (2000) 3507.
R. Bundschuh, C. Cassanello, D. Serban, M.R. Zirnbauer, Phys. Rev. B 59 (1999) 4382.
T. Senthil, M.P.A. Fisher, Phys. Rev. B 61 (2000) 9690.
N. Read, D. Green, Phys. Rev. B 61 (2000) 10267.
M. Bocquet, D. Serban, M.R. Zirnbauer, Nucl. Phys. B 578 (2000) 628.
I.A. Gruzberg, N. Read, A.W.W. Ludwig, Phys. Rev. B 63 (2001) 104422.
N. Read, A.W.W. Ludwig, Phys. Rev. B 63 (2001) 024404.
J.T. Chalker, N. Read, V. Kagalovsky, B. Horovitz, Y. Avishai, A.W.W. Ludwig, condmat/0009463.
F. Merz, J.T. Chalker, cond-mat/0106023.
R. Gade, F. Wegner, Nucl. Phys. B 360 (1991) 213.
R. Gade, Nucl. Phys. B 398 (1993) 499.
S. Hikami, M. Shirai, F. Wegner, Nucl. Phys. B 408 (1993) 413.
A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445;
A. Altland, B.D. Simons, J. Phys. A 32 (1999) L353.
M.R. Zirnbauer, hep-th/9905054.
S. Guruswamy, A. LeClair, A.W.W. Ludwig, Nucl. Phys. B 583 (2000) 475.
G. Parisi, N. Sourlas, J. Phys. Lett. (Paris) 41 (1980) L403.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183, hepth/9905111.
N. Berkovits, C. Vafa, E. Witten, JHEP 9903 (1999) 018, hep-th/9902098.
M. Bershadsky, S. Zhukov, A. Vaintrob, Nucl. Phys. B 559 (1999) 205, hep-th/9902180.
A. DAdda, M. Lscher, P. Di Vecchia, Nucl. Phys. B 146 (1978) 63.
E. Witten, Nucl. Phys. B 149 (1979) 285.
F. Wegner, Nucl Phys B 316 (1989) 663.
I. Affleck, Nucl. Phys. B 257 (1985) 397.
I. Affleck, Phys. Rev. Lett. 55 (1985) 1355.
J.M. Kosterlitz, D.J. Thouless, J. Phys. C 6 (1973) 1181;
J.M. Kosterlitz, J. Phys. C 7 (1974) 1046;
D.J. Amit, Y.Y. Goldschmidt, G. Grinstein, J. Phys. A 13 (1980) 585.
A.W.W. Ludwig, M.P.A. Fisher, R. Shankar, G. Grinstein, Phys. Rev. B 50 (1994) 7526.
A.A. Nersesyan, A.M. Tsvelik, F. Wenger, Nucl. Phys. B 438 (1995) 561.
C. Mudry, C. Chamon, X.-G. Wen, Nucl. Phys. B 466 (1996) 383;
C. Chamon, C. Mudry, X.-G. Wen, Phys. Rev. B 53 (1996) R7638.
A.A. Nersesyan, A.M. Tsvelik, F. Wenger, Phys. Rev. Lett. 72 (1994) 2628.
J.-S. Caux, N. Taniguchi, A.M. Tsvelik, Nucl. Phys. B 525 (1998) 671.
A. LeClair, D. Bernard, cond-mat/0003075.
A.W.W. Ludwig, cond-mat/0012189.
M.J. Bhaseen, J.-S. Caux, I.I. Kogan, A.M. Tsvelik, cond-mat/0012240.
V. Gurarie, Nucl. Phys. B 410 (1993) 535.

444

N. Read, H. Saleur / Nuclear Physics B 613 [FS] (2001) 409444

[48] P. di Francesco, H. Saleur, J.B. Zuber, J. Stat. Phys. 49 (1987) 57.


[49] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, New York, 1982,
Chapter 12.
[50] B. Nienhuis, J. Stat. Phys. 34 (1984) 731.
[51] I. Affleck, J. Phys. Condens. Matter 2 (1990) 405.
[52] N. Read, S. Sachdev, Nucl. Phys. B 316 (1989) 609.
[53] E.V. Ivashkevich, J. Phys. A 32 (1999) 1691.
[54] P.G. de Gennes, Phys. Lett. A 38 (1972) 339.
[55] E. Domany, D. Mukamel, B. Nienhuis, A. Schwimmer, Nucl. Phys. B 190 (1981) 279.
[56] B. Nienhuis, Phys. Rev. Lett. 49 (1982) 1062.
[57] H. Saleur, Nucl. Phys. B 382 (1992) 486.
[58] J. Des Cloizeaux, Phys. Rev. A 10 (1974) 1665.
[59] B. Duplantier, Phys. Rev. Lett. 81 (1998) 5489;
B. Duplantier, Phys. Rev. Lett. 82 (1999) 3940;
M. Aizenman, B. Duplantier, A. Aharony, Phys. Rev. Lett. 83 (1999) 1359.
[60] H. Saleur, B. Duplantier, Phys. Rev. Lett. 58 (1987) 2325.
[61] H. Saleur, J. Phys. A 19 (1986) L807.
[62] M. den Nijs, J. Phys. A 12 (1979) 1857.
[63] V. Gurarie, A.W.W. Ludwig, cond-mat/9911392.
[64] J.L. Cardy, H.W. Hamber, Phys. Rev. Lett. 45 (1980) 499.
[65] H.G. Kausch, hep-th/9510149.
[66] P. Goddard, D. Olive, G. Waterson, Commun. Math. Phys. 112 (1985) 591.
[67] G.H. Hardy, E.M. Wright, An Introduction to the Theory of Numbers, 5th edn., Clarendon
Press, Oxford, 1979.
[68] J.H. van Lint, R.M. Wilson, A Course in Combinatorics, Cambridge Univ. Press, Cambridge,
1992, Chapter 25.
[69] H. Saleur, M. Bauer, Nucl. Phys. B 320 (1989) 591.
[70] N. Read, H. Saleur, unpublished.
[71] L. Frappat, A. Sciarrino, P. Sorba, Dictionary on superalgebras, hep-th/9607161.
[72] V. Pasquier, H. Saleur, Nucl. Phys. B 330 (1990) 523.

Nuclear Physics B 613 [FS] (2001) 445471


www.elsevier.com/locate/npe

Models W Dn in the presence of disorder


and the coupled models
Vladimir S. Dotsenko, Xuan Son Nguyen, Raoul Santachiara
Universit Pierre et Marie Curie, Paris VI, Boite 126, Tour 16, 1er tage,
4 place Jussieu, F-75252 Paris Cedex 05, France 1
Received 26 April 2001; accepted 7 August 2001

Abstract
(p)

We have studied the conformal models W Dn , n = 3, 4, 5, . . . , in the presence of disorder which


couples to the energy operator of the model. In the limit of p  1, where p is the corresponding
minimal model index, the problem could be analyzed by means of the perturbative renormalization
group, with -expansion in = 1/p. We have found that the disorder makes to flow the model
(p)
(p1)
without disorder. In the related problem of N coupled regular
W Dn to the model W Dn
(p)
W Dn models (no disorder), coupled by their energy operators, we find a flow to the fixed point
(p1)
. But in addition we find in this case two new fixed points which could
of N decoupled W Dn
be reached by a fine tuning of the initial values of the couplings. The corresponding critical theories
realize the permutational symmetry in a nontrivial way, like this is known to be the case for coupled
Potts models, and they could not be identified with the presently known conformal models. 2001
Published by Elsevier Science B.V.
PACS: 12.40.Ee

1. Introduction
With the present knowledge of the effect of disorder on the critical behavior of statistical
systems it could be fair to say that any extra theoretical model will be interesting in which
these effects are nontrivial and can be studied reliably.
(p)
In this respect we find the set of conformal theory models W Dn , n = 3, 4, 5, . . ., to
be interesting. Here p is the index of the corresponding unitary series of minimal models.
These models are based on the corresponding W -chiral algebras. They have been defined
in the paper [1].
E-mail addresses: dotsenko@lpthe.jussieu.fr (V.S. Dotsenko), xuanson@lpthe.jussieu.fr (X.S. Nguyen),
santachi@lpthe.jussieu.fr (R. Santachiara).
1 Laboratoire associ No. 280 au CNRS.
0550-3213/01/$ see front matter 2001 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 3 9 2 - 3

446

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471


(p)

The simplest model in this series W D3 will be sufficient to give a preliminary


discussion of our approach and results.
The model itself, which has not much been used in statistical physics applications,
will be described in the next section. For the moment we shall observe that a particular
primary operator in this theory, which is naturally identified with the energy operator, (x),
possesses conformal dimension
=

1
2

(1.1)
(p)

ph

in the limiting theory W D3 , p . Physical dimension of will be twice bigger, =


2 = 1. For p finite, is of the form
=

1 5
,
2 2

(1.2)

where
=

1
.
p+1

(1.3)

If we couple disorder to (x)



d 2 x m(x)(x)

(1.4)

with m(x) being random valued, then the disorder produced interaction of the replicated
theory

g

d 2x

N


a (x)b (x)

(1.5)

a=b
(p)

will be marginal in the limit p , and will be slightly relevant in the model W D3 with
p large but finite. In this case the effect of disorder can reliably studied by the perturbative
renormalization group, by developing in .
We observe also that the problem formulated in this way will be defined better,
theoretically, compared to the much used Potts model, or the minimal conformal theory
models, with disorder [2,3]. In this latter case the disorder is marginal in the case of the
Ising model [4], and the parameter of the Potts model, in which, similarly, the disorder
is slightly relevant, corresponds to the deviation of the central charge, or of the number of
components, from the Ising case. The point of difference is that in this case there are no
unitary models close to Ising which would correspond to infinitesimal values of . Ising
model is not the accumulation point of the unitary series. Next to Ising, the 3-states Potts
model, lies at numerically small but finite distance in . For this reason, the perturbative
expansion in is likely to give numerical results, in case of Potts model with disorder, not
analytic ones.
(p)
This is different in case of the model W Dn where the disorder is marginal at the
accumulation point of unitary series, p = . In this case for any small value of 1/p
there would be a physical, unitary model behind. In this sense, will be a well defined

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

447

analytic parameter. One could expect that the perturbative expansion in would provide in
this case an analytic expansion in = 1/(p + 1) of unknown yet exact functions (central
charge, dimensions of operators). Like this is actually the case for the minimal model
perturbed with the operator 3,1 [5,6]. For these reasons we consider the model to be
interesting.
Another important point of difference with the Potts model is that in the operator product
decomposition of (x)(x ) there appears an extra slightly relevant operator, we shall note
it (x). The operator algebra has the form
(x)(x ) =

1
D
(I + ) +
((x) + )

2

|x x |
|x x |2

D
|x x |2

( (x) + ).

(1.6)

The detailed information on this decomposition shall be given in Section 3. For the moment
we shall observe only that one of the operators in (1.6), lets say (x), is irrelevant, so that
it can be dropped in the RG calculation. While another, (x), is slightly relevant =
1 A . This requires, in the RG calculation, to add this extra operator to the perturbative
action, so that the replicated theory will have the perturbation of the form

Apert = g

d 2x

N



a (x)b (x) +

a=b

d 2x

N


a (x)

(1.7)

a=1

with the initial values, in the RG sense, g = g0 , g0 > 0, which is due to disorder, and
= 0 = 0, initially zero, but produced further spontaneously by the interactions.
As will be shown in the next sections, the theory with Apert in (1.7) has a stable fixed
point at
= 0,

g = 0

(1.8)
(p1)
,
W D3

so that, in the limit N = 0 which


and it corresponds to N decoupled models
corresponds to problem with disorder, one gets a flow
(p)

W D3

(p1)

with disorder W3

regular, without disorder.

(1.9)

One gets this flow for N = 0, in the coefficients of the -function, and for the initial
conditions g0 < 0, 0 = 0. There are no other fixed points in this case (N = 0) in
addition to (1.8) and the trivial fixed point, unstable, g = = 0. Yet in the related problem
of N coupled regular models (no disorder), for N = 3 or bigger, the - function contains,
in addition to (1.8), two other fixed points, both stable in one direction and instable
in the other. They correspond to multicriticality. To reach them one has to have both
initial couplings g0 , 0 = 0, and, in addition their ratio have to be fine tuned. These
extra fixed points are nontrivial, in the sense of a nontrivial realization at these points
of a permutational symmetry. They are also nontrivial in the sense that the corresponding
(2) (2)
couplings (g(1) , (1)
), (g , ) are all nonvanishing. We expect that these points would
not be identified with any presently known conformal field theory, like this appears to be

448

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

the case for the coupled Potts models [2,7]. As has been argued above, we expect that in this
(p)
(p)
case, of coupled W D3 and more generally W Dn models, our calculations provide first
analytic corrections in 1/p to the development of yet unknown exact functions of ,
or p. In case of the Potts models, the corresponding corrections in have to be considered
as only numerical.
The rest of the paper is organized as follows. In Section 2 we introduce the Coulomb(p)
gas representation of the W D3 model. Then, in Section 3, we study the effects of weak
(p)
bond randomness on this model and the related problem of N coupled W D3 models.
The renormalization of the couplings and of the energy operators are computed at one loop
order. In Section 4 we study the behavior of the RG-flow in the case of disordered system
and of N coupled systems; we show in particular the presence of nontrivial fixed points in
both cases. We give the analytic expansion in of the central charge and of the dimension
(p)
of energy operators at these points. In Section 5 all the results we have found for the W D3
(p)
model are generalized for the whole family of W Dn models. We discuss what has been
obtained in the conclusions.
(p)

2. Coulomb gas of the model W D3

(p)

To make our presentation self-contained, and also because the models {W Dn } have
not much been used so far in the statistical physics applications, we shall reproduce in
this section the results of the paper [1], by giving the details on a particular model of our
(p)
interest, W D3 .
This model could be realized by a 3-component Coulomb gas, with the stress-energy
operator T (z) taking the form
1

(z):

+ i  0 2 .

T (z) = : (z)
(2.1)
4
Here (z)
 = {1 (z), 2 (z), 3 (z)} is a set of three free fields put into a vector, with the
correlation functions normalized as


1
ab
a (z, z )b (z , z ) = 2 log
|z z |2


1
1
= 2 log
+ 2 log
(2.2)
ab .
z z
z z
In the following, and in the above formula (2.1), we often suppress the dependence of
a (z, z ), on z , which will be implicit.
The vector 0 in (2.1) corresponds to a presence of the background charge operator,
V20 (R), puted at infinity, R [8].
In addition to T (z), the model contains two other operators in its chiral algebra:
W3 (z), W4 (z), with conformal dimensions 3 and 4. They could also be expressed in terms
of polynomials in derivatives of fields {a (z)} [1]. The expressions are longer, but they
will not actually be needed in the present analysis. Important is that W3 (z) and W4 (z)
exist and classify, together with T (z), all the fields (operators) of the model in terms of

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

449

primaries and descendants of the chiral algebra. Then the usual methods, combined with
the available Coulomb gas representation, define dimensions of primary operators (Kac
(p)
formula) and their correlation functions. In this sense the conformal theories of W Dn are
fully defined [1].
(p)
In the Coulomb gas of the model W D3 there are 3 screening operators +, and 3
screening operators :
+


,
Va+ (z) V a+ (z) = ei  a (z)

Va (z) V a (z) = e
 a

= ea ,

i  a (z)


(2.3)

(2.4)

(ea ) = 1.
2

(2.5)

The length of the screening operator vectors, , are fixed by the condition that the
conformal dimension of Va (z) is equal to 1. In general, for a vertex operator
V (z) = ei  

(2.6)

its conformal dimension with respect to the stress-energy operator T (z), Eq. (2.1), is given
by
 (V ) =  2 2
 0 = 2 20 cos .
(p)
Here is the angle between  0 and  . For the geometry of screenings of W D3
between {
a } and  0 is the same, Fig. 1. One gets, as a condition (Va ) = 1,

a2 2a 0 cos = 1,

(2.7)
the angle
(2.8)

(a )1,2 = 0 cos


02 cos + 1.

(2.9)

The unit vectors {ea } shown in Fig. 1 correspond to simple roots of the classical Lie
algebra D3 . As it was said above, the vector  0 is supposed to be equally distant from
e1 , e2 , e3 . Assuming this condition, for which we will not go into details here, cos will
be defined by the geometry in Fig. 1. One finds

cos = 1/ 10.
(2.10)
By formula (2.9), the lengths of the screenings in Eq. (2.5) will be functions of 0
(p)
only. This implies that in the model W D3 the orientational geometry is fixed, but there is
one free parameter in the lengths of the vectors: the screenings {
a() } and the background
charge  0 .
The primary operators of the model are represented by the vertex operators



V (z) = ei (z)
.

(2.11)

The allowed values of the vectors  are defined by the degeneracy condition of the modules
of V (z) with respect to the chiral algebra. They are found to be given by [1]
 = (n ,n1 )(n ,n2 )(n ,n3 ) =
1

3 

1 n

a=1


1 na
+
a .
2

(2.12)

450

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

(p)

Fig. 1. Screening geometry of the W D3

model.

Here {
 a } is a set of three vectors which are dual to the unit vectors of the screenings {ea }
(
 a , eb ) = ab .

(2.13)

Using the general formula for the dimensions of vertex operators, Eq. (2.7), one then gets
from (2.12) the Kac formula of the model which defines the set of dimensions of the
primary operators {(n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 ) (z)}:
(n ,n1 )(n ,n2 )(n ,n3 ) (z) V
1

(n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 )

(z),

(2.14)

(n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 )


2
= (...)
2
0 (...)


2 (
 a )2
b)
(
a ,
+2
u(n a , na )
=
u(n a , na )u(n b , nb )
4
4
a

(+ + )


a,b

a<b

u(n a , na )

(
a ,
b )
.
2

(2.15)

Here
u(n a , na ) = (1 n a ) + (1 na )+ .

(2.16)

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

In (2.15) we have used in addition the decomposition



2
0 = (+ + )

a

451

(2.17)

which is easily verified by multiplying both sides with a vector eb and using
2
0 eb = 20 cos = + +

(2.18)

by Eq. (2.9), and (


 a eb ) = ab , Eq. (2.13). The unit vectors {ea }, which are defined by
Fig. 1, when expressed in components will be given by
1
e1 = (0, 1, 1),
2

1
e2 = (1, 1, 0),
2

1
e3 = (1, 1, 2).
2

The dual vectors {


 a }, defined by the Eq. (2.13), are found to be equal to

1
1

 1 = 2(0, 0, 1),

 3 = (1, 1, 1).

 2 = (1, 1, 1),
2
2
Then one finds

2 1 1
1
3
(
a ,
b) = 1 2 2 .
1 12 23
(p)

Finally, the Kac formula of the model W D3

(2.19)

(2.20)

(2.21)

takes the following form:

(n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 )


1
3
3
= (u(n 1 , n1 ))2 + (u(n 2 , n 2 ))2 + (u(n 3 , n 3 ))2
2
8
8
1
1
1
+ u(n 1 , n1 )u(n 2 , n2 ) + u(n 1 , n1 )u(n 3 , n3 ) + u(n 2 , n2 )u(n 3 , n3 )
2
2
4 

3
3
(+ + ) 2u(n 1 , n1 ) + u(n 2 , n2 ) + u(n 3 , n3 ) .
2
2
u(n a , na ) are defined by Eq. (2.16). One checks that

(
a ,
 b ) = (4, 3, 3)

(2.22)

(2.23)

which we have used in the last term in (2.22).


As it was mentioned above, the model contains one free parameter. This will be either
0 , or + , or , or the central charge of the model, which, for T (z) in Eq. (2.1), will be
given by
c = 3 24
02 .

(2.24)

One observes that the lengths of the screenings + and the screenings , + and in
Eq. (2.9), are related
+ + = 20 cos ,

(2.25)

+ = 1.

(2.26)

452

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

Similar to the case of the basic conformal theory of [9], one gets a unitary set of models if
the parameter (+ )2 is restricted to the discrete values
(+ )2 =

p+1
.
p

(2.27)

By relation (2.26) one will have also


( )2 =

p
.
p+1

(2.28)

(p)

For W D3 , p = 5, 6, . . . [1]. In the following we shall restrict ourselves to this unitary


(p)
series. This explains for the extra index p of W D3 .
We shall next be interested in this model in the limit of p large, p  1. First, when we
take p , the Kac formula (2.22) simplifies and takes the form
+ = 1,

= 1,

u(n a , na ) = n a na ,

(2.29)
(2.30)

(n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 )


1
3
3
1
= (n 1 n1 )2 + (n 2 n2 )2 + (n 3 n3 )2 + (n 1 n1 )(n 2 n2 )
2
8
8
2
1
1


+ (n1 n1 )(n3 n3 ) + (n2 n2 )(n3 n3 ).
(2.31)
2
4
In general, not specifically for p large or infinite, the model contains a symmetry, in the set
of its primary operators and their dimensions, w.r.t. permutation of indexes 2 and 3. This
corresponds to the reflectional Z2 symmetry of the geometry of the screening vectors {ea }
in Fig. 1.
In the set of dimensions (n 1 ,n1 )(n 2 ,n2 )(n 3 ,n3 ) there is a subset which is not Z2 degenerate
and which corresponds to the singlet operators. For instance, singlet operators are
{(n 1 ,n1 )(1,1)(1,1)}. The lowest dimension operator in the singlet sector, next to the identity
operator I = (1,1)(1,1)(1,1) ((1,1)(1,1)(1,1) = 0, by Eq. (2.22)), will be the operator
(2,1)(1,1)(1,1)(x).

(2.32)

It is natural to identify it with the energy operator of the model,


(x) = (2,1)(1,1)(1,1)(x).

(2.33)

If we look now at the limiting case of p , we shall find, by formula (2.31), that
1
(p ) = .
2
For p finite, by formula (2.22), one finds
5 2
2.
= (2,1)(1,1)(1,1) =
2

(2.34)

(2.35)

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

453

Substituting, by Eq. (2.28),


2
=

1
p
=1
= 1 ,
p+1
p+1

(2.36)

where we have defined


1
=
p+1

(2.37)

one obtains
1 5
(2.38)
.
2 2
In our present study we are going to couple disorder to the energy operator (x), defined in
Eq. (2.33). In the related problem of coupled regular models, we shall couple the models
between themselves by their energy operators
N 

Apert = g
(2.39)
d 2 x a (x)b (x).
=

a=b

As has been discussed in the introduction, in both cases one gets a problem with a slightly
relevant perturbation, if is given by Eq. (2.38) with small, or p large, Eq. (2.37).
These problems will next be studied by the methods of the perturbative RG.
(p)

3. 1-loop RG equations for W D3

3.1. Renormalization of the couplings


(p)

We initially consider N regular W D3 models coupled by the energyenergy interaction (2.39)


N 
N


A(a)
+
g
d 2 x a (x)b (x).
A=
(3.1)
0
0
a

a=b
(p)

A(a)
model. Action (3.1)
0 being the conformal action corresponding to a single W D3
describes a conformal field theory perturbed by a slightly relevant term quadratic in the
energy operator; such a problem can be reliably studied by means of perturbative RG
with -expansion. The RG scheme requires that all the relevant terms produced by the
energyenergy interaction have to be added to (3.1) and the algebra of the enlarged set of
perturbing fields have not to present other relevant operators. As shown below, the OPE
of (x)(x ) contains, apart from identity, one and only one (slightly) relevant operator,
namely (1,1)(2,1)(2,1) , whose dimension is (1,1)(2,1)(2,1) = 14 . This implies that in
the perturbative computation there will be diagrams (generated by terms in (2.39) with the
same replica index) which produce, apart from trivial or irrelevant contributions, the term
N 

(3.2)
d 2 x a (x).
a

454

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

We therefore consider the more general action


N 
N 
N



(a)
A=
A0 + g0
d 2 x a (x)b (x) + 0
d 2 x a (x).
a=b

(3.3)

a
(p)

The problem of a single W D3 model perturbed by has been considered in [1], where
the existence of a nontrivial infrared fixed point to which the system flows has been shown.
(p1)
model.
The conformal field theory associated to this point corresponds to the W D3
In order to investigate the energy algebra, we study the four point correlation
function G(x) (0)(x)(1)(), which can be decomposed in a sum over s-channel
diagrams corresponding to insertions of different operators. Introducing the Coulomb gas
representation (cf. Section 2) for G(x) and defining for simplicity (2,1)(1,1)(1,1)  ,
0 (1,1)(2,1)(2,1)  we can write
2
0 (2,1)(1,1)(1,1)  , (1,1)(2,1)(2,1)  , and 2



G(x) . . .
V (0)V (x)V (1)V ()

(3.4)
V1 (1 )V1 (2 )V2 ( )V3 () d 2 1 d 2 2 d 2 d 2 ,
where the screenings Va , defined in (2.3), are integrated over the 2D plane and their
number is determined by the charge neutrality condition
2 + (m1 e1 + m2 e2 + m3 e3 ) = 0

(3.5)

which is satisfied for m1 = 2, m2 = 1, m3 = 1.


Each leading term in x 0 limit originates in the integral representation (3.4) when a
particular number qa of screenings Va (by Eq. (3.5) q1  2, q2  1, q3  1) approaches
the origin. Since in this limit

V (x)V (0) (Va )qa V2 +  qa ea (0) + ,
(3.6)
a

the intermediate state vertex V2 +  qa ea corresponds to a primary field which belongs
a
to the energy algebra. In particular we have noticed the presence of the vertex V2 + e1 ,
which corresponds to the primary operator , according to


3

2

.
+
2 + e1 =
(3.7)
2
2
In Appendix A we shall show in detail that is the only relevant operator in the energy
product decomposition and that the algebra of and , apart from irrelevant terms, is

(x)(0) =

D
I
+ 4 2 + ,
4

|x| |x|

D
I
+ 2 + ,
4
|x|
|x|

D
(x)(0) = 2 + ;
|x|

(x)(0) =

(3.8)
(p)

D and D are the associated structure constants of the W D3

model.

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

455

At one loop order the main RG characteristics of action (3.3), e.g., -functions or
renormalization of operators, are easily obtained from Eq. (3.8) (see Appendix C). The
renormalized coupling constants g(r) and (r) (r is the short distance cut-off) have been
computed, and the correspondent -functions take the form
g

g(r)

= (2 4 )g (N 2)g 2 4D
g,
ln(r)

D 2 N 1 2
(r)
= (2 2 )

D g ,
(3.9)
ln(r)
2
2
where the coupling constants have been redefined as g g/4 and /2 . In order
to study the coupling flow induced by Eq. (3.9) (Section 4) the structure constants have
been determined (see Appendix B)

5
8

D
(3.10)
+ 0( ),
D = + 0( ).
=
3
15
In Section 4 we will investigate the properties of Eqs. (3.9) for N = 0 (disorder) and
N  2 (coupled systems).

3.2. Renormalization of the energy operator


In order to study the effect of the perturbation on the energy operators, we need to
compute the renormalized operators a , which are expressed via the N N matrix [Z ]ab
by

a =
(3.11)
[Z ]ab b .
b

As usual, we proceed perturbatively: computing contributions from each coupling term and
rewriting bare quantities in term of renormalized ones, we have determined up to the first
order the matrix ab d ln[Z]ab /d ln(r) (see Appendix C). The new critical exponents at
the fixed points g , of the RG flow will be given by the eigenvalues of the dimension
matrix ab ab ab (g , ) which takes the form

1 + D
g
g

g

g
2 + D
g

g

g
g

+
D


ab =
.

N + D
g
(3.12)
It is straightforward to see that the symmetric combination s = 1 + 2 + + N and the
antisymmetric ones a1 = 1 2 , a2 = 2 3 , . . . , are eigenvectors of (3.12) and their
dimensions turn out to be

s (g , ) = + D
+ (N 1)g ,

g .
a1 (g , ) = a2 = = + D

(3.13)

456

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

In the next section we shall show that Eqs. (3.9) admit fixed points with g = 0 at which
N identical systems remain coupled; we expect therefore that the correspondent conformal
field theory realizes in a nontrivial way the permutation symmetry of N identical objects.
The splitting of dimensions in the energy sector s = a indicates that this is indeed the
case. Although a similar result was obtained for N coupled Potts models, in this case we
have access to the analytic expansion in of the dimensions of the energy operators and of
the central charge (see next section) of new conformal field theories.

4. RG-flow and central charge


4.1. Fixed point structure and couplings flow
The first step in the study of RG-flow is to find all points g , such that g (g , ) =
(g , ) = 0.
Eqs. (3.9) have been obtained for a generic number N of coupled models. We consider
firstly the quenched case which is obtained in the N 0 limit. The -functions, defined
in (3.9), take in this limit the form

5
2
g,
g = 10 g + 2g 2
3

4
1 5 2
= 8 2 +
(4.1)
g .
2
3
15
The RG flow has, apart from the trivial solution g = = 0, one stationary point

g = 0
= 2 15 ,

(4.2)

which is stable in all directions. This point has already been found in [1]: the associated
(p1)
model. Taking as initial conditions 0 =
critical theory is described by a pure W D3
0 and g0 < 0, and assuming that higher loop corrections will not change the qualitative
behavior of the flow near the two fixed points, the system flows toward point (4.2) as
supported by the numerical calculations shown in Fig. 2.
Initial conditions 0 = 0 and g0 < 0 correspond exactly to the disorder produced
(p)
couplings; so, the addition of a small bond randomness will drive the model W D3 to
(p1)
without disorder.
the model W D3
We consider now N  2, i.e., the case of coupled models: by (3.9) and (3.10) the functions are

5
2
g,
g = 10 g (N 2)g 2
3

4 2 N 1 5 2
= 8
(4.3)
g .
2
3
15
In addition to point (4.2), which remains a point of attraction of the RG-flow since the
respective RG-eigenvalues do not depend on N , we have two new fixed points

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

(p)

(1)

Fig. 2. RG-flow of the W D3

= 15 1 (N 2)


= 10

(2)




= 15 1 + (N 2)

= 10

model with disorder.


,

6
;
N 1 + 6N 2

g(1)

g(2)

6
N 1 + 6N 2

6
N 1 + 6N 2

6
.
N 1 + 6N 2

457

(4.4)

,
(4.5)

The stationary points (4.4) and (4.5) constitute node points of the flow and they can be
reached only by a fine tuning of the initial conditions. Studying in detail the flow diagram
of (4.3) (shown for N = 3 in Fig. 3) we see that with the initial conditions 0 = 0 and
g0 = 0 the system will flow far from our fixed points toward either a massive theory or
another fixed point which cannot be seen at this order in perturbation theory. 2 On the other
2 It could be mentioned that a particular case of N = 2 coupled W D (p) theories, with g = 0, = 0, have
0
0
n

458

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

(p)

Fig. 3. RG-trajectories of N = 3 coupled W D3

models.

hand, in the region > 0 and of order , the situation is rather different. There are two
solutions of Eq. (4.3), say g+ () > 0 and g () < 0, which are attracted respectively by
the node point (4.4) and (4.5); if we take 0 > 0 and g (0 ) < g0 < g+ (0 ), the system
will flow toward the stable point (4.2). In this case the infrared limit of the system will be
(p1)
models.
described by N decoupled W D3
Its important to note that, although quite similar, the flow is not symmetric along the
axis, and this will be at the origin of the difference between the values of the central
charge at the two fixed points. This asymmetry can be explained by a simple physical
argument. Indeed, when g > 0, the coupling of different models is always frustrated, while
for g < 0 the N -system can arrange itself in a kind of ferromagnetic configuration in order
to minimize the energy. For N = 2 (i.e., no frustration) we recover the symmetry g g,
and the central charge has the same value at the two new fixed points (see next section).
been considered previously in the paper [11], with a conclusion that the theory is integrable. We only want to
mention again that according to our RG analysis the second coupling, = 0, will have to be admitted eventually
into the action (see the corresponding flows in Fig. 3).
This is in contrast with two coupled Virasoro algebra minimal models Mp which were shown to be
integrable [12] and studied in detail in [11]. In this last case the original action, with a coupling term g0 only, is
stable with respect to RG evolution.
(p)
Then, if the coupling = 0 has to be taken into the action, for the two coupled W Dn theories, Eq. (3.3), the
natural question would be if the presence of the second perturbative term will modify essentially the analysis of
integrability.

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

459

Inserting the values of the structure constants (3.10) in Eq. (3.13), we have at the fixed
(1) (1)
(2) (2)
points (g , ) and (g , )



6
= + 5 1 N
,
(1),(2)
s
N 1 + 6N 2



6
(1),(2)
,
a
(4.6)
= + 5 1 N
N 1 + 6N 2
(1) (1)
(2)
(2) (2)
(1)
where (1)
s,a s,a (g , ) and s,a s,a (g , ). By Eq. (4.6) we have s =
(2)
(1)
(2)
a and a = s for all N . This result can be easily explained for N = 2: in this
case the two models are equivalent under the replacements g g and 1 1 . On
the other hand, for N > 2 no reason can be given, as it can be seen from the multiplicity
of the antisymmetric energy combinations. We believe therefore that these equalities are
accidental and they originate from the simplicity of the one-loop order computation.

4.2. Central charge


In the previous section we have shown that the RG-flow of the quenched system (N =
0) and of the coupled systems (N  2) exhibits nontrivial fixed points. A simple way
for computing the central charge of the associated conformal theories is given by the
Zamolodchikovs c-theorem [5]; the theorem provides us with a function of the couplings
c(g, ), to be defined below, whose value c(g , ) at the fixed point (g , ) is equal to
the central charge of the corresponding critical theory.
We define as (x) the trace of the stress-energy tensor. Its well known that (x), which
is zero at the fixed point, is proportional to the perturbing terms of the theory. In our case
we have from action (3.3)
N
N
g 

a (x)b (x) +
c (x),
(x) =
8
4 c

(4.7)

a=b

where the renormalization of the couplings g g/4 and /2 is taken into


account. The function c(g, ) is then completely determined by the following equations
[5]:


c(g, )
c(g, )
+
= 24 (0)(1) ,
g

c(0, 0) = cpure

(4.8)

with cpure and (0)(1) respectively the central charge and the field two-point
correlation function of the unperturbed theory. Inserting Eq. (4.7) and (3.9) into Eq. (4.8)
and using the following relations:
N 
N



a (1)b (1)c (0)d (0) = 2N(N 1),
a=b c=d

460

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471


N
N 



c (0)d (0)a (1) = 0,
c=d a=1
N 
N



a (1)b (0) = N,

(4.9)

a=1 b=1

its straightforward to verify that the function




3
(N 1)(N 2) 3 D 3
c(g, ) = cpure N
g 2 + 3 2
g

2
4
4

3
2
(N 1)D
g
4

(4.10)

(p)

satisfies Eq. (4.8). Function (4.10) has been obtained when a number N of W D3 models
(p)
(p)
is considered and so cpure = Nc(W D3 ), where c(W D3 ) is the central charge of a single
(p)
W D3 model, calculated in [1]:


(p)
c W D3 = 3 1


20
.
p(p + 1)

(4.11)

In order to compute the central charge in the case of the disordered problem, we normalize
the function c(g, ) by N and then we take the limit N 0; in fact, in terms of replicas this
is exactly the limit in which the related quenched free energy is obtained. Using the values
of the structure constants (3.10), the central charge cdis. at the infrared stable point (4.2)
turns out to be



c(0, 2 15 )
(p)
(p1)
= c W D3 120 3 c W D3
cdis. = lim
(4.12)
.
N0
N
This result is consistent with what has already been said in the previous section: the infra(p)
red behavior of a W D3 model with disorder coupled to the energy operator is described
(p1)
model.
by a W D3
In the related problem of N coupled models, the central charge ccoupl. at the fixed

(p1)
point (4.2) is ccoupl. = c(0, 2 15 ) = Nc(W D3
), i.e., the correspondent critical
(p1)
theory is described by N decoupled W D3
models.
Finally we have access to the analytic expansion up to the third order in of the central
(1) (1)
(2) (2)
charges c1 and c2 at the new fixed points (g , ) and (g , )
c1,2  Nc

(p)
W D3 N



60 1 (N 2)

6
N 1 + 6N 2


3.

(4.13)

Eq. (4.13) represents the first analytic result for a new series of conformal theories which
in addition to the W -symmetry present a nontrivial representation of the permutation
symmetry.

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471


(p)

5. Analysis of the W Dn

461

models

We shall show that the results we have obtained in the previous sections still hold in
(p)
the case of W Dn models. The construction of the Coulomb-gas representation of these
(p)
theories is a direct generalization of what has been presented in Section 2. The W Dn
model presents, in addition to the conformal symmetry, a series of additional symmetries
which are generated by a series of local currents {W2k (x)}k=1,...,n1 and Wn (x) with
dimension 2k = 2k and n = n, respectively. It can be represented by an n-component
Coulomb gas. Eqs. (2.3), (2.4) and (2.5) still define the screenings + and , with
unit vectors ea which lie on a n-dimensional space and correspond to simple roots of Lie
algebra Dn . The primary operators are represented by vertex operators V with
 (n ,n1 )(n ,n2 )...(n n ,nn ) =
1

n 

1 n

a=1


1 na
+
+
a ,
2

(5.1)

where the set of dual vectors {


 a }, defined by (2.13), have the quadratic form matrix Fab
 b ):
(
a ,
Fab = 2a,

a  b < n 1;

Fan1 = Fan = a,

a < n 1;
n
Fnn = Fn1n1 = ;
2
n2
.
Fn1m =
2
(p)

(5.2)
(p)

The central charge c(W Dn ) of the W Dn model has been calculated in [1]



(2n 2)(2n 1)
(p)
.
c W Dn = n 1
p(p + 1)

(5.3)

Using (2.15) and (5.2) the dimension of primary operators can be easily calculated. In
particular the dimension of the operator (2,1),(1,1),... ,(1,1) , naturally identified with the
energy operator (Section 2), is
1 2n 1

.
(5.4)
2
2
Therefore the disorder induced interaction ( ) is slightly relevant: we can study the
(p)
W Dn model with disorder coupled to the energy operator and the related problem of N coupled systems using the same technique as the one we have exploited in the case of a
(p)
W D3 model. The analogy with this case goes further: we show in Appendix A that the
operator (1,1)(2,1)(1,1)...(1,1) with dimension
(2,1),(1,1),...,(1,1) =

(1,1),(2,1),(1,1),...,(1,1) = 1 2(n 1)

(5.5)

is, apart from the identity, the only relevant operator in the OPE of (x )(x). The enlarged

algebra of the fields and is given by Eq. (3.8), where D and D are now the related
(p)

structure constant of the W Dn

model. We consider thus action (3.3), where in this case

462

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471


(p)

(a)

A0 describes a single W Dn model: the correspondent -functions take form (3.9) with
the values of the structure constants (up to the first order in ) calculated in Appendix A

4(n 1)
2n 1

+ O( ),
D =
D =
(5.6)
+ O( ).
n
n(2n 1)
In the case N = 0, in addition to the unstable trivial fixed point (0, 0), there is one stable
fixed point

g = 0,
(5.7)
= 2 n(2n 1) .
For N  2, there are two other supplementary fixed points, both node points of the RG
flow:




n(n 1)
(1)
= n(2n 1) 1 (N 2)
,
N 1 + n(n 1)N 2

n(n 1)
g(1) = +2(2n 1)
(5.8)
,
N 1 + n(n 1)N 2


n(n 1)
= n(2n 1) 1 + (N 2)
N 1 + n(n 1)N 2

n(n 1)
g(2) = 2(2n 1)
.
N 1 + n(n 1)N 2
(2)


,

(5.9)

In order to compute the central charges at the fixed points for the disorder problem and
for the coupled models, we can directly apply result (4.9), provided the new values of the
structure constants (5.6) and of the dimension of operators (5.4) and (5.5) are taken into
(n)
account. For the quenched problem, the central charge cdis.
at the stable fixed point (5.7) is


c(0, 2 n(2n 1) )
(p)
(n)
= c W Dn 4n(n 1)(2n 1) 3
cdis. = lim
N0
N

(p1)
.
c W Dn
(5.10)
(p1)

The correspondent field theory is described by the W Dn


model, as found already
(n)
(n)
in [1]. In the coupled model, the central charge ccoupl.
at the same fixed point is ccoupl.
=
(p1)

(p1)

) and the associated critical theory is described by N decoupled W Dn


Nc(W Dn
models.
(n)
At the new fixed points (5.8) and (5.9) the central charges c1,2
turn out to be


(p)
(n)
c1,2
= Nc W Dn 2Nn(n 1)(2n 1)



n(n 1)
3.
1 (N 2)
(5.11)
N 1 + n(n 1)N 2
As the RG-flow behavior is the same for all n  3, the results we have shown in Section 4.1
(p)
(p)
for the disordered W D3 model and for the system of N coupled W D3 models are valid
(p)
for the whole series of W Dn conformal theories.

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

Finally we give the direct generalization of Eq. (4.6):





n(n 1)
(1),(2)
s
= + (2n 1) 1 N
,
N 1 + n(n 1)N 2



n(n 1)
(1),(2)
.
= + (2n 1) 1 N
a
N 1 + n(n 1)N 2

463

(5.12)

6. Conclusions
As has been stressed throughout the paper, for every small value of
=

1
,
p+1

p1

(6.1)
(p)

we have a unitary model, W Dn , associated to it. When a set of N  2 of such models


are coupled and brought, by fine tuning of the couplings g and , to one of the two newly
found fixed points, (5.8) or (5.9), the corresponding critical theory should also be described
by a unitary conformal field theory. As we have a fixed point for every p, we should have a
unitary series of new conformal theories, with p being a parameter, accumulating towards
(p)
p = . Moreover, we should have a unitary series for every n, of W Dn , n = 3, 4, 5, . . . .
These new theories, presently unknown, have to incorporate into them the permutational
symmetry SN , N = 2, 3, . . . . This symmetry has to be incorporated into the chiral algebra
of these theories. As the symmetry is discrete, the natural suggestion would be that it should
be represented by parafermionic currents.
For N  3, the group SN is non-Abelian. Looking at the expression of the central charge,
Eq. (5.11), we observe that the correction term which we have calculated is nonrational for
N  3.
If we accept the idea of non-Abelian parafermionic conformal theories, we have to
accept also that these theories possess an infinite series of unitary models, labeled by p,
according to the arguments given above. And then formula (5.11) indicates that the central
charge for these unitary models takes nonrational values. This feature is unusual.
Being more precise, the argument for rational or nonrational values of the central
charge c, and the dimensions of the operators, could be given as follows.
If one assumes that c, which is a function of p, takes rational values for all integer
values of p, this then requires that the function c(p) should have a simple rational form,
like c(p) = Q(p)/P (p), where Q(p), P (p) are two polynomials of p with rational valued
coefficients. If the function c(p) of such a form is developed in a series of 1/p, the
coefficients will all be rational.
Still, it should be admitted that we do not know in fact if the perturbative expansion in
1/p, which we define in this paper, represents a convergent series. The series might
also be asymptotic, as it is often the case in perturbative expansions in field theory. This
then allows for two possibilities.

464

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

The first possibility will be that the series is convergent, representing an analytic function
c(p), like this is two case for the perturbed minimal models in [5].
In this case, according to the argument given above, one could actually judge on
rationalitynonrationality with the coefficients of the expansion.
Saying it again, for having rational values of c(p), for all integer p, a complicated
function c(p) will not be allowed. Then the coefficients of the expansion will also have
to be rational.
The second possibility will be that the series in is not convergent, that it is only
asymptotic. In this case our arguments do not apply and the exact values of the central
charge might well be rational, in spite of the irrational coefficients of the expansion.
In view of this second possibility, our conclusions on nonrationality should not be
viewed as definite.
Construction of the corresponding exact conformal field theories represents a theoretical
challenge.

Acknowledgements
Discussions with D. Bernard, J. Jacobsen, A. LeClair, M. Picco, have been stimulating.

Appendix A
A.1. Relevant operators in the OPE of (x )(x)
In Section 3 we have explained how to deduce the OPE of (x )(x) from the integral
representation of the energy four-point correlation function: within the set of vectors

{2 + 3a=1 qa ea } where q1 = 0, 1, 2, q2 = 0, 1, and q3 = 0, 1, we search for the
ones which decompose into positive integer numbers of { 2a }, basic vectors of the
representation lattice, i.e.,
2 +

3


qa ea =

a=1

3

1 n

a=1

a

(A.1)

with n a  1. In this case the vertex operator V2 + 3 qa ea belongs to the physical
a=1
sector of the Kac table and the correspondent primary operator (n ,1)(n ,1)(n ,1) appears in
1
2
3
the energy algebra. Using Eqs. (2.19) and (2.20), condition (A.1) is equivalent to:
n 1 = 3 2q1 + q3 + q2 ,
n 2 = 1 2q2 + q1 ,
n 3 = 1 2q3 + q1 .

(A.2)

Testing all the possible set of values (q1, q2 , q3 ), its easy to see that system (A.2) admits
only three solutions; they correspond to the identity operator (for q1 = 2, q3 = 1, q2 = 1),
to the operator (for q1 = 1, q3 = 0, q2 = 0) and to the operator (3,1)(1,1)(1,1) (for q1 =

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

465

q2 = q3 = 0). According to Eq. (2.31) (3,1)(1,1)(1,1) = 2: the only relevant operators in the
energy algebra are the identity and .
Now, in order to verify that the algebra formed by these two fields does not contain
any other relevant operator, we have to check the OPE of (x )(x). Following the
same procedure used before, its easy to verify that, apart from identity, the field is
the only relevant operator in the -algebra, as already discussed in [1]. We consider
in this case the integral representation of the four-point correlation function G (x)
(0)(x)(1)():




V (0)V (x)V (1)V ()V1 (1 )V1 (2 )V2 (1 )
G (x) . . .

V2 (2 )V3 (1 )V3 (2 ) ,
(A.3)
where the variables 1 , 2 , 1 , 2 , 1 , 2 are integrated over the 2D plane. The charge
neutrality is satisfied according to 2 + 2 (e1 + e2 + e3 ) = 0. As discussed above all
the operators (n 1 ,1)(n 2 ,1)(n 3 ,1) present in the OPE of (x )(x) are given by the condition
2 +

3


qa ea =

a=1

3

1 n

a=1

a ,

(A.4)

where q1 , q2 , q3 can assume the values 0, 1, 2; Eq. (A.4) is equivalent to:


n 1 = 1 2q1 + q3 + q2 ,
n 2 = 3 2q2 + q1 ,
n 3 = 3 + q1 2q3 .

(A.5)

By Eq. (A.5) we have determined, apart from the identity, four primary operators:
(1,1)(3,1)(3,1), (2,1)(3,1)(1,1), (2,1)(1,1)(3,1) and ; the only field whose dimension is
smaller than the unity is .
A.2. Relevant operators in the OPE of (x )(x) for the W Dn

(p)

model with n  4
(p)

As already discussed in Section 5, the energy operator of the W Dn model


is associated to the primary operator (2,1)(1,1)...(1,1) . We show below that in the
correspondent algebra there is, apart from the identity, only one relevant primary operator,
namely (1,1)(2,1)(1,1)...(1,1) . In the following we define for simplicity (2,1)(1,1)...(1,1)
 , 2
0 (2,1)(1,1)...(1,1)  , (1,1)(2,1)(1,1)...(1,1)  , and 2
0 (1,1)(2,1)(1,1)...(1,1)
 . The integral representation of the correspondent four-point correlation function
G(x) (0)(x)(1)() is



G(x)
V (0)V (x)V (1)V ()

V1 V1 V2 V2 Vn Vn ,
(A.6)
! "

! "
! " 
m1 times

m2 times

mn times

466

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

where the screenings Va with a = 1, . . . , n are integrated over the 2D plane. Using
quadratic form (5.2) its easy to see that the charge neutrality condition
2 +

n


ma ea = 0

(A.7)

imposes ma = 2 for a = 1, . . . , n 2 and mn1 = mn = 1. For the same arguments


discussed in the previous section, we search for the vectors (n ,1)(n ,1)...(n n ,1)) with n a  1
1
2
(see (5.1)) such that
2 +

n


qa ea = (n ,1)(n ,1)...(n n ,1) ,


1

(A.8)

a=1

where qa = 0, 1, 2 for a = 1, . . . , n 2, qn1 = 0, 1 and qn = 0, 1. Using Eq. (5.2),


condition (A.8) is equivalent to the following system of equations:
n 1 = 3 2q1 + q2 ,
n a = 1 2qa + qa1 + qa+1,

2  a < n 2,

n n2 = 1 2qn2 + qn3 + qn1 + qn ,


n n1 = 1 2qn1 + qn2 ,
n n = 1 2qn + qn2 .

(A.9)

Eq. (A.9) gives us all the possible primary operators (n 1 ,1)(n 2 ,1)...(n n ,1) present in the
energy algebra. Once again, testing all the possible set of values (q1 , . . . , qn ) its easy
to verify that, apart from the identity, there are only two solutions: they correspond to the
operator (3,1)(1,1)...(1,1) (for qa = 0), with dimension greater than unity, and to the operator
(1,1)(2,1)(1,1)...(1,1) (for q1 = 1 and qa = 0, a  2), whose dimension is given by
Eq. (5.5). In the operator product decomposition (x )(x), we have checked that the only
relevant field is itself: by studying the correspondent four-point correlation function we
obtain the following system:
n 1 = 1 2q1 + q2 ,
n 2 = 3 2q2 + q1 + q3 ,
n i = 1 2qi + qi1 + qi+1 ,

3  i  n 3,

n n2 = 1 2qn2 + qn3 + qn1


n n1 = 1 2qn1 + qn2 ,
n n = 1 2qn + qn2

+ qn ,

(A.10)

with the integers q1 , qn1 , qn = 0, 1, 2 and qa = 0, 1, 2, 3, 4 for a = 2, . . . , n 2.


According to Eq. (A.10), the field is the only operator with dimension smaller than
unity which appears in the -algebra.

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

467

(p)

Appendix B. Computation of the structure constants D and D for the W Dn


models

The structure constants D and D are determined by two three-point correlation


(p)

functions of the unperturbed W Dn model







D
= (0)(1)() ,
D = (0)(1)() ,

(B.1)
(p)

where corresponds to the primary operator (1,1)(2,1)(2,1) for the W D3 model and to the
(p)
primary operator (1,1)(2,1)(1,1)(1,1) for the W Dn model with n  4. In order to compute
their values using the Coulomb-gas representation, we have to take into account that the
vertex operators V , V , V and V can acquire nontrivial normalization factors N ,

N , N and N [8] such that


(x) = N1 V (x) = N1 V (x),
(x) = N1 V (x) = N1
V (x).

(B.2)

After imposing the fields normalizations (1)(0) = N N = 1 and (1)(0) =


N N = 1, we need to compute four integrals:
I1

(1) 
n

()mi

m(1)
i !

,
= N D

I2

n


()

(2)
mi

...

V (0)V (1)V () V1 V1 Vn Vn
 ! "  ! "
m(1)
1 times

m(1)
n times

(B.3)



...

(2)

mi !

V (0)V (1)V () V1 V1 Vn Vn
 ! "  ! "

(2)

(2)

m1 times

mn times

= N2 N1 D
,
(3) 

n



()mi
I3
. . . V (0)V (1)V () V1 V1 Vn Vn
(3)

!
"

!
"

mi !
i
(3)

(B.4)

(3)

m1 times

mn times

= N D ,
I4

n

()
i

m(4)
i

m(4)
i !

(B.5)



...

V (0)V (1)V2 () V1 V1 Vn Vn
 ! "  ! "
0
(4)

m1 times

= N2 ,

(4)

mn times

(B.6)

where the integration over the 2D plane of the screenings, whose number is fixed by the
charge neutrality condition, is intended. In Eq. (B.6) the vertex V2 corresponds to the
0
identity operator I and the relative normalization constant N2 = NI = 1. According to
0

468

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

the general form of the correlation function of N vertex operator Vi




 
 
V1 (1 ) VN (N ) =
|i j |4i j ,

(B.7)

i<j

and using the formula



C(, ) = d 2 | |2 | 1|2
=
and

=(1 + )=(1 + )=(1 )


=()=()=(2 + + )

(B.8)


K(, ) =

d 2 d 2 | |2 | 1|2 | |2 | 1|2 | |4
=(2)=(1 )  = 2 (1 + + i)=(1 2 (1 + i))
,
=()=(1 2)
= 2 ( i)=(2 + 2 + (1 + i))
1

= 2 2

(B.9)

i=0

where =(x) is the Gamma-function, integrals (B.3)(B.6) have been determined. Eq. (B.8)
and (B.9) represent special cases of a family of integrals calculated in [8].
In terms of these integrals, the structure constants are:


I2
I1 I2

D = I3
(B.10)
,
D =
.
I1 I4
I4
B.1. Computation of I1
The charge neutrality condition for I1 is
 +

n


m(1)
a = 0.
a e

(B.11)

a=1
(1)
(1)
(1)
It is satisfied for m(1)
1 = mn1 = mn = 1, ma = 2, a = 2, . . . , n 2. Using Eq. (B.7) and
taking into account all the scalar products, easily computed from the quadratic form (5.2),
the integral I1 takes the form


I1 = 23n


d 2 1

...

n2
2


d 2 a(k) d 2 n1 d 2 n

a=2 k=1

|1 |2
2

2


2
 
# 2 n3
# (k)
# 2#
# (k)
# 2
# 1 #2 # (k) 1#2
# (l) #2
2

n2


k=1

2
# (1)
# 2 
# (k)
#2 2 # (k)
#2 2
# (2) #4
#
# #
# ,
a
a
n2 n1
n2 n

a=2

a+1

a=2 k,l=1

k=1

(B.12)

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

469

2 = p/(p + 1) 1 . Integrating over the variables , and


where
n
n1 we obtain by
Eq. (B.8)


n2
2

 2

3n 2
2
2
I1 = 2 C ,
. . . d 1
d 2 a(k)
a=1 k=1
2
2
 
# 2#
# (k)
# 2
# 2 n3
2  # (k)
# 1 #2 # (k) 1#2
# (l) #2
|1 |2
a
2
2
a+1
k=1

n3


a=2 k,l=1

# (1)
# 2#
#
2
# (2) #4 # (1) (2) #44 .
a
a
n2
n2

(B.13)

a=2
(k)
Then using Eq. (B.9) we integrate in order over the couple of variables n2
, . . . , 2(k) and
over the variable 1 ; the result is

 
 2
 2

n3

2
2
2
2
C , 2
.
,
+ 2n
K
, i 1
I1 = 23n C 2

(B.14)

i=1

B.2. Computation of I2
According to
2 +  + e1 = 0,
m(2)
1

(B.15)

m(2)
a

= 1 and
= 0 for a = 2, . . . , n; the integral (B.4) reads



I2 = d 2 1 V (0)V (1)V ()V1 (1 )


2
2
= d 2 1 | |2 |1 1|2
 2

2
= C
,
.

we have

(B.16)

B.3. Computation of I3
Integral (B.5) must satisfy the charge neutrality condition (B.11); it takes the form


2
n2

3n
d 2 a(k) d 2 n1 d 2 n
. . . d 2 1
I3 = 2
a=2 k=1

2


2
 
# 2#
# (k)
# 2#
# 2 n3
# (k)
# 2
# 1 #2 # (k) 1#2 # (k) #2
# (l) #2
a
2
2
2
a+1

k=1

n2


a=2 k,l=1

# (1)
# 2
# (2) #4
a

a=2

2


# (k)
#2 2 # (k)
#2 2
#
# #
# .
n2 n1
n2 n

(B.17)

k=1

Following the same order of integrations used in the computation of integral (B.3), with the
difference that we integrate first over the variable 1 and then over the couple of variables

470

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471

2(1) , 2(2) , we obtain



 2

2
2 2n 5
2
I3 = 23n C 3
K
,
,
(n 2)
2

n4



 2 
2
.
K
, i 1

(B.18)

i=1

B.4. Computation of I4
According to the charge neutrality condition
2 +

n


m(4)
a = 0,
a e

(B.19)

a=1
(4)

(4)

(4)

we have ma = 2 for a = 1, . . . , n 2, mn1 = mn = 1. Integral (B.6) has the form



I4 = 22n

...

 n2
2


d 2 a(k) d 2 n1 d 2 n

a=1 k=1

2


2
 
# 2#
# (k)
# 2 n3
# (k)
# 2
# 1#2 # (k) #2
# (l) #2
1

k=1

n2


a+1

a=1 k,l=1
2
# (1)
# 2 
# (k)
#2 2 # (k)
#2 2
# (2) #4
#
# #
# .
a
a
n2 n1
n2 n

a=1

(B.20)

k=1
(k)

(k)

Integrating over the variables n , n1 , n2 , . . . , 1 , integral (B.6) assumes the value


 

 2
n2

2
2
2
,
K
, i 1
.
I4 = C 2

(B.21)

i=1

Developing in the integrals we have computed and using Eq. (B.10), the Eq. (5.6) are
obtained.

Appendix C. RG equations and renormalization of energy operators


Using the OPE (3.8) and the dimensions of the energy and fields (Eqs. (5.4) and (5.5)),
the operator algebra of the perturbing terms is
N

a=b

(a b )(x)

N

c=d

(c d )(y) 4(N 2)|x y|

2+2(2n1)

N

(a b )(y)
a=b

+ 4(N 1)D
|x y|2+4n

N

a=1

a (y) + ,

V.S. Dotsenko et al. / Nuclear Physics B 613 [FS] (2001) 445471


N


(a b )(x)

N


a=b
N


N

(a b )(y) + ,

a (y) 2D
|x y|2+2(n1)

a=b

a=1

a (x)

a=1

N


471

a (y) D |x y|2+2(n1)

b=1

N


a (y) + ,

(C.1)

a=1

where we have omitted the irrelevant terms. By Eq. (C.1) the 1-loop RG-equations can be
easily obtained (see for example [3] or [10]):

g,
g = 2(2n 1) g 4(N 2)g 2 4D

2
= 2(n 1) D 2 2(N 1)D
g .

(C.2)

With the redefinitions g g/(4) and /(2), we find Eq. (3.9).


Similarly the renormalized energy operators c (x) (3.11) can be computed at the first
order [3] from the following operator product decomposition
N


(a b )(x)c (y) 2|x y|

a=b
N


2+2(2n1)

N


a (y) + ,

a=c

a (x)c (y) D
|x y|2+2(n1) c (y) + .

(C.3)

References
[1] S.L. Lukyanov, V.A. Fateev, Sov. Sci. Rev. A. Phys. 15 (1990) 1117.
[2] A.W.W. Ludwig, Nucl. Phys. B 285 (1987) 97;
A.W.W. Ludwig, Nucl. Phys. B 330 (1990) 639.
[3] Vl.S. Dotsenko, M. Picco, P. Pujol, Phys. Lett. B 347 (1995) 113, hep-th/9405003;
Vl.S. Dotsenko, M. Picco, P. Pujol, Nucl. Phys. B 455 (1995) 701, hep-th/9501017.
[4] Vik.S. Dotsenko, Vl.S. Dotsenko, Sov. Phys. JETP Lett. 33 (1981) 37;
Vik.S. Dotsenko, Vl.S. Dotsenko, J. Phys. A 17 (1984) L301.
[5] A.B. Zamolodchikov, Pisma Zh. Eksp. Teor. Fiz. 43 (1986) 565, English translation: JETP
Lett. 43 (1986) 730;
A.B. Zamolodchikov, Sov. J. Nucl. Phys. 46 (1987) 1090.
[6] A.W. Ludwig, J.L. Cardy, Nucl. Phys. B 285 (1987) 687.
[7] Vl.S. Dotsenko, J.L. Jacobsen, M.A. Lewis, M. Picco, Nucl. Phys. B 546 (1999) 505, condmat/9812227.
[8] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 324 (1984) 312;
Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691;
Vl.S. Dotsenko, V.A. Fateev, Phys. Lett. B 154 (1985) 291.
[9] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
[10] D. Bernard, Lectures given at 1995 Cargese Summer School, hep-th/9509137.
[11] A. LeClair, A. Ludwig, G. Mussardo, Nucl. Phys. B 512 (1998) 523542, hep-th/9707159.
[12] I. Vaysburd, Nucl. Phys. B 446 (1995) 387.

Nuclear Physics B 613 [FS] (2001) 472496


www.elsevier.com/locate/npe

New spin CalogeroSutherland models related to


BN -type Dunkl operators
F. Finkel, D. Gmez-Ullate, A. Gonzlez-Lpez, M.A. Rodrguez,
R. Zhdanov 1
Departamento de Fsica Terica II, Universidad Complutense, 28040 Madrid, Spain
Received 23 March 2001; accepted 27 July 2001

Abstract
We construct several new families of exactly and quasi-exactly solvable BCN -type Calogero
Sutherland models with internal degrees of freedom. Our approach is based on the introduction
of a new family of Dunkl operators of BN type which, together with the original BN -type Dunkl
operators, are shown to preserve certain polynomial subspaces of finite dimension. We prove that a
wide class of quadratic combinations involving these three sets of Dunkl operators always yields a
spin CalogeroSutherland model, which is (quasi-)exactly solvable by construction. We show that all
the spin CalogeroSutherland models obtainable within this framework can be expressed in a unified
way in terms of a Weierstrass function with suitable half-periods. This provides a natural spin
counterpart of the well-known general formula for a scalar completely integrable potential of BCN
type due to Olshanetsky and Perelomov. As an illustration of our method, we exactly compute several
energy levels and their corresponding wavefunctions of an elliptic quasi-exactly solvable potential
for two and three particles of spin 1/2. 2001 Elsevier Science B.V. All rights reserved.
PACS: 03.65.Fd; 75.10.Jm; 03.65.Ge
Keywords: Spin CalogeroSutherland model; Dunkl operator; Quasi-exact solvability; Elliptic potential

1. Introduction
The completely integrable and exactly solvable models of Calogero [1] and Sutherland [2] describe a system of N quantum particles in one dimension with long-range pairwise interaction. These models and their subsequent generalizations (see [3] and references
therein for a comprehensive review) have been extensively applied in many different fields
of physical interest, such as fractional statistics and anyons [46], quantum Hall liquids [7],
YangMills theories [8,9], and propagation of soliton waves [10]. A significant effort has
E-mail address: artemio@eucmos.sim.ucm.es (A. Gonzlez-Lpez).
1 On leave of absence from Institute of Mathematics, 3 Tereshchenkivska St., 01601 Kyiv-4, Ukraine.

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 1 ) 0 0 3 7 8 - 9

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

473

been devoted over the last decade to the extension of scalar CalogeroSutherland models
to systems of particles with internal degrees of freedom or spin [1120]. These models
have attracted considerable interest due to their connection with integrable spin chains of
HaldaneShastry type [21,22] through the freezing trick of Polychronakos [23].
The exactly solvable and integrable spin models introduced in [11,14] generalize the
original rational (Calogero) and trigonometric (Sutherland) scalar models, and are invariant
with respect to the Weyl group of type AN . The exact solvability of both models can be
established by relating the Hamiltonian to a quadratic combination of either the Dunkl [24]
or the DunklCherednik [25] operators of AN type, whose relevance in this context was
first pointed out by Polychronakos [26]. We shall use the term Dunkl operators to
collectively refer to this type of operators. Up to the best of our knowledge, only two
BN -invariant spin CalogeroSutherland models have been proposed so far in the literature,
namely the rational and the trigonometric spin models constructed by Yamamoto in [16].
The exact solvability of the rational Yamamoto model was later proved in Ref. [19] using
the Dunkl operator formalism. The exact solvability of the trigonometric Yamamoto model
will be proved in this paper.
In a recent paper [20] the authors proposed a new systematic method for constructing
spin CalogeroSutherland models of type AN . One of the key ingredients of the method
was the introduction of a new family of Dunkl-type operators which, together with
the Dunkl operators defined in [24,25], preserve a certain polynomial module of finite
dimension. It was shown that a wide class of quadratic combinations of all three types
of Dunkl operators always yields a spin CalogeroSutherland model. In this way all
the previously known exactly solvable spin CalogeroSutherland models of AN type are
recovered and, what is more important, several new exactly and quasi-exactly solvable spin
models are obtained. By quasi-exactly solvable (QES) we mean here that the Hamiltonian
preserves a known finite-dimensional subspace of smooth functions, so that a finite
subset of the spectrum can be computed algebraically; see [2729] for further details.
If the Hamiltonian leaves invariant an infinite increasing sequence of finite-dimensional
subspaces, we shall say that the model is exactly solvable (ES).
In this paper we extend the method of Ref. [20] to construct new families of (Q)ES spin
CalogeroSutherland models of BCN type. To this end, we define in Section 2 a new set
of Dunkl operators of BN type leaving invariant a certain polynomial subspace of finite
dimension, which is also preserved by the original Dunkl operators of BN type introduced
in [19]. In Section 3, we show that a suitable quadratic combination of all three types
of Dunkl operators discussed in Section 2 can be mapped into a multi-parameter (Q)ES
physical Hamiltonian with spin. This approach is a generalization of the construction used
to prove the integrability of the AN spin CalogeroSutherland models, in which only a
single set of Dunkl operators is involved. Our method is also related to the so-called hidden
symmetry algebra approach to scalar N -body QES models [3032], where the Hamiltonian
is expressed as a quadratic combination of the generators of a realization of sl(N + 1). We
then show that the sets of Dunkl operators used in our construction are invariant under
inversions and scale transformations. This property is exploited in Section 4 to perform
a complete classification of the BCN -type (Q)ES spin CalogeroSutherland models that

474

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

can be constructed with the method described in this paper. The resulting potentials can be
divided into nine inequivalent classes, out of which only two (the rational and trigonometric
Yamamoto models) were previously known. In particular, we obtain four new families of
elliptic QES spin CalogeroSutherland models of BCN type. Section 5 is devoted to the
discussion of the general structure of the potentials listed in Section 4. We prove that all
the potentials in the classification are expressible in a unified way in terms of a Weierstrass
function with suitable (sometimes infinite) half-periods. This provides a natural spin
counterpart of Olshanetsky and Perelomovs formula for a general scalar potential related
to the BCN root system. Finally, in Section 6 we illustrate the method by exactly computing
several energy levels and their corresponding eigenstates for an elliptic spin 1/2 potential
in the two- and three-particle cases.

2. BN -type Dunkl operators


In this section we introduce a new family of BN -type Dunkl operators which will play a
central role in our construction of new (Q)ES spin CalogeroSutherland models.
Let f (z) be an arbitrary function of z = (z1 , . . . , zN ) RN . Consider the permutation
operators Kij = Kj i and the sign reversing operators Ki , whose action on the function f
is given by
(Kij f )(z1 , . . . , zi , . . . , zj , . . . , zN ) = f (z1 , . . . , zj , . . . , zi , . . . , zN ),
(Ki f )(z1 , . . . , zi , . . . , zN ) = f (z1 , . . . , zi , . . . , zN ),

(1)

where i, j = 1, . . . , N . It follows that Kij and Ki verify the relations


Kij2 = 1,

Kij Kj k = Kik Kij = Kj k Kik ,

Ki2 = 1,

Ki Kj = Kj Ki ,

Kij Kkl = Kkl Kij ,

Kij Kk = Kk Kij ,

Kij Kj = Ki Kij ,

(2)

where the indices i, j, k, l take distinct values in the range 1, . . . , N . The operators Kij , Ki
span the Weyl group of type BN , also called the hyperoctahedral group. We shall also
ij = Ki Kj Kij . Let us consider the following set of Dunkl
employ the customary notation K
operators:


 1
 1

ij ) + b (1 Ki ), (3)
Ji =
+a
(1 Kij ) +
(1 K
zi
zi zj
zi + zj
zi
j =i
j =i


 zi zj

m a  zi + zj
0
ij ) ,
Ji = zi
(4)
+
(1 Kij ) +
(1 K
zi
2
2
zi zj
zi + zj
j =i
j =i


 zi zj
 zi zj

ij )
Ji+ = zi2
m zi + a
(1 Kij )
(1 K
zi
zi zj
zi + zj
j =i
j =i


b zi 1 (1)m Ki ,
(5)
where a, b, b are nonzero real parameters, m is a nonnegative integer, and i = 1, . . . , N . In

Eqs. (3)(5), the symbol j =i denotes summation in j with j = 1, . . . , i 1, i + 1, . . . , N .

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

475

In general, any summation or product index without an explicit range will be understood
in this paper to run from 1 to N , unless otherwise constrained. It shall also be clear in each
case whether a sum symbol with more than one index present denotes single or multiple
summation.
The operators Ji in Eq. (3) have been used by Dunkl [19] to construct a complete set
of eigenvectors for Yamamotos BN rational spin model [16]. The operators Ji0 were also
introduced by Dunkl in Ref. [19]. To the best of our knowledge, the operators Ji+ have not
been considered previously in the literature.
The operators (3)(5) obey the commutation relations

 0 0 a 2 
ij )(Kj k + K
j k Kik K
ik ), (6)
Ji , Jj = 0,
(Kij + K
Ji , Jj =
4
k=i,j







Kij Ji = Jj Kij ,
Ki Ji = (1) Ji Ki ,
Kij , Jk = 0,
Ki , Jj = 0,
(7)
where  = , 0, and the indices i, j, k take distinct values in the range 1, . . . , N . The
operators Ji (respectively, Ji+ ), i = 1, . . . , N , together with Kij and Ki , i, j = 1, . . . , N ,
span a degenerate affine Hecke algebra, see [25]. The operators Ji0 do not commute, but


ij ) a j >i (Kij + K
ij ) do, it can be shown that the latter
since Ji0 + a2 j <i (Kij + K
2
operators, together with Kij and Ki , also define a degenerate affine Hecke algebra.
It is well-known [19] that the operators Ji and Ji0 preserve the space Pn of polynomials
in z1 , . . . , zN of degree at most n, for all n N. Moreover, for any nonnegative integer n,

l
the space Rn spanned by the monomials i zii with 0  li  n is also invariant under the
action of both Ji and Ji0 . Let us prove this assertion in the case of Ji0 . Since (zi z i
m
2 )Rn Rn , it suffices to show that
l
l
zi zj
zi + zj
ij )
(1 Kij )
zkk Rn , and
(1 K
zkk Rn ,
(8)
zi zj
zi + zj
k

for any pair of indices 1  i = j  N . For the first inclusion we note that
l
zi + zj
(1 Kij )
zkk
zi zj
k


|l l |
|l l |
l
zi i j zj i j
min(li ,lj )
k
=
zk (zi + zj )(zi zj )
sign(li lj )
zi zj

=

k=i,j


k=i,j

where


l
zkk

|li lj |1

(zi + zj )(zi zj )

min(li ,lj )

sign(li lj )

|l lj |1k k
zj ,

zi i

(9)

k=0

1, p < 0,
p = 0,
sign(p) = 0,

1,
p > 0.
The resulting polynomial thus belongs to Rn . Indeed, it is a linear combination of


monomials k zlk with lk = lk for k = i, j , and li , lj  max(li , lj ). Likewise, the second
k

476

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

inclusion in (8) follows from the identity


l
zi zj
ij )
(1 K
zkk
zi + zj
k


|l l |
|l l |
l
zj i j (1)li +lj zi i j
=
zkk (zi zj )(zi zj )min(li ,lj ) s(li , lj )
zi + zj

=

k=i,j


l
zkk

|li lj |1

(zi zj )(zi zj )min(li ,lj ) s(li , lj )

k=i,j

(zi )|li lj |1k zjk ,

(10)

k=0

where

1,
s(p, q) = 0,

(1)p+q ,

p < q,
p = q,
p > q.

We omit the analogous proof for the operators Ji . Unlike the previous types of Dunkl
operators, the operators Ji+ in Eq. (5) do not preserve the polynomial spaces Pn and Rk
with k = m. However, the space Rm is invariant under the action of Ji+ . In fact, the operator
zi2




m zi b zi 1 (1)m Ki ,
zi

preserves Rm , and, just as we did in the case of Ji0 , one can show that both
zi zj
(1 Kij )
zi zj

and

zi zj
ij )
(1 K
zi + zj

preserve Rn for any nonnegative integer n.

3. BCN -type spin many-body Hamiltonians


In Section 2 we have shown that all three sets of BN -type Dunkl operators (3)(5)
preserve the finite-dimensional polynomial space Rm . In this section we shall use this
fundamental property to construct several families of (Q)ES many-body Hamiltonians with
internal degrees of freedom.
Let S = span{|s1 , . . . , sN | si = M, M + 1, . . . , M; M 12 N} be the Hilbert space
of the particles internal degrees of freedom or spin. We shall denote by Sij and Si , i, j =
1, . . . , N , the spin permutation and spin reversing operators, respectively, whose action on
a spin state |s1 , . . . , sN is defined by
Sij |s1 , . . . , si , . . . , sj , . . . , sN = |s1 , . . . , sj , . . . , si , . . . , sN ,
Si |s1 , . . . , si , . . . , sN = |s1 , . . . , si , . . . , sN .

(11)

The operators Sij and Si are represented in S by (2M + 1)N -dimensional Hermitian
matrices, and obey identities analogous to (2). The notation Sij = Si Sj Sij shall also be
used in what follows.

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

477

We shall deal in this paper with a system of N identical fermions, so that the physical
states are completely antisymmetric under permutations of the particles. A physical state
must therefore satisfy 0 = , where 0 is the antisymmetrisation operator defined
by the relations 20 = 0 and ij 0 = 0 , j > i = 1, . . . , N , with ij = Kij Sij . Since
Kij2 = 1, the above relations are equivalent to Kij 0 = Sij 0 , j > i = 1, . . . , N . For the
lowest values of N , the antisymmetriser 0 is given by
1
N = 2: 0 = (1 12 ),
2
1
N = 3: 0 = (1 12 13 23 + 12 13 + 12 23 ).
6
Our aim is to construct new (Q)ES Hamiltonians symmetric under the Weyl group of
type BN generated by the permutation operators ij and the sign reversing operators Ki Si .
The corresponding algebraic eigenfunctions will be antisymmetric under a change of sign
of both the spatial and spin variables of any particle, and therefore satisfy = , where
is the projection on states antisymmetric under permutations and sign reversals. The
total antisymmetriser is determined by the relations 2 = and
Kij = Sij ,

Ki = Si ,

j > i = 1, . . . , N.

(12)

It may be easily shown that




1
= N
(1 Ki Si ) 0 .
2
i

Following closely the procedure outlined in [20], we shall consider a quadratic


combination of the Dunkl operators (3)(5) of the form


 2
 2
 2
c++ Ji+ + c00 Ji0 + c Ji + c0 Ji0 ,
H =
(13)
i
2
2 + c2
where c++ , c00 , c , c0 are arbitrary real constants such that c++
+ c00
= 0.
The second-order differential-difference operator (13) possesses the following remarkable
properties. First, it is a quasi-exactly solvable operator, since it leaves invariant the
polynomial space Rm . In particular, if c++ = 0 the operator H preserves Rn (and Pn )
for any nonnegative integer n, and is therefore exactly solvable. Secondly, H commutes
with Kij , Ki , Sij , and Si for all i, j = 1, . . . , N . This follows immediately from the
commutation relations (7). Note that none of the terms





Ji , Ji0 ,
Ji , Ji0 ,
Ji
i

commute with Ki , and for that reason they have not been included in the definition of H .


We have also discarded the term i {Ji+ , Ji } because it differs from 2 i [(Ji0 )2 + (b
b)Ji0] by a constant operator.
Since H preserves the polynomial module Rm , commutes with , and acts trivially on
S, the module
m = (Rm S)
R

(14)

478

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

is also invariant under H . It follows from Eqs. (12) that the action of the operators Kij and
m coincides with that of the spin operators Sij and Si , respectively.
Ki on the module R
 obtained from H by the formal substitutions Kij
Therefore, the differential operator H
m . For the same reason, if
Sij , Ki Si , i, j = 1, . . . , N , also preserves the module R



the coefficient
 c++ in Eq. (13) vanishes, the operator H leaves the modules Rn and Pn =
Pn S invariant for any nonnegative integer n. Using the formulae (3)(5) in the
appendix for the squares of the Dunkl operators, we get the following explicit expression
:
for the gauge spin Hamiltonian H

 zi P (zi )

=
H
zi
P (zi )z2i + Q(zi )zi + R(zi ) + 4a
z2 zj2
i
i=j i

 b c



2
m

(1 + Si ) + b c++ zi 1 + (1) Si
zi2
i



1 + Sij
1 + Sij
a
P (zi )
+
(zi zj )2 (zi + zj )2
i=j

ac++ 

+
(15)
(zi + zj )2 (1 + Sij ) + (zi zj )2 (1 + Sij ) + C,
2
i=j

where
P (z) = c++ z4 + c00 z2 + c ,


Q(z) = 2c++ 1 m b + 2a(1 N) z3



2bc
,
+ c0 + c00 1 m + 2a(1 N) z +
z
R(z) = c++ m(m 1 + 2b )z2 ,



2
2 

Nm
a
 = c00
+
[4 (Sij + Sij )(Sik + Sik )] + 6
(1 Si Sj )
C
4
12
i,j,k
i=j

Nmc0
a
.
(2 + Sij + Sij )
+
2
2

(16)

i=j


Hereafter, the symbol i,j,k denotes summation in i, j, k with i = j = k = i.
One of the main ingredients of our method is the fact that the gauge spin Hamiltonian

H can be reduced to a physical spin Hamiltonian

H =
(17)
x2i + V (x),
i

where V (x) is a Hermitian matrix-valued function, by a suitable change of variables z =


(x), x = (x1 , . . . , xN ) and a gauge transformation with a scalar function (x), namely
|z= (x) 1 = H.
H

(18)

We emphasize that in general there is no (matrix or scalar) gauge factor and change of
coordinates reducing a given matrix second-order differential operator in N variables to a

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

479

physical Hamiltonian of the form (17); see [33,34] and references therein for more details.
The quadratic combination H has precisely been chosen so that such a gauge factor
. For instance, we have omitted the
and change of variables can be easily found for H
 +
otherwise valid term i [Ji , Ji ] because it involves first-order derivatives with matrixvalued coefficients, which are usually very difficult to gauge away. The gauge factor and
change of variables z = (x) in Eq. (18) are, respectively, given by


zi

  Q(yi )
a
zi2 zj2
P (zi )1/4 ,
= exp
(19)
dyi
2P (yi )
i

i<j

and
xi =

zi
(zi ) =

dy
,

P (y)

i = 1, . . . , N.

(20)

The physical spin potential V reads





a + Sij
a + Sij
V =a
P (zi )
+
(zi zj )2 (zi + zj )2
i=j



c++ 
(zi + zj )2 (1 + Sij ) + (zi zj )2 (1 + Sij )
2



 b c
(1
+
S
)
+
W
(z
)
+ C,
+
b c++ zi2 1 + (1)m Si +
i
i
zi2
i

(21)

where






1
1
P
3P 
P 

W (z) =
Q
+
Q
Q
+ c z2 ,
2
2
4P
2
2

and

(22)




c00 

a(2N 1) + 3(m 1) C,
C = aN(N 1) c0
3


c = c++ 2a 2(N 1)(2N 1) + 4a(N 1)(b + m 1) + m(2b + m 1) .

(23)
Note that the change of variables (20), and hence the potential V (x), are defined up
to an arbitrary translation in each coordinate xi , i = 1, . . . , N . The hermiticity of the
potential (21) is a consequence of the Hermitian character of the spin operators Sij and Si .
m under the gauge spin Hamiltonian H
 and Eq. (18)
The invariance of the module R
imply that the finite-dimensional module
 

Mm = (x) Rm S z= (x) .
(24)
is invariant under the physical spin Hamiltonian H . Therefore, any quadratic combination
H of the form (13) leads to a (quasi-)exactly solvable spin many-body potential (21)(23)
(provided of course that the module Mm is not trivial). In particular, if the coefficient c++
vanishes, the spin Hamiltonian H with potential (21) is exactly solvable, since it leaves

480

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

invariant the infinite chains of finite-dimensional modules Mn and Nn = (x)[(Pn


S)]z= (x) , n N.
Our goal is to obtain a complete classification of the (Q)ES spin potentials of the
form (21)(23). The key observation used to perform this classification is the fact that
 may yield the same physical potential. This follows
different gauge spin Hamiltonians H
from the form invariance of the linear spaces span{Ji (z), Ji0 (z), Ji+ (z)}, i = 1, . . . , N ,
under projective (gauge) and scale transformations, given respectively by






1
m


m

(w) =
Ji (z)  J
zj
zj ,
Ji (z)
zj  wj = ,
i
zj
j

j = 1, . . . , N,  = , 0,

(25)

and
zj  wj = zj ,

 (w) = J  (z),
Ji (z)  J
i
i

where = 0 is real or purely imaginary. Indeed, we get




+


J
Ji0 (w) = Ji0 (w),
i (w) = Ji (w) b b ,

j = 1, . . . , N,  = , 0,

(26)


+


J
i (w) = Ji (w) bb ,

 (w) =  J  (w) for the scale transformafor the projective transformations (25), and J
i
i
 is still of the form (13),
tions (26). This implies that the resulting quadratic combination H
with (in general) different coefficients c++ , c00 , c , and c0 . Using these transformations,
we can reduce the polynomial P (z) in (15) to one of the following seven canonical forms:
1.

1,

2.

z2 ,


1 + z2 ,

2
1 z2 ,



e2i z2 e2i z2 ,



1 z2 1 k 2 z2 ,



1 z2 1 k 2 + k 2 z2 ,

3.
4.
5.
6.
7.

(27)

where > 0, 0 < k < 1, and 0 <  /4.


4. Classification of QES spin CalogeroSutherland models
We present in this section the complete classification of all the (Q)ES spin Calogero
Sutherland models that can be constructed applying the procedure described in the previous
sections. To further simplify the classification, we note that the scaling (c , c0 ) 
(c , c0 ) induces the mapping

V (x; c , c0 )  V (x; c , c0 ) = V ( x; c , c0 )


(28)
of the corresponding potentials. For this reason, in Cases 27 we shall only list the potential
for a suitably chosen value of the parameter . Note furthermore that in Cases 24 and 6

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

481

once the potential has been computed for a positive value 0 of the parameter , its
counterpart for the opposite value = 0 can be immediately obtained using (28), namely
V (x; 0, c0 ) = V (ix; 0 , c0 ).
For the models constructed to be symmetric under the Weyl group of type BN spanned
by the operators Kij Sij and Ki Si , 1  i < j  N , the change of variables z = (x) should
be an odd function of x, since only in this case x  x corresponds to z  z. In all
cases except the second one, this has essentially the effect of fixing the arbitrary constants
on which the change of variables (20) depends. For example, in Case 7 with = 4 the
change of variables is of the form zi = cn(2xi + i | k) cn(2xi + i ), 1  i  N .
Imposing that zi be an odd function of xi for all i and using the identity
cn(2xi + i ) + cn(2xi + i ) =

2 cn i cn(2xi )
,
1 k 2 sn2 i sn2 (2xi )

we obtain the condition


i = (2li 1)K,

li Z, 1  i  N,

where K K(k) is the complete elliptic integral of the first kind


/2


K(k) =
0

d
1 k 2 sin2

Since cn(2xi K + 2li K) = (1)li cn(2xi K), symmetry under exchange of the particles
requires that li be independent of i, so that zi = cn(2xi K) for all i = 1, . . . , N . Taking
into account that both the potential V and the gauge function are even functions of z
by Eqs. (19)(22), we see that the change of variables in this case can be taken as zi =
cn(2xi K), 1  i  N .
In the classification that follows, we have routinely discarded constant operators of the
form



Si + 2
Si Sj + 3
(Sij + Sij )
V0 = 0 + 1
i

+ 4



i<j

(Sij + Sij )(Sik + Sik ),

i<j

i R.

(29)

i,j,k

This is justified, since the operator V0 commutes with (it actually commutes with Kij ,
Sij , Ki , and Si for 1  i < j  N ) and therefore preserves the spaces Mn and Nn for all n.
All the potentials in the classification presented below are singular on the hyperplanes
xi = xj , 1  i < j  N , where they diverge as (xi xj )2 . In some cases there may be
other singular hyperplanes, near which the potential behaves as the inverse squared distance
to the hyperplane. We shall accordingly choose as domain of the functions in the Hilbert
space of the system a maximal open subset X of the open set
xN < xN1 < < x2 < x1

(30)

482

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

containing no singularities of the potential. In all cases except Case 2b, we shall take as
boundary conditions defining the eigenfunctions of H their square integrability on the
region X and their vanishing on the boundary X of X faster than the square root of the
distance to the boundary. Since the algebraic eigenfunctions that we shall construct are in
all cases regular inside X, when this set is bounded the square integrability of the algebraic
eigenfunctions on X is an automatic consequence of their vanishing on X. In Case 2b,
the potential is regular and periodic in each coordinate in an unbounded domain. Therefore
the square integrability of the eigenfunctions should be replaced by a Bloch-type boundary
condition in this case.
For each of the potentials in the classification, we shall list the domain chosen for its
eigenfunctions and the restrictions imposed by the boundary conditions discussed above on
the parameters on which the potential depends. In particular, the singularity of the potential
at xi = xj , 1  i < j  N , forces the parameter a to be greater than 1/2. Similarly, in all
cases except for the second one the potential is also singular on the hyperplanes xi = 0, 1 
1
i  N , and the vanishing of the algebraic eigenfunctions on these hyperplanes as |xi | 2 +
with > 0 requires that b > 1/2. The conditions
1
1
b>
a> ,
2
2
shall therefore be understood to hold in all cases. For similar reasons, in Cases 4b, 5, and 6b
we must also have
1
b > .
2
The potential in each case will be expressed as

V (x) = Vspin(x) +
U (xi ),
i

where the last term, which does not contain the spin operators Sij and Si , can be viewed as
the contribution of a scalar external field.
We shall use in the rest of this section the convenient abbreviations
1
= a(N 1) + (b + b + m).
xij = xi xj ,
2
Case 1 P (z) = 1. Change of variables: z = x. Gauge factor:

a b 1 x 2
(x) =
xij xij+
xi e 2 i .
i<j

(31)

Scalar external potential:


U (x) = 2 x 2 .
Spin potential:
Vspin(x) = 2a



(32)
xij

2


 2

(a + Sij ) + xij+ (a + Sij ) + b
xi2 (b + Si ).

i<j

Parameters: = 12 c0 > 0. Domain: 0 < xN < < x1 .

(33)

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

483

Case 2a P (z) = 4z2 . Change of variables: z = e2x . The most general change of variables
in this case is zi = e2xi , 1  i  N , but the following formulas are independent of the
choice of sign in the exponent and the value of the constant . Gauge factor:


 a
sinh 2xij .
(x) =
(34)
i<j

Scalar external potential: U (x) = 0. Spin potential:




Vspin(x) = 2a
sinh2 xij (a + Sij ) cosh2 xij (a + Sij ) .

(35)

i<j

Parameters: c0 = 4m. Domain: xN < < x1 .


Case 2b P (z) = 4 z2 . Change of variables: z = e2ix . Again, the most general change of
variables is zi = e2ixi , 1  i  N , but the following formulas do not change when this
is taken into account. Gauge factor:
 
 a
sin 2xij .
(x) =
(36)
i<j

Scalar external potential: U (x) = 0. Spin potential:




Vspin(x) = 2a
sin2 xij (a + Sij ) + cos2 xij (a + Sij ) .

(37)

i<j

Parameters: c0 = 4m. Domain: xN < < x1 < xN + /2.


Both potentials in this case are invariant under a simultaneous translation of all the
particles coordinates. The choice c0 = 4m, which simplifies the form of the gauge factor,
amounts to fixing the center of mass energy of the system. Note also that the potentials in
this case do not possess BN symmetry, due to the fact that the change of variables cannot
be made an odd function of x for any choice of the arbitrary constants. In fact, the sign
change zk  zk corresponds to the translation xk  xk + i/2 or xk  xk + /2, which
(as any overall translation) leaves the potential invariant. The potentials in this case are
therefore best interpreted as AN -type potentials depending both on spin permutation and
sign reversing operators.
For the hyperbolic potential 2a, none of the algebraic formal eigenfunctions are true
eigenfunctions, since they are not square integrable on their domain. On the other hand, the
algebraic eigenfunctions of the periodic potential 2b are clearly periodic in each coordinate
and regular on their domain, and thus qualify as true eigenfunctions.
Case 3a P (z) = 4(1 + z2 ). Change of variables: z = sinh(2x). Gauge factor:




 a
(x) =
sinh 2xij sinh 2xij+
[sinh(2xi )]b [cosh(2xi )] .
i<j

(38)

Scalar external potential:


U (x) = 4( 1) cosh2 (2x).

(39)

484

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

Spin potential:
Vspin(x) = 2a




sinh2 xij cosh2 xij+ (a + Sij )

i<j




+ sinh2 xij+ cosh2 xij (a + Sij )

+ 4b
sinh2 (2xi )(b + Si ).

(40)

Parameters:





m
c0
a(N 1) + b +
=
< 2a(N 1) + b + m .
8
2

Domain: 0 < xN < < x1 .


Case 3b P (z) = 4(1 + z2 ). Change of variables: z = i sin(2x). Gauge factor:
 
 
 a
sin 2xij sin 2xij+
(x) =
[sin(2xi )]b [cos(2xi )] .
i<j

(41)

Scalar external potential:


U (x) = 4( 1) cos2 (2x).
Spin potential:
Vspin(x) = 2a



(42)


sin2 xij + cos2 xij+ (a + Sij )

i<j




+ sin2 xij+ + cos2 xij (a + Sij )

+ 4b
sin2 (2xi )(b + Si ).

(43)

Parameters:
=

c0
m
+ a(N 1) + b +
8
2


>

1
2

or = 0.

Domain: 0 < xN < < x1 < /4, if > 1/2 and = 1;


0 < xN < < x1 < 2 x2 , if = 0, 1.
Case 4a P (z) = (1 z2 )2 . Change of variables: z = tanh x. Gauge factor:

a cosh(2x )

i (sinh x )b (cosh x )b +m .
sinh xij sinh xij+
(x) =
e
i
i
i<j

(44)

Scalar external potential:


U (x) = 2 2 cosh(4x) + 4(1 + 2) cosh(2x).
Spin potential:
Vspin(x) = 2a


i<j

sinh2 xij (a + Sij ) + sinh2 xij+ (a + Sij )

(45)

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

+b

sinh2 xi (b + Si ) b



cosh2 xi b + (1)m Si .

485

(46)

Parameters: = 18 (c0 + 2(b b )) < 0, or = 0 and < 0.


Domain: 0 < xN < < x1 .
Alternatively, we could have taken the change of variables as z = tanh(x i2 ) = coth x.
The gauge factor, external potential and spin potential become, respectively,

a cosh(2x )

i
sinh xij sinh xij+
e
(cosh xi )b (sinh xi )b +m ,
(x) =
(47)
i<j

U (x) = 2 cosh(4x) 4(1 + 2) cosh(2x),


2

and
Vspin(x) = 2a



sinh2 xij (a + Sij ) + sinh2 xij+ (a + Sij )

i<j

(48)

cosh2 xi (b + Si ) + b



sinh2 xi b + (1)m Si .

(49)

Case 4b P (z) = (1 z2 )2 . Change of variables: z = i tan x. Gauge factor:



a cos(2x )

i (sin x )b (cos x )b +m .
sin xij sin xij+
e
(x) =
i
i
i<j

(50)

Scalar external potential:


U (x) = 2 2 cos(4x) 4(1 + 2) cos(2x).

(51)

Spin potential:
Vspin(x) = 2a



sin2 xij (a + Sij ) + sin2 xij+ (a + Sij )

i<j

+b

sin2 xi (b + Si ) + b

= 18 (c0



cos2 xi b + (1)m Si .

(52)

+ 2(b

Parameters:
b)). Domain: 0 < xN < < x1 < /2.
The change of variable can also be taken as z = i tan(x /2) = i cot x. Since this is the
result of applying an overall real translation to the particles coordinates, we shall not list
the corresponding formulas for the potential and gauge factor.
x dn x
Case 5 P (z) = (e2i z2 )(e2i z2 ). Change of variables: z = sn cn
x , where the
modulus of the elliptic functions is k = cos . We shall
also
use
in
what
follows the


2
customary notation k for the complementary modulus 1 k . Gauge factor:



 sn xij dn xij sn xij+ dn xij+ a
k
(x) =
exp arctan  cn(2xi )
k
1 k 2 sn2 xij sn2 xij+
i<j
i


[sn(2xi )]b [1 + cn(2xi )] 2 (b b+m) [dn(2xi )] .


1

(53)

486

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

Scalar external potential:


2

U (x) = 4k dn



k
2
(2x) ( + 1)  (1 + 2) cn(2x) .
k

(54)

Spin potential:
Vspin(x) = 2a

 dn2 xij


sn2 xij

i<j


+
+b

dn2 xij+

sn2 xij+


k2k 2

k2k 2


(a + Sij )
+

sn2 xij+
dn2 xij



ij )
(a
+
S

sn2 xij

dn2 xij
2
 sn xi dn xi 2 

cn xi

(b + Si ) + b
b  + (1)m Si .
sn xi dn xi
cn xi

(55)
Parameters:

1
= 8kk
 (c0

+ 2(k 2

k 2 )(b b )).

Domain: 0 < xN < < x1 < K.

An alternative form for the change of variables in this case is


z=

cn x
sn(x K) dn(x K)
=
.
cn(x K)
sn x dn x

The resulting potential is obtained from the previous one by applying the overall real
translation xi  xi K, i = 1, . . . , N . Note also that, although zi is singular at the zeros
of cn xi , the algebraic eigenfunctions satisfy the appropriate boundary condition on these
hyperplanes on account of the inequality b > 1/2 and the identity
2 cn2 x
.
1 k 2 sn4 x

1 + cn(2x) =

Case 6a P (z) = 4(1 z2 )(1 k 2 z2 ). Change of variables: z = sn(2x). Here, as in the


remaining cases, the Jacobian elliptic functions have modulus k. Gauge factor:
(x) =

(sn xij cn xij dn xij sn xij+ cn xij+ dn xij+ )a


(1 k 2 sn2 xij sn2 xij+ )2a

i<j

[sn(2xi )]b [cn(2xi )] [dn(2xi )] .

(56)

Scalar external potential:




U (x) = 4k  2 ( 1) cn2 (2x)  (  1) dn2 (2x) .
Spin potential:
Vspin(x) = 2a

 cn2 xij dn2 xij


i<j

sn2 xij

+k

4

sn2 xij+


(a + Sij )
+

cn2 xij+ dn2 xij

(57)

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496


+

cn2 xij+ dn2 xij+



ij )
(a
+
S

sn2 xij

+ k 4

487

sn2 xij+
cn2 xij dn2 xij




+ 4b
sn2 (2xi )(b + Si ) + 4k 2 b
sn2 (2xi ) b + (1)m Si .
i

(58)

Parameters:



1
1 
2

or = 0,
c
(b

b
+
4
1
+
k
)
>
0
8k  2
2



1 
 =  2 c0 + 4 1 + k 2 (b b ) .
8k
Domain: 0 < xN < < x1 < K/2, if > 1/2 and = 1;
0 < xN < < x1 < K x2 , if = 0, 1.
=

Case 6b P (z) = 4(1 z2 )(1 k  2 z2 ). Change of variables:


sn(2x)
.
cn(2x)
Gauge factor:
z=i

(x) =

(sn xij cn xij dn xij sn xij+ cn xij+ dn xij+ )a


(1 k 2 sn2 xij sn2 xij+ )2a

i<j

[sn(2xi )]b [cn(2xi )]b +m [dn(2xi )] .

(59)

Scalar external potential:




U (x) = 4 ( 1)k 2 sn2 (2x)  (  1)k  2 dn2 (2x) .

(60)

Spin potential:
Vspin(x) = 2a



dn2 xij
sn2 xij cn2 xij

+k

sn2 xij+ cn2 xij+


(a + Sij )

dn2 xij+



sn2 xij cn2 xij
dn2 xij+
4
ij )
(a
+
S
+
k
+
sn2 xij+ cn2 xij+
dn2 xij




sn2 (2xi )(b + Si ) + 4k  2 b
cn2 (2xi ) b  + (1)m Si . (61)
+ 4b
i<j

Parameters:



1 
c0 + 4 1 + k  2 (b  b) ,
2
8k



1 
 = 2 c0 + 4 1 + k  2 (b b) .
8k
Domain: 0 < xN < < x1 < K/2.
In spite of the singularity of zi at the zeros of cn(2xi ), the vanishing of the gauge factor
on these hyperplanes clearly implies that the algebraic eigenfunctions fulfill the appropriate
boundary condition.
=

488

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

Case 7 P (z) = 4(1 z2 )(k  2 + k 2 z2 ). Change of variables:


z = cn(2x K) = k 

sn(2x)
.
dn(2x)

Gauge factor:
(x) =

(sn xij cn xij dn xij sn xij+ cn xij+ dn xij+ )a


i<j

(1 k 2 sn2 xij sn2 xij+ )2a

[sn(2xi )]b [cn(2xi )] [dn(2xi )]b +m .

(62)

Scalar external potential:




U (x) = 4  (  1)k 2 sn2 (2x) + ( 1)k  2 cn2 (2x) .

(63)

Spin potential:
Vspin(x) = 2a



cn2 xij
sn2 xij dn2 xij

i<j

cn2 xij+

sn2 xij+ dn2 xij+


cn2 xij+

sn2 xij dn2 xij


(a + Sij )




(a + Sij )

+
cn2 xij
sn2 xij+ dn2 xij+




sn2 (2xi )(b + Si ) 4k  2 b
dn2 (2xi ) b + (1)m Si .
+ 4b
+

(64)
Parameters:


 
c0 1  2
1
2
=
+ k k (b b) + >
or = 0,
8
2
2

c0 1  2
 =
+ k k  2 (b b) .
8
2
Domain: 0 < xN < < x1 < K/2, if > 1/2 and = 1;
0 < xN < < x1 < K x2 , if = 0, 1.

5. Discussion
The BCN -type potentials constructed in the previous section can be expressed in a
unified way that we shall now describe. In the first place, apart from irrelevant constant
operators of the form (29), the spin potential can be written as
  


v xij + v xij+ + P1 (a + Sij )
Vspin(x) = 2a
i<j

  


+ v xij+ + v xij + P1 (a + Sij )


v(xi ) + v(xi + P1 ) (b + Si )
+b
i

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

+ b




v(xi + P2 ) + v(xi + P1 + P2 ) b + (1)m Si ,

489

(65)

where v is a (possibly degenerate) elliptic function, and P1 and P2 are suitably chosen
primitive half-periods of v (see Table 1). In particular, when one of the periods Pi of v
goes to infinity, expressions like v(x + Pi ) with x R finite are defined as zero. In Case 1,
both periods are infinite and v(x + P1 + P2 ) is also defined as zero. Furthermore, the
constant K  K  (k) in the elliptic Cases 57 is the complete elliptic integral of the first
kind defined by K  (k) = K(k  ). Using this notation, it is easy to verify that in the nondegenerate elliptic Cases 57 the scalar external potential U (x) can be written as
 



1
1
U (x) = ( 1) v x + P1 + v x P1
2
2



 
1
1
 
+ ( 1) v x + P1 + P2 + v x P1 + P2 ,
(66)
2
2
where = and  =  in Cases 67, while = + i and  = i in Case 5.
Formula (66) holds also in Case 3, with = and  = 0. In Cases 1 and 4 Eq. (66) cannot
be directly applied, since in these cases all the terms in (66) are either indeterminate or
zero. However, the potentials in Cases 4a and 4b can be obtained from that of Case 5 in
the limits 0 and /2, respectively. Likewise, applying the rescaling xi  xi
(i = 1, . . . , N , > 0) to the potential of type 3a or 3b one obtains the potential of type 1
by taking = /(4 2 ) and letting 0.
Table 1
Function v(x) and its primitive half-periods Pi (see Eq. (65)) for
each of the BCN -type potentials in Section 4
Case

v(x)

P1

P2

x 2

3a
3b

sin2 x

i
2

sinh2 x

4a

sinh2 x

4b

sin2 x

dn2 x
sn2 x
2
cn x dn2 x
sn2 x
dn2 x
2
sn x cn2 x
cn2 x

K + iK 

5
6a
6b
7

sn2 x dn2 x

K
iK 
K

i
2

2
K
iK 
2
K
2
1
(K + iK  )
2

490

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

The function v(x) that determines the potential V (x) in the elliptic Cases 57 according
to Eqs. (65) and (66) can be expressed in a systematic way in terms of the Weierstrass
function (x; 1, 3 ) with primitive half-periods 1 = K and 3 = iK  . Indeed, dropping
inessential constant operators we have
v(x) = [ (x; 1, 3 ) + (x + 2P2 ; 1 , 3 )],

(67)

where P2 is the primitive half-period of v listed in Table 1, and  = 1 for Cases 67


while  = 1/2 for Case 5 (the only case in which 2P2 = 2K is a period of ). Since
in Cases 67 P1 and 2P2 are primitive half-periods of , the well-known second-order
modular transformation of the Weierstrass function [35] applied to Eq. (67) leads to the
equality
v(x) = (x; P1, P2 ),

(68)

where the primitive half-periods P1 and P2 are listed in Table 1, and we have dropped an
irrelevant additive constant. Substituting Eq. (68) into Eqs. (65) and (66) and applying once
again a modular transformation to the one-particle terms we readily obtain the following
remarkable expression for the potential V (x) in Cases 57:
 



xij ; P1 , P2 + xij+ + P1 ; P1 , P2 (a + Sij )
V (x) = 2a
i<j




 
+ xij+ ; P1 , P2 + xij + P1 ; P1 , P2 (a + Sij )

(2xi ; P1 , 2P2 )(b + Si )
+ 4b
i

+ 4b



(2xi + 2P2 ; P1 , 2P2 ) b + (1)m Si

+4



( 1) (2xi + P1 ; P1 , 2P2 )


+  ( 1) (2xi + P1 + 2P2 ; P1 , 2P2 ) .

(69)

One of the main results in this paper is thus the fact that the potential (69) is QES provided
that the ordered pair (P1 , P2 ) is chosen from Cases 57 in Table 1. In fact, the remaining
BCN -type (Q)ES spin potentials listed in Section 4 can be obtained from the potentials in
Eqs. (69) by sending one or both of the half-periods of the Weierstrass function to infinity.
This is of course reminiscent of the analogous property of the integrable scalar Calogero
Sutherland models associated to root systems [3].
The potentials in Cases 1, 2, and 3 are ES for all values of the parameters. (In Case 3, the
dependence of the parameter on m through can be absorbed in the coefficient c0 .) The
potentials of type 4 are also ES for = 0. The elliptic potentials in Cases 57 are always
QES.
All the potentials presented in Section 4 are new, except for Cases 1 and 4. Case 1 is the
rational BN -type model introduced by Yamamoto [16] and studied by Dunkl [19]. Case 4b
for = 0 is Yamamotos BN -type trigonometric potential with 1 = b (in the notation of

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

491

Ref. [16]), and either 1 = b  < 1/2 for m even or 1 = b > 1/2 for m odd. Our results
thus establish the exact solvability of the trigonometric Yamamoto model when |1 | > 1/2.
It should be noted that the method developed in this paper admits a number of
straightforward generalizations. In the first place, the algebraic states could be chosen
symmetric under sign reversals. The resulting Hamiltonians would coincide with the ones
presented in Section 4 with Si replaced by Si . In particular, if b = b = 0 one can obtain
algebraic eigenfunctions of both types (symmetric and antisymmetric under sign reversals)
for the same Hamiltonian. The construction can also be applied to a system of N identical
bosons, just by replacing the antisymmetriser 0 by the projector on states symmetric
under permutations of the particles. Choosing a system of fermions is motivated by the
fact that the internal degrees of freedom can be naturally interpreted as the physical spin of
the particles when M = 1/2.
The procedure described in Section 3 relies on the algebraic identities analogous to (2)
satisfied by the spin operators Sij and Si , and not on the particular realization (11). For
instance, replacing the operators Sij by new operators Sij spanning one of the anyon-like
realizations introduced by Basu-Mallick [17] would yield further families of (Q)ES spin
CalogeroSutherland models.

6. Exact solutions for an elliptic QES model


As an illustration of the procedure described in the previous sections, we shall now
compute the algebraic sector of the spectrum for the elliptic QES potential of type 6a in
Eqs. (57)(58) in the case of two and three particles of spin 1/2 (N = 2, 3 and M = 1/2),
for m = 1, 2, 3. Note that in the spin 1/2 case, the spin permutation and sign reversing
operators Sij and Si can be expressed in terms of the usual one-particle SU(2) spin
operators i = (i1 , i2 , i3 ) in the more familiar way
1
Si = 2i1 .
Sij = 2 i j + ,
2
The operator H corresponding to the potential (57), (58) reads
N 


 2
 2

 2
 + E0 ,
H =
4k 2 Ji+ 4 1 + k 2 Ji0 + 4 Ji + c0 Ji0 + C

(70)

i=1

 is the constant operator obtained by replacing Sij by Kij and Si by Ki in the


where C
 and the scalar constant E0 is given by
expression (16) for C,


1

E0 = c0 N a(N 1) + b + m +
2



2N 1 + k 2 2a(N 1)(2b + m) + 2b (b + m + 1)

2
+ m + (N 1)(2N 1)a 2 .
3

492

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

Let us first consider the two-particle case (N = 2), for which the spin space S is spanned
1 is the oneby the four spin states | | 12 12 . For m = 1 the polynomial module R
dimensional space span{1 }, with
1 = (z1 z2 )(|++ | ) + (z1 + z2 )(|+ |+ ).

(71)

Therefore, the spin state


1 (x) =


(x)

+
+

+ sn x12 cn x12 dn x12 (|++ | )


2
2
2
1 k sn x12 sn x12

+

cn x12
dn x12
(|+ |+ )
+ sn x12

(72)

is an eigenfunction of the Hamiltonian of type 6a, where the gauge factor (x) is given in
Eq. (56). The corresponding eigenvalue is E1 = E0 c0 .
2 is the three-dimensional space
If m = 2, the antisymmetrised polynomial module R
span{1 , 2 , 3 }, where 1 is given by Eq. (71) and


2 = z12 z22 (|++ + | |+ |+ ),
3 = z1 z2 (z1 z2 )(|++ | ) + z1 z2 (z1 + z2 )(|+ |+ ).

(73)

 (or
The matrix of the gauge spin Hamiltonian H
in the basis {1 , 2 , 3 } is given by

2
E0 c0 4(1 + k )
0
8(2b + 1)

,
0
0
E0 2c0

2
2
0
E0 3c0 4(1 + k )
8(2b + 1)k
H )

whose eigenvalues are



E1,3 = E0 2c0 4 1 + k 2 ,

E2 = E0 2c0 ,

where = [c02 + 64k 2 (2b + 1)(2b + 1)]1/2. The corresponding physical wavefunctions
are

(x)

+
+
1,3 (x) =
sn x12
cn x12
dn x12

+
1 k 2 sn2 x12 sn2 x12


c0 + 8k 2 (2b + 1) sn(2x1) sn(2x2) (|++ | )
+

cn x12
dn x12
+ sn x12


c0 8k 2 (2b + 1) sn(2x1) sn(2x2) (|+ |+ ) ,

2 (x) = (x)

+
+
+
sn x12
cn x12
dn x12
sn x12
cn x12
dn x12

+ 2
(1 k 2 sn2 x12
sn2 x12
)

(|++ + | |+ |+ ).

(74)

3 is spanned by the spin functions


For m = 3 the antisymmetrised polynomial module R
1 , . . . , 6 , where




4 = z13 z23 (|++ | ) + z13 + z23 (|+ |+ ),


5 = z1 z2 z12 z22 (|++ + | + |+ + |+ ),


6 = z12 z22 z1 z2 (|++ | ) + z12 z22 (z1 + z2 )(|+ |+ ).
(75)

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

 E0 in the basis {1 , . . . , 6 } is
The matrix representing H
c0 8(1 + k2 )
0
8(2b + 1)
8(4a + 2b + 3)

0
8(2b + 3)k 2
8(2b + 1)k 2
0
0

2c0 8(1 + k 2 )
0
0
0
0

0
3c0 16(1 + k 2 )
0
0
8(2b + 1)k 2

0
16a(1 + k 2 )
3c0 + 16a(1 + k 2 )
0
8(4a + 2b + 3)k 2

0
0
0
0
4c0 8(1 + k 2 )
0

493

0
0

8(2b + 3)

8(2b + 1)

0
5c0 8(1 + k 2 )

 and
Clearly, E0 2c0 8(1 + k 2 ) and E0 4c0 8(1 + k 2 ) belong to the spectrum of H
hence of H , with corresponding eigenfunctions given, respectively, by 2 (x) in Eq. (74)
and
5 (x) = (x)

+
+
+
cn x12
dn x12
sn x12
cn x12
dn x12
sn x12

+ 2
(1 k 2 sn2 x12
sn2 x12
)

sn(2x1) sn(2x2)

(|++ + | + |+ + |+ ).

(76)

The remaining algebraic levels are the roots of a fourth degree polynomial, whose
expression is too long to display here. For instance, if a = b = b = 1, k 2 = 1/2, and c0 =
14 (so that = 0), the algebraic levels are approximately E1 = 327.4, E2 = 288,
E3 = 281.4, E4 = 262.2, E5 = 260, and E6 = 201.0.
In the three-particle case, the spin space S is spanned by the eight states | . If
2
1 is trivial. For m = 2, the antisymmetrised space R
m = 1, the antisymmetrised space R
is spanned by the single state

+ +
z13 z23 (|+++ + | ) z12
z13 z23 (|++ + |+ )
1 = z12
+ +
+ +
z12
z13 z23 (|+ + |++ ) + z12
z13 z23 (|++ + |+ ),

(77)

= zi zj . Consequently, 1 (x) = (x)1 (z), with zi = sn(2xi ), is an


where zij
eigenfunction of H with eigenvalue E0 3c0 4(1 + k 2 ).
3 is given by the function 1 in Eq. (77) and
When m = 3, a basis for R

2 = z12
z13 z23 (z1 + z2 + z3 )(|+++ | )
+ +
+ z12
z13 z23 (z1 + z2 z3 )(|+ |++ )
+ +
+ z12
z13 z23 (z1 + z2 + z3 )(|+ |++ )
+ +
z13 z23 (z1 z2 + z3 )(|++ |+ ),
+ z12

z13 z23 (z1 z2 + z1 z3 + z2 z3 )(|+++ + | )
3 = z12
+ +
+ z12
z13 z23 (z1 z2 + z1 z3 + z2 z3 )(|+ + |++ )
+ +
+ z12
z13 z23 (z1 z2 + z1 z3 z2 z3 )(|+ + |++ )
+ +
z13 z23 (z1 z2 z1 z3 + z2 z3 )(|++ + |+ ),
z12

+ +
4 = z1 z2 z3 z12
z13 z23 (|+++ | ) + z12
z13 z23 (|++ |+ )

+ +
+ +
+ z12 z13 z23 (|++ |+ ) + z12 z13 z23 (|+ |++ ) .

 in the basis {1 , . . . , 4 } reads


The matrix of H

(78)

494

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

E0 3c0 16(1 + k 2 )
0

8(2b + 1)k 2
0

0
E0 4c0 + 8(1 + k 2 )(2a 1)
0
8(4a + 2b + 3)k 2

8(4a + 2b + 3)
0
E0 5c0 + 8(1 + k 2 )(2a 1)
0

0
8(2b + 1)

0
2
E0 6c0 16(1 + k )

The eigenvalues of this matrix are




E1,3 = E0 4c0 + 4 1 + k 2 (2a 3) ,


E2,4 = E0 5c0 + 4 1 + k 2 (2a 3) + ,
where

2


1/2
+ = c0 + 4 1 + k 2 (2a + 1) + 64k 2 (2b + 1)(4a + 2b + 3)
,


1/2

2
.
= c0 4 1 + k 2 (2a + 1) + 64k 2 (2b + 1)(4a + 2b + 3)
The corresponding eigenfunctions are





1,3 (x) = (x) c0 4 1 + k 2 (2a + 1) 1 (z) + 8k 2 (2b + 1)3 (z) ,





2,4 (x) = (x) c0 + 4 1 + k 2 (2a + 1) + 2 (z) + 8k 2 (4a + 2b + 3)4(z) ,
with zi = sn(2xi ).

Acknowledgements
This work was partially supported by the DGES under grant PB98-0821. R. Zhdanov
would like to acknowledge the financial support of the Spanish Ministry of Education and
Culture during his stay at the Universidad Complutense de Madrid.

Appendix A
In this appendix we present a list of identities satisfied by the Dunkl operators (3)(5)
used to compute the gauge spin Hamiltonian (15).
 2 

Ji =
z2i + 4a
i

z2
i=j i

+a

1
zi
zi + 2b
z
2
zi i
zj
i

 Kij 1
 K
 Ki 1
ij 1
+
a
+
b
,
(zi zj )2
(zi + zj )2
zi2
i=j

i=j

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

495

 2 



Ji0 =
zi2 z2i + 1 m + 2a(1 N) zi zi
i

+ 4a
a

zi3

z2
i=j i

zj


i=j

zi zj
(Kij 1)
(zi zj )2

2
zi zj
ij 1) + Nm
(
K
(zi + zj )2
4

i=j

a2

+a
2 zi





ij )(Kik + K
ik ) + 6
4 (Kij + K
(1 Ki Kj ) ,

12
i,j,k
i=j
 2 
 
 3 
+
4 2
Ji
zi zi 2 b + m 1 + 2a(N 1) zi zi
=
i

+ 4a

zi5

i=j

zi2 zj

+a
2 zi


i=j

zi2 zj2
(zi zj )2

(Kij 1)

zi2 zj2

ij 1)
(K
(zi + zj )2
 


+ b
zi2 (1)m Ki 1 + m(m 1 + 2b )
zi2 ,

+a

i=j

Ji0 =

The symbol



i,j,k

Nm
zi zi
.
2
means summation in i, j, k with i = j = k = i.

References
[1] F. Calogero, J. Math. Phys. 12 (1971) 419.
[2] B. Sutherland, Phys. Rev. A 4 (1971) 2019;
B. Sutherland, Phys. Rev. A 5 (1972) 1372.
[3] M.A. Olshanetsky, A.M. Perelomov, Phys. Rep. 94 (1983) 313.
[4] A.P. Polychronakos, Nucl. Phys. B 324 (1989) 597.
[5] Z.N.C. Ha, Phys. Rev. Lett. 73 (1994) 1574;
Z.N.C. Ha, Nucl. Phys. B 435 (1995) 604.
[6] S.B. Isakov, G. Lozano, S. Ouvry, Nucl. Phys. B 552 (1999) 677.
[7] H. Azuma, S. Iso, Phys. Lett. B 331 (1994) 107.
[8] A. Gorsky, N. Nekrasov, Nucl. Phys. B 414 (1994) 213.
[9] E. DHoker, D.H. Phong, Nucl. Phys. B 513 (1998) 405.
[10] A.P. Polychronakos, Phys. Rev. Lett. 74 (1995) 5153.
[11] Z.N.C. Ha, F.D.M. Haldane, Phys. Rev. B 46 (1992) 9359.
[12] K. Hikami, M. Wadati, Phys. Lett. A 173 (1993) 263.
[13] D. Bernard, M. Gaudin, F.D.M. Haldane, V. Pasquier, J. Phys. A: Math. Gen. 26 (1993) 5219.
[14] J.A. Minahan, A.P. Polychronakos, Phys. Lett. B 302 (1993) 265.
[15] J.A. Minahan, A.P. Polychronakos, Phys. Lett. B 326 (1994) 288.
[16] T. Yamamoto, Phys. Lett. A 208 (1995) 293.

496

[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]

F. Finkel et al. / Nuclear Physics B 613 [FS] (2001) 472496

B. Basu-Mallick, Nucl. Phys. B 482 (1996) 713.


V.I. Inozemtsev, Int. J. Mod. Phys. A 12 (1997) 195.
C.F. Dunkl, Commun. Math. Phys. 197 (1998) 451.
F. Finkel, D. Gmez-Ullate, A. Gonzlez-Lpez, M.A. Rodrguez, R. Zhdanov, Commun.
Math. Phys. 221 (2001) 447.
F.D.M. Haldane, Phys. Rev. Lett. 60 (1988) 635.
B.S. Shastry, Phys. Rev. Lett. 60 (1988) 639.
A.P. Polychronakos, Phys. Rev. Lett. 70 (1993) 2329.
C.F. Dunkl, Trans. Amer. Math. Soc. 311 (1989) 167.
I. Cherednik, Invent. Math. 106 (1991) 411;
I. Cherednik, Adv. Math. 106 (1994) 65.
A.P. Polychronakos, Phys. Rev. Lett. 69 (1992) 703.
A.V. Turbiner, Commun. Math. Phys. 118 (1988) 467.
M.A. Shifman, Int. J. Mod. Phys. A 4 (1989) 2897.
A.G. Ushveridze, Quasi-Exactly Solvable Models in Quantum Mechanics, Inst. Phys. Publ.,
Bristol, 1994.
A. Minzoni, M. Rosenbaum, A. Turbiner, Mod. Phys. Lett. A 11 (1996) 1977.
X. Hou, M. Shifman, Int. J. Mod. Phys. A 14 (1999) 2993.
D. Gmez-Ullate, A. Gnzalez-Lpez, M.A. Rodrguez, J. Phys. A: Math. Gen. 33 (2000) 7305.
A. Gonzlez-Lpez, N. Kamran, P.J. Olver, Commun. Math. Phys. 159 (1994) 503.
F. Finkel, N. Kamran, Nonlin. Math. Phys. 4 (34) (1997) 278.
D.F. Lawden, Elliptic Functions and Applications, Springer-Verlag, Berlin, 1989.

Nuclear Physics B 613 (2001) 497500


www.elsevier.com/locate/npe

CUMULATIVE AUTHOR INDEX B611B613

Abel, S.A.
ALPHA Collaboration
Arndt, D.
Astier, P.
Autiero, D.

B611 (2001) 43
B612 (2001) 3
B612 (2001) 171
B611 (2001) 3
B611 (2001) 3

Bais, F.A.
Baldisseri, A.
Baldo-Ceolin, M.
Banner, M.
Barenboim, G.
Bartels, M.
Baseilhac, P.
Bassompierre, G.
Becher, T.
Benedetti, R.
Beneke, M.
Benslama, K.
Bernard, D.
Besson, N.
Bird, I.
Blumenfeld, B.
Bobisut, F.
Botella, F.J.
Bouchez, J.
Boyd, S.
Brandhuber, A.
Brecher, D.
Bruzzo, U.
Bueno, A.
Bunyatov, S.
Buras, A.J.

B612 (2001) 229


B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 284
B612 (2001) 413
B612 (2001) 373
B611 (2001) 3
B611 (2001) 367
B613 (2001) 329
B612 (2001) 25
B611 (2001) 3
B612 (2001) 291
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 284
B611 (2001) 3
B611 (2001) 3
B611 (2001) 179
B613 (2001) 218
B611 (2001) 205
B611 (2001) 3
B611 (2001) 3
B611 (2001) 488

Caffo, M.
Camilleri, L.
Cardini, A.
Cattaneo, P.W.
Cavasinni, V.
Cervera-Villanueva, A.
Chaichian, M.
Chukanov, A.
Ciafaloni, M.

B611 (2001) 503


B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 383
B611 (2001) 3
B613 (2001) 381

0550-3213/2001 Published by Elsevier Science B.V.


PII: S 0 5 5 0 - 3 2 1 3 ( 0 1 ) 0 0 4 4 7 - 3

Ciafaloni, P.
Collazuol, G.
Comelli, D.
Conforto, G.
Conta, C.
Contalbrigo, M.
Cousins, R.
Cvetic, M.
Czarnecki, A.
Czyz, H.

B613 (2001) 381


B611 (2001) 3
B613 (2001) 381
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 167
B611 (2001) 488
B611 (2001) 503

DallAgata, G.
Daniels, D.
Degaudenzi, H.
Degrassi, G.
Delgado, A.
Del Prete, T.
Demichev, A.
De Santo, A.
Dhar, A.
Dib, C.O.
Diehl, H.W.
Dienes, K.R.
Dignan, T.
Di Lella, L.
Do Couto e Silva, E.
Dotsenko, V.S.
Dumarchez, J.
Drr, S.

B612 (2001) 123


B611 (2001) 3
B611 (2001) 3
B611 (2001) 403
B613 (2001) 49
B611 (2001) 3
B611 (2001) 383
B611 (2001) 3
B613 (2001) 105
B612 (2001) 492
B612 (2001) 340
B611 (2001) 146
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 445
B611 (2001) 3
B611 (2001) 281

Ellis, M.
Espinosa, O.

B611 (2001) 3
B612 (2001) 492

Falkowski, A.
Farzan, Y.
Feldman, G.J.
Feldmann, Th.
Feng, J.L.
Ferrari, F.
Ferrari, R.
Ferrre, D.
Finkel, F.

B613 (2001) 189


B612 (2001) 59
B611 (2001) 3
B612 (2001) 25
B613 (2001) 365
B612 (2001) 151
B611 (2001) 3
B611 (2001) 3
B613 (2001) 472

498

Nuclear Physics B 613 (2001) 497500

Flaminio, V.
Foerster, A.
Fraternali, M.
Fucito, F.

B611 (2001) 3
B612 (2001) 461
B611 (2001) 3
B611 (2001) 205

Gaillard, J.-M.
Gambino, P.
Gangler, E.
Garousi, M.R.
Gehrmann, B.
Geiser, A.
Geppert, D.
Gersdorff, G.v.
Gibin, D.
Gies, H.
Gninenko, S.
Godley, A.
Gomez-Cadenas, J.-J.
Gmez-Ullate, D.
Gomis, J.
Gonzlez-Lpez, A.
Gosset, J.
Gling, C.
Gouanre, M.
Gould, M.D.
Grant, A.
Graziani, G.
Guadagnini, E.
Guan, X.-W.
Gubser, S.S.
Guglielmi, A.
Gukov, S.

B611 (2001) 3
B611 (2001) 338
B611 (2001) 3
B611 (2001) 467
B612 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 49
B611 (2001) 3
B613 (2001) 352
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 472
B611 (2001) 179
B613 (2001) 472
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 461
B611 (2001) 3
B611 (2001) 3
B613 (2001) 329
B612 (2001) 461
B611 (2001) 179
B611 (2001) 3
B611 (2001) 179

Hagner, C.
Hatsuda, M.
Hatzinikitas, A.
Hebecker, A.
Hernando, J.
Herrmann, C.
Hubbard, D.
Hurst, P.
Hyett, N.

B611 (2001) 3
B611 (2001) 77
B613 (2001) 237
B613 (2001) 3
B611 (2001) 3
B612 (2001) 123
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3

Iacopini, E.

B611 (2001)

Joseph, C.
Joshipura, A.S.
Juget, F.

B611 (2001) 3
B611 (2001) 227
B611 (2001) 3

Kamimura, K.
Kazama, Y.
Khorsand, P.
Kim, J.E.
Kirsanov, M.
Kitazawa, Y.
Klimov, O.

B611 (2001) 77
B613 (2001) 17
B611 (2001) 239
B613 (2001) 305
B611 (2001) 3
B613 (2001) 105
B611 (2001) 3

Kokkonen, J.
Kovzelev, A.
Krasnoperov, A.
Kurth, S.
Kustov, D.
Kuznetsov, V.E.
Kyae, B.

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 305

Lacaprara, S.
Lachaud, C.
Lakic, B.
Lalak, Z.
Langfeld, K.
Lanza, A.
La Rotonda, L.
Laveder, M.
Lee, H.M.
Letessier-Selvon, A.
Levy, J.-M.
Lima-Santos, A.
Linssen, L.
Ljubic, A.
Long, J.
L, H.
Lukyanov, S.
Lupi, A.

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 189
B613 (2001) 352
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 305
B611 (2001) 3
B611 (2001) 3
B612 (2001) 446
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 167
B612 (2001) 391
B611 (2001) 3

Ma, J.P.
Mack, G.
Maggiore, N.
March-Russell, J.
Marchionni, A.
Martelli, F.
Martinelli, G.
Matchev, K.T.
Maxwell, C.J.
Mchain, X.
Mendiburu, J.-P.
Meyer, J.-P.
Mezzetto, M.
Miao, Y.-G.
Minasian, R.
Minasian, R.
Mirjalili, A.
Mishra, S.R.
Misiak, M.
Misiak, M.
Mizoguchi, S.
Moorhead, G.F.
Mller-Kirsten, H.J.W.
Muramatsu, T.

B611 (2001) 523


B612 (2001) 413
B613 (2001) 34
B613 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 311
B613 (2001) 365
B611 (2001) 423
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 215
B613 (2001) 87
B613 (2001) 127
B611 (2001) 423
B611 (2001) 3
B611 (2001) 338
B611 (2001) 488
B611 (2001) 253
B611 (2001) 3
B612 (2001) 215
B613 (2001) 17

Naumov, D.
Ndlec, P.
Nefedov, Yu.
Neubert, M.

B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 367

Nuclear Physics B 613 (2001) 497500


Nguyen, X.S.
Nguyen-Mau, C.
Nishino, H.
NOMAD Collaboration
Nomura, Y.

B613 (2001) 445


B611 (2001) 3
B612 (2001) 98
B611 (2001) 3
B613 (2001) 147

Olum, K.D.
Orestano, D.

B611 (2001) 125


B611 (2001) 3

Palma, G.
Park, D.K.
Paschos, E.A.
Pastore, F.
Peak, L.S.
Pennacchio, E.
Peres, O.L.G.
Pessard, H.
Petrov, A.A.
Petti, R.
Placci, A.
Pokorski, S.
Polesello, G.
Pollmann, D.
Polyarush, A.
Pope, C.N.
Popov, B.
Portugal, R.
Poulsen, C.
Prenajder, P.

B612 (2001) 413


B612 (2001) 215
B611 (2001) 227
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 59
B611 (2001) 3
B611 (2001) 367
B611 (2001) 3
B611 (2001) 3
B613 (2001) 189
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 167
B611 (2001) 3
B613 (2001) 237
B611 (2001) 3
B611 (2001) 383

Quirs, M.

B613 (2001) 49

Rajpoot, S.
Read, N.
Refolli, A.
Regnault, N.
Remiddi, E.
Rico, J.
Riemann, P.
Roda, C.
Rodejohann, W.
Roditi, I.
Rodrguez, M.A.
Rolf, J.
Rossi, G.C.
Rubbia, A.
Rupp, C.
Russo, J.G.

B612 (2001) 98
B613 (2001) 409
B613 (2001) 64
B612 (2001) 291
B611 (2001) 503
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 227
B612 (2001) 461
B613 (2001) 472
B612 (2001) 3
B611 (2001) 311
B611 (2001) 3
B612 (2001) 313
B611 (2001) 93

Sachrajda, C.T.
Saffin, P.M.
Saleur, H.
Salvatore, F.
Santachiara, R.
Santambrogio, A.
Santiago, J.

B611 (2001) 311


B613 (2001) 218
B613 (2001) 409
B611 (2001) 3
B613 (2001) 445
B613 (2001) 64
B611 (2001) 447

499

Schahmaneche, K.
Scharf, R.
Schmidt, B.
Schmidt, T.
Sconza, A.
Seidel, D.
Serban, D.
Servant, G.
Sevior, M.
Sfetsos, K.
Shadmi, Y.
Sharpe, S.
Shatashvili, S.L.
Sheikh-Jabbari, M.M.
Shpot, M.
Sibold, K.
Siemens, X.
Sillou, D.
Slavich, P.
Slingerland, J.K.
Smirnov, A.Yu.
Smith, D.
Smye, G.E.
Soler, F.J.P.
Sozzi, G.
Stanishkov, M.
Steele, D.
Stiegler, U.
Stipc, M.
Stolarczyk, Th.

B611 (2001) 3
B612 (2001) 313
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B612 (2001) 25
B612 (2001) 291
B611 (2001) 43
B611 (2001) 3
B612 (2001) 191
B613 (2001) 365
B611 (2001) 311
B613 (2001) 87
B611 (2001) 383
B612 (2001) 340
B612 (2001) 313
B611 (2001) 125
B611 (2001) 3
B611 (2001) 403
B612 (2001) 229
B612 (2001) 59
B613 (2001) 147
B613 (2001) 259
B611 (2001) 3
B611 (2001) 3
B612 (2001) 373
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3

Talevi, M.
Tani, T.
Tanzini, A.
Tanzini, A.
Tareb-Reyes, M.
Taylor, G.N.
Taylor, T.R.
Tereshchenko, V.
Terzi, N.
Testa, M.
Toropin, A.
Touchard, A.-M.
Tovey, S.N.
Tran, M.-T.
Travaglini, G.
Tsesmelis, E.
Tseytlin, A.A.
Tsimpis, D.
Tsvelik, A.M.
Tureanu, A.

B611 (2001) 311


B611 (2001) 253
B611 (2001) 205
B613 (2001) 34
B611 (2001) 3
B611 (2001) 3
B611 (2001) 239
B611 (2001) 3
B613 (2001) 64
B611 (2001) 311
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 205
B611 (2001) 3
B611 (2001) 93
B613 (2001) 127
B612 (2001) 479
B611 (2001) 383

Ulrichs, J.
Urban, J.

B611 (2001) 3
B611 (2001) 488

Vacavant, L.
Valdata-Nappi, M.

B611 (2001)
B611 (2001)

3
3

500

Nuclear Physics B 613 (2001) 497500

Valuev, V.
Vanhove, P.
Vannucci, F.
Varvell, K.E.
Veltri, M.
Vercesi, V.
Vidal-Sitjes, G.
Vieira, J.-M.
Vinogradova, T.
Vives, O.

B611 (2001) 3
B613 (2001) 87
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B611 (2001) 3
B613 (2001) 284

Weber, F.V.
Weiner, N.
Weisse, T.
Wilson, F.F.

B611 (2001) 3
B613 (2001) 147
B611 (2001) 3
B611 (2001) 3

Winton, L.J.
Wolff, U.

B611 (2001)
B612 (2001)

3
3

Yabsley, B.D.
Yndurin, F.J.

B611 (2001) 3
B611 (2001) 447

Zaccone, H.
Zagermann, M.
Zanon, D.
Zhdanov, R.
Zhou, H.-Q.
Zuber, K.
Zuccon, P.
Zwirner, F.

B611 (2001) 3
B612 (2001) 123
B613 (2001) 64
B613 (2001) 472
B612 (2001) 461
B611 (2001) 3
B611 (2001) 3
B611 (2001) 403

Anda mungkin juga menyukai