Nucl - Phys.B v.797 PDF

Nuclear Physics B 797 (2008) 126
www.elsevier.com/locate/nuclphysb
KaluzaKlein excitations of gauge bosons at the LHC

Abdelhak Djouadi a , Grgory Moreau a, , Ritesh K. Singh a,b
a Laboratoire de Physique Thorique, CNRS and Universit Paris-Sud, Bt. 210, F-91405 Orsay cedex, France
b Laboratoire de Physique Thorique, LAPTH, F-74941 Annecy-le-Vieux, France
Received 27 July 2007; accepted 18 December 2007

Available online 31 December 2007
Abstract
We consider the RandallSundrum extra-dimensional model with fields propagating in the bulk based
on an extended electroweak gauge symmetry with specific fermion charges and localizations that allow to
explain the LEP anomaly of the forwardbackward asymmetry for b-quarks, AbFB . We study the manifestations of the strongly-interacting and electroweak gauge boson KaluzaKlein excitations VKK at the LHC,
with masses of the order of a few TeV, which dominantly decays into top and bottom quark pairs. We first
analyze the two-body tree-level production processes pp t t and bb in which the KaluzaKlein (KK)
excitations of gauge bosons are exchanged. We find that the additional channels can lead to a significant
excess of events with respect to the Standard Model prediction; characteristic top quark polarization and
angular asymmetries are quantitatively studied where the polarization is the probe the chiral structure of
couplings to excited states. We then analyze higher order production processes for the gauge boson excitations which have too weak or no couplings to light quarks and, in particular, the loop induced process
gg VKK t t and bb in which the anomalous ggVKK four-dimensional vertex has to be regulated. The
t tt t, bbb
b and in the related
RS effects in this process, as well as in the four-body reactions pp t tbb,
in which the VKK excitations are mainly radiated off the heavy quarks,
three-body reactions gb bt t, bbb,
are shown to be potentially difficult to test at LHC, due to small phase space and low parton density for
MKK 3 TeV.
2007 Elsevier B.V. All rights reserved.
1. Introduction
Among the extra-dimensional models which have been proposed in the recent years as extensions of the Standard Model (SM), the one of Randall and Sundrum (RS) [1] seems to be
* Corresponding author.
E-mail address: moreau@th.u-psud.fr (G. Moreau).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.12.024
A. Djouadi et al. / Nuclear Physics B 797 (2008) 126
particularly attractive. First, it addresses the gauge hierarchy problem without introducing a new
energy scale in the fundamental theory. Moreover, the variant in which the SM fermions and
gauge bosons are propagating in the bulk allows for the unification of the gauge coupling constants at high energies [2] and provides a viable candidate of KaluzaKlein (KK) type for the
Dark Matter of the universe [3]. This version of the RS model with bulk matter offers also the
possibility, through a simple geometrical mechanism, of generating the large mass hierarchies
prevailing among SM fermions [46]. Indeed, if the various fermions are placed along the extra
dimension, their different wave functions overlap with the Higgs boson, which remains confined
on the so-called TeV-brane for its mass to be protected, generate hierarchical patterns among the
effective four-dimensional Yukawa couplings.
If this extra-dimensional model is to solve the gauge hierarchy problem, the masses of the
first KK excitations of the SM gauge bosons, MKK , must be in the vicinity of the TeV scale.
The direct experimental search for KK gluon excitations at Tevatron Run II leads to the bound
MKK 800 GeV [7,8]. The high-precision measurements impose strong lower bounds on MKK
as the exchanges of the KK excitations lead to new and unacceptably large contributions to the
electroweak observables [9]. Nevertheless, it was shown [10] that if the SM gauge symmetry is
enhanced to the leftright structure SU(2)L SU(2)R U (1)BL , with B and L being respectively the baryon and lepton numbers, the high-precision data can be nicely fitted while keeping
the masses down to an acceptable value1 of MKK 3 TeV, this lower limit being quite independent of the geometry of the extra dimension [12].
Instead of the U(1)BL group in the extended gauge symmetry as originally proposed in
Ref. [10], other U(1)X groups with different fermion charges can be considered [13,14]. An important motivation is that with specific charges of the new Abelian group and for specific fermion
localizations, different from the ones considered in Ref. [10], the three standard discrepancy between the forwardbackward asymmetry AbFB in Z decays to bottom quarks measured at LEP
and elsewhere [15,16] and the SM prediction [17] is naturally resolved, while keeping all the
other electroweak precision observables, including the partial decay width of the Z boson into
b-quarks Rb (Z bb )/ (Z hadrons), unaffected and in agreement with the data. More
precisely, in the framework presented in Refs. [13,14], the measured values of the asymmetry
AbFB ( s ) can be reproduced both at LEP1 [15], i.e. on the Z pole, and at energies outside the Zpole [16], while with the charge assignments of U(1)BL , the measured value of the asymmetry
cannot be reproduced at energies lower than LEP1; see Ref. [14] for a detailed discussion.
The resolution of the AbFB LEP anomaly is essentially due to the fact that since the third generation fermions should be localized closer to the TeV-brane to get higher masses, their couplings
to the KK gauge bosons, which are typically located near the TeV-brane, are larger and generate
more important contributions to electroweak observables in the b sector. One needs also to take
into account the tension between generating a large mass to the top quark, which requires the lefthanded doublet (t, b)L to be near the TeV-brane, and satisfying the bounds from precision data
that constrain the ZbL bL coupling to be almost SM-like, which forces the (t, b)L doublet [via the
mixing of the Z boson with the KK states] to have small couplings to the KK gauge bosons and
thus to be far from the TeV-brane. However, for a suitable choice of the (t, b)L representation under SU(2)R , the KK corrections to the coupling ZbL bL can be suppressed [13,18]. Alternatively,
1 The severe indirect bounds on M

KK from precision data could also be softened down to a few TeV in scenarios with
brane-localized kinetic terms for fermions and gauge bosons [11].
the tension is softened in the framework of Ref. [14] summarized above, in which the ZbL bL
and ZbR bR couplings are chosen so that the electroweak precision data are reproduced.2
Thus and, generically speaking, in the context of the RS model with SM fields in the bulk,
the KK gauge bosons dominantly couple to heavy SM fermions as they are localized toward the
TeV-brane (this typical feature can only be avoided in some particular situations [22]). In this
case, the processes involving the third generation b and t quarks are those which are expected to
be significantly affected by the presence of the new vector states. In Refs. [14,23], the impact of
the KK excitations of the electroweak gauge bosons has been discussed in the context of heavy
flavor production at high energy e+ e colliders (ILC). Because of the high precision which can
be achieved in the measurement of the production rates and polarization/angular asymmetries in
the e+ e t t, bb processes at the ILC, KK vector excitations with large masses can be probed,
provided that they have non-vanishing couplings
also to the initial light leptons. However, be
cause of the limited energy of the ILC, s = 0.51 TeV, and the expected large VKK masses,
MKK a few TeV, the particles are accessible only indirectly through their virtual exchange.
At the Large Hadron Collider (LHC), the energy is principle sufficient to produce directly
the KK excitations of the gluons and the electroweak gauge bosons with masses in the multiTeV range. The most direct manifestation of these KK states3 would be their production in the
DrellYan processes, pp VKK = g (1) , (1) , Z (1) , Z (1) . However, in contrast to the (standard)
production at the LHC of extra Z bosons from Grand Unified Theories (GUTs) [26] and to some
universal extra-dimensional models [27]4 , there are many difficulties in the present scenario.
The first one is that, contrary to the new Z s of GUTs, the KK excitations present in this model
have rather tiny couplings to the light fermions and large couplings to the heavy quarks. On the
one hand, the production cross sections in the DrellYan process q q VKK are much smaller,
the probability of finding a heavy quark in the proton being tiny; in fact, in the case of the
excitations of the extra Z , which is a superposition of the ZR and ZX bosons of the extended
SU(2)R U (1)X group, the couplings to light fermions are almost negligible and DrellYan
production does not occur [except through bb Z (1) which gives a small contribution due to
the reduced b-quark density in the proton]. On the other hand, because of their large couplings
to heavy fermions, the new KK excitations will decay almost exclusively into t t and bb final
states; these topologies have a less striking signature than the GUTS Z + with = e,
channel and are subject to a much more severe background at the LHC. Another difficulty, that is
related to the previous one, lies in the fact that because of the strong couplings to heavy fermions,
the KK excitations have large total decay widths and are not anymore narrow resonances. This
the
will complicate the searches for these states since in the processes p p VKK t t or bb,
corresponding QCD backgrounds become also large as they have to be integrated in wider bins
[29,30].
There are two other possibilities of observing the heavy KK states. The first one is single VKK production in the gluongluon fusion mechanism gg VKK which is mediated by
2 Note that there are also constraints on the top and bottom quark couplings from flavour changing neutral current
(FCNC) processes, but those can be satisfied for MKK values around the TeV scale Refs. [1921].
3 Another possibility to probe this variant of the RS scenario at LHC is to look for the production of KK gravitons [24],
or, the new quarks that are present in the model [25]. In the present analysis, we do not discuss these possibilities and we
assume that the new fermionic states are too heavy and do not end up as decay products of the KK gauge bosons.
4 In Ref. [27], this mechanism has been studied at the LHC in RS models with bulk matter but with the hypothesis of a
universal fermion profile [made in order to totally suppress FCNC effects] which are not compatible with an interpretation
of fermion mass hierarchies through the wave function overlap mechanism.
heavy Q = t, b quark loops5 ; because of the chirality dependence of VKK tt and VKK bb fourdimensional effective couplings, the one-loop effective ggVKK vertex is anomalous and the
chiral anomaly has to be regulated by considering the Stckelberg mixing term (for a review on
the topic, see e.g. Ref. [31]). The second possibility is associated production with heavy quark
KK with Q = t, b, leading to bbt
t, bbb
b and t tt t final states; anpairs, pp gg/q q QQV
other mechanism, which is related to the later class of processes would be the two-body process
bt t where the initial bottom parton is taken from the proton sea.6 In fact,
gb bVKK bbb,
when the couplings of the KK states to light quarks are small and the DrellYan process is not
effective, these mechanisms are the only possibilities which give access to the new states; this
is the case for the production of the KK excitation of the new Z boson for instance. However,
because these processes are of higher order in perturbation theory, the production cross sections
are not substantial despite the fact that the couplings which are involved are large. Nevertheless, in view of the interesting final state signature, a better control of the corresponding QCD
backgrounds might be possible.
In the present paper, we perform a complete study of the DrellYan and gluon fusion
t t as well as the associated production q q/gg
t, bbb
b,
t tt t and
processes q q,
gg bb,
bbt
bt t processes at the LHC, in the RS model with bulk matter. We will base our analygb bbb,
sis on the framework which resolves the AbFB anomaly [14], but the results that we obtain can
be easily generalized to other scenarios. In a parton-level analysis which includes only the dominant QCD backgrounds, we show that for a set of characteristic points of the parameter space,
the exchange of KK gauge bosons [excitations of the gluon as well as of the electroweak bosons
, Z and new Z ] can lead to visible deviations with respect to the SM production rates in
t t. For the latter reaction, the signal rates are large
the DrellYan process pp VKK bb,
and some top polarization and asymmetries allow to test the chiral couplings to the KK modes.
In the case where the KK excitations have too small couplings to the light quarks, the loop induced gluon fusion mechanism gg t t and bb as well as the associated production mechanisms
KK bbt
t and gb bt t, with an excess in the top or bottom pair invariant mass
pp QQV
distributions can be in principle exploited. However, because of the large mass of the KK excitations assumed in this study, the signal significance turns out to be small, except potentially, for
the gb bt t process.
Note that while the present analysis was on-going, the on-shell production of KK gluons
decaying into top quarks, pp g (1) t t has been studied at the LHC [28,29] in the framework
developed in Ref. [10] with the leftright extended gauge structure. The pp bb process has
not been considered as it is expected to be less significant than in the framework [14] considered
here where the right-handed b quark is closer to the TeV-brane, with an increased coupling to
KK gauge bosons; also in this case, the cross section for t t production would be different as b
coupling variations modify e.g. the KK gluon total width as well as the bb parton initial state
contribution (especially as here bR is closer to the TeV-brane so that it can get a non-vanishing
coupling to the Z ). Note also that the production of the excitations of the electroweak gauge
bosons as well as higher order processes (including 3 and 4 quark final states) have not been
5 The process gg V
KK t t has been first suggested in Ref. [10] and recently mentioned in Ref. [28]; however,
because of Yangs theorem the vector part of the ggVKK coupling will give zero contribution while the axial part of the
coupling will give a finite contribution only if the VKK state is virtual and, thus, if its total width is included.
6 This second class of processes have also been discussed in Ref. [32] for rather general scalar and vector resonances
which couple to top and bottom quarks with the same strength.
considered in Refs. [28,29] (in [28] the emphasis was put on the determination of excitation spin
and methods of highly energetic top identification).
The paper is organized as follows. In Section 2, we briefly describe the theoretical framework in which we will work and the considered scenarios which will be used for illustration;
the decays of the KK states will also be summarized. In Section 3, the signal of the KK excitations at the LHC in the DrellYan processes pp t t, bb and the corresponding irreducible
QCD backgrounds are analyzed; relevant polarization and angular asymmetries are studied in
details. In Section 4, we discuss the gluon fusion mechanism gg VKK t t, bb and briefly
the associated production processes with t t/bb pairs and those with a b-quark, as well as their
corresponding QCD backgrounds. A brief conclusion will be given in Section 5.
2. Physical framework
2.1. The specific RS extension
In this paper, we consider the RS model in which SM fields propagate along the extra spatial dimension, like gravity, but the Higgs boson remains confined on the TeV-brane. In the RS
scenario, the warped extra dimension is compactified over a S 1 /Z2 orbifold. While the gravity scale on the Planck-brane is MP = 2.44 1018 GeV, the effective scale on the TeV-brane,
M = ekRc MP , is suppressed by a warp factor which depends on the curvature radius of the
anti-de Sitter space 1/k and the compactification radius Rc . The product kRc 11 leads to
M = O(1) TeV, thus addressing the gauge hierarchy problem.
The fermion mass values are dictated by their wave function localization. In order to control
these localizations, the five-dimensional fermion fields i , with i = 1, 2, 3 being the generation
index, are usually coupled to distinct masses mi in the fundamental theory. If mi = sign(y)ci k,
where y parameterizes the fifth dimension and ci are dimensionless parameters, the fields de
(n) i
compose as i (x , y) =
n=0 i (x )fn (y), where n labels the tower of KK excitations and
f0i (y) = e(2ci )k|y| /N0i with N0i being a normalization factor. Hence, as ci increases, the wave
function f0i (y) tends to approach the Planck-brane at y = 0.
In order to protect the electroweak observables against large deviations and, at the same
time, to resolve the anomaly in the forwardbackward asymmetry for b-quark production AbFB ,
the electroweak gauge symmetry is enhanced to SU(2)L SU(2)R U(1)X with the following
fermion representations/charges under the group gauge [13,14]:
QiL (2, 1) 1 ;
6
dRi (1, 2) 5
with
1
ui
with I3RR = + ,
2
uiR (1, 2) 1
di
I3RR
1
=+ ,
2
1
(1)
2
respectively for the left-handed SU(2)L doublets of quarks, right-handed up/down type quarks,
doublets of leptons and right-handed charged leptons. The usual symmetry of the SM is recovered after the breaking of both SU(2)R and U(1)X on the Planck-brane, with possibly a small
breaking of the SU(2)R group in the bulk. Note the appearance of a new Z boson [but without a
zero-mode] which is a superposition of the state W 3 associated to the SU(2)R group and B associated to the U(1)X factor; the orthogonal state is the SM hypercharge B boson. Up-type quarks
LiL (2, 1) 1 ,
2
iR (1, 2) 1
i
with I3RR =
acquire masses through a Yukawa interaction of type (2, 1) 1 (2, 2)0 (1, 2) 1 , the Higgs boson be6
6
ing embedded in a bidoublet of SU(2)L SU(2)R . Because of their specific SU(2)R isospin
di
assignment I3RR , down-type quarks become massive via another kind of invariant operator as in
Refs. [13,14].
2.2. EW constraints and the fermion couplings
From the theoretical point of view, solving the gauge hierarchy problem forces the masses
of the first KK excitations of the SM gauge bosons, MKK = M (1) = Mg (1) MZ (1) MZ (1) ,
to be of order the TeV scale. Indeed, one has MKK = 2.45kM /MP M = O (TeV) since the
theoretical consistency bound on the five-dimensional curvature scalar leads to k < 0.105MP .
More precisely, the maximal value of MKK is fixed by this theoretical consistency bound and the
kRc value. One could consider a maximal value of MKK 10 TeV which corresponds to kRc =
10.11. From the experimental point of view, as mentioned previously, the electroweak (EW)
high-precision data force the KK mass to be larger than MKK 3 TeV. Given these theoretical
and experimental considerations, we will fix the masses of the KK excitations to a common value
MKK = 3 TeV in the present study.
With light SM fermions [leptons and first/second generation quarks] characterized by clight >
0.5, c being the parameter which determines the fermion localization as discussed in the subsection above, the bulk custodial isospin gauge symmetry insures an acceptable fit of the oblique
corrections to electroweak observables for MKK 3 TeV [10]. The large value of the top quark
mass requires ctR < 0.5 and cQ3 < 0.5, with cQ3 = ctL = cbL [as the states bL and tL belong to
L
L
the same SU(2)L multiplet] so that the top and bottom quarks have to be treated separately in the
studies of the electroweak precision data. In the framework developed in Ref. [14], the precision
data in the b sector, that is, the b-quark forwardbackward asymmetry AbFB and the bb partial
(Z hadrons), are correctly reproduced with

decay width of the Z boson Rb (Z bb)/
MKK = 3 TeV for certain values of the c parameters depending on the coupling constant gZ of
the new Z boson.
For instance, if the coupling

gZ is equal to 0.6 4 , the best fit of the observables Rb
measured at LEP1 and AbFB ( s ) measured at various center-of-mass energies is obtained for
2 14 and substantially improves the one
cQ3 0.36 and cbR 0.438 [the fit corresponds to RS
L
2 24]. For this c

in the SM, SM
bR value, one obtains the values Q(cbR ) 0.55 and Q (cbR )

0.75, where Q(c) (Q (c)) is the ratio of the four-dimensional effective coupling between the
g (1) / (1) /Z (1) (Z (1) ) boson and the SM fermions, over the coupling of the gluon/photon/Z
(would be Z ) boson zero-mode. In fact, the value chosen above for the gZ coupling is close to
the typical limit obtained from the perturbativity condition, 2kRc gZ2 /16 2 < 1, for the coupling of the KK excitation of the Z boson, when quantum corrections are included. Indeed, the
effectivefour-dimensional coupling of the Z (1) boson to SM fields is increased by a factor as
large as 2kRc relatively to gZ for SM fields near the TeV-brane, like for example the Higgs
boson.7
For the choice gZ = 0.3 4 , the best fit of Rb and AbFB ( s ) [still corresponding to a good
fit of the data, 2 14] is obtained for the c values cQ3 0.36 and cbR 0.135, leading to the
L
7 A bound of the same order can be derived from considerations on the strong coupling regime of the five-dimensional
theory: perturbativity imposes that the five-dimensional loop expansion parameter, estimated at an energy k to be
5D )2 k/16 2 = kR g 2 /16 2 , should be smaller than unity [33].
(gZ
c Z

Table 1
The four scenarios labeled E1,...,4 to be considered in this study and their respective c assignments, Q(c), Q (c) charges
and gZ couplings. The other (common) parameters are cQ3 = 0.36, which leads to Q(cQ3 ) = 1.34 and Q (cQ3 ) =
L
1.52), and clight 0.5, for which Q(clight ) = 0.2 and Q (clight ) = 0; MKK = 3 TeV
E1
E2
E3
E4
gZ
0.64
0.34
0.64
0.3 4
cbR
Q(cbR )
Q (cbR )
ctR
Q(ctR )
Q (ctR )
0.438
0.135
0.438
0.135
0.55
3.04
0.55
3.04
0.75
3.19
0.75
3.19
0.3
0.3
+0.1
+0.1
4.90
4.90
3.25
3.25
5.01
5.01
3.40
3.40
charges Q(cbR ) 3.0 and Q (cbR ) 3.2. For this value of cQ3 , one obtains for instance a top
L
quark mass mt 90 GeV or mt 80 GeV with, respectively, ctR 0.3 or ctR +0.1, which
are in the correct order of magnitude. In order to determine the set of parameters that would
reproduce exactly the measured mt value, one need to study the full three-flavour mass matrix
[by fixing also each Yukawa coupling constant] and include the effect of mixing between t and
its KK excitations, which is beyond the scope of this work. For cQ3 0.36 and cbR 0.438, the
L
bottom mass has also a correct order of magnitude. The bottom and top quark Yukawa couplings
have different gauge structures in the present context and the bottom Yukawa coupling constant
could be taken smaller (while choosing a smaller cbR , in order to still generate a satisfactory
bottom mass) as proposed in Ref. [13] in order to have a weak breaking of the custodial symmetry
subgroup protecting the ZbL bL coupling.
In this study, we will consider four typical scenarios which are defined by the c assignments
for the top and bottom quarks and the value of the coupling gZ ; these parameters and their
corresponding Q(c) and Q (c) charges are summarized in Table 1. In the four points of parameter
space, labeled E1,2,3,4 , the assignment for the left-handed (t, b) doublet, cQ3 = 0.36 which leads
L
to the charges Q(cQ3 ) = 1.34 and Q (cQ3 ) = 1.52, has been adopted in order to reproduce the
L
experimental values of AbFB and Rb . For the light fermions, i.e. the leptons and the first and
second generation quarks, we will set clight 0.5, as necessary to generate sufficiently small
masses. This leads to a small Q charge for these fermions, Q(clight ) 0.2, and a zero Z
charge, Q (clight ) = 0. This means that the couplings of the KK excitations of the gluon, photon
and Z boson have small couplings to light fermions compared to the top and bottom quarks [an
order of magnitude smaller in general], while the KK excitations of the Z boson do not couple
to these light fermions at all.
2.3. Decays of the KK gauge bosons
The KK excitations of the gauge bosons will decay into pairs of SM fermions, VKK f f;
as mentioned previously, we will assume that all the KK excitations of the fermions, as well as
the zero-modes of the new fermions that are present in the model, are too heavy, 2mf MVKK ,
so that they do not appear in the decays of the KK gauge bosons. The partial decay width of VKK
into a fermion species f is given by
(VKK f f)
=

MVKK f CfVKK 3 + f2
24
2
gfVLKK
VKK 2 3(f2 1) VKK VKK

+
+ gfR
gfL gfR
2

(2)
Table 2
The total decay widths (in GeV) and the bb and t t branching ratios of the first KK excitations of the gluon, photon, Z
BR(VKK t t)
and Z bosons in the four selected scenarios E1 , . . . , E4 for MKK = 3 TeV: VKK / BR(VKK bb)/
E1
E2
E3
E4
g (1)
(1)
Z (1)
Z (1)
627/0.08/0.91
828/0.30/0.69
328/0.14/0.83
530/0.47/0.52
137/0.02/0.96
149/0.10/0.89
68.0/0.04/0.93
80.0/0.18/0.79
75/0.28/0.68
79/0.31/0.65
54/0.39/0.56
58/0.43/0.53
3360/0.02/0.98
1080/0.29/0.71
1580/0.04/0.96
661.0/0.47/0.53
where mf is the mass of the fermion f , f =

1 4m2f /MV2KK its velocity in the rest frame
of the KK gauge boson and CfVKK the color factor: CqVKK = 3, CVKK = 1 for electroweak boson
g (1)
g (1)
couplings to, respectively, quarks and leptons and Cq = 12 , C = 0 for the coupling of the KK
gluon to quarks and colorless fermions. The left- and right-handed couplings of the KK gauge
boson VKK to the fermion f are denoted by gfVLKK and gfVRKK ; in terms of the couplings of the
KK
corresponding zero-mode, gfVL/R , they are given by gfVL/R
= gfVL/R Q() (cfL/R ), with Q(cfL/R ) and
Q (cfL/R ) as defined in the previous subsection and given in Table 1. The total decay width of the

KK excitation is simply VKK = f (VKK f f), where the sum runs over all SM fermions.
Because the KK gauge boson couplings to fermions are directly proportional to the charges
Q(c) and Q (c), the partial decay widths (VKK f f) (gfVLKK )2 + (gfVRKK )2 in the limit
MVKK mf , are typically two orders of magnitude larger in the case of final state top and bot
tom quarks than for the light fermions. The branching ratios BR(VKK t t ) and BR(VKK bb),
(1)
(1)

(1)
(1)
and g , are displayed in Table 2 in the four scenarios E1,...,4 introwith VKK = , Z , Z
duced previously. As can be seen, the sum of these two branching ratios is close to unity which
means that the KK excitations decay almost exclusively into the heavy t, b quarks and that little
room is left for decays into light quarks and leptons [in fact, there is no room at all for these
decays in the case of the Z (1) boson as the charge Q (clight ) is zero]. In our analysis, we will
therefore concentrate on the t, b final state decay products of the KK gauge bosons.
Finally, the total decay widths of the KK excitations VKK are also given in Table 2 for the
four scenarios. As they grow proportionally to the mass MVKK , they are rather large for the value
MKK = 3 TeV. For instance, the decay width of the KK gluon is of the order of a few hundred
GeV and is between 10% and 20% of the g (1) mass; the KK state can be thus considered as a
relatively narrow resonance. Due to the significant gZ values taken, the total decay widths are
smaller for the KK excitations of the photon and Z boson than for the KK excitation of the Z
boson. In the latter case, the total width is comparable to the Z (1) mass and one can hardly
consider the state as a true resonance.
3. Top and bottom quark pair production

3.1. General features
The most straightforward way to produce the KK excitations of the gauge bosons8 VKK at the
LHC is via the DrellYan process,
Q = t, b
pp q q VKK QQ,
(3)
8 From now on, we will restrict ourselves to the first KK excitations, V
(1) , Z (1) , Z (1) and g (1) .
KK =
(a)
(b)
Fig. 1. Feynman diagram for heavy Q = t, b quark pair production at the LHC through q q annihilation, where q stands
for initial state quarks and KK for the KK excitations of neutral gauge bosons, namely (n) , g (n) , Z (n) and Z (n) with
n = 0, 1, 2, . . . being the KK level; mainly the zero-modes n = 0, except for Z , and the first KK excitations n = 1
contribute to the signal. For bottom quark production there is an additional t -channel diagram exchanging neutral KK
gauge bosons with bottom partons in the initial state (b).
with the gauge bosons VKK subsequently decaying into top and bottom quarks; Fig. 1(a). As
discussed previously, VKK decays into the light fermion states [leptons and first/second generation quarks] will be neglected as their branching ratios are very small in the framework that
we are considering here. The relatively small couplings of the initial quarks q u, d, s, c to
VKK lead to smaller production rates compared to, for instance, the production of Z bosons
from GUTs which have full-strength couplings to light fermions. Because the VKK couplings to
bottom quarks are large, the partonic cross section from the subprocess bb VKK would, in
principle, be more substantial than for the q q VKK subprocesses but, after folding with the
smaller b-parton density, the corresponding hadronic cross section is reduced. Note that since
the KK gauge bosons have different couplings to left- and right-handed fermions, one expects
the produced t/b quarks to be polarized and to have a forwardbackward asymmetry.
Because the KK excitations have substantial total decay widths [see Table 2], the narrow width
approximation in which the production and decay processes are factorized and considered separately is not sufficient. Indeed, one needs to consider the virtual exchange of the KK excitations
in which the total width is included in a BreitWigner form, together with the exchange of the
zero modes. The full interference between the zero and first modes of the gluon, on the one hand,
and of the electroweak bosons, on the other hand, should be taken into account [because of color
conservation, there is no interference between the amplitudes in which the gluon and the electroweak gauge bosons are exchanged]. For the bb production, the subprocesses are also initiated
by bottom partons and one should also consider the channel in which the KK gauge bosons are
exchanged in the t-channel; Fig. 1(b).
The signal for the KK gauge bosons q q VKK QQ and the main SM background q q
QQ and gg QQ have to be considered simultaneously. For the signal reaction, we have
calculated the matrix element squared of the process pp QQ with polarized final state quarks
and incorporated the exchange, including the t -channel contributions, of all the SM gauge bosons
as well as their KK excitations and those of the Z boson. For the background processes, the
contributions of both the q q and gg initial states have been calculated and the exchange of the
KK excitations of the gauge bosons were switched off. To obtain the cross sections at the
hadronic level, we use the CTEQ5M1 set of parton distributions [34] with the factorization and
a renormalization scales set to the invariant mass of the QQ system, F = R = mQQ /2.
The significance S of the signal in the RS model can be then defined as
RS+SM SM
L,
(4)
SM
where L denotes the total integrated LHC luminosity. In our analysis, we will chose two examples
for the luminosity: a lower value L = 10 fb1 that is expected in the first years of the LHC
SL =
10
running and a high value L = 100 fb1 which is expected to be reached after a few years of
running.
In order to enhance the signal, which is peaked at the invariant mass of the QQ system (as
will shown in details later), and to suppress the continuum background, one needs to select events
near the KK resonance. Throughout this analysis, we impose a cut on the invariant mass of the
QQ final state
|mQQ MKK | VKK .
(5)
To further reduce the backgrounds, we also impose the following cuts on the transverse momenta
of the two final jets and their rapidity
Q,Q
pT
200 GeV,
|Q,Q | 2
(6)
as in the signal, the pT of the jets is peaked close to 12 MVKK and the production is central,
while in the background, the jets are peaked in the forward and backward directions and the
bulk of the cross section is for low pT jets. These cuts can certainly be improved to optimize
the signal to background ratio but we will refrain from performing such an optimization here.
In fact, in this preliminary and simple parton-level investigation, we will simply compare the
signal and the main corresponding physical background to determine if, grossly, the exchanges
of the KK states could give rise to a promising and possibly detectable signal. A more detailed
Monte Carlo study, including precise experimental effects, initial and final state radiation, the
correct identification of the t/b states and the measurement of their momenta, more efficient
and realistic cuts and detection efficiencies, which is beyond our scope here, will be needed.
Nevertheless, the large significances that we will present are expected to remain at an interesting
level after implementation of these experimental effects.
3.2. The cross sections
The invariant mass distributions d/dmt t of the process pp t t, are shown in Fig. 2 for
the four scenarios E1 to E4 ; the cuts on the transverse momenta and the rapidity of the final
t -jets given in Eq. (6) have been applied. Shown are the distributions in the case of the SM
background, qq/gg t t, in the case where the KK excitation of the gluon is also exchanged,
q q g, g (1) t t, and in the case where all the KK excitations of the electroweak gauge bosons
are also exchanged q q g, g (1) , (1) , Z (1) , Z (1) t t. As already mentioned, we have used a
common mass for the KK gauge bosons, MKK = 3 TeV, with the couplings given in Table 1 that
t t branching ratios and total decay widths displayed in Table 2.
lead to the bb,
As can be seen from the figure, there is a substantial contribution of the KK excitations to the
invariant mass distribution, in particular around the peak mt t 3 TeV. At higher invariant masses
the KK contribution becomes small, while at invariant masses lower than MKK it is significant
even for mt t 2 TeV; only for mt t 1 TeV the KK contribution becomes negligible. Outside
the KK mass peak, the RS effect is mostly due to the interference between the excited state and
SM contributions; this interference is positive below and negative above the peak [where the real
part of the propagator of the KK state changes sign]. The dominant contribution compared to the
SM case is by far due to the exchange of the excitation of the strongly interacting gluon which
has the largest (QCD versus EW) couplings to the initial state partons. The contributions of the
KK photon and KK Z boson increase the peak only slightly. In turn, the KK excitation of the Z
boson has a negligible impact on t t production as it does not couple to the initial light quarks,
Q (clight ) = 0, and the parton density of the heavier bottom quark in the proton is small.
11
(a)
(b)
(c)
(d)
Fig. 2. The invariant mass distribution of the cross section for the pp t t process for all four scenarios E1,2,3,4 defined
in Table 1. We use the CTEQ5M1 set of parton distribution functions and include the cuts of Eq. (6) on the pTt,t and t,t of
the final quarks. The solid curve is for the SM background and the long-dashed one for the RS model with contributions
of all the first KK excitations of the gauge bosons g (1) , (1) , Z (1) included; the short-dashed curve is for the SM case
along with only g (1) exchange; the dotted curve is for the SM case along with only (1) exchange; the dasheddotted
curve is for the SM case along with only Z (1) exchange.
The total cross sections of the pp t t process are displayed in Table 3 first for the SM
background only and then in the RS model in which all contributions of the KK excitations
[gluon, photon and Z boson] are added; the cuts on the invariant mass, transverse momentum
and rapidity of the t and t jets, Eqs. (5) and (6), are applied. The significance of the full signal
with a common mass, MKK = 3 TeV, is also shown for the integrated luminosities L = 10 and
100 fb1 . As can be seen, the significance of the excess of events in the RS scenarios when all
12
Table 3
Left: The pp t t cross sections [in fb] in the SM and in the RS model with the exchange of all the KK excitations as
well as the significance of the signal for L = 10 fb1 and L = 100 fb1 . Right: the significance for L = 100 fb1 of
the signal when only the separate contributions of the KK excitations of the gluon, photon and Z bosons are added to the
SM contribution. These numbers are obtained after the cuts of Eqs. (5) and (6) are applied
E1
E2
E3
E4
g (1)
(1)
(1)
SM
SM + RS
RS
S10
RS
S100
S100
S100
Z
S100
47.0
90.5
16.6
34.1
135
181
83.4
88.7
41
30
52
30
128
95
164
94
124
91
157
88
5.7
5.2
7.8
6.4
7.4
7.0
7.0
6.4
RS 30 for a moderate luminosity and

KK excitations are included is large in the four cases, S10
RS 90 for an high luminosity.
S100
To illustrate the impact of each KK state separately, the right-hand side of Table 3 shows the
significance of the signal in the case where only one KK excitation is exchanged in the pp t t
process, that is, when the mass of the two other excitations is (artificially) set to high values.
As one might have expected, since the excess over the SM background is mainly due to the exg (1)
change of the KK gluon, the significance in this case, S100 , is almost the same as when the full
RS . In the case where only the first KK excitation of the photon or the Z
signal is considered, S100
boson is considered, the significance is much smaller. This is a mere consequence of the fact that
the electroweak couplings of the (1) , Z (1) bosons are much smaller than the QCD couplings of
the g (1) , leading to limited production cross sections. The smaller rates are, however, partly compensated by the smaller total decay widths and one obtains in some scenarios significance at high
(1) ,Z (1)
luminosities that are large enough for the effects to be detectable at the LHC, S100
5. For
our 4 sets of parameters, the contributions of the neutral EW KK gauge bosons represent 47
above the KK gluon contribution (compare the fourth and fifth column in Table 3), indicating
that these contributions are significant. Nevertheless, experimentally the KK gluon would be difficult to distinguish from other KK gauge boson since their resonant mass are close, relatively
to their large widths, and their respective absolute coupling to quarks depends on the unknown c
parameters.
The significance of the signal strongly depends on the chosen scenario which fixes the couplings of the KK excitations to the t, b quarks. If the couplings to top quarks are large, the
pp VKK t t production rate is substantial but the total decay width is also large leading to
a more diluted signal. The signal is even more diluted when the couplings of the KK excitations
to b quarks is at the same time large: on the one hand, this increases the total decay width of
the excitation and, on the other hand, it lowers the branching ratio into the t t final states that we
are looking at. Taking the example of the g (1) state, the best significance is obtained in scenario
E3 in which the couplings to the b quarks are not too strongly enhanced (see Table 1) so that
the g (1) total decay width is the smallest, g (1) 330 GeV (see Table 2), and the branching ratio
for decays into top quarks is among the largest ones, BR(g (1) t t ) 83%. In scenario E1 ,
where the parameters are as in scenario E3 except for the Q(ctR ) and Q (ctR ) charges which are
larger, the total decay width of g (1) doubles, g (1) 630 GeV, while the fraction for decays into
top quarks is not significantly affected, BR(g (1) t t ) 91%, leading to a smaller significance,
RS = 41 versus S RS = 52. In scenarios E and E , the total decay widths of g (1) are larger and
S10
2
4
10
the t t branching ratios smaller so that the significance of the signals is reduced compared to the
two previous scenarios.
13
We have been focusing so far on MKK = 3 TeV (in particular in Table 3) which is the lower
bound set by precision EW measurements. Here, let us describe quantitatively the effect on the
t t signal of increasing MKK up to 3.5 TeV, namely slightly above the limit. In this case, the
excess of events in the t t production remains observable at high luminosity (100 fb1 ) as the
significance of the whole RS signal is reduced only down to 81, 59, 103 and 58 for scenarios
E1, . . . , E4 respectively. This decrease in the significance is obviously due to both the enhancement of all KK gauge boson widths (making the shape of resonance less prominent over the SM
background) and the deterioration of the total cross section (the dominant resonant contribution
occurring at larger, and thus disfavoured, values of initial parton momenta).
The discussion for the production process pp bb is quite similar to that of pp t t except
for the fact that the t -channel contribution to the bb bb subprocess has to be added. The
smallness of the parton distribution function of the bottom quark can be partly compensated by
the potentially larger KK gauge couplings of the bL,R states [for small cbR values as shown
in Table 1] compared to light quarks. For instance, this bb contribution reaches the level of
10% in the SM + RS cross section in scenario E4 . However, this contribution does not peak
at a given invariant mass value. The invariant mass distribution d/dmbb for bottom quark pair
production is shown in Fig. 3 for mbb between 2 and 4 TeV. As previously, the cuts on the
transverse momenta and rapidity of the final jets given in Eq. (6) have been applied. Here, we
simply show the SM background and the signal excess in the case where the contributions of
all KK excitations are simultaneously included; again, this excess is largely dominated by the
exchange of the KK excitation of the gluon.
The signals are less striking than in the pp t t case, the main reason being due to the
fact that the branching ratios BR(g (1) bb ) are much smaller than BR(g (1) t t ), due to the
hierarchy between the b and t couplings (see Table 1), except in scenario E4 and, to a lesser extent
scenario E2 , in which they are comparable. The cross section in the SM and in the RS model, as
well as the significance of the signal for the two possible luminosities L = 10 and 100 fb1 , are
shown in Table 4 in the four scenarios when the cuts of Eqs. (5) and (6) are applied. One can see
RS 5, to allow for the detection of the signal
that in all cases the significance is large enough, S10
even at moderate luminosities. For example, in the parameter set E4 , the cbR value is smaller
than for E3 so that the KK gauge coupling to b is larger, leading to higher bb production cross
section and significance. Comparing now E4 to E2 , ctR is larger which induces smaller widths
for KK gauge bosons and thus a better significance, for same reasons as before.
Note that when combining the pp t t and pp bb processes, one would in principle be
able to have access to the couplings of the KK states g (1) to top and bottom quarks. However,
it would have only a small contamination from the various overlapping electroweak KK resonances, since the major part of the signal is due to the contribution of the gluon excitation. As
the couplings are parity violating, an additional and important information is provided by some
polarization and forwardbackward asymmetries to which we turn our attention now.
3.3. Polarization and forwardbackward asymmetries
Each of the KK excitations of the gauge bosons, g (1) , (1) and Z (1) , has a different coupling to
the right- and left-handed top quarks [which are themselves different from the SM ones]. These
couplings appear in the forwardbackward asymmetry as well as in the polarization of the produced top quarks. While the enhancement in the production rate due to a single KK gauge boson
VKK 2
VKK 2
) + (gQ
) , the polarization and forward
is proportional to the sum of squared couplings (gQ
L
R
VKK 2
VKK 2
backward asymmetries are proportional to the difference (gQ
) (gQ
) . Thus, a combined
L
R
14
(a)
(b)
(c)
(d)
Fig. 3. The invariant mass distribution of the cross section for the pp bb process for all four parameter sets E1,2,3,4
defined in Table 1. We use the CTEQ5M1 set of parton distribution functions and include the cuts of Eq. (6) on the pTb,b
and b,b of the final quarks. The solid curve is for the SM background and the long-dashed one for the RS model with
contributions of all the first KK excitations of the gauge bosons g (1) , (1) , Z (1) , Z (1) included.
measurement of the cross section together with asymmetries would determine the couplings of
the vector boson VKK . However, since the process is mediated by the exchange of several KK
gauge bosons with differing right- and left-handed couplings, it will not be possible to measure
these couplings for any of the electroweak KK excitations. This is particularly true as the major
is coming from g (1) , as the contribution from
contribution to the total rate for (pp QQ)
(1)
(1)

(1)
is relatively small. Nevertheless, the measurethe electroweak excitations , Z and Z
15
Table 4
The pp bb cross sections [in fb] in the SM and in the RS model with the exchange of all the KK excitations as well
as the significance of the signal for L = 10 fb1 and L = 100 fb1 ; the cuts of Eqs. (5) and (6) have been applied
E1
E2
E3
E4
SM
SM + RS
RS
S10
RS
S100
55.3
109
17.0
39.7
67.6
159
32.8
90.0
5.2
15
12
25
17
48
38
80
ment of the polarization and forwardbackward asymmetries for top quarks, on which we will
concentrate here, will be instrumental in establishing the presence of parity violating KK gauge
bosons.
The polarization of the produced top quarks is defined as:
N R NL
(7)
,
N R + NL
where NR and NL are the numbers of events with right- and left-handed helicity for the top quark
with a given set of kinematical cuts.9 The statistical error in the measurement of this polarization,
in terms of the rate and the LHC luminosity L, is given by

1
1
Pt =
(8)
1 Pt2 =
1 Pt2 .
N R + NL
L
The polarization of the top quark is shown in Fig. 4 in the case of the SM and for the RS scenario E1 as a function of the invariant mass mt t; again the cuts of Eqs. (6) have been applied and
the contributions of all virtual KK states has been included.
The polarization is a parity odd quantity and can assume a non-zero value only in the presence of parity violating interactions. In the SM case, the only source of parity violation is the
contribution of the Z boson, which leads to a small negative polarization of order 103 . The
dominant contribution for mt t near MKK is due to g (1) which has a larger coupling to tR than
to tL in the chosen RS scenarios; this leads to a large polarization roughly of the size
Pt =
(gtVRKK )2 (gtVLKK )2
(gtVRKK )2 + (gtVLKK )2
KK
GV+
(9)
coming from the q q fusion channel. However, there is a dilution of this polarization due to the
parity-conserving SM production of top pairs through gg fusion and q q annihilation with gluon
exchange. For instance, in the case of scenario E1 , one has for the first gluon excitation g (1) ,
g (1)
G+ = 0.86, while the degree of polarization, Pt = 0.505, is diluted by the SM QCD processes.
Similar results are obtained in the other scenarios and one has: Pt 0.37, 0.52, 0.38 in scenarios E2 , E3 , E4 , respectively. Furthermore, there is some subleading contribution from the (1)
and Z (1) states which is, for instance, visible as a small dip just before the peak in the mt t
distribution of the polarization Pt in Fig. 4.
9 We note here that even for extra-dimensional models in which the KK excitations have the same couplings as their
SM counterparts, the Z (1) state will always have parity violating couplings and hence, the produced top quarks will be
polarized. However, in this case, following the structure of the SM, the left chiral top quark has larger coupling to the
Z (1) excitation, leading to a negative polarization. On the other hand, in the RS model that we consider here, the right
chiral top quarks have larger couplings to the KK gauge bosons, leading to a positive polarization.
16
Fig. 4. The top quark polarization as a function of the invariant mass mt t in the reaction pp t t in the SM (solid) and
in the RS model (dashed) for the set E1 of parameters.
The polarization of the top quark is not a direct observable. Nevertheless, it is reflected for
instance in the angular distribution of the lepton coming from the decaying top quark, t bW
b with = e, . In the rest frame of the t -quark, the leptons issued from these decays are
isotropically distributed for unpolarized top quarks. For positive Pt , the lepton distribution is
peaked in the direction of the would be boost or the direction of quantization axis, and for
negatively polarized top quarks, the peak is in the opposite direction. In the laboratory frame,
the decay products of the fast moving top quark are collimated in the forward direction due to
the boost. Thus, for the positively polarized top quarks, there will be additional focusing of the
leptons in the direction of the top quark momenta. Accordingly, there is a defocusing of the
lepton distribution for the negatively polarized top quarks. Thus, the simplest quantity to look at
is the azimuthal distribution (or the focusing) of the leptons with respect to the top production
plane [35]. We thus define the following asymmetry in the laboratory frame10 :
A =
(cos > 0) (cos < 0)

,
(cos > 0) + (cos < 0)
(10)
where is the azimuthal angle of the decay leptons with respect to the t -quark production plane,
or with respect to the transverse momentum of the top quark. Again, the statistical error in the
measurement of the asymmetry and the corresponding significance is given by

1
A =
1 A2
L
and S =
ARS+SM
ASM

ASM
(11)
respectively. The values of the asymmetries in the various RS scenarios and the significance in
their measurements are shown in the right-hand side of Table 5.
Since we are dealing with heavy KK bosons, the top quarks are highly boosted in the center
of mass frame, which leads to highly collimated decay products even for unpolarized top quarks.
This explains the large value of the asymmetry, A 0.96, for the SM in the case of E1 cuts for
example; see Table 5. A modification of the top polarization changes the collimation of the decay
10 A similar polarization asymmetry (P
LR ) was studied in [29] but this one was defined in the rest frame of the top
quark.
17
Table 5
Left: The polarization and forwardbackward asymmetries, A and AtFB , for the process pp t t in the SM and in the
RS model with the exchange of all the KK excitations as well as the significance of their measurements for L = 10 fb1
and L = 100 fb1
E1
E2
E3
E4
ASM
ARS+SM
S10

S100
0.957
0.952
0.963
0.960
0.982
0.972
0.987
0.980
1.5
1.3
1.2
1.0
4.7
4.1
3.8
3.2
ASM
FB
0.00178
0.00737
0.00666
0.00407
RS+SM
AFB
S10FB
FB
S100
0.195
0.157
0.232
0.181
7.2
6.7
6.7
5.4
23
21
21
17
Fig. 5. The laboratory frame lepton asymmetry A versus the mt t invariant mass for the process pp t t in SM and in
the RS scenario E1 .
products and, thus, affects this SM asymmetry. For instance, again in scenario E1 , one expects
a large and positive polarization for the top quarks, which leads to a slightly larger asymmetry
value, A 0.98. The significance S calculated according to Eq. (11), indicates that this asymmetry should be observable at the LHC but only for a high luminosity run in a post-discovery
analysis [the discovery being clearly possible at a low luminosity] as shown in Table 5. Therefore,
the associated distribution displayed in Fig. 5 for the parameter set E1 , which exhibits a specific
shape, should help in identifying the RS scenario. In particular, the fact that this A asymmetry is
enhanced by the RS contribution proves that the exchanged KK gauge bosons couple effectively
more to the tR than to the tL state.
Let us now turn our attention to the forwardbackward asymmetry. The s-channel diagrams
exchanging gauge bosons with parity violating couplings lead to a forwardbackward asymmetry
in the top quark pair production in the center of mass frame with respect to the quark direction.11
However, since at the LHC there are two identical protons colliding, there is no simple forward
backward asymmetry in the laboratory frame due to the symmetric distribution of the initial
state quarks and anti-quarks. But, since the proton is mainly composed of quarks and with only a
small proportion of anti-quarks in the sea, the fusing q q pairs are highly boosted in the laboratory
frame, mainly in the direction of the incoming quarks. Thus, one can construct a kind of forward
11 The forwardbackward asymmetry is non-vanishing only if the vector coupling to both initial and final fermions are
parity violating.
18
Fig. 6. The laboratory frame asymmetry AtFB as a function of the invariant mass mt t for the reaction pp t t in the SM
and in the RS scenario E1 .
backward asymmetry with respect to the direction of the boost in the center of mass frame. In
the laboratory frame, we first arbitrarily label the proton beams as 1 and 2, and then define a
quantity

x 1 x2
C=
(12)
cos tlab
x 1 + x2
where xi are the momenta fractions of the colliding partons and tlab the t -quark polar angle with
respect to the beam 1; the quantity in the brackets is the boost of the center of mass frame in
the laboratory frame. With respect to this quantity we define a forwardbackward asymmetry
with respect to the direction of boost:
AtFB =
(C > 0) (C < 0)
.
(C > 0) + (C < 0)
(13)
This quantity is sensitive to any change in the angular distribution in the center of mass frame.
The statistical error in the measurement of this asymmetry is given by

2

1
AtFB =
(14)
1 AtFB .
L
We should note that the above asymmetry is not a pure probe of parity violation, i.e. it can
be non-zero and positive even for parity-conserving processes and can receive contribution even
from an additional parity-conserving channel. The reason for this is the boost of the center of
mass frame with respect to the quark direction, which leads to focusing of particles in the forward direction. This explains the large value of AtFB for the SM as shown in Fig. 6 by a solid line.
The presence of additional channel, i.e., VKK , leads to an increase in the number of events with
large transverse momentum. This change in the angular distribution in the c.m. frame leads to a
changes AtFB , which is shown in Fig. 6 by a dashed line for the RS scenario E1 . We see a significant contribution to AtFB from the g (1) excitation. Once more, there is some finite contribution
from the electroweak KK modes which appears as a nipple peak in the mt t distribution of AtFB ,
Fig. 6. However, the integrated contribution to AtFB from EW KK gauge bosons is of order 0.01
and is thus negligible compared to the KK gluon contribution (cf. Table 5). Hence, no correction
(a)
19
(b)
Fig. 7. The Feynman diagram for QQ (Q = b, t ) production, via VKK exchange, at the LHC through the quark loop induced gluongluon fusion process (a) and the counter-term with the Stckelberg mixing that regulates the gauge anomaly
of the ggVKK vertex (b).
due to EW KK gauge bosons is expected to exceed the statistical uncertainties on ARS

FB , even at
high luminosities.
We emphasize here that g (1) coupling to light quarks in the proton is parity conserving, hence
the change in AtFB signals the additional resonance and not that of parity violation. Though there
is small forwardbackward asymmetric contribution coming from Z (1) excitation appearing as
a nipple peak, it cannot be decoupled from the contribution coming from g (1) excitation. The
forwardness of the signal events can be used to separate signal from the background. The
obtained values of AtFB , and their significance given in Table 5 for all scenarios, show that it is
AFB
potentially visible at the LHC even at low luminosity with S10
5 and can be used to enhance
the event selection for the signal as compared to the background.
Note that we have also calculated the asymmetry AbFB for the pp bb production process, but
we do not present the corresponding results as the significance is rather low in all the considered
scenarios. Furthermore, we will not discuss the polarization of the produced bottom quarks as it
will not be experimentally measurable.
4. The higher order processes
4.1. The gluongluon fusion mechanism
The higher order gluongluon fusion contribution to the heavy quark pair production at LHC,
via KK gauge boson exchange, should be favored by the large VKK couplings to heavy quarks.
Such a reaction occurs through a loop involving mainly the heavy top and bottom quarks and a
VKK exchanged in the s-channel, as shown in Fig. 7(a). Indeed, the main reaction is via VKK =
g (1) which receives no contributions from the light quarks in the loop as those have quasi identical
g (1) couplings for the left and right chirality (fixed by a common Q() (clight ) value as explained
at the end of Section 2.2) which makes the final amplitude vanishing. This reaction represents
the dominant two-body production mechanism for KK excitations which do not couple to light
quarks.

The effective one-loop vertex g(k1 )g(k2 )VKK (k3 ), where k1 , k2 , k3 denote the respective
momenta (k1 and k2 being taken as incoming with respect to the flow in the loop), is given in
momentum space by,

ij k
= t ij k A1 (k1 , k2 ) B2 (k1 , k2 ) k1 k2

B1 (k1 , k2 )
+
k k k k1 ,
2k1 .k2 2 1 2
(15)
20
with,
t ij k = gs2 Tr[i j k ]

gqVL Q( ) (cqL ) gqVR Q( ) (cqR ) ,
(16)
gs being the strong interaction coupling constant, i [i = 1, . . . , 8] the Gell-Mann matrices (only
two such matrices must be taken in the case of colorless KK excitations: VKK = g (1) ),
the purely antisymmetric tensor and qL/R the quarks running in the loop. Finally, the A1 (k1 , k2 )
function, entering Eq. (15), represents the usual amplitude a priori divergent. The finite functions
B1,2 (k1 , k2 ) are defined later.
Indeed, the amplitude Eq. (15) contains an anomaly in the third current. As a matter of fact,
only two of the three Ward identities that the vertex must obey are satisfied12 :
ij k
k1
= 0,
ij k
k3
ij k
k2
= 0,

ij k
= k1 + k2
= t ij k 2 A1 (k1 , k2 ) B2 (k1 , k2 ) k2 k1 .
(17)
The first two relations allow to define the ambiguous amplitude A1 (k1 , k2 ) in terms of finite
functions:
B1 (k1 , k2 )
(18)
= 0.
2
However, the ggVKK vertex receives another contribution from the exchange of the gauge
field fifth component V5 , as drawn in Fig. 7(b). There, the ggV5 coupling originates from the
so-called ChernSimons term in the action (see Ref. [38] for the case of warped orbifolds) which
can be written in terms of the gauge covariant field strengths as [39,40]:

SCS = d 4 x dy (y) MNP QR

2
Tr AM FN P FQR + iAM AN AP FQR AM AN AP AQ AR ,
(19)
5
A1 (k1 , k2 ) B2 (k1 , k2 )
where the Roman letter indices M, N, . . . run over all dimensions. While the mixing between
(n)
(n)
A5 and A (the Greek letter index is running over the first four dimensions only), appearing in the diagram of Fig. 7(b), and proportional to the gauge KK mass, comes from the
Stckelberg term; see for instance Refs. [39,41]. Following Ref. [41], one finds that the diagram
of Fig. 7(b) gives rise to a vertex contribution of the form,

k1 + k2
ij k
= 2t ij k A1 (k1 , k2 )
k2 k1 .
CS
(k1 + k2 )2
(20)
The whole amplitude is then free from any gauge anomaly as one has now,

ij k
ij k
k1
= 0,
+
CS

ij k
ij k
k2 + CS = 0,

ij k
ij k
k3
= 2t ij k B2 (k1 , k2 ) k2 k1 ,
+
CS
12 See Refs. [36,37] for the generic computations of gauge anomalies in the context of S 1 /Z orbifold models.
2
(21)
21
Fig. 8. The mt t distribution of the cross-section corresponding to the reaction shown in Fig. 7 for VKK = g (1) for the E2
choice of parameters. The whole SM distribution is also shown for comparison.
where B2 is proportional to the mass of fermions running in the loop. Therefore, as expected,
there is a mechanism responsible for the cancellation of the gauge anomaly.
The B functions entering above equations read as,
B1 (k1 , k2 ) =
2
3(2)2
1
dz
0
2
B2 (k1 , k2 ) =
3(2)2
1
0
1z
dx
0
1z
dz
dx
0
x(x + z 1)
,
(m2f /(2k1 . k2 )) xz
(1 z)m2f
m2f (2k1 . k2 )xz
(22)
In conclusion, both the vertex contributions Eqs. (15) and (20) have to be taken into account
in the calculation of the quark contribution to the ggVKK vertex. Note that the correction Eq. (20)
vanishes in the on-shell regime limit for VKK where k3 3 = 0, 3 being its polarization vector.
This is consistent with Yangs theorem according to which the entire production cross section
vanishes if the two initial zero-mode gluons as well as the produced KK gauge boson are simultaneously on-shell. This is illustrated by the mt t invariant mass distribution that we will discuss
now.
The mt t distribution for the above said process is shown in Fig. 8, along with that for the
SM background, for the excitation g (1) and the E2 choice of RS parameters. The dip in the
distribution near the mass mt t = mKK is due to the vanishing coupling for on-shell VKK bosons
with a pair of gluons. The shape of the distribution depends upon the total decay width of the
VKK . However, the rates for the signals are several orders of magnitude smaller than the SM
background. The cross sections are even smaller for the KK excitations of the electroweak gauge
bosons. One can then conclude that the gluongluon fusion mechanism for the production of KK
excitations will not be relevant at the LHC.
22
(a)
(b)
Fig. 9. Typical Feynman diagrams for associated VKK production with Q , Q = t, b quarks at the LHC: the four-body
Q
(a) and the three-body gb bQQ
(b) processes.
gg QQQ
4.2. Associated production with heavy quarks

In this section, we briefly discuss the associated production of the KK excitations of the gauge
bosons with heavy quarks
gg/q q Q Q VKK Q Q QQ,

gb bVKK bQQ
(23)
(24)
Q/Q
stand for the final state heavy

where q stands for the light quarks present in the proton and
t and/or b quarks. Some generic Feynman diagrams are shown in Fig. 9. In the first case, a pair
of b or t quarks is produced mainly in the gg fusion subprocess with a small contribution from
q q annihilation, followed by the emission of a KK gauge boson from one of the heavy quark
lines, while in the second case a KK gauge boson is emitted from the b-quark line in gb fusion
with the initial b-quark taken from the proton sea. In both cases the produced KK gauge boson
decays predominantly into heavy quark pairs, leading to topologies with multi b and/or t final
states.
In the 2 4 body reaction of Eq. (23), since the SM production of quark pairs is dominated
by the gg fusion mechanism, the largest contribution to associated production of KK gauge
bosons with heavy quarks comes also from the same channel. After radiation from the heavy
quark line, VKK can either decay into a heavy quark pair of the same flavour as it was produced
in association with, or of the other heavy flavour. To avoid the combinatorial background, we
KK with
choose to look at t t + bb final states which originate from the two channels pp bbV
Since the gg bb cross section is much larger
VKK t t and pp t tVKK with VKK bb.
than the pp t t cross section and that the t t branching ratio of VKK is in general larger than
the bb branching ratio, the final t tbb sample from the signal is dominantly originating from the
KK bbt
t.
subprocess pp bbV
As for the kinematics of this production channel, the KK gauge boson is produced almost at
rest owing to its large mass and hence decays into top quarks with a wide range of transverse
momenta with an excess of events with large pT . On the other hand, the b quarks produced in the
hard process of gg fusion dominantly have small transverse momenta. The picture is reversed for
the second channel pp t tVKK t tbb which, as already mentioned, gives a smaller contribu
tion to the signal cross section. For the QCD background that we have considered, pp t tbb,
and that we have calculated using the code CompHEP [42], the dominant contribution comes
from the subprocesses gg/q q t t production followed by the emission of a virtual gluon splitting into a bb pair. To reduce this QCD background without significantly affecting the signal, we
impose the following set of kinematical cuts
pTt,t 300 GeV,
pTb,b 100 GeV,
|Q,Q | 2.
(25)
23
Table 6
The signal cross section [fb], the signal to background ratio and the signal statistical significance for L = 100 fb1 for
the four-body process pp t tbb in the RS scenarios E1,2,3,4 after the cuts discussed in the text are applied
E1
E2
E3
E4
RS (pp t tbb)
RS / SM
ttbb
S100
0.06
0.26
0.04
0.16
0.13
0.58
0.08
0.35
0.8
3.9
0.6
2.3
We have used large cuts on the top quarks transverse momenta to reduce the background as
much as possible; one can further optimize these cuts to improve the significance of the signal,
but as the number of events will turn out to be rather small, we do not perform that analysis here.
Additionally, we impose an opening angle cut bb 60 which reduces the gluon splitting into
bb in the background by more than an order of magnitude and does not affect the signal much.
The cross-sections for t tbb production are shown in Table 6 in all four RS scenarios E1 to E4 .
The numbers for the RS model correspond to the signal only and do not include the background
events or the interference with the background amplitudes [which should occur in the case where
the KK excitation of the gluon is produced]. The rather low cross sections for signal events,
compared to a SM background of SM 0.4 fb, indicate that even with a high luminosity of
L = 100 fb1 , one will not be able to reach a significance of 5 . The best significance, 4 ,
is obtained in the case of scenario E2 . Hence, no discovery is possible with the t tbb channel of
associated production, if a reasonable luminosity is assumed. We expect the trend for the other
topologies to be similar. Note that here, we considered only the production of the KK excitation
of the gluon; the significance for the production of the KK excitations of the electroweak gauge
bosons [and in particular of the Z (1) which cannot be produced in the DrellYan processes
discussed in Section 3] is even smaller.
Next we turn to the associated production of the KK excitations with a b-quark in gb fusion.
Though the density of b/b quarks in the proton is small, the large coupling of the b-quark with the
KK gluon for instance could partially compensate for it. Looking at the reaction gb bg (1)
bt t, where b stands for both the b and the anti-b quark, we use the same kinematical cuts as those
introduced in Eq. (25). In fact, because now we are dealing with a 2 3 production process,
the signal cross sections are expected to be one order of magnitude larger than that of the 2 4
process discussed previously. Here again, the b quark produced in the hard process has low transverse momentum, while the top quarks can have large transverse momenta in the signal events.
The irreducible QCD background bg bt t, that we calculate again using CompHEP [42], is
also more than one order of magnitude larger compared to the previous process, SM 9 fb.
The signal production cross sections for a top quark pair and a b-type jet with the kinematical
cuts of Eq. (25) is shown in of Table 7 for the pure RS signal in the four selected scenarios
E1,...,4 . Again the points E2 and E4 have large signal cross sections owing to the large coupling
of the b quarks to the KK gluon. For the other two points, the couplings of b quark are small and,
hence, lower production rates of the KK gluon are obtained. With an integrated luminosity of
L = 100 fb1 , one obtains a significance which is larger than 8 for scenarios E2 and E4 , while
it is about 3 or less for the other two scenarios. We again have considered only the associated
production of a KK gluon and not accounted for the interference between the RS signal and the
background.
Thus, at least in the 3-body production process gb bt t, the cross section in some RS scenarios can be significantly larger than the QCD background cross section. However, to reach a
24
Table 7
The signal cross section [fb], the signal to background ratio and the signal statistical significance for L = 100 fb1 for
the three-body process gb t tb/t tb in the four RS scenarios E1,2,3,4 after the cuts discussed in the text are applied
E1
E2
E3
E4
RS (pp t tb/t tb)
RS / SM
ttb
S100
0.91
4.12
0.61
2.50
0.10
0.47
0.07
0.28
3.0
13.8
2.0
8.4
signal significance which is important, large integrated luminosities, L 100 fb1 , are needed.
In this case, the top and bottom quark jets cannot be identified with an excellent efficiency and
more background processes with potentially much larger rates need to be considered. Detailed
experimental analyzes need thus to be performed in order to assess the viability of this signal.
5. Conclusions
In the framework of the RandallSundrum extra-dimensional model with SM fields in the
bulk, we have studied the production of the KK excitations of gauge bosons at LHC which then
dominantly decay into heavy top and bottom quark pairs. For illustration, we have considered a
specific version of the model in which the SM gauge group is enhanced to a leftright symmetry
and concentrated on quark geometrical localizations and gauge quantum numbers that allow
to solve the LEP anomaly on the forwardbackward asymmetry AbFB . We have selected some
characteristic points of the parameter space of this model and assumed masses MKK 3 TeV
for the KK excitations of the gluon and the electroweak gauge bosons as dictated by constraints
from high-precision electroweak data.
We have shown that the contribution of the DrellYan process q q VKK t t to the pp
q q/gg
t t cross section, in which we have taken into account the total width of the excitation
as well as the interference between the signal and SM the background, can be substantial and
would allow for the detection of the gluon KK excitation with a high statistical significance; the
signal significance is smaller for the excitations of the electroweak gauge bosons as a result of the
smaller couplings. Besides the invariant mass distributions and the sizeable total widths of the
KK resonances, the polarization of the top quark and the forwardbackward asymmetry would
allow to characterize the signal, while the polarization would be instrumental to probe the chiral
structure of the fermion couplings of the KK excitations. In the case of the q q VKK bb
process, a large signal significance compared to the SM background is also obtained.
We have also calculated the cross sections for the higher order processes in which the KK
excitations couple only to the heavy top and bottom quarks and compared them to their corresponding irreducible SM backgrounds. We have first studied the gluongluon fusion mechanism,
which is induced by a heavy quark loop and in which the anomalous ggVKK
gg VKK t t, bb,
vertex has to be regulated via the Stckelberg mixing term. We have then analyzed the four-body
t tt t and bbb
b as well as the three-body reactions gb bt t and bbb,
in
reactions pp t tbb,
which the KK excitations are radiated from the heavy quark lines. The cross sections for these
higher order processes at the LHC, except potentially for the bg g (1) b process, turn out to be
too small compared to the SM background, to allow for a detection of heavy KK excitations with
MKK 3 TeV.
In our parton-level analysis, we took into account only the irreducible SM QCD background
and did not attempt to include experimental effects such as detection efficiencies, b-quark tag-
25
ging, reducible light jet backgrounds, etc. Only refined and realistic Monte Carlo analyses, which
take into account all these effects, would allow to assess the viability of the KK gauge boson
signals. These experimental analyses are currently being performed by some members of the ATLAS and CMS Collaborations [43] including in particular the analysis on identification of highly
boosted top and bottom quarks.
Acknowledgements
We thank Fawzi Boudjema and Emilian Dudas for useful discussions on the anomaly cancellation in the gg VKK process, as well as Rohini Godbole and Franois Richard for useful
suggestions. This work is supported by the CEFIPRA project No. 3004-B and by the French
ANR project PHYS@COL&COS.
References
[1] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370;
M. Gogberashvili, Int. J. Mod. Phys. D 11 (2002) 1635.
[2] A. Pomarol, Phys. Rev. Lett. 85 (2000) 4004;
L. Randall, M.D. Schwartz, JHEP 0111 (2001) 003;
L. Randall, M.D. Schwartz, Phys. Rev. Lett. 88 (2002) 081801;
W.D. Goldberger, I.Z. Rothstein, Phys. Rev. D 68 (2003) 125011;
K. Choi, I.W. Kim, Phys. Rev. D 67 (2003) 045005;
K. Agashe, A. Delgado, R. Sundrum, Ann. Phys. 304 (2003) 145.
[3] K. Agashe, G. Servant, Phys. Rev. Lett. 93 (2004) 231805;
K. Agashe, G. Servant, JCAP 0502 (2005) 002;
G. Blanger, A. Pukhov, G. Servant, hep-ph/0706.0526.
[4] T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141.
[5] S.J. Huber, Q. Shafi, Phys. Lett. B 498 (2001) 256;
S.J. Huber, Q. Shafi, Phys. Lett. B 544 (2002) 295;
S. Chang, et al., Phys. Rev. D 73 (2006) 033002;
G. Moreau, J.I. Silva-Marcos, JHEP 0601 (2006) 048.
[6] Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361;
T. Appelquist, et al., Phys. Rev. D 65 (2002) 105019;
T. Gherghetta, Phys. Rev. Lett. 92 (2004) 161601;
G. Moreau, Eur. Phys. J. C 40 (2005) 539.
[7] M. Guchait, F. Mahmoudi, K. Sridhar, hep-ph/0703060.
[8] B. Lillie, J. Shu, T.M.P. Tait, arXiv: 0706.3960 [hep-ph].
[9] G. Burdman, Phys. Rev. D 66 (2002) 076003;
C.S. Kim, et al., Phys. Rev. D 67 (2003) 015001;
J.L. Hewett, F.J. Petriello, T.G. Rizzo, JHEP 0209 (2002) 030;
S.J. Huber, Q. Shafi, Phys. Rev. D 63 (2001) 045010;
S.J. Huber, C.A. Lee, Q. Shafi, Phys. Lett. B 531 (2002) 112;
C. Csaki, et al., Phys. Rev. D 66 (2002) 064021.
[10] K. Agashe, A. Delgado, M.J. May, R. Sundrum, JHEP 0308 (2003) 050.
[11] M. Carena, E. Pontn, T. Tait, C. Wagner, Phys. Rev. D 67 (2003) 096006;
M. Carena, et al., Phys. Rev. D 68 (2003) 035010;
M. Carena, et al., Phys. Rev. D 71 (2005) 015010;
F. del Aguila, M. Perez-Victoria, J. Santiago, JHEP 0302 (2003) 051;
M. Carena, T. Tait, C. Wagner, Acta Phys. Pol. B 33 (2002) 2355.
[12] A. Delgado, A. Falkowski, hep-ph/0702234.
[13] K. Agashe, R. Contino, L. Da Rold, A. Pomarol, Phys. Lett. B 641 (2006) 62.
26
[14] A. Djouadi, G. Moreau, F. Richard, Nucl. Phys. B 773 (2007) 43.

[15] LEP Collaborations (ALEPH, DELPHI, L3 and OPAL Collaborations), LEP Electroweak Working Group, SLD
Heavy Flavour Group, A combination of preliminary electroweak measurements and constraints on the Standard
Model, Phys. Rep. 427 (2006) 257, hep-ex/0509008, http://lepewwg.web.cern.ch/LEPEWWG.
[16] For a review, see K. Monig, Rep. Prog. Phys. 61 (1998) 999.
[17] A. Djouadi, J. Khn, P.M. Zerwas, Z. Phys. C 46 (1990) 411.
[18] G. Cacciapaglia, C. Cski, G. Marandella, J. Terning, Phys. Rev. D 75 (2007) 015003;
M. Carena, E. Pontn, J. Santiago, C. Wagner, Nucl. Phys. B 759 (2006) 202;
M. Carena, E. Pontn, J. Santiago, C. Wagner, hep-ph/0701055;
M. Carena, et al., arXiv: 0712.0095 [hep-ph].
[19] S.J. Huber, Nucl. Phys. B 666 (2003) 269;
G. Moreau, J.I. Silva-Marcos, JHEP 0603 (2006) 090.
[20] K. Agashe, et al., Phys. Rev. Lett. 93 (2004) 201804;
K. Agashe, et al., Phys. Rev. D 71 (2005) 016002.
[21] Z. Ligeti, M. Papucci, G. Perez, Phys. Rev. Lett. 97 (2006) 101801;
K. Agashe, et al., hep-ph/0509117;
K. Agashe, et al., Phys. Rev. D 74 (2006) 053011;
K. Agashe, et al., Phys. Rev. D 75 (2007) 015002.
[22] F. Ledroit, G. Moreau, J. Morel, hep-ph/0703262.
[23] E. De Pree, M. Sher, Phys. Rev. D 73 (2006) 095006.
[24] K. Agashe, et al., hep-ph/0701186;
A. Liam Fitzpatrick, et al., hep-ph/0701150.
[25] C. Dennis, et al., hep-ph/0701158.
[26] For GUTs Z physics at hadron colliders, see for instance: J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193;
A. Leike, Phys. Rep. 317 (1999) 143;
For a more recent analysis, see: M. Dittmar, et al., Phys. Lett. B 583 (2004) 111.
[27] H. Davoudiasl, J. Hewett, T. Rizzo, Phys. Rev. D 63 (2001) 075004.
[28] B. Lillie, L. Randall, L.-T. Wang, hep-ph/0701166.
[29] K. Agashe, et al., hep-ph/0612015.
[30] D. Choudhury, R.M. Godbole, R.K. Singh, K. Wagh, arXiv: 0705.1499 [hep-ph].
[31] C.A. Scrucca, M. Serone, Int. J. Mod. Phys. A 19 (2004) 2579.
[32] T. Han, G. Valencia, Y. Wang, Phys. Rev. D 70 (2004) 034002;
T. Han, G. Valencia, Y. Wang, Int. J. Mod. Phys. A 20 (2005) 3214.
[33] K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165.
[34] J. Pumplin, D. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.K. Tung, CTEQ Collaboration, JHEP 0207 (2002) 012.
[35] R.M. Godbole, S.D. Rindani, R.K. Singh, JHEP 0612 (2006) 021.
[36] P. Bintruy, C. Deffayet, E. Dudas, P. Ramond, Phys. Lett. B 441 (1998) 163.
[37] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 516 (2001) 395.
[38] T. Hirayama, K. Yoshioka, JHEP 0401 (2004) 032.
[39] C.T. Hill, Phys. Rev. D 73 (2006) 085001.
[40] C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169.
[41] P. Anastasopoulos, M. Bianchi, E. Dudas, E. Kiritsis, JHEP 0611 (2006) 057.
[42] A. Pukhov, et al., hep-ph/9908288.
[43] G. Brooijmans, et al., for ATLAS Collaboration, C.H. Shepherd-Themistocleous, et al., for CMS Collaboration,
work in progress. We thank these authors for discussions about these processes during the Les Houches Workshop.

Electroweak corrections to hadronic production

of W bosons at large transverse momenta
Johann H. Khn a , A. Kulesza b , S. Pozzorini c , M. Schulze a,
a Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
b Deutsches Elektronen-Synchrotron DESY, Notkestrasse 85, D-22607 Hamburg, Germany
c Max-Planck-Institut fr Physik, Fhringer Ring 6, D-80805 Munich, Germany
Received 3 August 2007; accepted 14 December 2007

Available online 5 January 2008
Abstract
To match the precision of present and future measurements of W -boson production at hadron colliders
electroweak radiative corrections must be included in the theory predictions. In this paper we consider their
effect on the transverse momentum (pT ) distribution of W bosons, with emphasis on large pT . We evaluate
the full electroweak O() corrections to the processes pp W + jet and p p W + jet including virtual
and real photonic contributions. We present the explicit expressions in analytical form for the virtual corrections and provide results for the real corrections, discussing in detail the treatment of soft and collinear
singularities. We also provide compact approximate expressions which are valid in the high-energy region,
2 . These expressions dewhere the electroweak corrections are strongly enhanced by logarithms of s /MW
scribe the complete asymptotic behaviour at one loop as well as the leading and next-to-leading logarithms
at two loops. Numerical results are presented for protonproton collisions at 14 TeV and protonantiproton
collisions at 2 TeV. The corrections are negative and their size increases with pT . At the LHC, where transverse momenta of 2 TeV or more can be reached, the one- and two-loop corrections amount up to 40%
and +10%, respectively, and will be important for a precise analysis of W production. At the Tevatron,
transverse momenta up to 300 GeV are within reach. In this case the electroweak corrections amount up to
10% and are thus larger than the expected statistical error.
PACS: 12.15.-y; 12.15.Ji; 12.15.Lk
Keywords: Gauge boson; W ; Transverse momentum; High energy behavior; Correction, higher-order
E-mail address: schulze@particle.uni-karlsruhe.de (M. Schulze).

doi:10.1016/j.nuclphysb.2007.12.029
28
J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777
1. Introduction
After the startup of the Large Hadron Collider (LHC) hard scattering reactions will be explored with high event rates and momentum transfers up to several TeV. In order to identify new
phenomena in this region, the predictions of the Standard Model have to be understood with
adequate precision.
The study of gauge-boson production has been among the primary goals of hadron colliders,
starting with the discovery of the W and Z bosons more than two decades ago [1]. The investigation of the production dynamics, strictly predicted by the electroweak theory, constitutes one
of the important tests of the Standard Model. Differential distributions of gauge bosons, in rapidity as well as in transverse momentum (pT ), have always been the subject of theoretical and
experimental studies. This allows to search for and set limits on anomalous gauge-boson couplings, measure the parton distribution functions and, if understood sufficiently well, use these
reactions to calibrate the luminosity. For gauge-boson production at large pT the final state of
the leading-order process consists of an electroweak gauge boson plus one recoiling jet. Being,
in leading order, proportional to the strong coupling constant, these reactions could also lead to
a determination of S in the TeV region.
The high center-of-mass energy at the LHC in combination with its enormous luminosity will
allow to produce gauge bosons with transverse momenta up to 2 TeV or even beyond. In this
kinematic region the electroweak corrections are strongly enhanced, with the dominant terms
2 ), next-toin L-loop approximation being leading logarithms (LL) of the form L log2L (s /MW
2L1
2
L
leading logarithms (NLL) of the form log
(s /MW ), and so on. These corrections, also
known as electroweak Sudakov logarithms, may well amount to several tens of percent [211].
(A recent survey of the literature on electroweak Sudakov logarithms can be found in Ref. [12].)
Specifically, the electroweak corrections to the pT -distribution of photons and Z bosons at hadron
colliders were studied in Refs. [710]. In Refs. [810], it was found that at transverse momenta
of O(1 TeV) the dominant two-loop contributions to these reactions amount to several percent
and must be included to match the precision of the LHC experiments. This is quite different
from the production of on-shell gauge bosons with small transverse momenta [13], where the
electroweak corrections are not enhanced by Sudakov logarithms. With this motivation in mind
we study the electroweak corrections to hadronic production of W bosons in association with a
()
jet, pp Wj , at large pT .
As a consequence of the non-vanishing W charge, QED corrections cannot be separated from
the purely weak ones and will thus be included in our analysis. Thus, in comparison with Z-boson
production, several new aspects arise. Real photon emission must be included to cancel the infrared divergencies from virtual photonic corrections. Collinear singularities, a consequence of
radiation from massless quarks, must be isolated and absorbed in the parton distribution functions (PDFs) in the case of initial-state radiation. We regularize soft and collinear singularities
in two different schemes: using small quark and photon masses which are set to zero at the end
of the calculation and, alternatively, dimensional regularization. In events with real radiation, the
pT of the W boson is balanced both by the pT of the recoiling parton (quark or gluon) and the
photon. Configurations involving a small-pT parton and a hard photon are better described as
W final states. We thus define the Wj cross section imposing a lower limit on the jet transverse momentum, which is chosen independent of the W -boson pT . In order to avoid final-state
collinear singularities, we recombine collinear photonquark final states.
29
The virtual EW corrections to Wj production are formally connected with the real emission of W and Z bosons, which leads to W Vj final states with V = W, Z. Both contributions
are of O( 2 S ). If integrated over the full phase space, the real emission of gauge bosons produces large Sudakov logarithms that partially cancel those resulting from virtual gauge bosons.
However, in exclusive measurements of pp Wj , the available phase space for gauge boson
emission is strongly suppressed by the experimental cuts. We thus expect that real emission provides relatively small contributions while the bulk of electroweak effects originates from virtual
corrections. In fact, for pp Zj it was shown that, in presence of realistic (and relatively less
exclusive) experimental cuts, the contribution of real emission is about five times smaller than the
virtual corrections [11]. Moreover, real emission can be further reduced with a veto on additional
jets, which suppresses multiple-jet events resulting from the hadronic decay of the radiated gauge
bosons. Therefore we will restrict ourselves to the investigation of virtual electroweak corrections
(and photon bremsstrahlung). The real emission of W and Z bosons can be non-negligible and
certainly deserves further detailed studies, however we do not expect a dramatic impact on our
results.
W q ( ) with q =
The partonic reactions qq
W g( ), q g W q( ) and qg
u, d, s, c, b are considered. All of them are, however, trivially related by CP- and crossingsymmetry relations. Quark-mass effects are neglected throughout, which allows to incorporate
the effect of quark mixing through a simple redefinition of parton distribution functions (see Section 2.1). Our conventions for couplings, kinematics and two- as well as three-body phase space
are also collected in Section 2. The calculation of the virtual corrections is described in Section 3.
We present analytic expressions for the one-loop amplitude, specify the counterterms in the G
renormalization scheme and isolate the infrared singularities. The high-energy limit is studied
2 , keeping
in detail in Section 4. The analytic one-loop result is investigated in the limit s MW
quadratic and linear logarithms as well as constant terms. These results are compared to those
derived in the NLL approximation [14]. In view of their numerical importance we also derive
the dominant (NLL) two-loop terms, using the formalism of Refs. [15,16]. The calculation of
the real corrections is performed using the dipole subtraction formalism [1719], as discussed in
Section 5. The checks which we carried out in order to ensure the correctness of the results are
described in Section 6.
The numerical results are presented in Section 7. After convolution with parton distribution
functions, we obtain radiatively corrected predictions for pT -distributions of W bosons at the
LHC and the Tevatron. The quality of the one-loop NLL and NNLL approximations is investigated and the size of the dominant two-loop terms is compared with the expected statistical
precision of the experiments. Concerning perturbative QCD, our predictions are based on the
lowest order. To obtain realistic absolute cross sections, higher-order QCD corrections [20] must
be included. However, the relative rates for W + , W and Z production are expected to be more
stable against QCD effects. Therefore, the impact of the electroweak corrections on these ratios is
presented in Section 7. Our conclusions and a brief summary can be found in Section 8. Explicit
analytic results are collected in Appendices AE.
A short description of the method of calculation and the main results for LHC have been
given in Ref. [21]. After completion of this work, Hollik, Kasprzik and Kniehl [22] reported
results on hadronic W -boson production at large pT qualitatively similar to those of Ref. [21]
and the present paper.
30
2. Definitions and conventions

2.1. Hadronic cross section
The pT -distribution of W bosons in the reaction h1 h2 W j ( ) is given by

d abW k( )
d h1 h2
=
dx1 dx2 (x1 x2 min )fh1 ,a x1 , 2 fh2 ,b x2 , 2
,
dpT
dpT
1
a,b,k 0
(1)
where min depends on the kinematic configuration of the final state and is specified at the end
of Section 2.2. The indices a, b denote initial-state partons and fh1 ,a (x, 2 ), fh2 ,b (x, 2 ) are the
corresponding parton distribution functions (PDFs). abW k( ) is the partonic cross section for
the subprocess ab W k( ). The sum in (1) runs over all a, b, k combinations corresponding
to the subprocesses
dn um W + g( ),
um g W + dn ( ),
um dn W + g( ),
dn g W + u m ( ),
W+
gum W + dn ( ),
g dn W + u m ( ),
(2)
production, and similarly for

production.
The dependence of the partonic cross sections on the family indices m, n amounts to an overall
factor |Vum dn |2 . This factor can be easily absorbed by redefining the parton distribution functions
as
for
fh,dm =
3
|Vum dn |2 fh,dn ,
fh,dm =
n=1
fh,um = fh,um ,
3
|Vum dn |2 fh,dn ,
n=1
fh,u m = fh,u m ,
fh,g = fh,g .
(3)
The hadronic cross section (1) can be computed using the trivial CKM matrix Vui dj = ij and
the redefined PDFs (3). Since we do not consider initial or final states involving (anti-)top
quarks, only the contributions of the first two quark families (m = 1, 2) have to be included.
The corresponding redefined PDFs (fh,q with q = u, d, c, s) automatically include the (small)
contributions associated with initial- and final-state bottom quarks.
2.2. Kinematics
For the 2 2 subprocess ab W k the Mandelstam variables are defined in the standard
way,
s = (pa + pb )2 ,
t = (pa pW )2 ,
u = (pb pW )2 .
(4)
2
2
The momenta pa , pb , pk of the partons
are assumed to be massless, whereas pW = MW . In terms
of x1 , x2 , pT and the collider energy s we have
M 2 s
M 2 s
t = W
(1 cos ),
u = W
(1 + cos ),
s = x1 x2 s,
(5)
2
2

2 )2 corresponding to the cosine of the angle between the mowith cos = 1 4pT2 s /(s MW
menta pa and pW in the partonic center-of-mass frame.
31
The pT -distribution for the unpolarized partonic subprocess ab W k reads

d abW k
2
= Nab d2 MabW k FO,2 (2 ),
(6)
dpT
1
where
= 4 pol col involves the sum over polarization and color as well as the average
factor 1/4 for initial-state polarization. The factor Nab is given by
Nab =
(2)4
,
2s Nab
(7)
2
2
where Nqq
= Ng q = Nc (Nc 1), with Nc = 3, account for
= Nq q = Nc , Ngq = Nqg = Nqg
the initial-state colour average. The phase-space measure d2 is given by
d2 =
d 3 pW
d 3 pk
0 (2)3 2p 0
(2)3 2pW
k
4 (pa + pb pW pk ),
(8)
while the function FO,2 defines the observable of interest, i.e. the W -boson pT -distribution in
presence of a cut on the transverse momentum of the jet,

min
FO,2 (2 ) = (pT pT,W ) pT,j pT,j
(9)
.
In the 2-particle phase space the jet is identified with the parton k and momentum conservation
implies pT,j = pT,k = pT,W . In practice, since we always consider the pT -distribution in the
min > p min , the cut on p
region pT > pT,W
T,j in (9) is irrelevant. The phase-space integral in (6)
T,j
yields two contributions originating from kinematic configurations in the forward and backward
hemispheres with opposite values of cos in the center-of-mass frame,
d abW
dpT
abW
d fwd
dpT
abW
d bkwd
dpT
abW
d fwd
dpT
abW
d bkwd
dpT
(10)
with

pT
MabW k 2 ,
8Nab s |t u|
k
abW

d

= fwd
.

dp
=
tu
(11)
For the 2 3 subprocess ab W k we define the following five independent invariants

s = (pa + pb )2 ,
t = (pa pW )2 ,
t = (pa p )2 ,
u = (pb p )2 ,
u = (pb pW )2 ,
(12)
and the four dependent invariants

2
s = (pk + p )2 = s + t + u MW
,
s = (pW + pk )2 = s + t + u ,
2
t = (pa pk )2 = MW
s t t ,
2
u = (pb pk )2 = MW
s u u .
The pT -distribution for this subprocess reads

2
d abW k
= Nab d3 MabW k FO,3 (3 ),

dpT
(13)
(14)
32
where
d3 =
d 3 pW
d 3 pk
0
(2)3 2pW
(2)3 2pk0
d 3 p
4 (pa + pb pW pk p ).
(2)3 2p0
(15)
In the 3-particle phase space, the W -boson pT -distribution in Wj production is defined by the
observable function

min
FO,3 (3 ) = (pT pT,W ) pT,j pT,j
(16)
.
The cut on the jet transverse momentum rejects events where the W -boson pT is balanced by
an isolated photon plus a parton with small transverse momentum. This observable is thus free
from singularities associated with soft and collinear quarks or gluons. When applying the cut on
the jet momentum in the 3-particle phase space, care must be taken that the definition of the jet
pT is collinear-safe. In general the jet cannot be identified with the parton k, since in presence of
collinear photon radiation the transverse momentum of a charged parton is not a collinear-safe
quantity. Thus we identify the jet with the parton k only if k is a quark well separated from the
photon or a gluon. Otherwise, i.e., for collinear quark-photon configurations, the recombined
momentum of the quark and photon is taken as momentum of the jet. In practice, we define the
separation variable

R(q, ) = (q )2 + (q )2 ,
(17)
where i is the pseudo-rapidity and i is the azimuthal angle of a particle i. If R(q, ) < Rsep ,
then the photon and quark momenta are
recombined by simple four-vector addition into an effective momentum pj and then pT,j =
(pT,q + pT, )2 , otherwise pT,j = pT,q . We note that, in
min . This
the collinear region, lowest-order kinematics implies pT,j = pT,q + pT, = pT,W > pT,j
means that the recombination procedure effectively removes the cut on pT,q inside the collinear
cone R(q, ) < Rsep . For instance the recombined gq W q cross section is given by

gq W q

gq W q
min
rec.
d
=
d gq W q +
pT,q pT,j
. (18)
R(q, )<Rsep
R(q, )>Rsep
In contrast, for the case of final-state gluons, we do not perform photon-gluon recombination and
the cut on pT,g is imposed in the entire phase space.
This procedure has the advantage to avoid both collinear-photon and soft-gluon singularities. However it implies a different treatment of quark and gluon final states and can thus be
regarded as an arbitrary cut-off prescription for the final-state collinear singularity. Moreover,
the recombined cross section (18) has a logarithmic dependence on the cut-off parameter Rsep .
These aspects are discussed in detail in Appendix A. There we compare the recombination procedure with a realistic experimental definition of exclusive pp Wj production, where final-state
min ) within the entire phase
quarks are subject to the same cut as final state gluons (pT,q > pT,j
space. Describing the exclusive gq W q cross section,

gq W q

gq W q
min
excl.
(19)
d
= pT,q pT,j
,
by means of quark fragmentation functions, we find that the quantitative difference between the
two definitions (18) and (19) amounts to less than two permille. Moreover, we show that the recombined cross section is extremely stable with respect to variations of the parameter Rsep . This
33
means that the recombination procedure used in our calculation provides a very good description
of exclusive pp Wj production.
Another treatment of the singularities, which does not require recombination and treats quarkand gluon-induced jets uniformly, has been proposed in Ref. [22]. There, contributions from Wj
production and W production to a more inclusive observable, i.e., high-pT W production, are
both calculated. All soft and collinear singularities in the final state cancel in the approach of
Ref. [22] as a result of the more inclusive observable definition than associated production of the
W boson together with a jet, considered in this work. The comparison of our results with those of
Ref. [22] seems to indicate that these differences in the jet definitions have a quite small impact
on the size of the electroweak corrections.
The quantity min in (1) is related to the minimum partonic energy that is needed to produce
min and p
min
final states with pT,j > pT,j
T,W > pT,W ,

2

min
min 2 + M 2
+ pT,W
.
s min = pT,j
(20)
W
When we evaluate the 2 2 contributions to the hadronic cross section (1), after analytic intemin = p min = p in (20).
gration of the phase space in (6), we can set pT,j
T
T,W
2.3. Crossing symmetries
The unpolarized squared matrix elements for the 2 2 processes in (2) are related by the
crossing-symmetry relations

W g 2
Mgq W q 2 = Mqq
,
s t

q 2
W g 2
MqgW
= Mqq
,

MbaW k 2 =
MabW k 2
s u
.
tu
(21)
Moreover, due to CP symmetry, the unpolarized partonic cross section for the production of
positively and negatively charged W bosons are related by

g 2
MduW + g 2 =
Md uW
.
(22)
Eqs. (21) and (22) permit to relate the six processes for W + production in (2) and the six charge
conjugate ones to a single process. Hence the explicit computation of the unpolarized squared
matrix element needs to be performed only once. In the following we will present explicit results
for the process qq
W g.
Similarly, for the unpolarized squared matrix elements for the 2 3 processes in (2) we have

W g 2
Mgq W q 2 = Mqq
,
{s u ,ts ,t s }

q 2
W g 2
MqgW
= Mqq
,
{s t ,u
s ,u s }

MbaW k 2 =
MabW k 2
(23)
{tu,
t u }
and

g 2
MduW + g 2 =
Md uW
.
It is thus enough to perform calculations only for the qq

W g subprocess.
(24)
34
2.4. Couplings and Born matrix element

For gauge couplings we adopt the conventions of Ref. [23]. With this notation the gq q vertex
and the V q q vertices with V = A, Z, W read
= igS t a ,
= ie
IqV q ,
(25)
=R,L
where are the chiral projectors

1
R = (1 + 5 ),
2
1
L = (1 5 ),
2
(26)
t a are the Gell-Mann matrices and I V are matrices in the weak isospin space. For diagonal
matrices such as I A and I Z we write IqV q = qq IqV . In terms of the weak isospin Tq3 and the
weak hypercharge Yq we have
IqZ =
cW 3
sW Yq
T
,
sW q cW 2
IqA = Qq = Tq3
Yq
,
2
(27)
with the shorthands cW = cos W and sW = sin W for the weak mixing angle W . The eigenvalues of isospin, hypercharge and SU(2) Casimir operators for left-handed fermions are
1
1
3
Tu3L = Td3L = ,
YuL = YdL = ,
CF = ,
CA = 2.
2
3
4
The only non-vanishing components of the generators associated with W bosons are
1
+
IuWL dL = IdWL uL =
.
2sW
(28)
(29)
The triple gauge-bosons vertices read

=
e Va Vb Vc 1 2
g
(k1 k2 )3 + g 2 3 (k2 k3 )1
sW

+ g 3 1 (k3 k1 )2 ,
(30)
where the totally anti-symmetric tensor V1 V2 V3 is defined through the commutation relations

i
I V1 , I V2 =
sW
V1 V2 V3 I V3 ,
(31)
V3 =A,Z,W
+
and has components ZW W = icW and AW W = isW .

To lowest order in and S , the unpolarized squared matrix element for the qq
Wg
process reads
2
2
qq

2

M W g 2 = 8 2 S N 2 1 I W 2 t + u + 2MW s ,
c
0
qL qL
tu
(32)
where = e2 /(4) and S = gS2 /(4) are the electromagnetic and the strong coupling constants.
35
Fig. 1. Tree-level Feynman diagrams for the process qq

W g.
3. Virtual corrections
In this section we present the virtual electroweak corrections to the qq
W g process.
The algebraic reduction to gauge-coupling structures, standard matrix elements and one-loop
scalar integrals is described in Section 3.2. The renormalization of ultraviolet divergences and
the subtraction of infrared singularities originating from soft and collinear virtual photons are
discussed in Sections 3.3 and 3.4, respectively. In Section 3.5 we summarize the one-loop result
for the unpolarized squared matrix element.
3.1. Preliminaries
As discussed in the previous section, the twelve different processes relevant for Wj production
are related by CP and crossing symmetries. It is thus sufficient to consider only one of these
processes. In the following we derive the one-loop corrections for the qq
W g process. The
matrix element
qq
W g
M1
qq
W g
= M0
qq
W g
+ M1
(33)
is expressed as a function of the Mandelstam invariants

s = (pq + pq )2 ,
t = (pq pW )2 ,
u = (pq pW )2 .
(34)
qq
W g
results from the t - and u-channel diagrams of Fig. 1. The

The Born contribution M0
loop and counterterm diagrams contributing to the corrections,
qq
W g
M1
qq
W g
= M1,loops
qq
W g
+ M1,CT
(35)
are depicted in Figs. 2 and 3, respectively.

The quarks that are present in the loop diagrams of Fig. 2 are treated as massless, and the
regularization of the collinear singularities that arise in this limit is discussed in Section 3.4. The
only quark-mass effects that we take into account are the mt -terms that enter the counterterms
through gauge-boson self-energies.
Our calculation has been performed at the matrix-element level and provides full control over
polarization effects. However, at this level, the analytical expressions are too large to be published. Explicit results will thus be presented only for the unpolarized squared matrix element.
3.2. Algebraic reduction
The matrix element (33) has the general form
qq
W g
M1
L,
= iegS t a v(p
q )M1
L u(pq ) (pW ) (pg ).
(36)
36
Fig. 2. One-loop Feynman diagrams for the process qq

W g. The diagrams s1, s2, v1 and v2 receive contributions
from neutral and charged gauge bosons, V = A, Z, W . The diagrams v3, v4, b1 and b2 involve only neutral gauge
bosons, V = A, Z. The remaining diagrams, v5, v6 and b3 involve two contributions with one charged and one neutral
gauge boson: (V1 , V2 ) = (V , W ) and (W , V ) with V = A, Z.
Fig. 3. Counterterm diagrams for the process qq

W g.
L,
Since we neglect quark masses, M1

consists of terms involving an odd number of matrices
with = 0, . . . , 3. The 5 -terms are isolated in the chiral projector L defined in (26). The
polarization dependence of the quark spinors and gauge-boson polarization vectors is implicitly
37
understood. In analogy to (33) and (35) we write

L,
M1
L,
= M0
L,
+ M1
L,
L,
L,
= M1,loops + M1,CT .
M1
(37)
Following the approach adopted in Ref. [9], we isolate the SU(2) U(1) couplings that appear
in the Feynman diagrams and reduce the one-loop amplitude to a sum of contributions associated
with independent coupling structures. As we will see, besides an Abelian and a non-Abelian
contribution that are related to the ones found for Zj production [9], for Wj production we have
two additional coupling structures.
The coupling structure of the Born amplitude is trivial and consists simply of the qL qL component of the SU(2) generator,
L,
M0

= IqWL q S0
L
S
= 0 ,
2sW
(/
(/
pW p
/ q )
pg p
/ q )
.
+
u
t
S0 =
(38)
The contribution of the loop diagrams of Fig. 2 can be written as

W V V

L,

I
I I q q D1 MV2 + I V I V I W q q D2 MV2
M1,loops =
L L
L L
4
V =A,Z,W

i W V W V W
MV2 +
I I
qL qL
sW
V =A,Z

i V W W W V

2
2
.
+
I
D4 MV2 , MW
I q q D4 MW
, MV2
L L
sW
IV IW
IV
qL qL
D3
(39)
In the following, treating the electroweak gauge couplings as isospin matrices and using grouptheoretical identities (see Appendix B of Ref. [23]), we express the above amplitude in terms
of the eigenvalues of isospin, hypercharge and SU(2) Casimir operators for left-handed fermions (28).
The tensors D1 (MV2 ) and D2 (MV2 ) in (39) describe the contributions of the diagrams s1,
v1 and s2, v2, respectively. These diagrams may involve charged or neutral virtual bosons. In the
former case (V = W ), the corresponding couplings read1
IW
IW IW
IW IW
IW
CF (T 3 )2 W
I
.
2
sW
In the latter case (V = A, Z) the coupling factors read

3 2
2
SU(2) (T )
U(1) Y
W V V
3
I
I I = V V
+ XV T Y + V V
IW ,
2
2
sW
4cW

3
2
2
SU(2) (T )
U(1) Y
V V W
3
I I I
= V V
XV T Y + V V
IW .
2
2
sW
4cW
(40)
(41)
Here
SU(2)
V V
= (UV W 3 )2 ,
XV =
U V W 3 UV B
,
sW cW
U(1)
V V = (UV B )2 ,
(42)
1 The following identities have to be understood as matrix identities, where the q q indices of the SU(2) generators
L L
are implicitly understood.
38

SU(2)
where U is the electroweak mixing matrix. For V = A, Z we have AA

U(1)
2 , and SU(2) = c2 , X = 1, U(1) = s 2 . We note that
AA = cW
Z
W
W
ZZ
ZZ
U(1)

SU(2)
V V =
V V = 1,
XV = 0.
V =A,Z
V =A,Z
2 , X = 1,
= sW
A
(43)
V =A,Z
The tensor D3 (MV2 ) in (39) corresponds to the diagrams v3, v4, b1 and b2. These diagrams
receive contributions from neutral virtual gauge bosons only (V = A, Z). For the corresponding
couplings we have

2
SU(2) CF CA /2
U(1) Y
V W V
I I
(44)
I = V V
+ V V
IW .
2
2
sW
4cW
Finally, D4 (MV21 , MV22 ) represents the diagrams v5, v6 and b3. These diagrams involve a neutral
gauge boson (V = A, Z) and a W boson. The coupling factors yield

i W V W V W
SU(2) CA
3
I I
= V V
X
T
Y
IW ,
V
2
sW
4sW

i V W W W V
SU(2) CA
3
I
I = V V
+ XV T Y I W .
(45)
2
sW
4sW
Using the above identities we express the one-loop amplitude (39) for W -boson production in a
form that is analogous to the one adopted in Refs. [9,10] to describe the production of neutral
gauge bosons. To this end we define2

A1,A MV2 = D1 MV2 + D2 MV2 + D3 MV2 ,
1

2

2
A1,N MV2 = D4 MV2 , MW
+ D4 MW
, MV2 D3 MV2 ,
2

A1,X MV2 = D1 MV2 + D2 MV2 ,

2

2
D4 MW
, MV2 + D2 MV2 D1 MV2 .
A1,Y MV2 = D4 MV2 , MW
(46)
The tensor A1,A (MV2 ) is identical to the Abelian tensor defined in Ref. [9], and A1,N (MV2 )
2 . The remaining two tensors,
is equal to the non-Abelian tensor of Ref. [9] for MV2 = MW
2
2
A1,X (MV ) and A1,Y (MV ), are new. Using (40)(46) we can write the one-loop amplitude
(39) as

2

L,

SU(2) CF
U(1) YqL
M1,loops =
+
V V
A1,A MV2
VV
2
2
sW
4cW
4 2sW V =A,Z
(Tq3L )2

SU(2) CA
A1,X MV2 + V V
A1,N MV2
2
2sW

3
2
CF (TqL )

2
M
.
A
XV Tq3L YqL A1,Y MV2 +
W
1,X
2
sW
SU(2) CF
V V
2
sW
(47)
2 In our notation we emphasize the dependence of the form factors A on M , whereas the dependence on the
V
1,I
external momenta as well as the MW -dependence (for A1,N and A1,Y ) is implicitly understood.
39
This amplitude has been reduced algebraically using the Dirac equation, the identity p (p) =
0 for gauge-boson polarization vectors and Dirac algebra. Moreover, tensor loop integrals have
been reduced to scalar ones by means of the PassarinoVeltman technique [24]. The result has
been expressed in the form
10

ij

A1,I MV2 =
FI MV2 Si Jj MV2 ,
(48)
i=1 j
ij
for I = A, N, X, Y. The quantities FI (MV2 ) are rational functions of Mandelstam invariants and
masses. Explicit expressions for the tensors Si and the scalar loop integrals Jj (MV2 ) are provided in Appendices B and C.
3.3. Renormalization
While the tensors A1,X and A1,Y are ultraviolet finite, the Abelian and the non-Abelian
tensors give rise to the ultraviolet singularities

UV S ,
UV S ,
A1,N MV2 UV = 2
A1,A MV2 UV =
(49)
0
0
where S0 is the tensor structure of the Born amplitude (38), and

2

42 (1 + ) 1
UV =
= E + ln(4) + ln
+ O()
2
MZ
MZ2
(50)
in D = 4 2 dimensions. These singularities are cancelled by the counterterm diagrams depicted in Fig. 3 and the results are independent of the scale of dimensional regularization. The
counterterms that are responsible for the contributions of diagrams c1, c2, c3 and c4 read
= i/
pL ZqL ,
= igS t a L ZqL .
(51)
Since there is no O() contribution to the renormalization of the strong coupling constant gS ,
these counterterms depend only on the wave-function renormalization constants for left-handed
W g process, i.e., the sum of the diquarks, ZqL . Their combined contribution to the qq
agrams c1, c2, c3 and c4, vanishes. The renormalization of the qq
W g process is thus
provided by the diagrams c5 and c6, which originate from the W qq
counterterm,

= ie L IqWL q C A + C N ,
L
with
1
C = (ZuL + ZdL ),
2
A

g22
1
C =
ZW + 2 ,
2
g2
N
(52)
(53)
and yields
L,

L,
M1,CT = C A + C N M0 .
(54)
40
The wave-function renormalization constants of massless left-handed quarks and on-shell W

bosons are related to the corresponding self-energies by

TW (p 2 )
ZqL = Re q,L (0) ,
(55)
ZW = Re
2 2,
p 2
p =M
W
and have been evaluated using the explicit results of Ref. [25].
For the definition and the renormalization of the SU(2) coupling constant,
g22 =
4
,
2
sW
g22
g22
2
sW
2 ,
sW
(56)
we adopt the G -scheme, where the electromagnetic coupling constant is expressed in terms
of the Fermi constant G , and the weak mixing angle is related to the on-shell masses MZ , MW
of the gauge bosons,
2 s2
2G MW
W
2
2
2
=
(57)
= 1 cW
= 1 MW
/MZ2 .
,
sW
The counterterm / in the G -scheme can be derived from the on-shell counterterm
(0)/(0) for the fine-structure constant in the Thompson limit. Using the one-loop relation
= (0)[1 + r] and requiring + = (0) + (0) we have
(0)
=
r.
(0)
(58)
Combining the relations (56)(58) and using the explicit one-loop expression for r [26,27], we
obtain
W
2

2 ) W (0)
2
T (MW
MW
g22
7 12sW
1
= Re
ln
6+
2 UV +
. (59)
2
2
2
4
g2
MW
sW
2sW
MZ2
The above conterterms yield the ultraviolet singularities

UV
C A UV =
8

C N
q=u,d V =A,Z,W
UV .
=

UV
2
2sW

qL
IV IV

Y2
UV CF + qL ,

2
2
4
sW
4cW
(60)
Using (43) one can easily verify that these singularities cancel those resulting from the loop
diagrams [see (47) and (49)].
3.4. Soft and collinear singularities
Loop diagrams and wave-function renormalization constants involve singularities originating
from soft and collinear virtual photons (for brevity denoted in the following as IR singularities).
In order to isolate these singularities and check that they are cancelled by corresponding ones
originating from real photon bremsstrahlung, we split the wave-function renormalization con
stants and the photon contributions to (47), i.e. the terms A1,I (MA2 ), in IR-singular (IR) and
IR-finite (fin) parts:
41
ZqL = ZqIRL + ZqfinL ,

IR
fin
ZW = ZW
+ ZW
,

IR,
fin,
A1,I MA2 = A1,I + A1,I .
(61)
The singular parts depend on the scheme adopted to regularize IR singularities. The remaining
parts are scheme-independent and free from IR singularities, but can contain ultraviolet poles.
For the regularization of IR singularities we use, alternatively, two different schemes:
In the first scheme, which we denote as mass-regularization scheme (MR), we use infinitesimal quark masses m and a photon-mass regulator, MA = with 0 < m. Since the
quark-mass dependence disappears in the final result, we perform the computation using the
same mass m for all quarks. To denote quantities evaluated in this scheme we use the label
MR;
In the second scheme we perform the calculation using massless fermions and photons,
MA = m = 0, and we evaluate IR singularities in dimensional regularization (DR). To denote
quantities evaluated in this scheme we use the label DR.
The singular parts of the wave-function renormalization constants read
ZqIRL =
IR
ZW
=
4
42
2
MW
42
2
MW

(1 + )Q2q hIR
q ,

(1 + )hIR
W,
(62)
with
hIR
q,MR
2
2
MW
= ln
2 ln
4,
2
m
m2
hIR
W,MR
2
= 2 ln
,
2
MW
(63)
in the MR scheme and

1
hIR
q,DR = ,
2
hIR
W,DR = ,
(64)
in the DR scheme. The splitting of the loop contributions A1,I (MA2 ) into IR-singular and IRfinite parts is performed at the level of the scalar loop integrals Ji (MA2 ):

Ji MA2 = JiIR + Jifin .
(65)
Explicit expression for the IR-singular and IR-finite parts of individual loop integrals are presented in Appendix D. Combining all singular contributions JiIR we obtain

IR,
A1,I
42
2
MW
(1 + )fIIR S0 ,
(66)
42
i.e., the IR singularities factorize3 with respect to the Born amplitude (38). The IR-singular part
of the renormalized amplitude can be expressed in terms of the electromagnetic charges of the
external particles as

L,

42
L,
M1,IR =
(67)
(1 + ) Qq Qq f1IR + Qq f2IR Qq f3IR M0 ,
2
4 MW
where = 1 is the charge of the W boson and
f1IR = fAIR hIR
q ,

1 IR
IR
fX fYIR hIR
q hW ,
2

1 IR
IR
IR
IR
IR
f3 = fA fN + fX + fYIR hIR
q hW .
2
In the MR scheme we obtain
2
2
2
2
s
m
IR
2 m
f1,MR = 2 ln
ln
ln
+ 3 ln
+ 2 ln
,
2
2
2
2
m
m2
MW
MW
MW
2 2

t
1 2 m2
m
IR
f2,MR = ln
ln
2 ln 1 2
ln
2
2
2
2
MW
MW
MW
MW
2
2
1
m
+ ln
+ 2 ln
,
2
2
2
MW
MW

f IR = f IR
,
(69)
and in the DR scheme

2
1
s
IR
= 2 2 ln
f1,DR
3
+ 4,
2
MW

1
1
t
5
IR
+ 2,
f2,DR = 2 2 ln 1 2
MW

f IR = f IR
.
(70)
f2IR = fAIR fNIR +
3,MR
3,DR
2,MR tu
2,DR tu
(68)
The splitting (61) has been performed in such a way that in the high-energy limit (s , |t|, |u|

2 ) the IR-finite part of the amplitude has the same logarithmic behaviour as the virtual corMW
rections regularized by a photon mass MA = MW . Indeed the IR-singular parts fiIR correspond
exactly to the contribution called purely electromagnetic in Ref. [14]. This implies that, up to
terms that are not logarithmically enhanced at high energies, the IR-finite part of the corrections
corresponds to the symmetric electroweak contribution of Ref. [14], which is constructed by
setting the photon mass equal to MW . This property is evident in the asymptotic high-energy
expressions (80) for the IR-finite part of the diagrams involving virtual photons.
2
2
3 To be precise, the tensors A
1,X (MA ) and A1,N (MA ) contain also non-factorizable IR divergences. However these
non-factorizable singularities are related by
non-fact,
A1,X
2 = 2Anon-fact, M 2 ,
MA
A
1,N
and due to the identity CF (Tq3L )2 = CA /4, which relates the coupling structures associated with the X- and N-terms
in (47), they cancel.
43
3.5. Result
Let us summarize our result for the unpolarized squared matrix element for the qq
Wg
2
process. To O( S ),

qq
qq
qq

W g
qq
W g
M W g 2 =
M W g 2 + 2 Re
M
M
.
1
(71)
Using (36) and summing over the polarizations we can express the interference term as
2 Re

= 2
qq
W g
qq
W g
M0
M1
S Nc2

pW pW
L,
L,
g g
1 Re Tr p
/ q M0 p
/ q M1
,
2
pW

(72)
= 0 M 0 . Combining the contributions of the bare one-loop diagrams (47) and the
where M
counterterms (54) yields
2
qq
qq

M W g 2 = 1 + 2 Re C A + C N
M W g 2 + 2 S N 2 1
c
1
0
2
sW

2

CF
U(1) YqL
Re
+
VSU(2)
H1A MV2
V
VV
2
2
sW
4cW
V =A,Z
(Tq3L )2

SU(2) CA
H1X MV2 + V V
H1N MV2
2
2sW

CF (Tq3L )2 X 2
2
3
Y
H1 MW .
XV TqL YqL H1 MV +
2
sW
SU(2) CF
V V
2
sW
(73)
The unpolarized Born contribution is given in (32), the counterterms C A and C N are presented
in Section 3.3, and the coupling factors are specified by (28) and (42). The functions H1I (MV2 )
represent the contributions resulting from the loop diagrams of Fig. 2. They are related to the
tensors A1,I (MV2 ) in (47)(48) by

1

pW pW

/ q A1,I MV2 g g
/ q S0 p
.
H1I MV2 = Tr p
2
8
pW
(74)
The couplings associated with H1A , H1N and H1X are the same for q = u and q = d. Thus the
crossing and CP symmetry relations (21) and (22) imply that these functions are symmetric with
respect to the transformation t u.
In contrast, H1Y is antisymmetric with respect to t u exchange since the corresponding coupling is proportional to Tq3L and has thus opposite signs for
q = u and q = d. The functions H1I (MV2 ) are presented in Appendix E as linear combinations of
scalar loop integrals. We note that, in contrast to the definition adopted in the case of Zj production [9], here we do not include the contributions of the fermionic wave-function renormalization
constants in H1A (MV2 ).
44
For the IR-singular part of the renormalized one-loop correction we obtain

qq

M W g 2 = Re Qq Qq f IR + Qq f IR Qq f IR
1
2
3
1,IR
2

qq

42
M W g 2 .
(1
+
)
0
2
MW
(75)
The IR-singular functions fiIR in the MR and DR schemes are presented in Section 3.4.
4. High-energy limit
In this section we provide compact analytic expressions that describe the behaviour of the
2 /
s 0 with t/s and u/
s
IR-finite part of the virtual electroweak corrections in the limit MW
constant. In this limit, which is applicable for transverse momenta of O(100 GeV) or beyond,
2 ).
the electroweak corrections are dominated by logarithmic contributions of the type ln(s /MW
In Section 4.1 we present the asymptotic expansion of the one-loop corrections, including leading and next-to-leading logarithms, as well as terms that are not logarithmically enhanced at
high energies. In Section 4.2 we present the two-loop corrections to next-to-leading logarithmic
accuracy.
4.1. Next-to-next-to-leading approximation at one loop
In this section we discuss the high-energy behaviour of the IR-finite part of the one-loop
corrections to the qq
W g process, obtained by subtracting the IR divergence (75) from the
renormalized one-loop result (73),
qq
qq
qq

M W g 2 =
M W g 2
M W g 2 .
1,fin
1
1,IR
(76)
In the following we present explicit asymptotic expressions for the unrenormalized loop contributions, i.e., for the IR-finite parts H1I,fin of the functions H1I in (73). Using the general results
of Ref. [28], we evaluate the functions H1I,fin to next-to-next-to-leading logarithmic (NNLL) accuracy. This approximation accounts for all contributions that are not suppressed by powers of
2 /
s . It includes double and single logarithms as well as terms that are not logarithmically
MW
enhanced in the high-energy limit. To simplify non-logarithmic functions of the ratio MZ /MW
2 = 1 M 2 /M 2 , keeping only terms up to the first order4
we have performed an expansion in sW
W
Z
I,fin
2
in sW . The NNLL expansion of H1 (MV2 ) has the general form

NNLL

t2 + u 2

t2 u 2

H1I,fin MV2 = Re g0I MV2
+ g1I MV2
+ g2I MV2 .
tu
tu
(77)
which has the same angular behaviour as the

It involves the rational function (t2 + u 2 )/tu,
squared Born amplitude (32) in the high-energy limit, and two other rational functions, which
describe different angular dependencies. The functions giI consist of logarithms of the kinematical variables and constants. The loop diagrams involving Z and W bosons, with mass
4 In practice we find that all terms of O(s 2 ) cancel in the result.
W
45
MV = MZ , MW , yield

2
2

MV
MZ
s
2
= 2 UV + ln
+ ln
+ ln
2
2
MW
MW
MV2

1
2 t
2 u
2 u
2 t
+
ln
+
ln
+
ln
+
ln
ln2
2
2
2
2
2
u
MV
MW
MV
MW

2
3
20
2
t
u
ln2
+ ln2
+ 4,
2
s
s
9
3

1
u
t
g1N MV2 = ln2
ln2
,
2
s
s

2

MV
t
u
t
u
g2N (MV2 ) = 2 ln2
+ ln2
+ ln
+ ln
+ 2 ln
4 2 ,
2
s
s
s
s
MW

t
u
3 2 t
s
s
2 u
ln
+
ln
+
ln
+
ln
+
3
ln
+
g0A MV2 = ln2
2
s
s
s
s
MV2
MV2
g0N
MV2

7 2 5
+ g0A,UV MV2 ,
3
2

3
u
t
2
g1A MV2 = g1N MW
+ ln
ln
,
2
s
s

2
g2A MV2 = g2N MW
,

g0X MV2 = 0,

g1X MV2 = 0,

s
u
t
g2X MV2 = 2 2 ln
+
ln
3
,
+
ln
s
s
MV2

2 t
2 u
2 u
ln
ln
+
ln
,
g0Y MV2 = ln2
2
2
MW
MV2
MW
MV2

g1Y MV2 = 0,

t
g2Y MV2 = 2 ln
,
u
+
(78)
UV is defined in (50) and, in order to facilitate the comparison with Ref. [9], we have
where
isolated the term
2
MZ
1
A,UV 2
MV = UV + ln
g0
(79)
.
2
2
MV
If we included the fermionic wave-function renormalization constants in the definition of the
function H1A , as we had done for the case of Zj production in Ref. [9], this term would cancel
and the function g0A (MV2 ) would be identical to the one obtained in Ref. [9].
For the loop diagrams involving photons (MV = MA ), after subtraction of the IR-singular
parts, we obtain
46

2 7 2
2
g0N MA2 = g0N MW

+ ,
9
3
2
2
A
A
2
g0 MA = g0 MW + ,

2
g0I MA2 = g0I MW
for I = X, Y,
2
2
I
I
g1 MA = g1 MW
for I = A, N, X, Y,
2
2
I
I
g2 MA = g2 MW
for I = A, N, X, Y.
(80)
The contribution of the counterterms C A and C N to the IR-finite part of the renormalized result (76) is obtained by subtracting from (53) the IR-divergent part of the wave-function
renormalization constants (62). This contribution, consisting of on-shell self-energies and their
derivatives, does not depend on the scattering energy. Therefore we evaluate the IR-finite parts
of the counterterms in numerical form without applying any approximation. Using the input parameters specified in Section 7 we obtain

1 fin
ZuL + ZdfinL = C A UV + 5.57 104 ,
2

g22
1
N,fin
fin
=
ZW + 2 = C N UV 1.49 103 .
C
2
g2
C A,fin =
(81)
UV -terms in (78)(80).
The UV divergences C A,N |UV [see (60)] cancel against the
The results (78)(80), for the qq

W g process, are valid for arbitrary values of the Mandelstam invariants and can easily be translated to all other processes in (2) by means of the
relations (21)(22). Logarithms with negative arguments in (78)(80) are defined through the
usual i prescription, r r + i for r = s , t, u.
In next-to-leading logarithmic (NLL) approximation, i.e., retaining only double and single
logarithms that grow with energy, the above results assume a particularly compact form. In this
approximation the counterterms do not contribute,
NLL
NLL
C A = C N = 0,
(82)
H1I,fin (MV2 ),
neglecting logarithms of MZ /MW , we obtain

2
|t|
|s |
t + u 2
N,fin 2 NLL
2
2 |u|
2
H1
MV = ln
+
ln
ln
,
2
2
2
tu
MW
MW
MW

2
|s |
|s |
t + u 2
A,fin 2 NLL
2
MV = ln
3
ln
,
H1
2
2
tu
MW
MW

NLL
|s |
,
H1X,fin MV2 = 4 ln
2
MW

NLL
H1A,fin MV2 = 0,
and for the functions
NLL
(83)
for V = A, Z, W . We note that, owing to H1X,fin (MZ ) = H1X,fin (MA ) and (43), the NLL contribution of the function H1X,fin (MV ) cancels in (73). Thus the NLL corrections (83) are propor which describes the angular dependence of the Born
tional to the rational function (t2 + u 2 )/tu,
cross section.
47
4.2. Next-to-leading logarithms up to two loops

Let us now present our results for the NLL asymptotic behaviour of the electroweak corrections up to two loops. For a discussion of the calculation we refer to Ref. [8], where the same
class of corrections has been computed for Zj production. The results have been obtained in
the MZ = MW approximation. As in the previous section, we present results for the IR-finite
part of the electroweak corrections, obtained after subtraction of IR singularities. As discussed
in Section 3.4, at one loop this subtraction is performed in such a way that, to NLL accuracy,
the IR-finite part corresponds to the complete electroweak correction regularized with a fictitious
photon mass MA = MW . The same prescription is adopted at the two-loop level.
The unpolarized squared matrix element for qq
W g, including NLL terms up to the
two-loop level, has the general form

2

2
qq

2

(2)
M W g 2 = 8 2 S N 2 1 t + u A(0) + A(1) +
A
. (84)
c
2
2
2
tu
The Born contribution reads
A(0) =
1
.
2
2sW
At one loop, the NLL part consists of double- and single-logarithmic terms and reads

CA 2

1
(1) NLL
ew 2
2
2
= 2 CqL Ls 3Ls + 2 Lt + Lu Ls .
A
2sW
2sW
(85)
(86)
2 ) for the logarithms and C ew = Y 2 /(4c2 ) +

Here we used the shorthand Lkr = lnk (|r |/MW
qL
qL
W
2 are the eigenvalues of the electroweak Casimir operator for left-handed quarks. This exCF /sW
pression is consistent with the process-independent results of Ref. [14] as well as with the NLL
part of the one-loop asymptotic expressions presented in Section 4.1. At two loops we obtain

CA 4

CA
1
NLL 1
ew
ew 4
3
4
4
A(2) =
L
+
L
+
6L
+
L
L
C
C
qL
s
u
s
s
t
2
2
2
2 qL
2sW
2sW
2sW

CA
1 b1 YqL 2 b2
(87)
+
+
Ls3 ,
+
C
F
2
2
6 cW
2
2
sW
2 ) and b = 19/(6s 2 ) are the one-loop -function coefficients associated
where b1 = 41/(6cW
2
W
with the U(1) and SU(2) couplings, respectively. The LLs as well as the angular-dependent subset of the NLLs in (87), i.e., all contributions of the form L4r with r = s , t, u,
have been derived
from Ref. [15]. There, by means of a diagrammatic two-loop calculation in the spontaneously
broken electroweak theory, it was shown that such two-loop terms result from the exponentiation of the corresponding one-loop corrections. The additional NLLs of the form L3s in (87) have
been obtained via a fixed-order expansion of the process-independent resummed expression proposed in Ref. [16]. This resummation [16] relies on the assumption that effects from spontaneous
breaking of the SU(2) U(1) symmetry can be neglected in the high-energy limit.
Our NNLO predictions include only the LL and NLL terms. Thus they are affected by
a potentially large theoretical uncertainty, due to missing subleading contributions of order
2 ) with k = 2, 1, 0. For four-fermion scattering it was found that, at s 1 TeV2 , the
2 lnk (s /MW
two-loop logarithmic expansion has an oscillating behaviour characterized by large cancellations
48
Fig. 4. Tree-level diagrams for the process qq

Wg .
between leading and subleading terms [3]. In this case the subleading terms play a very important role and the NLL approximation yields misleading results. In contrast, in the case of Wj
production, the relative weight of the LL, NLL and NNLL contributions at one loop indicates a
fairly good convergence of the logarithmic expansion. Indeed, as can be seen from our numerical
results in Section 7, the one-loop corrections are clearly dominated by the negative LL contributions, while the NLL terms are relatively small and the NNLL contributions almost negligible.
A similar convergence is expected also at two-loops, owing to the exponentiation property of
the logarithmic corrections. Thus our NLL two-loop predictions can be regarded as a plausible
estimate of the size of the two-loop electroweak effects in high-pT W -boson production.
5. Real corrections
In order to cancel IR singularities from the virtual-photon corrections, real emission corrections need to be calculated. As discussed in Section 2.3, all relevant partonic reactions are related
to the qq
W g process through crossing and CP symmetry. The tree-level diagrams for this
process are shown in Fig. 4.
The squared matrix element for the 2 3 process ab W k , summed over polarization
and colour as well as averaged over initial-state polarization, can be written in a general form5
abW k 2

M
= 4 Qq Qq H 1,ab (s , t, u,
t , u )
r
0

+ Qq Hr2,ab (s , t, u,
t , u ) Qq Hr3,ab (s , t, u,
t , u ) ,
(88)
with the kinematical invariants defined in Section 2.2. In the limit of soft and/or collinear photon
emission, the squared matrix element (88) exhibits IR singularities. To combine these singularities with those originating from virtual corrections we have to extract them in analytic form.
This is done with the help of the dipole subtraction formalism [1719]. Within this framework
the partonic differential cross section can be schematically written as

d Aab
d abW k
(89)
= Nab d3 Mab (3 ) Mab
,
sub (3 ) +
dpT
dpT
with Nab given in (7). The quantity Mab reads
abW k 2
M
FO,3 (3 ).
Mab (3 ) =
0
5 The expressions for the functions H i,qq

in (88) are available from the authors upon request.
r
(90)
49
ab
The auxiliary function Mab
sub is chosen such that it has the same singular behaviour as M in the
ab
ab
soft and collinear limits. This ensures that the difference M Msub can be integrated numerically. To compensate for the subtraction, the integral of the auxiliary function Mab
sub , denoted here
d Aab /dpT , is then added back. The analytical form of d Aab /dpT is obtained after performing
integration over the subspace of the radiated photon. The result of this one-particle subspace
integration contains singular contributions which must be combined with those in virtual corrections. The algorithms for constructing the auxiliary subtraction function and its integrated
counterpart have been developed both for the case of photon radiation off massless or massive
fermions [17] and QCD radiation off massless [18] or massive partons [19]. In Sections 5.1 and
5.2 we discuss the application of both formalisms to calculate the O() real corrections to the
Wj production process. In both approaches we use expressions for the emission off a massive
fermion to describe the emission off a W boson, since only soft singularities are present in this
case and they depend only on the charge of the external particle and not on its spin.
After adding the real and virtual corrections, collinear singularities remain. Final-state singularities are avoided by recombining collinear photonquark configurations as discussed in Section 2.2. Initial-state singularities are absorbed in the definition of PDFs using the MS scheme.
5.1. Mass regularization

The formalism of Ref. [17] employs small photon and fermion masses to regularize soft and
collinear singularities. The subtraction term for the squared matrix element is constructed from
the appropriate dipole factors. Keeping the original notation of Ref. [17] we can write for the
process ab W k (where a (b) can be q,
q , g)

sub
Qa Qb gab,
Mab
(pa , pb , p )Ma0 b ( 2,ab )
sub (3 ) = 4
=

sub
(pa , pW , p )Ma0 b ( 2,aW )
Qa gaW,
sub
a b
Qa gW
a, (pW , pa , p )M0 (2,W a )

sub
Qa Qk gak,
(pa , pk , p )Ma0 b ( 2,ak )

sub
Qk Qa gka,
(pk , pa , p )Ma0 b ( 2,ka ) + (a b) {a =a,b =b}
sub
+ Qk gkW,
(pk , pW , p )Mab
0 (2,kW )

sub
ab
+ Qk gW
k, (pW , pk , p )M0 (2,W k ) ,
(91)

MabW k ( 2,nm )2 FO,2 ( 2,nm ).
0
(92)
with
Mab
0 (2,nm ) =
Due to Qg = 0 the dipole terms with gluon indices do not contribute to (91) and for each subsub
process the subtraction term Mab
sub is constructed from six dipole terms, characterised by the g
functions. Expressions for these functions are taken directly from Ref. [17]. In Appendix F (see
Table 1) we list all the functions which are used to calculate (91), together with the corresponding
equation numbers in Ref. [17].
For each subprocess the six dipole terms fall into three groups, each containing two dipole
terms and coming with a specific charge combination, either Qq Qq or Qq , or Qq . The
ab in (88) and the IR-singular part of the
subtraction term Mab
sub has then the same structure as M
50
virtual corrections (75). Thus the cancellation of singularities can be analyzed for each charge
combination separately.
The construction of the reduced phase space 2,nm follows the prescriptions of Ref. [17].
Generally 2,nm is a mapping from the 3-particle phase space into a 2-particle phase space. The
mapping respects all mass shell conditions. For different types of dipoles, different mappings
are necessary. In Table 1 we list numbers of equations in Ref. [17] which we used to perform
mapping for the dipole terms appearing in our calculations. In particular, the observable-defining
function FO,2 ( 2,nm ) in (92) is then

min
,
FO,2 ( 2,nm ) = (pT p T,W ) p T,j pT,j
(93)
with p T,W and p T,j belonging to 2,nm .
The expression for the subtraction term integrated over the phase space of the photon reads

abW k
d fwd
d Aab
sub
=
(s , pT )
2Qa Qb GI,I (rab )
dpT
2
dpT
1
+
dx
abW k
d fwd
sub
GI,I (rab , x) +
(x s , pT ) + (a
dpT

b)

abW k
d fwd
(
r
)
(s , pT )
Qa Gsub
I,FM aW
dpT
1
+
0
sub
d abW
raW (x), x + fwd
dx GI,FM
dpT

(x s , pT ) + (a b)

abW k
d fwd
sub
(s , pT )
Qb GI,FM (rbW )
dpT
1
+
dx
sub
GI,FM

d abW
rbW (x), x + fwd
dpT

(x s , pT ) + (a b)

abW k
d fwd
(
r
)
(s , pT )
Qa Qk Gsub
I,F ak
dpT
1
+
0
sub
d abW
rak (x), x + fwd
dx GI,F
dpT

(x s , pT ) + (a b)

abW k
d fwd
sub
(s , pT )
Qb Qk GI,F (rbk )
dpT
1
+
dx
0
sub
GI,F

d abW
rbk (x), x + fwd
dpT

(x s , pT ) + (a b)

abW k
d fwd
sub
(s , pT ) + (a b) .
+ Qk GF,FM (rkW )
dpT
(94)
51
The relevant invariants in (94) are defined as rab = (pa + pb )2 = s , rkW = (pk + pW )2 = s , and
raW (x) = (xpa pW )2 ,
rak (x) = (xpa pk )2 ,
rnl = rnl (1).
(95)
abW k
/dpT in (94) represent the contributions originating from

The terms proportional to d fwd
the forward hemisphere in the 2-particle phase space [see (10)(11)]. The (a b) terms are the
contributions from the backward hemisphere, and
abW
d fwd
dpT

d abW
= bkwd
dpT
ab
(96)
abW k /dp in (94) directly indicates the x-dependence of the

Note that the first argument of d fwd
T
actual values of the t, u invariants defined in (5). The plus-distributions appearing in Eq. (94) are
evaluated according to the prescription
1
0

d

dx G sub r (x), x +
(x s , pT )
dpT
1
=
0

d

d
sub
sub
r (x), x
r (1), x
dx G
(x s , pT ) (x ) G
(s , pT ) ,
dpT
dpT
where = (pT +
(97)

2 )2 /
pT2 + MW
s guarantees the minimal center-of-mass energy to produce the
final state. The expressions for the functions Gsub and G sub follow directly6 from the results in
Ref. [17]. For the functions Gsub they read

IR
1
r
r
2 2
2
Re f1,MR + ln
3 ln
+4 ,
2
2
2
3
MW
MW

IR
r
r
2
2

Gsub
(
r
)
=
Re
f
+
2
ln
1
ln
2
I,FM
2,MR tr
2
2
MW
MW

4

2
2
3MW
MW
MW
r
2
Li
+ ln 1 2
2
2 r
r
r 2
MW
2MW

2
MW
2 1
r
r
+
+ ,
2
Li
+
+ 2 Li2
2
2
2
r
6
2
2MW r
2MW r

IR
r
r
2 1

r ) = Re f1,MR
+ ln2
+ ,
3
ln
Gsub
I,F (
s r
2
2
3
2
MW
MW
Gsub
r) =
I,I (
(98)
(99)
(100)
6 The function Gsub (r ) has been derived from Eq. (4.10) in Ref. [17] by taking the limit of an infinitesimal quark
F,FM
mass.
52

2
MW
r
2
1
+
ln
1
2
r
MW

r MW
3
r
r
1
7
ln
ln
+ ln
2
2
2
2
2
MW
MW
r + MW

2

2
MW
M2
MW
2 2
+ 4 Li2
4 Li2
+ W
+ 3.
r
r
2r
3
IR

Gsub
r ) = Re f2,MR
+ ln2
F,FM (
tr
(101)
IR cancel against those in the virtual corrections, cf. Section 3.4.

The IR-singular functions fi,MR
The explicit forms of the functions G sub are

1 + x2
r
sub
(r , x) = (x) +
ln 2
+ 2 ln(1 x) + 1 x,
GI,I
(102)
1x
QED

2

MW r
1 + x2
sub
GI,FM (r , x) = (x) +
ln
+ ln (1 x) 1 z1 (r , x)
1x
x2QED

2 x
4MW
z1 (r , x) 1
+
+ 1 x,
3 + z1 (r , x)
(103)
2 )(1 x)
2(1 x)
(r MW

1 + x2
r
sub
GI,F (r , x) = (x) +
ln
+ ln(1 x)
1x
x2QED
3
+ 1 x,
2(1 x)
(104)
with
z1 (r , x) =
and
2 x
MW
(105)
2 (1 x)
MW
r
2

QED
1 + x2
ln
(x) =
2 ln(1 x) 1 ,
1x
m2
(106)
where QED is the factorization scale and m stands for the quark-mass regulator. The functions
(x) are singular. These singularities are related to the collinear photon radiation off an initialstate quark and are absorbed in the definition of the PDFs, yielding the hadronic cross section
finite. The procedure bears complete analogy to absorbing collinear QCD singularities into the
definition of the PDFs. In the MS factorization scheme, the redefinition is achieved by replacing [5]
fh,q x, 2QCD
fh,q x, 2QCD , 2QED
Q
2 q
1

dz
x 2
fh,q
, QCD , 2QED (z) + .
z
z
(107)
5.2. Dimensional regularization
In an independent calculation we used the results of Refs. [18,19] to evaluate the dipole subtraction terms and their integrated counterparts. The formalism of [18,19] is concerned with QCD
53
radiation and expressions for dipoles are given as matrices in colour and helicity space. Since we
consider photon emission off a fermion line, the colour and helicity structure disappears and the
dipole matrices reduce to simple expressions. More precisely, to adapt the formalism Refs. [18,
19] for the calculation of QED corrections, we make use of expressions describing gluon radiation off a fermion line in Refs. [18,19] and replace
S ,
Ti i Qi ,
CF Q2i ,
TR 1,
CA 0,
(108)
where Ti indicates the colour of the emitting parton, Qi is the electric charge in units of the
positron charge for this parton, and i = +1 (1) for incoming (outgoing) partons. Adopting
notation analogous to Refs. [18,19], the subtraction term for the process ab W k can be
then written
a ,b

a
a
a
a
Mab
sub (3 ) = DQED + DW,QED + D W,QED + Dk,QED + D k,QED + (a b)
+ D k,W,QED + D W,k,QED ,
(109)
where

I
= FO,2 ( 2,nm )DFI (pW , pk , p ; pa , pb )replacements of Eq. (108) .
DF,QED
(110)
It is understood in Eq. (109) that dipole subtraction terms with a gluon index do not contribute
ab
to Mab
sub . In a complete analogy to Eq. (91), for any initial state ab the expression for Msub is conI
structed from six dipole subtraction terms DF,QED , each associated with one of the three possible
charge combinations Qq Qq , Qq or Qq . The dipole subtraction functions DFI are taken
directly from Refs. [18,19]. A list of the functions DFI used to calculate the subtraction term
Mab
sub in (109), together with the corresponding equation numbers in Refs. [18,19], is presented
in Table 2, Appendix F. Additionally, for each dipole subtraction term appearing in (109) we
include a description of its type. The mappings from 3 to 2,nm agree between the formalism
of Refs. [18,19] and [17]. However, for the sake of completeness, Table 2 contains numbers of
equations which provide mapping formulae in Refs. [18,19]. The function FO,2 in (110) is given
by expression (93).
Moreover, apart from the final-state emitter, final-state spectator case, i.e., the dipoles
D k,W,QED and D W,k,QED , there is a direct correspondence between the dipole subtraction terms
in the two formalisms of the form

a ,b 0
sub
DQED Qa Qb 4
gab,
(pa , pb , p )Mab
0 (2,ab ),
=
0
a
DW,QED Qa 4
Da W,QED Qa 4
a
Dk,QED Qa Qk 4
0
sub
gaW,
(pa , pW , p )Mab
0 (2,aW ),
Da k,QED Qa Qk 4
sub
ab
gW
a, (pW , pa , p )M0 (2,W a ),
sub
gak,
(pa , pk , p )Mab
0 (2,ak ),
sub
gka,
(pk , pa , p )Mab
0 (2,ka ).
(111)
54
The subtraction term integrated over the photon phase space is constructed according to

1
ab

d Aab
ab
ab
=
I (s , pT ) + dx K (x, s , pT ) + P (x, s , pT ) ,
(112)
dpT
2
0
where the expressions for I, K and P functions follow from results for the integrated dipole functions in Refs. [18,19] after performing replacements of Eq. (108). For the photonic corrections
to any of the subprocesses ab W k we can write

d abW k
sab ) fwd
I ab (s , pT ) = Qa Qb I(
(s , pT ) + (a b)
dpT

abW k
d fwd

Qa I (saW )
(s , pT ) + (a b)
dpT

d abW k
Qb I (sbW ) fwd
(s , pT ) + (a b)
dpT
k

abW
d fwd
Qa Qk I(sak )
(s , pT ) + (a b)
dpT

d abW k
sbk ) fwd
Qb Qk I(
(s , pT ) + (a b)
dpT
k

abW
d fwd

(s , pT ) + (a b) ,
+ Qk I (skW )
(113)
dpT
where snm = 2pn pm , and we make use of Qg = 0. As in (94), the terms proportional to
abW k /dp originate from the forward hemisphere and the (a b) terms from the backd fwd
T
ward hemisphere. The integrated dipole functions in (113) read

2
IR
snm
2 snm

snm ) = 4
(1
+
)
Re
f
ln
I(
+
3
ln
1,DR s =snm
2
2
2
MW
MW
MW
4 2
6,
3

IR
42

2 snm

(1 + ) Re f2,DR t=M 2 + s ln
I (snm ) =
2
n m nm
W
MW
M2

W
snm
snm
snm
ln
ln
3 ln
2
2
2
MW
MW + snm
MW + snm

2
2
MW
MW
snm
+ ln
ln
+ 3 ln 1
2 + s
2 + s
2 + s
MW
MW
MW
nm
nm
nm

2
2
M
MW
snm
3MW
+ W ln
+ 2Li2
+
2
2
snm
MW + snm
MW + snm
2 +M
snm + MW
W
+
+ 2 6.
(114)
IR in (114) is kept the same as in (75) to manifestly show

The structure of the singular terms fi,DR
cancellation of singularities between virtual and real corrections. For the x-dependent functions
55
we have

abW k

d fwd
Kab (x, s , pT ) = Q2a + Q2b K(x)

(x s , pT ) + (a b)
dpT
k

abW
d fwd
+ 2Qa Qb K(x)
(x s , pT ) + (a b)
dpT

d abW k
Qa K (x, saW ) fwd
(x s , pT ) + (a b)
dpT
k

abW
d fwd

(x s , pT ) + (a b)
Qb K (x, sbW )
dpT

d abW k
(Qa Qk + Qb Qk ) K (x) fwd
(x s , pT ) + (a b) ,
dpT

d abW k
P ab (x, s , pT ) = 2Qa Qb P(x, sab ) fwd
(x s , pT ) + (a b)
dpT

abW k
d fwd
Qa P(x, saW )
(x s , pT ) + (a b)
dpT

d abW k
Qb P(x, sbW ) fwd
(x s , pT ) + (a b)
dpT

abW k
d fwd
Qa Qk P(x, sak )
(x s , pT ) + (a b)
dpT

d abW k
Qb Qk P(x, sbk ) fwd
(x s , pT ) + (a b) ,
dpT
with

2
1x
1x
+ (1 x) +
ln
K(x)
= Preg (x) ln
x
1x
x
+

2
(1 x) 5 ,

ln(1 x)
2
K(x)
= Preg (x) ln(1 x) 2
(1 x) ,
1x
3
+

ln(1 x)
ln(2 x)
K (x, snm ) = 2
+2
1x
1x
+

2
MW
1x
2
1
+
ln
1
x
+
2 /
snm
2(1 x + MW
snm )2 1 x
+

2
M
(2 x)snm
2
ln 2 x + W + ln
2
1x +
snm
(2 x)snm + MW

(1 x)snm
Preg (x) ln
2
(1 x)snm + MW
(115)
56

2
MW
3 M2
3MW
(1 x) + W ln
+
2
2
snm
snm + MW
2 +M
snm + MW
W

s 2M

2
2
2
nm
W snm + MW + 2MW
MW
3
1
+ ln
+
,
2
2
snm
2 snm + MW

3
1

+ (1 x) ,
K (x) =
2 1x +

2

QED
1 + x2
ln
P(x, snm ) =
,
1x +
x snm
and

Preg (x) =
1 + x2
1x

+
2
1x
3
(1 x).
2
+
(116)
(117)
Note that in contrast to Eq. (113), the quantity snm in (115) can be implicitly dependent on the
fraction x. More precisely, it is the case if snm involves the momentum of a final-state particle.
The final-state momentum belongs then to the phase space for which the squared center-of-mass
energy is x s = 2xpa pb [19].
The evaluation of the terms involving the plus-distribution is carried out as indicated in
Ref. [19], i.e., according to
1
0

d
dx R x, snm (x) +
(x s , pT )
dpT
1
=
0

d

d
dx R x, snm (x)
(x s , pT ) (x ) R x, snm (1)
(s , pT ) ,
dpT
dpT
(118)

2 )2 /
s.
with = (pT + pT2 + MW
In the formalism of Refs. [18,19] the collinear counterterms associated with PDF renormalization are included in the expressions for integrated dipole functions, i.e., the final results which
we use are free from collinear singularities. The expressions presented here are calculated using
the MS factorization scheme.
As can be seen from the presented formulae, the explicit expressions for the integrated dipole
functions in the two formalisms are different. In particular, the expressions for the end-point contributions have different forms due to specific conventions wrt. calculating the plus-distribution
terms in the two formalisms. However we have checked that, after subtraction of the IR singularities, for each charge combination apart from Qk the integrated dipole contributions to
dAab /dpT in the two formalisms are equivalent.
6. Checks
Every part of the presented calculation has been performed in two completely independent
ways. The algebraic reductions were done using two different M ATHEMATICA [29] codes. For
the numerical evaluation we have implemented the results in two independent F ORTRAN programs. Comparing the results at numerical level we find agreement within the statistical errors.
57
Furthermore, in order to control the correctness of our results we performed various consistency checks. On the side of virtual corrections we have verified that the one-loop corrections
(48) satisfy the Ward Identity

(pW )pg v(p
(119)
q ) A1,I MV2 u(pq ) = 0 for I = A, N, X, Y.
A similar Ward identity holds for the lowest-order amplitude.7 The cancellation of the ultraviolet
divergencies has been verified analytically and numerically. For the numerical evaluation of the
loop integrals we use a set of routines by A. Denner and, alternatively, the FF library [30]. The
NLL approximation that was derived from the full one-loop calculation, has been checked against
results from the general derivation of NLL terms [14]. Also the IR-singular contributions in the
high-energy limit have been reproduced within this framework.
The squared matrix element for the real corrections was checked numerically against M AD G RAPH [31]. The cancellation of IR singularities between real and virtual corrections was done
analytically using the dipole formalism. The subtraction terms were derived and implemented
in two different ways, using the mass regularization of IR singularities and the dimensional regularization. The phase-space integration for the real corrections was performed with adaptive
Monte Carlo integration using VEGAS [32]. Detailed comparisons at analytical and numerical
level were performed, and the agreement between the predictions generated within two different
regularization schemes provided a strong check on the calculation of the real corrections.
7. Numerical results
In this section we present numerical predictions for the large-pT production of W bosons
at the hadron colliders LHC and Tevatron. The following input parameters are used: G =
1.16637 105 GeV2 , MW = 80.39 GeV, MZ = 91.19 GeV, mt = 171.4 GeV, MH =
120 GeV. For the numerical values of elements in the CKM quark mixing matrix we refer to [33].
The hadronic cross sections are obtained using LO MRST2001 PDFs [34]. We choose
2QCD = pT2 as the factorization scale and, similarly, as the scale at which the strong coupling constant is evaluated.8 We also adopt, in agreement with the value used in the PDF analysis, the value
S (MZ2 ) = 0.13 and use the one-loop running expression for S (2QCD ). In our calculations of
2 . We note
the real corrections we choose the MS factorization scheme with the scale 2QED = MW
that in order to consistently include O() corrections in a calculation of a hadronic cross section,
PDFs that are used in the calculation need to take into account QED effects. Such PDF analysis
has been performed in [35] and the O() effects are known to be small for QED 100 GeV,
both concerning the change in the quark distribution functions (below O(1%) [36]) and the size
of the photon distribution function. Moreover, the currently available PDFs incorporating O()
7 We note that the Abelian one-loop contribution satisfies two additional Ward identities

pW (pg )v(p
q ) A1,A MV2 u(pq ) = 0,

pW pg v(p
q ) A1,A MV2 u(pq ) = 0.
Similar identities for the N-, X- and Y-form factors exist but are less trivial due to the non-vanishing contributions
from would-be Goldstone bosons on the right-hand side. This means that the calculation of the unpolarized cross section
requires the use of the exact expression for the W -boson polarization sum. Instead, owing to (119), the gluon polarization
sum can be implemented as g .
8 Note that when calculating the contribution to the hadronic cross section coming from the subtraction term in the real
corrections, we take the transverse momentum of the W boson in the reduced phase space, p T,W , as the factorization
scale and the argument of S .
58
corrections, MRST2004QED [35], include QCD effects at the NLO in S . Since our calculations are of the lowest order in QCD, and QED effects on PDFs are estimated to be small for
QED 100 GeV, we prefer to use a LO QCD PDF set without QED corrections incorporated,
rather than MRST2004QED, and we set QED = MW .9 Moreover we do not include photoninduced contributions, which are parametrically suppressed by a factor /S . However, in the
concurrent to this paper (and subsequent to Ref. [21]), work of Ref. [22], it has been reported
that photon-induced contributions are of numerical significance for large pT W -boson production at the LHC. Estimates of the exact size of these effects are obscured by large theoretical
uncertainty on the photons PDF, as demonstrated in Ref. [22].
min = 100 GeV for LHC and p min = 50 GeV
We choose the following values of the pT -cuts: pT,j
T,j
for Tevatron. The value of the separation parameter below which the recombination procedure is
applied is taken to be Rsep = 0.4. The dependence of our predictions on Rsep is negligible. We
have verified that the shift of the transverse-momentum distribution induced by variations of this
parameter in the range 0.1 Rsep 1.0 does not exceed a few permille.
Our lowest-order (LO) predictions result from (32). The next-to-leading order predictions
(NLO) include the LO + virtual contributions (73) and the real bremsstrahlung (89). We also
study the relative importance of the IR-finite parts of the virtual (NLOvirt ) and real (NLOreal )
contributions. These IR-finite parts are constructed by subtracting the IR divergence (75) from
the virtual corrections and adding it to the real ones. The next-to-leading-logarithmic (NLL)
and next-to-next-leading-logarithmic (NNLL) predictions10 are obtained adding to the LO the
approximations (86) and (77) for the NLOvirt part and neglecting the NLOreal part of the corrections. As we will demonstrate, for the case of fully inclusive photon radiation neglecting this
piece provides a good approximation of the complete calculation. The next-to-next-to-leading
order predictions (NNLO) include the full NLO results plus the two-loop NLL corrections (87).
The LO transverse-momentum distributions for pp W + j and pp W j at the LHC are
shown in Fig. 5(a). In Figs. 5(b) and 5(c) we plot the relative size of the NLO, one-loop NLL,
one-loop NNLL and NNLO corrections wrt. the LO predictions for W + and W production,
respectively. The behaviour of the relative corrections to W + and W production is very similar.
As expected, the importance of the NLO contribution increases significantly with pT and leads
to a negative correction ranging from 15% at pT = 500 GeV to 43% at pT = 2 TeV. We also
observe that the one-loop NLL and NNLL approximations are in good agreement (at the 12%
level) with the full NLO result for pT 100 GeV. The difference between NLO and NNLO
curves is significant. The two-loop terms are positive and amount to +3% at pT = 1 TeV and
+9% at pT = 2 TeV. This shifts the relative corrections for W + production up to 25% at
pT = 1 TeV and 34% at pT = 2 TeV.
9 The use of different factorization scales,
QCD = pT and QED = MW , is due to the fact that QCD and QED
play a different role in our calculation. The dependence on QCD is due to the LO evolution of the PDFs and represents
an effect of O(S ln(QCD /0 )), where 0 is the scale at which the PDF evolution starts. This dependence would be
compensated by NLO QCD contributions of O(S ln(pT /QCD )) and, although QCD corrections are not included in our
calculation, choosing QCD = pT we can absorb large NLO QCD logarithms of the scale pT in the LO PDF evolution.
In contrast, the QED dependence of our predictions is due to O( ln(pT /QED )) terms in the photon bremsstrahlung
corrections. This dependence is not compensated by the PDF evolution since we use a PDF set that does not include
QED effects, assuming that these effects are negligible. This approach makes sense only if the scale QED is chosen in
such a way that the (potential) impact of QED effects on the PDFs is very small. In Ref. [36] it was shown that the QED
corrections to the PDFs grow with QED but do not exceed one percent for QED 100 GeV. This motivates our choice
QED = MW for the QED factorization scale.
10 For details concerning the treatment of angular-dependent logarithms at the NLL level we refer to Ref. [8].
59
Fig. 5. Transverse-momentum distribution for W -boson production at the LHC. (a) LO distribution for pp W + j
and pp W j . (b) Relative NLO (dotted), NLL (thin solid), NNLL (squares) and NNLO (thick solid) electroweak
correction wrt. the LO distribution for pp W + j . (c) Relative NLO (dotted), NLL (thin solid), NNLL (squares) and
NNLO (thick solid) electroweak correction wrt. the LO distribution for pp W j .
The IR-finite parts of the virtual (NLOvirt ) and real (NLOreal ) corrections to W + production
at the LHC are shown separately in Fig. 6(a). Fig. 6(b) shows the relative size of the NLOvirt and
NLOreal corrections wrt. the LO predictions. The NLOvirt contribution dominates the full NLO
correction and amounts up to 42% at pT = 2 TeV. The NLOreal part contributes with a smaller
and nearly constant correction of about 1% in the entire pT -range. This means that, for the case
of fully inclusive photon radiation, the NLOvirt part represents a good approximation of the full
NLO correction.
The high-energy behaviour of the NLOvirt part is described by the compact NLL and NNLL
approximations presented in Section 4. The quality of these approximations is shown in Fig. 7.
We observe that the NLL approximation works well differing from the exact NLOvirt result by
60
Fig. 6. IR-finite parts of the virtual (NLOvirt ) and real (NLOreal ) contributions to the pT -distribution of W bosons in the
process pp W + j at s = 14 TeV.
Fig. 7. Relative precision of the high-energy approximations at one loop in the process pp W + j at s = 14 TeV as
a function of pT : NNLL (solid) and NLL (dashed) wrt. the IR-finite part of the exact one-loop result (LO + NLOvirt ).
less than 1% for pT 200 GeV. The quality of the NNLL approximation is of the order of one
permille or better in the entire pT -range. For less inclusive observables where a veto on hard
photons is imposed, the NLOreal contribution can become important. Fig. 8 shows the relative
61
Fig. 8. Relative size of the real correction (NLOreal ) wrt. the LO for fully inclusive photon radiation and the case where
visible photons are rejected, plotted as a function of pT for the process pp W + j at s = 14 TeV.
Fig. 9. Relative NLO (dotted) and NNLO (solid) electroweak corrections wrt. the LO and statistical error (shaded area)
for the unpolarized integrated cross section for (a) pp W + j at s = 14 TeV and (b) pp W j at s = 14 TeV as
cut
a function of pT (W ).
62
Fig. 10. Ratio of the pT -distributions for the processes (a) pp W + j and pp W j and (b) pp W + j and
pp j at s = 14 TeV: LO (thin solid), NLO (dotted) and NNLO (thick solid) predictions.
NLOreal corrections for W + production. We compare the fully inclusive photon radiation with
the case where visible photons with pT, > 10 GeV and R( , j ) > 0.4 are rejected. This veto
leads to a significant enhancement of the (absolute size of the) NLOreal part, which can exceed
5% for pT 1 TeV.
To underline the relevance of the large electroweak corrections for W production at the LHC,
in Figs. 9(a) and 9(b) we present the relative NLO and NNLO corrections to the W + and W
cross sections integrated over pT starting from pT = pTcut , as a function of pTcut . This is compared
with the statistical error, defined as stat / = 1/ N with N = L BR LO . The branching

ratio BR = 2/9 accounts for the full efficiency of W -detection in the ee and modes (for
this estimate we ignore experimental efficiencies and cuts) and we assume a total integrated
luminosity L = 300 fb1 for the LHC [37]. It is clear that the size of the NLO corrections is
much bigger than the statistical error. Indeed, already for L = 3 fb1 and pT 800 GeV they
correspond to a two standard deviation effect. Also the difference between the NNLO and NLO
corrections, due to two-loop logarithmic effects, is significant. In terms of the estimated statistical
error, these two-loop contributions amount to 13 standard deviations for pT of O(1 TeV).
Ratios of pT -distributions for W + , W , Z bosons [9] and photons [10], in contrast to the
distributions themselves are expected to be relatively insensitive to QCD corrections and theoretical uncertainties associated with S and PDFs. These ratios lead to important experimental
tests of W and Z couplings in the high-energy region. For W + and W the ratio is presented in
Fig. 10(a). The LO value increases from 1.5 at pT = 100 GeV to 3.4 at pT = 2 TeV. As already
observed, the (relative) electroweak corrections to the W + - and W -boson pT -distributions are
63
Fig. 11. Ratio of the pT -distributions for the processes (a) pp W + j and pp Zj and (b) pp W j and pp Zj
at s = 14 TeV: LO (thin solid), NLO (dotted) and NNLO (thick solid) predictions.
almost identical. In consequence, the LO, NLO and NNLO curves in Fig. 10(a) overlap. In contrast, the impact of the electroweak corrections on the W + / ratio (Fig. 10(b)) at the LHC is
clearly visible. The LO prediction, ranging from 1.4 to 2.5, receives a negative NLO correction
that grows with pT and amounts to 0.5 for pT = 1 TeV. At pT = 2 TeV the difference between
the NNLO and NLO curves is about 0.2.
The ratios of pT -distributions for W + /Z and W /Z are shown in Figs. 11(a) and 11(b), respectively. For the W + /Z ratio the LO prediction ranges from 1.5 to 2. For pT 1 TeV it is
reduced by 0.09 to 0.18 by the NLO electroweak corrections. The logarithmic two-loop correc/Z ratio.
tions are small. A qualitatively similar behaviour is observed for the W
The results of a similar analysis for W + production at the Tevatron ( s = 2 TeV) are shown
in Figs. 1215 (the pT -distributions for W + and W production are obviously identical). The
LO pT -distribution is shown in Fig. 12(a), the relative NLO, NLL, NNLL and NNLO corrections
in Fig. 12(b). The NLO corrections grow with pT and reach 11% at pT = 400 GeV. The oneloop NNLL and NLL approximations describe the exact NLO results with 1% and 3% precision,
respectively. The dominant two-loop effects have little impact on the size of the corrections.
The quality of the high-energy approximations wrt. the IR-finite part of the virtual corrections
(NLOvirt ) is shown in Fig. 13. In the pT -range under consideration both approximations are less
precise than at the LHC, nevertheless the NNLL approximation in sufficient for all practical
purposes.
In Fig. 14 the relative NLO and NNLO corrections to the integrated cross section for pT pTcut
are compared with the expected statistical error for an integrated luminosity L = 7 fb1 [38]. The
64
Fig. 12. Transverse-momentum distribution for W -boson production at the Tevatron. (a) LO distribution for
pp W +() j (solid). (b) Relative NLO (dotted), NLL (thin solid), NNLL (squares) and NNLO (thick solid) electroweak correction wrt. the LO distribution for pp W +() j .
Fig. 13. Relative precision of the high-energy approximations at one loop in the process pp W +() j at s = 2 TeV
as a function of pT : NNLL (solid) and NLL (dashed) wrt. the IR-finite part of the exact one-loop result (LO + NLOvirt ).
size of the NLO electroweak corrections is above the statistical error for a significant range of
pT -values. Therefore they should be included in the analysis when considering precision measurements. In contrast, the impact of the dominant two-loop corrections is negligible.
65
Fig. 14. Relative NLO (dotted) and NNLO (solid) electroweak corrections wrt. the LO and statistical error (shaded area)
for the unpolarized integrated cross section for p p W +() j at s = 2 TeV as a function of pTcut (W ).
Fig. 15. Ratio of the pT -distributions (a) for the processes p p W +() j and p p Zj and (b) for the processes
pp W +() j and p p j at s = 2 TeV: LO (thin solid), NLO (dotted) and NNLO (thick solid) predictions.
The effect of the electroweak corrections on the ratios of pT -distributions for W/Z and for
W/ is shown in Figs. 15(a) and 15(b), respectively.
66
8. Summary
In this work the electroweak corrections to large transverse momentum production of W
bosons at the hadron colliders Tevatron and LHC were evaluated. The contributions from real
and virtual photons cannot be separated in a gauge-invariant manner from purely weak corrections and were thus included in our analysis. Soft and collinear singularities were regulated by
introducing a small quark mass and a small photon mass and, alternatively, by using dimensional
regularization. The real photon radiation was evaluated using the dipole subtraction formalism.
The agreement between the results derived in the two regularization schemes has been an important cross check of the calculation. Numerous additional tests were performed to ensure the
correctness of the result.
At the Tevatron, pT -values up to around 300 GeV can be reached with reasonable event rates.
In this region the O() electroweak corrections reach up to 10% and are thus of relevance
for precision measurements. Two-loop electroweak corrections are negligible at the Tevatron.
With pT below 400 GeV the relative rates for W , Z and production are hardly affected by
electroweak corrections.
In contrast, for transverse momenta in the TeV region accessible at the LHC, electroweak
corrections play an important role. The O() corrections lead to a reduction of the cross section
by about 15% at transverse momenta of 500 GeV and reach more than 40% at 2 TeV. The
logarithmically dominant terms were extracted from the exact expression of the virtual corrections and agreement was found with the predictions based on the process-independent analysis
of electroweak Sudakov logarithms. If no cuts on real photons are applied, the contribution of
the real photon emission is numerically small (about 1%) and almost independent of pT . Numerically the NLL and NNLL approximations give a good description of the full O() result with
an accuracy of about 12%. Considering the large event rate at the LHC, leading to a fairly good
statistical precision even at transverse momenta up to 2 TeV, we evaluated also the dominant
(NLL) two-loop terms. In the high-pT region, these two-loop logarithmic effects increase the
cross section by 510% and thus become of importance in precision studies. We also studied the
relative rates for W + , W , Z and production, which are expected to be stable with respect
to QCD effects. The electroweak corrections cancel almost completely in the W + /W ratio. In
contrast, their impact on the W + /Z and the W + / ratios is significant and leads to a shift of
O(10%) for pT 1 TeV.
Acknowledgements
We would like to thank S. Dittmaier, B. Jger and P. Uwer for helpful discussions. This work
was supported in part by BMBF Grant No. 05HT4VKA/3, the Sonderforschungsbereich Transregio 9 and the DFG Graduiertenkolleg Hochenergiephysik und Teilchenastrophysik.
Appendix A. Recombination and exclusive Wj cross section
As discussed in Section 2.2, the recombination prescription that we use to regularize photon
quark final-state collinear singularities implies a different treatment of final-state quarks and
gluons. While for final-state gluons we apply a cut on pT,g within the entire phase space, for
final-state quarks the recombination effectively removes the cut on pT,q inside the collinear cone
R(q, ) < Rsep . As a consequence the recombined gq W q cross section (18) has a loga-
67
rithmic dependence on the cut-off parameter Rsep . In order to quantify this Rsep -dependence, let
us consider the contribution of real photon radiation inside the recombination cone. To this end,
assuming that the cone is sufficiently small (Rsep 1), we adopt a collinear approximation

d
gq W q
gq W q
1
dz Fq (z),
(A.1)
R(q, )<Rsep
where11 z = pT, /(pT,q + pT, ) = 1 pT,q /pT,W is the photon momentum fraction and [18]
Q2q
(42 )
Fq (z) =
Pq (z, )
2
(1 )
=
Q2q
2
Pq (z)
2
k,max

0
2
dk
2 )1+
(k

(4)
+ Fq z, 2 ,
(1 )
(A.2)
with

Q2q

2
2
Fq z, =
Pq (z) ln 2
z .
2
k,max
(A.3)
Here Pq (z, ) = Pq (z) z with Pq (z) = [1 + (1 z)2 ]/z is the q splitting function
in 4 2 dimensions, k is the photon transverse momentum wrt. the photonquark system,
and k,max = z(1 z)Rsep pT,W . The 1/ collinear singularity resulting from inclusive photon
radiation, i.e., integrating over the complete energy spectrum 0 z 1, cancels against the
virtual corrections.
The Rsep -dependence of the recombined cross section (18) is due to the fact that, inside the
min (or equivalently photons with z > 1 p min /p
recombination cone quarks with pT,q < pT,j
T,W )
T,j
are not rejected. Thus the variation of rec. induced by a rescaling Rsep sep Rsep amounts to
gq W q
rec.
gq W q
Q2q
2
1
2
ln sep
dz Pq (z)
min
with zmin = 1 pT,j
/pT,W .
(A.4)
zmin
min ) a rescaling of R
For relatively small transverse momenta (pT,W 2pT,j
sep by a factor sep = 10
shifts the gq W q( ) cross section by less than 2 (0.5) permille for up- (down-) type quarks.
Moreover it is obvious that at high pT,W , where zmin 1, this effect tends to disappear.
Let us now compare the recombination procedure with a realistic definition of exclusive pp
Wj production, where final-state quarks (a = q) and gluons (a = g) are subject to the same cut
min within the entire phase space (including collinear quark-photon configurations).
pT,a > pT,j
Since the recombination procedure does not affect final-state gluons, only channels involving
final-state quarks need to be considered. The difference between the recombined gq W q
cross section (18) and the exclusive cross section (19) corresponds to the contribution of hard
collinear photons with R(q, ) < Rsep and zmin z 1. This collinear hard-photon radiation
11 Here we assume lowest-order kinematics, i.e., p

T,q + pT, = pT,W in the collinear region.
68
can be described by means of quark fragmentation functions [4247] as

gq W q
excl. = rec.
gq W q
excl.
gq W q
1
dz Dq (z).
(A.5)
zmin
Here the effective quark fragmentation function Dq (z) = Fq (z) + Dq (z) consists of the perturbative contribution Fq and the bare fragmentation function Dq . The collinear singularity
resulting from the perturbative contribution is factorized into the bare fragmentation function at
the scale , such that in the MS scheme [42]

Dq (z) = Fq z, 2 + D q z, 2 ,
(A.6)
and the renormalized fragmentation function D q can be extracted from experimental measurements. Using the parametrization [43,47]

Q2q

D q z, 20 =
Pq (z) ln(1 z)2 13.26 ,
2
obtained by the ALEPH collaboration at 0 = 0.14 GeV, we arrive at

Q2q
zRsep pT,W 2
Dq (z) =
+ z 13.26 .
Pq (z) ln
2
0
(A.7)
(A.8)
With this expression we derive a conservative upper bound for excl. . To this end we consider
min p
Qq = 2/3, Rsep 1, and a wide range of transverse momenta, 2pT,j
T,W 2 TeV. With
these parameters we obtain
excl.
2 103 .
(A.9)
We conclude that, for Rsep O(1), the recombined cross section has a negligible dependence on
the recombination parameter Rsep and provides a fairly precise description of exclusive pp
Wj production at high transverse momentum.
Appendix B. Standard matrix elements
The algebraic expressions involving external momenta, Dirac matrices, spinors and gaugeboson polarization vectors have been reduced to a set of 10 standard matrix elements
Si = v(p
q )Si L u(pq ) (pW ) (pg ),
(B.1)
with
S1 = ( p
/W p
/ q ) ,
S2 = ( p
/W p
/ g )g ,
S6 = pq ,
/W p
/ g )pq pq ,
S8 = ( p
S9 = ( p
/W p
/ g )pg pq ,
S4 = pg ,
S5 = pq ,
S3 = pW
,
S7 = ( p
/W p
/ g )pg pW
,
S10 = ( p
/W p
/ g )pq pW
.
(B.2)
These algebraic expressions correspond to the massless subset of the standard matrix elements
of Ref. [25].
69
Appendix C. Scalar loop integrals

In this appendix we list the scalar loop integrals Jj (MV2 ) that contribute to (48). The symbols
Jj are chosen in analogy with Ref. [9]. For convenience, to denote constant terms we define

J0 MV2 = 1.
(C.1)
For the scalar integrals A0 , B0 , C0 and D0 we adopt the notation of F EYN C ALC [39]. However,
we choose their normalization according to Ref. [25], i.e., we include the factor (2)4D which
is omitted in the conventions of F EYN C ALC.
The UV-divergent one- and two-point functions are denoted as

2
2
J1a MV2 = B0 m2 ; MV2 , m2 ,
J1b MV2 = B0 m2 ; MW
,
, m2 = J1a MW
2
2

2
2

2
2
2
2
J3 MV = B0 pW ; MW , MV ,
J2 MV = B0 pW ; m , m ,
2

J5a MV2 = B0 u;
MV2 , m2 ,
J4 MV = B0 s ; m2 , m2 ,

2

2
MW
,
J6a MV2 = B0 t; MV2 , m2 ,
, m2 = J5a MW
J5b MV2 = B0 u;

2
2
.
, m2 = J6a MW
J6b MV2 = B0 t; MW
(C.2)
The remaining loop integrals are free from UV singularities. The following three-point functions
are finite if MV and the W -boson transverse momentum are non-vanishing:

J7 MV2 = C0 s , m2 , m2 ; m2 , m2 , MV2 ,

2
pW
, m2 ; MV2 , m2 , m2 ,
J8 MV2 = C0 u,

2
2
pW
, m2 ; m2 , MW
, MV2 ,
J9a MV2 = C0 u,

2
2
pW
= J9a MV2 M 2 M 2 ,
, m2 ; m2 , MV2 , MW
J9b MV2 = C0 u,
V
W

2

, m2 ; MV2 , m2 , m2 = J8 MV2 tu ,
J10 MV2 = C0 t, pW

2

2
, m2 ; m2 , MW
, MV2 = J9a MV2 tu ,
J11a MV2 = C0 t, pW

2

2
= J9b (MV2 )tu .
, m2 ; m2 , MV2 , MW
J11b MV2 = C0 t, pW
(C.3)
In addition, the box diagrams b1b3 in Fig. 2, provide the following combinations of three- and
four-point functions

2
J12 MV2 = D0 m2 , 0, pW
, m2 , u,
s ; MV2 , m2 , m2 , m2
1
s u + (t + u)M
V2

2
2
C0 u,
pW
0 u,
0, m2 ; MV2 , m2 , m2
u pW
, m2 ; MV2 , m2 , m2 + uC

2
2
+ s pW
C0 s , pW
, 0; m2 , m2 , m2 ,

J13 MV2 = J12 MV2 tu ,

2

2
, m2 , 0, m2 , t, u;
MV2 , MW
, m2 , m2
J14a MV2 = D0 pW
2 , m2 , m2 )
tC0 (t, 0, m2 ; MV2 , m2 , m2 ) + uC
0 (u,
0, m2 ; MW
,
2 uM
V2
tu tMW

J14b MV2 = J14a MV2 M 2 M 2 = J14a MV2 tu .
(C.4)
70
For non-vanishing MV and W -boson transverse momentum, the functions J12 J14b are finite.
The fact that the scalar four-point functions in (C.4) appear always in combination with threepoint functions is due to the cancellation of the collinear singularities that are associated with the
gq q vertex [9]. Although these singularities are present in individual D0 and C0 functions, they
always cancel in the complete result for box diagrams.
Appendix D. Infrared singularities
The scalar integrals Ji (MV2 ) in Appendix C contain soft and collinear singularities that appear
when MV = MA 0 and m 0. As discussed in Section 3.4, these integrals are split into
IR-singular (IR) and IR-finite (fin) parts,

Ji MA2 = JiIR + Jifin .
(D.1)
The IR-singular parts depend on the scheme adopted to regularize soft and collinear singularities.
The IR-finite parts are scheme independent and free from soft-collinear singularities, but can
contain ultraviolet poles.
Let us start with the two-point functions (C.2). Here only J1a (MA2 ) gives rise to IR singularities. This integral is split into
2
m
IR
J1a,MR = ln
+ 1,
2
MW

42 (1 + )
IR
J1a,DR
=
1,
2
MW

42 (1 + )
fin
J1a
(D.2)
=
+ 1.
2
MW
We note that within dimensional regularization the UV and IR singularities cancel each other and
IR
fin = 0. The three-point functions J (M 2 )
+ J1a
the massless two-point function vanishes, J1a,DR
9b
A
2
and J11b (MA ) are free from IR singularities and the singularities originating from J8 (MA2 ) and
J10 (MA2 ) do not need to be considered since the coefficients associated with these scalar integrals
are of order MA2 (see Appendix E). The remaining three-point functions in (C.3) contain soft and
collinear singularities. For them we find

2
2
MW
1
1 2 MW
s
IR
+ ln
ln
,
J7,MR = ln
s
2
m2
2
m2

42 (1 + ) 1
1
s
IR
=
ln
,
J7,DR
2
2
2
s
MW
MW

1 1 2 s
2
fin
J7 =
(D.3)
ln
,
2
s 2
6
MW
and
IR
J9a,MR
2 2
2
MW
MW
MW
1
1
1
=
ln
ln
ln
2
2
2
2
2
m
m2
u MW

2
MW
u
ln 1 2
,
+ ln
2
MW
71

42 (1 + ) 1
1
u
ln
1
,
2
2
2
2 2
MW
u MW
MW

1
u
u
fin
2
J9a =
(D.4)
ln 1 2 + Li2
,
2
2
u MW
MW
MW
x
where Li2 (x) = 0 dt ln(1 t)/t. The finite and singular parts for J11a (MA2 ) = J9a (MA2 )|u
t
are constructed in the same way.
The singular parts of the subtracted four-point functions (C.4) can be related to the ones of
the three-point functions,
IR
J9a,DR
=

1
1 IR
IR
IR
IR
IR
J12
= J13
= J7IR ,
J14a
= J14b
u = J9a ,
tu
t
u
t
in both regularization schemes. This implicitly defines the remainders as
(D.5)

1
fin
fin
= J13
= J12 MA2 J7IR ,
J12
tu
u

1 IR
fin
fin
J14a
(D.6)
= J14b
= J14a MA2 J9a
.
tu
t
Using the explicit analytic expressions for the infrared singular four-point and three-point functions [40,41] we obtain

2
2
MW
MW
1 1 2 s
2
fin
2 s
J12 =
ln
2 Li2 1
2 Li2 1
,
ln
2
s u 2
u
s
u
2
MW

1
1 2 t
t
u
fin
=
ln
ln
2
ln
1
J14a
2 )
2
2
2
2
MW
MW
MW
t(u MW

2
u
+ Li2
(D.7)
.
2
2
MW
Appendix E. Explicit result for the virtual corrections
In this appendix we present explicit analytic expression for the functions H1I (MV2 ) defined
in (74). These functions describe the contribution of the unrenormalized Feynman diagrams of
Fig. 2 to the unpolarized cross section. They consist of linear combinations of the scalar integrals
defined in Appendix C,

I 2 2
H1I MV2 =
(E.1)
for I = A, N, X, Y.
Kj MV Re Jj MV
j
The coefficients of the function H1A (MV2 ) read

4s 2 + 3(t2 + u 2 )
5 5
1
1
4
K0A MV2 =
+
+
+ s
,
s + t s + u u
tu
t t + u

2
3s
1
3s
1
A
MV = MV2
+
+
+
K1a
2
s + t s + u
(s + u)
2
(s + t)

2
s + u s + t
2
2s (2s + t + u)
(s + t )2 + (s + u)
2
+
+
+
4
,
u 2
t2
tu(
s + t )(s + u)
tu
72
2
A
MV = 0,
K1b

2
K2A MV2 = pW
6s MV2
3
6s MV2
3
2s MV2
2
2s MV2
2
4(s + t + u)
3
t

(s + u)
t
(s + t)
(s + t ) u (s + u)
(t + u)
2
2
2
2
s
+
u
2M
s
(2
s
+
t
+
u)(2M
s)
2
s
+
t
2M
3
V
V
V + 3
+
,
+
2
2
u
tu(
s + t )(s + u)
(s + u)
(s + t )

K3A MV2 = 0,

4s (s + 2t + 2u)
,
K4A MV2 =
2
(t + u)
2
6MV2 s u
MV2 (2u 5s ) s u 2MV2 (s + t + u)
MV2 + 4s + u
A
MV =
+
+
,
K5a
3
2
u 2
s + t
(s + t )
(s + t )
2
A
MV = 0,
K5b

2
A
A
MV2 = K5a
MV tu ,
K6a

A
MV2 = 0,
K6b

s
K7A MV2 =
+ t2 + u 2 ,
2 s + MV2 (t + u)
tu
2 M2

pW
V
2
2tMV2 (u s t ) 4pW
s s + t + MV2 ,
K8A MV2 =
2
u(
u pW )3
2
2
A
A
K9a MV = K9b
MV = 0,
2
2
A
A
K10 MV = K8 MV tu ,
2
2
A
A
MV = K11b
MV = 0,
K11a
2

M 2 (t + u)
+ s u
A
K12
MV = V
2 s + MV2 s + MV2 + t + t2 ,
tu
2
2
A
A
MV = K12
MV tu ,
K13
2
2
A
A
MV = K14b
MV = 0.
K14a
H1A (MV2 )
(E.2)
The only difference between

and the equally named function in Ref. [9] is due to the
A
2
fact that H1 (MV ) in Ref. [9] includes the contribution of the fermionic wave-function renormalA (see Eqs. (54) and (55) in Ref. [9]).
ization constants, which modify the coefficients K0A and K1a
N
2
For the coefficients of the function H1 (MV ) we obtain

2 4s
2
1
1
t u
N
+
+
+
2s
+2
,
K0 MV = (s + t + u)
s + u s + t t + u
u
tu
t

2

M 2 4s
1
1
N
K1a
MV = V
+
+ tu 2s (s + t + u)
2 tu
s + t )2
t(s + u)
2 u(
2
2
s (4s + 3t + 3u)
+ t + u
,
t u

N
N
MV2 = K1a
MV2 M 2 M 2 ,
K1b
V
W

K2N MV2 = K2A MV2 ,
73

1
1
1
+
2
tu (s + t )2 (s + u)
(s + u)
3 (s + t )3

4
2
s + t + u)
+ MW
+ MV2
s 2t2 u 2 + s 2 (t + u)(2

s 2 (s + t + u)
tu(2
s t u)
,

K4N MV2 = K4A MV2 ,

2 2s (s + t ) 2tu
s + t
u(2
s + t )
2s
2s + t
N
2 1
K5a MV =
MV
2
u
u 2(s + t )2
s + t 2(s + t )2
2(s + t )

2

2s
u(2
s t )
2
+ 2MV MW
+
,
(s + t )2
(s + t )3

2 2s (s + t ) 2tu
s + t
u(2
s + t )
2s
2s + t
N
2 1
+
K5b MV =
+ MW
2
u
u 2(s + t )2
s + t 2(s + t )2
2(s + t )

u(2
s t )
2s
2
MV
+
,
(s + t )2
(s + t )3

N
N
MV2 = K5a
MV2 tu ,
K6a
2
2
N
N
MV = K5b
MV tu ,
K6b

K7N MV2 = K7A MV2 ,

K8N MV2 = K8A MV2 ,

2
2s 2 + ts + t2 t s
s 2 s t s 2 + s t 2
N
K9a
MV = u
+ MW + MV2
+
2u
2t
2tu
2t
2 2

2
2
2
2
MV MW s
s t ) 2(MW + MV2 )s
2MV tu
MV u(2
+
2
2
s + t
s + t
tu
(s + t )
(s + t )
2
2
N
N
MV = K9a
MV M 2 M 2 ,
K9b
V
W

2
N
A
K10
MV2 = K8N MV2 tu = K10
MV ,

N
N
MV2 = K9a
MV2 tu ,
K11a

N
N
K11b
MV2 = K9b
MV2 tu ,
2
2
N
A
MV = K12
MV ,
K12
2
2

N
N
K M =K M
= K A M 2 ,

2

K3N MV2 = MW
+ MV2
13
12
tu
3s t
3s u
13

2
M 2 t + MV2 u tu 2 2
N
MV = W
MW
+ MV2 2tu
2MW MV + (2s + t + u)
K14a
2tu
s (t + u)
,
2
2
N
N
M 2
K
M =K
2 .
14b

2
14a
MV MW
(E.3)
For MV = MW the function H1N (MV2 ) is identical to the equally named function in Ref. [9].
The only non-vanishing coefficients of the function H1X (MV2 ) read

t2 + u 2 + s (t + u)
K0X MV2 =
,
tu
74
2
X 2
2
X
X
MV = K5a
MV + K6a
MV ,
K1a
2 2(MV2 u)(
s + t )
X
K5a
MV =
,
u 2

X
X
K6a
MV2 = K5a
MV2 tu .
Finally, for H1Y (MV2 ) we obtain

(u t )(s + t + u)
K0Y MV2 =
,
t u

2
3s
1
1
2
3s
Y
2 2
+

K1a MV = MV
+
2 s + u
t (s + t )2 s + t u (s + u)

2(s + u)
2(s + t )
,
+ MV2
2
u 2
t

2
3s
1
1
2
3s
Y
2 2
K1b MV = MW
+

+
2 s + u
t (s + t )2 s + t u (s + u)

2(s + u)
2(s + t )
,
u
t

K2Y MV2 = 0,
2 M 2 )(t u)(

2(MW

s + t + u)

V
K3Y MV2 =
7s 3 + tu 6s 2 (t + u)
3
3
(s + t ) (s + u)
2

s 2t tu + 2u 2 ,

K4Y MV2 = 0,

2
4s
2s 2 + 3tu + 2s (t + u)
2u(2
s t )
Y
2
K5a
MV =
M
+
W
(s + t )2
(s + t )2
(s + t )3

2s + t
2 2(s + t )
,

MV2
u 2
(s + t )2 u

2 2s 2 + 3tu + 2s (t + u)
4s
2u(2
s t )
Y
2
K5b MV =
+ MV
+
(s + t )2
(s + t )2
(s + t )3

2s + t
2(s + t )
2
2
+ MW
2
u
u
(s + t )
2
Y
Y
K6a
MV tu ,
(MV2 ) = K5a

Y
Y
MV2 = K5b
MV2 tu ,
K6b

K7Y MV2 = 0,

K8Y MV2 = 0,
2

4 2
1
Y
MV =
K9a
2MV tu 2s tu + 2s (t + u)
3
(s + t ) tu

MV2 (s + t ) (s + t )2 2s 2 + s t + t2 + (s t )(s + t )2 u 4t2 u 2

2
+ 2MW
s s (s + t ) 2tu (s + t )3 s 2 (t u)
+ 2tu 2

2
2s 2 + t(t u)
+ s t(t + u)
+ MW
+ s (t + u)
(E.4)
75
Table 1
Dipole subtraction terms from Ref. [17] used to calculate Mab
sub in (91) for the massive regularization (IS = initial-state,
FS = final-state)
Dipole
Type (emitter, spectator)
Eq. No.
2,nm
sub
gab,
sub
gaW,
sub
gW
a,
sub
gak,
sub
gka,
sub
gkW,
sub
gW
k,
massless IS, massless IS
(3.22)
(3.25)(3.27)
massless IS, massive FS
(A.1)
(3.12)
massive FS, massless IS
(A.1)
(3.12)
massless IS, massless FS
(3.9)
(3.12)
massless FS, massless IS
(3.9)
(3.12)
massless FS, massive FS
(4.4)
(4.5)
massive FS, massless FS
(4.4)
(4.5)
Table 2
Dipole expressions from Refs. [18,19] used to calculate Mab
sub in (109) for the dimensional regularization
Dipole
a ,b
DQED
a
DW,QED
Da W,QED
a
Dk,QED
Da k,QED
D k,W,QED
D W,k,QED
Type (emitter, spectator)
Eq. Nos.
2,nm
massless IS, massless IS
(5.136), (5.145) in Ref. [18]
(5.137), (5.139), (5.140) in Ref. [18]
massless IS, massive FS
(5.71), (5.81) in Ref. [19]
(5.73), (5.74) in Ref. [19]
massive FS, massless IS
(5.40), (5.50) in Ref. [19]
(5.42), (5.43) in Ref. [19]
massless IS, massless FS
(5.61), (5.65) in Ref. [18]
(5.62)(5.64) in Ref. [18]
massless FS, massless IS

massless FS, massive FS
massive FS, massless FS
(5.36), (5.39) in Ref. [18]

(5.2), (5.16) in Ref. [19]
(5.2), (5.16) in Ref. [19]
(5.37), (5.38) in Ref. [18]

(5.3), (5.7), (5.9) in Ref. [19]
(5.3), (5.7), (5.9) in Ref. [19]
2
2
Y
Y
MV = K9a
MV M 2 M 2 ,
K9b
V
W

Y
K10
MV2 = 0,
2
2
Y
Y
MV = K9a
MV tu ,
K11a

Y
Y
K11b
MV2 = K9b
MV2 tu ,
2
Y
K12
MV = 0,
2
Y
K13 MV = 0,
2
2
Y
N
K14a
MV = 2K14a
MV ,
2
2
2
Y
N
Y
K14b MV = 2K14b MV = K14a
MV M 2 M 2 .
V
(E.5)
Appendix F. Real corrections

In Tables 1 and 2 we list the dipoles that were used to calculate the subtraction terms in (91)
for the massive regularization and in (109) for the dimensional regularization, respectively. We
give references to the explicit formulae for the dipole terms and the phase-space mappings in the
original paper [17] and [18,19].
References
[1] G. Arnison, et al., UA1 Collaboration, Phys. Lett. B 122 (1983) 103;
76
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
G. Arnison, et al., UA1 Collaboration, Phys. Lett. B 126 (1983) 398;

M. Banner, et al., UA2 Collaboration, Phys. Lett. B 122 (1983) 476;
P. Bagnaia, et al., UA2 Collaboration, Phys. Lett. B 129 (1983) 130.
J.H. Khn, A.A. Penin, V.A. Smirnov, Eur. Phys. J. C 17 (2000) 97, hep-ph/9912503;
M. Ciafaloni, P. Ciafaloni, D. Comelli, Phys. Rev. Lett. 100 (2000) 4810, hep-ph/0001142;
A. Denner, S. Pozzorini, Eur. Phys. J. C 18 (2001) 461, hep-ph/0010201;
J.H. Khn, et al., Nucl. Phys. B 616 (2001) 286, hep-ph/0106298.
B. Jantzen, et al., Nucl. Phys. B 731 (2005) 188, hep-ph/0509157;
B. Jantzen, et al., Nucl. Phys. B 752 (2006) 327, Erratum.
V.S. Fadin, et al., Phys. Rev. D 61 (2000) 094002, hep-ph/9910338;
M. Melles, Phys. Rev. D 63 (2001) 034003, hep-ph/0004056;
W. Beenakker, A. Werthenbach, Nucl. Phys. B 630 (2002) 3, hep-ph/0112030;
A. Denner, M. Melles, S. Pozzorini, Nucl. Phys. B 662 (2003) 299, hep-ph/0301241;
S. Pozzorini, Nucl. Phys. B 692 (2004) 135, hep-ph/0401087.
S. Dittmaier, M. Krmer, Phys. Rev. D 65 (2002) 073007, hep-ph/0109062.
U. Baur, O. Brein, W. Hollik, C. Schappacher, D. Wackeroth, Phys. Rev. D 65 (2002) 033007, hep-ph/0108274;
E. Accomando, A. Denner, S. Pozzorini, Phys. Rev. D 65 (2002) 073003, hep-ph/0110114;
W. Hollik, C. Meier, Phys. Lett. B 590 (2004) 69, hep-ph/0402281;
U. Baur, D. Wackeroth, Phys. Rev. D 70 (2004) 073015, hep-ph/0405191;
E. Accomando, A. Denner, A. Kaiser, Nucl. Phys. B 706 (2005) 325, hep-ph/0409247;
E. Accomando, A. Denner, C. Meier, Eur. Phys. J. C 47 (2006) 125, hep-ph/0509234;
J.H. Khn, A. Scharf, P. Uwer, Eur. Phys. J. C 45 (2006) 139, hep-ph/0508092;
J.H. Khn, A. Scharf, P. Uwer, hep-ph/0610335;
S. Moretti, M.R. Nolten, D.A. Ross, Phys. Lett. B 639 (2006) 513, hep-ph/0603083;
C.M. Carloni Calame, G. Montagna, O. Nicrosini, A. Vicini, JHEP 0612 (2006) 016, hep-ph/0609170;
W. Bernreuther, M. Fuecker, Z.G. Si, Phys. Rev. D 74 (2006) 113005, hep-ph/0610334.
E. Maina, S. Moretti, D.A. Ross, Phys. Lett. B 593 (2004) 143, hep-ph/0403050;
E. Maina, S. Moretti, D.A. Ross, Phys. Lett. B 614 (2005) 216, Erratum.
J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, Phys. Lett. B 609 (2005) 277, hep-ph/0408308.
J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, Nucl. Phys. B 727 (2005) 368, hep-ph/0507178.
J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, JHEP 0603 (2006) 059, hep-ph/0508253.
U. Baur, Phys. Rev. D 75 (2007) 013005, hep-ph/0611241.
A. Denner, B. Jantzen, S. Pozzorini, Nucl. Phys. B 761 (2007) 1, hep-ph/0608326.
D. Wackeroth, W. Hollik, Phys. Rev. D 55 (1997) 6788, hep-ph/9606398;
U. Baur, S. Keller, D. Wackeroth, Phys. Rev. D 59 (1999) 013002, hep-ph/9807417.
A. Denner, S. Pozzorini, Eur. Phys. J. C 21 (2001) 63, hep-ph/0104127.
A. Denner, M. Melles, S. Pozzorini, Nucl. Phys. B 662 (2003) 299, hep-ph/0301241.
M. Melles, Phys. Rep. 375 (2003) 219, hep-ph/0104232;
M. Melles, Eur. Phys. J. C 24 (2002) 193, hep-ph/0108221.
S. Dittmaier, Nucl. Phys. B 565 (2000) 69, hep-ph/9904440.
S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291, hep-ph/9605323;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1998) 503, Erratum.
S. Catani, S. Dittmaier, M.H. Seymour, Z. Trocsanyi, Nucl. Phys. B 627 (2002) 189, hep-ph/0201036.
R.K. Ellis, G. Martinelli, R. Petronzio, Nucl. Phys. B 211 (1983) 106;
P.B. Arnold, M.H. Reno, Nucl. Phys. B 319 (1989) 37;
P.B. Arnold, M.H. Reno, Nucl. Phys. B 330 (1990) 284, Erratum;
R.J. Gonsalves, J. Pawlowski, C.F. Wai, Phys. Rev. D 40 (1989) 2245;
W.T. Giele, E.W.N. Glover, D.A. Kosower, Nucl. Phys. B 403 (1993) 633, hep-ph/9302225;
J. Campbell, R.K. Ellis, D.L. Rainwater, Phys. Rev. D 68 (2003) 094021, hep-ph/0308195.
J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, hep-ph/0703283;
J.H. Khn, A. Kulesza, S. Pozzorini, M. Schulze, Phys. Lett. B 651 (2007) 160.
77
[22] W. Hollik, T. Kasprzik, B.A. Kniehl, arXiv: 0707.2553;

W. Hollik, T. Kasprzik, B.A. Kniehl, Nucl. Phys. B 790 (2008) 138.
[23] S. Pozzorini, Doctoral thesis, Universitt Zrich, 2001, hep-ph/0201077.
[24] G. Passarino, M.J.G. Veltman, Nucl. Phys. B 160 (1979) 151.
[25] A. Denner, Fortschr. Phys. 41 (1993) 307.
[26] A. Sirlin, Phys. Rev. D 22 (1980) 971;
W.J. Marciano, A. Sirlin, Phys. Rev. D 22 (1980) 2695;
W.J. Marciano, A. Sirlin, Phys. Rev. D 31 (1985) 213, Erratum;
W.J. Marciano, A. Sirlin, Nucl. Phys. B 189 (1981) 442.
[27] M. Bhm, A. Denner, H. Joos, Gauge Theories of the Strong and Electroweak Interaction, Teubner, Stuttgart,
Germany, 2001, 784 p.
[28] M. Roth, A. Denner, Nucl. Phys. B 479 (1996) 495, hep-ph/9605420.
[29] Wolfram Research, Inc., Mathematica, Version 4.2, Champaign, IL, 2002.
[30] G.J. van Oldenborgh, Comput. Phys. Commun. 66 (1991) 1.
[31] T. Stelzer, W.F. Long, Comput. Phys. Commun. 81 (1994) 357.
[32] G.P. Lepage, Cornell preprint CLNS 80-447.
[33] S. Eidelman, et al., Particle Data Group, Phys. Lett. B 592 (2004) 1.
[34] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216, hep-ph/0201127.
[35] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 39 (2005) 155, hep-ph/0411040.
[36] M. Roth, S. Weinzierl, Phys. Lett. B 590 (2004) 190, hep-ph/0403200.
[37] V.A. Mitsou, hep-ph/0004161.
[38] R. Dixon, Presentation at the Fermilab Users Meeting, Batavia, IL, June 2007. http://www.fnal.gov/orgs/fermilab_
users_org/users_mtg/2007/agenda-for-web.htm.
[39] R. Mertig, M. Bhm, A. Denner, Comput. Phys. Commun. 64 (1991) 345;
http://www.feyncalc.org/.
[40] W. Beenakker, A. Denner, Nucl. Phys. B 338 (1990) 349.
[41] S. Dittmaier, Nucl. Phys. B 675 (2003) 447, hep-ph/0308246.
[42] E.W.N. Glover, A.G. Morgan, Z. Phys. C 62 (1994) 311.
[43] D. Buskulic, et al., ALEPH Collaboration, Z. Phys. C 69 (1996) 365.
[44] L. Bourhis, M. Fontannaz, J.P. Guillet, Eur. Phys. J. C 2 (1998) 529, hep-ph/9704447.
[45] M. Fontannaz, J.P. Guillet, G. Heinrich, Eur. Phys. J. C 21 (2001) 303, hep-ph/0105121.
[46] M. Klasen, Rev. Mod. Phys. 74 (2002) 1221, hep-ph/0206169.
[47] A. Gehrmann-De Ridder, T. Gehrmann, E. Poulsen, Eur. Phys. J. C 47 (2006) 395, hep-ph/0604030.

Phantom-like behaviour in dilatonic brane-world

scenario with induced gravity
Mariam Bouhmadi-Lpez
Centro Multidisciplinar de Astrofsica-CENTRA, Departamento de Fsica, Instituto Superior Tcnico,
Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal
Received 27 June 2007; accepted 21 December 2007
Abstract
The Dvali, Gabadadze and Porrati (DGP) model has a self-accelerating solution, the positive branch,
where the brane is asymptotically de Sitter. A de Sitter spacetime can be seen as a boundary between
quintessence-like behaviour and phantom-like behaviour. We show that in a 5D dilatonic bulk, where the
dilaton has an exponential potential, with an induced gravity term on the brane, whose matter content
corresponds only to vacuum energy, the positive branch solution undergoes a phantom-like stage where
it faces a curvature singularity in its infinite future. The singularity can be interpreted as the big rip
singularity pushed towards an infinite future cosmic time. The phantom-like behaviour on the brane occurs
without violating the null energy condition. There is another solution, the negative branch, where the brane
can undergo an early-epoch (transient) inflationary phase induced by the dilaton field.
PACS: 95.36.+x; 98.80.-k; 04.50.+z; 98.80.Es
Keywords: Dark energy; Phantom energy; Inflation; Cosmology with extra dimensions; Future singularities
1. Introduction
The supernova Ia (SNIa) observations [1] and the cosmic microwave background (CMB)
anisotropy data [2] suggest that the expansion of our universe seems to be accelerating. A possible explanation for this evolution is the usual vacuum energy represented by a cosmological
constant providing a negative pressure [3]. However, the observational value of is about 120
orders of magnitude smaller than that established from field theory methods [3]. So far, alternaE-mail address: mariam.bouhmadi@fisica.ist.utl.pt.
doi:10.1016/j.nuclphysb.2007.12.025
M. Bouhmadi-Lpez / Nuclear Physics B 797 (2008) 7892
79
tive phenomenological models have been proposed to describe the late-time acceleration of the
universe [4].
One approach is to consider an effective dark energy component in the energy momentum tensor. For example, this component can be described by a scalar field as in quintessence models [5]
or tachyonic models [6]. Dark energy can also be described effectively by a perfect fluid with
a linear equation of state or a more general barotropic equation of state like in Chaplygin gas
models [7]. Motivated initially by SNIa observations, phantom energy models [811] have also
been proposed to account for the late-time acceleration of the universe. A recent analysis [12] of
the current equation of state for dark energy based on CMB anisotropies, SNIa and X-ray galaxy
cluster data concluded that the current equation of state is compatible with a phantom-like behaviour of dark energy; i.e., w, the ratio of the pressure and the energy density of dark energy,
could be less than 1. A similar conclusion is reached in [13]; i.e., observational data do not
seem incompatible with phantom-like behaviour of dark energy. In phantom energy models the
null energy condition is not satisfied. Hence, the energy density is an increasing function of the
scale factor in an expanding FriedmannLematreRobertsonWalker (FLRW) universe. This
may lead to the occurrence of a big rip singularity in the future evolution of a phantom energy
dominated universe [911].
An alternative approach to account for the late-time acceleration of the universe is to consider
a generalised Einstein theory of gravity like brane-world models (for reviews, see Ref. [14])
where the observable four-dimensional (4D) universe is a brane (hyper-surface) embedded
in a higher dimensional space (bulk). For example, the Dvali, Gabadadze and Porrati (DGP)
model [15] has a self-accelerating solution at late-time which is asymptotically de Sitter [16,17].
In this model the bulk is a 5D Minkowski spacetime and the brane action contains an induced
gravity term [1524] which is proportional to the 4D Ricci scalar curvature of the brane. Dilatonic
brane-world models [2528] can also account for late-time acceleration of the brane through a
quintessence-like behaviour driven by a bulk dilaton [29]. One aim of this paper is to show that a
dilatonic brane-world model with an induced gravity term in the brane can mimic a phantom-like
behaviour without including matter on the brane that violates the null energy condition.
In an induced gravity brane-world model the Friedmann equation has two solutions (depending on the embedding of the brane in the bulk [16]). One of these solutions (the self-accelerating
solution or the positive branch) can account for the late-time evolution of the universe [16,17].
The other solution (the negative branch) can describe the early-time evolution of the universe1
[2024] and in particular corresponds to a correction to RandallSundrum (RS) model [30]. In
the dilatonic brane-world model with induced gravity that we will analyse, the negative branch
may undergo a transient inflationary epoch.
The layout of this paper is as follows. In Section 2, we present our dilatonic brane-world model
with induced gravity. The bulk scalar field potential is a Liouville potential. The matter content
of the brane is coupled to the dilaton field. We deduce the modified Friedmann equation for
both branches, the junction condition of the dilaton across the brane, which constrains the brane
tension, and the energy balance on the brane. In Section 3, we analytically derive the solutions
of a vacuum brane for both branches; i.e., whose matter content is described through the brane
tension. We show that the brane tension has a phantom-like energy density behaviour on the
positive branch in the sense that the brane tension grows as the brane expands. The brane hits
a singularity in its future evolution which may be interpreted as a big rip singularity pushed
1 For an alternative use of the negative branch of the DGP scenario see Ref. [18].
80
towards an infinite cosmic time. The negative branch can undergo an early-epoch (transient)
inflationary phase induced by the dilaton field through the brane tension. Finally, we conclude
and summarise in Section 4.
2. The model
We consider a brane, described by a 4D hyper-surface (h, metric g), embedded in a 5D bulk
spacetime (B, metric g (5) ), whose action is given by

1
1
1
S = 2 d 5 X g (5) R g (5) + L5 + d 4 X g 2 K + L4 ,
(2.1)
2
5
5
B
where 52 is the 5D gravitational constant, R[g (5) ] is the scalar curvature in the bulk and K
the extrinsic curvature of the brane in the higher dimensional bulk, corresponding to the York
GibbonsHawking boundary term [31].
We consider that there is a dilaton field living on the bulk and we choose to be dimensionless. Then, the 5D Lagrangian L5 can be written as
1
L5 = ()2 V ().
2
The 4D Lagrangian L4 corresponds to

L4 = R[g] () + 4 Lm 2 g .
(2.2)
(2.3)
The first term on the right-hand side (rhs) of the previous equation corresponds to an induced
gravity term [1524], where R[g] is the scalar curvature of the induced metric on the brane
and is a positive parameter which measures the strength of the induced gravity term and has
dimensions of mass squared. The term Lm in Eq. (2.3) describes the matter content of the brane
and () is the brane tension. We allow the brane matter content to be non-minimally coupled
(5)
on the (5D) Einstein frame but to be minimally coupled respect to a conformal metric g AB =
(5)
2 gAB , where = ().
We are mainly interested in the cosmology of a homogeneous and isotropic brane, with an
induced gravity term [22,23], and embedded in a bulk with a dilaton field [25]. It is known that
for an expanding FLRW brane the unique bulk spacetime in Einstein gravity (in vacuum) is a 5D
Schwarzschildanti de Sitter spacetime [32,33]. This property in principle cannot be extended
to a 5D dilatonic bulk. On the other hand, we stress that the presence of an induced gravity term
in the brane-world scenario affects the dynamics of the brane, through the junction conditions at
the brane, and does not affect the bulk field equations. Consequently, in order to study the effect
of an induced gravity term in a brane-world dilatonic model, it is possible to consider a bulk
corresponding to a dilatonic 5D spacetime and later on impose the junction conditions at the
brane. The junction conditions will then determine the dynamics of the brane and constrain the
brane tension. This is the approach we will follow.
2.1. The bulk
From now on, we will restrict our attention to a 5D dilatonic solution obtained by Feinstein
et al. [28] and analysed by Kunze and Vzquez-Mozo in the context of quintessence brane-world
81
models [29] without an induced gravity term on the brane. The 5D dilatonic solution reads [29]
ds52 =

1 2/3(k 2 3) 2
r
dr + r 2 dt 2 + ij dx i dx j ,
2
(2.4)
where ij is a 3D spatially flat metric. The bulk potential corresponds to a Liouville potential;
i.e.,

V () = exp (2/3)k .
(2.5)
The parameters k and in Eq. (2.4) measure the magnitude of the 5D cosmological constant
in the Liouville potential

1 2
(2.6)
k 12 2 .
2
On the other hand, the 5D scalar field scales logarithmically with the radial coordinate r [29]
=
= k log(r).
(2.7)
We will consider only the case k > 0 (the main conclusions of the paper do not depend on the
sign of k); i.e., the scalar field is a growing function of the coordinate r. The metric (2.4) has a
naked singularity at r = 0: the Ricci scalar diverges for vanishing r,
4

2
R g (5) = 2 2k 2 15 r (2/3)k ,
3
and the singularity is not surrounded by an event horizon.
(2.8)
2.2. The brane

We consider a homogeneous and isotropic brane embedded in the previous 5D dilatonic solution Eq. (2.4) and whose trajectory on the bulk is described by the following parametrisation
t = t ( ),
r = a( ),
xi = constant,
i = 1, . . . , 3,
(2.9)
where corresponds to the proper time of the brane. Then the brane metric reads
ds42 = g dx dx = d 2 + a 2 ( )ij dx i dx j .
(2.10)
For simplicity we will assume a Z2 -symmetry at the brane (which is also motivated by specific
M-theory constructions [34,35]). For an induced gravity brane-world model [16,24], there are
two physical ways of embedding the brane on the bulk when a Z2 -symmetry across the brane is
assumed. We will refer to a brane as the positive branch when the location of the brane r = a( )
is such that
ds52 =

1 2/3(k 2 3) 2
r
2
r > a( ).
(2.11)
On the other hand, we will refer to a brane as the negative branch when the location of the brane
r = a( ) is such that
ds52 =

1 2/3(k 2 3) 2
r
2
r < a( ).
(2.12)
82
The brane tension is constrained by the scalar field junction condition across the brane [25]

52
() (ln ) ()T , T = T ,
2
where a prime stands for derivative respect to the scalar field and T is defined as
n a a =
(2.13)

2

T =
g 4 Lm .
g g
(2.14)
We will consider that T is described effectively by a perfect fluid with energy density and
pressure p. For simplicity, we will consider that the matter content of the brane is minimally
(5)
(5)
coupled respect to the conformal metric g AB = exp(2b)gAB ; i.e., = exp(b), where b is a
constant.
The evolution equation for the scale factor of the brane, a, is given by the Israel junction
conditions [25,26]

52
1
K =
(2.15)
S Sg , S = S ,
2
3
where S is the energymomentum tensor of the brane

2

S =
gL4 ,
g g
(2.16)
and can be split as

S = ()g + T 2G .
(2.17)
The last term on the rhs of the previous equation corresponds to the effect of the induced gravity
term on the energy momentum tensor of the brane. This term is proportional to the Einstein
tensor of the brane G . We would like to stress that although the induced gravity effect does
not appear explicitly in the junction condition for the scalar field Eq. (2.13) (because there is
no coupling between the scalar field and the induced gravity term), the brane tension will feel the
induced gravity effect through the modified Friedmann equation.
The ij components of Israel junction condition Eq. (2.15) imply:

2
2 2
(2.18)
2 a 3 k + H 2 = 5 () + 6H 2 ,
6
where = 1. From now on, = 1, 1 refer to the positive and negative branches, respectively.
The tt component of Eq. (2.15) results on:
2 2
2

2 a 3 k + ( k3 + 1)H 2 + H
2
1
2

= 5 () + + p + 2 H 2 + 2H
,
2 2
2
3
3
2a 3 k + H 2
where a dot stands for derivative respect the cosmic time .

Using the junction condition of the scalar field at the brane Eq. (2.13), we obtain

2
2 2
k 2 a 3 k + H 2 = 5 () b( + 3p) .
2
(2.19)
(2.20)
83
The energy balance on the brane due to the scalar field comes from the Codacci equation [26],
the junction conditions (2.13), (2.15) and reads

S = () bT .
(2.21)
Substituting the explicit expression of the energy-momentum of the brane Eq. (2.17) in the previous expression, we obtain
T = bT .
(2.22)
Consequently
d
.
(2.23)
da
The energy density on the brane is not conserved as is usual in dilatonic brane-world models
[25,26]. Indeed, the presence of an induced gravity term on the brane action does not modify the
energy balance on the brane because the induced gravity term is not coupled to the bulk scalar
field; i.e., is constant. Then, if we suppose that the perfect fluid satisfies a constant equation
of state; i.e., p = w with w constant, the energy density scales with the scale factor as in the
model analysed in [29]:

() = 0 exp (m/k) , m = 3(w + 1) + (3w 1)kb.
(2.24)
+ 3H ( + p) = ( 3p)ba
However, the Friedmann equation of the brane is modified by the presence of the induced gravity
term as we will see shortly. Moreover, the differential equation satisfied by the brane tension
involves the Hubble parameter H (see Eq. (2.20)). Consequently, the solutions for () obtained
in [29] are no longer valid in our model.
Friedmann equation of the brane follows from Eq. (2.18) and reads
2
54
+ 6H 2 .
36
There are two solutions of the Friedmann equation

1
2
3
2
2
+ + 4 1 + 1 + 454 2 2 a 2k /3 + 54 ( + ) ,
H =
6
3
5
H 2 = 2 a (2/3)k +
2
(2.25)
(2.26)
where = 1 describes the cosmological evolution of the positive branch and = 1 describes
the cosmological evolution of the negative branch. The positive branch has not a well defined
limit when approaches zero. This branch generalises the DGP model [15]. Indeed, for =
= 0 and for a constant scalar field and vanishing potential on the bulk; i.e., k = 0 and = 0,
we recover the self-accelerating solution of the DGP model [16]. The negative branch has a well
defined limit where we recover the Friedmann equation obtained in [29]. Moreover, this branch
generalises the RS model [30]. In fact, for a vanishing induced gravity parameter, a constant
scalar field on the bulk and a constant potential on the bulk fine tuned with the brane tension
such that 36 2 = 54 2 , we recover RS model.
It remains to deduce the constraint on the brane tension. Using Eqs. (2.18), (2.20) and (2.24)
we obtain

1
k
3
2

H =
(2.27)
( + ) + ( + ) + (w + 1) .
2k
3
k
84
Finally, equating the last two equations, we deduce the constraint satisfied by the brane tension
for each of the branches

3
2
k
2

( + ) + (w + 1) = 4 1 + 1 + 454 2 2 a 2k /3 + 54 ( + ) .
(2.28)
k
3
5
Again, in the limiting case where the induced gravity parameter vanishes the last equation is well
defined for the negative branch. This is not true for the positive branch.
3. Vacuum brane solutions
In what follows, we will restrict our analysis to the case of a vacuum brane; i.e., p = .
Then the constraint equation of the brane tension (2.28) reduces to

k
2 4
2

4
2
2
(2/3)k
( + ) =
(3.1)
1 + 1 + 5 ( + ) + 4 5 a
.
3
54
If there is no matter on the brane then the brane tension satisfies the same constraint Eq. (3.1)
with = 0. This is not surprising as we have restricted the matter content on the brane to satisfy
a vacuum equation of state. Consequently, the energy density for w = 1 results in a rescaling
of the brane tension . For the sake of simplicity, we will refer to + (with w = 1) simply
as the brane tension.
Before getting the general solution of Eq. (3.1), let us see if there are maximally symmetric
branes (we restrict to Minkowski and de Sitter spacetimes). Combining Eqs. (2.27) and (3.1), it
can be shown that there are maximally symmetric branes as long as the brane tension scales with
the scale factor as

6
2 2
2
= 6H 2 H 2 + 2 a 3 k
(3.2)
5
and satisfy the differential equations (2.27) and (3.1). This is only possible for a flat brane;
i.e., H = 0. Indeed in this case the brane tension is fine-tuned with the cosmological constant.
This kind of solutions was already noticed in [36]. A flat brane in the negative branch can be
embedded in the bulk given by Eq. (2.4) for any value of the scalar field, , at the brane. However,
a flat brane in the positive branch can only be embedded on the bulk, if 3/(2k) log(d) (see
Eqs. (2.26) and (3.2)).
There are no brane solution with de Sitter geometry for an evolving bulk scalar field. However,
Eq. (3.2) satisfies Eqs. (2.27) and (3.1) for very small scale factors on both branches; i.e., the
positive and negative branches can be asymptotically de Sitter in the past. On the other hand,
it can also be checked that the negative branch is asymptotically Minkowski in the future; i.e.,
Eq. (3.2) satisfies Eqs. (2.27) and (3.1) for very large scale factors.
In order to solve the differential equation (3.1) it is helpful to rewrite it as:

dy
= 1 + 1 + y + dex ,
dx
in terms of the following dimensionless quantities
2
y = 54 ,
3
2
2
x = k = k 2 ln(a),
3
3
(3.3)
d = 4 2 54 2 .
(3.4)
85
The solution of the previous equation can be written as a 2D surface involving the dimensionless
brane tension y and the rescaled dimensionless scalar field x

z z2 dex
2
x
2 z de + ln
(3.5)
= 2c, z = 1 + 1 + y + dex ,
z + z2 dex
where c is an integration constant. We will show latter on that the initial value of the Hubble parameter is fixed by c. TheMinkowski solution we discussed earlier corresponds to z2 = d exp(x)
or equivalently y = 2 d exp(x/2).
Although, we are not able to get an explicit relation between the brane tension and the scale
factor (or between y and x), we can rewrite the constraint (3.5) in a parametric form as2

1
1
z() =
(3.6)
2c ln
,
2
1+

dex () = 1 2 z2 ,
(3.7)

2
y() = z 2 + z .
(3.8)
Finally, substituting Eqs. (3.6)(3.8) into Friedmann equation (2.26), the Hubble parameter
reads:
1
1
H=
h=
z,
252
252
(3.9)
where, for latter convenience, we have introduced the dimensionless Hubble parameter, h, which
in term of reads

2
1
h=
(3.10)
2c ln
.
4
1+
The range of variation of the parameter and the constant c depends crucially on the specific
branch. This will result in very different physical properties of the two branches.
3.1. Positive branch = 1
In the positive branch and for a vacuum brane, the brane tension is a growing function of the
scalar field or equivalently of the scale factor of the brane (see Eqs. (2.7) and (2.20)). We would
like to stress that in a standard 4D FLRW expanding universe filled with a phantom energy
component, the energy density behaves in this way too: the energy density is a growing function
of the scale factor [9,10].
The dimensionless variable z which essentially measures the growth of the brane tension
with respect to the scalar field is greater than 1 (see Eq. (3.5)). This implies [0, 1) and the
integration constant c is such that c 0. In addition, there is a one to one correspondence between
the scale factor and the parameter . Indeed, the scale factor is an increasing function3 of .
A detailed analysis of Eqs. (3.6)(3.8) for c > 0 shows that at high energy (a 0 or 0),
the brane tension reaches infinite negative values. On the other hand, for very large values of the
scale factor ( 1), the brane tension acquires infinite positive values. The larger is the value
2 The parametric solution given by Eqs. (3.6)(3.8) does not include the Minkowski solution, we discuss earlier.
3 The function dex = da 2/3k 2 is a decreasing function of .
86
of the integration constant c, the sooner (for smaller scale factor) the brane tension becomes
positive.
The behaviour of the solution given by Eqs. (3.6)(3.8) for c = 0 is different from the
behaviour of the solution with c > 0. The scale factor is bounded from below (for 0,
d exp(x) 1) when c = 0. The brane tension is also bounded from below (for 0,
y 2). On the other hand, the brane tension acquires very large positive values when the
scale factor is very large (or 1). Moreover, the solution given by Eqs. (3.6)(3.8) for c = 0
can be extended to 1 < < 0. In fact, the solution is bouncing around = 0 where the scale
factor acquires its minimum value. For simplicity, we will restrict to 0 < 1 when c = 0.
The dimensionless Hubble parameter h is an increasing function of and consequently an
increasing function of the scale factor a (see footnote 3). At high energy (small values of the
scale factor or approaching 0), the dimensionless parameter h reaches a constant positive value
proportional to the integration constant c. On the other hand, at very large values of the scale
factor (or approaching 1), the dimensionless Hubble parameter diverges. The divergence of the
Hubble parameter for very large values of the scale factor might point out the existence of a big
rip singularity in the future evolution of the brane; i.e., the scale factor and Hubble parameter
blow up in a finite cosmic time in the future evolution of the brane. However, as we will next
show the divergence of H and a (and also of ) occur in an infinite cosmic time in the future
evolution of the positive branch.
We introduce a dimensionless cosmic time for the brane defined as
=
1
252
(3.11)
Then using Eqs. (3.4), (3.7) and (3.9), we obtain

3
= 2
2k

c
g()
d,
(1 ) ln(1 )
(3.12)
where
g() =

1
d
(1 ) ln(1 ) ln 1 2 z2 ,
z
d
(3.13)
and c is a constant. The function g() is continuous on [0, 1], approaching 2 for large value
of the scale factor, i.e., = 1, and approaching 2/c for very small scale factor4 ; i.e., = 0.
Consequently, g() has a minimum on [0, 1] which we denote min(g). On the other hand, it can
be proven that g() is positive and different from zero on [0, 1] for any c > 0. Then min(g) > 0.
Consequently, the dimensionless cosmic time of the brane is bounded from below as follows
3
> 2 min(g)
2k
1
(1 ) ln(1 )
d.
(3.14)
Integrating the rhs of the previous inequality, it can be easily proven that the cosmic time on the
brane diverges for approaching 1 where the scale factor reaches infinite values. This proof is
only valid for c > 0. For c = 0 a straightforwards extension of the previous proof implies the
4 For c = 0, g() vanishes when approaches zero.
87
same result; i.e., diverges when approaches 1 where the scale factor and Hubble parameter
blow up. Another way of showing that there is no big rip singularity on the future evolution of
the brane is by noticing that the asymptotic behaviour of the dimensionless tension of the brane
at large value of the scale factor is y x 2 /4. Therefore,
H
k2
ln(a).
52
(3.15)
Consequently, the Hubble rate does not grow as fast as in phantom energy models with a constant
equation of state where a big rip singularity takes place on the future evolution of a homogeneous
and isotropic universe [9,10].
In summary, we have proven that in the positive branch filled only with vacuum energy there
is a singularity in the future evolution of the brane. The singularity is such that for large value of
the cosmic time, the scale factor and the Hubble parameter diverge. This kind of singularity can
be interpreted as a big rip singularity pushed towards an infinite cosmic time of the brane. We
would like to stress that this singularity appears only on the brane and not in the bulk. Indeed,
large values of the scale factor, a, correspond to large values of the extra coordinate r (where the
bulk is asymptotically flat) and the only bulk singularity is located at r = 0. It can also be proven
that the time derivative of the Hubble parameter,

2 4
dh dh d 1
=
2 2 5 H =
d
d d
2

2k
1 2
1 1
=
,
2c ln
2 2c + ln
12
1+
1+
(3.16)
acquires infinite positive values for very large scale factors or cosmic time. In fact, the whole evolution of the positive branch is super-inflationary, i.e., H > 0. The curvature singularity we have
found has some similarities with the one in 4D phantom energy models with constant equation
of state [9,10]: H and H diverges for large values of the scale factor. Although, in these models
[9,10] the scale factor, Hubble parameter and cosmic time derivative of the Hubble parameter
diverge on a finite future cosmic time.
Finally, we point out that there is no big bang singularity for the solutions we have analysed.
Despite the fact that the scale factor vanishes at earlier time for c > 0, the Hubble parameter is
finite and its time derivative vanishes (see Eq. (3.16)). Indeed, for c > 0 the brane geometry is
asymptotically de Sitter in the past. For c = 0, the initial scale factor is non vanishing, the Hubble
parameter vanishes and its time derivative is finite. As we mentioned before, the scale factor is
bouncing around = 0 when c = 0.
3.2. Negative branch = 1
In the negative branch and for a vacuum brane, the brane tension is positive and a decreasing
function of the scalar field or equivalently of the scale factor of the brane (see Eqs. (2.7), (2.18)
and (2.20)). Consequently, the dimensionless variable z defined in Eq. (3.5) is negative. This
implies ( tanh(c), 0] and the integration constant satisfies c > 0. It can be shown that there
is a one to one correspondence between the scale factor and the parameter . Indeed, the scale
88
factor is a decreasing function5 of . An analysis of Eqs. (3.6)(3.8) shows that at high energy
(a 0 or 0), the brane tension reaches infinite positive values. On the other hand, for very
large value of the scale factor; i.e., tanh(c), the brane tension vanishes.
The dimensionless Hubble parameter h given in Eq. (3.10) is an increasing function of and
consequently a decreasing function of the scale factor. At high energy, the dimensionless Hubble
parameter h reaches a constant positive value proportional to the integration constant c. On the
other hand, at very large values of the scale factor, the dimensionless Hubble parameter vanishes.
Unlike the positive branch, the negative branch is never super-inflating, although the negative
branch always undergoes an inflationary period, as we will next show. The second temporal
derivative of the scale factor satisfies

1 z
d x 1
2 4a
f (),
ln de
5 =
(3.17)
a
2 1 2 d
where
f () = z() 1 +
k2
.
3
(3.18)
The brane is inflating when f is negative.6 The brane behaves in two different ways depending
on the values acquired by k 2 . For k 2 3, the brane is eternally inflating as f is always negative
(see Figs. 1, 2). A similar behaviour was found in [29] in a dilatonic brane-world model without
an induced gravity term. On the other hand, the brane undergoes an initial stage of inflation
when k 2 > 3 until f vanishes. Later on the brane starts decelerating (see Figs. 1, 2). This second
behaviour was not found in [29] for a vacuum brane. Then, the inclusion of an induced gravity
term on a dilatonic brane-world model with an exponential potential on the bulk allows inflation
to take place in the negative branch, even if it does not for a vanishing induced gravity parameter
(for k 2 > 3). This behaviour has some similarity with steep inflation [37], where high energy
corrections to the Friedmann equation in RS scenario [30] permit an inflationary evolution of
the brane with potentials too steep to sustain it in the standard 4D case, although the inflationary
scenario introduced by Copeland et al. in [37] is supported by an inflaton confined in the brane
while in our model inflation on the brane is induced by a dilaton field on the bulk.
Finally, we stress that there is no big bang singularity for the vacuum solution on the negative
branch. Although, the scale factor vanishes at early times, the Hubble parameter is finite and
its time derivative vanishes (see Eq. (3.16)). Like the positive branch, the negative branch is
asymptotically de Sitter in the past. The vacuum negative branch also does not face a curvature
singularity in its future. Indeed, for large values of the scale factor the Hubble parameter and its
cosmic derivative vanish and the brane is asymptotically Minkowski.
4. Conclusions
In this paper we study the behaviour of a dilatonic brane-world model with an induced gravity
term on the brane with a constant induced gravity parameter . We assume a Z2 -symmetry across
5 The function dex = da 2/3k 2 is a increasing function of and consequently the scale factor is a decreasing function
of (see Eq. (3.4)).

6 The term multiplying f on the rhs of Eq. (3.17) is negative because dex is positive and a decreasing function of ,
1 < tanh(c) < 0 and the brane tension is positive (z is negative). Consequently, the brane is inflating whenever
f is negative.
89
Fig. 1. This figure shows the behaviour of the function f (defined in Eq. (3.18)) as a function of . The negative branch
undergoes an inflationary expansion whenever f is negative. For k 2 3, the negative branch is eternally inflating. On the
other hand, for k 2 > 3 the brane is initially inflating until f () vanishes. The solid (darker grey), dotted and dashed-dotted
(lighter grey) lines correspond to f () for k 2 = 2, 3, 15 respectively and c = 1.
Fig. 2. This figure shows the behaviour of the dimensionless acceleration parameter given by 2 54 a/a
(introduced
in Eq. (3.17)) as a function of the parameter . The solid (darker grey), dotted and dashed-dotted (lighter grey) lines
correspond to the acceleration parameter for k 2 = 2, 3, 15 respectively and c = 1. For k 2 = 2, 3 the negative branch is
eternally inflating. On the other hand, for k 2 = 15 the brane undergoes an initial transient inflationary epoch.
the brane. The dilatonic potential is an exponential function of the bulk scalar field and the matter
content of the brane is coupled to the dilaton field. We deduce the modified Friedmann equation
for the positive and negative branch (which specifies the way the brane is embedded in the bulk),
the junction condition for the scalar field across the brane and the energy balance on the brane.
We obtain the vacuum solutions Eqs. (3.6)(3.8); i.e., the matter content of the brane is specified
by the brane tension, for a FLRW brane.
90
In the vacuum positive branch, the brane tension is a growing function of the scale factor
and, consequently, mimics the behaviour of a phantom energy component on the brane. This
phantom-like behaviour is obtained without including a phantom fluid on the brane. In fact, the
brane tension does not violate the null energy condition. The expansion of the brane is superinflationary; i.e., the Hubble parameter is a growing function of the cosmic time. At high energy
(small scale factors), the brane is asymptotically de Sitter for c > 0, where the parameter c is proportional to the initial Hubble parameter. There is a another solution which is a bouncing solution
(c = 0) around a minimum non-vanishing scale factor. The brane faces a curvature singularity
in its future evolution, where the Hubble parameter, brane tension and scale factor diverge. The
singularity happens in an infinite cosmic time. Therefore, the singularity can be interpreted as a
big rip singularity pushed towards an infinite future cosmic time.
In the vacuum negative branch, the brane tension is a decreasing function of the scale factor.
Unlike the positive branch, the branch is not super-inflating. However, it always undergoes an
inflationary expansion (see Eq. (3.17)). The inflationary expansion can be eternal (k 2 3) or
transient (k 2 > 3). For large values of the scale factor, the negative branch is asymptotically
Minkowski.
It remains to be seen how the results obtained in this paper are modified when the brane has
matter contents in addition to the brane tension. On the other hand, we have chosen a specific
dilatonic bulk and consequently we do not know how general are the brane properties we have
found. We leave these interesting questions for future work.
Acknowledgements
We thank D. Wands for very enlightening discussions. The author is also grateful to A. Ferrera,
P.F. Gonzlez-Daz, K. Koyama and R. Maartens for comments. The author thanks R. Lazkoz
for encouragements. MBL acknowledges the support of FCT (Portugal) through the fellowship
SFRH/BPD/26542/2006 and the research grant FEDER-POCI/P/FIS/57547/2004. The initial
work of MBL was funded by MECD (Spain) and also partly supported by DGICYT under Research Project BMF2002 03758. This paper is dedicated to all the people who helped me to
recover the mobility of my arm, in particular, the traumatology team of San Juan de la Cruz
Hospital, Cati, Conchi and Juana.
References
[1] A.G. Riess, et al., Astron. J. 116 (1998) 1009, astro-ph/9805201;
S. Perlmutter, et al., Astrophys. J. 517 (1999) 565, astro-ph/9812133;
J.L. Tonry, et al., Astrophys. J. 594 (2003) 1, astro-ph/0305008;
A.G. Riess, et al., Astrophys. J. 607 (2004) 665, astro-ph/0402512;
A.G. Riess, et al., astro-ph/0611572.
[2] D.N. Spergel, et al., Astrophys. J. Suppl. 148 (2003) 175, astro-ph/0302209;
M. Tegmark, et al., Phys. Rev. D 69 (2004) 103501, astro-ph/0310723;
D.N. Spergel, et al., WMAP Collaboration, Astrophys. J. Suppl. 170 (2007) 377, astro-ph/0603449.
[3] S. Weinberg, Rev. Mod. Phys. 61 (1989) 1;
P.J.E. Peebles, B. Ratra, Rev. Mod. Phys. 75 (2003) 559;
T. Padmanabhan, Phys. Rep. 380 (2003) 235.
[4] For a review and an extensive list on dark energy models see for example: E.J. Copeland, M. Sami, S. Tsujikawa,
Int. J. Mod. Phys. D 15 (2006) 1753, hep-th/0603057.
[5] C. Wetterich, Nucl. Phys. B 302 (1988) 668;
B. Ratra, P.J.E. Peebles, Phys. Rev. D 37 (1988) 3406;
R.R. Caldwell, R. Dave, P.J. Steinhardt, Phys. Rev. Lett. 80 (1998) 1582.
91
[6] G.W. Gibbons, Phys. Lett. B 537 (2002) 1, hep-th/0204008;

T. Padmanabhan, Phys. Rev. D 66 (2002) 021301, hep-th/0204150.
[7] A.Y. Kamenshchik, U. Moschella, V. Pasquier, Phys. Lett. B 511 (2001) 265, gr-qc/0103004;
N. Bilic, G.B. Tupper, R.D. Viollier, Phys. Lett. B 535 (2002) 17, astro-ph/0111325;
M.C. Bento, O. Bertolami, A.A. Sen, Phys. Rev. D 66 (2002) 043507, gr-qc/0202064;
M. Bouhmadi-Lpez, P. Vargas Moniz, Phys. Rev. D 71 (2005) 063521, gr-qc/0404111.
[8] R.R. Caldwell, Phys. Lett. B 545 (2002) 23;
P.F. Gonzlez-Daz, Phys. Rev. D 68 (2003) 021303, astro-ph/0305559;
G.W. Gibbons, hep-th/0302199;
S. Nojiri, S.D. Odintsov, Phys. Lett. B 571 (2003) 1, hep-th/0306212;
L.P. Chimento, R. Lazkoz, Phys. Rev. Lett. 91 (2003) 211301, gr-qc/0307111;
M.P. Dabrowski, T. Stachowiak, M. Szydowski, Phys. Rev. D 68 (2003) 103519, hep-th/0307128;
B. Boisseau, G. Esposito-Farse, D. Polarski, A. Starobinsky, Phys. Rev. Lett. 85 (2000) 2236, gr-qc/0001066;
J. Martin, C. Schimd, J.P. Uzan, Phys. Rev. Lett. 96 (2006) 061303, astro-ph/0510208.
[9] R.R. Caldwell, M. Kamionkowski, N.N. Weinberg, Phys. Rev. Lett. 91 (2003) 071301, astro-ph/0302506;
P.F. Gonzlez-Daz, Phys. Lett. B 586 (2004) 1, astro-ph/0312579.
[10] M. Bouhmadi-Lpez, J.A. Jimnez Madrid, JCAP 0505 (2005) 005, astro-ph/0404540;
M. Bouhmadi-Lpez, P.F. Gonzlez-Daz, P. Martn-Moruno, Phys. Lett. B 659 (2008) 1, gr-qc/0612135.
[11] S. Nojiri, S.D. Odintsov, S. Tsujikawa, Phys. Rev. D 71 (2005) 063004, hep-th/0501025;
C. Cattoen, M. Visser, Class. Quantum Grav. 22 (2005) 4913, gr-qc/0508045.
[12] D. Rapetti, S.W. Allen, J. Weller, Mon. Not. R. Astron. Soc. 360 (2005) 555, astro-ph/0409574.
[13] W.J. Percival, S. Cole, D.J. Eisenstein, R.C. Nichol, J.A. Peacock, A.C. Pope, A.S. Szalay, Mon. Not. R. Astron.
Soc. 381 (2007) 1053, arXiv: 0705.3323 [astro-ph].
[14] D. Langlois, Astrophys. Space Sci. 283 (2003) 469, astro-ph/0301022;
P. Brax, C. van de Bruck, Class. Quantum Grav. 20 (2003) R201, hep-th/0303095;
R. Maartens, Living Rev. Rel. 7 (2004) 7, gr-qc/0312059;
P. Brax, C. van de Bruck, A.C. Davis, Rep. Prog. Phys. 67 (2004) 2183, hep-th/0404011.
[15] G.R. Dvali, G. Gabadadze, M. Porrati, Phys. Lett. B 485 (2000) 208, hep-th/0005016;
For a review on the DGP model, see for example: A. Lue, Phys. Rep. 423 (2006) 1, astro-ph/0510068;
G. Gabadadze, Nucl. Phys. B (Proc. Suppl.) 171 (2007) 88, arXiv: 0705.1929 [hep-th].
[16] C. Deffayet, Phys. Lett. B 502 (2001) 199, hep-th/0010186.
[17] C. Deffayet, G.R. Dvali, G. Gabadadze, Phys. Rev. D 65 (2002) 044023, astro-ph/0105068;
C. Deffayet, S.J. Landau, J. Raux, M. Zaldarriaga, P. Astier, Phys. Rev. D 66 (2002) 024019, astro-ph/0201164;
V. Sahni, Y. Shtanov, JCAP 0311 (2003) 014, astro-ph/0202346;
V. Sahni, Y. Shtanov, Int. J. Mod. Phys. D 11 (2000) 1515, gr-qc/0205111;
U. Alam, V. Sahni, astro-ph/0209443;
R.G. Vishwakarma, P. Singh, Class. Quantum Grav. 20 (2003) 2033, astro-ph/0211285.
[18] V. Sahni, Y. Shtanov, JCAP 0311 (2003) 014, astro-ph/0202346;
A. Lue, G.D. Starkman, Phys. Rev. D 70 (2004) 101501, astro-ph/0408246.
[19] H. Collins, B. Holdom, Phys. Rev. D 62 (2000) 105009, hep-ph/0003173;
Y. Shtanov, hep-th/0005193;
S. Nojiri, S.D. Odintsov, JHEP 0007 (2000) 049, hep-th/0006232;
N.J. Kim, H.W. Lee, Y.S. Myung, Phys. Lett. B 504 (2001) 323, hep-th/0101091;
G. Kofinas, JHEP 0108 (2001) 034, hep-th/0108013;
H.s. Zhang, R.G. Cai, JCAP 0408 (2004) 017, hep-th/0403234;
R.G. Cai, H.S. Zhang, A. Wang, Commun. Theor. Phys. 44 (2005) 948, hep-th/0505186.
[20] E. Kiritsis, N. Tetradis, T.N. Tomaras, JHEP 0203 (2002) 019, hep-th/0202037.
[21] E. Papantonopoulos, V. Zamarias, JCAP 0410 (2004) 001, gr-qc/0403090.
[22] K.i. Maeda, S. Mizuno, T. Torii, Phys. Rev. D 68 (2003) 024033, gr-qc/0303039.
[23] M. Bouhmadi-Lpez, D. Wands, Phys. Rev. D 71 (2005) 024010, hep-th/0408061.
[24] M. Bouhmadi-Lpez, R. Maartens, D. Wands, Phys. Rev. D 70 (2004) 123519, hep-th/0407162.
[25] H.A. Chamblin, H.S. Reall, Nucl. Phys. B 562 (1999) 133, hep-th/9903225.
[26] K.i. Maeda, D. Wands, Phys. Rev. D 62 (2000) 124009, hep-th/0008188.
[27] A. Mennim, R.A. Battye, Class. Quantum Grav. 18 (2001) 2171, hep-th/0008192;
D. Langlois, M. Rodrguez-Martnez, Phys. Rev. D 64 (2001) 123507, hep-th/0106245;
K. Koyama, K. Takahashi, Phys. Rev. D 68 (2003) 103512, hep-th/0307073;
92
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
E. Leeper, K. Koyama, R. Maartens, Phys. Rev. D 73 (2006) 043506, hep-th/0508145;

J.E. Lidsey, D. Seery, Phys. Rev. D 72 (2005) 103507, hep-th/0509176.
A. Feinstein, K.E. Kunze, M.A. Vzquez-Mozo, Phys. Rev. D 64 (2001) 084015, hep-th/0105182.
K.E. Kunze, M.A. Vzquez-Mozo, Phys. Rev. D 65 (2002) 044002, hep-th/0109038.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
J.W. York, Phys. Rev. Lett. 28 (1972) 1082;
G.W. Gibbons, S.W. Hawking, Phys. Rev. D 15 (1977) 2752.
P. Bowcock, C. Charmousis, R. Gregory, Class. Quantum Grav. 17 (2000) 4745, hep-th/0007177.
S. Mukohyama, T. Shiromizu, K.i. Maeda, Phys. Rev. D 62 (2000) 024028, hep-th/9912287;
S. Mukohyama, T. Shiromizu, K.i. Maeda, Phys. Rev. D 63 (2001) 029901, Erratum.
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Nucl. Phys. B 552 (1999) 246, hep-th/9806051.
S. Kachru, M.B. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021, hep-th/0001206.
E.J. Copeland, A.R. Liddle, J.E. Lidsey, Phys. Rev. D 64 (2001) 023509, astro-ph/0006421.

Effects of lightest neutrino mass in leptogenesis

E. Molinaro, S.T. Petcov 1 , T. Shindou 2 , Y. Takanishi
SISSA and INFN-Sezione di Trieste, Trieste I-34014, Italy
Received 12 September 2007; accepted 20 December 2007
Abstract
The effects of the lightest neutrino mass in flavoured leptogenesis are investigated in the case when
the CP-violation necessary for the generation of the baryon asymmetry of the Universe is due exclusively
to the Dirac and/or Majorana phases in the neutrino mixing matrix U . The type I see-saw scenario with
three heavy right-handed Majorana neutrinos having hierarchical spectrum is considered. The orthogonal
parametrisation of the matrix of neutrino Yukawa couplings, which involves a complex orthogonal matrix R,
is employed. Results for light neutrino mass spectrum with normal and inverted ordering (hierarchy) are
obtained. It is shown, in particular, that if the matrix R is real and CP-conserving and the lightest neutrino
mass m3 in the case of inverted hierarchical spectrum lies the interval 5 104 eV m3 7 103 eV,
the predicted baryon asymmetry can be larger by a factor of 100 than the asymmetry corresponding to
negligible m3
= 0. As consequence, we can have successful thermal leptogenesis for 5 106 eV m3
5 102 eV even if R is real and the only source of CP-violation in leptogenesis is the Majorana and/or
Dirac phase(s) in U .
PACS: 98.80.Cq; 14.60.Pq; 14.60.St
Keywords: Thermal leptogenesis; Seesaw mechanism; Lepton flavour effects
E-mail addresses: molinaro@sissa.it (E. Molinaro), testuo.shindou@desy.de (T. Shindou), yasutaka@sissa.it

(Y. Takanishi).
1 Also at: Institute of Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia, Bulgaria.
2 Current address: Deutsches Elektronen-Synchrotron DESY, Notkestrae 85, D-22603 Hamburg, Germany.
doi:10.1016/j.nuclphysb.2007.12.033
94
E. Molinaro et al. / Nuclear Physics B 797 (2008) 93116
1. Introduction
In the present article we continue to investigate the possible connection between leptogenesis
[1,2] (see also, e.g., [3,4]) and the low energy CP-violation in the lepton (neutrino) sector (for
earlier discussions see, e.g., [58] and the references quoted therein). It was shown recently [9]
that the CP-violation necessary for the generation of the observed baryon asymmetry of the Universe in the thermal leptogenesis scenario can be due exclusively to the Dirac and/or Majorana
CP-violating phases in the PontecorvoMakiNakagawaSakata (PMNS) [10] neutrino mixing
matrix, and thus can be directly related to the low energy CP-violation in the lepton sector (e.g.,
in neutrino oscillations, etc.). The analysis performed in [9] (see also [11,12]) was stimulated by
the progress made in the understanding of the importance of lepton flavour effects in leptogenesis
[1318]. It led to the realisation that these effects can play crucial role in the leptogenesis scenario
of baryon asymmetry generation [1517]. It was noticed in [16], in particular, that Scenarios in
j
which 1 = 0 while 1 = 0 entail the possibility that the phases in the light neutrino mixing maj
trix U are the only source of CP violation, 1 and 1 being respectively the individual lepton
number and the total lepton number CP violating asymmetries.
As is well-known, the leptogenesis theory [1] is based on the see-saw mechanism of neutrino
mass generation [19]. The latter provides a natural explanation of the observed smallness of neutrino masses (see, e.g., [2022]). An additional appealing feature of the see-saw scenario is that
through the leptogenesis theory it allows to relate the generation and the smallness of neutrino
masses with the generation of the baryon (matterantimatter) asymmetry of the Universe, YB .
The non-supersymmetric version of the type I see-saw model with two or three heavy righthanded (RH) Majorana neutrinos is the minimal scheme in which leptogenesis can be implemented. In [9] the analysis was performed within the simplest type I see-saw mechanism of
neutrino mass generation with three heavy RH Majorana neutrinos, Nj , j = 1, 2, 3. Taking into
account the lepton flavour effects in leptogenesis it was shown [9], in particular, that if the heavy
Majorana neutrinos have a hierarchical spectrum, i.e., if M1 M2,3 , Mj being the mass of Nj ,
the observed baryon asymmetry YB can be produced even if the only source of CP-violation is
the Majorana and/or Dirac phase(s) in the PMNS matrix3 UPMNS U . Let us recall that in the
case of hierarchical spectrum of the heavy Majorana neutrinos, the lepton flavour effects can be
significant in leptogenesis provided the mass of the lightest one M1 satisfies the constraint [15
17] (see also [23]): M1 1012 GeV. In this case the predicted value of the baryon asymmetry
depends explicitly (i.e., directly) on U and on the CP-violating phases in U . The results quoted
above were demonstrated to hold both for normal hierarchical (NH) and inverted hierarchical
(IH) spectrum of masses of the light Majorana neutrinos (see, e.g., [20]). In both these cases
they were obtained for negligible lightest neutrino mass and CP-conserving elements of the orthogonal matrix R, present in the orthogonal parametrisation [24] of the matrix of neutrino
Yukawa couplings. The CP-invariance constraints imply [9] that the matrix R could conserve the
CP-symmetry if its elements are real or purely imaginary. As was demonstrated in [9], for NH
spectrum and negligible lightest neutrino mass m1 one can have successful thermal leptogenesis
with real R. In contrast, in the case of IH spectrum and negligible lightest neutrino mass (m3 ),
the requisite baryon asymmetry was found to be produced for CP-conserving matrix R only if
certain elements of R are purely imaginary: for real R the baryon asymmetry YB is strongly
suppressed [8] and leptogenesis cannot be successful for M1 1012 GeV (i.e., in the regime in
3 The same result was shown to hold also for quasi-degenerate in mass heavy RH Majorana neutrinos [9].
95
which the lepton flavour effects are significant). It was suggested in [9] that the observed value of
YB can be reproduced for M1 1012 GeV in the case of IH spectrum and real (CP-conserving)
elements
of R if the lightest
neutrino mass m3 is non-negligible, having a value in the interval
2
2
8.0 105 eV2 is the
10
m m3 0.5 m2 , where m2 = m2 m2 m2 =

21
mass squared difference responsible for the solar neutrino oscillations, and m1,2 are the masses
of the two additional
light Majorana neutrinos. In this case we still would have m3 m1 , m2

2
since m1,2 = |m |

= 5.0 102 eV, |m2 | m2 m2
= m2 m2 being the mass squared
A
difference associated with the dominant atmospheric neutrino oscillations.

It should be noted that constructing a viable see-saw model which leads to real or purely
imaginary matrix R might encounter serious difficulties, as two recent attempts in this direction
indicate [18,25]. However, constructing such a model lies outside the scope of our study.
In the present article we investigate the effects of the lightest neutrino mass on flavoured
(thermal) leptogenesis. We concentrate on the case when the CP-violation necessary for the
generation of the observed baryon asymmetry of the Universe is due exclusively to the Dirac
and/or Majorana CP-violating phases in the PMNS matrix U . Our study is performed within
the simplest type I see-saw scenario with three heavy RH Majorana neutrinos Nj , j = 1, 2, 3.
The latter are assumed to have a hierarchical mass spectrum, M1 M2,3 . Throughout the
present study we employ the orthogonal parametrisation of the matrix of neutrino Yukawa couplings [24]. As was already mentioned earlier, this parametrisation involves an orthogonal matrix
R, R T R = RR T = 1. Although, in general, the matrix R can be complex, i.e., CP-violating, in
the present work we are primarily interested in the possibility that R conserves the CP-symmetry.
We consider the two types of light neutrino mass spectrum allowed by the data (see, e.g., [20]):
(i) with normal ordering (m2A > 0), m1 < m2 < m3 , and (ii) with inverted ordering (m2A < 0),
m3 < m1 < m2 . The case of inverted hierarchical (IH) spectrum and real (and CP-conserving)
matrix R is investigated in detail. Results for the normal hierarchical (NH) spectrum are also
presented.
Our analysis is performed for negligible renormalisation group (RG) running of mj and of
the parameters in the PMNS matrix U from MZ to M1 . This possibility is realised (in the
class of theories of interest) for sufficiently small values of the lightest neutrino mass min(mj )
[26,27], e.g., for min(mj ) 0.10 eV. The latter condition is fulfilled for the NH and IH neutrino
mass spectra, as well as for spectrum with partial hierarchy (see, e.g., [28]). Under the indicated
condition mj , and correspondingly m2A and m2 , and U can be taken at the scale MZ , at
which the neutrino mixing parameters are measured.
Throughout the present work we use the standard parametrisation of the PMNS matrix:

c12 c13
s12 c13
s13 ei
31
21

i
i
U = s12 c23 c12 s23 s13 e
c12 c23 s12 s23 s13 e
s23 c13 diag 1, ei 2 , ei 2 ,
s12 s23 c12 c23 s13 ei c12 s23 s12 c23 s13 ei c23 c13
(1)
where cij cos ij , sij sin ij , ij = [0, /2], = [0, 2] is the Dirac CP-violating (CPV)
phase and 21 and 31 are the two Majorana CPV phases [29,30], 21,31 = [0, 2]. All our
numerical results are obtained for the current best fit values of the solar and atmospheric neutrino
oscillation parameters [3133], m2 , sin2 12 and m2A , sin2 223 :
m2 = m221 = 8.0 105 eV2 ,
sin2 12 = 0.30,

m2 = m2
= 2.5 103 eV2 ,
sin2 223 = 1.
A
31(32)
(2)
(3)
96
In certain cases the predictions for |YB | are very sensitive to the variation of sin2 12 and sin2 223
within their 95% C.L. allowed ranges:
0.26 sin2 12 0.36,
0.36 sin2 23 0.64,
95% C.L.
(4)
We also use the current upper limit on the CHOOZ mixing angle 13 [31,32,34]:
sin2 13 < 0.025(0.041),
95%(99.73%) C.L.
(5)
2. Baryon asymmetry from low energy CP-violating Dirac and Majorana phases in
UPMNS
Following [9] we perform the analysis in the framework of the simplest type I see-saw scenario. It includes the Lagrangian of the Standard Model (SM) with the addition of three heavy
right-handed Majorana neutrinos Nj (j = 1, 2, 3) with masses 0 < M1 < M2 < M3 and Yukawa
couplings j l , l = e, , . We will work in the basis in which (i) the Yukawa couplings for the
charged leptons are flavour-diagonal, and (ii) the Majorana mass term of the RH neutrino fields
is also diagonal. The heavy Majorana neutrinos are assumed to possess a hierarchical mass spectrum, M1 M2 M3 .
In what follows we will use the well-known orthogonal parametrisation of the matrix of
neutrino Yukawa couplings [24]:
=
1
MR mU ,
v
(6)
where R is, in general, a complex orthogonal matrix, RR T = R T R = 1, M and m are diagonal matrices formed by the masses of Nj and of the light Majorana neutrinos j , M
Diag(M1 , M2 , M3 ), m Diag(m1 , m2 , m3 ), Mj > 0, mk 0, and v = 174 GeV is the vacuum
expectation value of the Higgs doublet field. We shall assume that the matrix R has real and/or
purely imaginary elements.
In the case of hierarchical heavy Majorana neutrinos Nj , the CP-violating asymmetries,
relevant for leptogenesis, are generated in out-of-equilibrium decays of the lightest one, N1 . The
asymmetry in the lepton flavour l (lepton charge Ll ) is given by [1517]:

1/2 3/2
3M1 Im( j k mj mk Ulj Ulk R1j R1k )

l =
(7)
.
2
16v 2
i mi |R1i |
Thus, for real or purely imaginary elements R1j of R, e + + = 0.
There are three possible regimes of generation of the baryon asymmetry in the leptogenesis
scenario [1517]. At temperatures T M1 > 1012 GeV the lepton flavours are indistinguishable
and the one flavour approximation is valid. The relevant asymmetry is = e + + and
in the case of interest (real or purely imaginary CP-conserving R1j ) no baryon asymmetry is
produced. For 109 GeV T M1 1012 GeV, the Boltzmann evolution of the asymmetry
in the -flavour (lepton charge L of the Universe) is distinguishable from the evolution of the
(e + )-flavour (or lepton charge Le + L ) asymmetry e + . This corresponds to the socalled two-flavour regime.4 At smaller temperatures, T M1 109 GeV, the evolution of the
4 As was suggested in [9] and confirmed in the more detailed study [23], in the two-flavour regime of leptogenesis the
flavour effects are fully developed at M1 5 1011 GeV.
97
-flavour (lepton charge L ) and of also become distinguishable (three-flavour regime). The
produced baryon asymmetry is a sum of the relevant flavour asymmetries, each weighted by the
corresponding efficiency factor accounting for the wash-out processes.
In the two-flavour regime, 109 GeV T M1 1012 GeV, the baryon asymmetry5 predicted
in the case of interest is given by [17] (see also [9]):

417
390
12
YB
m
2 +
m
2
=
37g
589
589

12
390
417
=
(8)

m

m
2 ,
37g
589
589
where the second expression corresponds to real and purely imaginary R1j R1k . Here g = 217/2
is the number of relativistic degrees of freedom, 2 = e + , m
2 =m
e +m
, m
l is the wash-out
mass parameter for the asymmetry in the lepton flavour l [1517],

2

1/2
m
l =
(9)
R1k mk Ulk , l = e, , ,
k
and (390m
/589)
) and (417m
2 /589)
2 ) are the efficiency factors for
= (0.66m
= (0.71m
generation of the asymmetries and 2 . The efficiency factors are well approximated by the
expression [17]:

1.16
1

X
8.25 103 eV
(X)
(10)
.
+
=
X
2 104 eV
At T M1 109 GeV, the three-flavour regime is realised and [17]

151
344
344
12
m
e +
m
+
m
.
e
YB =
37g
179
537
537
(11)
For real or purely imaginary R1j R1k of interest, j = k, it proves convenient to cast the asymmetries l in the form [9]:

3M1
k
j >k mk mj (mj mk )kj |R1k R1j | Im(Ulk Ulj )

,
l =
2
16v 2
i mi |R1i |
if Im(R1k R1j ) = 0,
(12)

3M1
k
j >k mk mj (mj + mk )kj |R1k R1j | Re(Ulk Ulj )

l =
,
2
2
16v
i mi |R1i |
if Re(R1k R1j ) = 0,
(13)
where we have used R1j R1k = j k |R1j R1k | and R1j R1k = ij k |R1j R1k |, j k = 1, j = k. Note
that real (purely imaginary) R1k R1j and purely imaginary (real) Ulk Ulj , j = k, implies violation
of CP-invariance by the matrix R [9]. In order for the CP-symmetry to be broken at low energies, we should have both Re(Ulk Ulj ) = 0 and Im(Ulk Ulj ) = 0 (see [9] for further details).
Note also that if R1j , j = 1, 2, 3, is real or purely imaginary, as the condition of CP-invariance
requires [9], of the three quantities R11 R12 , R11 R13 and R12 R13 , relevant for our discussion,
not more than two can be purely imaginary, i.e., if, for instance, R11 R12 = i12 |R11 R12 | and
R12 R13 = i23 |R12 R13 |, then we will have R11 R13 = 13 |R11 R13 |.
5 The expression we give is of the baryon asymmetry normalised to the entropy density, see, e.g., [9].
98
3. Effects of lightest neutrino mass: Real R1j

We consider next the possible effects the lightest neutrino mass min(mj ) can have on (thermal) leptogenesis. We will assume that the latter takes place in the regime in which the lepton
flavour effects are significant and that the CP-violation necessary for the generation of the baryon
asymmetry is provided only by the Majorana or Dirac phases in the PMNS matrix UPMNS . In the
present Section we analyse the possibility of real elements R1j , j = 1, 2, 3, of the matrix R. The
study will be performed both for light neutrino mass spectrum with normal and inverted ordering.
We begin with the more interesting possibility of spectrum with inverted ordering (hierarchy).
3.1. Light neutrino mass spectrum with inverted ordering
The case of inverted hierarchical (IH) neutrino mass spectrum, m3 m1 < m2 , m1,2 =
2
|mA |, is of particular interest since, as was already mentioned in the Introduction, for real
R1j , j = 1, 2, 3, IH spectrum and negligible lightest neutrino mass m3
= 0, it is impossible to
generate the observed baryon asymmetry YB
= 8.6 1011 in the regime of flavoured leptogenesis [9], i.e., for M1 1012 GeV, if the only source of CP violation are the Majorana
and/or Dirac phases in the PMNS matrix. For m3 m1 < m2 and real R1j , the terms pro
portional to m3 in the expressions for the asymmetries l and wash-out mass parameters m
l,
2 + R 2 = 1. The
l = e, , , will be negligible if m3
= 0, or if R13 = 0 and R11 , R12 = 0, R11
12
main reason for the indicated negative result lies in the fact that if m3 = 0, or m3 m1 < m2 and
R13 = 0, the lepton asymmetries l are suppressed by the factor m2 /(2m2A )
= 1.6 102 ,
6
while |R11 |, |R12 | 1, and the resulting baryon asymmetry is too small.
In what follows we will analyse the generation of the baryon asymmetry YB for real R1j ,
j = 1, 2, 3, when m3 is non-negligible. We will assume that YB is produced in the two-flavour
regime, 109 GeV M1 1012 GeV. Under these conditions the terms m3 in l will be
dominant provided [9]

3
2
m2A 4 |R13 |
m3
2
1.
|R11(12) |
m2
m2
(14)
This inequality can be fulfilled if R11 0, or R12 0, and if m3 is sufficiently large. The
neutrino mass spectrum will be of the IH type if m3 still obeys
m3 m1,2 . The latter condition
can be satisfied for m3 having a value m3 5 103 eV
|m2A |. Our general analysis will
be performed for values of m3 from the interval 1010 eV m3 5 102 eV.

Consider the simple possibility of R11 = 0. We will present later results of a general analysis,
performed without setting R11 to 0. For R11 = 0 the asymmetry = (e + ) of interest is
given by:

3M1
m3
=
(15)
m3 m2 1
23 r Im U2 U 3 ,
m2
16v 2
6 This suppression is present also in the one-flavour regime of Y generation, i.e., in the sum + + , when
e
B
R13 = 0 and the product R11 R12 has non-trivial real and imaginary parts [8].
where
m2 =
r=

m23 + m2A ,
99
(16)
|R13 R12 |
,
3
2
|R12 |2 + m
m2 |R13 |
(17)
and

Im(U2 U 3 ) = c23 c13 Im ei(31 21 )/2 c12 s23 + s12 c23 s13 ei .
The two relevant wash-out mass parameters are given by:
2
2
|U 2 |2 + m3 R13
|U 3 |2 + 2 m2 m3 23 |R12 R13 | Re(U2 U 3 ),
m
= m2 R12
2
m
2 m
e +m
= m2 R12
2
+ m3 R13
m
,
(18)
(19)
(20)
where 23 = sgn(R12 R13 ).

2 + R 2 + R 2 = 1, which in the case under
The orthogonality of the matrix R implies that R11
12
13
2
2
consideration reduces to R12 + R13 = 1. It is not difficult to show that for R12 and R13 satisfying
this constraint, the maximum of the function r, and therefore of the asymmetry | |, takes place
for
m3
m2
2
2
2
2
R12
(21)
=
,
R13
=
,
R12
< R13
.
m3 + m2
m3 + m2
At the maximum we get

1
|m2 |
1
2
1 m2 2 1
A
max |r| =
,
=
2 m3
2
m3
(22)
and

3M1
3M1
Im(U U 3 )
| |
(23)
m2A Im(U2 U 3 ).
(m
m
)
=
=
2
3
2
32v 2
32v 2
The second
approximate equalities in Eqs. (22) and (23) correspond to IH spectrum, i.e., to
m3 m2 = |m2 |. Thus, the maximum of the asymmetry | | thus found (i) is not suppressed
A
by the factor m2 /(m2A ), and (ii) practically does not depend on m3 in the case of IH spectrum.
Given the fact that
|m2 |

m2 m3
M1
A
5.0 108
Im(U U 3 ),
| | =
(24)
2
9
10 GeV
|m2 | 0.05 eV
A
max(| Im(U2 U 3 )|)

= 0.46, where we have used sin2 223 = 1, sin2 12 = 0.30 and sin2 13 <
0.04, and that max(|(0.66m
) (0.71m
2 )|)
= 7 102 , we find the absolute upper bound on
the baryon asymmetry in the case of IH spectrum and real matrix R (real R1j R1k ):

|m2 |

M1
A
|YB | 4.8 1012
(25)
.
0.05 eV
109 GeV
This upper bound allows to determine the minimal value of M1 for which it is possible to reproduce the observed value of |YB | lying in the interval 8.0 1011 |YB | 9.2 1011 for IH
100
spectrum, real R and R11 = 0:

M1 1.7 1010 GeV.
(26)
The values of R12 , for which | | is maximal, can differ, in general, from those maximising
|YB | due to the dependence of the wash-out mass parameters and of the corresponding efficiency
factors on R12 . However, this difference, when it is present, does not exceed 30%, as our calculations show, and is not significant. At the same time the discussion of the wash-out effects
for the maximal | | is rather straightforward and allows to understand in a rather simple way
the specific features of the generation of |YB | in the case under discussion. For these reasons in
our discussion of the wash-out mass parameters we will use R12 maximising | |. All our major
numerical results and most of the figures are obtained for R12 maximising |YB |.
For R12 (R13 ), which maximises the ratio |r| and the asymmetry | |, the relevant wash-out
mass parameters are given by:

m2 m3
|U 2 |2 + |U 3 |2 + 223 Re(U2 U 3 ) ,
m
=
(27)
m3 + m2
m2 m3
m
2 =2
(28)
m
.
m3 + m2
Eqs.(24), (27) and (28) suggest that in the case of IH spectrum with non-negligible m3 ,
m3 |m2A |, the generated baryon asymmetry |YB | can be strongly enhanced in comparison
with the asymmetry |YB | produced if m3
= 0. The enhancement can be by a factor of 100.
Indeed, the maximum of the asymmetry | | (with respect to |R12 |), Eq. (23), does not contain
the suppression factor m2 /(2m2A )
= 1.6 102 and its magnitude is not controlled by m3 ,

and m
2 , Eqs. (27)
but rather by |m2A |. At the same time, the wash-out mass parameters m
and (28), are determined by m2 m3 /(m2 + m3 ) = m3 . The latter in the case under discussion can
) and (0.71m
2 ),
take values as large as m3 5 103 eV. The efficiency factors (0.66m
which enter into the expression for the baryon asymmetry, Eq. (8), have a maximal value (X)
=
(67) 102 when X
= (0.71.5) 103 eV (weak wash-out regime). Given the range of values
of m3 for IH spectrum extends to 5 103 eV, one can always find a value of m3 in this
2 take a value maximising (0.66m
) or (0.71m
2 ), and |(0.66m
)
range such that m
or m
(0.71m
2 )|. This qualitative discussion suggests that there always exists an interval of values
of m3 for which the baryon asymmetry is produced in the weak wash-out regime. On the basis
of the above considerations one can expect that we can have successful leptogenesis for a nonnegligible m3 in the case of IH spectrum even if the requisite CP-violation is provided by the
Majorana or Dirac phase(s) in the PMNS matrix. This is confirmed by the detailed (analytic and
numerical) analysis we have performed, the results of which are described below.
3.1.1. A. Leptogenesis due to Majorana CP-violation in UPMNS
We will assume first that the Dirac phase has a CP-conserving value, = 0; . For
+
c23 s12 s13 )| sin 32 /2| and correspondingly
= 0(), we have | Im(U2 U 3 )| = c23 c13 (s23 c12 ()
0.36| sin 32 /2| | Im(U 2 U 3 )| 0.46| sin 32 /2|, where 32 = 31 21 and we have used
the best fit values of sin2 223 and sin2 12 , and the limit sin2 13 < 0.04. For s13 = 0 we get:
| Im(U2 U 3 )|
= 0.42| sin 32 /2|. The terms proportional to s13 have a subdominant effect on the
magnitude of the calculated | | and |YB |.
,2 remain
It is easy to check that the asymmetry | | and the wash-out mass parameters m
invariant with respect to the change 23 23 , 32 2 32 . Thus, the baryon asymmetry
101
|YB | satisfies the following relation: |YB (23 , 32 )| = |YB (23 , 2 32 )|. Therefore, unless
otherwise stated, we will consider the case of 23 = +1 in what follows.
The absolute maximum of the asymmetry |YB | with respect to 32 is not obtained for 32 =
for which | | is maximal,7 but rather for 32 having a value in the interval 32
= (/2 2/3)
if 23 = +1, or in the interval 32
=
1.
The
maximal
value of |YB | at
(4/3
3/2)
if
=
23
32 = is smaller at least by a factor of 2 than the value of |YB | at its absolute maximum (see
further Fig. 3). As can be easily shown, for 32 there is a rather strong mutual compensation
between the asymmetries in the lepton charges L and (Le + L ) owing to the fact that, due to
and m
2 have relatively close values and |(0.66m
) (0.71m
2 )| 102 .
Re(U2 U 3 ) = 0, m
2 )|
Actually, in certain cases one can even have |(0.66m
) (0.71m
= 0, and thus |YB |
= 0,
for 32 lying in the interval 32 ( 4/3) (see further Fig. 3). Similar cancellation can occur
for s13 = 0.2 at 32 /6. Obviously, we have |YB | = 0 for 32 = 0; 2 .
We are interested primarily in the dependence of |YB | on m3 . As m3 increases from the value
of 1010 eV up to 104 eV, in the case of R11 = 0 under discussion the maximal possible |YB |
for a given M1 increases monotonically, starting from a value which for M1 1012 GeV is much
smaller than the observed one, max(|YB |) 8.6 1011 . At approximately m3
= 2 106 eV,
11
11
for M1
we have max(|YB |)
= 8.6 10
= 5 10 GeV. As m3 increases beyond 2 106 eV,
max(|YB |) for a given M1 continues to increase until it reaches a maximum. This maximum
occurs for m3 such that 0.71m
2
2 ) is maximal, (0.71m
2)
=
= 9.0 104 eV and (0.71m
2
) is considerably smaller. As can be shown, for 23 = +1, it always
6.8 10 , while (0.66m
takes place at 32
= /2. For 32 = /2, s13 = 0 and 23 = +1, the maximum of |YB | in question
is located at m3
= 7 104 eV. It corresponds to the CP-asymmetry being predominantly in the
) (0.71m
2 )| and correspondingly |YB |,
(e + )-flavour. As m3 increases further, |(0.66m
rapidly decrease. At certain value of m3 , typically lying in the interval m3 (1.52.5)103 eV,
one has |(0.66m
) (0.71m
2 )|
= 0 and |YB | goes through a deep minimum: one can have
even |YB | = 0. This minimum of |YB | corresponds to a partial or complete cancellation between
the asymmetries in the -flavour and in the (e + )-flavour.8 In our example of 32 = /2,
s13 = 0 and 23 = +1, the indicated minimum of |YB | occurs at m3
= 2.3 103 eV. As m3
increases further, |(0.66m
) (0.71m
2 )| and |YB | rapidly increase and |YB | reaches a second
maximum, which in magnitude is of the order of the first one. This maximum corresponds to
the CP-asymmetry being predominantly in the -flavour rather than in the (e + )-flavour: now
)
2 ) is substantially
(0.66m
) is maximal having a value (0.66m
= 6.8 102 , and (0.71m
smaller. For 23 = +1, s13 = 0 or s13 = 0.2 and = 0, it takes place at a value of 32 close to
/2, 32
= /2, while for s13 = 0.2 and = , it occurs at 32
= 2/3. In the case of 23 = +1,
s13 = 0 and 32 = /2, the second maximum of |YB | is located at m3
= 7 103 eV. As m3
increases further, |YB | decreases rather slowly monotonically.
These features of the dependence of |YB | on m3 are confirmed by a more general analysis in
which, in particular, the value of R11 was not set to zero a priori. The results of this analysis are
presented in Fig. 1, while Fig. 2 illustrates the dependence of |YB | on m3 in the case of R11 = 0.
In Fig. 1 we show the correlated values of M1 and m3 for which one can have successful
leptogenesis in the case of neutrino mass spectrum with inverted ordering and CP-violation due
to the Majorana and Dirac phases in UPMNS . The figure was obtained by performing, for given
7 We would like to recall that in the case of = , = 0; , and real R R , the requisite violation of CP-symmetry
32
1j 1k
in leptogenesis is provided by the matrix R [9].
8 A general discussion of the possibility of such a suppression of the baryon asymmetry Y in the case of flavoured
B
leptogenesis is given in [9].
102
Fig. 1. Values of m3 and M1 for which the flavoured leptogenesis is successful, generating baryon asymmetry
|YB | = 8.6 1011 (red/dark shaded area). The figure corresponds to hierarchical heavy Majorana neutrinos, light
neutrino mass spectrum with inverted ordering (hierarchy), m3 < m1 < m2 , and real elements R1j of the matrix R. The
minimal value of M1 at given m3 , for which the measured value of |YB | is reproduced, corresponds to CP-violation
due to the Majorana phases in the PMNS matrix. The results shown are obtained using the best fit values of neutrino
oscillation parameters: m2 = 8.0 105 eV2 , m2A = 2.5 103 eV2 , sin2 12 = 0.30 and sin2 223 = 1. See text
for further details. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
Fig. 2. The dependence of |YB | on m3 in the case of IH spectrum, real R1j R1k , Majorana CP-violation, R11 = 0,
32 = /2, s13 = 0, M1 = 1011 GeV, and for (i) sgn(R12 R13 ) = +1 (left panel), and (ii) sgn(R12 R13 ) = 1 (right
panel). The baryon asymmetry |YB | was calculated for a given m3 , using the value of |R12 |, for which the asymmetry
| | is maximal. The horizontal dotted lines indicate the allowed range of |YB |, |YB | = (8.0 9.2) 1011 .
m3 from the interval 1010 m3 0.05 eV, a thorough scan of the relevant parameter space
searching for possible enhancement or suppression of the baryon asymmetry with respect to
that found for m3 = 0. The real elements of the R-matrix of interest, R1j , j = 1, 2, 3, were allowed to vary in their full ranges determined by the condition of orthogonality of the matrix R:
2 + R 2 + R 2 = 1. In particular, R was not set to zero. The Majorana phases
R11
11
21,31 were
12
13
varied in the interval [0, 2]. The calculations were performed for three values of the CHOOZ
angle 13 , corresponding to sin 13 = 0; 0.1; 0.2. In the cases of sin 13 = 0, the Dirac phase
was allowed to take values in the interval [0, 2]. The heavy Majorana neutrino mass M1
was varied in the interval 109 GeV M1 1012 GeV. For given m3 , the minimal value of the
mass M1 , for which the leptogenesis is successful, generating |YB |
= 8.6 1011 , was obtained
103
Fig. 3. The dependence of |YB | on 32 (Majorana CP-violation), in the case of IH spectrum, real R1j R1k , R11 = 0,
M1 = 1011 GeV, and for (i) s13 = 0.2, = 0(), |R12 | = 0.34(0.38), m3 = 6.7(4.3) 103 eV, sgn(R12 R13 ) = +1
(left panel, red (blue) line), and (ii) s13 = 0, sgn(R12 R13 ) = 1, |R12 | = 0.41, m3 = 4.2 103 eV (right panel). The
values of m3 and |R12 | used maximise |YB | at (i) 32 = /2(2/3) and (ii) 32 = 3/2. The horizontal dotted lines
indicate the allowed range of |YB | = (8.0 9.2) 1011 . (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
for the values of the other parameters which maximise |YB |. The min(M1 ) thus found does not
exhibit any significant dependence on s13 . If m3 2.5 107 eV, leptogenesis cannot be successful for M1 1012 GeV: the baryon asymmetry produced in this regime is too small. As
m3 increases starting from the indicated value, the maximal |YB | for a given M1 1012 GeV,
increases monotonically. Correspondingly, the min(M1 ) for which one can have successful leptogenesis decreases monotonically and for m3 5 106 eV we have min(M1 ) 5 1011 GeV.
The first maximum of |YB | (minimum of M1 ) as m3 increases is reached at m3
= 5.5 104 eV,
32
= /2 (21
= 0.041, 31
= 1.65), R11
= 0.061, R12
= 0.099, and R13
= 0.99. At the maximum we have |YB | = 8.6 1011 for M1
= 3.4 1010 GeV. The second maximum of |YB | (or
minimum of M1 ) seen in Fig. 1 corresponds to m3
= 5.9 103 eV, 32
= /2 (21
= 0.022,
31
= 1.45), R11
= 0.18, R12
= 0.29, and R13
= 0.94. The observed value of |YB | is reproduced in this case for M1
= 3.5 1010 GeV.
Similar features are seen in Fig. 2, which shows the dependence of |YB | on m3 for R11 = 0,
fixed M1 = 1011 GeV, 32 = /2, s13 = 0 and 23 = +1; (1). In the case of 32 = /2, s13 =
0.2, = 0 and 23 = +1, the absolute maximum of |YB | is obtained for m3
= 6.7 103 eV and
|R12 | = 0.34 (Fig. 3). At this maximum we have (0.66m

) = 0.067, (0.71m
2)
= 0.013, and

|m2 |

M1
A
12
|YB | = 2.6 10
(29)
.
0.05 eV
109 GeV
Correspondingly, the observed baryon asymmetry |YB |, 8.0 1011 |YB | 9.2 1011 ,
can be reproduced if M1 3.0 1010 GeV. If s13 = 0, the same result holds for M1
3.5 1010 GeV. The minimal values of M1 thus found are somewhat smaller than min(M1 )
=
5.3 1010 GeV obtained in the case of negligible m3
= 0 (R13 = 0) and purely imaginary
R11 R12 [9]. The dependence of the baryon asymmetry on 32 in the case of s13 = 0; 0.2 discussed
above is illustrated in Fig. 3. One obtains similar results in the case of R12 = 0, R11 , R13 = 0,
we have also analysed. The corresponding formulae can be obtained
from those derived for
R11 = 0 by replacing R12 with R11 , U 2 with U 1 and m2 with m1 = m23 + |m2A | m2
=

m23 + |m2A | = m2 . In this case we have, in particular, | | | Im(U1 U 3 )| = |c23 c13 (s12 s23
c12 c23 s13 ) sin 31 /2|, where the minus (plus) sign corresponds to = 0(). Evidently, the
104
relevant Majorana phase9 is 31 /2. Numerically we get 0.19| sin 31 /2| | Im(U1 U 3 )|
0.35| sin 31 /2|, while for s13 = 0 we find | Im(U1 U 3 )|
= 0.27| sin 31 /2|. Thus, the maximal
value of | | for R12 = 0 is smaller approximately by a factor of 1.3 than the maximal value of
| | when R11 = 0. As a consequence, the minimal M1 for which we can have successful leptogenesis can be expected to be bigger by a factor of 1.3 than the one we have obtained in the case
of R11 = 0. This is confirmed by the numerical calculations we have performed. For example,
for s13 = 0.2, = , sgn(R12 R13 ) = 1, and values of |R11 | = 0.38 and m3 = 4.5 103 eV
(which maximise |YB | at 31 = 2/3), we get

max |YB |
= 2.2 1012
|m2 |

M1
A
.
0.05 eV
109 GeV
(30)
Therefore the observed value of |YB | can be reproduced for M1 3.7 1010 GeV.
We would like to stress that the results we have obtained in the case of R1j = 0, j = 1, 2, 3,
which are shown in Fig. 1, are very different from the results for, e.g., R11 = 0 and R12 , R13 = 0.
Along the line of minimal values of M1 in Fig. 1, for which we can have successful leptogenesis,
2 104 eV, or m
2 103 eV and m
2
we find that either m

2 103 eV and m
103 eV, practically for any m3 from the interval10 1010 eV m3 5.0 102 eV. This
explains why one can have successful leptogenesis for min(M1 ) 5 1011 GeV even when
m3
= 5 106 eV. If R11 = 0, we get for m3 m2 , as it follows from Eqs. (27) and (28),
2 m3 . Consequently, for m3 103 eV, one also has m
,m
2 103 eV, and for
m
m
12
M1 < 10 GeV the baryon asymmetry generated under these conditions is strongly suppressed,
|YB | 8.6 1011 .
3.1.2. B. Dirac CP-violation in UPMNS and leptogenesis
We will assume that 21 = 31 = 0 (or 21 = 2k, 31 = 2k , k, k = 0, 1, 2, . . .)
and11 analyse the simple possibility of R11 = 0. For R11 = 0 and 32 = 0 we have: | |
2 c s s | sin | 0.054| sin |, where we have used c2 = 0.5 and12 s
| Im(U2 U 3 )| = c23
13 12 13
12 =
23
0.55. Thus, for given M1 the maximal asymmetry |YB | we can obtain will be smaller by a factor
(78) than the maximal possible asymmetry |YB | in the corresponding case of CP-violation
, corresponding to R12
due to the Majorana phase(s) in UPMNS . The wash-out mass parameter m
maximising | |, is given by
9 Note that the Majorana phase (R = 0) or (R = 0), relevant for leptogenesis in the case of IH spectrum
32
11
31
12

and real matrix R, does not coincide with the Majorana phase 21 , which together with |m2A | and sin2 12 determines
the values of the effective Majorana mass in neutrinoless double beta decay (see, e.g., [28,35,36]).
10 As our numerical calculations show, at m 109 eV, for instance, there is a very narrow interval of values of m
3
3
around m3 2103 eV, in which both m
2 and m
increase rapidly monotonically from 103 eV and 2104 eV
2
to 7 103 eV and 2 103 eV, respectively. As m3 increases from 2.5 103 eV to 5.0 102 eV, m
continues to increase monotonically, while m
remains practically constant.
11 If in the case of real R R and, e.g., R = 0 (R = 0), the Majorana phase
1j 1k
11
12
32(31) entering into the expression
for takes the CP-conserving value 32(31) = , the CP-symmetry will be violated not only by the Dirac phase = k ,
k = 0, 1, 2, . . . , but also by the matrix R [9].
12 Given the 2 allowed range of values of c2 = (0.360.64), it is clear that the asymmetry | | can be bigger (smaller)
23
by a factor 1.3 (0.72).
105
Fig. 4. The dependence of |YB | on m3 in the case of spectrum with inverted ordering (hierarchy), real R1j R1k and Dirac
CP-violation, for R11 = 0, = /2, s13 = 0.2, 32 = 0, M1 = 2.5 1011 GeV and sgn(R12 R13 ) = +1(1) (red lines
(blue dashed line)). The baryon asymmetry |YB | was calculated for a given m3 , using the value of |R12 |, for which the
CP-asymmetry | | is maximal. The results shown for sgn(R12 R13 ) = +1 are obtained for sin2 23 = 0.50; 0.36; 0.64
(red solid, dotted and dash-dotted lines), while those for sgn(R12 R13 ) = 1 correspond to sin2 23 = 0.5. See text for
further details. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version
of this article.)
m2 m3
2 2 2
(c12 s23 23 c13 c23 )2 + s12
m

s13 c23
=
m3 + m2

+ 2s12 s13 c23 (c12 s23 23 c13 c23 ) cos ,
(31)
while m
2 is determined by Eq. (28). Depending on the value of 23 , there are two quite different
cases to be considered.
2 s 2 c2 and 2s s c cos in the expression for m
If 23 = 1, the terms s12
, Eq. (31), are
12 13 23
13 23
13
and m
2 practically do not depend on and we
subdominant and can beneglected. Thus, m
have for c23 = s23 = 1/ 2: m

= 1.66m2 m3 /(m2 + m3 ),
= 0.5(c12 + c13 )2 m2 m3 /(m2 + m3 )
m
2
= 0.34m2 m3 /(m3 + m2 ). Both the asymmetry | | and |YB | are maximal for = /2 + k ,
k = 0, 1, . . . . The dependence of |YB | on m3 is analogous to that in the case of CP-violation
due to the Majorana phase(s) in UPMNS : we have two similar maxima corresponding to the CPasymmetry being predominantly respectively in the -flavour and in the (e + )-flavour. The two
maxima are separated by a deep minimum of |YB | (Fig. 4). The maxima occur at m3
= 7.5
104 eV (|R12 |
= 0.12) and at m3
= 0.30), i.e., at values of m3 which
= 4.9 103 eV (|R12 |
differ by a factor14 of 7. At the first (second) maximum we have (0.66m
) (0.71m
2)
=
0.044(0.046) and

|m2 |

sin
M1
A
13
|YB |
.
= 3.5(3.7) 1013 | sin |
0.2
0.05 eV
109 GeV
(32)
13 The term 2s s c cos , for instance, gives a relative contribution to m

not exceeding 10%.
12 13 23
14 The positions of the two maxima and of the deep minimum of |Y | under discussion exhibit very weak dependence
B
on sin2 23 when the latter is varied within its 2 allowed range, while the value of |YB | at each of the two maxima
changes according to |YB | cos2 23 .
106
Thus, the measured value of |YB |, 8.0 1011 |YB | 9.2 1011 , can be reproduced for
M1 2.3(2.2) 1011 GeV. The flavour effects in leptogenesis are fully developed for M1
5 1011 GeV. This upper bound and the requirement of successful leptogenesis in the case of
breaking of CP-symmetry due to the Dirac phase in UPMNS , lead to the following lower limit on
| sin 13 sin | and thus on sin 13 :
| sin 13 sin | 0.087,
sin 13 0.087.
(33)
The preceding lower bound corresponds to

|JCP | 0.02,
(34)
where JCP is the rephasing invariant associated with the Dirac phase , which controls the magnitude of CP-violation effects in neutrino oscillations15 [38]. Values of s13 in the range given
in Eq. (33) can be probed in the forthcoming Double CHOOZ [39] and future reactor neutrino
experiments16 [40]. CP-violation effects with magnitude determined by |JCP | satisfying (34)
are within the sensitivity of the next generation of neutrino oscillation experiments, designed to
search for CP- or T-symmetry violations in the oscillations [41]. Since in the case under discussion the wash-out factor |B | |(0.66m
) (0.71m
2 )| in the expression for |YB | practically
does not depend on s13 and , while both |YB | |s13 sin | and |JCP | |s13 sin |, there is a direct
relation between |YB | and |JCP | for given m3 (or m2 ) and M1 :
|m2 |

m2 m3
M1
A
10

1.8
10
|J
||
|
|YB |
(35)
,
=
CP
B
109 GeV
|m2 | 0.05 eV
A
where B = B (m2 m3 /(m2 + m3 ), 12 , 23 ) and we haveused the best fit values of sin2 12 and
sin2 23 . In the case of IH spectrum we have (m2 m3 )/ |m2 |
= 1 and m2 m3 /(m2 + m3 )
=
A
m3 . Similar relation between |YB | and |JCP | holds in an analogous case of normal hierarchical
light neutrino mass spectrum [9].
We get somewhat different results if 23 = +1. Now there is a strong compensation between
the terms in the round brackets in the expression (31) for m
and we have m
m2 m3 /(m2 +
m3 ). Correspondingly, one has m
2
. Thus, m
2 practically does not
= 2m2 m3 /(m2 + m3 )
m
depend on and on the neutrino mixing angles. The two wash-out mass parameters m
2 and
2 = c2 = 0.5 and s = 0.2 and s 2 = 0.30 one
m
can differ by a factor 100. Indeed, for s23
13
23
12
finds m
/m
2
= 0.5(0.0162 0.0156 cos ). For fixed sin2 12 = 0.30, the magnitude of the ratio
m
/m
2 (which is practically independent of m2 m3 /(m2 + m3 )) is very sensitive to the value
2 = 0.64 we get m
2 /0.04 0.0107(s /0.2) cos ),
of 23 : for s23
/m
2
= 0.5(0.0066 + 0.0043s13
13
2
2 /0.04 0.0494(s /0.2) cos ).
while if s23 = 0.36 one obtains m

/m
2 = 0.5(0.0794 + 0.0077s13
13
The maxima of the asymmetry |YB | take place at = /2 + k , k = 0, 1, 2, . . . . For = /2,
2 = 0.64; 0.5; 0.36 we get: m
s13 = 0.2 and s23
/m
2
= 0.52 102 ; 0.81 102 ; 4.36 102 .
Therefore the two maxima of |YB | as a function of m3 , corresponding to the CP-asymmetry
being predominantly in the (e + )-flavour and in the -flavour, can be expected to occur at
15 As is well-known, the Majorana phases, in contrast to the Dirac phase, do not affect the flavour neutrino oscillations
[29,37].
16 If we use M 1012 GeV [1517] for the maximal value of M in this discussion, we get, obviously, | sin sin |
1
1
13
0.044, sin 13 0.044 and |JCP | 0.01.
107
2 = 0.36; 0.5; 0.64 and s 2 = 0.30 would differ by a factor of m

values of m3 , which for s23
2 /m

12
20; 120; 190. The position of the deep minimum of |YB | between the two maxima would also be
2 = 0.36 and s 2 = 0.5(0.64). Obviously, the relative position on the m axis
very different for s23
3
23
of two maxima and the minimum of |YB | under discussion will depend not only on the precise
value of sin2 23 , but also on the precise value of sin2 12 .
To be more concrete, the maximum of |YB | (as a function of m3 ), associated with the CPasymmetry being predominantly in the (e + )-flavour, takes place at m3
= 7.5 104 eV, i.e.,
2 ) is maximal, (0.71m
2)
in the region of IH spectrum. At this value of m3 , (0.71m
= 0.068,
2 ), and we have:
while (0.66m
) 0.005 (0.71m

|m2 |

sin 13
M1
A
13
|YB |
(36)
5.1
10
|
sin
|
.
=
0.2
0.05 eV
109 GeV
The position of this maximum does not depend on 12 , 23 , 13 and . Thus, the measured value
of |YB |, 8.0 1011 |YB | 9.2 1011 , can be reproduced for a somewhat smaller value of
M1 1.6 1011 GeV than the corresponding value of M1 we have found for 23 = 1 (compare
Eqs. (32) and (36)). In the vicinity of the maximum there exists a correlation between the values
of |YB | and |JCP | similar to the one given in Eq. (35). Now the requirement of successful leptogenesis leads for M1 5 1011 GeV to a somewhat less stringent lower limit on | sin 13 sin |,
and thus on sin 13 and |JCP |:
| sin 13 sin |, sin 13 0.063,
|JCP | 0.015.
(37)
The second maximum of |YB |, related to the possibility of the CP-asymmetry being predominantly in the -flavour, takes place, as it is not difficult to convince oneself, at m2 m3 /(m2 +m3 )
102 eV, i.e., for values of m3 1.2 102 eV in the region of neutrino mass spectrum
with partial inverted hierarchy. In this case the factor in|YB |, which determines the position
of the maximum as a function of m3 , is ((m2 m3 )/ |m2A |)(0.66m

), rather than just
(0.66m
). Taking this observation into account, it is not difficult to show that for = /2
2 = 0.30 and, e.g., s 2 = 0.36(0.50), the maximum occurs at
and s13 = 0.2 maximising |YB |, s12
23
1.8(5.0) 102 eV. If M1 = 1011 GeV and |m2 | = 5.0 102 eV, the value of |YB | at
m3 =
A
2 = 0.64 we get for the same values of the

this maximum reads: |YB |
= 4.4(1.1) 1011 . For s23
11
other parameters max(|YB |)
= 0.6 10 . Obviously, if m3 102 eV, the observed value of
2 < 0.50. The position of the deep min|YB | can be reproduced for M1 5 1011 GeV only if s23
imum of |YB | at m3 103 eV, as we have already indicated, is also very sensitive to the value
2 : for = /2, s = 0.2 and s 2 = 0.30, it takes place at m 2 103 eV if s 2 = 0.36,
of s23
13
3=
12
23
2 = 0.50. These features of the dependence of |Y | on m
and at m3
= 102 eV in the case of s23
B
3
are illustrated in Fig. 4.
One can perform a similar analysis in the case of real R1j R1k , R12 = 0 and R11 , R13 = 0. In
2 c c s | sin | 0.082| sin |, and
this case we have | | | Im(U1 U 3 )| = c23
13 12 13
m1 m3
2 2 2
m

(s12 s23 + 13 c13 c23 )2 + c12
c23 s13
=
m3 + m1

2s13 c12 c23 (s12 s23 + 13 c23 c13 ) cos ,
(38)

, where 13 sgn(R11 R13 ) = 1 and m1
m
2 = 2m1 m3 /(m3 + m1 ) m
= m2 = m23 + |m2A |.
For 13 = +1, the two maxima of |YB | (as a function of m3 ) have the same magnitude. They
108
occur at
= 3/4, s13 = 0.2 and m3
= 7.5 104 (3.5 103 ) eV. The maximal baryon asym2 . For s 2 = 0.36(0.50), M = 5
metry, max(|YB |), exhibits rather strong dependence on s23
1
23

2
2
11
2
10
10 GeV and |mA | = 5.0 10 eV, we get max(|YB |) = 1.7(0.9) 10 . If s23 > 0.50,
however, it is impossible to reproduce the observed value of |YB | for M1 5 1011 GeV. The
2 from the interval [0.360.64] if s 0.10.
same negative result holds for any s23
13
2 in the region of the maximum of |Y | at
In the case of 13 = 1, we have |YB | c23
B
m3
= 7.5 104 eV, associated with the CP-asymmetry being predominantly in the (e + )flavour. The baryon asymmetry |YB | is maximal for = /2
which maximises the CP-asymmetry
2 = 0.5, M = 5 1011 GeV and
| |. For s13 = 0.2, c23
|m2A | = 5.0 102 eV, we find
1
max(|YB |)
= 4.5 1010 :

|m2 |

sin
M1
A
13
|YB |
(39)
.
= 9.0 1013 | sin |
0.2
0.05 eV
109 GeV
Thus, the observed value of the baryon asymmetry can be reproduced for relatively small values
of | sin 13 sin |, and correspondingly of sin 13 and |JCP |:
| sin 13 sin |, sin 13 0.036,
|JCP | 0.0086.
(40)
In contrast, the position (with respect to m3 ) of the maximum of |YB |, associated with the CPasymmetry being predominantly in the -flavour, and its magnitude, exhibit rather strong depen2 . For s 2 = 0.36; 0.50; 0.64, the maximum of |Y | is located at m 0.7; 1.5; 3.0
dence on s23
B
3=
23
2
10 eV. For M1 51011 GeV, the measured value of |YB |, 8.01011 |YB | 9.21011 ,
2 = 0.36; 0.50; 0.64.
can be reproduced provided | sin 13 sin | 0.046; 0.053; 0.16 if s23
3.2. Neutrino mass spectrum with normal ordering
We can get different results for light neutrino mass spectrum with normal ordering. The case
of negligible m1 and real (CP-conserving) elements R1j of R was analysed in detail in [9].17 It
was found that if the only source of CP-violation is the Dirac phase in the PMNS matrix, the observed value of the baryon asymmetry YB
= 8.6 1011 can be reproduced if [9] | sin 13 sin |
0.09. Given the upper limit | sin 13 sin | < 0.2, this requires M1 2 1011 GeV. The quoted
lower limit on | sin 13 sin | implies that we should have sin 13 0.09 and that |JCP | 2 102 .
The indicated values of sin 13 and of |JCP | are testable in the reactor neutrino and long baseline
neutrino oscillation experiments under preparation [3942]. If, however, the Dirac phase has a
CP-conserving value,
= k , k = 0, 1, 2, . . . , and the requisite CP violation is due exclusively
to the Majorana phases 21,31 in U , the observed YB can be obtained for [9] M1 4 1010 GeV.
For M1 = 5 1011 GeV, for which the flavour effects are fully developed, the measured value
of YB can be reproduced for a rather small value of | sin 32 /2|
= 0.15, where 32 31 21 .
In searching for possible significant effects of non-negligible m1 in leptogenesis we have
considered values of m1 as large as 0.05 eV, m1 0.05 eV. For 3 103 eV m1 0.10 eV,
the neutrino mass spectrum is not hierarchical; the spectrum exhibits partial hierarchy (see, e.g.,
[28]), i.e., we have m1 < m2 < m3 .
2
2
17 Results for quasi-degenerate (QD) spectrum, m
1 = m2 = m3 , mj
|mA |, which requires mj 0.10 eV, were
also obtained in [9].
109
We are interested in the dependence of the baryon asymmetry on the value of m1 . In the case
of neutrino mass spectrum with normal ordering we have:

m2 = m21 + m2 ,
(41)
m3 = m21 + m2A .
We will illustrate the characteristic features of the possible effects of m1 in leptogenesis by
analysing two simple possibilities: |R11 | 1 and |R12 | 1. Results of a more general analysis
performed without making a priori assumptions about the real parameters R11 and R12 will also
be presented.
We first set R11 = 0. The asymmetry is given by:

3M1 m2A m3
12 m2A

(42)
23 r Im(U2 U 3 ),
=
m2
m2 + m3
16v 2
where
r=
|R12 R13 |
,
3
2
|R12 |2 + m
m2 |R13 |
(43)
and Im(U2 U 3 ) is given in Eq. (18). The ratio in (43) is similar to the ratio in Eq. (17). Note,
however, that the masses m2,3 present in Eqs. (15) and (17) are very different from the masses
2 + R 2 = 1 and the results given in
m2,3 in Eqs. (42) and (43). Using again the fact that R12
13
Eq. (21), it is easy to find that r has a maximum for
2
R12
=
m3
,
m2 + m3
2
R13
=
m2
,
m2 + m3
2
2
R13
< R12
,
(44)
where m2 and m3 are defined now by Eq. (41). At the maximum

max(r) =

1
1 m2 2
.
2 m3
(45)
For the value of R12 (R13 ), which maximises the ratio |r| and correspondingly the asymmetry
| |, the relevant wash-out mass parameters m
and m
2 are given by Eqs.
(27) and (28) with m2
and m3 determined by Eq. (41). Since in the case of interest m2 m2
= 0.9 102 eV,

m3 m2A
= 5.0 102 eV, we have m2 m3 /(m2 + m3 ) 0.7 102 eV. The lightest
neutrino mass m1 can have any effect on the generation of
the baryon asymmetry YB only
if m21
m2 and if m1 is non-negligible with respect to
m2A . Correspondingly, for non-
negligible m1 of interest we will have m2 m3 /(m2 + m3 ) 102 eV and the baryon asymmetry
will be generated in the strong wash-out regime, unless there is a strong cancellation between
the first two and the third terms in the expression for m
, see Eq. (27). Obviously, the possibility
of such a cancellation depends critically on the sgn(R12 R13 ) = 23 . It should also be clear from
and m
2 (see
Eq. (41) and the dependence on m2,3 of max(| |) and of the corresponding m
Eqs. (42), (45), (27) and (28)) that with the increasing of m1 beyond 102 eV the predicted
baryon asymmetry decreases.
3.2.1. Leptogenesis due to Majorana CP-violation in UPMNS
Suppose first that the Dirac phase in the PMNS matrix has a CP-conserving value, =
k, k = 0, 1, 2, . . . , and that the only source of CP-violation are the Majorana phases 21,31
110
Fig. 5. Values of m1 and M1 for which the flavoured leptogenesis is successful and baryon asymmetry
YB = 8.6 1011 can be generated (red shaded area). The figure corresponds to light neutrino mass spectrum with
normal ordering. The CP-violation necessary for leptogenesis is due to the Majorana and Dirac phases in the PMNS matrix. The results shown are obtained using the best fit values of neutrino oscillation parameters: m2 = 8.0 105 eV2 ,
m2A = 2.5 103 eV2 , sin2 12 = 0.30 and sin2 223 = 1. See text for further details. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
in the PMNS matrix U . In the specific case of R11 = 0 we are considering the relevant CPviolating parameter is the difference of the two Majorana phases 32 31 21 . In this case
2 c | sin /2| 0.42| sin /2|, where we have neglected the possible
| | Im(U2 U 3 )
= c23
=
12
32
32
subleading corrections due to terms proportional to sin 13 . Even for sin 13 = 0.2 the latter have
practically no influence on the results we are going to obtain and for simplicity we set sin 13 = 0
in the discussion of Majorana CP-violation which follows. For the wash-out mass parameter m
we find:

m3
32
2 2
2
s23 + c23
223 c23 s23 c12 cos
m

(46)
.
c12
= m2
m2 + m3
2
If cos 32 /2
= 0, the baryon asymmetry YB is produced in the strong wash-out regime and for
M1 < 1012 GeV it is impossible to reproduce the observed value of YB
= 8.6 1011 : the
calculated asymmetry is too small. Actually, the maximum of |YB | in the case under discussion occurs, as can be shown, for 32
= /2 + k, k = 0, 1, 2, . . . . There are two distinctive
possibilities to be considered corresponding to the two possible signs of 23 sgn(cos 32 /2). If

23 sgn(cos 32 /2) = +1, we have m
= 0.25m2 m3 /(m2 + m3 ), the asymmetry in the -flavour
((e + )-flavour) is produced in the weak (strong) wash-out regime and we get for, e.g.,
m1 = 2 102 (5 102 ) eV:

m2

M1
A
12
|YB | = 1.20(0.36) 10
(47)
,
32
= /2 + k.
0.05 eV
109 GeV
Thus, for m1 = 2 102 (5 102 ) eV the measured value of YB
= 8.6 1011 can be obtained
10
11
for M1 7.2 10 (2.4 10 ) GeV.
These results are illustrated in Fig. 5, showing the correlated values of M1 and m1 for
which one can have successful leptogenesis. The figure was obtained using the same general method of analysis we have employed to produce Fig. 1. For given m1 from the interval
111
1010 m1 0.05 eV, a thorough scan of the relevant parameter space was performed in the
calculation of |YB |, searching for possible non-standard features (enhancement or suppression)
of the baryon asymmetry. The real elements R1j of interest of the matrix R, were allowed to vary
2 + R 2 + R 2 = 1. The
in their full ranges determined by the condition of orthogonality of R: R11
12
13
Majorana and Dirac phases 21,31 and were varied in the interval [0, 2]. The calculations
were performed again for three values of the CHOOZ angle, sin 13 = 0; 0.1; 0.2. The relevant
heavy Majorana neutrino mass M1 was varied in the interval 109 GeV M1 1012 GeV. For
given m1 , the minimal value of the mass M1 , for which the leptogenesis is successful generating YB
= 8.6 1011 , was obtained for the values of the other parameters which maximise
|YB |. The min(M1 ) thus calculated did not show any significant dependence on s13 . For m1
7.5 103 eV we did not find any noticeable effect of m1 in leptogenesis: the results we have obtained practically coincide with those corresponding to m1 = 0 and derived in [9]. The minimal
2 0.85, R 2 0.15 and /2
value of M1
= 41010 GeV seen in Fig. 1 corresponds to R12
=
32 =
13 =
3
2
(23 sgn(cos 32 /2) = +1); it does not depend on m1 . For 7.5 10 eV m1 5 10 eV, as
our calculations and Fig. 1 show, the predicted baryon asymmetry YB for given M1 is generically
smaller with respect to the asymmetry YB one finds for m1 = 0. Thus, successful leptogenesis is
possible for larger values of min(M1 ). The corresponding suppression factor increases with m1
and for m1
= 5 102 eV values of M1 1011 GeV are required.
If, however, 23 sgn(cos 32 /2) = 1, both the asymmetries in the -flavour and in the (e +
)-flavour are generated under the conditions of strong wash-out effects. Correspondingly, it is
impossible to have a successful leptogenesis for M1 < 1012 GeV if m1
= 5 102 eV. If m1 has
2
a somewhat lower value, say m1 = 2 10 eV, the wash-out of the (e + )-flavour asymmetry
is less severe (m
2
= 8.6 103 eV) and the observed YB can be reproduced for 32 = /2 + k
11
if M1 2.5 10 GeV.
3.2.2. Leptogenesis due to Dirac CP-violation in UPMNS
If the Majorana phases 21,31 have CP-conserving values and the only source of CP-violation
2 c s s sin | 0.054| sin |. The factor
is the Dirac phase in UPMNS , one has | | |c23
13 12 13
2
c23 c13 s12 s13 in | | is smaller by at least approximately an order of magnitude than the analogous
2 c
factor c23
12 in | | corresponding to Majorana CP-violation we have considered above. The
study of this case of Dirac CP-violation we have performed shows that as a consequence of the
indicated suppression it is impossible to generate the observed value of the baryon asymmetry
|YB | for M1 5 1011 GeV.
We have investigated also the possibility of |R13 | being sufficiently small, so that the term
R11 R12 in the expression for is the dominant one. We have found that if R13 = 0, it is
impossible to have successful leptogenesis for m1 0.05 eV and M1 < 1012 GeV if the requisite
CP-violation is due to the Majorana and/or Dirac
phases in UPMNS . The same conclusion is valid
for normal hierarchical spectrum, m1 m2 = m2 .

3.2.3. The case of R12 = 0

We get very different results if R12 = 0, while R11 R13 = 0. In this case the expression for the
asymmetry can be obtained formally from Eq. (42) by replacing m2 with m1 , 23 with 13 ,
U2 with U1 and the ratio r with
r=
|R11 R13 |
,
3
2
|R11 |2 + m
m1 |R13 |
2
2
R11
+ R13
= 1.
(48)
112
Fig. 6. The dependence of |YB | on m1 in the case of neutrino mass spectrum with normal ordering and real R1j R1k , for
R12 = 0, s13 = 0, M1 = 1.5 1011 GeV and sgn(R11 R13 ) = 1. The red solid, the blue dotted, and the green dashed
lines correspond to 31 = 2/3, /2, and /3 respectively. The figure is obtained for 23 = /4. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)
As in the similar cases discussed earlier, the ratio r, and the asymmetry | |, take maximal values
for
m3
m1
2
2
R11
(49)
=
,
R13
=
,
m1 + m3
m1 + m3
1
and max(r) = 0.5(m1 /m3 ) 2 . The expression of the asymmetry | | at the maximum (with respect to R11 ) reads:

3M1 m2A m2A

Im(U U 3 ).
| |
(50)
=
1
32v 2 m1 + m3
The wash-out parameters m
and m
2 , corresponding to the maximum | |, are given by

m1 m3
m
=
|U 1 |2 + |U 3 |2 + 213 Re(U1 U 3 )
m1 + m3

31
m1 m3
2 2
2
,
=
s s + c23 + 213 c23 s23 s12 cos
m1 + m3 12 23
2
(51)
(52)
2 = 2m1 m3 /(m1 + m3 )
where we have set s13 = 0 in obtaining the second expression, and m
m
. Note that if m1 m3
= 5 102 eV, the asymmetry | | practically does not depend on
m1 , while m
,2 O(m1 ). This implies that the dependence of max(|YB |) on m1 as the latter
increases, will exhibit the same features as in the case of IH spectrum we have already discussed:
|YB | will have two maxima, corresponding to the CP-asymmetry being predominantly in the
-flavour and in the (e + )-flavour, separated with a deep minimum. The analysis of the similar
case of IH spectrum suggests, that for s13 = 0, the largest baryon asymmetry |YB | is obtained for
31 = (2k + 1) and 13 sgn(cos 31 /2) = 1. These features are confirmed by our numerical
calculations and are illustrated in Fig. 6. In Fig. 6 we have displayed the dependence of |YB |
on m1 in the case of neutrino mass spectrum with normal ordering, R12 = 0 and real R11 R13 .
The results shown are obtained for 13 = 1, sin 13 = 0, M1 = 3 1011 GeV, and three CPviolating values of the Majorana phase 31 , relevant for the calculation of |YB |: 2/3; /2; /3.
113
The two maxima and the deep minimum of |YB | are evident in the figure. The maximal values of
|YB | are reached for 31
= 2/3. Actually, in what concerns the dependence of |YB | on 31 and
13 in the case of s13 = 0, the following relation holds: |YB (13 , 31 )| = |YB (13 , 2 31 )|.
The two maxima in question occur at m1
= 7.7 104 eV and at m1
= 5.5 103 eV. At these
maximum points we have (0.66m
) (0.71m
2)
= (0.044) and (+0.047), respectively. The
complete compensation between (0.66m
) and (0.71m
2 ), leading to |YB |
= 0, takes place at
m1 1.5 103 eV. For 31 = 2/3, the baryon asymmetry at the two maxima reads:

|m2 |

M1
A
12
|YB | = 1.5(1.4) 10
(53)
.
0.05 eV
109 GeV
Thus, one can have successful leptogenesis for M1 5.3 1010 GeV.
Similar analysis can be performed assuming that the only source of CP-violation in leptogenesis is the Dirac phase in the neutrino mixing matrix. The results one obtains in this case
are analogous to those derived in the last part of Section 3.1 (see Eqs. (38)(40) and the related
discussion).
4. Conclusions
In the present article we have investigated the dependence of the flavoured (thermal) leptogenesis on the lightest neutrino mass in the case when the CP-violation necessary for the
generation of the observed baryon asymmetry of the Universe is due exclusively to the Majorana
and/or Dirac CP-violating phases in the PMNS neutrino mixing matrix UPMNS . The two possible
types of light neutrino mass spectrum allowed by the data were considered: (i) with normal ordering (m2A > 0), m1 < m2 < m3 , and (ii) with inverted ordering (m2A < 0), m3 < m1 < m2 .
The study was performed within the simplest type I see-saw scenario with three heavy Majorana neutrinos Nj , j = 1, 2, 3, having a hierarchical mass spectrum with masses M1 M2,3 .
Throughout this analysis we used the orthogonal parametrisation of the matrix of neutrino
Yukawa couplings, involving an orthogonal matrix R, R T R = RR T = 1. The latter, in general,
can be complex, i.e., CP-violating. In the present work we were primarily interested in the possibility that R is real, conserves the CP-symmetry, and the violation of CP-symmetry necessary
for leptogenesis is due exclusively to the CP-violating phases in UPMNS . In the case of hierarchical heavy Majorana neutrinos N1,2,3 , M1 M2 M3 , the CP-violating lepton charge (flavour)
asymmetries l , l = e, , , relevant in leptogenesis, are produced in the decays of the lightest
one, N1 . As a consequence, the generated baryon asymmetry YB depends (linearly) on the mass
of N1 , M1 , and on the elements R1j of the matrix R, j = 1, 2, 3, present in the neutrino Yukawa
couplings of N1 .
Our analysis was performed under the condition of negligible RG running from MZ to M1 of
mj and of the parameters in UPMNS . This condition is satisfied for sufficiently small values of
the lightest neutrino mass, min(mj ) 0.10 eV. The latter requirement is fulfilled for the normal
hierarchical (NH), m1 m2 < m3 , and inverted hierarchical (IH), m3 m1 < m2 , spectra, and
for the spectra with partial hierarchy. Therefore the neutrino masses mj , the solar, atmospheric
and CHOOZ mixing angles, 12 , 23 and 13 , and the Majorana and Dirac CP-violating phases,
21,31 and , present in UPMNS , are taken at the scale MZ , at which the neutrino mixing
parameters are measured.
We have investigated in detail the case of light neutrino mass spectrum with inverted ordering
and real (and CP-conserving) matrix R, considering values of the lightest neutrino mass m3 in the
114

range 1010 eV m3 0.05 eV. The case of IH spectrum, m3 m1 < m2 , m1,2
= |m2A |, is
of major interest since, for real elements R1j of R, IH spectrum, m3
= 0 and CP-violation due
to the Majorana and/or Dirac phases in UPMNS , it was shown [9] to be impossible to generate the
observed baryon asymmetry YB
= 8.6 1011 in the regime of flavoured leptogenesis (i.e.,
12
for M1 10 GeV): the resulting baryon asymmetry is too small. Therefore our investigation
was focused primarily on the effects in leptogenesis of a non-negligible
m3 , having a value in the
6
3
range of the IH spectrum: 10 eV m3 5 10 eV, m3 m1,2 = |m2A |

= 5 102 eV.
0 or R12 =
0.
These effects can be particularly large if R11 =

We have found that in the case of IH spectrum with non-negligible m3 , m3 |m2A |, the
generated baryon asymmetry |YB | can be strongly enhanced in comparison with the asymmetry
|YB | produced if m3
= 0. The enhancement can be by a factor of 100, or even by a larger factor.
As a consequence, one can have successful leptogenesis for IH spectrum with m3 5 106 eV
even if the elements R1j of R are real and the requisite CP-violation is provided by the Majorana
or Dirac phase(s) in the PMNS matrix. The dependence of |YB | on m3 has the following characteristic features. As m3 increases from the value of 1010 eV up to 104 eV, the maximal possible
|YB | for a given M1 increases monotonically, starting from a value which for M1 1012 GeV is
much smaller than the observed one, max(|YB |) 8.6 1011 . At m3
= few 106 eV, we
11
11
for M1
have max(|YB |)
= 8.6 10
= 5 10 GeV. As m3 increases beyond few 106 eV,
max(|YB |) for a given M1 continues to increase until it reaches a maximum (Figs. 1, 2 and 4).
This maximum is located typically at m3
= (7.0 7.5) 104 eV. It corresponds to the CPasymmetry being predominantly in the (e + )-flavour. As m3 increases further, |YB | rapidly
decreases. At certain value of m3 , typically lying in the interval m3 (1.52.5) 103 eV or
m3 (2.010.0) 103 eV, depending on whether the CP-violation in leptogenesis is caused
by the Majorana or Dirac phases in UPMNS , |YB | goes through a deep minimum: one can have
even |YB | = 0 (Figs. 1, 2 and 4). This minimum of |YB | corresponds to a partial or complete
cancellation between the asymmetries in the -flavour and in the (e + )-flavour. The position
of the minimum in the case of Dirac CP-violation from UPMNS is very sensitive to the precise value of the atmospheric and solar neutrino mixing angles 23 and 12 . The dependence on
sin2 23 is particularly strong (Fig. 4). As m3 increases further, |YB | rapidly increases reaching
a second maximum. In magnitude the latter is of the order of the first one in the case of Majorana CP-violation from UPMNS . If the CP-violation is due to the Dirac phase in UPMNS , the
second maximum of |YB | can be of the order of the first one, or can be significantly smaller,
depending on the precise value of sin2 23 (Fig. 4). This maximum corresponds to the CPasymmetry being predominantly in the -flavour, rather than in the (e + )-flavour. For Majorana
(Dirac) CP-violation from UPMNS , the second maximum of |YB | occurs typically in the interval
m3
= (4.08.0) 103 eV (m3
= (0.77.0) 102 eV). As m3 increases further, |YB | decreases
rather slowly monotonically. If CP-symmetry is violated by the Majorana phases in UPMNS , we
can have successful leptogenesis for M1 3.0 1010 GeV. A somewhat larger values of M1
are typically required if the CP-violation is due to the Dirac phase : M1 1011 GeV. The requirement of successful flavoured leptogenesis in the latter case leads to the following lower
limits on | sin 13 sin |, and thus on sin 13 and on the rephasing invariant JCP which controls the
magnitude of CP-violation effects in neutrino oscillations: | sin 13 sin |, sin 13 (0.040.09),
|JCP | (0.0090.020), where the precise value of the limit within the intervals given depends
on the sgn(R11 R13 ) (or sgn(R12 R13 )) and on sin2 23 .
The results we have obtained for light neutrino mass spectrum with normal ordering, m1 <
m2 < m3 , depend on whether R11
= 0 or R12
= 0. If R11
= 0, we did not find any signif-
115
icant enhancement of the baryon asymmetry |YB |, generated within flavoured leptogenesis
scenario with real matrix R and CP-violation provided by the neutrino mixing matrix UPMNS ,
when the lightest neutrino mass was varied in the interval 1010 eV m1 0.05 eV: for
m1 7.5 103 eV, the produced asymmetry |YB | practically coincides with that corresponding
to m1 = 0 (Fig. 5). For m1 102 eV, the lightest neutrino mass m1 has a suppressing effect
on the asymmetry |YB | (Fig. 5). If, however, R12
= 0, the dependence of |YB | on m1 exhibits
qualitatively the same features as the dependence of |YB | on m3 in the case of neutrino mass
spectrum with inverted ordering (hierarchy) we have summarised above: |YB | possesses two
maxima separated by a deep minimum (Fig. 6). Quantitatively, max(|YB |) is somewhat smaller
than in the corresponding IH spectrum cases. As a consequence, it is possible to reproduced the
observed value of YB if the CP-violation is due to the Majorana phase(s) in UPMNS provided
M1 5.3 1010 GeV.
The results obtained in the present article show that the value of the lightest neutrino mass
in the cases of neutrino mass spectrum with inverted and normal ordering (hierarchy) can have
dramatic effect on the magnitude of the baryon asymmetry of the Universe, generated within the
flavoured leptogenesis scenario with hierarchical heavy Majorana neutrinos.
Acknowledgements
This work was supported in part by the INFN under the program Fisica Astroparticellare,
and by the Italian MIUR (Internazionalizzazione Program) and Yukawa Institute of Theoretical
Physics (YITP), Kyoto, Japan, within the joint SISSA-YITP research project on Fundamental
Interactions and the Early Universe (S.T.P.).
References
[1] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[2] V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
[3] W. Buchmller, P. Di Bari, M. Plmacher, Nucl. Phys. B 643 (2002) 367;
W. Buchmller, P. Di Bari, M. Plmacher, Ann. Phys. 315 (2005) 305.
[4] G.F. Giudice, et al., Nucl. Phys. B 685 (2004) 89.
[5] H.B. Nielsen, Y. Takanishi, Phys. Lett. B 507 (2001) 241;
W. Buchmller, D. Wyler, Phys. Lett. B 521 (2001) 291;
J. Ellis, M. Raidal, T. Yanagida, Phys. Lett. B 546 (2002) 228;
S. Davidson, A. Ibarra, Nucl. Phys. B 648 (2003) 345.
[6] M. Hirsch, S.F. King, Phys. Rev. D 64 (2001) 113005;
G.C. Branco, et al., Nucl. Phys. B 640 (2002) 202;
J. Ellis, M. Raidal, Nucl. Phys. B 643 (2002) 229;
M.N. Rebelo, Phys. Rev. D 67 (2003) 013008.
[7] S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Rev. D 68 (2003) 093007.
[8] S.T. Petcov, W. Rodejohann, T. Shindou, Y. Takanishi, Nucl. Phys. B 739 (2006) 208.
[9] S. Pascoli, S.T. Petcov, A. Riotto, Phys. Rev. D 68 (2003) 093007;
S. Pascoli, S.T. Petcov, A. Riotto, Nucl. Phys. B 739 (2006) 208.
[10] B. Pontecorvo, Zh. Eksp. Teor. Fiz. 33 (1957) 549;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 34 (1958) 247;
B. Pontecorvo, Zh. Eksp. Teor. Fiz. 53 (1967) 1717;
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
[11] G.C. Branco, R. Gonzalez Felipe, F.R. Joaquim, Phys. Lett. B 645 (2007) 432.
[12] S. Blanchet, P. Di Bari, JCAP 0703 (2007) 018.
[13] R. Barbieri, P. Creminelli, A. Strumia, N. Tetradis, Nucl. Phys. B 575 (2000) 61.
[14] H.B. Nielsen, Y. Takanishi, Nucl. Phys. B 636 (2002) 305.
116
[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]
A. Abada, et al., JCAP 0604 (2006) 004.

E. Nardi, Y. Nir, E. Roulet, J. Racker, JHEP 0601 (2006) 164.
A. Abada, et al., JHEP 0609 (2006) 010.
S. Antusch, S.F. King, A. Riotto, JCAP 0611 (2006) 011.
P. Minkowski, Phys. Lett. B 67 (1977) 421;
M. Gell-Mann, P. Ramond, R. Slansky, in: P. Van Nieuwenhuizen, D. Freedman (Eds.) Proceedings of the Supergravity Stony Brook Workshop, New York, 1979;
T. Yanagida, in: A. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified Theories and Baryon
Number in the Universe, Tsukuba, Japan, 1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
S.T. Petcov, Nucl. Phys. B (Proc. Suppl.) 143 (2005) 159, hep-ph/0412410.
V. Lobashev, et al., Nucl. Phys. A 719 (2003) 153c;
K. Eitel, et al., Nucl. Phys. B (Proc. Suppl.) 143 (2005) 197.
S. Hannestad, H. Tu, Y.Y.Y. Wong, JCAP 0606 (2006) 025.
S. Blanchet, P. Di Bari, G.G. Raffelt, JCAP 0703 (2007) 012.
J.A. Casas, A. Ibarra, Nucl. Phys. B 618 (2001) 171.
J.A. Casas, A. Ibarra, F. Jimnez-Albuquerque, hep-ph/0612289.
J.A. Casas, J.R. Espinosa, A. Ibarra, I. Navarro, Nucl. Phys. B 573 (2000) 652;
S. Antusch, et al., Phys. Lett. B 519 (2001) 238;
T. Miura, T. Shindou, E. Takasugi, Phys. Rev. D 66 (2002) 093002.
S.T. Petcov, T. Shindou, Y. Takanishi, Nucl. Phys. B 738 (2006) 219.
S.M. Bilenky, S. Pascoli, S.T. Petcov, Phys. Rev. D 64 (2001) 113003.
S.M. Bilenky, J. Hosek, S.T. Petcov, Phys. Lett. B 94 (1980) 495.
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227;
M. Doi, et al., Phys. Lett. B 102 (1981) 323.
A. Bandyopadhyay, et al., Phys. Lett. B 608 (2005) 115;
A. Bandyopadhyay, et al., 2005, unpublished.
T. Schwetz, Phys. Scr. T 127 (2006) 1.
G.L. Fogli, et al., Prog. Part. Nucl. Phys. 57 (2006) 71.
M. Apollonio, et al., Phys. Lett. B 466 (1999) 415.
S.M. Bilenky, S.T. Petcov, Rev. Mod. Phys. 59 (1987) 67.
S.T. Petcov, New J. Phys. 6 (2004) 109, http://stacks.iop.org/1367-2630/6/109;
S.T. Petcov, Phys. Scr. T 121 (2005) 94, hep-ph/0504166;
S. Pascoli, S.T. Petcov, hep-ph/0308034;
C. Aalseth, et al., hep-ph/0412300;
A. Morales, J. Morales, Nucl. Phys. B (Proc. Suppl.) 114 (2003) 141.
P. Langacker, et al., Nucl. Phys. B 282 (1987) 589.
P.I. Krastev, S.T. Petcov, Phys. Lett. B 205 (1988) 84.
F. Ardellier, et al., Double Chooz Collaboration, hep-ex/0606025.
See, e.g., K.M. Heeger, Talk given at Neutrino06 International Conference, Sant Fe, USA, 1319 June 2006.
See, for instance, G. De Lellis, et al., Neutrino factories and superbeams, in: Proceedings, 7th International Workshop, NuFact05, Frascati, Italy, 2126 June 2005.
C. Albright, et al., physics/0411123;
Y. Itow, et al., hep-ex/0106019;
D.S. Ayres, et al., hep-ex/0503053.

Flat-brane compactifications
in supergravity induced by scalars
Alex Kehagias , Constantina Mattheopoulou
Department of Physics, National Technical University of Athens, GR-15773 Zografou, Athens, Greece
Received 26 November 2007; accepted 21 December 2007
Abstract
We discuss flat compactifications of supergravities in diverse dimensions in the presence of branes. The
compactification is induced by the scalar fields of supergravity and it is such that there is no relic cosmological constant on the brane, rendering this way the latter flat. We discuss in particular the D = 4,
N = 2, 4 and D = 8, N = 1 supergravities with n = 1, 2, 3 vector multiplets where the scalar manifolds
are Grassmannian cosets of the form SO(2, n)/SO(2) SO(n). By introducing branes at certain points in
the transverse space, finite energy solutions to the field equations are constructed. Some of the solutions we
present may be interpreted as intersecting branes.
PACS: 04.65.+e; 11.25.Mj
Keywords: Supergravity; Compactification; Stringy cosmic strings; Modular group; Intersecting branes
1. Introduction
Codimension-two brane solutions of gravitational theories have attracted much interest the recent years. The distinguishing features of such solutions are: (i) the brane worldvolume is always
Ricci-flat, irrespective of any vacuum energy induced on the brane, and (ii) the internal space
has the same local geometry as it would have in the absence of branes, apart from conical deficit
angles proportional to the brane tensions [116]. The first property motivates the study of such
solutions in the context of six-dimensional theories, since the corresponding Ricci-flat 3-branes
E-mail addresses: kehagias@central.ntua.gr (A. Kehagias), conmat@central.ntua.gr (C. Mattheopoulou).

doi:10.1016/j.nuclphysb.2007.12.027
118
A. Kehagias, C. Mattheopoulou / Nuclear Physics B 797 (2008) 117136
provide a new perspective for a possible solution of the cosmological constant problem [1723].
More generally, codimension-two brane solutions can be examined in gravitational theories in
diverse dimensions, where they correspond to defects of dimension lower or higher than four.
The codimension-two solutions mentioned above may be triggered by matter fields appearing
in the theory, such as p-form gauge fields [2,57] or, most importantly, by scalar fields. On the
other hand, for the case of sigma models with a compact target space, solutions of this type
have been found in [12,13]; however, sigma models with such scalar manifolds do not occur
in supergravity. For the case of non-compact sigma models, there are two prototype solutions.
The first type of solutions [10] generalize the teardrop solution of [24] to account for the
presence of branes; here the internal 2-dimensional manifold is a non-compact space of finite
volume [2527] and the geometry has a naked singularity at its boundary which, however, may
be rendered harmless by imposing appropriate boundary conditions. These boundary conditions
guarantee that the conservation laws of the theory are not spoiled and energy, momentum angular
momentum, etc. do not leak from the boundary. The second type of solutions are based on the
stringy cosmic string of [28]. In this case, the internal geometry can be non-singular provided
that the brane tensions are restricted to a certain range, and, in fact, correspond to a compact
manifold of Euler number 2 provided that the brane tensions are appropriately fine-tuned [31].
Moreover, the existence of modular symmetries in the non-compact case guarantees that the
scalars and the metric are actually single-valued, unlike the compact case where this issue is not
clear.
In this paper we present codimension-two solutions of supergravity models in diverse dimensions, in the presence of branes. In particular, we consider D-dimensional supergravity theories
coupled to nonlinear sigma models, with the sigma-model target spaces being non-compact Khler manifolds. We seek exact solutions of the form M Dn K, where M Dn is a flat Minkowski
space and K is an n-dimensional internal space. As concrete examples, we consider the cases of
N = 4 and N = 2 supergravity in 4 dimensions (with the solutions corresponding to strings), as
well as the cases of minimal supergravities coupled to vector multiplets in 8 dimensions (with
the solutions corresponding to parallel or intersecting five-branes). For all of the above cases, the
SO(2,n)
scalar manifold is special Khler of the form SL(2,R)
U (1) SO(2)SO(n) [3234] or a Grassmannian
SO(2,n)
coset SO(2)SO(n)
[35]. To find the explicit solutions, we follow the guidelines of [28], employing
a holomorphic ansatz for the scalars that restricts the latter to lie in the fundamental domain of the
modular groups and allowing modular SL(2, Z), SL(2, Z) SL(2, Z) or Sp(4, Z) jumps around
certain points in the internal space. This leads to scalar field configurations with finite energy per
unit volume. Explicit solutions may be presented only if n 3, since it is only for these cases that
the modular forms required to construct the solutions are explicitly known. The solutions under
consideration generically possess singularities appearing in the form of conical deficit angles; in
order to arrive at non-singular solutions, one has to arrange the total deficit angle to be 4 in
which case the internal space compactifies to S2 .
The solutions described above can be generalized to include brane configurations as well.
In this case, the branes are introduced at the points where the scalar fields diverge leading to
extra delta-function contributions to the scalar energy-momentum tensor. The branes induces
further deficit angles, proportional to their tensions in the internal space. In such a scenario, the
requirement for the absence of conical singularities may be fulfilled by suitably tuning [1,12,31]
the brane tensions so that the total deficit angle equals 4 .
The structure of this paper is as follows. In Section 2, we review the solution of the equations
of motion for a gravitational theory coupled to a Khler sigma model, we examine the prototype
stringy cosmic string solution, and we also comment of the issue of codimension-four solutions.
119
In Section 3, we describe the essential aspects of the Khler sigma models under consideration
and we state the explicit form of the Khler potentials in terms of the supergravity fields. In
Section 4, we apply the above results to the cases N = 4 and N = 2 supergravity in 4 dimensions
and we present the corresponding stringy cosmic string solutions, while we also consider the
case of minimal supergravity in 8 dimensions and we present the corresponding 5-brane solution
as well as a four-dimensional intersecting-brane solution. Finally, in Section 5 we summarize our
main results.
2. Khler sigma models
Our general setup corresponds to a D-dimensional theory of gravity coupled to a Khler sigma
model. This is a generic situation in almost all supergravity theories in four or higher dimensions.
Without specifying the scalar sigma model in detail at this stage, we assume that its target space
is a Khler manifold M spanned by the complex coordinates ( i , j ) and characterized by the
Khler potential K( i , j ) and the metric Ki j ( i , i ) = i j K( i , j ). The dynamics of this

system is described by the action

i j
D
D2 1
i M j
R Ki j , M ,
S = d x gM
(2.1)
2
where MD2 is the D-dimensional Planck mass. The equations of motion as follow from (2.1)
are the scalar equation of motion

1
M gKi j M i j Ki k M i M k = 0,
g
and the Einstein equation

RMN = Ki j M i N j + M j N i .
(2.2)
(2.3)
We are looking for solutions of the form M Dn K, where M Dn is a flat (D n)-dimensional

Minkowski spacetime parametrized by the coordinates x and K is an internal complex man
ifold, parametrized by the complex coordinates (za , z b ) and metric ka b (za , z b ). To solve (2.2),
(2.3), we assume that the scalars depend only on the internal coordinates, so that the solution we
are after is

i = i za , z b .
ds 2 = dx dx + ka b za , z b dza d z b ,
(2.4)
Inserting this ansatz in Eqs. (2.2) and (2.3), and making use of standard Khler identities such as
[k Ki]j = 0 and [c ka]b = 0, we find that these equations reduce to
and
k a b a b i + K i j k Kl j k a b a k b l = 0,
(2.5)

a b ln det ka b = Ki j a i b j + a j b i ,
(2.6)
respectively. Starting from the scalar equation (2.5), we see that it is automatically satisfied when
the i are holomorphic or antiholomorphic functions. Restricting for definiteness to the holomorphic case,

i = i za ,
(2.7)
120
we find that the Einstein equation (2.6) reduces to
a b ln det ka b = Ki j a i b j = a b K,
(2.8)
and hence it is solved by

2
i j

det ka b za , z b = eK( , ) F za ,
(2.9)
where F (za ) is an arbitrary holomorphic function. Given a specific scalar manifold M and a
specific ansatz for the moduli i (za ), the choice of F (za ) is dictated by the symmetries of the
moduli space and by geometric properties such as the non-degeneracy of the metric and the
absence of curvature singularities.
Generalizing our solution, we may also consider a (D n 1)-brane located at za = 0 to
which the scalar fields are not coupled. These branes contribute an additional energymomentum
tensor to the right-hand side of Einstein equations of the form1
T = g T0 (n) (z),
Tab = Ta b = Ta b = 0,
(2.10)
where , = 0, . . . , D n 1 and T0 is the tension of the brane located at the origin. Then,
(2.8) is changed to
a b ln det ka b = Ki j a i b j ka b
T0
MD2
(n) (z),
(2.11)
which can be solved under certain conditions, as we will see below.

2.1. Complex dimension one
We will first consider the case where the internal space has complex dimension one, which is
relevant for seeking codimension-two brane solutions with an internal compact or non-compact
space. Parametrizing the transverse space by the complex coordinate z, the explicit form of the
solution (2.9) reads

2
ds 2 = dx dx + eK F (z) dz d z .
i = i (z),
(2.12)
The energy density of such configurations can be written in the BPS-like form

i
i
2
i j
d z Ki j =
K,
E=
2
2
K
(2.13)
(K)
where and are Dolbeault operators and in the second integral, the domain of integration has
been pulled back to the image (K) of the internal manifold K. In the presence of brane with
tension T0 , the above relations are modified to

2
D2
ds 2 = dx dx + eK F (z) |z|T0 /M dz d z
(2.14)
and
i
E=
2
1 Where
j+
d 2 z Ki j i
n
d z det ka b (n) (z) = 1.
T0
MD2
(2.15)
121
Although the first form of the integral in (2.13) may appear to give a zero answer due to the
fact that the integration domain is a compact surface, the Khler potential is not globally welldefined on M and hence the integral may contain jumps that render it nonzero. This is made
more explicit in the second form of the integral in terms of the Khler potential. Indeed, if there
exist symmetries of M (such as modular invariance for example) that result in a bounded (K)
of finite volume, the integral may give a nonzero result due to boundary terms. The energy per
unit volume is alternatively expressed as
E = 2,
(2.16)
where is the Euler characteristic of K.

2.1.1. Stringy cosmic strings
An example of particular importance, which will serve as the prototype of the solutions to
be constructed later on, refers to the case where the moduli space consists of a single toroidal
modulus = 1 + i2 , such as the type IIB axion-dilaton or the complex-structure or Khler modulus of an internal compactification torus. In this case, the scalar manifold is the SL(2, R)/U (1)
space, which is a Khler manifold with Khler potential

K = ln i( ) ,
(2.17)
so that the effective action (2.1) takes the explicit form

D
D2 1
R+
.
S = d x gM
2
( )2
The solution to the equations of motion for the above action as we have seen are

2
= (z),
ds 2 = dx dx + 2 F (z) dz d z ,
whose energy per unit volume reads

i
E=
d 2 z ln 2 .
2
(2.18)
(2.19)
(2.20)
To discuss the possible choices of (z), we first note that the effective action (2.15) has a
symmetry under the SL(2, R) group, acting as
a + b
(2.21)
,
ad bc = 1.
c + d
In the full theory, this symmetry group is broken to the modular group PSL(2, Z), which is
interpreted as a local symmetry. As a result, the space of inequivalent choices of the modulus
is the quotient of the complex plane by the PSL(2, Z) group, which can be taken to be the
fundamental domain F1 specified by the conditions
1
|1 | ,
(2.22)
| | 1,
1 0 if | | = 1.
2 > 0,
2
The above considerations imply that an arbitrary holomorphic ansatz for (z) is generally inconsistent since is restricted to live on F1 while z covers the whole Riemann sphere; this is
alternatively verified by noting that for such a naive choice, e.g. (z) = zn , the energy per unit
volume diverges. To construct consistent, finite-energy solutions we need a holomorphic function that provides a one-to-one mapping fundamental domain F1 to the Riemman sphere. This
122
mapping is provided by the modular function j ( ) or, equivalently, by Kleins absolute invariant
J ( ) = j ( ) 744 =
41E4 ( )3 + 31E6 ( )2
,
72( )
(2.23)
where E4 ( ), E6 ( ) are the Eisenstein series of weight 4 and 6 respectively, and ( ) =

(E4 ( )3 E6 ( )2 )/1728 is the cusp form of weight twelve. The modular invariant has the asymptotic behavior
J ( ) q 1 ,
as 2 .
(2.24)
Finite-energy solutions can then be constructed by equating J ( (z)) to a holomorphic function

of z. Doing so, pulling back the integral in (2.20) from the z-plane to F1 , and converting it into
a line integral over the boundary of F1 , we indeed obtain the finite expression
iN
d 2 ln 2 = N,
E=
(2.25)
2
6
F1
where N is the number of times the z-plane covers F1 . Here, we will consider the rational maps

P (z)
,
J (z) =
Q(z)
(2.26)
where P (z) and Q(z) are polynomials in z of degrees p and q respectively. For this choice, the
integer N is equal to q if p q (in which case J ( ) approaches a constant value at infinity and
diverges as (z zi )1 at the zeros of Q(z) which are identified with the cores of the solutions)
and equal to p if p > q (in which case J ( ) diverges as zpq at infinity). In what follows we will
consider the p < q case where J = b/(z zi ). Given this ansatz for (z), it remains to choose
the function F (z) in (2.19) in such a way that the metric is modular-invariant and non-degenerate.
The first requirement is fulfilled by noting that the PSL(2, Z) transformation of 2 is given by
2
2
,
|c + d|2
(2.27)
and hence can be compensated by multiplying 2 by |f ( )|2 where f ( ) is an PSL(2, Z) modular

form of weight 1. This modular form is explicitly given in terms of the square of Dedekind eta
function
( ) = q
1/24

1 q n = 1/24 ,
q = e2i .
(2.28)
n=1
So the combination 2 |( )|4 is modular invariant. To fulfil the second requirement we note
that, near the zeros zi , we can write q 1 J ( ) (z zi )1 which
implies from (2.28) that

1/12 leads to the
2 |( )|4 |(z zi )1/12 |2 . Therefore, the choice F (z) = ( )2 N
(z
i=1 zi )
modular-invariant, non-degenerate solution
N
2

4

ds 2 = dx dx + 2 ( ) (z zi )1/12 dz d z ,

i=1

P (z)
J (z) =
.
Q(z)
(2.29)
123
As |z| , the internal metric approaches kzz (zz)N/12 . Hence, by standard arguments, the
solution has a deficit angle = 6 N , which is equal to the energy as expected. So, at infinity, the
internal space has conical singularities which signify the geometry of a non-compact space. For
N = 12 the internal space is cylindrical, while for N = 24 the internal space compactifies to S2 .
The above considerations are quite general and apply in all cases where toroidal moduli with
PSL(2, Z) symmetry exist. For instance, we can consider the case of D = 4 theories arising e.g.
from string compactifications on an internal space containing a T2 . Then the resulting solution
(2.29) corresponds to a configuration of N strings carrying charge under the toroidal modulus
field. This solution is the stringy cosmic string of [28]. It has been employed to describe the D7
brane configuration [29] as well as the D7/D3 brane system [30].
There is also the possibility to add (D 3)-brane sources at the points zi where the scalar
field diverges. In this case, an energymomentum tensor of the form
T =
Ti (2) (z zi ),
Tzz = 0,
(2.30)
i=1
should be introduced, where Ti is the tension of the brane located at the point zi . These branes
cause additional deficit angles equal to their tensions. Then, from (2.14) we find that the modularinvariant metric is
N
2

4

2
1/12
ds = dx dx + 2 ( ) (z zi )

i=1
N
2

D2

(2.31)
(z zi )Ti /2M dz d z .

i=1
At infinity the total deficit angle turns out to be

Ti
,
= N+
6
MD2
N
(2.32)
i=1
which is equal to the energy (2.16) as expected. In order for the internal space to compactify to
must be equal to 4 . This amounts to a fine-tuning condition on the
S2 , the above deficit angle
D2 (1 N ).
brane tensions Ti , namely N
i Ti = 4M
24
At the vicinity of each brane (z zi ), where J ( ) , the internal metric becomes kzz
D2
2 (z zi )Ti /M . Then, contracting (2.11) with k zz one deduces that the Ricci scalar is not
singular for Ti > 2MD2 . But the true condition for the absence of curvature singularities
follows when the previous condition and Eq. (2.32) are both satisfied. These conditions restrict
the number of branes to N = 1, in accordance with the result of [31]. Note that in the absence of
the extra term involving the tension in kzz there is a curvature singularity.
2.2. Complex dimension two
We next proceed to the case where the internal space has complex dimension two [36],
which is relevant for seeking codimension-four brane solutions with an internal compact or noncompact space. Now, Eq. (2.9) takes the form

2
k11 k22 k12 k21 = eK F (z, w) ,
(2.33)
124
which is a highly nonlinear differential equation. Eq. (2.33) is very difficult to be solved given an
explicit form of the Khler potential K of the scalar manifold and a general holomorphic ansatz
for the fields i (z, w). However, for the special case where K decomposes as the sum

K i , j = K (1) A , A + K (2) B , B ,
(2.34)
with the first term involving a subset ( A , A ) of the ( i , i ) and the second term involving the
remaining fields ( B , B ), we can easily solve this equation by assuming an ansatz of the form
A = A (z),
A
B = B (w),
k = k (1) (z, z ) + k (2) (w, w),
(2.35)
where the
and
depend only on z and w and the metric is the sum of two terms involving
(z, z ) and (w, w),
respectively. Writing also F (z, w) = F (1) (z)F (2) (w), Eq. (2.33) simplifies to
2
2

(1)
(2)
(1) (2)
k11 k22 = eK F (1) (z) eK F (2) (w) ,
(2.36)
and k2(2)
equal to the first and second terms in brackets respecand is easily solved by taking k1(1)
1
2
tively. The final solution, which generalizes (2.31) then reads
A = A (z),
B = B (w),
N
2
1

(1)
D2
2

F (1) (z) (z zi )Ti /2M dz d z
ds 2 = dx dx + e

i=1
N
2
2

(2)
2

Tj /2MD2
K (2) (2)

+e
F (w) (w wj )
dw d w.

K (1)
(2.37)
j =1
We will discuss in Section 4.3 the interpretation of such a solution.

3. Special Khler and Grassmannian
The construction of solutions of the type described in the previous section carries over to more
complicated moduli spaces. Here, we will construct solutions of this form for the cases where
the classical moduli space is a special Khler manifold of the form
SL(2, R)
SO(2, n)
,
(3.1)
U (1)
SO(2) SO(n)
or a Khler manifold of the form
SO(2, n)
.
Kn =
(3.2)
SO(2) SO(n)
In what follows, we will give a brief description of the geometry of these manifolds, using the
formalism of special geometry, and we will state the corresponding Khler potentials for the
cases of interest. The geometry of the special Khler manifold SKn+1 is completely specified by
a holomorphic symplectic section [32,33]
I
X
=
(3.3)
,
FI
SKn+1 =
in terms of which the Khler potential is given by

ln i X I FI X I FI .
K = ln i|
(3.4)
125
In the above, X I , I = 0, . . . , n + 1 are a set of complex parameters, while FI are usually specified
as the derivatives of a holomorphic prepotential F (X) with respect to the X I . In the present case,
it is convenient to employ the so-called symplectic gauge in which is written as
I

X
XI
=
(3.5)
=
,
FI
SI J X J
where I J = diag(+1, +1, 1, . . . , 1) is the SO(2, n) invariant metric and S parametrizes the
SL(2,R)
SO(2,n)
I
U (1) factor in the usual way. X parametrize the SO(2)SO(n) factor and are required to satisfy
the SO(2, n) orthogonality condition
I J X I X J = 0.
(3.6)
Although this gauge choice makes it impossible to specify FI by means of a prepotential,

Eq. (3.4) for the Khler potential is perfectly valid, leading to the result [37]
K = K1 + K2 ,
(3.7)
where
K1 = ln(S S)
(3.8)
is the standard Khler potential for

K2 = ln I J X I X J
SL(2,R)
U (1)
and
(3.9)
SO(2,n)
is the Khler potential of SO(2)SO(n)
. The latter can be verified by parametrizing X I in terms of
the independent CalabiVesentini coordinates y a , a = 1, . . . , n, according to
1
2
2 (1 + y )
X I (y) = i (1 y 2 ) .
(3.10)
2
a
y
Then, it is straightforward to see that the familiar formula

2

K2 = ln 1 2y y + y 2 ,
(3.11)
is recovered.
We are particular interested in Khler manifolds of the form (3.2) with n = 1, 2, 3 for which
the modular forms required to construct our solutions are explicitly known. In what follows,
we give the explicit parametrizations of XI in terms of supergravity fields and we state the
corresponding Khler potentials for these particular cases.
n = 1. For this case, the SO(2, 1; R) vector X I is parametrized in terms of a single complex
field T as [33]
1 (1 T 2 )
2
.
X I (T ) =
(3.12)
2T
1
2
(1 + T )
2
Inserting this into (3.9), we find the Khler potential

K2 (T ) = 2 ln(T T ).
(3.13)
126
n = 2. For this case, the SO(2, 2) vector X I is parametrized in terms of two complex fields
T and U as [33,38]
1
(1 T U )
2
1 (T + U )
2
I
X (T , U ) = 1
(3.14)
,
(1 + T U )
2
1 (T
2
U)
and the Khler potential reads

K2 (T , U ) = ln (T T )(U U ) .
(3.15)
n = 3. Now, the SO(2, 3) vector X I can be parametrized in terms of three complex fields T ,
U and V as
1
(1 T U + V 2 )
2
1 (T + U )
I
1
2
X (T , U, V ) = (1 + T U V )
(3.16)
,
2
1
(T U )
2
2V
and the Khler potential reads

K2 (T , U, V ) = ln (T T )(U U ) (V V )2 .
(3.17)
To summarize, the Khler potential for the special Khler manifolds (3.1) is given by K = K1 +
K2 where K1 is given in (3.8) and K2 is given in (3.9), while the Khler potential for the Khler
manifolds (3.2) is simply K = K2 . Explicit expressions for K2 for the cases n = 1, n = 2 and
n = 3 are given in Eqs. (3.13), (3.15) and (3.17), respectively.
4. Application to supergravity theories
We may apply the results of the previous sections to construct solutions in the context of supergravity theories where scalar Khler manifolds of the sort discussed earlier appear. In particular,
we will discuss two classes of solutions. The first class corresponds to stringy-cosmic-string solutions of D = 4 supergravities with N = 4 or N = 2 supersymmetry, arising from appropriate
heterotic string compactifications. The theories under consideration possess modular symmetries
that may be exploited to construct stringy cosmic string solutions according to the guidelines of
Section 2. Moreover, for these theories, the quantum corrections to the Khler potential are under
control and thus one can extend the classical solutions to solutions that are exact to all orders in
perturbation theory. The second class of solutions corresponds to five-brane solutions of minimal
D = 8, N = 1 supergravity and four-dimensional intersections thereof.
4.1. String solutions in D = 4, N = 4 supergravity
We first consider the case of the N = 4 theories [34] arising from compactifications of the
E8 E8 or SO(32) heterotic string theories on T4 T2 or, equivalently, from compactifications
of N = 1 six-dimensional supergravity on T2 . For these models, the moduli space consists of
127
three factors involving (i) the axion-dilaton S, (ii) the moduli T and U corresponding to the
complex and Khler structure moduli of T2 , and (iii) the moduli of T4 . In what follows, we will
consider only the first two types of moduli, which parametrize the space

SL(2, R)

SO(2, 2)
M = PSL(2, Z)
(4.1)
SO(2, 2; Z)
,
U (1)
SO(2) SO(2) T ,U
S
with the isomorphism SO(2, 2; Z)
= PSL(2, Z) PSL(2, Z) implying that the duality group is
given by the product PSL(2, Z)S PSL(2, Z)T PSL(2, Z)U . The moduli S, T and U are given
in terms of six-dimensional fields as

det gmn
g45
T = B45 + i det gmn ,
U=
+i
,
S = + ie2 ,
(4.2)
g55
g55
where and are the dilaton and axion while gmn and B45 is the metric and B-field on T2 . The
effective action for these fields follows from the Khler potential in Eq. (3.15), namely
ln(T T ) ln(U U ).
K = ln(S S)
(4.3)
It is invariant under the duality group, as well as under string/string/string triality [39] which
interchanges S, T and U .
The solution for the moduli for this case is readily obtained by taking S, T and U to be
holomorphic functions restricted to the fundamental domains of PSL(2, Z)S , PSL(2, Z)T and
PSL(2, Z)U , respectively, by relations of the form (2.26), and by inserting the Khler potential (4.3) in Eq. (2.12). This leads to a stringy cosmic string solution with transverse metric

2
d 2 = S2 T2 U2 F (z) dz d z .
(4.4)
To determine F (z), we impose the requirements of modular invariance and non-degeneracy of
the metric as before. The first requirement leads to a factor of |(S)(T )(U )|4 while the second
requirement leads to a factor of |(z zi )1/12 |2 for each string. Letting NS , NT and NU be the
number of strings carrying charge with respect to the S, T and U moduli respectively, we finally
find
N N N
2
S
U
T

4

1/12
2

d = S2 T2 U2 (S)(T )(U )
(z zi )(z zj )(z zk )
dz d z . (4.5)

i=1 j =1 k=1
Imposing string/string/string triality leads to NS = NT = NU = N . As |z| for each string

we have a deficit angle = 6 , and the energy of the solution is E = 6 (NS + NT + NU ) = 2 N .
Therefore, to compactify the transverse space to S2 , we need N = 8.
In the above we have assumed that each string is charged with respect to only a single modulus
so zi
= zj
= zk . However, string/string/string triality also allows us to consider STU-strings
that are charged under all three moduli. Such configurations may give rise to orbifold singularities
on the transverse space; in order for this to occur, we need deficit angles of the form = 2(n
1)/n where n > 1 is an integer. To discuss this, we first write the transverse metric (for the case
N = 8) as
8
2

4

2
1/4
d = S2 T2 U2 (S)(T )(U ) (z zi )
(4.6)
dz d z .

i=1
128
For the generic case where the locations zi of the strings are different, we have a deficit angle of
/2 for each string and hence no orbifold singularities occur. However, when some of the zi are
identified, such singularities appear. For example, consider the case where the eight zi coalesce
into three points z1 , z2 and z3 , of orders three, three and two respectively. Then the transverse
metric turns to be

4
2
d 2 = S2 T2 U2 (S)(T )(U ) (z z1 )3/4 (z z2 )3/4 (z z3 )1/2 dz d z ,
(4.7)
and one recognizes the deficit angles of 3/2, 3/2 and around z1 , z2 and z3 respectively.
The transverse space is thus a T2 /Z4 orbifold as we can see from for the n = 4 value. Another
example is obtained by taking the eight zi to coalesce into four points z1 , . . . , z4 , of order two
each. Then the transverse metric turns to
4
2

4

d 2 = S2 T2 U2 (S)(T )(U ) (z zi )1/2 dz d z ,
(4.8)

i=1
and one recognizes a deficit angle of for each string. The transverse space is thus a T2 /Z2
orbifold.
We may now consider sting sources located at the points zi , zj , zk where the scalar fields S,
T , U diverge with energy-momentum tensors of the form

N
NU
NT
S

(2)
(2)
(2)
T =
Ti (z zi ) +
Tj (z zj ) +
Tk (z zk ) ,
j =1
i=1
k=1
Tzz = 0.
(4.9)
In this case, the solution (4.5) changes to

N N N
2
S
U
T

4

1/12
2
d = S2 T2 U2 (S)(T )(U )
(z zi )(z zj )(z zk )

i=1 j =1 k=1
N N N
2
S
U
T

Ti /2M2
Tj /2M2
Tk /2M2

(z zi )
(z zj )
(z zk )
dz d z . (4.10)

i=1 j =1 k=1
So at infinity the transverse space compactifies to S2 , if

4 =
S
U
T

Tj
Ti
Tk
+
+
.
(NS + NT + NU ) +
2
2
6
M
M
M2
i=1
j =1
k=1
(4.11)
Note that by imposing string/string/string triality we are led again to take NS = NT = NU = N

in (4.11). Then again there are no curvature singularities when NS = NT = NU = 1 and T >
2M2 .
4.2. String solutions in D = 4, N = 2 supergravities
We next consider the case of the N = 2 theories arising from compactifications of heterotic
string theories on K3 T2 or, equivalently, from compactifications of minimal N = 1 sixdimensional supergravity coupled to vector multiplets on T2 [39]. For these models, the moduli
space consists of (i) the vector-multiplet moduli space MV parametrized by the axion-dilaton S,
129
the moduli T and U corresponding to combinations of the complex and Khler structure moduli
T i of T2 , and the Wilson line moduli V a , and (ii) the hypermultiplet moduli space MH parametrized by the moduli of K3 and of the vector bundle. Restricting to the vector multiplet moduli
space, its classical geometry is locally of the form

SL(2, R)

SO(2, n)
SO(2, n; Z)
,
MV = PSL(2, Z)
(4.12)
U (1)
SO(2) SO(n)
where n = p + 2 with p being the number of Wilson line moduli. Here, the PSL(2, Z) and
SO(2, n; Z) are the S- and T-duality groups [40]. This moduli space falls into the class of special
Khler manifolds, considered in Section 3.
For the construction of solutions of interest in the models considered here, there are two points
that need special attention. First, the PSL(2, Z) S-duality is no longer expected to be a symmetry
of the full quantum theory and so consistent solutions can be constructed only by fixing the S
modulus to a constant value and demanding invariance only under the T-duality group. Second,
the prepotential is renormalized both perturbatively and nonperturbatively, where the N = 2
non-renormalization theorems guarantee that the perturbative corrections enter only at one-loop
order. Perturbatively exact solutions can thus be constructed by taking account of the one-loop
corrections which, at the level of the Khler potential, amount to the shift

S2 S2 + VGS T i , V a ,
(4.13)
where VGS (T i , V a ) is the GreenSchwarz term. Note that S and VGS (T i , V a ) transform under
T-duality in such a way that the corresponding transformation of K is a Khler transformation.
Given these observations, we may proceed to construct stringy cosmic string solutions for the
special cases n = 1, 2, 3 where the modular forms used for the construction of invariant solutions
are explicitly known.
4.2.1. The n = 1 ST model
The ST model corresponds to the case where Wilson line moduli are absent and only the T
modulus of the torus is turned on [41]. It is obtained from the general case by setting n = 1. The
T-duality group is then
SO(2, 1; Z)
= PSL(2, Z),
(4.14)
and the classical Khler potential is read off from Eqs. (3.13), (3.8)
2 ln(T T ).
K(S, T ) = ln(S S)
(4.15)
In the quantum theory, the above relation is modified by setting S2 S2 + VGS (T ).

As remarked earlier on, the stringy cosmic string solutions of interest are constructed by
fixing the S modulus to some constant value and imposing invariance under PSL(2, Z)T and
non-degeneracy of the metric. The former requirement now leads to a factor of |(T )|8 while
the second requirement leads to a factor of |(z zi )1/6 |2 for each string (the different powers
are due to the factor of two appearing in the Khler potential). Therefore, our solution for the
transverse metric reads
N
2

8

2
2
2
2
1/6
ds = dt + dx + (S2 + VGS )T2 (T ) (z zi )
(4.16)
dz d z .

i=1
= 3 ,
and the total energy is E = 3 N , i.e. the

At infinity for each string we have a deficit angle
energy per string is twice that of the N = 4 solution. The generalized solution becomes as (2.31).
130
4.2.2. The n = 2 STU model

The STU model corresponds to the case where Wilson line moduli are absent and both moduli
of the torus are turned on [39,42]. It is obtained by the general case by setting n = 2. The classical
T-duality group is in this case
SO(2, 2; Z)
= PSL(2, Z)T P SL(2, Z)U ,
and the classical Khler potential is read off from Eqs. (3.15), (3.8)

ln (T T )(U U ) .
K(S, T , U ) = ln(S S)
(4.17)
(4.18)
There is also a Z2 symmetry corresponding to the exchange T U . In the quantum theory,

Eq. (4.18) is similarly modified by setting S2 S2 + VGS (T , U ), while the Z2 symmetry mentioned above is broken.
The stringy cosmic string solution is constructed as before, and the result for the transverse
metric is
N N
2
U
T

4

d 2 = (S2 + VGS )T2 U2 (T )(U )
(4.19)
(z zi )1/12 (z zj )1/12 dz d z .

i=1 j =1
As |z| for each string we have a deficit angle = 6 , the total energy is E = 6 (NT + NU )
and due to the fact that the Z2 exchange symmetry is broken, the numbers NT and NU may be
different. Regularity of the solution requires NT + NU = 24. The generalized solution becomes
as in (4.10) with the factors corresponding to the NS strings omitted.
4.2.3. The n = 3 STUV model
The final case we will consider here is the STUV model [43], which corresponds to turning
on a single Wilson line modulus in addition to the two moduli of the torus. It is obtained by the
general case by setting n = 3 so that classical T-duality group is
SO(2, 3; Z)
= Sp(4, Z).
(4.20)
In the genus two case the moduli space is the quotient of the Siegel upper half space by the
modular group PSp(4, Z), which can be taken to be the fundamental domain F2 , and is parametrized by the period matrix , which transforms according to (A + B)(C + D)1 .
This matrix is specified as

T V
=
.
(4.21)
V U
A Siegel modular form Fw of weight w is defined as a holomorphic function of that transforms
as

w
Fw () det(C + D) Fw ().
(4.22)
Any such form admits a Laurent expansion in the parameters q = e2iT , r = e2iV and s = e2iU .
The graded ring of Siegel modular forms is generated [44,45] by four forms of weight 4, 6, 10
and 12, namely by the two Eisenstein series 4 and 6 and the two cusp forms 10 and 12 . In
the degeneration limit 0, where Sp(4, Z) degenerates to SL(2, Z) SL(2, Z), the genus two
surface can be constructed from two tori with modular parameters q1 = e2iT and q2 = e2iU
[46,47]. These two tori are joined by excising a disk of radius || from each torus and making
131
an appropriate identification of two annular regions around the excised disk. In this limit the
relations between the parameters T , U , V and T , U , are as follows [47]

T = T + O 2 ,
(4.23)
U = U + O 2 ,
V = + O 3 .
Turning now to the classical Khler potential, this is read off Eqs. (3.17), (3.8)

ln (T T )(U U ) (V V )2
K(S, T , U, V ) = ln(S S)
ln det( ).
= ln(S S)
(4.24)
Again, in the quantum theory, Eq. (4.24) is modified by setting S2 S2 + VGS (T , U, V ).

The stringy cosmic string solution for the model under consideration is obtained by generalizing the standard procedure to the Sp(4, Z) case. First, the space of inequivalent choices for
is, as said, the fundamental domain F2 specified by the conditions
1
|T1 |, |U1 |, |V1 | ,
0 |2V2 | T2 U2 ,
2

det(C + D) 1 for all A B Sp(4, Z).
C D
(4.25)
To construct finite-energy solutions, we need a set of holomorphic functions that provide a map
from the variable , which is restricted to live on F2 according to (4.25), to the Riemann sphere,
i.e. the Sp(4, Z) counterparts of the J -function. Such functions exist (known as Igusa invariants
[48]) and are explicitly given in terms of the Sp(4, Z) Eisenstein series 4 , 6 and the cusp forms
10 , 12 as follows
x1 =
2
4 10
2
12
x2 =
3
6 10
3
12
x3 =
6
10
5
12
Using the Khler potential (4.24), we find the transverse metric

2
d 2 = (S2 + VGS ) det 2 F (z) dz d z ,
(4.26)
(4.27)
where 2 equals to Im and now the function F (z) must be chosen so as to enforce Sp(4, Z)
modular invariance and non-degeneracy of the metric. To ensure modular invariance, we note
that the Sp(4, Z) transformation of det 2 reads
det 2
det 2
| det(C + D)|2
(4.28)
and hence can be compensated by multiplying by |f ()|2 , where f () is an Sp(4, Z) modular

form of weight 1 as follows from (4.22) with no zeros on the fundamental domain F2 . The
unique form with these properties is given by the twelfth root of the cusp form 12 , i.e. f () =
1/12
12 (). In order to have non-degenerate metric, we note that the poles of Igusa invariants are
determined by the zeros of the cusp form 12 , as one can see from Eq. (4.26). This cusp form
has zeros in the z-plane at the locus q = s = 0, where the locations of the T -, U -string cores zi ,
zj exist. As we go around such a string, should undergo an Sp(4, Z) transformation generated
by the Sp(4, Z) matrices

122
si
Ti =
(4.29)
,
0
122
132
where
s1 =

1 0
,
0 0

s2 =

0 0
,
0 1

s3 =

0 1
.
1 0
(4.30)
This leads to the Sp(4, Z) jumps + si , or in terms of T , U, V ,

T T + 1,
U U + 1,
V V + 1.
(4.31)
These monodromies and holomorphicity require that near the core of the string, we will have
T
1
ln(z zi ),
2i
1
ln(z zj ),
2i
1
ln(z zk ),
2i
(4.32)
r = e2iV (z zk ).
(4.33)
so that
q = e2iT (z zi ),
s = e2iU (z zj ),
Note that, due to (4.25), V should degenerate together with T and/or U , i.e., zk should coincide
with zi and/or zj .
Turning to 12 , its full expansion is given in [46], which to leading order reads
12 = 96qs + .
(4.34)
Then, from (4.33) and (4.34) follows that the form of F (z) in the transverse metric is determined
1/12 NT NU
1/12 (z z )1/12 . This leads to the modular invariant,
to be F (z) = 12
j
i=1 j =1 (z zi )
non-degenerate solution
N N
2
U
T

d 2 = (S2 + VGS ) det 2 |12 |1/6
(4.35)
(z zi )1/12 (z zj )1/12 dz d z .

i=1 j =1
As |z| for each string the deficit angle is =
and the energy is indeed finite,
(4.36)
(NT + NU ).
6
Regularity of the solution demands that NT + NU = 24.
In the degeneration limit 0, Sp(4, Z) degenerates to SL(2, Z) SL(2, Z). In this limit,
the Eisenstein series 4 , 6 and the cusp forms 10 , 12 take the form

4 = E4 (q1 )E4 (q2 ) + O 2 ,

6 = E6 (q1 )E6 (q2 ) + O 2 ,

10 = 2 (q1 )(q2 ) + O 4 ,

12 = (q1 )(q2 ) + O 2 ,
(4.37)
E=
where E4 (qi ), E6 (qi ), (qi ), i = 1, 2 are the weight 4 and 6 Eisenstein series and the cusp form
of weight 12 respectively for each SL(2, Z) factor. Then a linear combination of Igusa invariants
x1 , x2 , x3 , gives again a modular invariant form. In particular, using the linear combination
(x1 )3 + (x2 )2 x3 (4 )3 + (6 )2 12
=
,
x3
12
(4.38)
133
31
where = 41
72 , = 72 , = 732096 and substituting the expressions (4.37) for 4 , 6 , 12 ,
which are valid in the limit 0, leads to the modular invariant form J (q1 )J (q2 ) corresponding
to the SL(2, Z) SL(2, Z) case.
Now the zeros of the cusp form 12 are easily found, as for 0 12 (q1 )(q2 ),
so that 12 0 for (q1 , ) (0, 0) and (q2 , ) (0, 0). This implies that near the zeros zi
1/12
and zj , we can write det 2 |12 |2 |(z zi )1/12 (z zj )1/12 |2 . Therefore, the appropriate
1/12 NT NU
1/12 (z z )1/12 . Then using the fact that
choice for F (z) is F (z) = 12
j
i=1 j =1 (z zi )
12 (q1 )(q2 ) and det 2 = T2 U2 , one recovers the solution of the STU model appeared in
the previous section.
4.3. Brane solutions in D = 8, N = 1 supergravity

SO(2,n)
Another situation where Khler manifolds of the type Kn = SO(2)SO(n)
examined in Section 3
occur is N = 1 supergravity coupled to n vector multiplets in eight dimensions [35]. Each vector
multiplet contains 2 scalars so that the total 2n scalars parametrize the coset Kn . For this case,
we can construct codimension-two solutions for n = 1, 2, 3 corresponding to five-branes. The
n = 1, 2 cases also appear as solutions to minimal D = 9 and D = 7 supergravities [49] coupled
to two vector multiplets.
Starting from codimension-two solutions, these can be constructed by considering the
SO(2,n)
SO(2)SO(n) Khler potential (K2 in the notation of Section 3). The resulting transverse metrics
are readily obtained from those of Section 4.2 by simply discarding the S modulus. Therefore,
for the case n = 1 where there exists a single modulus T , we obtain the solution
N
2

8

2
2
2
1/6
2
ds = dt + dx5 + T2 (T ) (z zi )
(4.39)
dz d z ,

i=1
dxp2
the spatial metric on the world-volume of a p-brane. The generalwhere we will denote by
ized solution is like (2.31). In the D = 9, N = 1 supergravity coupled to n vector multiplets the
scalars parametrize the coset SO(1, n)/SO(n). It is clear that for two vector multiplets coupled
to gravity the codimension-two solution is like (4.39).
For the case n = 2 where there exist two moduli T and U , we find
N N
2
U
T

4

2
2
2
1/12
1/12
ds = dt + dx5 + T2 U2 (T )(U )
(z zi )
(z zj )
dz d z . (4.40)

i=1 j =1
Finally, for the case n = 3 where there exist the three moduli T , U and V , we have
N N
2
U
T

2
2
2
1/6
1/12
1/12
(z zi )
(z zj )
ds = dt + dx5 + det 2 |12 |
dz d z .

(4.41)
i=1 j =1
The generalized solution is as (4.10) by discarding the NS strings. In the D = 7, N = 2 supergravity coupled to n vector multiplets, 3n scalars parametrize the coset SO(3, n)/SO(3) SO(n).
When the number of vector multiplets is two then the codimension-two solution is like (4.41).
Turning to codimension-four solutions, these can be constructed according to guidelines of
Section 2.2. The simplest possible situation is when the Khler potential K decomposes as in
Eq. (2.34) and is realized when n = 2, in which case we have
K(T , U ) = K (1) (T ) + K (2) (U ) = ln(T T ) ln(U U ).
(4.42)
134
Then, setting T = T (z) and U = U (w) we obtain k11 = T2 |F (1) (z)|2 and k22 = U2 |F (2) (w)|2 .
Determining the functions F (1) (z) and F (2) (w) in the usual manner, we finally obtain the metric
(2.37), in the presence of tensions
N
2 N
2
T
T

4
(1)
2
2
2
1/12
Ti /2M6

ds = dt + dx3 + T2 (T ) (z zi )
(z zi )
dz d z

i=1
i=1
N
2 N
2
U
U

(2)

4
6

+ U2 (U ) (w wj )1/12 (w wj )Tj /2M dw d w.

(4.43)

j =1
j =1
The deficit angles at infinity in the z, w-plane are

T

Ti
,
NT +
6
M6
1 =
2 =
i=1
(2)
U T

j
,
NU +
6
M6
(1)
(4.44)
j =1
respectively. With i = 4 , we get an S2 S2 compactification of the D = 8, N = 1 supergravity.

We may easily interpret the solution (4.43) by calculating the corresponding energy
NT (i)
NU (j )
+
momentum tensor TMN . We may write TMN = TMN

i T1,MN +
j T2,MN , where TMN
(i)
is the scalar energymomentum tensor and TMN is the contribution of the brane located at the
point zi . Then, by going to real coordinates z = x 4 + ix 5 , w = x 6 + ix 7 we find that
(i)
(1)
T1, = Ti (2) (z zi ),
(j )
T2, = Tj(2) (2) (w wj ),
(i)
(1)
(i)
T1,mn = gmn Ti (2) (z zi ),

(j )
T2,mn = 0,
(j )
T1,rs = 0,
(2)
T2,rs = grs Tj (2) (w wj ).

(4.45)
where (m, n = 4, 5) and (r, s = 6, 7) in (4.45) represents intersecting five-branes of tensions T (1) ,
T (2) with world-volumes extended across the (012345) and (012367) directions. Their common
world-volume in the (0123) direction is the 4D Minkowski intersection.
5. Conclusions
We have presented here codimension-two solutions of supergravity models in diverse dimensions, with or without brane sources. We have considered in particular D-dimensional supergravity theories coupled to a set of scalar fields forming a nonlinear sigma model targeted on
some non-compact manifold. The scalar manifolds employed are special Khler of the form
SL(2,R)
SO(2,n)
SO(2,n)
U (1) SO(2)SO(n) [3234] or the Grassmannian cosets SO(2)SO(n) [35]. The solutions we
found are of the general form M Dn K, where M Dn is a flat Minkowski space and K is an
n-dimensional internal space. We tried to keep the discussion as general as possible. However,
for concreteness we have considered the cases of N = 4 and N = 2 supergravity in 4 dimensions as well as minimal supergravities coupled to vector multiplets in 8 dimensions. In the first
case, the solution presents a string and the 4D spacetime is compactified down to two dimensions
by a number of such strings. In the former case, the solution presents a five-brane or intersecting five-branes along four-dimensional flat space, compactifying this way the eight-dimensional
supergravity down to 4D Minkowski spacetime.
The explicit solutions were found by employing a holomorphic ansatz for the scalars. The latter were restricted to lie in the fundamental domain of the modular groups and allowing modular
135
SL(2, Z), SL(2, Z) SL(2, Z) or Sp(4, Z) jumps around certain points in the internal space. This
modular jumps permit scalar field configurations with finite energy per unit volume and explicit
solutions presented only for those cases where the modular forms required to construct the solutions were explicitly known. Note that in order the solutions to have finite energy, one has to
arrange the total deficit angle produced by the scalar configurations to be 4 in which case the
internal space compactifies to S2 .
All the solutions we have described have be generalized to include brane configurations as
well, with the only requirement that the scalars of the theory do not couple to the branes. The
latter induces further deficit angles, proportional to their tensions in the internal space. The requirement for the absence of conical singularities may be fulfilled by suitably tuning the brane
tensions so that the total deficit angle equals 4 and leading to a smooth sphere compactification.
It should also be noted that configurations of this type might be relevant for the solution of the
cosmological constant problem as the world-volume of the branes are always flat irrespectively
of any bulk dynamics.
Acknowledgements
This work is cofunded by the European Social Fund (75%) and National Resources (25%)
(EPEAEK II)PYTHAGORAS. We would like to thank S. Randjbar-Daemi for extensive correspondences and detailed comments on the paper.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
R. Sundrum, Phys. Rev. D 59 (1999) 085010, hep-ph/9807348.

S.M. Carroll, M.M. Guica, hep-th/0302067.
C. Csaki, J. Erlich, C. Grojean, T.J. Hollowood, Nucl. Phys. B 584 (2000) 359, hep-th/0004133.
J. Vinet, J.M. Cline, Phys. Rev. D 70 (2004) 083514, hep-th/0406141.
I. Navarro, JCAP 0309 (2003) 004, hep-th/0302129.
M. Peloso, L. Sorbo, G. Tasinato, Phys. Rev. D 73 (2006) 104025, hep-th/0603026.
Y. Aghababaie, C.P. Burgess, S.L. Parameswaran, F. Quevedo, Nucl. Phys. B 680 (2004) 389, hep-th/0304256;
Y. Aghababaie, et al., JHEP 0309 (2003) 037, hep-th/0308064;
C.P. Burgess, Ann. Phys. 313 (2004) 283, hep-th/0402200;
C.P. Burgess, arXiv: 0708.0911 [hep-ph];
G.W. Gibbons, R. Guven, C.N. Pope, Phys. Lett. B 595 (2004) 498, hep-th/0307238.
F. Leblond, R.C. Myers, D.J. Winters, JHEP 0107 (2001) 031, hep-th/0106140.
J.M. Cline, J. Descheneau, M. Giovannini, J. Vinet, JHEP 0306 (2003) 048, hep-th/0304147.
A. Kehagias, Phys. Lett. B 600 (2004) 133, hep-th/0406025.
V.P. Nair, S. Randjbar-Daemi, JHEP 0503 (2005) 049, hep-th/0408063;
S.L. Parameswaran, S. Randjbar-Daemi, A. Salvio, arXiv: 0706.1893 [hep-th].
S. Randjbar-Daemi, V.A. Rubakov, JHEP 0410 (2004) 054, hep-th/0407176.
H.M. Lee, A. Papazoglou, Nucl. Phys. B 705 (2005) 152, hep-th/0407208.
T. Kobayashi, Y.i. Takamizu, arXiv: 0707.0894 [hep-th].
E. Papantonopoulos, A. Papazoglou, V. Zamarias, arXiv: 0707.1396 [hep-th].
I. Navarro, Class. Quantum Grav. 20 (2003) 3603, hep-th/0305014.
S. Weinberg, Rev. Mod. Phys. 61 (1989) 1, astro-ph/0005265;
S.M. Carroll, Living Rev. Rel. 4 (2001) 1, astro-ph/0004075;
P.J.E. Peebles, B. Ratra, Rev. Mod. Phys. 75 (2003) 559, astro-ph/0207347;
T. Padmanabhan, Phys. Rep. 380 (2003) 235, hep-th/0212290.
V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 125 (1983) 139.
N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, Phys. Lett. B 480 (2000) 193, hep-th/0001197.
S. Kachru, M.B. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021, hep-th/0001206.
136
[21] 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.
[22] A. Kehagias, K. Tamvakis, Mod. Phys. Lett. A 17 (2002) 1767, hep-th/0011006.
[23] J.W. Chen, M.A. Luty, E. Ponton, JHEP 0009 (2000) 012, hep-th/0003067.
[24] M. Gell-Mann, B. Zwiebach, Phys. Lett. B 147 (1984) 111;
M. Gell-Mann, B. Zwiebach, Nucl. Phys. B 260 (1985) 569.
[25] H. Nicolai, C. Wetterich, Phys. Lett. B 150 (1985) 347.
[26] A. Kehagias, J.G. Russo, JHEP 0007 (2000) 027, hep-th/0003281;
A. Kehagias, hep-th/9911134;
A. Kehagias, Phys. Lett. B 469 (1999) 123, hep-th/9906204.
[27] A.G. Cohen, D.B. Kaplan, Phys. Lett. B 470 (1999) 52, hep-th/9910132.
[28] B.R. Greene, A.D. Shapere, C. Vafa, S.T. Yau, Nucl. Phys. B 337 (1990) 1.
[29] G.W. Gibbons, M.B. Green, M.J. Perry, Phys. Lett. B 370 (1996) 37, hep-th/9511080.
[30] M. Grana, J. Polchinski, Phys. Rev. D 65 (2002) 126005, hep-th/0106014.
[31] V.P. Nair, S. Randjbar-Daemi, JHEP 0503 (2005) 049, hep-th/0408063.
[32] B. de Wit, A. Van Proeyen, Nucl. Phys. B 245 (1984) 89;
B. de Wit, P.G. Lauwers, A. Van Proeyen, Nucl. Phys. B 255 (1985) 569;
E. Cremmer, C. Kounnas, A. Van Proeyen, J.P. Derendinger, S. Ferrara, B. de Wit, L. Girardello, Nucl. Phys. B 250
(1985) 385;
E. Cremmer, A. Van Proeyen, Class. Quantum Grav. 2 (1985) 445;
S. Cecotti, S. Ferrara, L. Girardello, Int. J. Mod. Phys. A 4 (1989) 2475.
[33] A. Ceresole, R. DAuria, S. Ferrara, A. Van Proeyen, Nucl. Phys. B 444 (1995) 92, hep-th/9502072.
[34] S. Ferrara, L. Girardello, C. Kounnas, M. Porrati, Phys. Lett. B 192 (1987) 368;
J.H. Schwarz, A. Sen, Nucl. Phys. B 411 (1994) 35, hep-th/9304154;
A. Sen, Int. J. Mod. Phys. A 9 (1994) 3707, hep-th/9402002.
[35] A. Salam, E. Sezgin, Phys. Lett. B 154 (1985) 37.
[36] L.Y. Hung, JHEP 0703 (2007) 071, hep-th/0612070.
[37] L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fre, T. Magri, J. Geom. Phys. 23 (1997) 111,
hep-th/9605032;
P. Fre, M. Trigiante, A. Van Proeyen, Class. Quantum Grav. 19 (2002) 4167, hep-th/0205119.
[38] C. Angelantonj, R. DAuria, S. Ferrara, M. Trigiante, Phys. Lett. B 583 (2004) 331, hep-th/0312019.
[39] M.J. Duff, J.T. Liu, J. Rahmfeld, Nucl. Phys. B 459 (1996) 125, hep-th/9508094.
[40] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77, hep-th/9401139.
[41] V. Kaplunovsky, J. Louis, S. Theisen, Phys. Lett. B 357 (1995) 71, hep-th/9506110.
[42] B. de Wit, V. Kaplunovsky, J. Louis, D. Lust, Nucl. Phys. B 451 (1995) 53, hep-th/9504006.
[43] G. Lopes Cardoso, G. Curio, D. Lust, Nucl. Phys. B 491 (1997) 147, hep-th/9608154.
[44] H. Klingen, Introductory Lectures on Siegel Modular Forms, Cambridge Univ. Press, 1990.
[45] E. Freitag, Siegelsche Modulformen, Grundlehren der Mathematischen Wissenschaften, vol. 254, Springer-Verlag,
1983.
[46] M.P. Tuite, math/9910136.
[47] G. Mason, M.P. Tuite, Commun. Math. Phys. 270 (2007) 587, math/0603088.
[48] J. Igusa, Am. J. Math. 84 (1962) 175200;
J. Igusa, Am. J. Math. 86 (1964) 392412.
[49] S.J. Gates, H. Nishino, E. Sezgin, Class. Quantum Grav. 3 (1986) 21;
E. Bergshoeff, I.G. Koh, E. Sezgin, Phys. Rev. D 32 (1985) 1353.

Thermal/quantum effects and induced superstring

cosmologies
Tristan Catelin-Jullien a , Costas Kounnas a , Herv Partouche b, ,
Nicolaos Toumbas c
a Laboratoire de Physique Thorique, Ecole Normale Suprieure, 1 24 rue Lhomond, F-75231 Paris cedex 05, France
b Centre de Physique Thorique, Ecole Polytechnique, 2 F-91128 Palaiseau, France
c Department of Physics, University of Cyprus, Nicosia 1678, Cyprus

Abstract
We consider classical superstring theories on flat four-dimensional spacetimes, and where N = 4 or
N = 2 supersymmetry is spontaneously broken. We obtain the thermal and quantum corrections at the
string one-loop level and show that the back-reaction on the spacetime metric induces a cosmological
evolution. We concentrate on heterotic string models obtained by compactification on a T 6 torus and
on T 6 /Z2 orbifolds. The temperature T and the supersymmetry breaking scale M are generated via the
ScherkSchwarz mechanism on the Euclidean time cycle and on an internal spatial cycle respectively. The
effective field theory corresponds to a no-scale supergravity, where the corresponding no-scale modulus
controls the SUSY-breaking scale. The classical flatness of this modulus is lifted by an effective thermal
potential, given by the free energy. The gravitational field equations admit solutions where M, T and the
inverse scale factor 1/a of the universe remain proportional. In particular the ratio M/T is fixed during
time evolution. The induced cosmology is governed by a FriedmannHubble equation involving an effective radiation term 1/a 4 and an effective curvature term 1/a 2 , whose coefficients are functions of the
complex structure ratio M/T .
Research partially supported by the EU (under the contracts MRTN-CT-2004-005104, MRTN-CT-2004-512194,

MRTN-CT-2004-503369, MEXT-CT-2003-509661), INTAS grant 03-51-6346, CNRS PICS 2530, 3059 and 3747, ANR
(CNRS-USAR) contract 05-BLAN-0079-01 and INTERREG IIIA Crete/Cyprus.
E-mail addresses: catelin@lpt.ens.fr (T. Catelin-Jullien), costas.kounnas@lpt.ens.fr (C. Kounnas),
herve.partouche@cpht.polytechnique.fr (H. Partouche), nick@ucy.ac.cy (N. Toumbas).
1 Unit mixte du CNRS et de lEcole Normale Suprieure associe lUniversit Pierre et Marie Curie (Paris 6), UMR
8549.
2 Unit mixte du CNRS et de lEcole Polytechnique, UMR 7644.
doi:10.1016/j.nuclphysb.2007.12.030
138
T. Catelin-Jullien et al. / Nuclear Physics B 797 (2008) 137178

PACS: 11.25.Mj
Keywords: Quantum cosmology; Finite temperature; Heterotic string
1. Introduction
It is important to develop a string theoretic framework for studying cosmology. The ultimate goal of this task is to determine whether string theory can describe basic features of our
Universe. Despite considerable effort towards this direction over the last few years (see for example [110]), still very little is known about the dynamics of string theory in time-dependent,
cosmological settings. The purpose of this work is to provide a new class of non-trivial string
theory cosmological solutions, where some of the difficult issues can be explored and analyzed
concretely.
At the classical string level, it seems difficult to obtain exact cosmological solutions [8]. Indeed, after extensive studies in the framework of superstring compactifications (with or without
fluxes), the obtained results appear to be unsuitable for cosmology. In most cases, the classical
ground states correspond to static anti-de Sitter or flat backgrounds but not to cosmological ones.
The same situation appears to be true in the effective supergravity theories. Naively, the results
obtained in this direction lead to the conclusion that cosmological ground states are unlikely to
be found in superstring theory.
From our viewpoint this conclusion cannot be correct for two reasons:
The first follows from the fact that already exact (to all orders in ) cosmological solutions
exist, which are described by a two-dimensional worldsheet conformal field theory based on
SL(2,R)
a gauged WessZuminoWitten model at negative level |k|: U (1)|k| M, [24].
The second is that quantum and thermal corrections are neglected in the classical string/
supergravity regime.
The first class of stringy cosmological models was studied recently in [5], where it was shown
how to define a normalizable wave-function for this class of backgrounds, realizing the Hartle
Hawking no-boundary proposal [11] in string theory. Explicit calculable examples were given
for small values of the level |k|. As it was shown in [5], these models are intrinsically thermal
with a temperature below but still close to the Hagedorn temperature. The disadvantage of small
level |k|, however, is the absence of a semi-classical limit with |k| arbitrarily large, which prevents us from obtaining a clean geometrical picture and studying issues such as back-reaction
and particle production in a straightforward way.
Another direction consists of studying the quantum and thermal cosmological solutions,
which are generated dynamically at the quantum level of string theory [6,12]. Although this
study looks to be hopeless and out of any systematic control, it turns out that in certain cases the
quantum and thermal corrections are under control thanks to the special structure of the underlying effective supergravity theory in its spontaneously broken supersymmetric phase. An effective
field theory study has already been initiated in [6,12]. (See also [13,14].)
In order to see how cosmological solutions arise naturally in this context, consider the case of
a supersymmetric flat string background. At finite temperature the thermal fluctuations produce a
non-zero energy density that is calculable perturbatively at the full string level. The back-reaction
139
on the spacetime metric and on certain of the moduli fields gives rise to a specific cosmological
evolution. For temperatures below the Hagedorn temperature, the evolution of the universe is
known to be radiation dominated. (See for instance [15,16] for some earlier work in this case
and [16] for a review on string gas cosmology.)
More interesting cases are those where spacetime supersymmetry is spontaneously broken at
the string level either by geometrical [17] or non-geometrical fluxes. In the case where the geometrical fluxes are generated via freely acting orbifolds [1823], the stringy quantum corrections
are under control in a very similar way as the thermal ones. The back-reaction of the quantum
and thermal corrections on the spacetime metric and the moduli fields results in deferent kinds
of cosmologies depending on the initial amount of supersymmetry (N = 4, N = 2, N = 1).
In this work we restrict attention to four-dimensional backgrounds with initial N = 4 or N = 2
spacetime supersymmetry, obtained by toroidal compactification of the heterotic superstring
on T 6 and T 6 /Z2 -orbifolds. The spontaneous breaking of supersymmetry is implemented via
freely acting orbifolds (as in [1823]). The quantum and thermal corrections are determined
simultaneously by considering the Euclidean version of the model where all coordinates are
compactified: ST1 T 3 (for the four-dimensional spacetime part) M 6 (for the internal manifold). Apart from being interesting in their own right, these examples may give us useful hinds
on how to handle the phenomenologically more relevant N = 1 cases. The N = 1 cases will be
studied elsewhere [31].
The thermal corrections are implemented by introducing a coupling of the spacetime fermion
number QF to the string momentum and winding numbers associated to the Euclidean time
cycle ST1 . The breaking of supersymmetry is generated by a similar coupling of an internal Rsymmetry charge QR to the momentum and winding numbers associated to an internal spatial
1 , e.g., the X coordinate cycle.
cycle SM
5
We stress here, that the thermal and supersymmetry breaking couplings correspond to string
theoretic generalizations of ScherkSchwarz compactifications. Two very special mass scales
appear both associated with the breaking of supersymmetry. These are the temperature scale
T 1/(2R0 ) and the supersymmetry breaking scale M 1/(2R5 ), with R0 and R5 the radii
1 , respectively. The initially
of the Euclidean time cycle, ST1 , and of the internal spatial cycle, SM
degenerate mass levels of bosons and fermions split by an amount proportional to T or M,
according to the charges QF and QR . This mass splitting is the signal of supersymmetry breaking
and gives rise to a non-trivial free energy density, which incorporates simultaneously the thermal
corrections and quantum corrections due to the supersymmetry breaking boundary conditions
1 .
along the spatial cycle SM
At weak coupling, the free energy density can be obtained from the one-loop Euclidean string
partition function [2022]. The perturbative string amplitudes are free of the usual ultraviolet ambiguities that plague a field theoretic approach towards quantum gravity and cosmology. For large
enough R0 , R5 , the Euclidean system is also free of tachyonsthe presence of tachyons would
correspond to infrared instabilities, driving the system towards a phase transition [2123]. Therefore, the corresponding energy density and pressure can be determined unambiguously, and we
can use them as sources in Einsteins equations to obtain non-trivial cosmological solutions. This
perturbative approach breaks down near the initial space-like singularity. We speculate whether
this breakdown of perturbation theory can be associated with an early universe phase transition.
The paper is organized as follows. Section 2 is mainly a review, where we also fix most of
our notations and conventions. We first consider the four-dimensional heterotic string models at
finite temperature. We obtain the one-loop thermal partition function at the full string level, and
then we discuss the effective field theory limit at large radius R0 . We also review the analogous
140
computation of the one-loop string partition function at zero temperature and in the case where
1 , [1823]. In
SUSY-breaking boundary conditions are placed along the internal spatial cycle SM
the large radius limit, the Einstein frame effective potential is proportional to the fourth power
of the gravitino mass scale, and it can be positive or negative depending on the choice of the
SUSY-breaking operator QR .
In Section 3, we consider the case where thermal and quantum corrections due to the supersymmetry breaking are present simultaneously. For the simplest choice QR = QF , the corresponding one-loop string partition function is invariant under the T M exchange, manifesting
the underlying temperature/gravitino mass scale duality of the models. This duality is broken by
the other allowable choices for the SUSY-breaking operator QR , which we classify for both the
N = 4 and the N = 2 orbifold cases.
In the large radii R0 , R5 limit, the pressure consists of two pieces: the purely thermal part
which scales as nT T 4 , with the coefficient nT being the number of all massless boson/fermion
pairs in the initially supersymmetric theory, and another potential-like piece which scales as
nV M 4 and with the coefficient nV being positive or negative depending on the choice of the
operator QR . In both pieces, the rest of the dependence on the scales T and M can be expressed
neatly in terms of non-holomorphic Eisenstein series of order 5/2 whose variable is the complex
structure-like ratio M/T . In addition, we incorporate the effects of small, continuous Wilson line
deformations in our computation. Wilson lines along any of the internal spatial cycles, other than
1 , introduce new mass scales, and pieces proportional to T 2 and M 2 arise in the effective
SM
thermodynamic quantities.
In Section 4 we present our ansatz for the induced cosmological solutions. These are homogeneous and isotropic cosmologies for which the SUSY-breaking scales T and M as well as the
inverse of the scale factor 1/a evolve the same way in time, and so the ratio of any two of these
quantities is constant. The form of this ansatz is dictated by the scaling properties of the effective energy density and pressure. The compatibility of the gravitational field equations with the
equation of motion of the scalar modulus controlling the size of the gravitino mass scale fixes
the ratio M/T . By solving the compatibility equations numerically, we find that in the absence
1 , non-trivial four-dimensional solutions exist when n is negative and
of Wilson lines along SM
V
the ratio |nV |/nT is small enough. These conditions are satisfied by various models we describe
1 , the value of the ratio M/T for
explicitly in the paper. When we include Wilson lines along SM
some of the solutions can be large or small, and so we can have models with a hierarchy for the
scales M and T .
Having solved the compatibility equations, the time-dependence of the system is governed
solely by the familiar FriedmannHubble equation. There is a radiation term, cr /a 4 , whose co 2 , can be generated
efficient cr is positive in our examples. An effective curvature term, k/a
by turning on Wilson line deformations. The sign of k can be positive or negative depending on
the model. When we turn on the kinetic terms of some of the extra flat moduli, we generate an
additional term that scales as cm /a 6 (with cm positive).
In Section 5, we solve the FriedmannHubble equation for the various possible cases, and we
elaborate on the properties of the cosmological solutions:
When cr > 0, we have standard hot big bang cosmologies with an intermediate radiation
dominated era. The late time behavior is governed by the spatial curvature of the models.
We also consider a priori possible exotic models characterized by cr < 0. A big bang occurs when cm > 0. The cosmological evolution always ends with a big crunch when k 0.
The case k < 0 however is more interesting. It involves either a first or second order phase
141
transition between the big bang cosmology and a linearly expanding universe. The first case
corresponds to a tunneling effect involving a gravitational instanton, while the transition is
smooth in the second case. If the first order transition does not occur, the universe ends in a
big crunch.
We finish with our conclusions and directions for future research.
2. Thermal and quantum corrections in heterotic backgrounds
Our starting point is the class of four-dimensional string backgrounds obtained by toroidal
compactification of the heterotic string on T 6 and T 6 /Z2 orbifolds. Initially the amount of space
time supersymmetry is N4 = 4 for the case of compactification on the T 6 torus and N4 = 2 for the
orbifold compactifications, and the four-dimensional spacetime metric is flat. Spacetime supersymmetry is then spontaneously broken by introducing ScherkSchwarz boundary conditions
on an internal spatial cycle and/or by thermal corrections. Due to the supersymmetry breaking,
the one-loop string partition function is non-vanishing, giving rise to an effective potential. Our
aim is to determine the back-reaction to the initially flat metric and moduli fields.
At the one-loop level, the four-dimensional string frame effective action is given by

1
R + 2 + VString ,
S = d 4 x det g e2
(2.1)
2
where is the 4d dilaton field and the ellipses stand for the kinetic terms of other moduli fields
(to be specified later). At zero temperature, the effective potential VString can be obtained from
the one-loop Euclidean string partition function as follows:
Z
= VString ,
V4
(2.2)
with V4 the 4d Euclidean volume. The absence of a dilaton factor multiplying the potential term
in the action is due to the fact that this arises at the one-loop level.
At finite temperature, the one-loop Euclidean partition function determines the free energy
density and pressure to this order
Z
= FString = PString .
V4
(2.3)
The subscript indicates that these densities are defined with respect to the string frame metric. The
relevant Euclidean amplitude incorporates simultaneously the thermal corrections and quantum
corrections which arise from the spontaneous breaking of supersymmetry and which are present
even at zero temperature.
In order to determine the back-reaction of the (thermal and/or) quantum corrections, it is
convenient to work in the Einstein frame where there is no mixing between the metric and the
dilaton kinetic terms. We define as usual the complex field S,
S = e2 + i,
(2.4)
where is the axion field. Then after the Einstein rescaling of the metric, the one-loop effective
action becomes:

1
1
S = d 4 x det g R g KI J I J 2 VString (I , I ) ,
(2.5)
2
s
142
where Ki j is the metric on the scalar field manifold {I }, which is parameterized by various
compactification moduli including the field S. This manifold includes also the main moduli fields
TI , UI , I = 1, 2, 3, which are the volume and complex structure moduli of the three internal 2cycles respectively. We notice that in the Einstein frame the effective potential, VString , is rescaled
by a factor 1/s 2 , where s = Re(S) = e2 . Taking this rescaling into account, we have
1
VString .
(2.6)
s2
This relation will be crucial for our work later on. (We will always work in gravitational mass
1
units, with MG = 8G
= 2.4 1018 GeV.)
N
Keeping only the main moduli fields {S, TI , UI }, their kinetic terms are determined in terms
of the Khler potential K [24,25]:

K = log(S + S)
(2.7)
log(TI + TI )
log(UI + U I )
VEins =
with KI J = I J K. The classical superpotential depends on the way supersymmetry is broken.

Generically string backgrounds with spontaneously broken supersymmetry are flat at the classical level due to the no-scale structure of the effective supergravity theory [25]. Once the thermal
and/or quantum corrections are taken into account, we obtain in some cases interesting cosmological solutions.
2.1. Heterotic supersymmetric backgrounds at finite temperature
In order to fix our notations and conventions, we first consider the case of an exact supersymmetric background at finite temperature [2123]. For definiteness we choose the heterotic string
with maximal spacetime supersymmetry (N4 = 4). All nine spatial directions as well as the
Euclidean time are compactified on a ten-dimensional torus. At zero temperature, the Euclidean
string partition function is zero due to spacetime supersymmetry. At finite temperature however
the result is a well-defined finite quantity. Indeed, at genus one the string partition function is
given by:

4

(10,26) ab
d d 1
a
a+b+ab
Z=
(2.8)
()
,
b ( )12 (
4 Im 2
)24
F
a,b
where (10,26) b is a shifted Narain lattice (which we specify more precisely below). The nonvanishing of the partition function is due to the non-trivial coupling of the lattice to the spin
structures (a, b). Here, the argument a is zero for spacetime bosons and one for spacetime
fermions. The spin/statistics connection and modular invariance require that the unshifted (1,1)
sub-lattice of the Euclidean time cycle
(1,1)
2 |m+n |2
Im
R0 (Im ) 2 eR0
(2.9)
m,n
be replaced as follows:
(1,1)

m,n
2 |m+n |2
Im
R0 (Im ) 2 eR0
ei(ma+nb+mn) .
(2.10)
143
Redefining
m 2m + g,
n 2n + h,
where g, h are integers defined modulo 2, and introducing the notation (1,1)
lattice,

2
1
h
2 |2m+g+(2n+h) |
Im
(1,1)
R0 (Im ) 2 eR0
,
=
g
m,n
h
g
(2.11)
for a shifted
(2.12)
the thermal partition function takes the form:

Z=
F
d d 1
4 Im 2
ga+hb+hg
()
a+b+ab
()
(a,b),(h,g)
h
4
(9,25) (1,1) g
a
.
b
( )12 (
)24
Defining a = a h and b = b g and using the Jacobi identity

a + h 4
1+h 4
1
a b
b+
()a+

=
,
b+g
1+g
2
(2.13)
(2.14)
(a,
b)
we obtain
Z=
F

d d
1+h 4
h
(9,25)
()g+h
(1,1)
.
1+g
4 Im
g ( )12 (
)24
(2.15)
(h,g)
The temperature in string frame is given by TString = 1/(2R0 ).

Since our aim is the study of induced cosmological solutions in 3 + 1 dimensions, we consider
the case for which the radii of three spatial directions are very large: Rx = Ry = Rz R 1. In
this case the three-dimensional spatial volume factorizes
3
3
V3
(Im ) 2 .
(3,3)
= R 3 (Im ) 2 =
3
(2)
(2.16)
Using the expression for the (1,1) [hg ] shifted lattice we obtain:
Z = (2R0 )V3 FString = V4 PString

|2
d d
(6,22)
1+h 4
V4
g+h R02 |2m+g+(2n+h)
Im
()
e
,
=
1 + g ( )12 (
(2)4
4 Im 3
)24
(n,m),(h,g)
F
(2.17)
where V4 = (2R0 )V3 is the four-dimensional spacetime volume, FString the free energy density and PString the pressure in string frame.
Before we proceed further, we make some comments:
The sector (h, g) = (0, 0) gives zero contribution. This is due to the fact that we started with
a supersymmetric background.
In the odd winding sector,
h = 1, the partition function diverges when R0 is between the
Hagedorn radius RH = ( 2 + 1)/2 and its dual 1/RH : R1H < R0 < RH . The divergence is
due to a winding state that is tachyonic when R0 takes values in this range, and it signals a
phase transition around the Hagedorn temperature [2123]. In this paper we study the regime
144
R0 > RH , where there is no tachyon and the winding sector is exponentially suppressed. The
high temperature regime and the cosmological consequences of the phase transition will be
examined in future work [32].
When R0 1, the contributions of the oscillator states are also exponentially suppressed,
provided that the moduli parameterizing the internal (6,22) lattice are of order unity.
2.2. The effective field theory in the large R0 limit
As we already mentioned, the h = 1 sector of the theory gives exponentially suppressed con2
tributions of order O(eR0 ). Also, the (h, g) = (0, 0) sector vanishes due to supersymmetry.
Thus for large R0 , only the sector (h, g) = (0, 1) contributes significantly. Using the identity:

0
0
1
1
+ (1,1)
+ (1,1)
+ (1,1)
(1,1) (R0 ) = (1,1)
(2.18)
0
1
0
1
and neglecting the h = 1 sectors, we may replace

1
0
0
(1,1) (R0 ) (1,1)
= (1,1) (R0 ) (1,1) (2R0 )
(1,1)
2
1
0
(2.19)
in the integral expression for Z. For each lattice term we decompose the contribution in modular
orbits: (m, n) = (0, 0) and (m, n)
= (0, 0). For (m, n)
= (0, 0), the integration over the fundamental domain is equivalent with the integration over the whole strip but with n = 0. The (0, 0)
contribution is integrated over the fundamental domain. Now the (0, 0) contribution of (1,1) (R0 )
cancels the one of 12 (1,1) (R0 ), and we are left with the integration over the whole strip:

d d R 2 (2m+1)2 1 4
(6,22)
V4
0
Im
.
Z=
(2.20)
0 ( )12 (
(2)4
4 Im 3 m
)24
||
The integral over 1 imposes the leftright level matching condition. The left-moving part
contains the ratio
4

10
= 24 + O e2 ,
(2.21)
12
which implies that the lowest contribution is at the massless level. Thus after the integration over
1 (2 t), the partition function takes the form
Z=
V4
(2)4

0

dt R 2 (2m+1)2 4
t2
0
t
e
D
+
D()e
2
,
0
2t 3 m
(2.22)
where D() denotes the multiplicity of the mass level and 24 D0 is the multiplicity of the
massless level. Changing the integration variable by setting t = R02 (2m + 1)2 x, we have:

1
dx 1 4
V4
2
2 2 2
e x 2 D0 +
D()ex (2m+1) R0 .
Z= 2
4
4
3
(2R0 ) m (2m + 1)
2x
0
(2.23)
Now the second term in the parenthesis is exponentially suppressed when the masses are of
order (or close) to the string oscillator mass scale. This will be the case when all of the internal
145
radii and the Wilson-line moduli of the (6,22) lattice are of order unity. For this specific case,
the partition function simplifies to
Z = 2 3 D0

V4
1
2 3 D0 2
1 n 2
V4
4
=
=
,
V4 TString
48 (2R0 )4 3 16
2 (2R0 )4 m (2m + 1)4
(2.24)
where n = 23 D0 is the number of the massless boson/fermion pairs in the theory. The free
energy density and pressure in string frame are given by
PString = FString =
2 4
1 n TString
.
3
16
(2.25)
In the Einstein frame, energy densities are rescaled by a factor 1/s 2 as in Eq. (2.6). Thus the
pressure and free energy density in this frame are given by
PEins = FEins =
2 4
1 n TString 1 n 2 T 4
=
,
3
3 16
16s 2
(2.26)
where T = TString / s is the proper temperature in the Einstein frame. This result is expected
from the effective field theory point of view. When only massless states are thermally excited,
the field theory expression for the pressure is given by

2 4
1
7
T
P=
(2.27)
nB + nF
,
3
8
30
where nB and nF are the numbers of massless bosonic and fermionic degrees of freedom respectively. When nB = nF = n , as in a supersymmetric theory, we recover Eq. (2.26).
2.3. Spontaneous breaking of supersymmetry at zero temperature
In this case we consider the same class of heterotic models, but now the breaking of supersymmetry arises due to the coupling of the spacetime fermion number to the momentum and
winding quantum numbers of an internal spatial cycle [1823]. Since the temperature is taken to
be zero, the spin structures (a, b) do not couple to the quantum numbers of the Euclidean time
cycle which will be taken to be very large. We also consider the case where three additional spatial directions are large. Following similar steps to the purely thermal case, the partition function
is given by

|2
V5
d d
(5,21)
1+h 4
g+h R52 |2m+g+(2n+h)
Im
Z=
() e
,
7
1
+
g
(2)5
(
)12 (
)24
4 Im 2 (n,m),(h,g)
F
(2.28)
where now V5 = V4 (2R5 ) is a five-dimensional volume and the (5,21) lattice parameterizes the
2
internal space. Here also, the h = 1 sectors give exponentially suppressed contributions O(eR5 ),
and the (h, g) = (0, 0) sector vanishes due to supersymmetry. The rest of the steps can be repeated as in the derivation above to find
V5
Z=
(2)5

0
dt
2t
7
2
2 (2m+1)2
t
eR5
24 D 0 +

2
D()et ,
(2.29)
146
which after the change of variables t = R52 (2m + 1)2 x gives

Z=
V5
5
2
(2R5 )5
1
|2m + 1|5

0
dx
2x
7
2

1
2
2 2 2
e x 24 D 0 +
D()ex (2m+1) R5 .
(2.30)
For of order unity, this simplifies to

(5) ( 52 ) V4

n
Z = 2 1 25
5
(2R5 )4
2
(2.31)
with n = 23 D0 .
This result was expected from the effective field theory point of view. Indeed in a theory with
spontaneously broken N = 4 supersymmetry, the one-loop effective potential receives a nonzero contribution proportional to the mass super-trace Str M4 , which in turn is proportional to
the fourth power of the gravitino mass. The super-traces Str Mn vanish for n < N = 4. In the
example of supersymmetry breaking we examined above, the masses of the states are shifted
according to their spin. For initially massless states, the mass after supersymmetry breaking
becomes:
2
MQ
Q2F
R52
(2.32)
This shows that the string frame gravitino mass is of order MString 1/R5 and thus Str M4
c/R54 . Including the contributions from all KaluzaKlein states, one obtains the result given in
formula (2.31). We obtain for the string frame effective potential:
VString =
(5) ( 52 )

1
Z
= 2 1 25
n
.
5
4
V4
(2R
5)
2
In the Einstein frame, we have VEins =
1
V
see
s 2 String
(2.33)
Eq. (2.6)so that
(5) ( 52 )

1
1
VEins = 2 1 25
n 2
= CV
= CV M 4 ,
5
4
2
s
(2R
)
(st
u
)
5
1 1
2
(2.34)
where t1 = Re(T1 ), u1 = Re(U1 ), and M = 1/(st1 u1 )1/2 is the gravitino mass scale in the Einstein frame.
We stress here that the one-loop effective potential depends only on the gravitino mass scale,
which in turn depends only on the product of the s, t1 and u1 moduli. This suggests to freeze all
moduli and keep only the diagonal combination
3 log z = log s + log t1 + log u1 .
(2.35)
The Khler potential of the diagonal modulus Z (with z = Re(Z)), takes the well-known
SU(1, 1) structure [25]
K = 3 log(Z + Z).
(2.36)
This gives rise to the kinetic term and gravitino mass scale,
g 3
Z Z
,
2
(Z + Z)
M 2 = 8eK =
8
.
3
(Z + Z)
(2.37)
147
Freezing Im Z and defining the field by

e2 = M 2 =
8
,
3
(Z + Z)
(2.38)
one finds the kinetic term

g 3
2
Z Z

=
g
.
2
3
(Z + Z)
(2.39)
The choice 2 = 3/2 normalizes canonically the kinetic term of the modulus . The potential
for this particular model is:

3
4
4
VEins () = CV M = CV e
(2.40)
, =
.
2
Observe that in this simple model the sign of the potential is negative. As we now explain, we
can construct models with a positive potential, but with the rest of the dependence on the modulus being the same. All we have to do is to couple the momentum and winding numbers of the
ScherkSchwarz cycle not only to the spacetime fermion number but also to another internal
charge. For example consider the E8 E8 heterotic string on T 6 and instead of coupling just
to QF , we couple to QF + QE8 + QE8 , where QE8 denotes the charge of an E8 representation
decomposed in terms of SO(16) ones, and similarly for QE8 . These charges take half integer values for the spinorial representations and integer values for the others. The initial SUSY-breaking
co-cycle gets modified as follows

()ag+bh+hg () a+ +

g+b++
h+hg
(2.41)
where as before the argument a is one for spacetime fermions and zero for spacetime bosons,
and ( , ) = (1, 1) for the spinorial representations of SO(16) SO(16) and (0, 0) for the
adjoint representations. This operation breaks explicitly the E8 E8 gauge group to SO(16)
SO(16) . Proceeding in similar way as in the previous example, one finds:
(5) ( 52 ) V4

n
,
Z = 2 1 25
5
(2R5 )4
2
where

n = 23 [2]X2,3 + [6]T 6 + [120 128]E8 + [120 128]E8
= 23 8 = 64.
(2.42)
(2.43)
In the previous example only positive signs appear in the above formula since there is no coupling
of the ScherkSchwarz lattice quantum numbers to the E8 E8 charges, giving the value n =
23 504. The reversing of sign for some representations indicates that it is for the bosons that
the masses are shifted and not for the fermions in the corresponding multiplet.
We note that in the N = 4 case, we cannot change the left-multiplicity since all of the leftmoving R-charges are equivalent as required by symmetry. This however is not true for the
N = 2 and N = 1 cases. Consider for instance the class of N = 2 supersymmetric backgrounds
obtained by compactifying the heterotic string on a T 4 /Z2 orbifold (e.g., the Z2 -orbifold limit
of the K3 CY-compactification). In this class of models (see for instance [26]) four internal
supercoordinates are twisted and the corresponding four internal R-charges are half-shifted. The
148
Euclidean partition function is given by

Z=
d d 1
4 Im 4
a+b+ab
()
a 2
a+H 2
b+G
b
( )4

(1,1) (R0 )(3,3) (space)
0
H
Z(2,2+n0 )
Z(4,4+nt )
.
0
G
( )2 (
)2
(a,b),(H,G)
(2.44)
Here Z(2,2+n0 ) is the contribution of two internal coordinates3 (X5 , X6 ) and n0 -right-moving
world-sheet bosons i . Before supersymmetry breaking, the corresponding (2, 2 + n0 )-lattice
is unshifted. Z(4,4+nt ) stands for the contribution of four internal coordinates (X7 , X8 , X9 , X10 )
all of which are Z2 -twisted by (H, G), and nt -right-moving world-sheet bosons I which can
be Z2 -twisted breaking part of the initial gauge group. The -function terms come from the
contribution of the left-moving world-sheet fermions. Four of them are Z2 -twisted by (H, G).
The contribution associated to the spacetime bosons is when a = 0, while the one associated to
the spacetime fermions is when a = 1.
From the above supersymmetric N = 2 partition function, the thermal partition function is
obtained in a way similar to the N = 4 example, by the following replacement of the Euclidean
time sub-lattice:

h1
(1,1) (R0 ) (1,1)
(2.45)
(R0 )()g1 a+h1 b+h1 g1 .
g1
In the case of ScherkSchwarz spontaneous supersymmetry breaking, the partition function can
be obtained by a similar replacement of the internal X5 coordinate lattice, either by utilizing the
same operator QF

h2
(1,1) (R5 ) (1,1)
(2.46)
(R5 )()g2 a+h2 b+h2 g2
g2
or by utilizing an R-symmetry operator associated to one of the twisted complex planes

h2
(1,1) (R5 ) (1,1)
(R5 )()g2 (a+H )+h2 (b+G)+h2 g2 .
g2
(2.47)
These are in fact the only two possibilities involving left-moving R-charges since all others are
equivalent choices. However, many
other choices exist by utilizing parity-like operators involving
the right-moving gauge charges i , as in the explicit example of SO(16) SO(16) spinorial
representations we gave above:

h2
(1,1) (R5 ) (1,1)

(2.48)
(R5 )()g2 (a+H + i )+h2 (b+G+ i )+h2 g2 .
g2
In the next section we examine representative examples in the case where thermal and spontaneous SUSY-breaking operations are present simultaneously.
3. Thermal and spontaneous breaking of SUSY
The most interesting situation for cosmological applications is the case where spontaneous
supersymmetry breaking and thermal corrections are taken into account simultaneously.
3 In our notations, the spacetime coordinates are X
0,...,3 , while the internal ones are X5,...,10 .
149
3.1. Untwisted sector

The untwisted sector of the N = 2 case, (H, G) = (0, 0) in Eq. (2.44),4 has an N = 4 structure
and thus all choices for the left R-symmetry operators are equivalent. The quantum numbers of
the Euclidean time cycle and the internal X5 -cycle are coupled to the spin structures (a, b) in the
same way. After performing the Jacobi theta-function identity the partition function becomes:

d d
1 V5
Zuntwist =
()g1 +g2 +h1 +h2
7
2 (2)5
4 Im 2 (n1 ,m1 ),(h1 ,g1 ) (n2 ,m2 ),(h2 ,g2 )
F
|2m +g +(2n +h ) |
R02 1 1 Im 1 1
2 |2m2 +g2 +(2n2 +h2 ) |

Im
eR5

(5,21)
1 + h1 + h2 4
.
1 + g1 + g2 ( )12 (
)24
(3.1)
The factor of 1/2 is due to the Z2 orbifolding of the N = 4 theory.

Proceeding as in the simpler examples before and neglecting the h1 = 1 and h2 = 1 sectors
for large R0 , R5 , the non-zero contributions to the partition function occur when g1 + g2 = 1.
Assuming also that all other moduli are of order unity, the only non-exponentially suppressed
contributions come from the zero mass left- and right-levels. We obtain
Zuntwist = 23 D0
V5 (1 ()g1 +g2 )
2
(2)5 g ,g
1
dt
2t
7
2
2 (2m1 +g1 )
t
eR0
R52
(2m2 +g2 )2
t
(3.2)
m1 ,m2
which after the change of variables t = (R02 (2m1 + g1 )2 + R52 (2m2 + g2 )2 )x gives
Zuntwist =
4D0 ( 52 ) V5
1
5
5
2
(2) m ,m (R (2m1 + 1)2 + R 2 (2m2 )2 ) 52
2
1 2
0
5
+
4D0 ( 52 ) V5
1
.
5
5
2
(2) m ,m (R (2m1 )2 + R 2 (2m2 + 1)2 ) 52
2
1 2
0
5
(3.3)
This expression is symmetric under the R0 R5 exchange. This is suggestive of a temperature/gravitino mass, T /M, duality. This duality will be broken when the supersymmetry breaking
arises due to the coupling to a different QR charge than QF .
To obtain the effective four-dimensional pressure, we must factorize out the spacetime volume V4 . To this extent it is convenient to define the complex structure-like ratio
u=
R0 M
= ,
R5
T
(3.4)
4 For h = h = 0 (even windings), the sector (H, G) = (0, 1) gives zero net contribution due to the identity
1
2
1 ()a+b+ab ()ag1 ()ag2
a 2
a
a = 0.
a,b
2
b+1
b1
b
150
and re-write the partition function in the following way

Zuntwist =
4D0 ( 52 )
5
2

V4
u4
(2R0 )4 m ,m |(2m1 + 1)iu + 2m2 |5
1
4D0 ( 52 )
5
2
V4
(2R5 )4
m1 ,m2
1
.
|2m1 iu + (2m2 + 1)|5
(3.5)
Define the function

f (u)

m1 ,m2
u4
,
|(2m1 + 1)iu + 2m2 |5
which we can express in terms of Eisenstein functions of order 5/2:

3
1
iu
1
f (u) = u 2
E
E
(iu)
,
5/2
5/2
5
2
25
22
(3.6)
(3.7)
where
Ek (U ) =

(m,n)
=(0,0)
Im U
|m + nU |2
k
.
(3.8)
Then the pressure in the Einstein frame can be written as

Puntwist = CTunt T 4 f (u) + CVunt M 4
f (1/u)
,
u
(3.9)
where
CTunt = CVunt = nunt
( 52 )
5
(3.10)
Here nunt = 4D0 is the number of massless boson/fermion pairs in the untwisted sector. It is
smaller by a factor of 1/2 from the corresponding number in the N = 4 case due to the Z2 orbifolding. In this particular model the coefficients CTunt and CVunt are equal due to the underlying
gravitino mass/temperature duality. For fixed u the first term stands for the thermal contribution
to the pressure while the second term stands for minus the effective potential.
We note that the coefficient CTunt is fixed and positive as it is determined by the number of
all massless boson/fermion pairs in the untwisted sector of the initially supersymmetric theory:
nunt = 4D0 . In general, the coefficient CVunt will depend on the precise way supersymmetry is
broken. As we have demonstrated in the previous section, it can take both negative and positive
values, depending on how the SUSY-breaking operator couples to the right movers: CVunt n unt .
Thus in general the temperature/gravitino mass scale duality will be broken.
Let us discuss the large u limit, which can be obtained by taking R5 to be small (but still
parametrically larger as compared to the string scale), while taking R0 to be much larger. In
this limit we expect to find a four-dimensional system at finite temperature, for which only
massless bosonic degrees of freedom are thermally excited. All fermions attain a mass from the
ScherkSchwarz boundary conditions along the X5 cycle, and this mass is much bigger than the
temperature for large u. Therefore they can be integrated out giving a temperature-independent
151
contribution to the pressure of order 1/R54 . Setting u = 1/u, we have in the limit u 0 (u ):
f (u) = f (1/u)

m1
and
dx
((2m1 + 1)2 + 4x 2 ) 2
1
2
1 4
=
, (3.11)
3 m (2m1 + 1)4 3 24
1

f (1/u)
1
dx
1
+ 5 4
5
5
2
u
|2m
+
1|
2
u
2
1
m1
m2
=0 (m2 + x ) 2

1 1
= 2 1 25 (5) +
24u4
m4
m
=0 2
2

= 2 1 25 (5) +
1
4
.
90
12u4
(3.12)
Using these results, we find

Puntwist =
(5) ( 52 )

1 23 D0 2 T 4
(4D0 )M 4
+ 2 1 25
5
3 2
30
2
(3.13)
for the first two leading terms for small u = 1/u. We have used the relation ( 52 ) = 3 /4. The
first contribution arises from the thermally excited massless bosons. As compared to Eq. (2.27)
with nB = 23 D0 and nF = 0, it is off by a factor of 1/2 due to the Z2 -orbifolding. Similarly, the
second term is off by a factor of 1/2 as compared to Eq. (2.31) due to the orbifolding. For large u,
the potential term is dominant. For generic values of u both fermions and bosons contribute to the
thermal piece as in Eq. (3.9), with the contribution depending on the number of massless states
at zero temperature and before the breaking of supersymmetry. Finally for small u, the system is
essentially a five-dimensional purely thermal system.
3.2. Twisted sector, H = 1
The contributions of the twisted sectors in the large R0 , R5 limit depend on the number of
the massless twisted states before the supersymmetry breaking, and can be determined in similar
way as before. However, there is a class of models where the Z2 -orbifolding acts freely, without
any fixed points, and therefore there are no massless states in the twisted sectors. For this class
of models, the whole contribution to the one-loop partition function, in the large R0 , R5 limit, is
that of the massless untwisted sector states we have already determined. One example with this
property is when the Z2 -twists (H, G) are accompanied with a shift of the (1,1) (R6 ) sub-lattice.
This operation leads to the modification of Z(2,10) in Eq. (2.44), where we also set (n0 , nt ) =
(8, 8), as follows:

8
(1,1) (R5 )(1,1) (R6 ) H
0
H
G 1
Z(2,10)
(3.14)
.
Z(2,10)
=
2
0
G
( )2 (
)2
(
)8
,
If R6 is sufficiently
large, the coupling of the (H, G)-shift of the lattice to the twisted partition
function Z(4,12) H
G ensures the absence of massless states in the twisted sector.
In other situations, there are massless states in the twisted sector. Before the supersymmetry
breaking, the number of massless bosons is equal to the number of massless fermions with a multiplicity ntwist . Proceeding as in the untwisted sector, and neglecting the h1 , h2 = 1 sectors, one
152
finds that there is only a non-zero contribution when g1 + g2 = 1. The relative sign of the thermal
part T 4 and the supersymmetry breaking part M 4 depends on the choice of the operators
QF and QR . When there is no coupling to the right-moving gauge charges we obtain:
Ptwist = CTtwist T 4 f (u) + CVtwist M 4
f (1/u)
,
u
(3.15)
where
CTtwist = ntwist
( 52 )
CVtwist = ntwist
,
5
2
( 52 )
5
2
= ()(QR QF ) .
(3.16)
Here, we have for the coefficient :

= 1 when QR = QF .
= 1 when QR
= QF .
In the later case, ()(QR QF ) = ()H = 1, see Eq. (2.47), and H = 1 in the twisted sector.
The change of sign indicates that it is the bosons that are becoming massive because of the
supersymmetry breaking. This is related to a mechanism for moduli stabilization induced by
geometrical fluxes [17].
Adding the contributions of the untwisted and twisted sectors together we obtain for the pressure
P = CT T 4 f (u) + CV M 4
f (1/u)
,
u
(3.17)
where CT = CTunt + CTtwist and likewise for CV . The sign of the thermal contribution is always
positive,
CT =
nT ( 52 )
5
(3.18)
nT = nunt + ntwist . The coefficient multiplying the supersymmetry breaking part is given by
CV =
nV ( 52 )
(3.19)
5
2
with nV = nunt + ntwist . In general nV can be positive or negative depending on the model.
3.3. An explicit example
As an example we consider the E8 E8 heterotic string on a T 4 /Z2 orbifold, whose initially
supersymmetric partition function is obtained by setting
8
(1,1) (R5 )(1,1) (R6 ) 1
Z(2,10) =
(3.20)
,
2
( )2 (
)2
(
)8
,
153
and
6
+H
H

+G
G
(4,4) H
H
G 1
=
Z(4,12)
G
( )4 (
)4 2
(
)8
(3.21)
in Eq. (2.44). We shall use an R-symmetry operator associated to one of the twisted complex
planes for breaking the supersymmetry, replacing the (1,1) (R5 ) lattice as in Eq. (2.47). In the
twisted sectors, (H, G)
= (0, 0), the internal (4,4) shifted lattice is given by
(4,4)

)6
24 ( )6 (
H
=
2
1+H 2 .
G
1+H
1+G 1+G
(3.22)
The orbifolding breaks the E8 E8 gauge group to E8 E7 SU(2). Under E8 E7 SU(2),

the 248-dimensional adjoint representation of E8 decomposes as
(1, 3) (56, 2) (133, 1).
(3.23)
The untwisted sector contains 23 504 massless states giving the value nunt = 4 504 for
the total number of boson/fermion pairs. These numbers arise as follows. In terms of world-sheet
left/right movers the number of bosonic degrees of freedom is given by

nunt = 4 504 = [4] 2,3,5,6 [4]X2,3,5,6 + [248]E8 + [133]E7 + [3]SU(2)

+ [4] 7,8,9,10 [4]X7,8,9,10 + [2]SU(2) [56]E7 .
(3.24)
The first line gives the bosonic content of a d = 6 supergravity multiplet, a tensor multiplet and
a vector multiplet in the adjoint of the E8 E7 SU(2) gauge group. The second line gives the
bosonic content of four uncharged and one charged hyper-multiplets. The number of fermionic
degrees of freedom follows by supersymmetry. At finite temperature and when SUSY is broken,
the contribution of the massless untwisted states is determined as before (see Eq. (3.5)).
Next we analyze the contribution of states in the twisted sectors, H = 1. For large R0 , R5 , we
may neglect the h1 = 1 and h2 = 1 sectors. Also for R6 of order unity we may set (1,1) (R6 )
= 1.
Setting H = 1, the partition function becomes in this limit
V5
Ztwist =
(2)5

||
d d
1
6 (
(
)
)18
4 Im
7
2
2
2 (2m1 +g1 )
Im
eR0
2 (2m2 +g2 )
Im
eR5
(m1 ,g1 ),(m2 ,g2 )
2

1
a
a+1 2
a+b+ab
ag1
(a+1)g2
()
() ()
b
b+G
4
(a,b,G)

1 81 6 +1
24
1
0 2
0 2 2
+ G
G
2
1+G
1+G
,
,
(3.25)
The non-vanishing contributions arise when g1 + g2 = 1. Non-exponentially suppressed contributions arise only at the zero mass level. To obtain them, we expand the integrand in powers of
q = e2i .
When g1 + g2 = 1, we have for the left movers
154
2

1
1
a
a+1 2
a+b+ab
ag1
(a+1)g2
()
() ()
0 2 2
b
b+G
( )6 1+G
(a,b)
2

1
1
a
a+1 2
g2
b+ab
=
() ()
0 2 2
b
b+G
( )6 1+G
(a,b)
=
()g2 22 32
6 42

g2
for G = 0 or

1 + O q 1/2 .
= 4()
()g2 22 42
6 32
for G = 1
(3.26)
For the right movers we have

1 81 6 +1
1
1

2 2
+G
G
2
(
)18 0
,
,
1+G
=
=
1
q 3/4

1 + ()G 4 q 1/2 + (1 + 240 q + ) q 1/4 ()G 2 + 56 q 1/2 +
()G 2 + 8 q 1/2 + 56 q 1/2 + O(q)

.
q 1/2
(3.27)
When we add the contributions of the (H, G) = (1, 0) and (H, G) = (1, 1) twisted sectors together, we find that the lowest right-mass level is at zero mass as it is the case for the lowest
left-mass level.
Using these results we obtain the contribution to the partition function of the massless twisted
states:

dt R 2 (2m1 +g1 )2 R 2 (2m2 +g2 )2
V5 ()g2 ()g1
6
0
5
Im
Im
e
e
Ztwist = 2 (56 + 8)
7
5
2
(2) g ,g
2t 2 (m1 ,m2 )
1 2
0

5
5
2 (56 + 8) ( 2 )
f (1/u)
V4
V4
=
f (u)
(3.28)
.
5
u
(2R0 )4
(2R5 )4
2
The contribution to the pressure is as in Eqs. (3.15), (3.16) with ntwist = 25 (56 + 8), the number
of massless boson/fermion pairs in the twisted sector, and = 1. The massless bosonic content
consists of 32 scalars in the (56, 1) representation of E7 SU(2) and 128 scalars in the representation (1, 2). The number of massless fermionic degrees of freedom follows by supersymmetry.
Adding the contributions of the untwisted and twisted sectors together, we obtain
nT = 4 504 + 16 128 = 4064,
nV = 4 504 16 128 = 32.
(3.29)
nT
nV
=
= 4064. We can also
In addition, we have the choice with = 1 in Eq. (3.16), giving
change nV by considering SUSY-breaking operators involving the right-moving gauge charges
as in Eq. (2.48).
3.4. Small mass scales from Wilson line deformations
A generic supersymmetric heterotic background may contain in its spectrum massive supermultiplets whose mass is obtained by switching on non-trivial continuous Wilson lines [2628].
155
This is a stringy realization of the Higgs mechanism, breaking the initial gauge group G to a
smaller one spontaneously. This statement is not absolutely correct for discrete Wilson lines corresponding to extended symmetry points where the gauge symmetry may enhance or even get
modified.
For our purposes, we restrict to arbitrary and small Wilson line deformations starting from a
given supersymmetric background where RI , I = 6, 7, . . . , 10, are of the order the string scale.
This restriction ensures that the contributions to the thermal partition function of the momentum
and winding states in these five internal directions will be exponentially suppressed in the limit
where R0 and R5 are large.
A systematic study of the effects of Wilson lines can be found in [2628]. In the zero winding
sector, a Wilson line just modifies the KaluzaKlein momenta, and the corresponding Kaluza
Klein mass becomes
m2I
RI2
(mI + yIa Qa )2
RI2
(3.30)
where Qa is the charge operator associated to the Wilson line yIa . We distinguish two different
situations according to the direction I :
(i) I = 5 where R5 is large,
(ii) I = 6, . . . , 10 where the RI are of order the string scale.
In the first case, I = 5, after a Poisson re-summation, the net modification to the partition
function is obtained by the following replacement in Eq. (3.5):
dt
7
t2
2 (2m1 +g1 )
t
eR0
R52
(2m2 +g2 )2
t
m1 ,m2
dt
t
7
2
2 (2m1 +g1 )
t
eR0
R52
(2m2 +g2 )2
t

a
e2i(2m2 +g2 )y5 Qa .
(3.31)
m1 ,m2
The term in the brackets can be replaced with

cos 2(2m2 + g2 )y5a Qa = 1 2 sin2 (2m2 + g2 )y5a Qa .
(3.32)
In the second case, I = 6, 7, . . . , 10, we can set the momentum and winding numbers to zero,
mI = nI = 0, so that the extra modification in the partition function is the insertion of the term:

a
y a Qa 2
t I ( yI Qa )2
I
R
I
e
(3.33)
1t
.
RI
I
Incorporating the effects of the Wilson lines up to quadratic order, we get for the overall
pressure:

( 5 ) u4 (nT 2 s sin2 (2m2 y5a Qsa ))
P = T 4 52
|(2m1 + 1)iu + 2m2 |5
2 m1 ,m2

3
u2 (MT2 2 s Ms2 sin2 (2m2 y5a Qsa ))
2 (2)
T
3
|(2m1 + 1)iu + 2m2 |3
2 m ,m
1
156
n 2
V
sign(s) sin2 ((2m2 + 1)y5a Qsa )

|2m1 iu + (2m2 + 1)|5
m1 ,m2

(2)
3
MV 2 s sign(s)Ms2 sin2 ((2m2 + 1)y5a Qsa )
2 (2)
M
.
3
|2m1 iu + (2m2 + 1)|3
2 m1 ,m2
+M
5
4 (2)
5
2
(3.34)
In this expression, we have defined

Ms2 =
10
1 (yIa Qsa )2
4
RI2
(3.35)
I =6
for the pair of boson/fermion states s and also introduced

(2)
MT2 =
Ms2 ,
MV =
sign(s)Ms2 ,
s
(3.36)
where sign(s) indicates whether the state s contributes positively or negatively to nV and MV ,
both being possibly negative.
The following comments are in order:
(2)
In the above expression, y5a Qsa is an effective number summarizing the contribution of the
R5 -Wilson line in the term corresponding to the pair of boson/fermion states s. It does not
introduce a new scale.
The Ms s introduce new mass scales in the theory, qualitatively different than T and M. The
masses Ms are supersymmetric mass scales rather than SUSY-breaking scales like T and M.
The first two terms (which arise from the (g1 , g2 ) = (1, 0) sector) can be identified in the effective field theory as the thermal contribution to the pressure, Pthermal . Again the number nT
is always positive being the number of the massless boson/fermion pairs in the initially supersymmetric background. This purely thermal piece is always positive.
The two last terms can be identified as minus the effective potential Veff . This is naturally
regularized in the infrared by the temperature scale T . This infrared regularization differs
from that considered in [29], and used in [6], which is valid at zero temperature.
The number nV can be either positive or negative depending of the way supersymmetry
is broken. This shows that the sign of the one-loop effective potential depends on the way
supersymmetry is broken, as it can be seen in supergravity by utilizing the super-trace arguments [30].
3.5. Scaling properties of the thermal effective potential
The final expression for P contains various mass scales: the two supersymmetry breaking
scales which are the temperature T and the gravitino mass scale M, as well as the supersymmetric
masses Ms which are generated by the Wilson lines in the directions 6, 7, 8, 9, 10. The first
identity follows immediately from the definition of P :

T
(3.37)
Ms
+M
+
P = 4P
T
M
Ms
s
which can be best seen by writing P as
P T 4 p4 (u) + T 2 p2 (u),
u=
M
,
T
(3.38)
157
where

( 52 )
F u, y5a + F u, y5a ,
5
2
and using the definitions
p4 =

F u, y5a =
m1 ,m2 ,s
p2 =
u4 cos(4m2 y5a Qsa )

[(2m1 + 1)2 u2 + 4m22 ]5/2

( 32 )
u, y a ,
G u, y5a + G
5
3
2
u4 sign(s) cos(2(2m2 + 1)y a Qsa )

5
F u, y5a =
;
2 u2 + (2m + 1)2 ]5/2
[4m
2
1
m1 ,m2 ,s

G u, y5a =
m1 ,m2 ,s
u2 Ms2 cos(4m2 y5a Qsa )

[(2m1 + 1)2 u2 + 4m22 ]3/2
(3.39)
(3.40)
u2 sign(s)Ms2 cos(2(2m2 + 1)y a Qsa )

5
u, y a =
G
.
5
2 u2 + (2m + 1)2 ]3/2
[4m
2
1
m1 ,m2 ,s
(3.41)
Using standard thermodynamic identities, we can obtain the energy density :

T
P
P = T 4 r4 (u) + T 2 r2 (u),
T
(3.42)
where
r4 = 3p4 up4 ,
r2 = p2 up2
(3.43)
and the primes stand for derivatives with respect to u. In the sequel, we allow the SUSY-breaking
scales T and M to vary with time while fixing the supersymmetric masses Ms , and investigate
the back-reaction to the initially flat metric and moduli fields.
4. Gravitational equations and critical solution
We assume that the back-reacted spacetime metric is homogeneous and isotropic,

a
ds 2 = dt 2 + a(t)2 dk2 ,
H=
,
a
(4.1)
where k denotes the three-dimensional space with constant curvature k and H is the Hubble
parameter.
From the fact that P plays the role of the effective potential and the relation between the
gravitino mass scale M and the no-scale modulus ,

3
M =e , =
,
2
we obtain the field equation for :

P
P
+ 3H =
(4.2)
= T 4 (r4 3p4 ) + T 2 (r2 p2 ) .
= u
u T
We have made use of Eq. (3.43).
158
For other flat moduli i , with independent kinetic terms, the equation of motion is straightforward to solve,
1 2 ci2
(4.3)
= 6,
2 i
a
where the ci s are integration constants.
Knowing the thermal effective potential P , the energy density as well as the field equation
for the modulus , we can derive the (one-loop) corrected spacetime metric by solving the
gravitational field equations. These are the FriedmannHubble equation,
1
1 2
3k
i + 2 ,
3H 2 = 2 +
(4.4)
2
2
a
i + 3H i = 0
and the equation that follows from varying with respect to the spatial components of the metric:
k
1
1 2
i .
2H + 3H 2 = 2 P 2
(4.5)
2
2
a
i
For our purposes, it will be useful to replace Eq. (4.5) by the linear sum of Eqs. (4.4) and (4.5),
so that the kinetic terms of and i drop out:
2k 1
H + 3H 2 = 2 + ( P ).
2
a
(4.6)
4.1. Critical solution

The fundamental ingredients in our analysis are the scaling properties of the thermal effective potential P = T 4 p4 T 2 p2 . These scaling properties suggest to search for a solution
where all varying mass scales of the system, M(), T and 1/a, remain proportional during time
evolution:
1
H = ,
M() = uT ,
e M() =
(4.7)
a
with and u fixed in time. Our aim is thus to determine the constants and u.
On the trajectory (4.7), the -equation is given by

M4
M2
H + 3H 2 = 2 (r4 3p4 ) 4 + (r2 p2 ) 2
u
u
(4.8)
and the gravity equation (4.6) by

1
M4 1
M2
H + 3H 2 = 2k 2 M 2 + (r4 p4 ) 4 + (r2 p2 ) 2 .
(4.9)
2
2
u
u
The compatibility of these two equations requires an identification of the coefficients of the
monomials in M. The quartic terms give an equation for u, while the quadratic terms determine
the sign of the parameter k and the magnitude of |k 2 |,

6 2 1
3
2
p4
r4 = 2
(4.10)
,
r4 = 4p4 for =
2
2 1

2 2 1
1
(r2 p2 )
3
2
for
=
(r2 p2 ) 2
.
2k 2 =
2k 2 =
(4.11)
2
2
u
u2
159
Eq. (4.10) reminds us of the equation of state for thermal radiation in five dimensions. In the
absence of Wilson lines, where r2 = p2 = 0, we have that = 4P , which is indeed the 5d state
equation for thermal radiation.
Next let us consider the FriedmannHubble equation (4.4) along the critical trajectory (4.7)
where we may set 2 = H 2 / 2 . It becomes
2

c2
6 1
3k
1 2
3k
i
2
4
2
(4.12)
=
+
T
r
+
T
r
+
.
3H
4
2
i
2
6 2
a2
a2
a6
i
Then by using the compatibility conditions (4.10) and (4.11), it is easy to check that Eq. (4.9)
follows. Our ansatz (4.7) allows to reduce the differential system of equations for the modulus
and the gravitational field to the compatibility equations for the constants , u and k and the
FriedmanHubble Eq. (4.12).
cr and cm , once
The dilatation factor in front of 3H 2 can be absorbed in the definition of k,
we take into account equations (4.14) and (4.15) below:
3H 2 =
3k
cr
cm
+
+ 6,
a2 a4
a
(4.13)
where

1 6 2 1 3(2 2 1)
(r2 p2 ) + r2 ,
3k = 2 2
4
6 1 u2
(4.14)
cr =
1 6 2 r4
1 6 2 p4
=
,
4 6 2 1 u4 4 2 2 1 u4
(4.15)
cm =
6 2 2
ci .
6 2 1
(4.16)
and
Recalling that 2 = 3/2 and the relation (3.43), Eq. (4.10) for u becomes
p4 + up4 = 0,
whose solution determines
9 2
9 p4
cm =
ci ,
cr = 4 4 ,
8
2 u
i
(4.17)
3 2p2 5up2
k =
.
16 2
u2
(4.18)
Clearly, a necessary condition for the curvature k not to vanish is to have non-trivial Wilson lines
in any of the directions 6, 7, 8, 9, 10.
We note that Eq. (4.13) also controls the dynamics of an FRW universe, where space has constant curvature k and is formally filled with a thermal bath of radiation (since the sign of cr can
a priori be positive as well as negative). There can be an extra contribution, arising from the kinetic terms of some extra flat moduli, that scales as 1/a 6 . Finally let us address a seeming puzzle.
We said that along the critical trajectory and in the absence of Wilson lines, our thermodynamic
quantities satisfy = 4P . How can this situation correspond to a 4d universe filled with thermal
radiation? The answer is that we must take into account the kinetic energy density of the modulus
field . When there are no Wilson lines and the kinetic terms of the other moduli are switched
off, the FriedmannHubble equation gives 2 = (r4 T 4 )/4 along the critical trajectory, and so
160
Fig. 1. The non-trivial root u of Eq. (4.17) as a function of the ratio nV /nT .
the total energy density and pressure satisfy

1
9r4 4
T ,
tot = r4 T 4 + 2 =
2
8
1
3r4 4
Ptot = p4 T 4 + 2 =
T .
2
8
Thus tot = 3Ptot which is the 4d equation of state for thermal radiation.
(4.19)
4.2. Numerical study

4.2.1. Without Wilson lines y5a
Let us consider the case where the Wilson lines along the X5 -direction are switched off. From
Eq. (3.40), we obtain
p4 (u) =
( 52 )
5
2

nT f (u) + nV f(u) ,
f(u) = u3 f (1/u),
(4.20)
thanks to the identities nT f (u) F (u, y5a = 0) and nV f(u) F (u, y5a = 0). Since f + uf
and f + uf vanish at the origin, Eq. (4.17) admits a universal solution u = 0 for arbitrary nV
in the range nT nV nT . This solution corresponds to M(t) 0 at some finite T , and is
associated to a 5-dimensional purely thermal system (R5 = ). The 4d effective description we
have considered is not valid in this case. We are thus looking for non-trivial solutions u > 0 of
Eq. (4.17), which we write in the form
nV
f + uf
=
.
nT
f + uf
(4.21)
A numerical study shows that such non-trivial solutions (and consequently, non-trivial cosmological evolutions) exist only for models satisfying
0.0666 . . . <
nV
< 0.
nT
(4.22)
The non-vanishing root of Eq. (4.21) is an increasing function of the ratio nV /nT satisfying
u + when nV /nT 0 and u 0 when nV /nT 0.0666 . . . . (See Fig. 1.)
161
Fig. 2. Phase diagram (nV /nT , MV /MT2 ) of possible models with vanishing Wilson lines in the direction 5 and arbitrary
ones in the directions 6, 7, 8, 9, 10. The quantity cr is everywhere positive, while the curvature k is negative above the
critical curve and positive elsewhere. There is no critical point u > 0 outside the rectangular domain.
(2)
The corresponding value of cr in Eq. (4.18) is finite and always positive. This can be seen
by noting that the quantity nT f + nV f appearing in the expression for cr equals u(nT f +
nV f ) > 0. The positivity follows since f and f are always negative and positive respectively.
When Wilson lines in the directions 6, 7, 8, 9, 10 are switched on, we need to determine the
(2)
Defining G(u, y a = 0) M 2 g(u) and G(u,
value of k.
y5a = 0) MV g(u),
Eq. (3.39) gives

T
5
p2 (u) =
( 32 )
3
2

(2)
MT2 g(u) + MV g(u)
g(u)
= ug(1/u),
(4.23)
so that the condition for the sign of ksee

Eq. (4.18)can be expressed as
k 0
(2)
1
MV
MT2
2g 5ug
.
2g 5ug
(4.24)
The function of u appearing in the RHS of the inequality is monotonically decreasing from +
(at u = 0) to 0 (at u = ). It takes the value 1, which is the maximum allowed value of the ratio
MV(2) /MT2 , for u 0.940 . . . . Thus a non-trivial solution of Eq. (4.17) satisfying u 0.940 . . .
always implies k 0. If instead u > 0.940 . . . , one has
k < 0
M
2g 5ug
< V2 < 1.

2g 5ug
MT
(2)
(4.25)
For the critical value u = 0.940 . . . , Eq. (4.21) requires that nV /nT 0.0348 . . . . Fig. 2 rep(2)
resents the two-parameter phase diagram (nV /nT , MV /MT2 ) of the models. It is divided in
negative above the critical curve and positive
two regions characterized by opposite signs for k:
elsewhere. Let us detail the models we considered in the previous sections by computing the
(2)
quantities nV , nT , MT2 and MV to determine where they are located on this phase diagram.
Model 1 We consider an N = 4 heterotic string model, for example the E8 E8 theory on T 6 ,
at finite temperature and when supersymmetry is spontaneously broken by choosing QR = QF .
162
For this model

nT = nV = 23 504.
(4.26)
Since nV > 0, there is no critical solution u > 0.

Model 2 In the same E8 E8 heterotic string model, the R-symmetry operator used to break
supersymmetry is chosen to be QF + QE8 + QE8 in order to have nV < 0 (see Eq. (2.43) for the
T = 0 case):
nT = 23 504;
nV = 23 (8).
(4.27)
nV /nT
= 1/63 0.0159, and so the model admits a critical solution u > 0. One finds
Then
numerically that u 1.46 and 4 cr 441.
Let us consider Wilson lines yIa in the directions I = 6, 7, 8, 9, 10 and define

a 2
1 yIa 2
=
, a = 1, . . . , 16,
Y
(4.28)
4
RI
I
where a = 1, . . . , 8 stand for the Cartan generators of the first E8 factor and a = 9, . . . , 16 for the
second. The derivation of the charges Qa of the initially massless states (see Appendix A) gives:
2
2

MT2 = 23 60 Y 1 + + Y 16 ,
(4.29)
and
2
2

MV(2) = 23 (4) Y 1 + + Y 16 .
(4.30)
(2)
MV /MT2
Their ratio is Wilson line independent,

= 1/15, which is negative, and so the effective curvature is positive. Numerically, one finds 2 k 67.2((Y 1 )2 + + (Y 16 )2 ).
Model 3 In the N = 2 orbifold model (see Section 3.3 for details), we set QR = QF + QH +
QE7 with QE7 being 1 for the spinorial representations of E7 , decomposed in terms of SO(12)
ones, and 0 for the vectorial ones. We find the following:
nT = 4064;
Since
nV /nT
nV = 992.
(4.31)
is positive, this model does not admit a critical point u > 0 at this stage.
Model 4 This is the N = 2 orbifold model constructed in Section 3.3, with

nT = 4064,
nV = 32.
nV /nT
(4.32)
103 ,
= 1/127 7.87
there is again a non-trivial critical solution u > 0.
Since
Numerically, we find u 1.90 and 4 cr 130.
Before switching on Wilson lines in the directions I = 6, 7, 8, 9, 10, the gauge group is E7
SU(2) E8 . Let us consider arbitrary Wilson lines, Y a , a = 1, . . . , 16: Y 1 , . . . , Y 6 for the Cartan
generators of the SO(12) subalgebra of E7 , Y 7 for the Cartan generator of the SU(2) subalgebra
of E7 , Y 8 for the SU(2) factor and Y 9 , . . . , Y 16 for the E8 . The mass square scales, computed in
Appendix A, are:
MT2 = 624
6

Ya
a=1
16

a 2
2
2
Y
+ 1248 Y 7 + 736 Y 8 + 240
a=9
(4.33)
163
and
(2)
MV = 144
6

Ya
16

a 2
2
2
Y .
288 Y 7 + 224 Y 8 + 240
a=1
(4.34)
a=9
(2)
Since MV has both definite positive and definite negative monomials, we can choose configu due to the
rations of Wilson lines giving rise to either positive or negative effective curvature k,
a
fact that 0.0348 . . . < nV /nT < 0. For instance, if any Y , a = 1, . . . , 7, is non-trivial (while
Y a = 0, a = 8, . . . , 16), one has MV2 /MT2 = 3/13, and so k is positive. Numerically, one finds
2 k 101((Y 1 )2 + + (Y 6 )2 + 2(Y 7 )2 ). On the contrary, if any Y a , a = 9, . . . , 16, is non(2)
trivial (while Y a = 0, a = 1, . . . , 8), one has MV /MT2 = 1 and so the representative point in
the phase diagram of Fig. 2 sits above the critical curve, i.e., k is negative. Numerically, one has
2 k 274((Y 9 )2 + + (Y 16 )2 ).
4.2.2. With Wilson lines y5a
(2)
Any model originally characterized by the quantities nT , nV , MT2 and MV can be deformed
a
by switching on the Wilson lines y5 , a = 1, . . . , 16. We are looking for solutions of Eq. (4.17)
written in terms of the functions defined in (3.40):

F u, y5a + uF u, y5a + F u, y5a + uF u, y5a = 0.
(4.35)
We observe that the thermal contribution described by F (u, y5a ) + uF (u, y5a ) vanishes at u = 0
and asymptotically at infinity. In all of the following examples, it is also positive. On the contrary,
the effective potential corrections F (u, y5a ) + uF (u, y5a ) vanish at u = 0 and diverge to +
or at infinity. Thus, in presence of arbitrary Wilson lines y5a , the universal solution u = 0
remains and we are looking for non-trivial ones u > 0. These can only arise when F (u, y5a ) +
uF (u, y5a ) takes negative values.
Model 1 Among the Wilson lines y5a (a = 1, . . . , 8 for the first E8 factor and a = 9, . . . , 16 for
the second), we choose to switch on either
(i) y51 (with y52 = = y516 = 0) or
(ii) y51 = = y58 (with y59 = = y516 = 0).
In these cases and at fixed Wilson lines, F (u, y5a ) + uF (u, y5a ) increases from 0 to +, and
so there is still no solution u > 0.
Model 2 For the Wilson lines defined in case (ii) above, one finds that the critical solution u > 0
present before deformation is slightly shifted. However, in case (i), the root u > 0 is sent to +
when y51 approaches the numerical value 0.227. . . from below. For y51 above this critical bound,
there is no non-trivial solution anymore.
Model 3 As said before, this model does not admit a critical solution u > 0 when all Wilson lines in the fifth dimension are switched off. We consider four patterns of deformations, by
switching on a Wilson line for a single Cartan generator:
(i) y51 (Cartan generator of SO(12) E7 ),
164
(ii) y57 (Cartan generator of SU(2) E7 ),

(iii) y58 (Cartan generator of SU(2)),
(iv) y59 (Cartan generator of E8 ).
In cases (i) and (iv), a large enough Wilson line deformation generates a non-trivial solution.
We find the critical bound y51 = 0.895 . . . for case (i), and y59 = 0.772 . . . for case (iv). In all
these cases, the phase transition occurs when the limit at u of the effective potential
contribution F (u, y5a ) + uF (u, y5a ) is switched from + to . In some sense, a solution
u = + appears at the transition, and then decreases for a larger Wilson line, and finally reaches
a non-zero minimal value. In cases (ii) and (iii), since F (u, y5a ) + uF (u, y5a ) is positive for any
value of the Wilson line, the deformation does not create a non-trivial solution.
Model 4 We consider the same four patterns of Wilson lines. This time, the critical solution
u > 0 exists before deforming the model. We find that in the above cases, switching on a large
enough Wilson line makes this solution disappear. However, two distinct behaviors are found:
In cases (i) and (ii), we find again that the phase transition corresponds to a change in the
behavior of the effective potential contribution at . In a similar way as before, the solution
u > 0 existing before deformation will increase when switching on the Wilson line, then go to
+ at the transition point, and disappear for a larger Wilson line. We find the critical bound
y51 = 0.105 . . . for case (i), and y57 = 0.074 . . . for case (ii).
For (iii) and (iv), switching on the Wilson line makes the critical solution u > 0 decrease
towards zero, so that we are left only with the universal solution when the Wilson line is above
some bound. In case (iii), this maximal value is y58 = 0.205 . . . , while it is y59 = 0.207 . . . for
case (iv).
Some remarks are in order:
In all cases presented above, when a critical solution u > 0 exists, the effective radiation
term cr given by Eq. (4.18) is strictly positive.
In some models, incorporating the Wilson lines in the fifth direction allows the critical solution u > 0 to be close to 0 or large. In other words, a hierarchy between the scales M and T
can be found by tuning the moduli y5a .
5. Cosmological evolutions
When a non-degenerate solution u > 0 exists, the model admits a well-defined low energy
description in four dimensions. The dynamics in this regime is controlled by the Friedmann
In all models
Hubble equation, whose behavior depends drastically on the signs of cr and k.
we considered here, with initial N = 4 or N = 2 supersymmetry, cr turns out to be positive.
However, this situation may not be generic in more complex stringy examples with initial N = 1
supersymmetry. For completeness, we briefly describe all possible cosmologies arising for any
positive or negative value of cr . In addition, we allow non-trivial time-dependent profiles for the
moduli i , giving cm > 0.
5.1. Solutions for cr > 0
For k = 0, cr > 0, cm 0.
165
Fig. 3. Cosmological evolution for the case k = 0, cr > 0, cm 0.
When Wilson lines in the directions 6, 7, . . . , 10 are absent, the curvature k vanishes. In real
time, the FriedmannHubble equation (4.13),
3H 2 =
cr
cm
+ 6,
4
a
a
(5.1)
can be used to express the time variable t as an integral function of the scale factor a,

3
cr
t (a) =
a

0
v 2 dv
v 2 + a02

(a 0),
a0 =
cm
.
cr
(5.2)
In this form, it is straightforward to draw a(t) (see Fig. 3). The explicit solution is

t (a) = t0

a2
a
a
1 + 2 arcsinh
,
a0
a
a0
0
t0 =
3 cm
,
2 cr3/2
(5.3)
describing an expanding universe that starts with a big bang. As can be seen from Eq. (5.1), the
slope a is infinite when a vanishes. At large t, the scale factor behaves as in the cm = 0 particular
case:

a(t) =
cr
3
1/4
2t
(t 0).
(5.4)
Since the transformation t t is a symmetry of the FriedmannHubble equation, the previous

expanding solutions have contracting counterparts and thus ending at t = 0 with a big crunch.
Finally, since the RHS of Eq. (5.1) is positive, there is no solution in Euclidean time.
The effective field theory description always breaks down before the occurrence of a spacelike singularity, when the temperature (in string frame) is of order the Hagedorn temperature. At
this temperature scale, new stringy dynamics must be taken into consideration which can result
into a phase transition, realizing the scenario of [16].
For k > 0, cr > 0, cm 0.
166
When we switch on Wilson lines, k > 0. Rewriting the FriedmannHubble equation (4.13) in
the form

c2 + 12kc
m cr

2 2
2
r
2
2
2
a a = k a + a a+ a , a =
(5.5)
,
6k
one expects a cosmological evolution satisfying 0 a a+ should exist, while a solution with
a scale factor greater than a+ should make sense in imaginary time only. In real time, one can
actually express t as a function of a as follows

a
v 2 dv
1

t (a) = ti +
(5.6)
(0 a a+ ),
2 )(a 2 v 2 )
k 0
(v 2 + a
+
where
1
ti =
k
a+
0
v 2 dv

.
2 )(a 2 v 2 )
(v 2 + a
+
(5.7)
From these expressions, one can see that the cosmological evolution starts with a big bang at
t = ti . It expands until t = 0 where the maximum size of the universe a+ is reached, and then it
contracts until a big crunch occurs at t = ti .
To find a Euclidean solution, one needs to consider a scale factor greater than a+ . It is then
possible to find it by proceeding as before, or use the fact that such a solution can be obtained by
analytic continuation of the expression (5.6) at t = 0, where a = a+ . One can write

a+
aE
v 2 dv
v 2 dv
1
1

+
t = ti +
2 )(a 2 v 2 )
2 )(v 2 a 2 )
k 0
k
(v 2 + a
(v 2 + a
+
+
a+
i,
(5.8)
from which we derive

aE
v 2 dv
1

(aE ) =
2 )(v 2 a 2 )
k a
(v 2 + a
+
+
(aE a+ ).
(5.9)
Fig. 4 represents the solutions a(t) and aE ( ). In the particular case where cm = 0, the solutions (5.6) and (5.9) are taking the explicit forms

2

cr
t
1 cr
a(t) = a+ 1
(5.10)
(ti t ti ), a+ =
, ti =
,
ti
3k
k 3
and

aE ( ) = a+ 1 +
2
,
ti
whose shapes are similar to the generic case with cm > 0.

For k < 0, cr > 0, cm 0.
(5.11)
167
Fig. 4. Cosmological evolution for the case k > 0, cr > 0, cm 0 (in bold line). A big bang and a big crunch are occurring
at t = ti and t = ti , respectively. This solution is connected to a Euclidean one at t = i = 0 that is asymptotically
linear.
Fig. 5. Cosmological evolution for the case k < 0, cr > 0, cm 0.
This case is easier to deal with. Eq. (4.13) can be rewritten as

m
c
cr2 12|k|c
2

2 2
r
2
2
a 2 + a
a + a+
, a =
,
a a = |k|
6|k|
(5.12)
and admits the expanding solution

1
t (a) =
k
a

0
v 2 dv
(5.13)
2 )(v 2 + a 2 )
(v 2 + a
+
(See Fig. 5.) After a big bang, the scale factor is growing linearly in time. The result for the
particular case cm = 0 can be written more explicitly. The solution takes the form

cr
1 cr
t + t0 2
, t0 =
1 (t 0), a+ =
.
a(t) = a+
(5.14)
t0
3
3|k|
|k|
These cosmological solutions do not admit a sensible Euclidean continuation.
168
Fig. 6. Cosmological evolution for the case k = 0, cr < 0, cm 0 (in bold line). A big bang and a big crunch are occurring
at t = ti and t = ti , respectively. This solution is connected to a Euclidean one at t = i = 0.
5.2. Exotic cosmologies with cr < 0

Although in all explicit models we presented before cr is positive, it is interesting to analyze
the exotic situation with negative cr , which is not a priori forbidden in more general cases with
N = 1 initial supersymmetry.
For k = 0, cr < 0, cm 0.
A cosmological evolution in real time only exists if cm is switched on. The scale factor satisfies
0 a a0 where

cm
.
a0 =
(5.15)
|cr |
Between a big bang at ti < 0 and a big crunch at ti > 0, a reaches a maximum a0 at t = 0. At
this time, an analytic continuation is allowed: A Euclidean solution satisfies aE a0 and goes to
infinity for large positive or negative Euclidean time (see Fig. 6).
For k > 0, cr 0, cm 0.
Most of the considerations of this case are identical to the one derived for k > 0, cr > 0,
cm 0. In particular, the FriedmannHubble equation is still given by Eq. (5.5) and both the
cosmological solution (5.6) and the Euclidean one (5.9) are valid. They are shown in Fig. 7. As
long as cm > 0, the only qualitative difference with the case cr > 0 is that the Euclidean solution
has two symmetric inflexion points. However, when cm vanishes, the evolution in real time ceases
to exist.
For k < 0, cr < 0, cm 0.
This case presents the most interesting features and involves either a first or second order
phase transition in the early universe. The former case is the only one considered in this paper,
where a Euclidean solution has a finite action and thus can be interpreted as an instanton involved
in a tunneling effect. These behaviors are qualitatively similar to the inflationary case studied in
169
Fig. 7. Cosmological evolution for the case k > 0, cr 0, cm > 0 (in bold line). A big bang and a big crunch are occurring
at t = ti and t = ti , respectively. This solution is connected to a Euclidean one at t = i = 0 that is asymptotically
linear and has two symmetric inflexion points.
Fig. 8. There are two cosmological evolutions (in bold lines) for the case k < 0, cr 0, cm 0, when > 0. The first one
starts with a big bang at t = ti and ends with a big crunch at t = ti . The second one has a contracting phase followed
by an expanding one. These two branches are connected to each other by a first order phase transition via an instanton.
[6,12]. To be more specific we consider the FriedmannHubble equation in the form

4 |cr |a 2 + cm ,
3 a 2 a = 3|k|a
(5.16)
and discuss various regimes, depending on the value of the discriminant of the RHS
m.
cr2 12|k|c
(i) When > 0, there are two critical values for the scale factor

|c | c2 12|k|c
m
r
r
a =
.
6|k|
(5.17)
(5.18)
Eq. (5.16) then admits two distinct cosmological evolutions. The first one satisfies 0 a a
and corresponds to the usual dynamics between a big bang at ti < 0 and a big crunch at ti . The
170
Fig. 9. Cosmological evolution for the case k < 0, cr 0, cm 0, when < 0. It describes a second order phase
transition occurring at tinf , between a phase that starts with a big bang to another phase that expands linearly in time.
scale factor reaches a maximum a at t = 0. The second one describes an asymptotically linear
contracting solution followed by an asymptotically linear expanding one. The two branches are
smoothly connected at t = 0, where the scale factor reaches a minimum value a+ . Therefore,
this solution is non-singular. (See Fig. 8.) The two cosmological evolutions are also related to
one another by a double analytic continuation: t = i and then = f + it. Between = 0
and = f , a Euclidean solution whose action can be shown to be finite is allowed. It is thus an
instanton between the two branches in real time and contributes to a first order phase transition.
We note that when cm = 0, the big bang/big crunch solution disappears since a vanishes.
(ii) Let us turn now to the second case where the discriminant (5.17) satisfies < 0. Eq. (5.16)
does not admit any critical point and the scale factor is never stationary. There is a single cosmological evolution (and no Euclidean solution). It increases from a big bang at t = 0, while for
large t, its time dependence becomes linear. Thus, close to the big bang, the evolution is similar
to the first solution occurring when > 0, while for large t its behavior is similar
to the second
expanding solution. Since there is an inflexion point at t = tinf when a = 2cm /|cr |, the cosmological evolution for < 0 can be interpreted as a second order phase transition between the
same initial and final states encountered in the first order phase transition for > 0. (See Fig. 9.)
(iii) In the critical case = 0, Eq. (5.16) admits two expanding cosmological evolutions
which are asymptotic to a static one, a a0 , where

|cr |
2cm
a0 a =
(5.19)
= ainf =
,
|cr |
6|k|
together with two contracting ones obtained by time reversal. The first expanding solution starts
with a big bang, while the second one is linear for large positive time. (See Fig. 10.)
6. Conclusions
We have obtained several cosmological solutions in a large class of four-dimensional heterotic
string compactifications with spontaneously broken N = 4 or N = 2 spacetime supersymmetry.
The cosmological evolution is induced once radiative quantum and thermal corrections are taken
into consideration. These corrections are calculated at the perturbative string level and shown to
possess universal scaling properties. The reason is an underlying duality between the temperature
and the supersymmetry breaking scale.
171
Fig. 10. There are two expanding (contracting) cosmological evolutions for the case k < 0, cr 0, cm 0, when = 0.
All are asymptotic to the static solution a a0 .
Our solutions correspond to homogeneous and isotropic FriedmannRobertsonWalker universes. They are characterized by the ratio of the supersymmetry breaking scale to the temperature, and this ratio remains constant during time evolution. Even though KaluzaKlein states
associated to the supersymmetry breaking cycle are thermally excited, the equation of state
governing cosmological evolution is identical to that of massless thermal radiation in four dimensions. This is due to the special relation between the no-scale modulus field associated to
the supersymmetry breaking scale and the Hubble parameter: 2 = 2H 2 /3. Universes with
spherical, toroidal or hyperbolic spatial sections can be found once we incorporate Wilson line
deformations.
In this paper we focused on the low temperature phase of the models. When the temperature is close to the Hagedorn temperature our effective field theory analysis breaks down and
new stringy dynamics must be taken into consideration. It would be interesting to investigate if
phase transitions can occur in these models as the temperature approaches the Hagedorn temperature, and whether such phase transitions result in non-singular time-dependent geometries. To
this extent it could prove useful to incorporate in our work the proposal of [23], where such a
phase transition is shown to occur in N = 4 heterotic string models, and study the cosmological
implications.
It would be interesting to extend our analysis to the N = 1 heterotic orbifold models, and for
the cases where supersymmetry is broken spontaneously. In this class of models, one expects to
find inflationary phases, once radiative and thermal corrections are properly taken into account.
The analysis of [6,12] reveals interesting transitions between such inflationary phases and radiation dominated phases with similar properties to those found in this work. In our examples, the
coefficient of the 1/a 4 term in the FriedmannHubble equation is positive. Perhaps among the
N = 1 examples it is possible to find models characterized by negative values of this coefficient.
Then non-trivial cosmological phenomena would occur, including first or second order phase
transitions that allow for the possibility to realize the proposal for the creation of a universe from
nothing [11] in string theory [6,12].
The relation between the supersymmetry breaking scale with the temperature is a key property
of our solutions. Suppose that such a scaling property persisted in an early universe epoch, and
that initially supersymmetry was broken around the string scale. During such epoch, the SUSYbreaking scale gets lower and lower as the universe expands and cools. At lower temperatures
172
new dynamics may become relevant that can stabilize this scale. Such a scenario can give us a
new perspective on how to handle the hierarchy and naturalness problems.
Acknowledgements
We are grateful to Constantin Bachas, Ramy Brustein, Dieter Lst, Marios Petropoulos, Jan
Troost and Fabio Zwirner for useful discussions. N.T. thanks the Ecole Normale Superieure and
C.K. and H.P. the University of Cyprus for hospitality.
The work of C.K. and H.P. is partially supported by the EU contract MRTN-CT-2004-005104
and the ANR (CNRS-USAR) contract 05-BLAN-0079-01 (01/12/05). N.T. and C.K. are supported by the EU contract MRTN-CT-2004-512194. H.P. is also supported by the EU contracts
MRTN-CT-2004-503369 and MEXT-CT-2003-509661, INTAS grant 03-51-6346, and CNRS
PICS 2530, 3059 and 3747, while N.T. is also supported by an INTERREG IIIA Crete/Cyprus
program.
Appendix A
In this appendix, we provide details on the derivation of the charges associated to the Cartan
generators of the gauge groups of the heterotic string models we considered. We also derive the
contributions to MT2 and MV(2) (see Eq. (3.34)) associated to Wilson lines in the internal direction
6 only. Since the result is linear in the (y6i )2 s, the result for arbitrary Wilson lines in the directions
6, 7, 8, 9, 10 is obtained under the replacement (y6i )2 /(4R62 ) (Y i )2 , where (Y i )2 is defined
in (4.28).
Consider the N = 4 heterotic model with gauge group E8 E8 . When built out from 32
worldsheet fermions as in the standard procedure, the vertex operators for the gauge fields associated to the Cartan generators of the first E8 factor are given by
2i1 2i ,
i = 1, . . . , 8,
(A.1)
whereas those associated to the Cartan generators of the second E8 are given by
2i1 2i ,
i = 9, . . . , 16.
(A.2)
In terms of representations of SO(16), the adjoint representation of E8 decomposes as

248 = 120 128,
(A.3)
where the 120 is the adjoint representation of SO(16) and the 128 is the spinorial representation
with positive chirality. The corresponding vertex operators can be written explicitly by bosonizing the 32 fermions into 16 free bosons Hi , i = 1, . . . , 16. This formalism has the advantage to
make the roots of the Lie algebra appear in a clear way. For the 120 we obtain

120 : ei(Hj Hk ) , (j
= k) iHj , j, k {1, . . . , 8}.
(A.4)
The 8 latter vertex operators correspond to the Cartan generators. For the 128 we have
i
128 : e 2 (1 H1 +2 H2 ++8 H8 ) ,

with the GSO constraint 8i=1 i = 1. Our goal here is to compute the quantity

1 s a 2
Qa y6 ,
MT2 y6a =
4R62 s248
(A.5)
(A.6)
173
where we denote by Qsa the charges of a state s with respect to the Cartan generators. We are going to consider the case of one E8 gauge group; the generalization to E8 E8 is straightforward.
To begin with, we recall that the Cartan states are neutral. Then we are interested in the 112
remaining states of the 120 adjoint representation. It is not hard to see that

2
j 2
Qsa y6a =
1 y6i + 2 y6 .
(A.7)
s120
i
=j {1,...,8} 1 ,2 =1
Therefore we get a quadratic polynomial in the y6i s. Noticing that this polynomial is invariant
j
under the transformations y6i y6i and y6i y6 , we obtain
8

2
i 2
Qsa y6a =
y6 .
s120
(A.8)
i=1
Computing the (y61 )2 term, we get = 28.

For the 128, we see that
7 2

1
2
Qsa y6a =
i y68 .
1 y61 + 2 y62 + +
4
s128
1 ,...,7 =1
(A.9)
i=1
If we set y68 = 0, the symmetries y6i y6i and y6i y6 , valid for i, j = 1, . . . , 7, guarantee that
this polynomial will be of the form
7

i 2
y6 .
(A.10)
i=1
Restoring y68
= 0 gives a (y68 )2 term and crossed terms y6i y68 . However, y68 has been artificially
isolated in the treatment of the GSO constraint: by isolating other y6i s and using the same arguments, we can show that our polynomial is of the form
8

2
i 2
Qsa y6a =
y6 .
s128
(A.11)
i=1
We obtain = 32.
It is then straightforward to evaluate the sums encountered before. We obtain
16

2
23
2
i
MT =
y6
60
,
4R62
i=1
and when coupling the ScherkSchwarz cycle to the helicity of the E8 representation

16

i 2
23
(2)
MV =
y6
4
.
4R62
i=1
(A.12)
(A.13)
In the N = 2 models, the orbifolding breaks E8 E7 SU(2), under which the adjoint
representation decomposes as
248 (133, 1) (56, 2) (1, 3).
(A.14)
174
The Cartan generators of E8 give the Cartan generators of E7 SU(2):

iH1 , . . . , iH6 , i(H7 H8 ) , i(H7 + H8 ).
(A.15)
y61 , . . . , y67 , y68 ,
we compute the charges of the various states step by step.

Switching on arbitrary
In the 133, we have 7 neutral Cartan operators, 60 ladder operators in the Adj(SO(12)) subalgebra
ei(Hj Hk ) ,
j
= k {1, . . . , 6},
(A.16)
2 ladders in the Adj(SU(2))

ei(H7 H8 ) ,
(A.17)
and 64 ladders in a spinorial representation

i
e 2 (H1 H6 (H7 H8 ))
(A.18)
obeying a GSO condition. We see that y67 has a particular role here. The latter states have charges
12 under the first six Cartan generators, and charges 1 under the seventh. Using the same
arguments as before, we see that the sum for the spinorial states is of the form
6

2 2
i
7
(A.19)
y6 + 2y6
.
i=1
The polynomial we are looking for is therefore

6

6

2 2

2
2
2
Qsa y6a = 20
y6i + 2 2y67 + 16
y6i + 2y67
s133
i=1
i=1
6

i 2
2
= 36
y6 + 72 y67 .
(A.20)
i=1
Note that if wewant to couple the ScherkSchwarz cycle to the helicity of the E7 , we have to
compute also s sign(s)(Qa y6a )2 , where states in the spinorial representations of the SO(12)
subgroup contribute with a minus sign. To get this sum we have to put a minus sign in front of
the 64 part, so that
6

6

2 2

i 2
s a 2
7 2
i
7
y6 + 2 2y6 16
y6 + 2y6
sign(s) Qa y6 = 20
i=1
s133
=4
i=1
6

i 2
2
y6 56 y67 .
(A.21)
i=1
For the (56, 2) representation, we begin with the states with vertex operators
eHi H7 ,
eHi H8 .
The corresponding Qsa y6a are respectively

y6i y67 y68 .
y6i y67 + y68 ,
(A.22)
(A.23)
175
Therefore the sum for these states equals

6
6

2
i 2
2
i 2
2

Qsa y6a = 4
y6 + 24 y67 + y68 + 4
y6 + 24 y67 y68
s
i=1
=8
i=1
6

i 2
2 2
y6 + 48 y67 + y68 .
(A.24)
i=1
The remaining states to be considered have vertex operators

i
e 2 (H1 H6 (H7 +H8 )) .
(A.25)
For these, we get

6
2 2
i
8
y6 + 2y6
.
16
(A.26)
i=1
Adding everything, we get the final result for the representation:

6

2
i 2
2
2
Qsa y6a = 24
y6 + 48 y67 + 112 y68 .
(A.27)
i=1
s(56,2)
If we couple to the E7 helicity, we also have

6

i 2
s a 2
2
2
y6 + 48 y67 16 y68 .
sign(s) Qa y6 = 8
(A.28)
i=1
s(56,2)
For the 3 of SU(2) the two states

ei(H7 +H8 )
(A.29)
have charges 2. So we get

2
2
Qsa y6a = 8 y68 .
(A.30)
s(1,3)
In the twisted sector, we encounter the representation (56, 1), whose sum is obtained by
switching off y68 in the result obtained for the (56, 2) representation and by dividing the result
by 2. One then obtains,
6

2
i 2
2
Qsa y6a = 12
y6 + 24 y67 ,
s(56,1)

s(56,1)
(A.31)
i=1
6

2
i 2

2
y6 + 24 y67 .
sign(s) Qsa y6a = 4
i=1
We also encounter the (1, 2) representation, where the sum equals 2(y68 )2 .
(A.32)
176
Application to models 3 and 4 For model 4, we set QR = QF + QH . If we consider Wilson

lines corresponding to the 16 Cartan generators of E8 E7 SU(2), the result is
16
6
6
2

i 2
i 2
7 2
4
2,(2)
i
y
y
y6
+
36
+
72
y
+
24
MT ,V =
60
6
6
6
2
4R6
i=9
i=1
i=1

7 2
8 2
8 2
+ 48 y6 + 112 y6 + 8 y6

6

i 2
7 2
8 2
1
y6 + 24 y6
.
32 12
+ 128 2 y6
4R62
i=1
So we get

16
6

1
2
2
2
2
y6i + 624
y6i + 1248 y67 + 736 y68
MT2 =
240
,
4R62
i=9
i=1

16
6

i 2
i 2
7 2
8 2
1
(2)
y6 144
y6 288 y6 + 224 y6
MV =
240
.
4R62
i=9
i=1
(A.33)
For model 3, we set QR = Qa + QH + QE7 . We get the same expression for MT2 , while
6
6
2

i 2
7 2
2
4
(2)
i
y6 56 y6 8
y6 + 48 y67
MV =
4
2
4R6
i=1
i=1

16

i 2
8 2
8 2
y6
16 y6 + 8 y6 + 60
i=9

i 2
7 2
8 2
1
y6 + 24 y6
32 4
+ 128 2 y6
4R62
i=1

6
16

i 2
i 2
7 2
8 2
1
=
y6 800 y6 288 y6 + 240
y6
112
.
4R62
i=1
i=9

6

(A.34)
References
[1] I. Antoniadis, C. Bachas, J.R. Ellis, D.V. Nanopoulos, An expanding universe in string theory, Nucl. Phys. B 328
(1989) 117.
[2] C. Kounnas, D. Lust, Cosmological string backgrounds from gauged WZW models, Phys. Lett. B 289 (1992) 56,
hep-th/9205046.
[3] C.R. Nappi, E. Witten, A closed, expanding universe in string theory, Phys. Lett. B 293 (1992) 309, hep-th/9206078.
[4] S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici, From big bang to big crunch and beyond, JHEP 0206 (2002) 017,
hep-th/0204189.
[5] C. Kounnas, N. Toumbas, J. Troost, A wave-function for stringy universes, JHEP 0708 (2007) 018, arXiv: 0704.1996
[hep-th].
[6] C. Kounnas, H. Partouche, Inflationary de Sitter solutions from superstrings, arXiv: 0706.0728 [hep-th].
[7] H. Liu, G.W. Moore, N. Seiberg, The challenging cosmic singularity, gr-qc/0301001.
[8] J.M. 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.
177
[9] P.K. Townsend, Quintessence from M-theory, JHEP 0111 (2001) 042, hep-th/0110072;
J. Sonner, P.K. Townsend, Recurrent acceleration in dilatonaxion cosmology, Phys. Rev. D 74 (2006) 103508,
hep-th/0608068;
J. Sonner, P.K. Townsend, Dilaton domain walls and dynamical systems, Class. Quantum Grav. 23 (2006) 441,
hep-th/0510115.
[10] K. Skenderis, P.K. Townsend, A. Van Proeyen, Domain-wall/cosmology correspondence in adS/dS supergravity,
JHEP 0708 (2007) 036, arXiv: 0704.3918 [hep-th].
[11] A. Vilenkin, Creation of universes from nothing, Phys. Lett. B 117 (1982) 25;
J.B. Hartle, S.W. Hawking, Wave function of the universe, Phys. Rev. D 28 (1983) 2960;
A. Vilenkin, Quantum creation of universes, Phys. Rev. D 30 (1984) 509;
R. Brustein, S.P. de Alwis, The landscape of string theory and the wave function of the universe, Phys. Rev. D 73
(2006) 046009, hep-th/0511093.
[12] C. Kounnas, H. Partouche, Instanton transition in thermal and moduli deformed de Sitter cosmology, arXiv:
0705.3206 [hep-th].
[13] N. Ohta, Accelerating cosmologies and inflation from M/superstring theories, Int. J. Mod. Phys. A 20 (2005) 1,
hep-th/0411230;
K.i. Maeda, N. Ohta, Inflation from superstring/M theory compactification with higher order corrections. I, Phys.
Rev. D 71 (2005) 063520, hep-th/0411093.
[14] M. Bouhmadi-Lopez, P. Vargas Moniz, Quantisation of parameters and the string landscape problem, JCAP 0705
(2007) 005, hep-th/0612149.
[15] N. Matsuo, Superstring thermodynamics and its application to cosmology, Z. Phys. C 36 (1987) 289.
[16] R.H. Brandenberger, C. Vafa, Superstrings in the early universe, Nucl. Phys. B 316 (1989) 391;
R.H. Brandenberger, String gas cosmology and structure formation: A brief review, hep-th/0702001.
[17] M. Grana, T.W. Grimm, H. Jockers, J. Louis, Soft supersymmetry breaking in CalabiYau orientifolds with D-branes
and fluxes, Nucl. Phys. B 690 (2004) 21, hep-th/0312232;
D. Lust, S. Reffert, S. Stieberger, Flux-induced soft supersymmetry breaking in chiral type IIb orientifolds with
D3/D7-branes, Nucl. Phys. B 706 (2005) 3, hep-th/0406092;
J.P. Derendinger, C. Kounnas, P.M. Petropoulos, F. Zwirner, Superpotentials in IIA compactifications with general
fluxes, Nucl. Phys. B 715 (2005) 211, hep-th/0411276;
L. Andrianopoli, M.A. Lledo, M. Trigiante, The ScherkSchwarz mechanism as a flux compactification with internal torsion, JHEP 0505 (2005) 051, hep-th/0502083;
G. DallAgata, N. Prezas, ScherkSchwarz reduction of M-theory on G2 -manifolds with fluxes, JHEP 0510 (2005)
103, hep-th/0509052.
[18] J. Scherk, J.H. Schwarz, Spontaneous breaking of supersymmetry through dimensional reduction, Phys. Lett. B 82
(1979) 60.
[19] R. Rohm, Spontaneous supersymmetry breaking in supersymmetric string theories, Nucl. Phys. B 237 (1984) 553.
[20] C. Kounnas, M. Porrati, Spontaneous supersymmetry breaking in string theory, Nucl. Phys. B 310 (1988) 355;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Superstrings with spontaneously broken supersymmetry and their
effective theories, Nucl. Phys. B 318 (1989) 75.
[21] J.J. Atick, E. Witten, The Hagedorn transition and the number of degrees of freedom of string theory, Nucl. Phys.
B 310 (1988) 291.
[22] C. Kounnas, B. Rostand, Coordinate dependent compactifications and discrete symmetries, Nucl. Phys. B 341
(1990) 641.
[23] I. Antoniadis, C. Kounnas, Superstring phase transition at high temperature, Phys. Lett. B 261 (1991) 369;
I. Antoniadis, J.P. Derendinger, C. Kounnas, Non-perturbative temperature instabilities in N = 4 strings, Nucl. Phys.
B 551 (1999) 41, hep-th/9902032.
[24] E. Witten, Dimensional reduction of superstring models, Phys. Lett. B 155 (1985) 151;
S. Ferrara, C. Kounnas, M. Porrati, General dimensional reduction of ten-dimensional supergravity and superstring,
Phys. Lett. B 181 (1986) 263;
M. Cvetic, J. Louis, B.A. Ovrut, A string calculation of the Khler potentials for moduli of ZN orbifolds, Phys.
Lett. B 206 (1988) 227;
L.J. Dixon, V. Kaplunovsky, J. Louis, On effective field theories describing (2, 2) vacua of the heterotic string, Nucl.
Phys. B 329 (1990) 27;
M. Cvetic, J. Molera, B.A. Ovrut, Khler potentials for matter scalars and moduli of ZN orbifolds, Phys. Rev. D 40
(1989) 1140.
[25] E. Cremmer, S. Ferrara, C. Kounnas, D.V. Nanopoulos, Naturally vanishing cosmological constant in N = 1 supergravity, Phys. Lett. B 133 (1983) 61;
178
J.R. Ellis, C. Kounnas, D.V. Nanopoulos, Phenomenological SU(1, 1) supergravity, Nucl. Phys. B 241 (1984) 406.
[26] E. Kiritsis, C. Kounnas, Perturbative and non-perturbative partial supersymmetry breaking: N = 4 N = 2
N = 1, Nucl. Phys. B 503 (1997) 117, hep-th/9703059.
[27] K.S. Narain, New heterotic string theories in uncompactified dimensions <10, Phys. Lett. B 169 (1986) 41;
K.S. Narain, M.H. Sarmadi, E. Witten, A note on toroidal compactification of heterotic string theory, Nucl. Phys.
B 279 (1987) 369.
[28] E. Kiritsis, String Theory in a Nutshell, Princeton Univ. Press, Princeton, USA, 2007, 588 p.
[29] E. Kiritsis, C. Kounnas, Curved four-dimensional spacetimes as infrared regulator in superstring theories, Nucl.
Phys. B (Proc. Suppl.) 41 (1995) 331, hep-th/9410212;
E. Kiritsis, C. Kounnas, Infrared regularization of superstring theory and the one loop calculation of coupling constants, Nucl. Phys. B 442 (1995) 472, hep-th/9501020.
[30] S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Effective super-Higgs and Str M 2 from four-dimensional strings,
Phys. Lett. B 194 (1987) 366.
[31] T. Catelin-Jullien, C. Kounnas, H. Partouche, N. Toumbas, in preparation.
[32] C. Kounnas, H. Partouche, N. Toumbas, J. Troost, in preparation.

Noncommutative D-branes from covariant

AdS superstring
Makoto Sakaguchi a , Kentaroh Yoshida b,
a Okayama Institute for Quantum Physics, 1-9-1 Kyoyama, Okayama 700-0015, Japan
b Theory Division, Institute of Particle and Nuclear Studies, High Energy Accelerator Research Organization (KEK),
Tsukuba, Ibaraki 305-0801, Japan

Received 26 December 2006; received in revised form 13 December 2007; accepted 31 December 2007
Abstract
We study noncommutative (NC) D-branes on AdS5 S5 from -invariance of covariant GreenSchwarz
action of an open string with a non-trivial world-volume flux. Finding boundary conditions to ensure the
-invariance, we can see possible configurations of the NC D-branes. With this method 1/4 BPS NC
D-branes are discussed. The resulting NC Dp-branes are 1/4 BPS at arbitrary position other than the p = 1
case. The exceptional D-string is 1/2 BPS at the origin and 1/4 BPS outside the origin. Those are reduced
to possible 1/4 BPS or 1/2 BPS AdS D-branes in the commutative limit. The same analysis is applied to
an open superstring in a pp-wave and leads to 1/4 BPS configurations of NC D-branes. These D-branes are
consistently obtained from AdS D-branes via the Penrose limit.
PACS: 11.25.Uv; 11.25.-w; 02.40.Gh
Keywords: D-branes; Noncommutative geometry; -symmetry
1. Introduction
Noncommutative (NC) geometry in string theory, which was initiated by Seiberg and Witten [1], has been one of the interesting subjects. When D-branes are considered in the presence
of a constant NSNS two-form B, a low-energy effective theory on the world-volume is realized
as a field theory on an NC space. The constant two-form flux may be taken as a magnetic flux
E-mail addresses: makoto_sakaguchi@pref.okayama.jp (M. Sakaguchi), kyoshida@post.kek.jp (K. Yoshida).

doi:10.1016/j.nuclphysb.2007.12.032
180
M. Sakaguchi, K. Yoshida / Nuclear Physics B 797 (2008) 179198
on the world-volume. D-branes with constant fluxes (i.e., gauge field condensates on the branes)
are often referred to as NC D-branes. NC D-branes in flat spacetime have been well studied (for
example, see [24]), but it seems not to be the case for those on curved backgrounds. Here we
are interested in AdS-branes with gauge field condensates, which are interesting objects to study
in the context of the AdS/CFT duality [5].
An AdS-brane gives a defect in the field theory side, and another field theory [6], which is
called a defect conformal field theory (dCFT), is realized on the defect [7]. When magnetic or
electric flux is turned on through the AdS-brane worldvolume, the flux should deform the dCFT
and it is interesting to study the deformed theory. One of the interesting issues is an integrable
open spin chain appearing as an anomalous dimension matrix of local composite operators in the
dCFT [8].1 When a flux is turned on, a boundary magnetic field would be turned on at each of
the end points of the spin chain. For this purpose, as the first step, we will study NC D-branes on
AdS5 S5 and pp-wave.2
In a series of our previous works we showed the power of -invariance of type IIB superstrings in a covariant classification of possible D-branes in AdS5 S5 [12,13] (for a short review
see [14]) and in the pp-wave [15].3 The -invariance of an open string requires that surface terms
under the -variation should be deleted by imposing some additional conditions [22], which
lead to the classification of the possible D-branes. This procedure has some advantages in comparison to the brane probe analysis with DiracBornInfeld (DBI) action on the AdS and the
pp-wave [18]. First, it is obvious that we do not need to analyze each of DBI actions of Dpbranes. All we have to consider is just a string action. Secondly, our procedure is covariant and
does not depend on the light-cone gauge. Finally we can show that the classification is valid
at full order of the fermionic variables [13]. This procedure is also applicable to Dirichlet
branes for an open supermembrane on AdS4/7 S7/4 background [23,24] and pp-wave [25,26],
and D-branes [27] of type IIA pp-wave string [28]. These results are consistently related via the
double dimensional reduction [29].
In this paper we study NC D-branes on AdS5 S5 by analyzing a covariant type IIB string
action in the GreenSchwarz formulation [30]. We apply the procedure developed in our previous
papers to the case with a constant two-form F composed of B and the field strength on the Dbrane. The boundary conditions to ensure that surface terms under the -variation vanish are
shown to lead to possible NC D-branes at the fourth order of . Then we find 1/4 BPS NC Dbranes. The D-branes other than D-string are 1/4 BPS anywhere, but the D-string is exceptional
and it is 1/2 BPS at the origin and 1/4 BPS outside the origin. All of them are reduced to possible
1/4 BPS or 1/2 BPS AdS D-branes in the commutative limit. The same analysis is applied also
to an open superstring in a pp-wave and derives configurations of 1/4 BPS NC D-branes. These
are consistently obtained from NC AdS D-branes via the Penrose limit [31].
Note that the result is obtained in the fourth order of but we may expect that it is still valid
even in the full order. The reason is that the higher-order contributions can be deleted by the
conditions obtained at the fourth-order level analysis. That was the case for the commutative
case actually [13]. Here we have not discussed the higher-order contributions but we naively
expect that it should be the case even for the present case. This would be discussed in another
place as a future work.
1 For other open spin chains, see [9] for D3D7O7 and [10] for a giant graviton.
2 NC D-branes in the pp-wave background are also discussed in [11].
3 For a classification of possible 1/2 BPS D-branes on a pp-wave with the light-cone gauge, see [1621].
181
This paper is organized as follows. In Section 2, we shall introduce a covariant action of

type IIB open string on AdS5 S5 background in the GreenSchwarz formulation, which includes a constant two-form F . Then we write down surface terms under the -variation of the
action, which play an important role in determining possible configuration of NC D-branes. In
Section 3, we discuss NC D-branes in flat spacetime. The well-known results obtained in the
different methods can be reproduced and this justifies the consistency of our method with the
others. In Section 4, we discuss 1/4 BPS NC D-branes on AdS5 S5 . In Section 5, after introducing a covariant action of a pp-wave superstring with a constant two-form, we find 1/4 BPS
NC D-branes in the pp-wave. Section 6 is devoted to a summary and discussions.
2. Open AdS superstring with a constant two-form
In this section we shall introduce the covariant action of type IIB string on AdS5 S5 [30],
including a constant two-form. The action has the -symmetry which ensures the consistency
of the theory. When we consider an open string rather than a closed string, the -variation of
the action gives surface terms and so we need to impose some boundary conditions in order to
preserve the -symmetry [22]. These conditions restrict possible configurations of D-branes on
AdS5 S5 . We will derive the surface terms here, and the boundary conditions will be considered
in Sections 3 and 4.
2.1. The action of AdS superstring
The action of type IIB superstring on a supergravity background is given by

S = d 2 [LNG + LWZ ],
(2.1)
where the NambuGoto (NG) part is4

LNG = g(X, ),
gij = EiM EjN GMN
= EiA EjB AB ,
and the WessZumino (WZ) part consists of
= i Z E A ,
M
the two parts5 :
EiA
= X ,
M
LWZ = L0WZ + L1WZ ,

1
L1WZ = 2i ij
A E j ,
dt E iA
(2.2)
(2.3)
(2.4)
(2.5)
1
1
A
B
j X N eN
FAB ,
L0WZ = ij eiA ejB FAB = ij i X M eM
(2.6)
2
2
where E = E|t and = 3 for an F-string while = 1 for a D-string. We will focus upon
a single F-string coupled to a constant F (the so-called dipole string) hereafter. The two-from
field strength F is defined by F = B da where a is a gauge field on the D-brane.
4 We follow the notation used in [12].
5 For alternative superstring actions in AdS S5 , see [3235] (see also [36]). For the brane charges in AdS and
5
pp-wave, see [32,33,3739]. For classical integrability of AdS superstring [40] in [35], see [41].
182
For the case of AdS5 S5 we are interested in, the supervielbein is given by [12]

2

sinh M
A
A
A sinh(M/2)
D,
E =
D ,
E = e + i
M/2
M
(2.7)
where we have introduced the following quantities:

2
1
M = i (A i2 ) A (AB ) AB i2 ,
2
1 AB
A
D = d + e A i2 + AB ,
2
4
A (a I, a J ),
AB (ab I, a b J ),
I = 01234 ,
J = 56789 .
(2.8)
Here e and are the vielbein and the spin connection, respectively, on AdS5 S5 . The parameter corresponds to the inverse of AdS5 radius (or S5 radius). When 0, the flat space is
realized.
In the presence of a non-trivial constant F along the D-brane world-volume (i.e., FA B = const
and others = 0), the boundary condition for Neumann directions is modified and becomes a
mixed boundary condition. The resulting boundary condition is as follows:
A
B
n X N e N
+ t X N eN
FB A = 0,
A
t X N eN = 0,
A i (i = 0, . . . , p) Neumann,
Ai (i = p + 1, . . . , 9) Dirichlet,
(2.9)
where t and n are derivatives on tangential and normal to the D-brane world-volume,
B)
are replaced
respectively. In the limit for a magnetic flux, F , Neumann directions (A,
by Dirichlet directions. On the other hand, for an electric flux F , no dimensional reduction of
the D-brane occurs because the range of the electric flux is restricted as 0 F 1.
2.2. The -variation of the open string action
According to our procedure explained in the introduction, let us examine the surface terms
under the -variation of the type IIB string action on AdS5 S5 up to the fourth order in . For
the case with F = 0, the vanishing conditions on the surface terms were examined in [12]. Note
that the NG part produces no surface term, and so the -variation surface terms emerge only from
the WZ part. Thus we will concentrate on the analysis of the WZ term below.
2
0
B
To extract the surface terms of the fourth order in from SWZ
d LWZ , we need
to know the expression of the -variation up to fourth order in . The -variation E A =
Z M E A = 0 in this background is given by [13]

M

A
A
= 1 + H + H 2 + B EB ,
X M eM
(2.10)
where
H
2

A sinh M/2
= i
[D]N eBN ,
M/2
It is easy to derive
EA
2

A sinh M/2
= i
.
M/2
(2.11)
i
A
X M e M
= i A A M2
12

1
A

B i2 + B CD CD B + O 6 .
2
4
183
(2.12)
B
By using this expression, we obtain the -variation surface term of SWZ

B
A
SWZ
= d t X M eM
FAB

i
i B B M2
12

B
CD E
C 1 B CD E
+ O 6 .
C i2
2
4
It is known that the -variation of the total action S gives surface terms only.
The surface terms under the -variation of the WZ term are given by

SWZ | = d L0 + L + Lspin ,
(2.13)
(2.14)
with

A
B
L0 = it X M eM
A + FAB

1
A t + A A t ,
+ A
2

M A
L = t X eM A C i2 + FAB B C i2 C
2

B
B A i2 + B B A i2
+
4

i
2
B
2
A M + FAB M ,
12

1 CD
A E
Lspin = E
t X M eM
A CD + FAB B CD
4

1 BC
D A
t X M eM
+ D
A BC + A A BC ,
8
(2.15)
(2.16)
(2.17)
where Lspin includes -dependent terms, L includes -dependent (but -independent)

terms and L0 includes -independent terms. These surface terms should vanish under some
conditions to ensure the -invariance. From the next section, we will examine the conditions
under which the surface terms vanish.
3. NC D-branes in flat spacetime
In this section we will examine the vanishing condition of L0 . This condition leads to the
classification of the possible NC D-branes in flat spacetime, since the AdS metric is reduced to
the flat spacetime metric when we consider the = 0 case.
NC D-branes in the GreenSchwarz formulation have already been studied before [24], but
our analysis here is covariant and so more general.
184
First, let us examine the first line of L0 , which is the second order in ,

A
it X M eM
A + FA B B

A
= it X M eM
(1 + F )A B 1 B 1 (1 F )A B 2 B 2 ,
(3.1)
where we have used the boundary condition (2.9). We should carefully consider the gluing conditions for the fermionic variable at boundaries in order to see that the surface terms vanish.
For this purpose we define the following matrix,
M g0
A 0 A p
[(p+1)/2]

gn + hn a 2n1 a 2n ,
(3.2)
n=1
and demand
2 = M 1 .
(3.3)
Because 1 and 2 are MajoranaWeyl spinors, p must be odd, p = 1, 3, 5, . . . . (a 2n1 , a 2n )

with n = 1, . . . , [ p+1
r =
a s if r = s. For odd p,
2 ] are pairs of Neumann directions such that a
{a 1 , . . . , a p+1 } is a permutation of {A 0 , . . . , A p }.
Let C be the charge conjugation matrix satisfying
A T
= C A C 1 ,
(3.4)
then one finds the following relation,
2 = 1 C 1 M T C = 1 M ,

p+1
M (1)p+1+[ 2 ] g0
gn hn a 2n1 a 2n A0 Ap .
(3.5)
(3.6)
n=1
If B {a 21 , a 2 }, then we find that

gn2 + h2n sn g2 h2 s + 2g h B C C ,
M B M = sg02
(3.7)
n=
where ( )a 21 a 2 = a 2 a 2 , ( )a 2 a 21 = a 21 a 21 and others are zeros. s and sk are defined
by s = ( A0 )2 ( Ap )2 and sk = ( a 2k1 )2 ( a 2k )2 . Hence the vanishing condition of the first

0
line of L is

(1 + F)A B = (1 F )A C sg02
(3.8)
gn2 + h2n sn g2 h2 s + 2g h C B .
n=
For the case without an electric F , since

0
Fa 21 a 2
0
F =
Fa 2 a 21
0
a 21 a 21
a 2 a 2
0
and sn = 1, (3.8) implies the following two equations:

1 = sg02
gn2 + h2n g2 h2 2g h F ,
n=
F = sg02
n=

gn2 + h2n 2g h g2 h2 F ,

= Fa 21 a 2
1 0
0 1

(3.9)
(3.10)
(3.11)
for 1 [ p+1
2 ]. These are solved by
g0 = s,
gn = cos n ,
hn = sin n ,
Fa 2n1 a 2n = tan n ,
185
(3.12)
where 0 n /2. Here we restricted the flux to be positive since we are interested in an Fstring, rather than an anti-F-string. We can argue the anti-F-string in the same way by considering
a negative flux. Note that we assume the positive flux hereafter.
For the case with an electric F , say Fa 1 a 2 and a 1 a 1 = a 2 a 2 = 1, since

0 1
0
Fa 1 a 2
1 =
(3.13)
,
= 1 Fa 1 a 2 ,
1 0
Fa 2 a 1
0
s1 = 1 and sn = 1 (n 2), (3.8) implies

1 = sg02
gn2 + h2n sn g2 h2 s 2g h F ,
n=
F = sg02
(3.14)

gn2 + h2n sn 2g h g2 h2 s F
(3.15)
n=
for
1 [ p+1
2 ].
g0 = s,
These are solved by
g1 = cosh 1 ,
gn = cos n ,
(3.16)
h1 = sinh 1 ,
hn = sin n ,
Fa 1 a 2 = tanh 1 ,
Fa 2n1 a 2n = tan n
where 0 1 and 0 n /2 (n 2).

In both cases, M is written as6
[(p+1)/2]

a 2n1 a 2n
M = s exp
n
A0 Ap .
(3.17)
(n 2),
(3.18)
(3.19)
n=1
It is clear from this expression that an NC Dp-brane is reduced to a commutative Dp-brane for
n 0. On the other hand, it is reduced to an NC D(p 2)-brane for a magnetic Fa 2n1 a 2n
(n /2), while remains to be an NC Dp-brane for an electric Fa 1 a 2 1 (1 ).
It is convenient to rewrite the boundary condition 2 = M 1 in the 2 2 matrix form as
follows:

1
0 M 1
= M =
(3.20)
,
M 2 = 1.
M
0
2
Here the matrix M is defined as

M = s exp
n a 2n1 a 2n 3 A0 Ap
(3.21)
n=1
and then the matrix is represented by 2 2 Pauli matrices, according to the value of p,

1
when p = 1 mod 4,
=
(3.22)
i2 when p = 3 mod 4.
6 For the case with an electric F
a a
a a
a 1 a 2 , we have used cosh 1 + 1 2 sinh 1 = exp(1 1 2 ). Because 1 =
1+
F
a a
arctanh Fa 1 a 2 = 12 log 1F 1 2 , we believe that the gluing matrix in the earlier literature is valid only for an electric F .
a 1 a 2
186
Table 1
The sequence of NC D-branes in flat spacetime
F
F 0
NC D9
C D9
NC D7
C D7
NC D5
C D5
NC D3
C D3
NC D1
C D1
C D(1)
The range of the parameter n depends on the type of the flux, namely magnetic or electric. For
the magnetic flux case n takes the value in the range of 0 n /2 and for the electric flux
case it takes in 0 n . As we have just seen above, the gluing matrix M satisfies

M A 3 + A 3 M + FA B M B + B M = 0,
(3.23)
or equivalently

M A 3 + FA B B = A 3 + FA B B M,
(3.24)
where M is defined as
M = C 1 M T C = M.
(3.25)
L0
vanishes.
It is obvious from (3.24) that the first line of
Then, by using (3.23), the second line of L0 , the fourth-order term in vanishes as follows:

1 A
A + A A t
2
A

1
B + B
A t = 0,
= FA B
2
where we have used
A
A = M
= A = 0.
(3.26)
(3.27)
In summary we have discussed boundary conditions and constructed the gluing matrix for
NC Dp-branes in flat spacetime. Taking a strong magnetic flux limit F on a Dp-brane,
an infinite number of D(p 2)-branes are dissolved on the Dp-branes. The commutative limit
F 0 leads to the standard commutative D-branes. The sequence of them via the two limits is
depicted in Table 1. All of these D-branes are 1/2 BPS. In the case of flat spacetime higher order
terms than the fourth order in do not appear and so our result is rigorous with respect to .
4. NC D-branes on AdS5 S5
In this section, we examine -dependent terms L and -dependent terms Lspin . These
terms include the parameter , which characterizes the AdS geometry. Hence the analysis here
is intrinsic to the AdS case and the boundary conditions lead to a classification of possible NC
AdS-branes.
It is straightforward to see L is rewritten with (3.23) as follows

C

A 3 + FA B B M C i2
2

+ B B 3 + FB C C M A i2
4
+ B 3 B A i2
4
i

A
+ A 3 + FA B B MM2 t X M eM
.
12
187
(4.1)
The first and the second lines in (4.1) vanish when

M C i2 = C i2 ,
(4.2)
which ensures that the third line vanishes as

B A i2 = B M A i2 = B A i2 = 0.
Among odd p, we find that this is satisfied for p = 1. Noting that

1 a 1 a 2 3 , C i2 = 0
(4.3)
(4.4)
is satisfied by (2, 0)- and (0, 2)-branes,7 we find that (4.2) is satisfied by the D1-branes. It is
straightforward to see that the fourth line vanishes for the D1-branes. As we will see soon, Lspin
vanishes for these branes sitting at the origin, and thus we find that NC D1-branes, (0, 2) and
(2, 0), sitting at the origin are 1/2 BPS.
To proceed further, we shall introduce an additional gluing matrix M 0 satisfying M 20 = 1,

s for p = 1 mod 4,
M 0 = 0 A0 Ap 1, =
(4.5)
s
for p = 3 mod 4,
and demand = M 0 . It satisfies
M 0 = C 1 M T0 C = (1)p+1+[
p+1
2 ]
M 0.
(4.6)
The gluing matrices, M and M 0 , commute each other, i.e., [M, M 0 ] = 0, and hence the branes
obtained below are 1/4 BPS. Here we should note that the most general 1/4 projections are
not always written in terms of mutually commuting gluing matrices only. Hence 1/4 BPS NC
D-branes we are considering is a part of possible 1/4 BPS D-branes.
By using the gluing condition = M 0 , we can derive the following equations:
0 C M 0 = 0,
C = M
when p = 3 mod 4, and

A 3 + FA B B M C i2 = 0,
(4.7)
(4.8)
when p = 1 (3) mod 4 and d, d = odd (even). Here d (d ) is the number of Dirichlet directions
in AdS5 (S5 ), respectively. It follows from these equations that the first and the second lines in
(4.1) vanish when one of the followings is satisfied
p = 3 mod 4,
p = 1 mod 4 and d = odd.
7 We use the notation for AdS Dp-brane like (m, n), where p + 1 = m + n and m (n) means the number of the Neumann
directions in AdS5 (S5 ).
188
On the other hand, by noting that

C 3 = 0 when p = 1 mod 4,
B A i2 = 0 when p = 1 (3) mod 4 and d, d = even (odd),
(4.9)
(4.10)
we find the third line vanishes when one of the followings is satisfied
p = 1 mod 4,
p = 3 mod 4 and d, d = odd.
In summary we find that the first, the second and the third lines vanish when d = odd and p =
1, 3 mod 4, namely we see that (even, even)-branes are possible.
Let us examine the fourth line. Noting that (4.7) and
B M A = 0 when A N (D), p = 1 (3) mod 4 and d, d = odd,
B MAB
= 0 when p = 1 mod 4,
B
(4.11)
(4.12)
MA B = MAB = 0 when p = 3 mod 4,
(4.13)
AB i2 = 0 when p = 1 (3) mod 4 and d, d = even (odd),

(4.14)
A
B
AB

i2 =
i2 = 0 when p = 1 (3) mod 4 and d, d = odd (even),
(4.15)
we find the fourth line vanishes for (even, even)-branes.
Next, we examine -dependent terms, Lspin , which are rewritten as

1 CD

A E
E t X M eM
A 3 + FA B B MCD
4

1 BC
E
M D A
D
MBC
t X eM A 3 + FA E
8
1 BC
D A
t X M eM
+ D
(4.16)
3 A BC .
8
Let us consider 1/2 BPS D1-branes, (2, 0)- and (0, 2)-branes, first. The vanishing condition for
the first and the second lines is
CD
CD
E
MCD = E CD
(4.17)
CD
which ensures that the third line vanishes. Since E
CD
= 0 8 and MC D = C D M for these
branes, the first and the second lines vanish if E = 0, i.e., if branes are sitting at the origin. As
BC
A B C = 0 for these branes, the last line vanishes if D = 0, i.e., if the branes are sitting at
the origin. Thus (2, 0)- and (0, 2)-branes sitting at the origin are 1/2 BPS.
8 The vielbein and spin connection in AdS are given in Appendix B in [12] as ea = sinh X ( a Xa Xm ) + Xa Xm
5
m
m
X
X2
X2
X
2
2
sinh
1
ab =
a b X b a ),
2 (X a b X b a ). It is straightforward to derive ab = em ab =
and m
(X
m
m
c
c m
c
c
X
2
2 cosh X
2
2

1
a b
b a
A
and similarly in S5 ca b = 2
(X c X c ). It is obvious from these expressions that the origin X = 0 is
cos X
2
a special point.
189
Table 2
1/4 BPS NC D-branes in AdS5 S5
D1
D3
D5
D7
D9
(2, 0), (0, 2)
(4, 0), (0, 4), (2, 2)
(4, 2), (2, 4)
(4, 4)
absent
Table 3
Commutative D-branes in AdS5 S5 sitting at the origin
SUSY
D(1)
D1
D3
D5
D7
D9
1/2
1/4
(0, 0)
(2, 0), (0, 2)
(3, 1), (1, 3)

(4, 0), (0, 4), (2, 2)
(4, 2), (2, 4)
(3, 5), (5, 3)

(4, 4)
absent
Next we consider 1/4 BPS D-branes. First we examine the first and the second lines. They
vanish when p = 3 mod 4 due to (4.7). For p = 1 mod 4, the terms proportional to CD vanish
CD
C D . It is natural to
due to (4.12). Noting that E = 0, we are left with the terms including E
expect that C D = 0 for FB C = 0 = FD E . This implies that (4.17) is satisfied for p = 1 mod 4.
Thus the first and the second lines vanish under the configuration even outside the origin. Next
we examine the last line. It vanishes for p = 1 mod 4 due to (4.9). Since
A BC
=0
when p = 3 mod 4
(4.18)

BC
and D = 0, we are left with terms proportional to B C . By the same reasoning above, we
derive

BC
BC
BC
D
A B C = D M A B C = D A B C M = 0,
(4.19)
vanishes under the

so that the last line vanishes even for p = 3 mod 4. We note that
configuration even outside the origin.
In summary, we find 1/4 BPS NC D-branes depicted in Table 2. (2, 0)- and (0, 2)-branes are
1/2 BPS when they are sitting at the origin, while 1/4 BPS when they move away from the
origin.
When F 0, 1/4 BPS NC Dp-branes with p = 3 mod 4 are reduced to 1/4 BPS commutative Dp-branes, while 1/4 BPS NC Dp-branes with p = 1 mod 4 become 1/2 BPS commutative
Dp-branes because surface terms vanish without M 0 in this limit [12]. We summarize commutative D-branes sitting at the origin in Table 3.
Lspin
4.1. Penrose limit

Now let us discuss the Penrose limit [31] of the above branes. The Penrose limit is taken
as follows [42]. First let us introduce the light-cone coordinates as X = 1 (X 9 X 0 ) and
= P with P = 12 + , and scale as
X+ 2 X+ ,
+ + .
(4.20)
Then the limit 0 is taken. We distinguish the cases depending on the boundary conditions
of the light-cone coordinate (X + , X ) as (N, N) for X Neumann directions and (D, D) for
X Dirichlet directions. We consider the Penrose limits of (N, N)- and (D, D)-cases. Then we
see that the NC AdS-branes obtained above are reduced to NC D-branes in the pp-wave as shown
in Table 4.
190
Table 4
Penrose limit of NC AdS-branes
DD
NN
DD
(4, 0) 1 (4, 0) 1
4
DD
NN
(2, 2) 1 (2, 2) 1 (+; 1, 1) 1

4
DD
NN
(2, 0) 1 (2, 0) 1
4
DD
NN
(4, 4) 1 (4, 4) 1 (+; 3, 3) 1

4
4
NN
DD
DD
NN
(0, 2) 1 (0, 2) 1
(4, 2) 1 (4, 2) 1 (+; 3, 1) 1

4
NN
(0, 4) 1 (0, 4) 1
DD
NN
(2, 4) 1 (2, 4) 1 (+; 1, 3) 1

4
In the next section, we will discuss NC D-branes in a pp-wave by studying the -invariance
of a covariant string action on the pp-wave. We will see that the possible NC D-branes are (+;
odd, odd)- and (even, even)-type configurations and all of them are consistently obtained in the
Penrose limit considered above.
5. NC D-branes on pp-wave
In this section we shall consider an open superstring on the maximally supersymmetric ppwave background. NC D-branes on the pp-wave have been well studied in [11]. It is however
done with the light-cone gauge, and as far as we know there is no covariant study of NC Dbranes on the pp-wave. Hence, by repeating the analysis in the case of the AdS superstring, we
shall classify possible NC D-branes on the pp-wave.
5.1. Open pp-wave superstring with a constant two-form
For the pp-wave background supervielbeins are given by [12]

2

i
sinh M
A
A
A sinh(M/2)
D,
E =
D .
E = e +
2
M/2
M
Here we have introduced several quantities:

2
+ i (m i2 ) m
M = i (f + g)i2
2
2

r
+ i (r + ) f i2
2

i (r + ) r gi2 i (mn ) mn i2 ,
2
2
2
D = d + e (f + g)i2 + em m i2 + em m + ,
2
2
2
mn = (rs + f, r s + g),
m = (r + f, r + g),
g = 5678 ,
2 m 2
dX ,
X
e+ = dX +
2
em = X m dX .
em = dX m ,
f = 1234 ,
e = dX ,
In this parametrization, the pp-wave metric becomes the standard form

2
2
2
2
ds 2 = 2e+ e + em = 2 dX + dX 2 X m dX + dX m .
The WZ action in the case of the pp-wave is given by

SWZ = d 2 L0WZ + L1WZ
191
(5.1)
with
A
B
L0WZ = ij i X M eM
j X N eN
FAB ,

i
A
A
A Dj i X M eM
A M2 Dj
L1WZ = ij ii X M eM
12

1 A
+ Di A Dj + O 6 .
4
(5.2)
(5.3)
The universal feature of -variation is

A
E A = X M EM
+ EA = 0
(5.4)
and for the pp-wave case it can be rewritten, at the fourth order of , as

i
i
1
A
A B + spin + O 6 ,
X M eM
= A A M2 B
B
2
24
4
(5.5)
spin
where A and A are defined, respectively, by

M
m
A = eA eM (f + g)i2 + eM m i2 ,
2
2
spin
A
M
= eA

2
m
eM m + .
2
(5.6)
It is straightforward to see that
2 m
spin
(5.7)
(f + g)i2 ,
X m + ,
m = m i2 ,
=
2
2
2
and the others vanish.
Under the -variation, SNG produce no surface term. The surface terms under the -variation
of SWZ are obtained as

SWZ | = d L0 + L + Lspin
(5.8)
=
with

A
B X M eM
L0 = i A + FAB

1
A + A A ,
+ A
4

1
L = C A C + FAB B C
2

1 B
+ B A + B B A
4

i
A
A M2 + FAB B M2 X M eM
,
12
(5.9)
(5.10)
192

1 C
A spin + FAB B spin
Lspin =
C
C
2
1 B
spin
spin
A
+ B A + B B A X M eM
.
4
(5.11)
In the next subsection we will examine these surface terms and determine boundary conditions
under which these vanish.
5.2. 1/4 BPS NC D-branes on pp-wave
In order to determine the boundary conditions for the fermionic variable , we shall introduce
the gluing matrix M given in (3.21) and demand that = M in the same way as in the AdS
case. Since it satisfies (3.23), L0 vanishes.
By using (3.23), L can be rewritten as

1 C

A + FA B B MC
2

1

B B + FB C C MA
4
1
+ B 3 B A
4

i
B
A
2
,
+ A + FA B MM X M eM
(5.12)
12
where we have used

1
B = M B + B M = 0.
2
The vanishing condition for the first and the second lines is
(5.13)
MC = C M,
(5.14)
which ensures that the third line vanishes. (5.14) means

Mm = m M
(5.15)
for D, while for N , (5.15) and

M = M.
(5.16)
Among other p, p = 1 is the special case. For p = 1 and D, (5.15) is satisfied when
n a 2n1 a 2n f = f n a 2n1 a 2n
and d = even,
(5.17)
(d )
where d
is the number of Dirichlet directions contained in {1, 2, 3, 4} ({5, 6, 7, 8}), respectively. This means that the condition is satisfied by (2, 0)- and (0, 2)-branes. On the other hand,
for N , (5.16) implies that
n a 2n1 a 2n (f + g) = (f + g)n a 2n1 a 2n
and
d = even,
(5.18)
which is not satisfied by the (+)-brane. Thus we find that for (0, 2)- and (2, 0)-branes the first
and the second lines vanish and so the third line vanishes. It is straightforward to see that the last
193
line vanishes for these branes. As will be seen below, Lspin vanishes for these branes sitting
even outside the origin. Thus we find that (0, 2)- and (2, 0)-branes sitting anywhere are 1/2 BPS.
For 1/4 BPS branes,9 we introduce an additional gluing matrix M 0 given in (4.5) which
commutes with M, [M, M 0 ] = 0, and satisfies
M 0 = C 1 M T0 C = (1)p+1+[
p+1
2 ]
M 0.
(5.19)
First we examine the first and the second lines in (5.12). By using the boundary condition =
M 0 ,
B = 0 for p = 3 mod 4
(5.20)
is derived and so they vanish for p = 3 mod 4. For p = 1 mod 4, since
B MC = M 0 B MC = B MM 0 C
(5.21)
and

(1)d f + (1)d g i2 M 0 ,
2

d
d
M 0 m = (1) r M 0 , (1) r M 0 ,
0
M 0 =
D,
N,
(5.22)
they vanish when one of the followings is satisfied

p = 3 mod 4,
p = 1 mod 4, D and d = even,
p = 1 mod 4, N and d = odd.
Next, we examine the third line. It vanishes when p = 1 mod 4 because
B 3 = 0 for p = 1 mod 4.
(5.23)
For p = 3 mod 4, since

B A = B M 0 A
(5.24)
and (5.22), the third line vanishes when one of the followings is satisfied
p = 1 mod 4,
p = 3 mod 4, D and d = even,
p = 3 mod 4, N and d = odd.
In summary, we find that the first, second and third lines vanish for (+; odd, odd)-branes
and (even, even)-branes. It is straightforward to see that the fourth line vanishes for these branes.
Next we examine Lspin , which is rewritten by using (3.23) as
9 We should note that 1/4 BPS branes we consider here is a part of general 1/4 BPS branes. The 1/4 projectors here
are composed of two of mutually commuting gluing matrices only.
194
Table 5
NC D-branes in the pp-wave
SUSY
D1
1/2
1/4
(0, 2), (2, 0)
D3
D5
D7
(4, 0), (2, 2), (0, 4)

(+; 1, 1)
(4, 2), (2, 4)

(+; 3, 1), (+; 1, 3)
(4, 4)
(+; 3, 3)
D9
absent

1 C

spin
A 3 + FA B B MC
2

1

spin
B B 3 + FC C MA
4
1
A
B spin
+ B 3 A X M eM
.
4
(5.25)
m+
In the presence of F on the D-brane world-volume, the spin connection dX =
dX
AB dX . Even in this case, it is natural to expect that

may develop additional components
spin
n a 2n1 a 2n , AB A B = 0
(5.26)
spin
as in the AdS case. For D, it vanishes as C = 0 only for C = , so we consider the case
with N below. The first and the second lines vanish for p = 3 mod 4 due to (5.20). Even for
p = 1 mod 4, they vanish since we derive

A 3 + FA B B C D C D = A 3 + FA B B C D M C D = 0,
(5.27)
and

2
2

0 A 3 + FA B B M X m m +
A 3 + FA B B M X m m + = M
2
2
= 0.
(5.28)
Next we examine the last line, which vanishes when p = 1 mod 4 due to (5.23). For p = 3 mod 4,
it vanishes since we derive assuming (5.26)

B C D M C D = 0,
B C D C D =
(5.29)
and
B
2 m
2
X m + = B M 0 X m m + = 0.
2
2
(5.30)
In summary, Lspin vanishes for (+; odd, odd)-branes and (even, even)-branes even if they
are sitting outside the origin. Finally, it is obvious that (0, 2)- and (2, 0)-branes are 1/2 BPS
even off the origin, because D for these branes. As a result, we have obtained NC D-branes
summarized in Table 5.
In the commutative limit F 0, NC D-branes above are reduced to commutative D-branes.
Combining the result obtained in the flat limit with that obtained in [12], we summarize commutative D-branes sitting at the origin of the pp-wave in Table 6.
195
Table 6
Commutative D-branes in the pp-wave (sitting at the origin)
SUSY
D(1)
D1
D3
D5
1/2
(0, 0)
(0, 2), (2, 0)
(1, 3), (3, 1)

(+; 0, 2)
(+; 2, 0)
(4, 0), (2, 2), (0, 4)
(+; 1, 1)
(2, 4), (4, 2)

(+; 1, 3)
(+; 3, 1)
1/4
D7
D9
absent
(+; 2, 4)
(+; 4, 2)
(4, 4)
(+; 3, 3)
6. Summary and discussions

We have considered some possible configurations of 1/4 BPS NC D-branes on AdS5 S5 and
pp-wave. The 1/4 BPS NC AdS-branes are allowed to exist at arbitrary position in the spacetime.
The D-string case is exceptional and it is 1/2 BPS at the origin and 1/4 BPS outside the origin.
We also have seen that all of them are reduced to 1/2 BPS and 1/4 BPS commutative AdSbranes in the commutative limit. The 1/4 BPS NC D-branes in the pp-wave background have
also been classified by applying the same analysis as in the AdS case to the pp-wave case. The
resulting possible D-branes are consistently reproduced from the NC AdS-branes via the Penrose
limit.
It would be possible to see that the result shown in this paper is still valid at full order in
by following our previous paper [13], though the possible NC D-branes have been classified at
fourth order in . This issue is more complicated and so we leave it as a future problem. We hope
that we could report on this issue in the near future.
It is also interesting to consider the interpretation of our results before taking the near-horizon
limit. According to the work of Skenderis and Taylor [18], an AdS-brane is surely related to a
supersymmetric intersection of D-brane with a stack of N D3-branes. Hence our result may also
be interpreted in terms of the intersecting D-branes. In particular, the T-dual picture would be
related to an intersecting D-branes at angles [43]. For this direction, it would also be useful to
generalize the work [18,44] by including a constant two-form. There the relation to dCFTs is
also discussed. As a further generalization, it would be interesting to consider an intersecting
AdS D-branes (for an intersecting D-branes on a pp-wave, see [45]), though we have discussed
a single AdS D-brane here.
As another generalization of our works, it would be interesting to consider oblique Dbranes [46] with gauge field condensates in the pp-wave background as discussed in [47,48].
It would also be nice to study the AdS origin of the oblique D-branes by following [49].
As an interesting application of our procedure, we can consider NC M-branes. For NC Mbranes, the classification of the possible Dirichlet branes is drastically modified and one can see
some interesting features. For the flat case see [50] and we will report for the AdS case in the
near future [51].
Acknowledgements
The authors thank Y. Hikida and Y. Susaki for useful discussions. This work is supported
in part by the Grant-in-Aid for Scientific Research (Nos. 17540262 and 17540091) from the
Ministry of Education, Science and Culture, Japan. The work of K.Y. is supported in part by
JSPS Research Fellowships for Young Scientists.
196
References
[1] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[2] C.S. Chu, P.M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hep-th/9812219.
[3] C.S. Chu, F. Zamora, Manifest supersymmetry in non-commutative geometry, JHEP 0002 (2000) 022, hep-th/
9912153.
[4] I. Rudychev, From noncommutative string/membrane to ordinary ones, JHEP 0104 (2001) 015, hep-th/0101039.
[5] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
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;
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[6] A. Karch, L. Randall, Localized gravity in string theory, Phys. Rev. Lett. 87 (2001) 061601, hep-th/0105108;
A. Karch, L. Randall, Open and closed string interpretation of SUSY CFTs on branes with boundaries, JHEP 0106
(2001) 063, hep-th/0105132.
[7] O. DeWolfe, D.Z. Freedman, H. Ooguri, Holography and defect conformal field theories, Phys. Rev. D 66 (2002)
025009, hep-th/0111135;
J. Erdmenger, Z. Guralnik, I. Kirsch, Four-dimensional superconformal theories with interacting boundaries or
defects, Phys. Rev. D 66 (2002) 025020, hep-th/0203020.
[8] O. DeWolfe, N. Mann, Integrable open spin chains in defect conformal field theory, JHEP 0404 (2004) 035, hep-th/
0401041;
Y. Susaki, Y. Takayama, K. Yoshida, Open semiclassical strings and long defect operators in AdS/dCFT correspondence, Phys. Rev. D 71 (2005) 126006, hep-th/0410139;
Y. Susaki, Y. Takayama, K. Yoshida, Integrability and higher loops in AdS/dCFT correspondence, Phys. Lett. B 624
(2005) 115, hep-th/0504209;
K. Okamura, Y. Takayama, K. Yoshida, Open spinning strings and AdS/dCFT duality, JHEP 0601 (2006) 112,
hep-th/0511139;
T. McLoughlin, I.J. Swanson, Open string integrability and AdS/CFT, Nucl. Phys. B 723 (2005) 132, hep-th/
0504203.
[9] B. Chen, X.J. Wang, Y.S. Wu, Integrable open spin chain in super-YangMills and the plane-wave/SYM duality,
JHEP 0402 (2004) 029, hep-th/0401016;
B. Chen, X.J. Wang, Y.S. Wu, Open spin chain and open spinning string, Phys. Lett. B 591 (2004) 170, hep-th/
0403004;
T.G. Erler, N. Mann, Integrable open spin chains and the doubling trick in N = 2 SYM with fundamental matter,
JHEP 0601 (2006) 131, hep-th/0508064.
[10] D. Berenstein, S.E. Vazquez, Integrable open spin chains from giant gravitons, JHEP 0506 (2005) 059, hep-th/
0501078;
D. Berenstein, D.H. Correa, S.E. Vazquez, Quantizing open spin chains with variable length: An example from giant
gravitons, Phys. Rev. Lett. 95 (2005) 191601, hep-th/0502172;
A. Agarwal, Open spin chains in super-YangMills at higher loops: Some potential problems with integrability,
hep-th/0603067.
[11] C.S. Chu, P.M. Ho, Noncommutative D-brane and open string in pp-wave background with B-field, Nucl. Phys.
B 636 (2002) 141, hep-th/0203186.
[12] M. Sakaguchi, K. Yoshida, D-branes of covariant AdS superstrings, Nucl. Phys. B 684 (2004) 100, hep-th/0310228.
[13] M. Sakaguchi, K. Yoshida, Notes on D-branes of type IIB string on AdS5 S5 , Phys. Lett. B 591 (2004) 318,
hep-th/0403243.
[14] M. Sakaguchi, K. Yoshida, D-branes of covariant AdS superstrings: An overview, hep-th/0408208.
[15] P. Bain, K. Peeters, M. Zamaklar, D-branes in a plane wave from covariant open strings, Phys. Rev. D 67 (2003)
066001, hep-th/0208038.
[16] A. Dabholkar, S. Parvizi, Dp branes in pp-wave background, Nucl. Phys. B 641 (2002) 223, hep-th/0203231.
[17] M. Billo, I. Pesando, Boundary states for GS superstrings in an Hpp wave background, Phys. Lett. B 536 (2002)
121, hep-th/0203028.
[18] K. Skenderis, M. Taylor, Branes in AdS and pp-wave spacetimes, JHEP 0206 (2002) 025, hep-th/0204054.
[19] P. Bain, P. Meessen, M. Zamaklar, Supergravity solutions for D-branes in Hpp-wave backgrounds, Class. Quantum
Grav. 20 (2003) 913, hep-th/0205106.
197
[20] O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background, JHEP 0303
(2003) 002, hep-th/0205183;
M.R. Gaberdiel, M.B. Green, The D-instanton and other supersymmetric D-branes in IIB plane-wave string theory,
Ann. Phys. 307 (2003) 147, hep-th/0211122.
[21] K. Skenderis, M. Taylor, Open strings in the plane wave background. I: Quantization and symmetries, Nucl. Phys.
B 665 (2003) 3, hep-th/0211011;
K. Skenderis, M. Taylor, Open strings in the plane wave background. II: Superalgebras and spectra, JHEP 0307
(2003) 006, hep-th/0212184;
K. Skenderis, M. Taylor, An overview of branes in the plane wave background, Class. Quantum Grav. 20 (2003)
S567, hep-th/0301221;
D.Z. Freedman, K. Skenderis, M. Taylor, Worldvolume supersymmetries for branes in plane waves, hep-th/0306046.
[22] N.D. Lambert, P.C. West, D-branes in the GreenSchwarz formalism, Phys. Lett. B 459 (1999) 515, hep-th/9905031.
[23] M. Sakaguchi, K. Yoshida, Dirichlet branes of the covariant open supermembrane in AdS4 S7 and AdS7 S4 ,
Nucl. Phys. B 681 (2004) 137, hep-th/0310035.
[24] M. Sakaguchi, K. Yoshida, Open M-branes on AdS4/7 S7/4 revisited, Nucl. Phys. B 714 (2005) 51, hep-th/
0405109.
[25] M. Sakaguchi, K. Yoshida, Dirichlet branes of the covariant open supermembrane on a pp-wave background, Nucl.
Phys. B 676 (2004) 311, hep-th/0306213.
[26] K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, Nucl. Phys. B 644 (2002) 113, hep-th/
0206070.
[27] S. Hyun, J. Park, H. Shin, Covariant description of D-branes in IIA plane-wave background, Phys. Lett. B 559
(2003) 80, hep-th/0212343.
[28] K. Sugiyama, K. Yoshida, Type IIA string and matrix string on pp-wave, Nucl. Phys. B 644 (2002) 128, hep-th/
0208029;
S. Hyun, H. Shin, N = (4, 4) type IIA string theory on pp-wave background, JHEP 0210 (2002) 070, hep-th/
0208074.
[29] 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.
[30] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S5 background, Nucl. Phys. B 533 (1998) 109,
hep-th/9805028.
[31] R. Penrose, Any spacetime has a plane wave as a limit, in: Differential Geometry and Relativity, Reidel, Dordrecht,
1976, pp. 271275;
R. Gueven, Plane wave limits and T-duality, Phys. Lett. B 482 (2000) 255, hep-th/0005061;
M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry, Class. Quantum Grav. 19 (2002) L87, hep-th/0201081.
[32] M. Hatsuda, K. Kamimura, M. Sakaguchi, Nondegenerate super-anti-de Sitter algebra and a superstring action,
Phys. Rev. D 62 (2000) 105024, hep-th/0007009.
[33] M. Hatsuda, M. Sakaguchi, WessZumino term for AdS superstring, Phys. Rev. D 66 (2002) 045020, hep-th/
0205092;
M. Hatsuda, M. Sakaguchi, WessZumino term for the AdS superstring and generalized InoenueWigner contraction, Prog. Theor. Phys. 109 (2003) 853, hep-th/0106114.
[34] A.M. Polyakov, Conformal fixed points of unidentified gauge theories, Mod. Phys. Lett. A 19 (2004) 1649, hep-th/
0405106.
[35] R. Roiban, W. Siegel, Superstrings on AdS5 S5 supertwistor space, JHEP 0011 (2000) 024, hep-th/0010104.
[36] N. Berkovits, C. Vafa, E. Witten, Conformal field theory of AdS background with RamondRamond flux,
JHEP 9903 (1999) 018, hep-th/9902098;
N. Berkovits, M. Bershadsky, T. Hauer, S. Zhukov, B. Zwiebach, Superstring theory on AdS2 S2 as a coset
supermanifold, Nucl. Phys. B 567 (2000) 61, hep-th/9907200.
[37] K. Sugiyama, K. Yoshida, Supermembrane on the pp-wave background, Nucl. Phys. B 644 (2002) 113, hep-th/
0206070.
[38] K. Kamimura, M. Sakaguchi, osp(1|32) and extensions of super-AdS5 S5 algebra, Nucl. Phys. B 662 (2003) 491,
hep-th/0301083.
[39] K. Peeters, M. Zamaklar, Anti-de Sitter vacua require fermionic brane charges, Phys. Rev. D 69 (2004) 066009,
hep-th/0311110;
P. Meessen, K. Peeters, M. Zamaklar, On nonperturbative extensions of anti-de Sitter algebras, hep-th/0302198.
[40] I. Bena, J. Polchinski, R. Roiban, Hidden symmetries of the AdS5 S5 superstring, Phys. Rev. D 69 (2004) 046002,
hep-th/0305116.
198
[41] M. Hatsuda, K. Yoshida, Classical integrability and super-Yangian of superstring on AdS5 S5 , Adv. Theor. Math.
Phys. 9 (2005) 703, hep-th/0407044.
[42] M. Hatsuda, K. Kamimura, M. Sakaguchi, From super-AdS5 S5 algebra to super-pp-wave algebra, Nucl. Phys.
B 632 (2002) 114, hep-th/0202190;
M. Hatsuda, K. Kamimura, M. Sakaguchi, Super-pp-wave algebra from super-AdS S algebras in eleven-dimensions, Nucl. Phys. B 637 (2002) 168, hep-th/0204002.
[43] See for example, N. Ohta, P.K. Townsend, Supersymmetry of M-branes at angles, Phys. Lett. B 418 (1998) 77,
hep-th/9710129, and references thereof.
[44] D. Arean, A.V. Ramallo, Open string modes at brane intersections, hep-th/0602174.
[45] N. Ohta, K.L. Panigrahi, S. Siwach, Intersecting branes in pp-wave spacetime, Nucl. Phys. B 674 (2003) 306,
hep-th/0306186.
[46] Y. Hikida, S. Yamaguchi, D-branes in pp-waves and massive theories on world-sheet with boundary, JHEP 0301
(2003) 072, hep-th/0210262;
M.R. Gaberdiel, M.B. Green, S. Schafer-Nameki, A. Sinha, Oblique and curved D-branes in IIB plane-wave string
theory, JHEP 0310 (2003) 052, hep-th/0306056.
[47] T. Mattik, Branes in the plane wave background with gauge field condensates, JHEP 0506 (2005) 041, hep-th/
0501088.
[48] B.H. Lee, J.w. Lee, C. Park, H.S. Yang, More on supersymmetric D-branes in type IIB plane wave background,
JHEP 0601 (2006) 015, hep-th/0506091.
[49] G. Sarkissian, M. Zamaklar, Diagonal D-branes in product spaces and their Penrose limits, JHEP 0403 (2004) 005,
hep-th/0308174.
[50] M. Sakaguchi, K. Yoshida, Noncommutative M-branes from covariant open supermembranes, Phys. Lett. B 642
(2006) 400, hep-th/0608099.
[51] M. Sakaguchi, K. Yoshida, Noncommutative M-branes in AdS and pp-wave, in preparation.

De Sitter space and eternity

A.M. Polyakov
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
Received 28 December 2007; accepted 7 January 2008
Abstract
This paper explores infrared quantum effects in the de Sitter space. The notion of eternal manifolds is
introduced and it is shown that in most cases the de Sitter space does not belong to this class. It is unstable
under small perturbations which may cause a breakdown of the de Sitter symmetry. The de Sitter string
sigma model is discussed. It is argued that the gauge theory at the complex coupling is dual to the matrix
elements of vertex operators in the de Sitter space, taken between the BunchDavies vacuum and the out
state without particles. The described infrared effects are likely to screen away the cosmological constant.
PACS: 11.25.-w
1. Introduction
Many years ago I conjectured [1] that the cosmological constant may be screened by the
infrared fluctuations of the metric, much like the electric charge in quantum electrodynamics.
This effect, if it exists, must be non-perturbative, related to the fluctuations of the metric g
near zero, and not near a classical background value. In the time-dependent picture the screening
is equivalent to the instability of spacetimes with constant curvature. In this picture the initially present curvature is gradually decaying. It is important, therefore to find out, whether the
de Sitter space carries the infrared seeds of its own destruction. This is the topic of the present
article.
It is not straightforward to find an appropriate framework for discussing this question. The
notion of fields and particles in the curved Lorentzian backgrounds is ambiguous and the Hamiltonian may not exist at all. We have to formulate a principle which describes quantum field
theories in these circumstances. With this goal in mind we will postulate that a free particle on
E-mail address: polyakov@princeton.edu.
doi:10.1016/j.nuclphysb.2008.01.002
200
A.M. Polyakov / Nuclear Physics B 797 (2008) 199217
a stable manifold must propagate with an amplitude

G++ (x, x ) =
eimL(Pxx )
(1)
(Pxx )
where the sum goes over all paths belonging to the manifold and connecting the points x and x ,
L(P ) is the length of the path, m is the mass of the particle and the subscripts will be explained
later. When we consider interacting particles, the Feynman prescription is to draw the diagrams
and to integrate the position of the vertices over the whole spacetime.
That seems to contradict causality, since the observables at a given moment of time seem to
depend on the interaction in the future. The resolution of this puzzle is well known. The use of
Feynmans diagrams presupposes that the vacuum is stable, that nothing new will ever happen.
This assumption restores the causality and leads to CPT symmetry. In this case the manifold is
eternal and the future is being prefixed. We define (perturbatively) eternal manifolds as the
ones for which the Feynman rules are valid and the vacuum loops do not contain an imaginary
parts.
If, however, the vacuum is unstable, the picture is quite different. To calculate the present we
must eliminate all the signals from the future. The way to do it was discovered by Schwinger [3]
and developed by a number of people [4]. According to [3], we must double the manifold and
consider the two copies (+/) with the chronological ordering from to on the first sheet
and back on the second. The vertices of Feynmans diagrams are now labeled by their position x
and an Ising variable = +/ indicating the choice of the sheet.
The eternity test is the condition that all diagrams in which a particle changes the sheet (thus
containing the G+ Green function) are zero. A slightly different way to put it is to require that
all diagrams containing spidersa () point surrounded by (+)s onlyare zero. A spider
represents spontaneous particle creation from the vacuum. If this is forbidden we return to the
ordinary Feynmans diagrams. We will explore this condition below and find that in the de Sitter
space in the certain cases it is violated. The next step will be to understand this from the point of
view of gauge/strings duality.
If de Sitter space turns out to be intrinsically unstable, it must be described by a non-unitary
conformal field theory. A natural candidate for such a theory would be a YangMills theory with
complex coupling obtained by the analytic continuation from negative to positive curvature. We
also should expect to find a description in terms of the world-sheet sigma model with the de Sitter
space as a target.
These statements are puzzling since the de Sitter space has been investigated in many works
and no instability of the so-called BunchDavies vacuum was ever found (see [9,13] for the
excellent reviews). My assertion here is that neither this nor any other vacuum passes the eternity test. It is perhaps appropriate to explain the meaning of this test, leaving technical details
(discussed below).
Let us consider the BunchDavies vacuum for a scalar field . This field in the de Sitter
space with the standard metric ds 2 = cosh2 t d 2 dt 2 can be expanded in partial waves. Each
partial wave m (m being angular momentum, see below) can be further expanded as m =

m , where s are the appropriate solutions of the KleinGordon equation (27)
am m (t) + am
and a, a are the creation and annihilation operators. The BunchDavies (aka HartleHawking,
aka Euclidean) vacuum E is defined as a|E = 0. There are also two other vacua |, defined
with the modes without positive frequencies in the future or negative in the past (we will also
call them in and out states when it is not ambiguous). These are the states without particles
at the past/future infinity respectively. They have the feature |E|| < 1, meaning that in the
201
Euclidean vacuum particles are created. By itself this is not yet an instability. One might think that
the Euclidean vacuum is eternal, but viewed from infinity, contains particles. Until we include
interactions this is a possible point of view.
Let us now take, say, the 4 interaction and turn it on adiabatically and then turn it off in
the same way. In the usual field theory the vacuum will acquire a phase after this procedure
but otherwise will remain unchanged.
Not so in the de Sitter space. From the formulae given
below it follows that E|T exp(i (t) 4 (dx))|E 0 as 0. In other words, with any
interaction |Eout is orthogonal to |Ein . Technically this adiabatic catastrophe occurs since
we can evaluate the above matrix element by the usual Feynman diagrams, and, as we will see,
they contain very strong infrared divergences.
One can try to study instabilities in a different way, by looking at the in/in matrix elements
and using the Schwinger prescriptions. This is a kind of a laissez-faire approach when we let
the system develop by itself. In this framework the instability means that small perturbations
breaking the de Sitter symmetry grow with time, while the divergences, requiring the IR cutoff (like in the above formula) go away. It is reasonable to expect that as a result of these
infrared phenomena, a small perturbations of the metric and other fields will be amplified with
time, destroying the de Sitter symmetry. This can be viewed as a kind of spontaneous symmetry
breaking (with large value of time playing the role of the large volume in statistical mechanics).
We must stress, however, that there is still no complete proof of the instability in the Schwinger
framework. It remains to be proved that in this case the vacuum polarization will lead to some
kind of gravitational instability. Below we will give some arguments in favour of this statement.
A closely related phenomenon happens in the Starobinsky model [10] where the instability occurs
due to the trace anomaly.
Another limitation of the present work is that it treats only interacting scalar fields and leaves
aside the tensor structure and gauge fixing of the gravitational oscillations. Hopefully the qualitative effects we are after will still be present in the full theory. The dS/CFT correspondence has
been discussed in the papers [2,8,11]. While the approaches of [2,8] are quite different from the
present paper, the ideas of [11] go in the same direction. Still, our conclusions below seem new.
Many formulae for the various propagators used below must be well known to the experts in the
field. Infrared divergences have also been investigated before [5,7], but we add some new results
and approaches. The relation between infrared divergences and stability and the relation to the
non-unitary gauge theory was discussed in [12] on a qualitative level; see also a recent discussion
of the Euclidean infrared problems [6]. The particle production, instabilities and back reaction
have been discussed from the various points of view in the papers [1416] and references therein.
A runaway particle production was studied in an interesting paper [17].
2. The composition principle and the Unruh detector
The formula (1) implies the asymptotic condition which uniquely fixes the propagator
G(x, x ) eimL(x,x )
L(x, x )
(2)
where
being the geodesic time-like length. Another way of putting it is to select
the Green function of the KleinGordon equation which, for the time-like separations is analytic
in mass if Im m < 0.
It is important to realize that the generic propagator would contain a superposition AeimL +
BeimL . This propagator would violate the following composition principle which lies at the
foundation of quantum field theory. We should be able to glue together two paths and obtain
202
a single one. This condition is equivalent to unitarity in the ordinary quantum field theory. In the
classical limit this gluing procedure is expressed by the relation L(x, x ) = L(x, y) + L(y, x )
where y must extremize this expression. In the quantum case one has to account for the entropy
on the world line (or the world sheet in the case of strings) and the relation is

dy G(x, y)G(y, x ) =
(3)
L(P )eimL(P )
G(x, x ).
m
(Pxx )
As a side remark, let us notice that in the case of strings the extra factor L in this formula is
replaced by the free energy of the self-avoiding path belonging to the surface. These gluing
relations in the flat space play the role of unitarity conditions for the corresponding field theory.
It seems reasonable to postulate that the right propagators must respect these conditions in
the general case. If the two exponents are present, we would get in the asymptotic the term
L(x,
x ) = L(x, y) L(y, x ), which, being minimized with respect to y, does not reproduce the
left-hand side.
The above condition has a curious relation to the behaviour of the Unruh detector. As well
known [13], the probability for absorbing energy by this detector is given by

w() ds1 ds2 ei(s1 s2 ) G x(s1 ), x(s2 )
where x(s) is a trajectory of the detector and s is its proper time. Let us assume that the trajectory is a geodesic. In this case G Aeim(s1 s2 ) + Beim(s1 s2 ) . We see that the B coefficient,
which violates the composition, simultaneously defines the absorption probability of the inertial
detector. If it is non-zero, the inertial detector can borrow energy from the vacuum indefinitely!
It is not surprising that such a vacuum will not live long.
Let us now turn to the de Sitter space. The standard approach uses the propagators obtained
by analytic continuation from the sphere (which define the BunchDavies or Euclidean vacuum).
Let us have a look at them, using for simplicity the 2d sphere. The Green function is given by
G(n, n ) =

(2l + 1)Pl (n n )
l=0
l(l
+ 1) + M 2
P (n n ),
sin
(4)
here n is a unit vector, is defined by a relation M 2 = ( + 1) and the above formula is

obvious since the left-hand side has the same poles and residues as the right one. It also has the
expected logarithmic singularity at the coinciding points (n n ) = 1. On a sphere it satisfies the
composition principle

1
1
(5)
dn1 P (n n1 )(n1 n ).
P (n n ) = 2
2 + 1 sin
sin
For the future references, let us point out that in the (d + 1)-dimensional case the only changes
in these formulae are
d
P (z) C2 (z);
l(l + 1) l(l + d);
2l + 1 2l + d
(6)
(where C are the Gegenbauer polynomials). We will obtain the dS/AdS spaces by analytic continuation from the sphere. But the procedure is nontrivial and is discussed in the next section.
203
3. Analytic connections of various spaces

The de Sitter space (or imaginary Lobachevsky space) is obtained from a sphere by an analytic
continuation n0 in0 and forms a one sheeted hyperboloid
n2k n20 = 1.
(7)
The metric has the form

= (dnk (dn0 Since even complex change of variables does
not change the scalar curvature, it remains constant and positive. The distance L between two
points is given by the formula cosh L = (nn ) and could be both time-like (real L) and space-like
(imaginary L). The KleinGordon (or Laplace) equation, defining a propagator is also unchanged
as we make the analytic continuation. Therefore we can obtain propagators on the de Sitter space
from the propagator on a sphere by the simple substitution. The only problem is that a physical
propagator in one case may be unphysical in another.
The Euclidean vacuum is defined by the propagator (4) in which we substitute (n n ) =
(nk nk ) n0 n0 . This is a nice, well-defined function with the expected singularity. However, it
does not satisfy our additivity requirement. Indeed, in the limit of large L we get the following
asymptotics, using the formulae P (z) Az + Bz1 (at large z) and = 12 + i; M 2 =
1
2
4 + ,
ds 2
G(n, n ) AeiL + BeiL .
)2
)2 .
(8)
So this Green function is unlikely to be the right starting point for the quantum theory in the
de Sitter space. The formula (5) is destroyed by the analytic continuation because on a hyperboloid it becomes divergent.
A different Green function which does satisfy our requirement is given by
G(n, n ) = Q (n n ).
(9)
It contains only one exponential in the time-like asymptotics and has a correct singularity at the
coinciding points. But it also has an unexpected singularity at the antipodal points defined by
(n n ) = 1. Strange as it is, this singularity must appear if we define the propagator as a sum
over trajectories. Indeed, the antipodal points can be connected by infinite number of geodesics.
The functional integral includes the integration over the zero mode describing the subspace of
these geodesics. If this subspace is non-compact, one expects and gets the divergence at the
antipodal points. We will discuss this singularity below.
Before discussing the physical consequences of this propagator, let us analyze the AdS spaces.
We can get them from a sphere by the continuation nk ink , so that in this case we have a twosheeted hyperboloid, n20 n2k = 1. The important difference from the above is that we also have
to change the sign of the metric on a sphere, ds 2 ds 2 . In this way we obtain the AdS space
with the euclidean signature (EAdS, or a real Lobachevsky space). Its metric is given by

ds 2 = (dnk )2 (dn0 )2 = y 2 (dx)2 + (dy)2 = (d)2 + sinh2 (d)2 + cosh2 (dt)2
(10)
where we introduced for the future use two standard parameterizations of the AdS. The scalar
curvature R = g R changes sign under the above transformation g g and hence the
space has constant negative curvature. More interestingly, under this transformation the covariant
Laplacian also changes sign. Hence we have the following relation between the Green functions

GdS n, n , M 2 = GAdS n, n , M 2 .
(11)
204
This formula implies that the positive mass in the de Sitter space corresponds to a tachionic
mass in AdS. That give us a first hint that the dS space may be unstable. However these analytic relations do not tell us which propagators are physical. This is determined by their spectral
properties.
4. Spectral features of various propagators
We see that the propagators in the spaces of constant curvature are analytically related. Let us
begin to analyze their spectral properties with the EAdS space. In the Poincar coordinates we
have the following expression
(x x )2 + (y y )2
2yy

1
cosh(t1 t2 ) cos(1 2 )
=1+
sin 1 sin 2
z = (n n ) = 1 +
(12)
where we introduce variable 0 , cosh = sin1 . This variable appears when the Poincar
upper half-plane is conformally mapped onto an infinite strip. We see that since z 1 and we
must make a change M 2 M 2 when passing from a sphere. The only propagator which does
not blow up in this space is
G(n, n ) = Q 1 + (z)
(13)
where M 2 = 14 + 2 . Its spectral decomposition is given by the formula

G(n, n ) =
tanh
0
P 1 +i (n n )
2
2 + 2
(14)
This is a precise analogue of (4) in which the angular momentum l is set to be 12 + i. It is

also instructive to look at the massless case, = 12 . Since the metric (10) is conformally flat, the
propagator in this case is just that of a free particle moving in the upper half plane or, equivalently,
in a strip with the Dirichlet boundary conditions

1 k|t|
z+1

G(n, n ) = log
(15)
sin k1 sin k2 .
=
e
z1
k
1
If we take a Fourier transform with respect to t , the kth term in this expression will give Gk
1
. The variable t is conjugate to the dilatation operator. It is easy to see that in the case of
2 +k 2
arbitrary mass, the answer will be

k (1 )k (2 )
G(, 1 , 2 ) =
.
2 + (k + )2
(16)
k=0
Here = + 12 is the lowest dimension which is increased by the raising operators; the eigenfunctions are expressed through the Legendre functions and form the space of the lowest weight
unitary representation of the symmetry group.
Our next task is to perform the Minkowskian continuation. It can be accomplished by taking
t it in the metric (10). It is also possible to take one of the xs to be imaginary in Poincars
205
metric. The fact that in the latter case the resulting coordinates do not cover the whole space
is of little consequence, since the propagators depend on the invariant distances only. There
are, nevertheless, a number of subtleties which we have to discuss. The variable z is given by
the formula (12) in which we have to replace cosh(t1 t2 ) cos(t1 t2 ). As a result, now
z . Hence, we have to deal with antipodal singularity at z = 1 in (13) and (15). We
might consider for a moment to replace the Q-function by the P -function which does not have
this singularity. However this choice is a disaster since the P -function blows up at infinity.
Another puzzle is related to the fact that any function of n n is periodic in t variable. That
means that they are defined on a space with the closed time-like geodesics and not on the universal covering space which we are primarily interested in. In other words these propagators have
singularities not only at t = t1 t2 = 0 (if 1 = 2 ) but also at t1 t2 = 2m.
The resolution of these puzzles follow from the composition principle. It dictates that while
changing i in order to go the Minkowskian AdS (MAdS), we must take the following
propagator
G(, 1 , 2 ) =

k=0
k (1 )k (2 )
.
2 (k + )2 + i0
(17)
If we
This formula gives the Feynman propagator for the universal covering space, the MAdS.
are interested in the original space on which t is a periodic variable, we must restrict to the
integer values. If we attempt to change the i0 prescription, say by taking a superposition of +i0
and i0 terms for a propagator, we will destroy the composition, since the convolution on the
right-hand side is divergent (the contour is pinched from the opposite sides).
In the coordinate space the function G is not periodic in time, while still having the form (13)
away from the singularities. This happens because the i0 prescriptions are different at t = 0 and
t = 2m. This is clear already from the expression (15) since the factor eik| | is not periodic in t .
In the first case we have a standard flat space singularity

2
G log t 2 (1 2 ) i0
(18)
which, under the action of the Laplace operator, generates a delta function at the coinciding
points. At the same time near t = 2m we have

G log (t 2m + i0)2 (1 2 )2
(19)
and there is no delta function at these points.
The singularities at z = 1 are physically necessary. They occur, according to (12), when
cos(t1 t2 ) = cos(1 + 2 ). All spacetime is represented by a strip 0 . The above
relation means that the two points are connected by a light geodesic which is reflected from
odd number of times. Let us stress again that our prescription requires to sum over all paths lying
inside the strip. We do not allow them to go to outside. The boundaries of the strip act as perfect
mirrors. If the path is a null geodesics, we expect and get singularities whenever L(Pxx ) = 0.
Let us turn now to the case of the de Sitter space. All we have to do is to interchange space and
time in the AdS propagator and take i (which inverts the sign of M 2 ). It is also important
to remember that in this case the space is periodic since we are dealing with the one-sheeted
hyperboloid. As a result, we get the following propagator
G(n, n ) = Q 1 +i (n n + i0)
2
(20)
206
which satisfies the composition principle. At the same time the standard propagator, proportional
to the P -function blows up and the composition integral diverges. It is instructive to analyze
these propagators in the Poincar coordinates, although they are geodesically incomplete. Their
advantage is simplicity. We have the following wave equation

M2
2 + p 2 + 2 gp (, ) = ( ).
(21)
Here we made a Fourier transform with respect to space variable and introduced the conju1
gate momentum p. The resulting Green function is given by Gp (, ) = ( ) 2 gp (, ). The
solution corresponding to the BunchDavies vacuum is given by
(1)
(2)
gp (, ) Hi
(p< )Hi (p> )
(22)
which, in the case of massless particles is simply

Gp (, ) p 1 eip| | .
(23)
The prefactor p 1 would be p d in d space dimension and is responsible for the scale invariant
spectrum of quantum fluctuations. It is obviously the propagator obtained by the sum over paths
lying in the whole plane. At the same time, the Poincar coordinates are defined on the upper
half-plane. Therefore, the paths in the above propagator are allowed to go to the wrong side of
infinity. The propagator (20) has a different behavior. It includes paths which are reflected from
infinity ( = 0), but never go beyond it. This results in the Dirichlet propagator in the massless
case
Gp (, )
1 ip| |

eip( + )
e
p
(24)
while in the general case

gp (, ) Ji (p< )Hi (p> ).
(25)
The relation of this propagator to the global one (20) follows from the formula

1
(1 2 ) 2 dp Ji (p< )Hi (p> )eipx = Q 1 +i (z)
2
(26)
where z is given by (12) with y i .

To understand it better, let us consider the spectral decomposition in the global coordinates.
In the de Sitter space these coordinates are obtained from the usual polar coordinates, ds 2 =
d 2 + sin2 d 2 on a sphere by the analytic continuation 2 + it (where 0 is
a polar angle, while the global time t . On a sphere the eigenmodes are given by
m Plm (cos ) (where l is the angular momentum and m is a magnetic quantum number). As
we make the above substitution, we obtain the BunchDavies modes, m Plm (i sinh t), where
this time l = 12 + i. These modes are selected by the condition that they are non-singular
at t = i
2 (the south pole of the Euclidean sphere). The reason for this requirement is based
on the HartleHawking geometry in which the euclidean half-sphere (presumably describing the
tunneling creation of the universe) is joined with the Lorentzian half-hyperboloid. It is far from
obvious that this centaur is capable of living. We will see in the next section that it is unstable in
the sense discussed in the Introduction.
207
Let us discuss other vacua, corresponding to the full hyperboloid. Generally, the eigenmodes
satisfy the KleinGordon equation

1
2
(m d1
2
2
2 ) 4
=0
(27)
+
t 2
cosh2 t
(where we returned to (d + 1)-dimensional case for the future references). From this equation
it is clear that the spectral decomposition of the Green function must contain both discrete and
continuous spectra. We can choose the in state by eit as t and, correspondingly
the out state with the same asymptotic but at t . These sets of modes (we also denote
them by ()) describe the vacua without particles at the past or future infinities. They give rise
to the Dirichlet propagator described above.
The in or modes (here d = 1 again, to simplify notations) are given by
1
Qm
l (i sinh t); l = 2 + i (these are the standard Jost functions for this problem, which contain
only one exponential at t ). This function must be understood as an analytic continuation from t > 0. The standard definition of the Q-function contains a cut from to 1.
As t becomes negative, we must go to the second sheet through the cut. As a result we have
m
Qm
l (i sinh t) + iPl (i sinh t) for t < 0, where the second term compensates the jump across
the cut, making an analytic function. These modes must be complemented with the discrete
states, which are given by the same formula but with 0 l m 1 (and being an integer). The
out modes are obtained by the CPT reflection. The presence of the discrete states reflects the
fact that on a hyperboloid there are closed geodesics (ellipses). The open geodesics (hyperbolae) correspond to the continuous spectrum. In the case of the HartleHawking geometry the
regularity on the south pole forces us to drop the discrete states, since the Q-function, unlike P
is singular there. To sum up, we have three types of modes (vacua), in, out and E. The
E/out propagator is given by the Q-function and satisfies the composition principle, the E/E
propagator is given by the P -function and does not; all other propagators are combinations of P
and Q. We will also need the expression for the Q-propagator in dSd+1 . It is given by

d1 d1
G(z) = const 1 z2 4 Q 21
2 +i
(z)
(28)
where this time M 2 = d4 + 2 and as before z = (n n ). It is interesting to notice that for even d
this expression reduces to the elementary functions. For example, for d = 2 the answer is
1 eil
(29)
4i sinh l
where l = log(z + z2 1 ) is the geodesic distance between the two points. Our convention is
that it is real for z 1 (time-like separations), imaginary for 1 z 1 (moderate space-like
separations). For z < 1 the geodesic distance becomes complex. This is a peculiar property of
the dS spacesuch points cannot be connected by a real geodesics.
G=
5. The composition principle vs. Hartle and Hawking

In a certain sense our proposal for the propagators on eternal manifolds is different from the
HartleHawking proposal for the wave functional. Indeed, we could calculate the correlation
function for the two points lying on the equator of the hyperboloid in the following way. The ddimensional geometry of the equator is given (according to Hartle and Hawking) by the sum over
208
Euclidean geometries bounded by the equator. In the semi-classical approximation this manifold
is simply a hemisphere. So, for a free scalar field we expect the wave functional of the form

( ) exp
d d D(, )( )( )
(30)
where the Poisson kernel D(, ) = G(, ) is expressed through the normal (with respect to the equator) derivatives of the Green function. In the case of a sphere we can always
decompose the functional integral in a following way

(31)
D = D
D
D
S
S+
where S are the hemispheres, E is an equator, and the integrals in the brackets are taken with
the boundary condition that the bulk fields approach a given value at the equator. We see that
the correlation function for the two points on the equator, calculated with HartleHawking wave
functional is the same on a sphere and in the de Sitter space. At the same time, the composition
principle dictates that this correlation is given by the Q-function in the de Sitter case and by the
P -function on a sphere. The difference is that in one case the amplitudes are given by the sum
over paths lying in the full Lorentzian manifold (a hyperboloid in our case), while in the other
the one half of the hyperboloid is replaced by an Euclidean sphere.
The natural question to ask in this situation (which was already touched in the introduction)
is what is wrong with the standard approach in which we postulate the BunchDavies (or the
HartleHawking) vacuum from the start and to discard the composition principle.
To answer this question, let us assume for a moment that the BunchDavies vacuum is the
correct and stable one. That means that the in and out states coincide. Therefore, if we
introduce some interaction, say 4 , we can use the standard Feynman rules. This leads to the
incurable infrared divergence even for the massive particles. For example, the first correction to
the Green function has the form

G(1) (n, n ) dn1 G(n, n1 )G(n1 , n1 )G(n1 , n )
(32)
where we integrate over the hyperboloid. If we use the P -propagator, the integrand contains
the interference terms behaving as z zd zd (where d is the number of space dimensions
and z = (n n1 )). Since the measure dn1 zd1 dz, we see that we have a logarithmic divergence log z L at large separations (L being a geodesic distance) even for the massive particles.
Clearly, in the higher orders we will be getting stronger divergences.
To understand the meaning of this phenomenon, let us digress and consider as an analogy the
case of a free scalar field in the flat spacetime but in a state with arbitrary occupation numbers,
f (p), where p is momentum. The Green function in this state is given by
G(p, t1 , t2 ) =
1
(1 + f )ei|t1 t2 | + f ei|t1 t2 |
2(p)
(33)

where = p 2 + m2 . If we assume for a moment that this state is stable in the above sense
(equality of in and out), the Feynman perturbation theory leads us to a disaster, since in any
loop diagram the interference terms proportional to f (1 + f ) will be linearly infrared divergent
due to the cancellation of the exponents. This divergence has a simple meaning: the interaction
is changing the occupation numbers and creates a non-zero time derivative of f (p). In the case
209
of weak coupling this time derivative satisfies the Boltzmann equation (written here for the 3
case to simplify the notations)

f
1 + f (p) f (p1 )f (p2 ) f (p) 1 + f (p1 ) 1 + f (p2 )

t
( 1 2 )(p p1 p2 ) dp1 dp2 .
(34)
The linear time divergence is absent only if the collision integral in (34) is zero. This is the steady
state condition. To recapitulate, without interaction we can have a stable state with arbitrary
occupation numbers f (p). This stability is destroyed by any, however weak, interaction, unless
the occupation numbers are given by Bose/Fermi distributions.
Analogously, in our problem the linear divergence is a signal that stability and the composition
principle are violated. Indeed, if we use the doubled spacetime we can easily derive the formula

G++ (n, n ) dn1 G++ (n, n1 )G++ (n1 , n ) G+ (n, n1 )G+ (n1 , n ) .
(35)
m
In this integral the linear divergence cancels. The non-zero contribution of the second term
(violating the composition) means that the vacuum we are working with is unstable. The Bunch
Davies (E) vacuum fails the eternity test (we define this test as the vanishing of the second term
in (35)).
Let us investigate this condition in more details. The presence of the spider diagrams imply
that instead of the vacuum state we must look for an excited states with the occupation numbers {f (p)} over the BunchDavies vacuum. In the small coupling limit these functions satisfy
a Boltzmann equation. The structure of this equation in this case is unusual, since it includes
particles creation from the vacuum. Leaving the detailed analyzes for another work, we will only
present these non-standard terms. They have the form

f
= dp1 dp2 (p + p1 + p2 )|A|2 1 + f (p) 1 + f (p1 ) 1 + f (p2 ) + O(f )
t
where

A
d
d+1
(1)
(1)
(1)
3d/2 Hi (p1 )Hi (p2 )Hi (p3 ).
This term describes the vacuum decay since it is non-zero even when f = 0. The terms not
explicitly displayed are the standard terms describing the evolution of the one particle states.
It seems plausible that the steady state can be reached only when the gravity from the created
excitation screens the anti-gravity of the cosmological constant leaving us with the Friedmanlike universe. I hope to discuss these fascinating questions elsewhere.
Let us explore the non-interacting theory in more details. We expand the partial wave of the
field
= a + a +
(36)
a+
where s are some solutions of (27) and a,

are the creation and annihilation operators. The
vacuum is defined by the a|vac = 0. Different choices of the solutions correspond to the different
vacua. The Green functions, vac| . . . |vac are given by
G++ (t1 , t2 ) = (t> ) (t< );
G+ (t1 , t2 ) = (t2 ) (t1 ),
etc.
(37)
210
Substituting these expressions (36) we find the eternity condition

dt 2 (t) = 0.
(38)
As we mentioned above, the modes corresponding to the E-vacuum are those which remain finite
at the south pole and have the form E Pm (cos ), = 2 + it. Let us notice also that in terms
of the conformal time (tanh t = cos ) the euclidean south pole corresponds to i. That
explains why in the Poincar coordinates the E-vacuum
requires the Hankel function (22). The

integral (38) can be written in the invariant form as C d sin (Pm (cos ))2 (the contour C goes
from zero to 2 and then to i). These integrals in can be expressed in terms of the scattering
data. Namely, let
(t) ()eit + ()eit ,
t .
(39)
Using the standard Wronskian identities, we find

T

dt 2 (t) = i
+ 2T .
(40)
We conclude that for a manifold to be eternal, we must have zero reflection amplitude. Eternity
is a rare thing.
We can also explore vacuum stability without doubling of the manifold. For this purpose we
introduce the in/out and in/E Green functions
G(n, n ) = out|T (n)(n )|inout|in1 ,
G(n, n ) = out|T (n)(n )|Eout|E
(41)
(42)
where the time ordering is taken with the respect of the time component of n. In the first case we
look at the development of the entire de Sitter spacethe full hyperboloid. In the second case
we are dealing with the centaurEuclidean hemisphere joined with the Lorentzian hyperboloid
along the equator. As we noticed before, they are given by the Q-propagator, the only one satisfying the composition principle. This is all we need, we do not have to know what are the in
and out states. Using the formula for the effective action, eiW = out|in we find the standard
formula

W
(43)
dn G(n, n).
Of course the Green function at the coinciding points is infinite, but its imaginary part is finite and
determines the imaginary part of the effective action (which will have also a trivial infinitythe
total volume of the dS space, since the integrand is independent of n). The result is
Im W (Vol) Im Q (1)
(44)
(where (Vol) is the infinite volume of the whole space). Using the formula
Q 1 +i (z + i0) Q 1 i (z i0) = i(tanh 1)P 1 +i (z)
2
(45)
we find

Im W (Vol) log 1 + e2 .
(46)
211
This method of finding the vacuum decay mimics the derivation of Schwinger of pair production in the strong electric fields. It is interesting that the effective action can also be expressed
in terms of the scattering data. Let us consider the effective action Wm for the given angular
momentum m. The partial in/out Green function is given by
Gm (t, t ) =
1
m (t> )m (t< )
2m ()
(47)
where and are the Jost functions for Eq. (27) (having the asymptotics eit at t
respectively, while 1 is the transmission coefficient). The partial effective action satisfies the
identity

Wm 1
1
log m ()
dt Gm (t, t)
dt m (t)m (t)
i
(48)
=
.
2
m
2
The last integral was again evaluated using the Wronskians. Collecting all factors, we find a
curious formula

1
eiWm =
(49)
= Tm
m
where Tm is the transmission amplitude. We see once again that the non-vanishing imaginary part
of the effective action is related to the presence of reflection. The reflection amplitude for Eq. (27)
can be extracted from the readers favorite text book on quantum mechanics (the Landau/Lifshits
in my case). But one must be careful in comparing the partial wave actions with the complete
one (44), since the summation on m diverges and can easily give a wrong result (instead of the
infinite invariant volume).
We come to the conclusion that the de Sitter space of even dimensionality is intrinsically
unstable. Whatever we do, we cannot build the eternal space. In the case of the odd dimensions
we do not get the imaginary part of the action. The reason for this can be traced to Eq. (27). If d
is odd, the potential in this Schrdinger equation can be written as n(n+1)
with integer n. This is
cosh2 t
a soliton of the KdV equation which gives reflectionless potential (the vanishing of the reflection
coefficient was noticed by the explicit calculations in [9]).
As a side remark let us notice that in the case of general FRW spaces with the scale factor a(t)
one may try to look at the more general reflectionless potentials, corresponding to the separated
solitons, in an attempt to find eternal backgrounds. This relation between eternal manifolds and
completely integrable systems seems quite intriguing.
Returning back to the odd-dimensional spaces we come to the following conclusion. If we
consider a full de Sitter space, with the past and future infinities it seems that at least in one
loop approximation we have a stable manifold. However, if we consider the HartleHawking
geometry, the amplitude = 0 and the stability is lost.
As another side remark, let us notice that in the case of AdS spaces the above analyzes indicates that the simply connected global AdS is stable, but the one with the closed time-like
geodesics is not (this is easily seen from the propagators (17) in which becomes discrete in the
multiply connected case). Such geodesics seem to destabilize a manifold in general by giving an
imaginary part to the in/out propagators in the formula (1).
6. The infrared effects in the de Sitter space
In this section we will discuss infrared interactions. We already saw how significant they are
in a number of examples. It is desirable to have a general estimate of their strength. In the flat
212
space such estimates are well known and very importantthey led to the notion of relevant and
irrelevant operators. In [1] I tried to evaluate the infrared corrections to the cosmological constant
in the Euclidean signature, while in the later work [5] and [7] a number of results were obtained
in the inin formalism.
Let us remember the infrared situation in the flat space, by considering the amplitude for
two interacting paths (with 4 interaction). The relevance of this interaction is determined by
the probability for these two Brownian paths to intersect. The standard estimate is to compare
1
the propagation without interaction which contributes G2 (R) (where G R D2
is the Green
4
D
function and R is a typical distance) with the interaction term which gives G (R)R (where the
second factor is the volume in which the interaction takes place). From here one finds the critical
dimension Dcr = 4.
Let us analyze the de Sitter case in a similar manner. The BunchDavies propagator has the
following asymptotic behavior

d
G(z i0) z 2 A()z + A()z
(50)

2
where z = (n1 n2 ) and = d4 M 2 . We have to distinguish the cases of the light particles
2
with M 2 d4 and the heavy ones with the opposite inequality. In the previous sections we dealt
mostly with the heavy particles for which we replaced i. Here we will be interested in
both cases. Let us consider the N interaction and begin with the Feynman perturbation theory.
Its significance is measured by the amplitude

F (n1 . . . nN ) (dn)G(nn1 ) G(nnN ).
(51)
To evaluate convergence of this integral we use the first term in (50) and the fact that at large z
(the infrared limit) the measure (dn) zd1 dz.
We see that for the sufficiently light particles the integral has a power like divergence. The
condition for it is given by
N
N 2
d.
2
(52)
2
d
For the 4 interaction this gives the condition M 2 3d
16 . The massless case, = 2 , is always
IR divergent. Because of the i0 prescription, in general these divergent terms have imaginary
parts (which may cancel in some special cases).
If the mass is large enough for the above integral to converge, it is easy to see that it is
dominated by the region (ni nj ) (nk n)2 1. In this case an easy estimate shows that the
interaction is always marginal. This can be seen in the Poincar coordinates in which the integral
(51) takes the form
d
d
d
N ( 2 )
d x d
F (1 N ) 2
(53)
.
d+1 ((x xk )2 ( k )2 ) d2
In this form (52) is just the divergence (convergence) condition at 0. As we compare the
value of F at small k with the non-interacting N -point function, we obtain the above result for
the convergent case.
In order to study back reaction we need to know
various polarization operators. The simplest
one appears if we add to the Lagrangian a term (dn)A(n)J (n) where J (n) = 2 (n). As usual
213
we define the back reaction (for the massless case) from the relation

2

J (n) = (dn )(n, n )A(n ) = (dn )G (nn )A(n ) (dn ) log2 (nn i0)A(n ).
(54)
In the case of in/in formalism the structure of the infrared divergences is slightly different
because of the partial cancellations between the () Green functions (see also [7]). To evaluate
them we need the relations

G+ (n, n ) = g nn + i sgn(n0 n0 ) ,
G++ (n, n ) = g(nn i0);
(55)

G+ (n, n ) = g nn i sgn(n0 n0 ) .
G (n, n ) = g(nn + i0);
(56)
The back reaction formula (54) is replaced by

J (n) = (dn ) G2++ (n, n ) G2+ (n, n ) A(n ).
(57)
This time the integration goes over the region inside the past light cone (outside the cone G++ =
G+ and the integrand is zero). Nevertheless we still have the long range effect since log2 (z
i0) log2 (z + i0) = 2i log|z|. Thus

J (n) (dn ) (n0 n0 ) (nn 1) log(nn )A(n ).
(58)
The formula (58) must be substituted into the equations of motion for the field A. The resulting
effective equations cannot be derived from any action principle, since the kernel in (58) is not
symmetric with respect to n and n . The long range correlations should generate the Jeans-like
instabilities. However, to make the discussion realistic, we must replace the scalar field with the
gravitational fluctuations, fix the gauge and account for the tensor structure. This may change the
answer. We leave this for the future work.
The contribution of the heavy particles (for which we have to change i) to the polarization operator is dominated by the interference term. As a result, in the Feynman case, we
get the kernel (nn )d . This creates a logarithmic divergence, which we already discussed in
connection with the composition principle.
7. The gauge/strings duality
Let us discuss now the gauge theory dual of the de Sitter space. It was pointed out in [12] that
the natural candidate for it is a gauge theory with the complex coupling constant. We can now
present more arguments in favor of this assertion.
Gauge/string duality is usually understood as a relation between the gauge theory correlators
and the string theory correlators in D + 1 dimension placed at infinity (which is D-dimensional
as it should). While it is often convenient, this point of view may cause trouble in the cases when
the manifold has a complicated infinity or no infinity at all. Perhaps a more general construction
is to consider the isomorphism between the vertex operators of string theory (a set of primary
operators with the world sheet dimensions (1, 1)) and a set of the field theory operators. In other
words we consider a 2d CFT on the world sheet with a critical central charge (the Liouville
field is assumed to be included in the menu). We couple it to the world sheet gravity, which (in
this case) amounts to selecting the above primaries and integrating them over the world sheet,
214
forming the vertex operators. The resulting objects are identified with the colorless field theory
operators and the world sheet OPE are transplanted to the spacetime.
The derivation of the gauge/strings duality from the first principles is a fundamental and unsolved problem. Still, there are a number of indirect qualitative arguments helping to find the
duality in each particular case. The simplest one runs as following. Suppose that we have a
string theory in the background ds 2 = d 2 + a 2 () dx 2 , where x are the coordinates of the
spacetime in which a gauge theory is located, while the is the Liouville direction. Let us
place the Wilson loop at the value = such that a( ) = . This Wilson loop defines the
open string amplitudes. The slopes of closed and open strings are different. In this case we have

open
a 2 ( )closed
0. But in the zero slope limit only the massless states survive, giving
the gauge theory. The closed string still has infinite number of states corresponding to the gauge
invariant operators, while the open string has finite number of massless states (gluons, etc.) The
key point of the argument is that the infinite blue shift sends all massive modes of the open string
to infinity.
We can now apply this argument to two possible cases. The first case is the centaur geometry
(half-sphere, half-hyperboloid). The topology in this case is the same as in the AdS, which has
the one component infinity, conformally equivalent to the flat space. Once again, the a factor
provides an infinite blue shift and we can expect that there is a gauge theory in d dimensions,
describing the (d +1)-dimensional strings in the centaur background. The single trace operator of
the gauge theory must be equal to the vertex operators, giving the usual relation for the generating
functions

log exp i dx
(59)
hn (x) Tr On = iW hn (x, t)
n
where the Tr On are the gauge-invariant operators, while W is the effective action in the hbackground for the strings on the centaur. The right-hand side is the generating function for
the vertex operators. This relation means that we can identify gauge and vertex operators in a
following way (in the Poincar coordinates)

Tr On (x) = d 2 Vn x + x( ), y( )
(60)
where the integration goes along the world sheet and the Vn ( )s are the primary operators of the
string sigma model which satisfy the on-shell condition (L0 1)Vn = 0. In the low curvature
limit this condition reduces to the KleinGordon equation. The expectation values Vn1 . . . Vnk
are given by the sum of tree diagrams in spacetime with the set of massive and massless string
states and the legs at infinity. In the AdS case the propagators in this diagrams are given by the
Q 1 + (z) function. To pass to the case of the positive curvature, all we have to do is to change
2
i. As a result we obtain the same tree diagrams but with the propagators Q 1 +i . As we
2
have learned, these propagators correspond to taking matrix elements out| . . . |E in the centaur
geometry. The new piece of information in these formulae is the statement that the analytic continuation from AdS to dS (or, more precisely, to the centaur) leads to the out/E matrix elements,
see also [11]).
Let us discuss first the N = 4 gauge theory. In the strong
the change i
coupling limit
2
corresponds to the change of the gauge coupling = gYM N . This is obviously a
non-unitary conformal theory with some of the anomalous dimensions being negative or complex, giving another manifestation of the intrinsic instability of the de Sitter space. Strictly
215
speaking, in this example we have not only the de Sitter background but also the RR fields
which become imaginary under the analytic continuation. Gauge theories with the complex coupling are expected to be unstableand so are the corresponding de Sitter spaces. Apart from other
things, the change of the sign of at large changes the sign of the Coulomb interaction of two
charges. In this case one expects Dysons instability, which potentially can terminate de Sitters
inflation.
So far we discussed the continuation of the maximally supersymmetric case, which is the
easiest but not terribly realistic. In the next section we look at the more general situation.
8. De Sitter sigma model
To discuss string theory we need a conformal sigma model on the world sheet with the de Sitter
target space. Let us begin with a well studied case of a sphere. It will be convenient for our
discussion to discretize one of the world-sheet directions, keeping the other continuous. In this
case the Hamiltonian of the model can be written as
H = 0

x
lx2 +
1
(nx nx+1 )2
0 x
(61)
where 0 is the bare coupling constant, n is a unit vector, and l is the corresponding angular
momentum operator. This model is not conformal and develops a mass gap. Let us remember
how to see this (without using the available exact solution of this model). First one calculates
the one loop beta function and find that this model is asymptotically free. That means that the
coupling grows as we go to the infrared and if the higher orders do not stop it, leads to the mass
gap m2 exp( const
0 ) at small coupling. If, on the other hand there is a zero of the beta function
at some finite value of , the theory will have no mass gap and will be conformal. To choose
between the possibilities we look at the strong coupling limit of (61). The second term can be
neglected and the model reduces to a collection of uncoupled rotators. The lowest excitation is
obtained by taking l = 1 at some site and has a mass gap 0 at large . We conclude that the
beta function has no zeroes and the theory is massive for all couplings. Of course it is possible
that the beta function would have two zeroes, but this seems unlikely and indeed, according to
the exact solution, does not happen.
We see that the outcome depends on two factorsfirst, the sign of the one-loop beta function which is determined by the sign of the curvature of the target space and the second
compactness/non-compactness of this space. The compactness in the above example was responsible for the discreteness of the spectrum and the possibility to drop the last term in (61).
Let us now consider generalizations of this method for dS and AdS spaces and their cosets.
Perhaps the simplest case is the one with constant negative curvature but compact target space.
This can be realized in two ways. Either one considers the so-called O(n) models with n < 2
(the curvature of an n-sphere n 2), or one factors AdS space by a discrete subgroup with the
compact fundamental domain. In both cases we have positive beta function at the week coupling
and the discrete spectrum in the strong coupling limit. Therefore we expect the beta function to
have a zero at some intermediate coupling. In the case of the O(n) models these conformal fixed
points are well knownthey are simply and explicitly described by the minimal models. In the
case of the compact cosets the conformal theory is still unknown.
As we return to the de Sitter space, we notice first of all that its beta function coincides
with that of a sphere in all orders of perturbation theory. This follows from the fact that the
216
transformation n0 in0 taking us from the sphere to the dS space does not change the any
Feynman diagram. Nevertheless these two theories are very different. We have already mentioned
that the angular momentum in the dS space (or in centaur geometry takes the values l = 12 + i
where is real). That means that the spectrum of kinetic energy in the strong coupling limit is
continuous and we expect no mass gap. This is a conjecture. What we said so far does not really
prove it because in the non-compact case we cannot simply drop the potential energy, since n can
become large. Some variational estimates show that by taking the slow varying n we can suppress
the potential energy and end up with the gapless spectrum. Intuitively the continuous spectrum
appears because the de Sitter space is non-compact, unlike a sphere. It is highly desirable to have
a proof of this statement.
If we believe this conjecture, we must conclude that the dS sigma model has an infrared fixed
point which resolves the conflict between the negative beta function at the weak coupling and the
gapless phase at the strong. Generally speaking the central charge at this fixed point is smaller the
critical one. However, adding real RR fields to the background allows to adjust it to the critical
value.
So we expect that the de Sitter space without any supersymmetry can be described by a string
sigma model. This model has a fixed point determined by the zero of the beta function. That
means that the curvature of the AdS is fixed. The dual gauge theory must also be at the fixed
point. At the moment we do not have explicit description of these fixed point. We will make the
following conjecture about its origin.
In 1974 t Hooft conjectured that planar diagrams become dense and can be described by a
strings world sheet. In many recent papers it is stated that AdS/CFT is a realization of this idea.
This statement is wrong because in the usual gauge/string duality Feynmans diagrams do not
become dense at all. The origin of strings in this case are electric flux lines, something quite
different from the propagators.
The planar diagrams have a finite radius of convergence and at some complex coupling constant really become dense. It is tempting to relate that to the fixed point of the de Sitter sigma
model described above. If this is correct, the t Hooft string describes gauge theory at the critical complex coupling and lives in the de Sitter space. However, at present we do not have the
necessary tools to check this conjecture.
So far we discussed the HartleHawking geometry. Next, it is natural to ask what is the gauge
theory dual in the complete de Sitter space. The main feature in this case is the existence of two
infinities, past and future. It is natural to conjecture that in this case we are dealing with two
interacting gauge theories living at these far away locations. The nature of this duality and the
possible forms of the tracetrace interactions remains to be clarified.
Acknowledgements
I would like to thank T. Damour for our conversations over many years if not decades about
infrared effects in cosmology. I am very grateful to J. Maldacena for sharing his (different) views
on dS space. Discussions with I. Klebanov and N. Nekrasov were, as always, very useful.
Finally, my deep gratitude goes to D. Makogonenko for constant encouragement, advice and
moral support.
This work was partially supported by the NSF grant 0243680. Any opinions, findings and
conclusions or recommendations expressed in this material are those of the authors and do not
necessarily reflect the views of the National Science Foundation.
217
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
A.M. Polyakov, Phase transitions and the universe, Sov. Phys. Usp. 25 (1982) 187, Usp. Fiz. Nauk 136 (1982) 538.
A. Strominger, The dS/CFT correspondence, JHEP 0110 (2001) 034, hep-th/0106113.
J.S. Schwinger, Brownian motion of a quantum oscillator, J. Math. Phys. 2 (1961) 407.
L.V. Keldysh, Zh. Eksp. Teor. Fiz. 47 (1964) 1515, Sov. Phys. JETP 20 (1965) 1018;
K.T. Mahanthappa, Multiple production of photons in quantum electrodynamics, Phys. Rev. 126 (1962) 329.
N.C. Tsamis, R.P. Woodard, Quantum gravity slows inflation, Nucl. Phys. B 474 (1996) 235, hep-ph/9602315.
R. Jackiw, C. Nunez, S.Y. Pi, Quantum relaxation of the cosmological constant, Phys. Lett. A 347 (2005) 47, hepth/0502215.
S. Weinberg, Quantum contributions to cosmological correlations, Phys. Rev. D 72 (2005) 043514, hep-th/0506236;
S. Weinberg, Quantum contributions to cosmological correlations. II: Can these corrections become large? Phys.
Rev. D 74 (2006) 023508, hep-th/0605244.
E. Witten, Quantum gravity in de Sitter space, hep-th/0106109.
R. Bousso, A. Maloney, A. Strominger, Phys. Rev. D 65 (2002) 104039, hep-th/0112218.
A.A. Starobinsky, A new type of isotropic cosmological models without singularity, Phys. Lett. B 91 (1980) 99.
J.M. Maldacena, Non-Gaussian features of primordial fluctuations in single field inflationary models, JHEP 0305
(2003) 013, astro-ph/0210603.
A.M. Polyakov, Beyond spacetime, in: Proceedings of the 23 Solvay Conference on Physics, hep-th/0602011.
N.D. Birrel, P.C.W. Davies, Quantum Fields in Curved SpaceTime, Cambridge Univ. Press, 1984.
I. Antoniadis, P.O. Mazur, E. Mottola, Cosmological dark energy: Prospects for a dynamical theory, New J. Phys. 9
(2007) 11.
Y. Kluger, J.M. Eisenberg, B. Svetitsky, F. Cooper, E. Mottola, Fermion pair production in a strong electric field,
Phys. Rev. D 45 (1992) 4659.
I. Antoniadis, E. Mottola, 4-D quantum gravity in the conformal sector, Phys. Rev. D 45 (1992) 2013.
N.P. Myhrvold, Phys. Rev. D 28 (1983) 2439.

Higher moments of heavy quark correlators

in the low energy limit at O(s2)
A. Maier, P. Maierhfer, P. Marquard
Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, 76128 Karlsruhe, Germany
Received 20 December 2007; accepted 31 December 2007
Abstract
We present the first 30 moments of the low energy expansions of the vector, axial-vector, scalar and
pseudo-scalar heavy quark correlation functions at three-loop order, including the singlet contribution which
appears for the first time at three loops. In addition we compare the behavior of the moments for large n
with the prediction from threshold calculations.
PACS: 12.38.Bx; 14.65.Dw; 14.65.Fy
Keywords: Heavy quarks
1. Introduction
5 ,
Correlators of two currents j[] (x), where j[] (x) is the scalar ,
pseudo-scalar i
5 current, are important for many interesting phenomeno , or axial-vector
vector
logical purposes. Their low energy expansions, in particular, are of interest for applications based
on sum rules [1,2] like the determination of the charm and bottom quark mass from R(s) [3,4].
In three-loop approximation, i.e. O(s2 ), the moments of the Taylor expansion for the four
correlators were evaluated up to (q 2 )8 in [5] using recursion algorithms originally suggested
in [6]. The series was evaluated in four-loop approximation up to the first physical moment
(q 2 )1 [79] employing the Laporta algorithm [10]. The three-loop singlet contributions have
E-mail address: peter.marquard@kit.edu (P. Marquard).

doi:10.1016/j.nuclphysb.2007.12.035
A. Maier et al. / Nuclear Physics B 797 (2008) 218242
219
been calculated in [11] up to the seventh and eighth moments for the axial-vector and the scalar
and pseudo-scalar correlators, respectively.
Recently, using a completely different technique based on master differential equations [12
14], the three-loop vector correlator has been calculated up to (q 2 )30 [15].
Higher moments of the pseudo-scalar correlator might be used to determine the impact of
non-perturbative effects in the determination of quark masses from lattice calculations [16].
In this work we calculate the low energy expansions for the four types of correlators in order
s2 up to (q 2 )30 , including the logarithms which arise from the singlet contribution, given by
double triangle diagrams.
This paper is organized as follows: In Section 2 we briefly present the methods used in the
calculation, paying special attention to the singlet parts. The results are presented in Section 3
where we also compare the explicitly calculated moments with the behavior for large powers of
q 2 derived from the threshold region. Section 4 contains a brief summary and conclusions.
2. Definitions and methods
The polarization functions are defined by

2
2

2
q g + q q q + q q L q = i dx eiqx 0|Tj (x)j (0)|0

q q2 = i
2
for = v, a,
(1)

dx e
iqx
0|Tj (x)j (0)|0
for = s, p,
(2)
with the currents

,
jv =
5 ,
ja =
j s = ,
5 .
j p = i
In the low energy limit the polarization functions can be written as a series in z =
is the mass of the heavy quark,

3 n
q2 =
Cn z ,
16 2
(3)
q2
4m2
, where m
(4)
n>0
with
Cn
= Cn(0),
2
s
s
(1),
+ CF Cn +
Cn(2), + ,
(2),
(2),
(2),
(5)
(2),
Cn(2), = CF2 CA,n + CF CA CN A,n + CF TF nl Cl,n + CF TF nh CF,n

(2),
+ CF TF nh CS,n .
(6)
(2),
The singlet contributions CS,n arise from the diagrams shown in Fig. 2. It has to be noted
2
q
that these are not simple power series in z but contain logarithms of the form log( m
2 ). This
2
logarithmic dependence on q reflects the presence of massless cuts in the diagrams.
The expansion in the limit q 2 0 can be achieved in two different ways. One possibility is to
first expand the propagators and later reduce the resulting massive tadpoles to master integrals.
For each order in the z-expansion two more powers of the propagators arise. For large n this
makes the reduction to master integrals through the reduction formalisms of Broadhurst or Laporta very cumbersome. The second possibility is the reduction of the full propagators to master
220
integrals. These master integrals have then to be known at least as an expansion in the external
momentum which can be calculated efficiently to high orders. The price to pay is the increased
difficulty in obtaining the necessary reduction to master integrals. In this paper we choose the
latter approach, which we present in more detail in the following.
In the first step, the diagrams are generated with QGRAF [17]. The occurring topologies are
identified with the help of q2e and exp [18,19]. The integrals are reduced to scalar propagatortype integrals of the form

[dk1 ][dk2 ][dk3 ]
P q 2 , a1 , . . . , a9 =
(7)
a .
D1a1 D2a2 . . . D9 9
using the computer algebra program FORM [20]. The denominators are given by Di = (l m2i +
i), where l is a linear combination of the loop momenta ki and the external momentum q,
and mi {0, m} is the mass of the corresponding propagator. Only one non-zero mass is taken
into account for the heaviest quark, lighter quarks are treated as massless. The powers ai of the
denominators are integers. Irreducible scalar products are expressed as propagators with negative
powers. The integrals are dimensionally regularized in spacetime dimension d = 4 2
.
In a next step, Integration-by-Parts [21] and Laportas algorithm [10,22] are used to reduce
all needed integrals of the form (7) to master integrals. This reduction is done with the program
Crusher [23]. Crusher uses GiNaC [24] for algebraic manipulations and Fermat [25] for
the simplification of the intermediate expressions using a special interface [26]. As a result of the
reduction one finds a total of 55 master integrals.
The master integrals have to be expanded for small external momenta around q 2 = 0. To
achieve this, we use the scaling equation

D
2
2
+m
Mi q 2 = 0
q
(8)
2
q 2
m2
for the master integrals Mi . D applied to Mi gives the mass dimension of Mi . Carrying out the
mass derivative produces integrals with additional powers of propagators, that are again mapped
to master integrals. This gives a system of coupled inhomogeneous linear differential equations
in q 2 . For the non-singlet part, the system is solved by a Taylor series

(k) k
Mi q 2 ,
Mi q 2 =
(9)
k=0
(k)
where the coefficients Mi have to be calculated. Diagrammatically, the coefficients are tadpole
diagrams. As boundary conditions, the three-loop master tadpoles depicted in Fig. 1(a)(c) are
(k)
chosen. The Mi can be expressed as linear combinations of these. For the singlet part, noninteger powers of q 2 have to be taken into account:
(k)
(k)
(k) 2
2 k
Mi + Mi,s q 2
q .
+ Mi,ss q 2
Mi q 2 =
(10)
k
(k)
(k)
In this case, the boundary conditions for the coefficients Mi,s and Mi,ss are the diagrams depicted
in Fig. 1(d) and (e), (f), respectively. Generation and solution of the system are implemented in
a Mathematica program. With our setup we can calculate the q 2 expansion to almost arbitrary depths. In this work we limit ourselves to the first 30 moments. This way of calculating
expansions of integrals was proposed in [12,13] and applied to the vector current in [15].
221
Fig. 1. Diagrams, which are chosen as boundary conditions for the expansion of the propagator-type master integrals
in q 2 . The diagrams (d)(f) contribute only to the singlet part.
To perform the renormalization we need the strong coupling s [27] and the mass renormalization constant at two-loop order [2830].
For the pseudo-scalar and axial-vector currents we have to pay special attention to the treatment of 5 . While in the non-singlet contributions 5 can simply be considered as anticommutating this is not possible for the singlet contributions. In the latter case the 5 matrix cannot easily
be removed from the problem and has to be treated properly to avoid problems with dimensional regularization. Therefore we follow the prescription proposed by Larin [31] and use the
definition 5 = 4!i
. Since the
-tensor is intrinsically a four-dimensional object
it cannot be used in dimensional regularization and has to be taken out of the calculation until
the renormalization has been performed. Taking the
-tensors out of the calculation the tensor
structure of the pseudo-scalar and axial-vector correlators can be cast into the form
p [
p = 1 g[ g g g ]] + 2 q [ q[ g g g ]] ,
a = 1a g[ g g ] + 2a q [ q[ g g ] ,
(11)
(12)
where [. . .] denotes total antisymmetrization. After renormalization these expressions can be

multiplied with the
-tensors to obtain the final result. The correlators are then given by
p = 1 + 2 ,
(13)
a = 1a + 2a ,
(14)
La
= 2a .
(15)
The pseudo-scalar and the longitudinal part of the axial-vector correlator are connected through
a Ward identity.

q 2 La q 2 = 4m2 p q 2 q 2 p q 2 /q 2 q 2 =0 .
(16)
In order to retain this Ward identity for the singlet contribution it is necessary to cancel the
axial-vector anomaly. For this reason we computed the singlet part ,S of the axial-vector
correlator for an isospin doublet (, ), where is taken to be heavy and light. The diagrams
contributing to the full singlet part are depicted in Fig. 2. Note that the completely massless
diagram does only contribute to the leading moment.
222
Fig. 2. Diagrams for the singlet contribution. The first diagram contributes to scalar, pseudo-scalar and axial-vector
correlators, while the latter two are only taken into account for the axial-vector to cancel the anomaly. Solid and dashed
lines denote massive and massless lines, respectively.
3. Results and asymptotic behavior for large n

The numerical results1 for the first 30 moments of the various currents are listed in
Appendix A, Tables A.1A.8. The results are given in both the MS and the onshell schemes
since both schemes have their own range of applications. For convenience we set the renormalization scale = m. Since the longitudinal part of the axial-vector correlator can be easily
obtained from the pseudo-scalar one by use of the Ward identity (16) we give only the results for
the transversal part in this case.
The results for the singlet contributions to the scalar, pseudo-scalar and axial-vector correq2
lators can be split into a constant and a logarithmic part proportional to log( m
2 ). These are
(2),
(2),
denoted by CS,n
[1] and CS,n
[L], respectively. In the case of the axial-vector correlator the singlet part includes contributions from heavy-light diagrams as explained in the previous section.
The corresponding analytical expressions for moments 912 in the MS scheme can be found
in Appendix B. The analytical results for all calculated moments both in the MS and onshell
scheme including all logarithms are available in computer readable form from http://wwwttp.particle.uni-karlsruhe.de/Progdata/ttp07/ttp07-32.
For large expansion depths, the moments can be compared to the asymptotic behavior, which
follows from the threshold behavior of heavy quark production from e+ e .
4
Ck,n =
9
2 n
1
n1

ds
4m
4
=
d 2 Rk () 1 2
,
Rk (s)
s
s
9
where
k = A, NA, l, F
(17)
4m2

and =
4m2
.
s
(18)
In view of the smooth behavior of the calculated moments, we expect good agreement for
n 30. Large n expansions can be found in Ref. [32] and derived from [5,33,34] by using
Eq. (17). A power j of the velocity in the R-ratio Rk () corresponds to a power n1j/2 of the
large n expansion.

9/2 1/2
(2),v
2 1
4 n +
n
23 4 + 8 2 6Hn+ 1 47 + 36 log(2)
CA,n =
2
6
144

+ 36(39 43 ) n3/2 + O n2 ,
1 While we agree in general with the results given in [15], we disagree with several single digits.
223
(2),v
Fig. 3. Analytical moments compared to the large n expansion to different depths for the CA,n part.

2
2
33Hn + 31 66 log(2) n1 +
36Hn+ 1 + 107 120 log(2)
2
36
72

302 4683 n3/2 + O n2 ,

2
11 3/2
(2),v
=
+ O n2 ,
3Hn + 5 6 log(2) n1 +
n
Cl,n
9
9

4
(2),v
2 11 n3/2 +
33 2 344 n5/2 + O n7/2 ,
CF,n =
(19)
9
54
where Hn is the harmonic number given by
(2),v
CN
A,n =
Hn =
n

1
i=1
(20)
for integer arguments or its generalization

Hn = E + 0 (n + 1)
(21)
for non-integer arguments, where E is Eulers constant and 0 is the digamma function. The
asymptotic behavior is in both cases given by

Hn E + log(n) + O n1 .
(22)
The comparison is shown for the vector current separately for each color factor in Figs. 36.
(2),v
The behavior of the leading order approximation is worst for the CA,n contribution since the
series starts with O(n1/2 ) compared with at least O(n1 ) for the other color factors. Comparing
identical orders in the n expansion of the various color structures shows roughly the same degree
(2),v
of convergence, but the CA,n is still penalized by the bad leading approximation. Overall we find
good agreement between the large n expansion and the analytical moments. Also at two-loops
this procedure shows nearly perfect agreement if one includes two terms in the large n expansion
as can be seen in Fig. 7.
224
(2),v
Fig. 4. Analytical moments compared to the large n expansion to different depths for the CN A,n part.
(2),v
Fig. 5. Analytical moments compared to the large n expansion to different depths for the Cl,n
part.
4. Conclusion
We calculated the low-energy expansion of the heavy quark correlator for the scalar, pseudoq 2 30
in three-loop approximation. For the scalar,
scalar, vector and axial-vector current up to ( 4m
2)
pseudo-scalar and axial-vector currents we included the singlet contributions arising from double
triangle diagrams. Furthermore, for the vector current we compared the large n behavior of the
moments with the asymptotic form derived from threshold behavior. Taking the lowest two or
225
(2),v
Fig. 6. Analytical moments compared to the large n expansion to different depths for the CF,n part.
(1),v
Fig. 7. Analytical moments compared to the large n expansion to different depths for the Cn
part.
three terms of the 1/n expansion into account, we find good agreement between the asymptotic
formulae and the explicit results.
Acknowledgements
We like to thank J.H. Khn, K. Chetyrkin and M. Steinhauser for fruitful discussions. This
work was supported by the DFG through SFB/TR 9. The work of Ph. M. was supported by the
Graduiertenkolleg Hochenergiephysik und Teilchenastrophysik.
226
Appendix A. Numerical results

In this appendix we give the numerical results for the four different currents defined in Section 2 up to n = 30. For each of the correlators the results are presented in both the MS and
onshell schemes in seperate tables.
Table A.1
Moments for the vector correlator in the MS scheme
(0),v
(1),v
(2),v
(2),v
(2),v
(2),v
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
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
1.067
0.4571
0.2709
0.1847
0.1364
0.1061
0.08558
0.07094
0.06006
0.05170
0.04512
0.03983
0.03550
0.03190
0.02887
0.02629
0.02408
0.02216
0.02048
0.01900
0.01770
0.01653
0.01549
0.01455
0.01371
0.01294
0.01224
0.01161
0.01102
0.01049
1.916
0.8322
0.3895
0.1523
0.007970
0.08688
0.1525
0.1995
0.2342
0.2602
0.2801
0.2954
0.3073
0.3166
0.3239
0.3296
0.3339
0.3373
0.3398
0.3417
0.3430
0.3438
0.3443
0.3444
0.3443
0.3439
0.3434
0.3427
0.3419
0.3410
0.1541
0.3025
0.2711
0.2785
0.3424
0.4535
0.6007
0.7750
0.9694
1.179
1.400
1.629
1.864
2.104
2.347
2.592
2.838
3.085
3.333
3.581
3.828
4.074
4.319
4.564
4.807
5.049
5.290
5.529
5.767
6.003
0.006971
0.2215
0.01532
0.2218
0.3775
0.4936
0.5813
0.6484
0.7004
0.7412
0.7734
0.7991
0.8195
0.8359
0.8490
0.8595
0.8677
0.8742
0.8793
0.8830
0.8858
0.8877
0.8888
0.8894
0.8894
0.8889
0.8881
0.8870
0.8855
0.8839
0.9934
0.6824
0.6433
0.6372
0.6362
0.6351
0.6329
0.6297
0.6257
0.6211
0.6162
0.6110
0.6056
0.6002
0.5948
0.5894
0.5841
0.5788
0.5737
0.5686
0.5636
0.5587
0.5540
0.5493
0.5448
0.5404
0.5360
0.5318
0.5277
0.5237
0.3956
0.01696
0.1026
0.1301
0.1396
0.1421
0.1415
0.1396
0.1369
0.1340
0.1310
0.1281
0.1252
0.1225
0.1198
0.1173
0.1149
0.1126
0.1105
0.1084
0.1065
0.1046
0.1028
0.1011
0.09953
0.09798
0.09650
0.09509
0.09373
0.09243
Table A.2
Moments for the vector correlator in the onshell scheme
n
1
2
3
4
5
6
7
8
Cn
(0),v
Cn
(1),v
CA,n
(2),v
CN A,n
(2),v
Cl,n
(2),v
CF,n
(2),v
1.067
0.4571
0.2709
0.1847
0.1364
0.1061
0.08558
0.07094
4.049
2.661
2.015
1.630
1.372
1.186
1.046
0.9356
5.075
6.393
6.689
6.685
6.574
6.426
6.267
6.108
7.098
6.311
5.398
4.699
4.165
3.746
3.409
3.132
2.339
2.174
1.896
1.671
1.494
1.353
1.239
1.143
0.7270
0.2671
0.1499
0.09947
0.07230
0.05566
0.04459
0.03677
227
Table A.2 (continued)

n
Cn
(0),v
Cn
(1),v
CA,n
(2),v
CN A,n
(2),v
Cl,n
(2),v
CF,n
(2),v
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
0.06006
0.05170
0.04512
0.03983
0.03550
0.03190
0.02887
0.02629
0.02408
0.02216
0.02048
0.01900
0.01770
0.01653
0.01549
0.01455
0.01371
0.01294
0.01224
0.01161
0.01102
0.01049
0.8469
0.7738
0.7126
0.6605
0.6156
0.5765
0.5422
0.5118
0.4847
0.4603
0.4383
0.4184
0.4002
0.3836
0.3683
0.3541
0.3411
0.3290
0.3177
0.3072
0.2974
0.2881
5.955
5.809
5.671
5.542
5.420
5.305
5.198
5.096
5.000
4.910
4.824
4.743
4.665
4.592
4.522
4.455
4.392
4.331
4.272
4.216
4.162
4.111
2.900
2.702
2.532
2.384
2.254
2.139
2.035
1.943
1.858
1.782
1.712
1.648
1.589
1.535
1.484
1.437
1.393
1.352
1.314
1.277
1.243
1.211
1.063
0.9941
0.9345
0.8822
0.8361
0.7950
0.7581
0.7248
0.6946
0.6671
0.6419
0.6187
0.5973
0.5775
0.5590
0.5419
0.5258
0.5108
0.4967
0.4834
0.4709
0.4590
0.03100
0.02661
0.02317
0.02041
0.01816
0.01629
0.01473
0.01340
0.01226
0.01127
0.01041
0.009654
0.008984
0.008389
0.007857
0.007379
0.006947
0.006556
0.006201
0.005876
0.005579
0.005306
Table A.3
Moments for the axial-vector correlator in the MS scheme
(0),a
(1),a
(2),a
(2),a
(2),a
(2),a
(2),a
(2),a
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
CS,n [1]
CS,n [L]
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
0.53330
0.15240
0.06772
0.03694
0.02273
0.01516
0.01070
0.007883
0.006006
0.004700
0.003760
0.003064
0.002536
0.002127
0.001804
0.001547
0.001338
0.001166
0.001024
0.0009049
0.0008043
0.0007188
0.0006454
0.0005821
0.6346
0.1062
0.009570
0.04310
0.05255
0.05367
0.05172
0.04869
0.04540
0.04219
0.03919
0.03644
0.03394
0.03168
0.02963
0.02779
0.02612
0.02460
0.02322
0.02196
0.02081
0.01975
0.01878
0.01789
0.5003
0.3025
0.2271
0.2071
0.2079
0.2160
0.2261
0.2363
0.2456
0.2539
0.2610
0.2671
0.2723
0.2767
0.2803
0.2833
0.2858
0.2879
0.2895
0.2909
0.2919
0.2926
0.2932
0.2935
0.8386
0.3370
0.2631
0.2263
0.1987
0.1764
0.1578
0.1422
0.1290
0.1177
0.1079
0.09949
0.09209
0.08558
0.07982
0.07469
0.07010
0.06597
0.06224
0.05886
0.05578
0.05296
0.05039
0.04802
0.6198
0.2883
0.2034
0.1602
0.1323
0.1124
0.09739
0.08563
0.07617
0.06841
0.06193
0.05646
0.05177
0.04772
0.04420
0.04109
0.03835
0.03591
0.03373
0.03176
0.02999
0.02837
0.02691
0.02556
0.1775
0.02960
0.04074
0.03578
0.02992
0.02505
0.02121
0.01820
0.01580
0.01388
0.01230
0.01099
0.009902
0.008976
0.008185
0.007502
0.006909
0.006389
0.005931
0.005525
0.005162
0.004838
0.004546
0.004282
1.970
0.6914
0.3832
0.1867
0.1447
0.08050
0.07290
0.04309
0.04308
0.02618
0.02813
0.01729
0.01966
0.01212
0.01443
0.008880
0.01098
0.006736
0.008609
0.005252
0.006905
0.004189
0.005646
0.003404
0.004690
0.002811
0.003949
0.002353
0.003364
0.001993
0.002895
0.001706
0.002514
0.001473
0.002200
0.001283
0.001938
0.001125
0.001719
0.0009933
0.001533
0.0008821
0.001375
0.0007876
0.001238
0.0007067
0.001120
0.0006370
(continued on next page)
228

(0),a
(1),a
(2),a
(2),a
(2),a
(2),a
(2),a
(2),a
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
CS,n [1]
CS,n [L]
25
26
27
28
29
30
0.0005272
0.0004793
0.0004372
0.0004002
0.0003674
0.0003382
0.01706
0.01630
0.01559
0.01493
0.01432
0.01375
0.2937
0.2937
0.2936
0.2934
0.2931
0.2927
0.04583
0.04381
0.04194
0.04020
0.03858
0.03707
0.02433
0.02320
0.02216
0.02119
0.02029
0.01946
0.004043
0.003825
0.003626
0.003444
0.003276
0.003121
0.001017
0.0009275
0.0008484
0.0007785
0.0007164
0.0006611
0.0005765
0.0005238
0.0004776
0.0004369
0.0004009
0.0003689
Table A.4
Moments for the axial-vector correlator in the onshell scheme
n
Cn
(0),a
Cn
(1),a
CA,n
(2),a
CN A,n
(2),a
Cl,n
(2),a
CF,n
(2),a
CS,n [1]
(2),a
CS,n [L]
(2),a
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
0.5333
0.1524
0.06772
0.03694
0.02273
0.01516
0.01070
0.007883
0.006006
0.004700
0.003760
0.003064
0.002536
0.002127
0.001804
0.001547
0.001338
0.001166
0.001024
0.0009049
0.0008043
0.0007188
0.0006454
0.0005821
0.0005272
0.0004793
0.0004372
0.0004002
0.0003674
0.0003382
1.701
0.7158
0.3968
0.2524
0.1748
0.1282
0.09805
0.07743
0.06270
0.05181
0.04353
0.03709
0.03199
0.02787
0.02450
0.02170
0.01936
0.01738
0.01569
0.01424
0.01297
0.01187
0.01091
0.01005
0.009298
0.008624
0.008021
0.007479
0.006991
0.006548
2.314
1.648
1.190
0.8998
0.7077
0.5741
0.4772
0.4045
0.3484
0.3040
0.2683
0.2390
0.2147
0.1942
0.1768
0.1618
0.1488
0.1375
0.1275
0.1187
0.1109
0.1038
0.09753
0.09183
0.08668
0.08198
0.07770
0.07377
0.07017
0.06685
2.714
1.693
1.090
0.7579
0.5583
0.4293
0.3409
0.2778
0.2310
0.1954
0.1675
0.1454
0.1275
0.1127
0.1005
0.09013
0.08136
0.07384
0.06734
0.06168
0.05672
0.05236
0.04849
0.04504
0.04196
0.03919
0.03669
0.03443
0.03238
0.03051
1.046
0.6638
0.4313
0.3015
0.2228
0.1717
0.1366
0.1114
0.09269
0.07843
0.06728
0.05840
0.05120
0.04529
0.04036
0.03621
0.03269
0.02967
0.02705
0.02478
0.02278
0.02103
0.01947
0.01808
0.01684
0.01573
0.01473
0.01382
0.01299
0.01224
0.3433
0.06509
0.02238
0.01013
0.005396
0.003202
0.002052
0.001393
0.0009894
0.0007283
0.0005520
0.0004287
0.0003399
0.0002742
0.0002246
0.0001864
0.0001566
0.0001329
0.0001138
0.00009826
0.00008549
0.00007488
0.00006599
0.00005849
0.00005211
0.00004664
0.00004194
0.00003786
0.00003431
0.00003120
1.970
0.3832
0.1447
0.07290
0.04308
0.02813
0.01966
0.01443
0.01098
0.008609
0.006905
0.005646
0.004690
0.003949
0.003364
0.002895
0.002514
0.002200
0.001938
0.001719
0.001533
0.001375
0.001238
0.001120
0.001017
0.0009275
0.0008484
0.0007785
0.0007164
0.0006611
0.6914
0.1867
0.08050
0.04309
0.02618
0.01729
0.01212
0.008880
0.006736
0.005252
0.004189
0.003404
0.002811
0.002353
0.001993
0.001706
0.001473
0.001283
0.001125
0.0009933
0.0008821
0.0007876
0.0007067
0.0006370
0.0005765
0.0005238
0.0004776
0.0004369
0.0004009
0.0003689
(2),s
Table A.5
Moments for the scalar correlator in the MS scheme
n
1
2
3
4
5
Cn
(0),s
Cn
(1),s
CA,n
(2),s
CN A,n
(2),s
Cl,n
(2),s
CF,n
CS,n [1]
(2),s
CS,n [L]
(2),s
0.8000
0.2286
0.1016
0.05541
0.03410
0.4519
0.3194
0.1152
0.02460
0.01630
0.03484
0.5536
0.4516
0.3448
0.2876
2.511
0.6269
0.3544
0.2797
0.2439
0.8815
0.3550
0.2345
0.1857
0.1570
0.7186
0.1201
0.008315
0.01803
0.02406
2.353
0.8228
0.3660
0.1923
0.1134
0.4444
0.2074
0.1089
0.06377
0.04057
229

(0),s
(1),s
(2),s
(2),s
(2),s
(2),s
(2),s
(2),s
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
CS,n [1]
CS,n [L]
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
0.02273
0.01605
0.01182
0.009009
0.007050
0.005640
0.004596
0.003803
0.003190
0.002707
0.002320
0.002006
0.001749
0.001536
0.001357
0.001206
0.001078
0.0009681
0.0008732
0.0007908
0.0007189
0.0006559
0.0006003
0.0005511
0.0005073
0.03517
0.04364
0.04691
0.04752
0.04673
0.04524
0.04342
0.04146
0.03949
0.03756
0.03571
0.03397
0.03232
0.03078
0.02934
0.02799
0.02674
0.02556
0.02447
0.02344
0.02249
0.02159
0.02075
0.01996
0.01922
0.2635
0.2577
0.2614
0.2698
0.2803
0.2913
0.3021
0.3124
0.3219
0.3305
0.3383
0.3454
0.3516
0.3572
0.3622
0.3665
0.3704
0.3738
0.3767
0.3793
0.3816
0.3835
0.3852
0.3866
0.3878
0.2193
0.1995
0.1827
0.1681
0.1553
0.1440
0.1339
0.1250
0.1170
0.1098
0.1033
0.09739
0.09205
0.08719
0.08274
0.07866
0.07491
0.07145
0.06826
0.06530
0.06255
0.05999
0.05761
0.05538
0.05329
0.1366
0.1210
0.1083
0.09788
0.08908
0.08156
0.07508
0.06943
0.06448
0.06010
0.05621
0.05273
0.04960
0.04678
0.04422
0.04189
0.03976
0.03781
0.03602
0.03437
0.03285
0.03144
0.03012
0.02890
0.02776
0.02428
0.02277
0.02083
0.01890
0.01713
0.01556
0.01417
0.01296
0.01190
0.01096
0.01014
0.009409
0.008759
0.008178
0.007656
0.007187
0.006763
0.006378
0.006027
0.005707
0.005415
0.005146
0.004898
0.004669
0.004458
0.07267
0.04956
0.03546
0.02635
0.02019
0.01586
0.01273
0.01040
0.008624
0.007246
0.006159
0.005287
0.004580
0.003998
0.003516
0.003111
0.002769
0.002478
0.002228
0.002012
0.001825
0.001661
0.001517
0.001390
0.001278
0.02746
0.01950
0.01438
0.01093
0.008525
0.006789
0.005505
0.004532
0.003782
0.003193
0.002723
0.002344
0.002034
0.001778
0.001565
0.001385
0.001233
0.001103
0.0009917
0.0008952
0.0008113
0.0007379
0.0006734
0.0006165
0.0005661
(2),s
Table A.6
Moments for the scalar correlator in the onshell scheme
(0),s
(1),s
(2),s
(2),s
(2),s
(2),s
(2),s
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
CS,n [1]
CS,n [L]
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
0.8000
0.2286
0.1016
0.05541
0.03410
0.02273
0.01605
0.01182
0.009009
0.007050
0.005640
0.004596
0.003803
0.003190
0.002707
0.002320
0.002006
0.001749
0.001536
0.001357
0.001206
0.4519
0.7765
0.5215
0.3571
0.2565
0.1922
0.1489
0.1186
0.09662
0.08017
0.06756
0.05769
0.04982
0.04345
0.03823
0.03388
0.03024
0.02715
0.02451
0.02224
0.02027
0.03484
1.426
1.526
1.327
1.115
0.9371
0.7952
0.6824
0.5921
0.5191
0.4592
0.4096
0.3681
0.3329
0.3028
0.2769
0.2544
0.2347
0.2174
0.2020
0.1884
2.511
0.8955
0.9989
0.8275
0.6646
0.5378
0.4417
0.3685
0.3119
0.2673
0.2317
0.2028
0.1790
0.1592
0.1426
0.1285
0.1164
0.1060
0.09695
0.08903
0.08206
0.8815
0.3591
0.4002
0.3336
0.2691
0.2185
0.1798
0.1502
0.1273
0.1092
0.09464
0.08285
0.07315
0.06508
0.05828
0.05251
0.04757
0.04330
0.03959
0.03635
0.03349
0.7186
0.1911
0.07144
0.03362
0.01832
0.01103
0.007146
0.004890
0.003494
0.002584
0.001967
0.001532
0.001218
0.0009850
0.0008084
0.0006722
0.0005654
0.0004804
0.0004119
0.0003561
0.0003101
2.353
0.4444
0.8228
0.2074
0.3660
0.1089
0.1923
0.06377
0.04057
0.1134
0.07267
0.02746
0.04956
0.01950
0.03546
0.01438
0.02635
0.01093
0.02019
0.008525
0.01586
0.006789
0.01273
0.005505
0.01040
0.004532
0.008624
0.003782
0.007246
0.003193
0.006159
0.002723
0.005287
0.002344
0.004580
0.002034
0.003998
0.001778
0.003516
0.001565
0.003111
0.001385
230

n
Cn
(0),s
Cn
(1),s
CA,n
(2),s
CN A,n
(2),s
Cl,n
(2),s
CF,n
(2),s
CS,n [1]
(2),s
CS,n [L]
(2),s
22
23
24
25
26
27
28
29
30
0.001078
0.0009681
0.0008732
0.0007908
0.0007189
0.0006559
0.0006003
0.0005511
0.0005073
0.01855
0.01703
0.01570
0.01451
0.01346
0.01252
0.01167
0.01090
0.01021
0.1762
0.1653
0.1554
0.1464
0.1383
0.1309
0.1241
0.1179
0.1122
0.07589
0.07041
0.06551
0.06112
0.05716
0.05359
0.05035
0.04739
0.04470
0.03097
0.02872
0.02672
0.02492
0.02330
0.02184
0.02051
0.01930
0.01820
0.0002719
0.0002398
0.0002127
0.0001896
0.0001698
0.0001528
0.0001380
0.0001251
0.0001138
0.002769
0.002478
0.002228
0.002012
0.001825
0.001661
0.001517
0.001390
0.001278
0.001233
0.001103
0.0009917
0.0008952
0.0008113
0.0007379
0.0006734
0.0006165
0.0005661
Table A.7
Moments for the pseudo-scalar correlator in the MS scheme
n
Cn
(0),p
Cn
(1),p
CA,n
(2),p
CN A,n
(2),p
Cl,n
(2),p
CF,n
(2),p
CS,n [1]
(2),p
CS,n [L]
(2),p
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
1.333
0.5333
0.3048
0.2032
0.1478
0.1137
0.09093
0.07488
0.06306
0.05405
0.04700
0.04136
0.03677
0.03296
0.02977
0.02707
0.02475
0.02274
0.02099
0.01945
0.01810
0.01689
0.01581
0.01484
0.01397
0.01318
0.01246
0.01181
0.01121
0.01065
2.333
1.548
0.9088
0.5346
0.3010
0.1458
0.03751
0.04085
0.09923
0.1437
0.1783
0.2055
0.2271
0.2446
0.2587
0.2702
0.2796
0.2874
0.2937
0.2989
0.3032
0.3067
0.3095
0.3118
0.3136
0.3150
0.3160
0.3168
0.3173
0.3176
2.712
3.399
2.797
2.198
1.774
1.506
1.358
1.300
1.310
1.369
1.467
1.595
1.745
1.913
2.095
2.288
2.490
2.699
2.913
3.131
3.353
3.577
3.803
4.030
4.258
4.487
4.715
4.944
5.173
5.401
1.858
0.02615
0.04435
0.1204
0.2777
0.4046
0.5039
0.5813
0.6422
0.6903
0.7286
0.7594
0.7842
0.8042
0.8205
0.8336
0.8442
0.8527
0.8595
0.8649
0.8690
0.8721
0.8744
0.8759
0.8768
0.8771
0.8770
0.8765
0.8757
0.8746
0.9259
0.4346
0.4017
0.4294
0.4600
0.4846
0.5029
0.5162
0.5256
0.5320
0.5362
0.5386
0.5398
0.5400
0.5394
0.5383
0.5367
0.5347
0.5324
0.5299
0.5273
0.5246
0.5217
0.5188
0.5159
0.5129
0.5099
0.5069
0.5039
0.5009
1.311
0.3307
0.06951
0.02758
0.07127
0.09305
0.1044
0.1104
0.1133
0.1144
0.1145
0.1139
0.1128
0.1116
0.1101
0.1086
0.1071
0.1056
0.1040
0.1025
0.1010
0.09956
0.09814
0.09677
0.09544
0.09415
0.09291
0.09170
0.09053
0.08940
5.629
3.065
1.952
1.375
1.034
0.8132
0.6612
0.5512
0.4686
0.4048
0.3542
0.3134
0.2798
0.2519
0.2283
0.2082
0.1908
0.1758
0.1626
0.1510
0.1407
0.1315
0.1233
0.1159
0.1092
0.1032
0.09765
0.09260
0.08797
0.08372
1.000
0.6667
0.4667
0.3471
0.2703
0.2180
0.1804
0.1525
0.1311
0.1142
0.1007
0.08963
0.08044
0.07273
0.06617
0.06054
0.05566
0.05141
0.04767
0.04436
0.04142
0.03879
0.03642
0.03429
0.03236
0.03060
0.02899
0.02752
0.02618
0.02493
(2),p
Table A.8
Moments for the pseudo-scalar correlator in the onshell scheme
n
1
2
(0),p
(1),p
(2),p
(2),p
(2),p
(2),p
(2),p
Cn
Cn
CA,n
CN A,n
Cl,n
CF,n
CS,n [1]
CS,n [L]
1.333
0.5333
2.333
2.615
2.712
7.040
1.858
3.578
0.9259
1.232
1.311
0.4964
5.629
3.065
1.000
0.6667
231

(0),p
(1),p
(2),p
(2),p
Cn
Cn
CA,n
CN A,n
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
0.3048
0.2032
0.1478
0.1137
0.09093
0.07488
0.06306
0.05405
0.04700
0.04136
0.03677
0.03296
0.02977
0.02707
0.02475
0.02274
0.02099
0.01945
0.01810
0.01689
0.01581
0.01484
0.01397
0.01318
0.01246
0.01181
0.01121
0.01065
2.128
1.754
1.483
1.282
1.129
1.008
0.9097
0.8292
0.7618
0.7045
0.6553
0.6125
0.5749
0.5418
0.5122
0.4858
0.4619
0.4403
0.4207
0.4027
0.3862
0.3710
0.3570
0.3440
0.3320
0.3207
0.3102
0.3004
8.273
8.466
8.331
8.090
7.821
7.554
7.300
7.062
6.842
6.639
6.451
6.276
6.115
5.965
5.825
5.694
5.571
5.456
5.348
5.246
5.150
5.059
4.972
4.890
4.812
4.737
4.666
4.598
4.104
3.939
3.659
3.381
3.130
2.910
2.718
2.550
2.402
2.271
2.154
2.050
1.956
1.871
1.793
1.722
1.657
1.597
1.542
1.490
1.443
1.398
1.357
1.318
1.281
1.247
1.214
1.183
Cl,n
(2),p
CF,n
(2),p
CS,n [1]
(2),p
CS,n [L]
(2),p
1.502
1.475
1.387
1.291
1.202
1.121
1.050
0.9878
0.9322
0.8828
0.8385
0.7987
0.7627
0.7301
0.7003
0.6730
0.6480
0.6248
0.6034
0.5836
0.5651
0.5478
0.5317
0.5166
0.5023
0.4889
0.4763
0.4643
0.2589
0.1618
0.1124
0.08353
0.06509
0.05249
0.04347
0.03674
0.03157
0.02751
0.02424
0.02157
0.01936
0.01749
0.01591
0.01455
0.01338
0.01235
0.01145
0.01065
0.009943
0.009309
0.008739
0.008225
0.007760
0.007337
0.006951
0.006597
1.952
1.375
1.034
0.8132
0.6612
0.5512
0.4686
0.4048
0.3542
0.3134
0.2798
0.2519
0.2283
0.2082
0.1908
0.1758
0.1626
0.1510
0.1407
0.1315
0.1233
0.1159
0.1092
0.1032
0.09765
0.09260
0.08797
0.08372
0.4667
0.3471
0.2703
0.2180
0.1804
0.1525
0.1311
0.1142
0.1007
0.08963
0.08044
0.07273
0.06617
0.06054
0.05566
0.05141
0.04767
0.04436
0.04142
0.03879
0.03642
0.03429
0.03236
0.03060
0.02899
0.02752
0.02618
0.02493
Appendix B. Analytical results

In this appendix we present the analytical results for the moments 9 to 12 of the various
currents in the MS-scheme. The expression for the moments up to 30 can be obtained from
http://www-ttp.particle.uni-karlsruhe.de/Progdata/ttp07/ttp07-32 in computer readable form in
the MS and onshell scheme. In the following expressions we use L = log 2 /m2 and Lq 2 =
(2),
log(q 2 /m2 ). The singlet contributions CS,n are always given separately.
B.1. Scalar current
12

Cn(0),s zn =
n=9
12

n=9
Cn(1),s zn
131072 9
262144 10
1048576 11
2097152 12
z +
z +
z +
z ,
14549535
37182145
185910725
456326325

4979043746471936
524288
=
L z9
104786039514531375 4849845

2888036521295872
3538944
+
L z10
61796895098313375 37182145
232
(2),s
C9
(2),s
C10
(2),s
C11

3145728
38132976301932544
L z11
+
842822916656645925 37182145

11534336
33183292389464276992
L z12 ,
+
764287144877276645625 152108775

1048576 2 25108082269650944
L +
L
=
1616615
34928679838177125
73746539708944133873833421163704138849
+
5771391802019474808577720320000

3409922386710494034767011
3 CF2
320780855251304448

5406868428560384
65536 2
L
L
+ CA
1322685
28578010776690375
102915497024490763942656449854922612500879
+
437529212511096385238276977459200000

1345087895588279678743561
3 C F
6873875469670809600

262144
262144
14418875171495936
nh +
n l L2 +
nh
+ TF
14549535
14549535
314358118543594125

370939064258803430677307
14418875171495936
nl L
nh
+
314358118543594125
488861314420492468224000

12790934874367656891392
125624633262191
nl +
n h 3 C F ,
+
130678984236690621356625
204096845905920

23887872 2 1743551373172736
=
L +
L
37182145
2288773892530125
1181283682867330857434847115794497652295927361
+
18741992087247811295709473066188800000

1856910365195457777240839095367
3 CF2
35414206419744011059200

147456 2
2883103404986368
L
L
+ CA
3380195
16853698663176375
19485093544688649029885182650128118344232413
+
20183683786266873703071740225126400000

738743868672823822714945309
3 C F
919849517395948339200

589824
589824
7789487453519872
nh +
n l L2 +
nh
+ TF
37182145
37182145
185390685294940125

4979656088382493508774203867
7789487453519872
+
nl L
nh
185390685294940125
6626352163198301909286912000

48054627678191205275648
332499096340061
+
nl +
n h 3 C F ,
539469652874543334318375
544258255749120

4718592 2
8933862862782464
L +
L
=
7436429
11237638888755279
7404906335674218474752347314237559396292812077443
24289621745073163439239477093780684800000

131072 2
61139862184252373166516579786803
3 CF2 + CA
L
241074161519130140737536
3380195
1880663970832384
L
12097936602726975
8830293913654550377524203305197260449225292386979
+
2237852825236511033113012799650733752320000

209352182103823255038597110761
3 C F
63776233206119084851200

524288
524288
nh +
n l L2
+ TF
37182145
37182145

97554230956294144
97554230956294144
nh +
nl L
+
2528468749969937775
2528468749969937775
202040916959120787173924674741
394357158107030025924608
nh +
nl
271479640140730429738057728000
4835000144023764273824265

2287436936895761
n h 3 C F ,
+
3773523906527232

31719424 2 19006459960269111296
L +
L
=
50702925
23160216511432625625
188473078860503880648883122776543550015991250654673059
+
129919789020675223866838798646389820620800000

15859712 2
23090275323409880205366729096347
3 CF2 + CA
L
19132869961835725455360
456326325
29590089938919374848
L
208441948602893630625
20886533237898744255234695965842642441637712801424593
+
1299197890206752238668387986463898206208000000

41752924406364782322130747883
3 C F
3121913513586248908800

5767168
5767168
nh +
n l L2
+ TF
456326325
456326325

81479567700203732992
81479567700203732992
nh +
nl L
+
2292861434631829936875
2292861434631829936875
88884847785166765220997026873
nh
120599609370209094749021798400

41147533572114438909275987968
77358608492779
+
nl +
n h 3 C F ,
548058254961784643538602765625
128642860449792
28771023505735526373476647
143600001918689
=
+
3
1454362410400965092966400000 3741775508275200
289369094656
L 2,
26467422856875 q
+
(2),s
C12
(2),s
CS,9
233
234
(2),s
1090589835840240580931001947
48559274491381
+
3
57980581427985141706260480000 1496710203310080
3471392768
L 2,
407191120875 q
4337059242964423
115469790307862708811970912146409
+
3
=
6584738671613416573296590192640000 156086670679080960
225667578742784
L 2,
33237789623663625 q
675994471643152871
6397059251470658815951587545516867
+
3
=
395084320296804994397795411558400000 28095600722234572800
2744536517378048
L 2.
498566844354954375 q
CS,10 =
(2),s
CS,11
(2),s
CS,12
B.2. Pseudo-scalar current

12

131072 9
262144 10
1048576 11
2097152 12
z +
z +
z +
z ,
2078505
4849845
22309287
50702925
n=9

12

524288
49751393148928
(1),p
L z9
Cn z n =
501368610117375 692835
n=9

557796221206528
1179648
+
L z10
3880964426464125 1616615

5242880
4091600265773056
L z11
+
22953132465087825 7436429

11534336
35891639960928256
L z12 ,
+
174694204543377519 16900975

24213861812657343488
23936615221766899739377
(2),p
C9
nl
nh
= +CF TR
46071664047183800025
13951233275688124416000
386477922523
365917995900928
n h 3 +
L(nl + nh )
+
289910292480
1504105830352125
262144 2
L (nl + nh )
+
2078505

2774476486497511252284870305976724044391
+ CF CA
9695938227392717678410570137600000

65536 2
1870074094390219597606687
151838577682432
3
L
L
+
7855857679623782400
136736893668375
188955

18248707238521284112217374676684189
+ CF2
1201215870545458762874880000

143085894538571009718455041
372897204109312
1048576 2
+
3 +
L+
L ,
11321677244163686400
167122870039125
230945

18023374701238546946048
1835086571234207640548161
(2),p
C10 = +CF TR
nl
nh
33879736653956827759125
941510679624652161024000
(0),p n
Cn
z =+
235
46149597022417
2917754077462528
n h 3 +
L(nl + nh )
30236569763840
11642893279392375
196608 2
L (nl + nh )
+
1616615

1539378756641920703135316484755839467328419
+ CF C A
1333422361938579459773796502732800000

12803248135021178751056537
1179975928287232
49152 2
+
3
L
L
13331152426028236800
1058444843581125
146965

27324314428644557274188086091814971713169
+ CF2
369369075329246387748974100480000
423117757374464
13536779710726154580166718023
+
3 +
L
219964015029465907200
143739423202375

7962624 2
L ,
+
1616615

3760136484063862872174592
(2),p
C11 = CF TR
nl
7013105487369063346138875
14436891835130136916846548977
nh
6626352163198301909286912000
186926938565003
17577073730093056
n h 3 +
L(nl + nh )
+
108851651149824
68859397395263475

2621440 2
+
L (nl + nh )
22309287

2196670741691390865449882336983942060570342639
+ CF C A
472298200598644844651878721267957760000
6968767912075264
5140924281484864972357892859
3
L
+
1328671525127480934400
6259945217751225

655360 2
L
2028117

19892582218578011106338578696247144162810751553
2
+ CF
56225976261743433887128419198566400000
796428906692608
821232360279025582272054039829
3 +
L
+
2790210202767709962240
218601261572265

39321600 2
+
L ,
7436429

2361650790215100407548239872
(2),p
C12 = CF TR
nl
4384466039694277148308822125
43084675796375645372731027301693
nh
17917656249288208362711810048000
3922441614750337
676219888715104256
n h 3 +
L(nl + nh )
+
2058285767196672
2620413068150662785
5767168 2
+
L (nl + nh )
50702925
+
236

70034794682581481813249178402809862391078239422033231
+ CF CA
3741689923795446447364957401016026833879040000
77778913328384051652725192513
1319456473035685888
3
L
+
4995061621737998254080
1191096849159392175

15859712 2
L
50702925

2267473277094876433590459813785417415724639080638329
+ CF2
1360218817724097152597410717251718348800000
66863488712130988880884620697226071
572835921739218944
+
3 +
L
48214832303826028147507200
132344094351043575

95158272 2
L ,
+
16900975
2454162037509
27352144660490483390633
263495168
(2),p
+
3
L 2,
CS,9 = +
174887254737970790400000 9448928051200
2010133125 q
46177240370863
99214339418590131445021
732934144
(2),p
+
3
L 2,
CS,10 = +
783494901226109140992000 199561360441344
6416344935 q
2787996822340736999889074473
53552348130925
(2),p
CS,11 = +
+
3
26799913193379798833115955200 257285720899584
21320382464
L 2,
211739382855 q
561179696637195
14641282147919288479322671871
(2),p
+
3
CS,12 = +
169262609642398729472311296000 2973079441506304
1212441755648
L 2.
13527793904625 q
B.3. Vector current
12

n=9
12

Cn(0),v zn =
524288 9
524288 10
8388608 11
4194304 12
z +
z +
z +
z ,
8729721
10140585
185910725
105306075

2103270735183872 262144
L z9
Cn(1),a zn =
8981660529816975 323323
n=9

524288
57014372204019712
L z10
+
219098082621292875 676039

13456105329705877504 12582912
L z11
+
48040906249428817725 16900975

8388608
249307596877594624
L z12 ,
+
843894528756654375 11700675

1769472 2 1422485887860736
(2),v
L +
L
C9
=
323323
332654093696925
381427200524285859827479342961645120691697
26317546617208805127114404659200000

4641207108486463789099278533
3 CF2
384937026301565337600

32768 2 3201458214330368
L
L
+ CA
88179
2449543780859175
4800616671655745922224061798696958246114657
21001402200532626491437294918041600000

62742999460179248975566471
3 C F
+
329946022544198860800

131072
131072
8171733794127872
nh +
n l L2 +
nh
+ TR
969969
969969
26944981589450925

7505222973653044996167293
8171733794127872
+
nl L
nh
26944981589450925
2893884388042736664576000

1716980884029425246777344
15641077379375
nl +
n h 3 C F ,
+
2744258668970503048489125
7653631721472

3932160
72590094055309312
(2),v
L2 +
L
C10 =
676039
14606538841419525
541634465543720570056096925789635688755714501
7906777911807670390377433949798400000

3503779464918218459698976337
3 CF2
+
61482997256500019200

720896 2 76685205430140928
+ CA
L
L
2028117
59754022533079875
100571880864387246559861014880105540717546397
110694890765307385465284075297177600000

358487787916859833012705187
3 C F
+
474297407407285862400

262144
262144
198611843579379712
2
nh +
nl L +
nh
+ TR
2028117
2028117
657294247863878625

56043284857896604271843338613
198611843579379712
nl L
nh
+
657294247863878625
19879056489594905727860736000

1187963065976173458415616
331575524130137
nl +
n h 3 C F ,
+
1912665132918835458037875
148434069749760

103809024 2 1643667546316865536
(2),v
L +
L
C11 =
16900975
291157007572295865
1070887924864138198352946851863557684684339508412651
3351967800820096554615047838941734502400000

9610834140751883729239482548466517
+
3 CF2
36161124227869521110630400

5767168 2 16505631011785080832
+ CA
L
L
16900975
13102065340753313925
10142587654701589960628097458296025022529793100251069
2806267442846584835523718050762020125409280000
+
237
238

1150547909207069411531224161671
3 C F
382657399236714509107200

2097152
2097152
nh +
n l L2
+ TR
16900975
16900975

43261806664333656064
43261806664333656064
+
nh +
nl L
144122718748286453175
144122718748286453175
85142270271685420818460662817
nh
27996337889512825566737203200
849053414353143539189776384
nl
+
1377975041046772818039915525

428230928483939
n h 3 C F ,
+
176883933118464

8388608 2 591048256550862848
(2),v
C12 =
L +
L
1300075
93766058750739375
2120116328634260041790541524636709954701152751347827221
1439111509152094787448060538852318013030400000

16883124936978831575637459733931549
3 CF2
+
13775666372521722327859200

11534336 2 284624595694944256
L
L
+ CA
35102025
230153053297269375
20238608949682386388157368187133321175530208319894383
1406650347291521220813893759780461215744000000

103631644405192190650944065895
3 C F
+
8658106811012530307072

4194304
4194304
nh +
n l L2
+ TR
35102025
35102025

753487939399450624
753487939399450624
nh +
nl L
+
2531683586269963125
2531683586269963125
213296190145039197864108250117409
nh
65438396736530847933382262784000

369724970712395017121284096
4949453959637945
nl +
n h 3 C F .
+
605143454135205716089734375
1899956092796928
+
B.4. Axial-vector current

12

262144 9
524288 10
2097152 11
4194304 12
z +
z +
z +
z ,
43648605
111546435
557732175
1368978975
n=9

12

131072
14273234676170752
(1),a n
L z9
Cn z =
314358118543594125 1616615
n=9

524288
101689470330109952
L z10
+
2410078908834221625 7436429
Cn(0),a zn =
239

1048576
14155800052957184
L z11
+
361209821424276825 16900975

8388608
83544812412930228224
L z12 ,
+
2292861434631829936875 152108775

209029528206436443461632
(2),a
nl
= CF TR
C9
2744258668970503048489125
721707504396631265775619523
1102011173546563
nh +
n h 3
1296460205843146025730048000
2449162150871040

35512855382474752
65536 2
L(nl + nh ) +
L (nl + nh )
+
943074355630782375
4849845

2044821895603527816981536356704130687706357
+ CF C A
3360224352085220238629967186886656000

668136450989003773285863131
16384 2
12933050525913088
3
L
L
1319784090176795443200
85734032330071125
440895

217311353924672252345666774738607455908207
+ CF2
5540536129938695816234611507200000
8434594159026176
4567332086889583465526764451
3 +
L
139977100473296486400
11642893279392375

884736 2
+
L ,
1616615

287854153300932624846848
209290713314001851931555887
(2),a
nl
nh
C10 = CF TR
4207863292421438007683325
382289547876825110151168000
181196559131413
243286941705469952
+
n h 3 +
L(nl + nh )
408193691811840
7230236726502664875

262144 2
+
L (nl + nh )
22309287

8649742991474532665515591016625431163403791601
+ CF CA
3542236504489836334889090409509683200000
87853979961663488
44535077782003477491448862123
3
L
21923080164603435417600
657294247863878625

65536 2
L
2028117

3633687108507123867092545337354886401661662813
+ CF2
19462837936757342499390606645657600000
117265192181399552
35752565032549034445184262371577
3 +
L
230192341728336071884800
160671927255614775

3932160 2
+
L ,
7436429

157211774390308655582420992
(2),a
C11
= CF TR
nl
2538375075612476243757739125
175561738935918272688195117821
nh
325775568168876515685669273600
240
9918778191302813
32831051515756544
n h 3 +
L(nl + nh )
22641143439163392
1083629464272830475

524288 2
L (nl + nh )
+
50702925

1157200807257694296198427317159640234150785132824303
+ CF CA
118158629172487782548367075821558742122496000
261883214808207508025819129858143
619633587273728
3
L
32143221535884018765004800
5184829972597275

1441792 2
L
50702925

325018733172338765882468493323383479610810052389621
+ CF2
370968768470208314344748377432286822400000
105426049549787146186606803725022377
1602332519923712
3 +
L
144644496911478084442521600
2189150432874405

8650752 2
L ,
+
16900975

92827398267438355820629508096
(2),a
C12 = CF TR
nl
1644174764885353930615808296875
104450315270262159531899064503
nh
196742892149046993786639482880
5341389426809173
188918503999998132224
n h 3 +
L(nl + nh )
+
12349714603180032
6878584303895489810625

4194304 2
L (nl + nh )
+
456326325

14691137807907519650408236687148333789997689179014563981
+ CF C A
374168992379544644736495740101602683387904000000
818111578538634207243886625532379
67346421666621587456
3
L
25046666131857676959744000
625325845808680891875

11534336 2
L
456326325

151997936918859535196394940736545376607752983761243261719
+ CF2
37416899237954464473649574010160268338790400000
3258783646497863020834421263009179937
3
964296646076520562950144000

8388608 2
186408134950156238848
L+
L ,
+
254762381625758881875
16900975
+
(2),a
24604386555649
54729591036013756991855587205067403
+
3
2933529196264874811502352306012160000 997806802206720
6722140930048
L 2,
997962281090775 q
CS,9 =
(2),a
=
CS,10
87798501391061
119621269712255836267447988400312983
+
3
7360491437900958617951356695085056000 4245214394843136
241
303225290752
L 2,
57733355104425 q
392233656487033430717559964717352391313
(2),a
CS,11 =
27547265098473410838449785029714247680000
2353323008444009
1219596674400256
3
L 2,
+
133788574867783680
291157007572295865 q
5473635073785826339847768690355966987576523
(2),a
CS,12 =
436094396712755996196381981470399397888000000
26173581704692859
466514873679872
3
L 2.
+
1728960044445204480
137042701251080625 q
References
[1] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, QCD and resonance physics: Applications, Nucl. Phys. B 147 (1979)
448518.
[2] L.J. Reinders, H. Rubinstein, S. Yazaki, Hadron properties from QCD sum rules, Phys. Rep. 127 (1985) 1.
[3] J.H. Kuhn, M. Steinhauser, C. Sturm, Heavy quark masses from sum rules in four-loop approximation, Nucl. Phys.
B 778 (2007) 192215.
[4] J.H. Kuhn, M. Steinhauser, Determination of s and heavy quark masses from recent measurements of R(s), Nucl.
Phys. B 619 (2001) 588602.
[5] K.G. Chetyrkin, J.H. Kuhn, M. Steinhauser, Heavy quark current correlators to O(s2 ), Nucl. Phys. B 505 (1997)
4064.
[6] D.J. Broadhurst, Three loop on-shell charge renormalization without integration: Lambda-MS (QED) to four loops,
Z. Phys. C 54 (1992) 599606.
[7] K.G. Chetyrkin, J.H. Kuhn, C. Sturm, Four-loop moments of the heavy quark vacuum polarization function in
perturbative QCD, Eur. Phys. J. C 48 (2006) 107110.
[8] R. Boughezal, M. Czakon, T. Schutzmeier, Charm and bottom quark masses from perturbative QCD, Phys. Rev.
D 74 (2006) 074006.
[9] B.A. Kniehl, A.V. Kotikov, Heavy-quark QCD vacuum polarisation function: Analytical results at four loops, Phys.
Lett. B 642 (2006) 68.
[10] S. Laporta, High-precision calculation of multi-loop Feynman integrals by difference equations, Int. J. Mod. Phys.
A 15 (2000) 50875159.
[11] K.G. Chetyrkin, R. Harlander, M. Steinhauser, Singlet polarization functions at O(s2 ), Phys. Rev. D 58 (1998)
014012.
[12] M. Caffo, H. Czyz, S. Laporta, E. Remiddi, The master differential equations for the 2-loop sunrise selfmass amplitudes, Nuovo Cimento A 111 (1998) 365389.
[13] E. Remiddi, Differential equations for Feynman graph amplitudes, Nuovo Cimento A 110 (1997) 14351452.
[14] A.V. Kotikov, Differential equations method: New technique for massive Feynman diagrams calculation, Phys. Lett.
B 254 (1991) 158.
[15] R. Boughezal, M. Czakon, T. Schutzmeier, Four-loop tadpoles: Applications in QCD, Nucl. Phys. B (Proc.
Suppl.) 160 (2006) 160164.
[16] P. Lepage, private communication.
[17] P. Nogueira, Automatic Feynman graph generation, J. Comput. Phys. 105 (1993) 279289.
[18] T. Seidensticker, Automatic application of successive asymptotic expansions of Feynman diagrams, PhD thesis,
1999.
[19] R. Harlander, T. Seidensticker, M. Steinhauser, Complete corrections of O(s ) to the decay of the Z boson into
bottom quarks, Phys. Lett. B 426 (1998) 125132.
[20] J.A.M. Vermaseren, New features of FORM, math-ph/0010025.
[21] K.G. Chetyrkin, F.V. Tkachov, Integration by parts: The algorithm to calculate beta functions in 4 loops, Nucl. Phys.
B 192 (1981) 159204.
[22] S. Laporta, E. Remiddi, The analytical value of the electron (g 2) at order 3 in QED, Phys. Lett. B 379 (1996)
283291.
[23] P. Marquard, D. Seidel, Crusher, an automated integral reduction setup, unpublished.
242
[24] C. Bauer, A. Frink, R. Kreckel, Introduction to the GiNaC framework for symbolic computation within the C++
programming language, J. Symbolic Comput. 33 (2002) 1.
[25] R.H. Lewis, Fermats user guide.
[26] M. Tentyukov, J.A.M. Vermaseren, Extension of the functionality of the symbolic program FORM by external
software, Comput. Phys. Commun. 176 (2007) 385.
[27] O.V. Tarasov, A.A. Vladimirov, A.Yu. Zharkov, The Gell-Mann-low function of QCD in the three loop approximation, Phys. Lett. B 93 (1980) 429432.
[28] N. Gray, D.J. Broadhurst, W. Grafe, K. Schilcher, Three loop relation of quark (modified) MS and pole masses,
Z. Phys. C 48 (1990) 673680.
[29] O.V. Tarasov, Anomalous dimensions of quark masses in three loop approximation, JINR-P2-82-900.
[30] S.G. Gorishnii, A.L. Kataev, S.A. Larin, Next-next-to-leading perturbative QCD corrections and light quark masses,
Phys. Lett. B 135 (1984) 457.
[31] S.A. Larin, The renormalization of the axial anomaly in dimensional regularization, Phys. Lett. B 303 (1993) 113
118.
[32] K.G. Chetyrkin, J.H. Kuhn, M. Steinhauser, Higher moments of heavy quark vacuum polarization, hep-ph/9607422.
[33] A. Czarnecki, K. Melnikov, Two-loop QCD corrections to the heavy quark pair production cross section in e+ e
annihilation near the threshold, Phys. Rev. Lett. 80 (1998) 25312534.
[34] K.G. Chetyrkin, J.H. Kuhn, M. Steinhauser, Three-loop polarization function and O(s2 ) corrections to the production of heavy quarks, Nucl. Phys. B 482 (1996) 213240.

Moduli stabilization in heterotic M-theory

Filipe Paccetti Correia a, , Michael G. Schmidt b
a Centro de Fsica do Porto, Faculdade de Cincias da Universidade do Porto, Rua do Campo Alegre, 687,
4169-007 Porto, Portugal

b Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, 69120 Heidelberg, Germany
Received 8 November 2007; accepted 8 January 2008

Abstract
We reconsider the ingredients of moduli stabilization in heterotic M-theory. On this line we close a gap in
the literature deriving the Khler potential dependence on vector bundle moduli and charged matter. Crucial
in this derivation is our superspace formulation of 5d heterotic M-theory taking into account the Bianchi
identities modified by brane terms. Likewise, we obtain the FayetIliopolous terms due to brane localised
anomalous U(1)s. After assembling perturbative and non-perturbative contributions to the superpotential,
we study supersymmetric (AdS) vacua. It is found that the susy condition decouples the bundle moduli
from the geometric moduli. We show that M-theory supersymmetric vacua without five-branes can be found,
albeit not at phenomenologically interesting values of the geometric moduli. This result is fairly independent
of the choice of vector bundle at the observable brane.
Keywords: Heterotic M-theory; Moduli stabilization
1. Introduction
The last few years witnessed a notable activity, following the work of KKLT [1], in the construction of stable de Sitter vacua in type II string theory, less so in heterotic string/M-theory.
One should recall, however, that the heterotic string is rather appealing in regard to particle
model building, being the ideal context for the construction of supersymmetric Grand Unified
Theories. Moreover, its strongly coupled limit, heterotic M-theory [2,3], has additional attractive
E-mail addresses: paccetti@fc.up.pt (F. Paccetti Correia), m.g.schmidt@thphys.uni-heidelberg.de (M.G. Schmidt).

doi:10.1016/j.nuclphysb.2008.01.005
244
F. Paccetti Correia, M.G. Schmidt / Nuclear Physics B 797 (2008) 243267
phenomenological features [4,5], including the fact that there is an intermediate regime where
the theory looks effectively five-dimensional [6,7].
The proposal of KKLT [1] for moduli stabilization, following the work of [8], and improvements thereof [921] have been implemented mainly for the type II string. In fact, although the
problem of moduli stabilization in heterotic M-theory has been object of some interest in recent
years [2229], it is fair to say that it is far from being a settled issue. It is well known that sometimes, in the phenomenologically interesting regime we are forced to approach the limit where
warping along the S 1 /Z2 direction is considerable. Then, we have to face several obstacles.
Firstly, there is a constraint on the validity range of the 11d supergravity approximation. Secondly, the reduction of this theory to 4d is complicated by the fact that the different zero-modes
localise differently along the bulk. Moreover, generally the zero-mode wave-functions are not
explicitely known. On the other hand, the low-energy gauge theory is determined by the VEVs
of the vector bundle moduli and charged scalars, all of them localised at the boundary branes.
The effect of these fields in moduli stabilisation must also be taken into account.
One of the problems with these brane chiral multiplets is the poor understanding of their contribution to the Khler potential. It is one of the aims of this paper to improve on this situation.
The reason the Khler potential dependence on vector bundle moduli and charged scalars C is
unknown is twofold. Firstly, the dependence of the gauge bundle on and C is in general not
known as this would in principle require an explicit knowledge of the solutions of the zero-slope
Hermitian YangMills equations on the CalabiYau three-fold. Unfortunately, this seems difficult, if not impossible to find. Secondly, even if we would know the gauge bundle explicitely we
lack sofar an explicit expression (i.e. in terms of holomorphic coordinates) for the Khler potential dependence on the gauge bundle. It is this gap that we will be able to close in this paper.
We will show in Section 3 that the Khler potential takes a form which is strongly reminiscent
of the DeWolfeGiddings Khler potential [30] describing a mobile D3 brane in type IIB compactifications, and that the bundle moduli and charged scalars enter the Khler potential through
k I , I = 1, . . . , h1,1 , dubbed little Khler potentials, which have the form
certain real functions
I
CY tr(A A) .
While in general this only provides an implicit knowledge of the way the Khler potential
depends on and C, the formal expressions found in Section 3 can be used to derive a number
of interesting results and, in particular, study supersymmetric stable vacua of heterotic M-theory.
This is due to the fact, shown in Section 6, that the F-flatness conditions decouple the bundle
moduli from the geometric moduli, even in the presence of brane instantons and gaugino condensates. In other words, the value of the gauge bundle in the supersymmetric vacuum is exactly
independent of the value of the geometric moduli in contrast with the expectations raised in the
literature. Then, the remaining supersymmetry conditions are used to determine the values of the
geometric moduli S and T I in the background of the gauge bundle. By solving these equations
with h1,1 = 1, we will see in Section 5 that the effect of membrane instantons wrapping holomorphic cycles dominates over the effect of gaugino condensation at the hidden brane in such a
way that the theory is driven to an phenomenologically uninteresting limit with a far too small
size of the M-theory orbifold direction, implying a too small 4d Planck mass.
In this work we will also briefly discuss the relevance for moduli stabilization in dS vacua
of potentials induced by FayetIliopolous terms due to anomalous U(1)s acting on the boundary
branes of heterotic M-theory. It has been recently emphasized [3133] that such U(1)s are generically present in attempts to obtain realistic heterotic model-building using complex line-bundles.
We point out in Section 5 that FI D-terms and membrane instantons in general exclude each
other in the following sense. Given a Khler modulus associated with an isolated holomorphic
245
curve, either the corresponding membrane instanton superpotential vanishes or its anomalous
U(1) charge vanishes.
We will perform the derivation of the bundle moduli and matter dependent Khler potentials using the N = 1 superfield description of 5d heterotic M-theory we developed in [34]. This
is a particular application of a general superfield formalism constructed for 5d gauged supergravity on the S 1 /Z2 orbifold in Refs. [3436] and [37,38]. Extending the results of [34] to
the non-standard embedding will require a careful study of certain Bianchi identities and their
dependence on the brane localised gauge bundles. It is worth emphasizing that, as we will see
below, the superfield formalism turns out to be a powerfull and economic technic and is crucial
for obtaining the results of this paper.
This article is organized as follows. Section 2 reviews the relevant couplings of 5d heterotic
M-theory and its reduction to four dimensions. A brief explanation of the superfield description
used herein is also presented. We derive explicitely the Khler potentials in 3 situations. One
with h1,1 = 1; one with h1,1 = 3; and h1,1 = 1 with a mobile five-brane in the bulk. But we also
discuss the structure of the Khler potential in the general case. This discussion is completed
in Appendix A. Section 3 contains the main technical results of this paper. We derive here the
brane (and bulk-brane) superfield couplings involving the gauge bundle degrees of freedom, beyond the standard embedding. This allows us to obtain the effective Khler potential dependence
on the bundle moduli and charged matter C. We draw some parallels with type IIB compactifications with mobile D3-branes. In Section 4 we discuss FayetIliopolous D-terms induced by
anomalous U(1)s and discuss the D-term potential in the strongly coupled case. Section 5 assembles perturbative and non-perturbative superpotentials and discusses the compatibility of the
latter with non-vanishing FI terms. Finally, in Section 6, we present a study of supersymmetric
vacua without mobile five-branes. We conclude with a brief discussion.
2. 5d heterotic M-theory and its 4d reduction
In the following paragraphs we review the relevant couplings of 5d heterotic M-theory. For
this purpose, we use the (conformal) superfield formalism of [3436] and [37,38], based on the
off-shell 5d supergravity construction of [3942]. The point of using the superfield description is
that the reduction to 4d, that we will perform below, is rather transparent, specially regarding the
N = 1 conformal superspace structure.
2.1. Relevant couplings of 5d heterotic M-theory
As any five-dimensional gauged supergravity, 5d heterotic M-theory contains in addition to
the gravitational sector a number of vector and hypermultiplets [6,7]. The bosonic components
of the vector multiplets are h1,1 Khler structure moduli {bI } and gauge fields {AI } obtained by
expanding the 11d 3-form potential C3 over (a basis of) harmonic (1, 1) forms. On the other hand,
the hypermultiplets include the dilaton universal hypermultiplet and h2,1 complex structure
moduli. We will not deal here with the complex structure moduli in much detail.
As we just mentioned, the Khler moduli are associated with 5d vector multiplets. Each of
these decomposes into a chiral superfield I and a vector superfield V I of N = 1 (4d) super-
246
symmetry as follows [35,37,42]1 :

1
i
I = ey5 bI + AIy + ,
2
2
V I = AI + .
(2.1)
(2.2)
Note that we take the WessZumino gauge. The scalars bI have to satisfy the constraint [7,42]
N (b) dI J K bI bJ bK = 1.
(2.3)
I
Note, however, that the corresponding chiral superfields

are not constrained [35], as they
depend on one additional degree of freedom, ey5 .
There is a combination of the above superfields that will play an important rle in this paper.
This is the gauge invariant
I
y
VyI I + I y V I = ey5 bI + F
+ ,
(2.4)
where Fy = y A Ay . Clearly, VyI is invariant under the gauge transformations

V I V I + I + I ,
I I + y I .
(2.5)
For the upcoming discussion, it is crucial to stress the fact that Fy is a component of VyI . In fact,
it is well known that the brane localised gauge fields lead to modifications of the 2-form F I ,
F I = dAI (y)(brane fields).
(2.6)
(The precise form of the brane term will be presented in Section 3.) We will have therefore to
I still holds. This will be done in Section 3.
modify Eq. (2.4) in such a way that VyI Fy
The crucial ingredient in 5d heterotic M-theory is the universal hypermultiplet. It consists
of four real scalars and their supersymmetric partners., and can be decomposed into two chiral
N = 1 superfields, S and . The so-called dilaton superfield S is associated with the volume of
the internal CalabiYau 3-fold measured at the position y along the orbifold. Its lowest component is
S(y) = VCY (y) + | |2 + i.
(2.7)
Here, is dual to the 5d 3-form potential C 3 arising from the 11d bulk in the HoravaWitten
setup, while is an S 1 /Z2 -odd complex field which appears in the expansion of the 11d 3-form
potential C3 (of the 4-form G) in harmonic forms,
1/3
C3 = C 3 + + + VCY AI I + .
(2.8)
Finally, let us mention that is an S 1 /Z2 -odd chiral superfield.

According to [6,7], the 11d G-flux induces a gauging of the shift symmetry + by a
suitable combination of the 5d vector multiplets2 :
S S 2(y)I I ,
(2.9)
1 The ellipsis in the following expressions stand for the fermionic and auxiliary components of the corresponding
superfields. For a discussion of these terms see [35,37,42].
2 This shift symmetry is a particular subgroup of the isometries of the universal hyperscalar manifold. As such, from
the viewpoint of 5d supergravity, this is not the most general gauging (see [41] and references therein).
V I V I + I + I ,
I I + y I .
247
(2.10)
The 5d superspace Lagrangian will have to be invariant under these transformations. As it was
explained in [34], the relevant bulk couplings, i.e. those couplings leading to non-vanishing 4d
moduli interactions, can be written in the form of a D-term Lagrangian in five dimensions

N (Vy ) 1/3
LD = 3 d 4

S0 + S0 + 2(y)I V 2I
y
1/3

(y) I + I
(2.11)
N (Vy ) dI J K VyI VyJ VyK
where
is determined by the CY intersection numbers dI J K appearing
also in (2.3). We note that at energies below the compactification scale, the superfields and
S0 can be replaced by their zero modes, which have no y-dependence. In fact, as we reduce to
4d becomes the chiral compensator superfield and S0 is the 4d chiral superfield defined by
S0 S(y = 0).
The superspace 5d Lagrangian above was derived in [34] from a 5d supergravity perspective.
Here, we will give a bottomup way of understanding this expression. In the limit of vanishing
G4 -flux there is no warping and the low energy 4d tree-level Lagrangian is just

exp(K/3),
LD,4d = 3 d 4
(2.12)
with Khler potential
ln N ( + ).
K = ln(S + S)
(2.13)
Now, consider uplifting this expression to 5d still without G-flux (and warping). We must care
about 5d Poincar and gauge invariance. As we said above, in 5d I + I is not a gauge invariant, while VyI = I + I y V I is. Therefore, in going to 5d we have to modify (2.12)
as

1/3 + ,
N (Vy ) 1/3 (S + S)
LD,4d LD,5d = 3 d 4
(2.14)
where the ellipsis stand for odd superfields which vanish in the low-energy limit. Of course, 5d
Poincar invariance must be also imposed both in the vector and dilaton sector. We will here
not care about the vector sector, see [34] for details. The only way of recovering 5d Poincar
invariance in the dilaton sector is to introduce an F-term Lagrangian which must have the form
(see e.g. [43])

LF,5d = d 2 3 y S + h.c. + ,
(2.15)
where is an odd chiral superfield.3
Let us now turn on the G-flux. We know that this corresponds to gauging the shift symmetry
S S + i by a suitable combination of the 5d vector multiplets. Imposing invariance of LF,5d
under the action of this transformation we find that y S must be replaced by y S + 2(y)I I .
3 In the notation of [34], is an odd combination of several superfields defined therein: 4 = hc / h + H c (h +
1
2 1
H )/(h1 h2 ).
248
by (S + S + 2(y)I V I ). Finally, we introduce S0

Likewise, in LD,5d we must replace (S + S)
as
y
S = S0 2I
(y) I .
(2.16)
To fully understand the meaning of S0 we must note that now LF,5d reads

LF,5d = d 2 3 y S0 + h.c. + .
(2.17)
Going to 4d the odd superfield acts as multiplier imposing that y S0 = 0 [34,38]. This means
that S0 is a 4d superfield and according to (2.16) it is the value of the dilaton measured at
the visible brane (y = 0). Finally, plugging (2.16) back into (2.14) we obtain the D-term Lagrangian (2.11).
2.2. Reduction to 4d and the effective Khler potential
In the preceding discussion we have seen how starting from the known 4d weakly coupled
heterotic theory, one can uplift to 5d and turn on the G-flux to obtain the superspace Lagrangian (2.11). We will now go the opposite way from 5d to 4d, but keeping the non-vanishing
warping. It is not difficult to see that the 4d Khler potential is obtained by performing the following integral (cf. (2.12))
eK/3 =
1/3

y

1/3
I
dy N ( + )
,
S0 + S0 2I (y) + I
(2.18)
where we retained only S 1 /Z2 -even quantities. The difficulty now is to evaluate this expression
explicitely in the warped background. As it was pointed out in [34], it is only in the h1,1 = 1 case
that this is straightforward. We obtain [34,44]

3
(S0 + S0 )4/3 (S + S )4/3 ,
(2.19)
4
where the modulus S S0 T measures the size of the CY 3-fold at the location of the hidden
brane, and4
K = 3 ln
dy =

dy ey5
+i
dy Ay + .
(2.20)
In the weak coupling limit Re(T ) Re(S) we recover the universal Khler potential
3 ln(T + T )
K = ln(S + S)
(2.21)
where we introduced the average CY volume modulus S (S0 + S )/2. Note that due to warping, in the strong coupling regime the nice direct product of the hypermultiplet sector and the
4 In this paper, the definition of the moduli T I differs from the one in [34] by a factor of 2 .
249
vector sector patent in (2.21) disappears. Moreover, unlike the weak coupling Khler potential (2.21), the exact Khler potential (2.19) describes a moduli space which has a curvature
singularity at the strong coupling limit Re(S ) = 0.
Before considering the addition of bulk M5 branes, let us comment on the structure of the
Khler potential in the more general h1,1 > 1 case. The first fact we want to stress is that K is
now a function of Re(S0 ) and of the 4d Khler moduli Re(T I ), where
I
T dy I ,
(2.22)
and, again, Re(S0 ) measures the size of the CY 3-fold at the visible brane. Note that, instead of,
say, T 0 one can use the modulus S S0 I T I which sets the size of the CY at the hidden
brane, or the average CY size modulus S defined above. As a non-trivial example, let us consider
the STU-model with h1,1 = 3, 0 = 0, 1 = 2 = 0 and d123 = 1. The Khler potential can be
determined using the methods of [34,38] and turns out to be
KSTU = ln

1
(S0 + S0 )2 (S + S )2 ln T 1 + T 1 T 2 + T 2 .
20
(2.23)
It is important to note that in the two examples presented in this section, where we are able to
explicitely calculate the low-energy Khler potential, the following set of relations is satisfied:

Ki X i + X i = 4,
(2.24)
i

i
Ki j X + X = Kj ,
(2.25)
Kj i Ki Kj = 4,
(2.26)
where the X i denote collectively the S and T moduli. Moreover, these expressions still hold in
the presence of M5 branes in the bulk. This leads us to conjecture that they should be valid also in
the general case, regardless of the explicit form of the intersection numbers dij k and the number
of five-branes in the 5d bulk. We show in Appendix A that this is true, at least for general dij k
and no five-branes.
2.3. Bulk M5 branes
Sometimes it is useful to introduce bulk five-branes to more easily satisfy the tadpole cancellation conditions [4548]. For simplicity we will specialize to the case with h1,1 = 1. There is
therefore only one 2-cycle the five-brane can wrap. The presence of the five-brane in the bulk
modifies the y-dependence of the G4 flux. This can be very easily accounted for by replacing the
coupling 0 (y) in Eq. (2.18) by (y)(y) with

0 , 0 < |y| < z,
(y) =
(2.27)
1 , z < |y| < ,
z
where z is a fixed reference distance of the five-brane to the visible boundary, dy ey5 being the
physical distance.
The Khler potential was determined in [34]. It is given by
K = 3 ln

3 1
0 (S0 + S0 )4/3 01 11 (S5 + S5 )4/3 11 (S + S )4/3 .
4
(2.28)
250
Here, Re(S5 ) is the volume of the internal CY measured at the position of the five-brane, which
can be rewritten in terms of the five-brane position modulus Z as5
S5 = S0 0 Z,
(2.29)
where [4951]
z
Z

dy ey5 + iAy iB,
(2.30)
and B originates from the KK reduction of the 2-form living on the five-brane. Finally, the
modulus S can also be expressed in terms of the five-brane modulus Z as
S = S0 1 T (0 1 )Z.
(2.31)
We note that (1 0 ) is the tension of the five-brane and therefore must be positive, that is
1 > 0 .
(2.32)
3. Brane localised couplings and the Khler potential

The requirement of global anomaly cancellation leads necessarily to non-trivial gauge bundles
and therefore to the breaking of the E8 gauge symmetries at least at one of the boundary planes.
The resulting low-energy brane field content as well as its interactions depend on the precise
choice of the vector bundle and also on the CY 3-fold. We will see now how by solving the
Bianchi identities one can derive the contribution of the brane multiplets to both the Khler and
superpotential. To illustrate this idea, we will first describe the brane localised couplings in the
simplest case of the standard embedding with universal moduli only. The general case of a nonstandard embedding with h1,1 = 1 will then be discussed below.
3.1. Standard embedding with universal moduli
In the standard embedding the gauge group at the visible brane reduces to an E6 E8 and, in
addition to the corresponding gauge fields, there are h1,1 charged multiplets C I transforming in
the 27 representation of the E6 (for simplicity we assume h2,1 = 0). We will consider the case
h1,1 = 1. Since we are dealing here with 5d supergravity, the kinetic terms for C must be encoded
in a D-term Lagrangian localised at the visible brane,

S + S;
Vy ).
=
3(y)
d 4 (C,
C;
LD
(3.1)
matter
This term can be determined by solving the Bianchi identity for the 5d 2-form F in the presence
of (a gradient of) the matter field C. In fact, the only natural and gauge invariant way of obtaining
such a term is by performing the following addition to (2.11):
Vy Vy = + y V (y)CC.
(3.2)
This modification is required to ensure the gauge invariance of the action and can be obtained
I . In fact, in the presence of brane matter, the 5d Bianchi
by noting that VyI = + i Fy
5 The definition of the modulus Z used in [34] differers from the present one by a factor of 2.
251
identities have brane localised sources, leading to

C + c.c.) dx dy.
F = dA + (y)(i CD
(3.3)
The above redefinition of Vy must be accompanied by a modification in the definition of the

superfield ,
=
i
1 5
ey b + (y)|C|2 + Ay + .
2
2
(3.4)
The rational behind this is to impose that the lowest component of the superfield in Eq. (3.1),
which is the coefficient of the EinsteinHilbert term [52], has no support on the brane. In this way
we ensure that no brane localised EinsteinHilbert terms exist. Likewise, the dilaton superfield
S0 (see (2.7)) is modified as
S0 = VCY (0) +
2
|C| + i (0)
2
(3.5)
(with (0) = 0) otherwise the bulk Lagrangian would depend on the brane field C.
To determine the low-energy Khler potential we have now to perform the following integral:
e
K/3

1/3
y

1/3
dy N + (y)C C
.
S0 + S0 2 (y)( + )

=
(3.6)
It is enough to calculate the lowest component of this superfield expression, since the rest can be
obtained by supersymmetrization [34]. (This is the reason we dropped the vector superfield V in
Eq. (3.6).) The lowest component of (3.6) is (with d111 = 1)
e
K/3

=2
dy ey5
0
1/3
y
2VCY (0) 2

ey5
0
3
3
=
(2VCY )4/3 (0)
2VCY (0) 2
4
4
4/3

dy ey5
(3.7)
Notice that, due to the definition of , Eq. (3.4), we have now

T=
dy =
1

dy ey5 + iAy + |C|2 .
2
(3.8)
We thus see that in terms of the holomorphic coordinates the Khler potential reads

K = K S0 + S0 |C|2 ; T + T |C|2
4/3
4/3
3
S0 + S0 (T + T )
.
= 3 ln
S0 + S0 |C|2
4
In the 0 limit we find the well known result

K = ln S + S |C|2 /2 3 ln T + T |C|2 .
(3.9)
(3.10)
252
It is important to stress that the combination T + T |C|2 , appearing in the Khler potential,
is still proportional to the undeformed volume of the 2-cycle:
2
5
J,
T + T |C| = dy ey dy g11
(3.11)
2-cycle
where J is the Khler 2-form of the CY 3-fold. By undeformed volume we mean the following.
To leading order, the effect of turning on the gauge bundles at the boundaries is just to produce an
11d metric which is a warped product of a CY 3-fold and a 5d spacetime. At this level of approximation, the only effect of the G-flux is the CY volumes dependence on y. Clearly, Eq. (3.11)
gives the volume of the cycle at this level of approximation. Notice, in particular, that it does
not depend on the value of C. In fact, this should be expected, as the warping is determined by
topological numbers and it therefore cannot change continuously as we change the gauge bundles at the boundaries. On the other hand, in general we do expect the volume of the cycles to be
deformed continuously in the presence of these gauge bundles, that is, it should depend on the
vector bundle moduli and charged matter C a . As we will argue in Section 6, while the Khler
potential depends on the undeformed volume of the cycles, the superpotential induced by brane
instantons will be a function of the deformed volume of the wrapped holomorphic cycles and
will thus depend on and C a . Note that the same reasoning applies to the dilaton superfields
S0 and S and the fact that the Khler potential depends only on the warped volume of the CY,
which is sensitive only to the topology of the gauge bundle.
3.2. Non-standard embedding
In the general case, that is for other choices of the gauge bundle, additional states will be
present at the 4d branes. In particular, in addition to the chiral multiplets C a charged under the
4d gauge groups, one finds vector bundle moduli whose number depends on the topological
properties of the vector bundle. The VEVs of the bundle moduli determine the configuration of
the gauge bundle and thus also the value of Yukawa couplings and matter kinetic terms. As such,
it is important to know how and at which values they are stabilized. As we have seen above,
their kinetic terms, determined by the way they enter the 4d Khler potential, can be obtained
by looking at the 5d Bianchi identities. A direct reduction from the 11d Bianchi identity for the
4-form G shows that

I
I
F = dA (y) dy dx trE8 A d A + (D A) A ,
(3.12)
cI
CY
where we decomposed the E8 gauge field into a component along the CY 3-fold A and a component along the macroscopic four dimensions A = A dx . Moreover, d is the exterior derivative
operator defined on the CY while D + [A , ]. The {cI } form a basis of h1,1 cycles such
that

(3.13)
J = I J = JI .
cI
CY
Note that above and in the following, for simplicity, we suppress similar terms that can be induced
by a non-trivial gauge bundle at the hidden brane.
253
In Eq. (3.12) we have two source terms. The first one will be discussed in Section 4 as it can
lead to FayetIliopolous D-terms. The second,

A I ,
(y) trE8 (D A)
(3.14)
CY
is manifestly gauge invariant in respect to the unbroken 4d gauge transformations, and it depends
We introduce now a set of (local) complex coordinates along
on the (covariant) derivative of A.
n
the CY 3-fold, {z } (n = 1, 2, 3), and write
A = An dzn + A n d z n .
I
;
and the fact that ,a An = 0, it follows that Fy
C, C]
Recalling that A = A[,
of the superfield VyI where now
C, C),
VyI = I + I y V I (y)k I (, ;
and the superfield k I is found to be6

I
k = i trE8 An dzn A n d z n I ,
(3.15)
is a component
(3.16)
(3.17)
CY
which can be easily checked to be real. As in the universal case discussed above, the 5d Khler
moduli I are modified as
i
1
I = ey5 bI + (y)k I + AIy ,
(3.18)
2
2
and therefore

1

I
T = dy ey5 bI + iAIy + k I .
(3.19)
2
0
Likewise, the lowest component of the dilaton superfield also depends on the brane fields as
+ i (0).
C, C)
S0 = VCY (0) + I k I (, ;
(3.20)
We can use now the same reasoning as before to show that in the general case the effective 4d
Khler potential reads

T I + T I k I (, ;
.
C, C);
C, C)
K = K S0 + S0 I k I (, ;
(3.21)
In the case of the standard embedding with universal moduli only we have k = |C|2 and
Eq. (3.21) reduces to Eq. (3.9).
As a non-trivial test of this result let us consider the weak coupling limit I 0. In this case
the Khler potential is

+ K1,1 T I + T I k I
K = ln(S + S)
+ K1,1 (T + T ) K1,1 k I ,
ln(S + S)
I
(3.22)
6 In fact, we are determining k I up to a sum of a holomorphic and an anti-holomorphic term which can be shifted into
the definition of T I without affecting our conclusions below.
254
where K1,1 (T + T ) is the Khler potential for the Khler moduli. The last term on the r.h.s. can
be regarded as the brane fields Khler potential and reads

1,1 I
Kbrane = KI k = i trE8 An dzn A n d z n KI1,1 I .
(3.23)
CY
Finally, this can be rewritten as

J J
tr(A A)
CY
nm
tr
gCY gCY
An A m ,
Kbrane 3i
2
J
J
CY
(3.24)
where J is the Khler 2-form of the CY 3-fold. This expression matches the Khler potential that
can be obtained by a direct reduction of the 10d YangMills action, see e.g. [24]. Note that also
the cross term between the T -moduli and the bundle moduli mentioned in this reference follows
directly from the above Khler potential. There is no need for an additional contribution to K to
explain this term, in contrast to the expectations expressed in [24].
Notice that we can also determine a one-loop correction to the bundle moduli and matter
Khler potential (3.24), which reads

1-loop
tr F 2 1 tr R2 ,
Kbrane KS0 I k I i
(3.25)
tr(A A)
2
S0 + S0
CY
and depends on the second Chern characters of the vector and tangent bundles.
A few remarks are now in order. Firstly, we would like to stress that the way K depends on
the brane fields is not unexpected. A similar expression, the DeWolfeGiddings Khler potential [30], has been proposed in the context of type IIB compactifications with mobile D3 branes,
where it can naturally be matched to the kinetic terms derived from the BornInfeld action for
D3-branes.7 An important feature of our Khler potential is that the vector bundle moduli cannot
be disentangled from the geometric moduli. It is therefore excluded that the Khler potential for
the former be
p ln( + ),
(3.26)
as advocated e.g. in [24]. Moreover, the above form for the Khler potential preserves an important property already mentioned in Section 2.2, namely that
Ki j Ki Kj = 4,
(3.27)
where i, j run over all moduli and matter, even in the presence of the brane fields. We prove
this assertion in Appendix A. This is the heterotic M-theory counterpart of the no-scale structure
To do
C, C).
property in the type IIB case. Finally, let us comment on the structure of k I (, ;
this, we note that the bundle moduli and charged matter fields are introduced as
A = A()
+ C,
(3.28)
7 A relevant difference between our case and the DeWolfeGiddings Khler potential is that the latter is only known
to apply when h1,1 = 1. In that case the little Khler potential k is just the Khler potential of the compactification CY
evaluated at the position of the mobile D3 brane [30]. It is natural to expect that this expression extends to h1,1 > 1,
although it is not clear what the k I will be in that case.
255
where A()
is a solution of the zero-slope limit of the Hermitian YangMills equations, and the
charged fluctuations C span the Dolbeault cohomology H 1 (CY3 , VS ), where VS is the vector
bundle in the representation S at the y = 0 brane. Let us introduce a set of 1-forms {u a } forming
a basis of H 1 (CY3 , VS ). Then, C = u a C a and we see that k I consists of two pieces
+ GI (, )
C b C a ,
k I = kI (, )
ab
(3.29)
where kI depends only on the bundle moduli and is obtained by evaluating Eq. (3.17) for solutions of the zero-slope limit of the Hermitian YangMills equations, while

= i trE8 (ub u a ) I .
GIa b (, )
(3.30)
CY
It would be interesting to evaluate these expressions explicitly, e.g. using the gauge instanton
solution studied in [53]. We leave this for future work.
4. Anomalous U(1)s and FayetIliopolous terms
Recent searches for realistic heterotic model-building using non-trivial line-bundles naturally
lead to the presence of anomalous U(1)s in the gauge group content of both the visible and the
hidden branes (see [3133] and references therein). The effects of such anomalous U(1)X factors
was studied long ago in [54] in the absence of bulk M5 branes. Their results were recently
extended to include M5 branes in [32]. One of the facts pointed out in [32] is that, in the effective
4d theory, not only the dilaton but also the Khler moduli and the five-brane moduli will be
charged under the anomalous U(1)X . We will show below how their charges relate to each other.
Important to us is the fact that the U(1)X induce moduli-dependent FayetIliopolous terms,
leading to positive contributions to the potential energy. As it was first pointed out in the context
of type IIB flux compactifications, these FI terms have the potential of helping with uplifting
possible AdS4 vacua [9,17]. This possibility was also explored in the heterotic M-theory setup
in Ref. [26]. However, as we will argue in this section, this reference did not include all terms
relevant for the analysis.
To study the effects of the anomalous U(1)X we have once again to take into account the modification to the 5d Lagrangian presented in the previous section. This modification is supported
at the orbifold branes where the U(1)X is acting. We will assume, for notational simplicity, that
there is just one U(1)X at the visible brane, none in the hidden one. The generalization to multiple
anomalous U(1)s in both branes is straightforward. The modified Bianchi identities for the bulk
gauge multiplets lead to [54]
F I = dAI q I (y)AX dy,
where, according to Eq. (3.12), the charges

q I d A X = c1 (A X ),
cI
(4.1)
qI
are the topological invariants

(4.2)
cI
determined by the first Chern class c1 (A X ) of the vector bundle. In our superspace formalism
this corresponds to rewriting VyI as
VyI = I + I y V I q I (y)VX ,
(4.3)
256
where VX is the 4d vector superfield (in WessZumino gauge) transforming under the U(1)X as
VX VX + X + X .
(4.4)
FI
VyI
Moreover, the condition that

is a gauge invariant implies that
has to be invariant under
I
this transformation. Thus, we find that shifts under the anomalous U(1)X ,
I I + q I (y)X .
(4.5)
In component notation this means that AIy AIy + q I (y)X (x). This shift is necessary to cancel
out the brane localised anomaly [54].
As we just pointed out, the only effect of the anomalous U(1)X on the bulk superspace Lagrangian is to rewrite Vy , introducing brane-localised terms. The latter induce tadpole terms for
V I proportional to VX , which source non-zero modes

1
V I = q I VX (y) 1 |y|/ + V I ,
2
leading to
(4.6)
qI
VyI = I + I y V I
VX .
2
(4.7)
Here, V I stands for the y-dependent fluctuations which, since the vector superfield V I is
S 1 /Z2 -odd, vanish at low energies. Going down to 4d, the above non-zero modes will contribute
VX -dependent terms to the effective Khler potential. We will see now what those terms are.
To determine the effective Khler potential, using Eqs. (2.11) and (4.6) we find that we have
to perform the following integral

1/3

q
eK/3 = dy N +
VX
2

1/3
y
I
q
S0 + S0 I q I VX 2I (y) I + I
,
(4.8)
VX
2
0
where we suppressed all odd superfields and possible brane chiral matter contributions. Now, we
know that for VX = 0 and no brane chiral matter, the Khler potential depends only on the 4d
moduli S0 and T I . Therefore, if K(S0 + S0 ; T I + T I ) is the Khler potential for VX = 0, then
we find that turning on VX the Khler potential is

K = K S0 + S0 I q I VX ; T I + T I q I VX .
(4.9)
To obtain the D-term potential due to the FI terms we will assume that VX is normalized such
that the anomalous gauge kinetic term is

1
d 2 S0 WX WX + c.c.
(4.10)
4
We find then
VD =
1 (q I I KS0 + q I KI )2
.
2
S0 + S0
(4.11)
257
As pointed out in [31], the D-flatness condition q I (I KS0 + KI ) = 0 can be re-interpreted as a

1-loop corrected DonaldsonUhlenbeckYau integrability condition. This is intimately related to
the appearance of a 1-loop correction to the brane kinetic term (3.25). To illustrate the relevance
of FI D-term potentials for moduli stabilization, we study the tractable h1,1 = 1 case. Using
Eq. (2.19) we find that the D-term potential reads
1
8 2 q 2
.
(4.12)
(S0 + S0 )1/3 [(S0 + S0 )4/3 (S + S )4/3 ]2
It is interesting to consider the weak-coupling limit Re(T ) Re(S0 ). We find to leading order
VD =
9q 2
1
,
(4.13)
2 (S + S)(T + T )2
which clearly differs from the result found in [26]. Note that this difference is not qualitative,
because the physical force towards increasing both Re(S) and Re(T ) is also present in the D-term
potential of [26]. This force drives the theory to the phenomenologically interesting limit of
strong coupling at the hidden brane and might be a means of obtaining (supersymmetry breaking)
stable dS vacua. We will explore this possibility in a forthcoming publication.
We close this section recalling how the above reasoning is changed once we take into account
the effects of matter charged under the anomalous U(1)X acting at the visible brane. We denote
by C the chiral multiplets with U(1)X charges QC . Note that we are suppressing non-Abelian
indices. The main effect of the presence of the charged matter at the visible brane is to induce
the following change in the D-term

I
I
2
QC GC C (, )|C| (I KS0 + KI ).
DX q +
(4.14)
VD
As expected, the inclusion of the charged matter fields induces a potential that eventually can be
canceled by giving suitable VEVs to the charged scalars. If this happens, then (4.12) is of no use
for moduli stabilization.
5. Superpotential
In this section we discuss the perturbative and non-perturbative superpotentials needed for
moduli stabilization.
The perturbative superpotential can be derived in several ways. One is to use the well-known
fact that in the presence of non-zero four-form flux G, an effective 4d superpotential [5558]

W = G
(5.1)
M7
(M7 = CY3 S 1 /Z2 ) is generated. We will study first the superpotential dependence on the
brane fields. Solving the Bianchi identities one finds the following brane induced G

2
Gbrane dy trE8 A d A + A A A ,
(5.2)
3
which then leads to the superpotential [59]

2
W trE8 A A + A A A .
(5.3)
3
CY
258
Note that a priori this superpotential includes not only the bundle moduli and charged matter
but also the massive modes. In addition, it also depends on the complex structure moduli. As
pointed out in [59], once the condition of holomorphicity F = 0 is satisfied, the bundle moduli
parameterize flat directions of the superpotential. In fact,
W 2 trE8 ( A F),
(5.4)
vanishes if F () = 0.
It is instructive to see what kind of terms one obtains expanding around a given gauge bundle
A(),
A = A()
+ C A()
+ ua C a .
We find that
Wpert

2
trE8 C DC + C C C ,
3
(5.5)
(5.6)
already taking into account that the background gauge bundle is holomorphic. We recall that the
= 0. Then, only the last term contributes to the
charged massless fluctuations should satisfy DC
superpotential which is a cubic function of the charged matter fields. It is important to stress that
it also depends on the complex structure moduli z and bundle moduli .
In addition to (5.6), the four-flux G generally also induces a flux superpotential Wflux (z)
through (5.1), depending only on the complex structure moduli z. In this paper we will assume
that all complex structure moduli are frozen by the effect of Wflux . After integrating out these
moduli, a constant8 superpotential W0 remains, which can in 5d be seen as a ScherkSchwarz
twist.
The perturbative effects discussed sofar cannot alone provide for a stable vacuum. Fortunately,
a number of non-perturbative effects have the potential to deform the vacuum structure of heterotic M-theory. In particular, non-perturbative superpotentials can be induced by: (i) euclidean
M2 branes wrapping CY 2-cycles and stretching between the orbifold planes (and possible bulk
M5 branes) [22,51,60,61]; (ii) gaugino condensation in strongly coupled gauge sector(s) [6264],
typically at the hidden brane [6569]; (iii) and euclidean M5 branes wrapping the whole CY
3-fold at the orbifold fix points [70].
Regarding the third effect (M5 brane instantons), let us point out its similarity to the superpotential induced by the gauge instantons, in that both are controlled by the Si moduli. One might
wonder whether these are two different descriptions of the same phenomenon. For that reason,
and since the detailed form of the euclidean M5 brane superpotential is not fully understood [70],
we will neglect it and assume that the non-perturbative superpotential reads [51],9

i
Wnp =
Ai eaT + BebS ,
(5.7)
i
where the first term at the r.h.s. arises from the brane instantons while the second one is due to
gaugino condensation at the hidden brane. Here, T i = Ii T I is the Khler modulus associated
with the wrapped ith holomorphic two cycle and a is determined by the M2 brane tension. The
8 See however Ref. [58], which indicates that the flux superpotential may also depend on the Khler moduli.
9 See, however, Ref. [71], which advocates another form for the non-perturbative superpotential W .
np
259
constant b is determined by the field content at the hidden brane. On the other hand, the prefactors
Ai and B are, in general, functions of additional moduli. That is
Ai = Ai (, C, z),
B = B(, C, z),
(5.8)
where , C denote brane localised bundle moduli and matter and z stands for complex structure
moduli. In case we have additional M5 branes in the 5d bulk, Eq. (5.7) includes additional terms
due to M2 brane instantons connecting the different branes.
The bundle moduli dependence of Ai is in general not known but there are a few results that
can be stated. Firstly, let us recall that for Ai to be non-vanishing, the M2 brane instanton must
be wrapping an isolated genus zero holomorphic curve (times the orbifold interval). Moreover,
the vector bundle pulled back to the wrapped curve must be trivial which implies that since
T i Ii T I , then [72]

Ii q I c1 (VX ) Ii I = 0,
(5.9)
must be satisfied. This has the interesting implication that Ai (, C) must be a singlet.10
The prefactor Ai is essentially determined by the Pfaffian of the chiral Dirac operator evaluated in the background of the gauge bundle pulled back to the holomorphic curve. In a series of
papers [7476], Buchbinder et al. were able to explicitly calculate the bundle moduli dependence
of Ai in a number of examples with vector bundles constructed on elliptically fibered CY 3-folds.
They found that Ai is a non-vanishing homogeneous polynomial function of certain bundle moduli, dubbed transition moduli since they trigger small instanton transitions at the ith curve as they
approach zero. A nice explanation of these and related issues can be found in [24].
Finally, let us point out that in this paper we assume the gauge bundle at the hidden brane to be
trivial. Therefore, in this case the gaugino condensate 1-loop determinant B is just a holomorphic
function of the complex structure moduli.
6. Supersymmetric (AdS) vacua
In this section we search for supersymmetric vacua. In view of the new understanding of the
Khler potential with brane multiplets obtained in Section 3.1, we expect our results to be distinct
from the ones in Ref. [24]. To keep our discussion as simple as possible we take h1,1 = 1 and
no bulk M5 branes. Furthermore, we consider the gauge bundle at the hidden brane to be trivial
and exclude the possibility of anomalous U(1) factors acting at the visible brane. We follow [24],
which in the spirit of KKLT [1] assumes that the complex structure moduli are stabilized in
values which are approximately independent of the other moduli. Then, the superpotential reads
W = W0 + Wpert (, C) + A(, C)eaT + BebS .
(6.1)
Supersymmetric anti-de Sitter vacua can be obtained whenever

Di W = 0
(6.2)
10 We see that this presents a solution to the possible incomparability between D-term potentials and non-perturbative
superpotentials different from the one found in the case of gaugino condensation on D7-branes in type IIB compactifications. In this case, it was shown by an explicit calculation [73] that the prefactor can be charged under the anomalous
U(1)X in such a way as to make the superpotential gauge invariant. In our case, the prefactor is always a singlet and
therefore must vanish in case the exponential factor in the superpotential is not U(1)X invariant.
260
can be satisfied, with i running over all moduli.

We observe first that DC a W = 0 can be satisfied at C = 0 in case A(, C) is regular at C = 0,
which we take for granted in the following. In fact the latter implies that C a W = 0. Then, since
(cf. (3.30))
C b
Ka C a K = KI GIa b (, )
(6.3)
also vanishes for C = 0, we find that Ka W = 0 at that point and DC a W = 0 as we wanted to

show.
Our search for supersymmetric (AdS) vacua reduces now to studying the superpotential
W = W0 + A()eaT + BebS ,
(6.4)
where A() = A(, C = 0). Recall that in Section 5 we pointed out that Wpert (, C = 0) would
not depend on the vector bundle moduli once we impose the vector bundle to be holomorphic
(F = 0). We will assume this to be the case in the following. It is then not difficult to find that,
due to the specific way the bundle moduli and the Khler moduli mix in Eq. (3.21), we have
D W = eaT ( ak )A() k DT W.
(6.5)
in a supersymmetric vacuum (with DT W = 0) is thus found by solving the

The value of
reduced problem11
D W() W() + k W() = 0,
(6.6)
A()1/a
with the reduced superpotential W() =

and the reduced Khler potential being
It is important to stress that the value of in the supersymmetric vacuum is fully
k(, ).
independent of the geometric moduli S and T . This is in contrast to the findings of [24].
Moreover, the above supersymmetry condition on the bundle moduli has an interesting physical interpretation, as we explain now. This relies on understanding the origin of the -dependence
of the prefactor A(). Adapting an argument given in [78,79], in the type IIB context, we find
that this dependence can be traced back to the correction that the volume of the 3-cycle wrapped
by the brane instanton suffers in the presence of the non-trivial brane localised gauge bundle.12
In the present case, with h1,1 = 1, the total volume reads
V3 dy V1,1 = V30 + V30 ,

(6.7)
with the undeformed volume being (cf. (3.11))
V30 = T + T k(, ),
(6.8)
and the correction

a 1 lnA()2 + const.
V30 = k(, )
(6.9)
Clearly, the supersymmetry condition on the gauge bundle, D W() = 0, now gains a suggestive
new face
= 0.
V 0 (, )
(6.10)
3
11 The analog result, in the context of type IIB supersymmetric vacua with D3-branes, was obtained in [77].
12 In the type IIB case [78,79], it is the volume of wrapped 4-cycles which is deformed in the presence of a D3-brane.
261
We thus see that in a supersymmetric vacuum, the bundle moduli extremize the volume of the
cycles wrapped by brane instantons.
In case a solution, = 0 to Eq. (6.6) can be found, we are left with a set of two (complex)
coupled equations involving just the moduli S and T . For simplicity, in the following we work
with the moduli S and S0 = S + T . The conditions Di W = 0 for a supersymmetric vacuum
can be combined to give

b
aA0
S + S 1/3
S + ln
1
,
S0 = 1 +
(6.11)
a
a
bB
S0 + S0
where A0 A(0 ), and

a
a
S + S 1/3
a
a
W0 = 1 +
+
(S0 + S0 )
+
(S + S )
b 4
b 4
S0 + S0
a
A0 e (S0 S ) .
(6.12)
Following [24], we have a 102 , b 5 and B 106 Mp3 . On the other hand, acceptable phenomenology imposes that the brane tension is bounded as 1 and Re(S0 ) 1. With these
values we find that there is an AdS4 supersymmetric vacuum with

b
Re(S ) 1
(6.13)
Re(S0 ) Re(S0 ),
a
if we fine tune the flux superpotential W0 to be
|W0 | |B|eb Re(S ) 108 Mp3 .
We note, however, that the value of Re(S ) is by far too large to be phenomenologically acceptable. In fact, the problem lies on the ratio (b/a) 5 102 being too small. That is, the force
exerted by the membrane instanton against the growth of the wrapped cycle is too strong when
compared to the effect of gaugino condensation. The situation would improve if we manage to
bring b/a to be of order one. It is conceivable to increase b by breaking down the E8 at the
hidden brane to some subgroup by using a non-trivial gauge bundle. To decrease a by the needed
amount, one would have to decrease the size of the holomorphic curve wrapped by the brane
instanton by one order of magnitude. It is interesting to note that the vacuum (6.13) depends
only very weakly on A0 = A(0 ) and therefore, in principle, we cannot choose the visible sector
gauge bundle in such a way as to move the susy vacuum into a phenomenologically interesting
corner of the moduli space.
Finally, it is straightforward to determine the value of the potential energy at the vacuum. We
find
3
a |W0 |2
1012 Mp4 ,
Vmin
(6.14)
b
Mp2
a rather deep AdS vacuum.
Before closing this section let us comment on the assumptions we made in obtaining the
above results. The first assumption was that the complex structure moduli can be fixed due to
the effect of a flux superpotential with values that do not depend on the other moduli , T , S .
Clearly, if this was not to be the case, the above results could be strongly modified. We refer the
reader to [12] for a nice discussion of this possibility. Secondly, we assumed that the charged
scalars can be stabilized at C = 0. This relies on the fact that A(z, , C) is a gauge singlet and
262
smooth at C = 0 and on the assumption that the gauge bundle at the hidden brane is trivial.
Non-trivial bundles may allow for more exotic possibilities, with a charged matter dependent
prefactor B, obstructing supersymmetric vacua. The third assumption was to exclude the possibility that Eq. (6.6) has no solution for holomorphic vector bundles satisfying (the zero-slope
limit of) the Hermitian YangMills equations. In fact, the possibility that non-perturbative corrections might destabilize heterotic vacua was raised long ago [8082] (see also [72,83]). However,
we should note that [24] gave reasonable arguments against that possibility at least for the class
of compactifications studied therein.
7. Discussion
In this paper we performed a discussion of the ingredients involved in moduli stabilization
in heterotic M-theory. That is, we derived expressions for the low-energy Khler potential and
superpotentials as functions of the dilaton S and Khler moduli T I , as well as of the bundle
moduli and charged matter C. For this purpose we used a superfield description of the relevant couplings of 5d heterotic M-theory, which in this paper we expanded beyond the standard
embedding.
One of the main achievements of this paper is the first time derivation of the way the bundle
moduli and charged matter enter the Khler potential. We found in Section 3 that the Khler
potential of heterotic M-theory (and its weak coupling limit, the heterotic string) is similar in
form to the De WolfeGiddings Khler potential [30] in type IIB string, but unlike the latter it
is defined also for h1,1 > 0. Using this and the perturbative and non-perturbative superpotentials
discussed in Section 5 we performed a study of moduli stabilization, searching for supersymmetric vacua (in a setup with h1,1 = 1 for simplicity). Let us recapitulate our main findings.
The configuration of the gauge bundle is determined by solving a reduced supersymmetry
condition, which involves only the bundle moduli . We have argued that this condition has a
simple physical interpretation. Namely, it picks extrema of the volume of the isolated holomorphic curves (wrapped by membrane instantons) in respect to the bundle moduli. If a solution
0 ) is found, we can determine the values of the geometric moduli S and T I in the vacuum,
A(
0 ) background. We found that, at least
by solving the corresponding susy conditions in the A(
in the h1,1 = 1 case without five-branes in the bulk, at the supersymmetric AdS4 vacuum the
geometric moduli take values far from the phenomenologically acceptable corner of the moduli
space.
Anomalous U(1)s are predicted in recent semi-realistic heterotic model-building. In this paper
we derived the FI D-terms induced by the anomalous U(1)s, extending to strong coupling the
results of [31]. As we pointed out in Section 4, the D-term potential drives the theory towards
strong coupling and may therefore be of use in moduli stabilization. We have also seen that the
appearance of a charged Khler modulus T i implies the vanishing of the membrane instanton
superpotential associated with T i . We note that, interestingly, in that case a contracting force
(the membrane instanton) is traded for an expanding one (the FI D-term).
To obtain realistic vacua, the AdS supersymmetric minima discussed in this paper have not
only to be uplifted to de Sitter vacua, but also the VEVs of the geometric moduli must be changed.
Several ways of achieving this have already been proposed in the literature, mostly in the realm
of type IIB compactifications. These include: D-terms from anomalous U(1)s [26] ([9,17,20] in
in KKLT [1]); de Sitter vacua from matter
the type IIB context); anti M5-branes [28,84] (D3
superpotentials [19]; and loop corrections [27] (or the combination of and loop corrections
[11,1316,18,20,21] in type IIB compactifications). We will investigate some of these and other
263
possibilities in a forthcoming publication, where we will address the construction of stable de

Sitter vacua in 5d heterotic M-theory (see [85] for an explicit example of dS stable vacua in 5d
gauged supergravity at one loop).
Since to obtain de Sitter vacua we will have to find supersymmetry breaking minima of the
scalar potential, the question arises as to whether the gauge bundle is still determined independently of the geometric moduli, as in the AdS vacua studied in Section 6. Remarkably, it can be
shown (when C a = 0) that if the scalar potential has an extremum in respect to S and T , i.e.
S V = T V = 0, then the reduced flatness condition, Eq. (6.6), will imply that V also has an
extremum in respect to the bundle moduli, V = 0, and we expect this to be the case also for
h1,1 > 1. We thus see that the gauge bundle in a de Sitter vacuum with C = 0 is exactly the same
as in the supersymmetric AdS vacuum, which is determined by the reduced flatness condition.
However, it remains to be seen what happens once we allow for vacua where the charged scalars
receive non-vanishing VEVs as in [25], or include anti M5 branes in the 5d bulk [28,84]. In
this case we cannot exclude that the configuration of the gauge bundle will be dependent on the
geometric moduli.
Acknowledgements
This work was started while F.P.C. was visiting the Institut fr Theoretische Physik, Heidelberg, Germany, with the support of Schwerpunktprogramm Stringtheorie. . . (5PP1096). We
are grateful to M. Olechowski and Z. Tavartkiladze for interesting discussions. F.P.C. is supported by FCT through the grant SFRH/BPD/20667/2004.
Appendix A. The structure of the Khler potential
It is well known that in the type IIB compactifications the no-scale structure is preserved even
in the presence of a mobile D3 brane. Here we prove the heterotic M-theory counterpart of this
result, namely that
Ki j Ki Kj = 4,
(A.1)
where i, j run over all moduli, even in the presence of the brane fields. Our sole assumption is
that in the absence of brane fields we have Ki (X i + X i ) = n, where X i = T I , S0 . In case of
interest we have n = 4, but our analysis also includes the type IIB case with n = 3. We will now
prove this assumption in the 5d heterotic M-theory case. The crucial fact is that
K = 3 ln ,
(A.2)
where

T I + T I
,
= 0 F 0
S0 + S0
and 0 is the well known weak coupling result

03 = (S0 + S0 )dI J K T I + T I T J + T J T K + T K .
(A.3)
(A.4)
In other words, F (0) = 1. Note that, without any loss of generality, we chose I = 0 for I = 0.
Using Eq. (A.3), it is then straightforward to find that

Ki X i + X i = 4.
(A.5)
264
As we will show now, once we include brane fields the same result follows due to the following
structure of the Khler potential

T I + T I k I (, ;
.
C, C);
C, C)
K = K S0 + S0 I k I (, ;
(A.6)
A word on the notation: We denote XI T I and X I =1 S0 , while i runs over all chiral
superfields. In addition, k I =1 I k I . We have thus that

KI X I + X I k I = n.
(A.7)
It follows then that

KI j X I + X I k I = Kj ,
(A.8)
and therefore

Ki j Kj = Ii X I + X I k I .
(A.9)
Using Eq. (A.7) we finally find that
Ki j Ki Kj = n,
(A.10)
as we wanted to show.
References
[1] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, Phys. Rev. D 68 (2003) 046005,
hep-th/0301240.
[2] P. Horava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys. B 460 (1996) 506,
hep-th/9510209.
[3] P. Horava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl. Phys. B 475 (1996) 94,
hep-th/9603142.
[4] E. Witten, Strong coupling expansion of CalabiYau compactification, Nucl. Phys. B 471 (1996) 135, hepth/9602070.
[5] T. Banks, M. Dine, Couplings and scales in strongly coupled heterotic string theory, Nucl. Phys. B 479 (1996) 173,
hep-th/9605136.
[6] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, The universe as a domain wall, Phys. Rev. D 59 (1999) 086001,
hep-th/9803235.
[7] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Heterotic M-theory in five dimensions, Nucl. Phys. B 552 (1999)
246, hep-th/9806051.
[8] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66 (2002)
106006, hep-th/0105097.
[9] C.P. Burgess, R. Kallosh, F. Quevedo, De Sitter string vacua from supersymmetric D-terms, JHEP 0310 (2003) 056,
hep-th/0309187.
[10] A. Saltman, E. Silverstein, The scaling of the no-scale potential and de Sitter model building, JHEP 0411 (2004)
066, hep-th/0402135.
[11] V. Balasubramanian, P. Berglund, Stringy corrections to Khler potentials, SUSY breaking, and the cosmological
constant problem, JHEP 0411 (2004) 085, hep-th/0408054.
[12] K. Choi, A. Falkowski, H.P. Nilles, M. Olechowski, S. Pokorski, Stability of flux compactifications and the pattern
of supersymmetry breaking, JHEP 0411 (2004) 076, hep-th/0411066.
[13] K. Bobkov, Volume stabilization via alpha corrections in type IIB theory with fluxes, JHEP 0505 (2005) 010,
hep-th/0412239.
[14] V. Balasubramanian, P. Berglund, J.P. Conlon, F. Quevedo, Systematics of moduli stabilisation in CalabiYau flux
compactifications, JHEP 0503 (2005) 007, hep-th/0502058.
[15] G. von Gersdorff, A. Hebecker, Khler corrections for the volume modulus of flux compactifications, Phys. Lett.
B 624 (2005) 270, hep-th/0507131.
265
[16] M. Berg, M. Haack, B. Kors, On volume stabilization by quantum corrections, Phys. Rev. Lett. 96 (2006) 021601,
hep-th/0508171.
[17] A. Achucarro, B. de Carlos, J.A. Casas, L. Doplicher, De Sitter vacua from uplifting D-terms in effective supergravities from realistic strings, JHEP 0606 (2006) 014, hep-th/0601190.
[18] S.L. Parameswaran, A. Westphal, De Sitter string vacua from perturbative Khler corrections and consistent
D-terms, JHEP 0610 (2006) 079, hep-th/0602253.
[19] O. Lebedev, H.P. Nilles, M. Ratz, De Sitter vacua from matter superpotentials, Phys. Lett. B 636 (2006) 126, hepth/0603047.
[20] E. Dudas, Y. Mambrini, Moduli stabilization with positive vacuum energy, JHEP 0610 (2006) 044, hep-th/0607077.
[21] A. Westphal, De Sitter string vacua from Khler uplifting, JHEP 0703 (2007) 102, hep-th/0611332.
[22] G. Curio, A. Krause, G-fluxes and non-perturbative stabilisation of heterotic M-theory, Nucl. Phys. B 643 (2002)
131, hep-th/0108220.
[23] G. Curio, A. Krause, Enlarging the parameter space of heterotic M-theory flux compactifications to phenomenological viability, Nucl. Phys. B 693 (2004) 195, hep-th/0308202.
[24] E.I. Buchbinder, B.A. Ovrut, Vacuum stability in heterotic M-theory, Phys. Rev. D 69 (2004) 086010, hep-th/
0310112.
[25] M. Becker, G. Curio, A. Krause, De Sitter vacua from heterotic M-theory, Nucl. Phys. B 693 (2004) 223, hep-th/
0403027.
[26] E.I. Buchbinder, Raising anti-de Sitter vacua to de Sitter vacua in heterotic M-theory, Phys. Rev. D 70 (2004)
066008, hep-th/0406101.
[27] L. Anguelova, D. Vaman, R 4 corrections to heterotic M-theory, Nucl. Phys. B 733 (2006) 132, hep-th/0506191.
[28] V. Braun, B.A. Ovrut, Stabilizing moduli with a positive cosmological constant in heterotic M-theory, JHEP 0607
(2006) 035, hep-th/0603088.
[29] G. Curio, A. Krause, S-track stabilization of heterotic de Sitter vacua, hep-th/0606243.
[30] O. DeWolfe, S.B. Giddings, Scales and hierarchies in warped compactifications and brane worlds, Phys. Rev. D 67
(2003) 066008, hep-th/0208123.
[31] R. Blumenhagen, G. Honecker, T. Weigand, Loop-corrected compactifications of the heterotic string with line bundles, JHEP 0506 (2005) 020, hep-th/0504232.
[32] R. Blumenhagen, S. Moster, T. Weigand, Heterotic GUT and standard model vacua from simply connected Calabi
Yau manifolds, Nucl. Phys. B 751 (2006) 186, hep-th/0603015.
[33] R. Tatar, T. Watari, Proton decay, Yukawa couplings and underlying gauge symmetry in string theory, Nucl. Phys.
B 747 (2006) 212, hep-th/0602238.
[34] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, 4D superfield reduction of 5D orbifold SUGRA and heterotic
M-theory, Nucl. Phys. B 751 (2006) 222, hep-th/0602173.
[35] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Superfield approach to 5D conformal SUGRA and the radion,
Nucl. Phys. B 709 (2005) 141, hep-th/0408138.
[36] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, (BPS) FayetIliopolous terms in 5D orbifold SUGRA, Phys.
Lett. B 613 (2005) 83, hep-th/0410281.
[37] H. Abe, Y. Sakamura, Superfield description of 5D supergravity on general warped geometry, JHEP 0410 (2004)
013, hep-th/0408224.
[38] H. Abe, Y. Sakamura, Roles of Z(2)-odd N = 1 multiplets in off-shell dimensional reduction of 5D supergravity,
Phys. Rev. D 75 (2007) 025018, hep-th/0610234.
[39] T. Kugo, K. Ohashi, Supergravity tensor calculus in 5D from 6D, Prog. Theor. Phys. 104 (2000) 835, hep-ph/
0006231.
[40] T. Kugo, K. Ohashi, Off-shell d = 5 supergravity coupled to matter-YangMills system, Prog. Theor. Phys. 105
(2001) 323, hep-ph/0010288.
[41] T. Fujita, T. Kugo, K. Ohashi, Off-shell formulation of supergravity on orbifold, Prog. Theor. Phys. 106 (2001) 671,
hep-th/0106051.
[42] T. Kugo, K. Ohashi, Superconformal tensor calculus on orbifold in 5D, Prog. Theor. Phys. 108 (2002) 203, hep-th/
0203276.
[43] J. Bagger, C. Xiong, N = 2 nonlinear sigma models in N = 1 superspace: Four and five dimensions, hep-th/
0601165.
[44] Z. Lalak, Low energy supersymmetry in warped brane worlds, hep-th/0109074.
[45] A. Lukas, B.A. Ovrut, D. Waldram, Non-standard embedding and five-branes in heterotic M-theory, Phys. Rev.
D 59 (1999) 106005, hep-th/9808101.
[46] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Non-perturbative vacua and particle physics in M-theory, JHEP 9905
(1999) 018, hep-th/9811168.
266
[47] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Holomorphic vector bundles and non-perturbative vacua in
M-theory, JHEP 9906 (1999) 034, hep-th/9901009.
[48] R. Donagi, B.A. Ovrut, T. Pantev, D. Waldram, Standard-model bundles on non-simply connected CalabiYau
threefolds, JHEP 0108 (2001) 053, hep-th/0008008.
[49] J.P. Derendinger, R. Sauser, A five-brane modulus in the effective N = 1 supergravity of M-theory, Nucl. Phys.
B 598 (2001) 87, hep-th/0009054.
[50] M. Brandle, A. Lukas, Five-branes in heterotic brane-world theories, Phys. Rev. D 65 (2002) 064024, hep-th/
0109173.
[51] G.W. Moore, G. Peradze, N. Saulina, Instabilities in heterotic M-theory induced by open membrane instantons,
Nucl. Phys. B 607 (2001) 117, hep-th/0012104.
[52] T. Kugo, S. Uehara, Improved superconformal gauge conditions in the N = 1 supergravity YangMills matter
system, Nucl. Phys. B 222 (1983) 125.
[53] J. Gray, A. Lukas, Gauge five brane moduli in four-dimensional heterotic models, Phys. Rev. D 70 (2004) 086003,
hep-th/0309096.
[54] A. Lukas, K.S. Stelle, Heterotic anomaly cancellation in five dimensions, JHEP 0001 (2000) 010, hep-th/9911156.
[55] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69, hep-th/9906070;
S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 608 (2001) 477, Erratum.
[56] K. Behrndt, S. Gukov, Domain walls and superpotentials from M theory on CalabiYau three-folds, Nucl. Phys.
B 580 (2000) 225, hep-th/0001082.
[57] M. Becker, D. Constantin, A note on flux induced superpotentials in string theory, JHEP 0308 (2003) 015, hepth/0210131.
[58] L. Anguelova, K. Zoubos, Flux superpotential in heterotic M-theory, Phys. Rev. D 74 (2006) 026005, hep-th/
0602039.
[59] E. Witten, New issues in manifolds of SU(3) holonomy, Nucl. Phys. B 268 (1986) 79.
[60] E. Lima, B.A. Ovrut, J. Park, R. Reinbacher, Non-perturbative superpotential from membrane instantons in heterotic
[61] E. Lima, B.A. Ovrut, J. Park, Five-brane superpotentials in heterotic M-theory, Nucl. Phys. B 626 (2002) 113,
hep-th/0102046.
[62] M. Dine, R. Rohm, N. Seiberg, E. Witten, Gluino condensation in superstring models, Phys. Lett. B 156 (1985) 55.
[63] V.S. Kaplunovsky, J. Louis, Model independent analysis of soft terms in effective supergravity and in string theory,
Phys. Lett. B 306 (1993) 269, hep-th/9303040.
[64] A. Brignole, L.E. Ibanez, C. Munoz, Towards a theory of soft terms for the supersymmetric Standard Model, Nucl.
Phys. B 422 (1994) 125, hep-ph/9308271;
A. Brignole, L.E. Ibanez, C. Munoz, Towards a theory of soft terms for the supersymmetric Standard Model, Nucl.
Phys. B 436 (1995) 747, Erratum.
[65] P. Horava, Gluino condensation in strongly coupled heterotic string theory, Phys. Rev. D 54 (1996) 7561, hep-th/
9608019.
[66] H.P. Nilles, M. Olechowski, M. Yamaguchi, Supersymmetry breaking and soft terms in M-theory, Phys. Lett. B 415
(1997) 24, hep-th/9707143.
[67] Z. Lalak, S. Thomas, Gaugino condensation, moduli potential and supersymmetry breaking in M-theory models,
Nucl. Phys. B 515 (1998) 55, hep-th/9707223.
[68] A. Lukas, B.A. Ovrut, D. Waldram, Gaugino condensation in M-theory on S 1 /Z2 , Phys. Rev. D 57 (1998) 7529,
hep-th/9711197.
[69] A. Lukas, B.A. Ovrut, D. Waldram, Five-branes and supersymmetry breaking in M-theory, JHEP 9904 (1999) 009,
hep-th/9901017.
[70] E.I. Buchbinder, Derivative F-terms from heterotic M-theory five-brane instanton, Phys. Lett. B 645 (2007) 281,
hep-th/0611119.
[71] L. Carlevaro, J.P. Derendinger, Five-brane thresholds and membrane instantons in four-dimensional heterotic
[72] J. Distler, B.R. Greene, Aspects of (2, 0) string compactifications, Nucl. Phys. B 304 (1988) 1.
[73] M. Haack, D. Krefl, D. Lust, A. Van Proeyen, M. Zagermann, Gaugino condensates and D-terms from D7-branes,
JHEP 0701 (2007) 078, hep-th/0609211.
[74] E. Buchbinder, R. Donagi, B.A. Ovrut, Vector bundle moduli and small instanton transitions, JHEP 0206 (2002)
054, hep-th/0202084.
[75] E.I. Buchbinder, R. Donagi, B.A. Ovrut, Superpotentials for vector bundle moduli, Nucl. Phys. B 653 (2003) 400,
hep-th/0205190.
267
[76] E.I. Buchbinder, R. Donagi, B.A. Ovrut, Vector bundle moduli superpotentials in heterotic superstrings and
M-theory, JHEP 0207 (2002) 066, hep-th/0206203.
[77] O. DeWolfe, L. McAllister, G. Shiu, B. Underwood, D3-brane vacua in stabilized compactifications, hep-th/
0703088.
[78] S.B. Giddings, A. Maharana, Dynamics of warped compactifications and the shape of the warped landscape, Phys.
Rev. D 73 (2006) 126003, hep-th/0507158.
[79] D. Baumann, A. Dymarsky, I.R. Klebanov, J. Maldacena, L. McAllister, A. Murugan, On D3-brane potentials in
compactifications with fluxes and wrapped D-branes, JHEP 0611 (2006) 031, hep-th/0607050.
[80] J.R. Ellis, C. Gomez, D.V. Nanopoulos, M. Quiros, World sheet instanton effects on no scale structure, Phys. Lett.
B 173 (1986) 59;
J.R. Ellis, C. Gomez, D.V. Nanopoulos, M. Quiros, World sheet instanton effects on no scale structure, Phys. Lett.
B 174 (1986) 465, Erratum.
[81] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string world sheet, Nucl. Phys. B 278
(1986) 769.
[82] M. Dine, N. Seiberg, X.G. Wen, E. Witten, Nonperturbative effects on the string world sheet. 2, Nucl. Phys. B 289
(1987) 319.
[83] J. Distler, Resurrecting (2, 0) compactifications, Phys. Lett. B 188 (1987) 431.
[84] J. Gray, A. Lukas, B. Ovrut, Perturbative anti-brane potentials in heterotic M-theory, hep-th/0701025.
[85] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Radion stabilization in 5D SUGRA, Nucl. Phys. B 763 (2007)
247, hep-th/0608058.

String junctions in curved backgrounds, their stability

and dyon interactions in SYM
Konstadinos Sfetsos , Konstadinos Siampos
Department of Engineering Sciences, University of Patras, 26110 Patras, Greece
Received 1 November 2007; accepted 8 January 2008
Abstract
We provide a systematic construction of string junctions in curved backgrounds which are relevant in
computing, within the gauge/gravity correspondence, the interaction energy of heavy dyons, notably of
quarkmonopole pairs, in strongly coupled SYM theories. To isolate the configurations of physical interest
we examine their stability under small fluctuations and prove several general statements. We present all
details, in several examples, involving non-extremal and multicenter D3-brane backgrounds as well as the
Rindler space. We show that a string junction could be perturbatively stable even in branches that are not
energetically the most favorable ones. We present a mechanical analog of this phenomenon.
1. Introduction
String junctions [1,2] are of particular interest in string theory for several reasons: Due to their
BPS nature they can be used to build supersymmetric string networks [37] and they are useful
for the description of BPS states in SYM theories [8] and for applications to gauge symmetry
enhancement in string theory [9]. The scattering of string modes at string junctions in flat spacetime has been analyzed in [10]. Particularly, important for the purposes in the present paper is
the fact that since string junctions connect strings of different type they can be used, within the
gauge/gravity correspondence [11], to compute the interaction energy of dyons and in particular
of heavy quarkmonopole pairs [12].
E-mail addresses: sfetsos@upatras.gr (K. Sfetsos), ksiampos@upatras.gr (K. Siampos).

doi:10.1016/j.nuclphysb.2008.01.008
K. Sfetsos, K. Siampos / Nuclear Physics B 797 (2008) 268294
269
The purpose of the present paper is to formulate and construct the string junctions appropriate for computing the interaction potentials of heavy dyons at strong coupling, within the
gauge/gravity correspondence. This is done for a large class of curved backgrounds away from
the conformal point and with reduced or no supersymmetry at all. Having constructed string
junction solutions does not imply that the corresponding dyon interaction potentials are physical.
One has to at least investigate perturbative stability which will appropriately restrict the parametric space in which the solutions are physically relevant. Such investigations were exhaustively
performed [13,14] for single string solutions useful in computing the heavy quarkantiquark
potential. Given these works and in comparison to them we will find that the stability analysis
of junctions leads to expected as well as unexpected results, the latter being counterintuitive to
stability arguments based on energy considerations.
The organization of this paper is as follows: In Section 2 we formulate string junctions in a
quite general class of curved backgrounds. We derive general formulae from which the interaction energy of a dyon pair is computed. In Section 3 we consider in great detail the stability
analysis under small fluctuations of these string junctions aiming at discovering the physically
relevant regions. We pay particular attention to the precise formulation of the boundary and
the matching conditions at the junction point and derive general statements that isolate the
boundaries of stability for all types of perturbative fluctuations. In Section 4, we present several
examples of string junctions using D3-black branes and multicenter D3-brane solutions which
within the AdS/CFT correspondence are relevant for N = 4 SYM at finite temperature and at
generic points in the Coulomb branch of the theory, respectively. In addition, we present the
details of string junctions in Rindler space, given its relevance in a variety of black hole backgrounds near the horizon. In Section 5, we apply the outcome of the general stability analysis
of Section 3 to the examples of Section 4. We present our conclusions in Section 5. In Appendix A work out the stability of a classical mechanical system which also exhibits an unexpected
perturbative stability similar to the one we found with string junctions.
2. The classical solutions
In this section we develop and present the general setup of the AdS/CFT calculation of the
static potential of a heavy static dyondyon pair with general NS and RR charges. These computations involve three strings and therefore we will be based to the well-known results for the
potential of heavy quarkantiquark pairs of [15] for the conformal case, as extended in [16]
for general backgrounds, and to the computation of the interaction energy of a heavy quark
monopole pair in the conformal case [12].
We consider a general diagonal metric of Lorentzian signature of the form
ds 2 = Gtt dt 2 + Gyy dy 2 + Guu du2 + Gxx dx 2 + G d 2 + .
(2.1)
Here, y denotes the (cyclic) coordinate along which the spatial side of the Wilson loop extends,
u denotes the radial direction playing the rle of an energy scale in the dual gauge theory and
extending from the UV at u down to the IR at some minimum value umin determined by
the geometry, x stands for a generic cyclic coordinate, stands for a generic coordinate on which
the metric components may depend on and the omitted terms involve coordinates that fall into
one of the two latter classes without mixing terms. It is convenient to introduce the functions
g(u, ) = Gtt Guu ,
fy (u, ) = Gtt Gyy ,
f (u, ) = Gtt G ,
h(u, ) = Gyy Guu .
fx (u, ) = Gtt Gxx ,

(2.2)
270
If the conformal limit can be taken (this is the case in all examples in the present paper except
the one involving the Rindler space) we can approximate the metric by that for AdS5 S 5 with
radii (in string units) R = (4gs N )1/4 , with gs being the string coupling. In this limit we have
the leading order expressions
g h 1,
fx fy u4 /R 4 ,
f u2 ,
as u .
(2.3)
A string junction consists of three co-planar strings joined at a point as depicted in Fig. 1. The
string labelled as 1 has (p, q) NS and RR charges, the second labelled as 2 has (p , q ) charges,
whereas the third straight string labelled by 3 has (m, n) = (p + p , q + q ) due to the charge
conservation at the junction point. From a microscopic point of view the strings that interact
are the fundamental F - and D-string with charges (1, 0) and (0, 1), respectively. Any other
charge combination should be achievable by performing in this pair an SL(2, R) transformation
which gives the condition pq qp = 1 [9]. The first two strings end at the boundary of
the spacetime at u = , whereas the straight third string ends at the minimum value umin of
the spacetime allowed by the geometry. All three strings meet at the junction point at u = u0 .
Within the framework of the AdS/CFT correspondence, the interaction potential energy of the
(p, q) dyon is given by

eiET = W (C) = eiSp,q [C] ,
(2.4)
where

Tp,q
d d det g , g = G x x ,
Sp,q [C] =
(2.5)
2
is the NambuGoto action for a string propagating in the dual supergravity background whose
endpoints
trace a rectangular contour along one temporal and one space direction, Tp,q =

p 2 + q 2 /gs2 and similar expressions hold for the strings 2 and 3. We will not consider a contribution from the WessZumino terms either because these terms do not exist all or have vanishing
contribution within our ansatz for solutions below.
We next fix reparametrization invariance for each string by choosing
t = ,
u = ,
(2.6)
we assume translational invariance along t , and we consider the embedding

y = y(u),
x = 0,
= 0 = const.,
rest = const.,
(2.7)
supplemented by the boundary condition

u(L1 ) = u(L2 ) = ,
(2.8)
appropriate for a (p, q) dyon placed at y = L1 and a (p , q ) dyon placed at y = L2 . The string
with charges (m, n) extending from the junction point to u = umin is straight as y = const. is
always a solution of the equations of motion (see (2.11) below). In the ansatz (2.7), the constant
value 0 of the non-cyclic coordinate must be consistent with the corresponding equation of
motion. This requires that [13]
g(u, )|=0 = fy (u, )|=0 = 0.
For the ansatz given above, the NambuGoto action reads

Tp,q T
Sp,q =
du g(u) + fy (u)y 2 ,
2
(2.9)
(2.10)
271
Fig. 1. Junction configuration: We denote by Fi , i = 1, 2, 3 (Fpq , etc., in the text) the forces exerted at each string at the
junction point u0 . The turning points ui , i = 1, 2 for the strings 1 and 2, are not depicted and lie at the right of u0 . We
always have umin < u0 , but the ui s may or may not be larger than umin .
where T denotes the temporal extent of the Wilson loop, the prime denotes a derivative with
respect to u while g(u) g(u, 0 ) and fy (u) fy (u, 0 ) are the functions in (2.2) evaluated at
the chosen constant value 0 of . Similar actions are also considered for the (p , q ) string, as
well as for the straight (m, n) string. Independence of the Lagrangian density from y implies that
the associated momentum is conserved, leading to the equation

fyi Fi
fy ycl
1/2

= fyi
ycl =
,
(2.11)
fy
g + f y 2
y cl
where ui are the values of u at the turning point for each string, fyi fy (ui ), fy0 fy (u0 ) and
ycl is the classical solution with the two signs corresponding to the lower (string 1) and upper
(string 2). The symbol Fi (not to be confused with the forces in Fig. 1) stands for the function
Fi =
gfy
,
fy fyi
i = 1, 2.
(2.12)
Note that, for the junction point u0 we have that u0 umin . For the turning points we have that
ui u0 , but they are not necessarily larger than umin , since the strings 1 and 2 are not actually
extending to their turning points. Integrating (2.11), we express the separation length as

F1
F2
1/2
1/2
du
+ fy2
du
.
L = L1 + L2 = fy1
(2.13)
fy
fy
u0
u0
ycl
Finally, inserting the solution for

into (2.10) and subtracting the divergent self-energy contribution of disconnected worldsheets, we write the interaction energy of the junction as

u0
1
du g ,
E=
Tp,q E1 + Tp ,q E2 + Tm,n
2
umin

Ei =
u0

du Fi
umin
du g,
i = 1, 2.
(2.14)
272
Ideally, one would like to evaluate the integrals (2.13) and (2.14) exactly, solve (2.13) for u0 and
insert into (2.14) to obtain an expression for the energy E in terms of the separation length L and
u1 , u2 . Then, minimizing this expression with respect to the ui s we will obtain the expression
of the E in terms of the separation length L, which is then identified with the interaction energy
of the heavy dyon pair.
However, in practice this cannot be done exactly, except for the conformal case, and
Eqs. (2.13) and (2.14) are to be regarded as parametric equations for L and E with parameters
u0 , u1 , u2 . A much easier way to find u1 , u2 is to impose that the net force at the string junction
is zero [2,4,10], an approach followed in the present context for the conformal case in [12]. The
infinitesimal lengths squared along each string are

1
d
2i = Gyy ycl 2 + Guu du2 =
Fi du2 ,
Gtt
i = 1, 2.
(2.15)
Hence from the action (2.10) we find that the tensions of the strings at the junction point u =
Tp,q
u0 are 2
Gtt and similarly for the other two string that meet at the junction. The angles
between the string and the u axis at the junction point are computed from

Gyy dy
sin i = fyi , i = 1, 2,
At u = u0 : tan i =
(2.16)
fy0
Guu du
where we used (2.11) and took into account that displacements along perpendicular axes are measured by a curved metric. The above explicitly shows that the angles depend on the parameters
u0 and ui . The forces exerted to each of the strings on the uy plane are
Tp,q
Gtt ( cos 1 , sin 1 ),
2
Tp ,q
Gtt (cos 2 , sin 2 ),
Fp ,q =
2
Tm,n
Fm,n =
(2.17)
Gtt (1, 0),
2
the first entry being the u and the second the y component. Demanding that the total force be
zero one finds that the angles at the equilibrium point are given by
Fp,q =
cos 1 =
2 +T2 T2
Tm,n
p,q
p ,q
2Tm,n Tp,q
cos 2 =
2 +T2
2
Tm,n
p ,q Tp,q
2Tm,n Tp ,q
(2.18)
From these expressions the angles are determined in terms of the NS and RR charges of the
strings and then from (2.16) one may express ui in terms of u0 and the string charges. Recalling
that u0 is determined by L, one ends with the interaction energy being solely a function of L and
of the strings charges as fixed parameters.
We can verify that the approach of fixing the parameters ui by utilizing the zero force condition at the junction point gives rise to a minimal energy. Since we cannot solve (2.13) for u0
in general, we shall recall that we have an implicit expression of u0 in terms of L, u1 , u2 , by inverting (2.13). Since we choose to minimize the energy with respect to ui , the physical length L
does not depend on them and we have
L
= 0,
ui
i = 1, 2,
(2.19)
273
which can be used together with (2.16) to express the derivative of u0 with respect to ui as

fyi
gfy
fy0
u0
1
(2.20)
=
du
, i = 1, 2.
ui
tan 1 + tan 2 2 g0 fyi
(fy fyi )3/2
u0
We have not used the no force condition to derive the above in the sense that the angles i are
not constant except for the equilibrium point. Next, using (2.20) and the zero force condition one
can check, after careful algebraic manipulations, that
E
= 0,
ui
i = 1, 2,
(2.21)
as advertized. Our proof is completely general within our ansatz (2.1) and nowhere we did need
to explicitly evaluate the various integrals by resorting to specific examples.1 Had we done so,
in the cases where this is possible, we would have ended up with complicated expressions and
our proof would have been equivalent to proving certain identities involving, in general, special
functions (for instance, for the conformal case see the appendix of [12]).
Let us mention that the expression for the interaction energy E in terms of length L for
the configuration of (p, q) and (q, p) dyons is invariant under the S-duality transformation
gs 1/gs , as it should be. However, this is not true for L and E separately as functions of
the auxiliary parameter u0 .
2.1. On the quarkmonopole interaction
The most important case arises for the interaction potential of a quark with a monopole with
charges (1, 0) and (0, 1), respectively. In that case it is easily seen from (2.18) that 1 + 2 = /2.
It will be useful for later purposes to consider the limit of gs 0. In that case the tension of the
(0, 1) string corresponding to the monopole becomes very large, so that we expected that it stays
almost straight, whereas in order for the forces to balance, the (1, 0) string should hit the straight
string almost perpendicularly. This is indeed reflected in the expansions

1
1
2 = gs gs3 + O gs5 .
gs + gs3 + O gs5 ,
(2.22)
2
3
3
Of course an equivalent expansion exists for gs , with 1 and 2 interchanged and
gs 1/gs . In the small string coupling limit we easily infer from (2.16) and (2.22) that the
turning points of the strings 1 and 2 are

u2 = uroot + O gs4 ,
u1 = u0 + O gs2 ,
(2.23)
1 =
where uroot is the largest root of the equation fy2 = 0. In the examples we present in detail below,
uroot = umin , except for one case in which it has a smaller value. Therefore the expressions for
the separation length (2.13) and the energy (2.14) for the quarkmonopole interaction, become
1/2
L = fy0

du
u0
F0
+ O(gs )
fy
(2.24)
1 The result is also valid even when in the upper limit of integration infinite is replaced by a finite value, as in the
example of the Rindler space below.
274
and
1
E=
2

du F0
du g + O(gs ),
u0
(2.25)
umin
where we have defined F0 as in (2.12) with fyi replaced by fy0 = fy (u0 ). Thanks to the S-duality
invariance of the general results (2.13) and (2.14) identical expressions to (2.24) and (2.25) hold
in the large string coupling expansion, where in the correction terms we just replace gs 1/gs .
Hence, in the limit of small and large string coupling, the energy and the length assume half of
their corresponding values for the single string for the quarkantiquark system. We emphasize
that this fact does note imply that the spectrum of small fluctuations around the each classical
configuration will be the same. Actually we will explicitly demonstrate that the opposite is true,
which implies that no matter how stiff two of the strings (the straight and one of the strings 1
or 2) become, its fluctuations do not decouple and affect those of the third string.
3. Stability analysis
We now turn to the stability analysis of these configurations, aiming at isolating in parametric
space the physically stable regions and pointing out the similarities and differences with the similar analysis for string configurations dual to the quarkantiquark potential. The small fluctuations
about the classical solutions we will discuss, fall into three types: (i) transverse fluctuations,
referring to cyclic coordinates transverse to the dyondyon axis such as x, (ii) longitudinal
fluctuations, referring to the cyclic coordinate y along the dyondyon axis, and (iii) angular
fluctuations, referring to the special non-cyclic coordinate . In [13] general results for single
string fluctuations of all of the above types for the class of backgrounds with metric (2.1) were
derived and used to analyze the stability of heavy quarkantiquark potentials. These are certainly
relevant for the investigations of the present paper, so that we review them below, but one has to
be careful by paying particular attention to the matching conditions at the junction point of the
three strings which affect the parametric space where the stability occurs, in crucial ways.
3.1. Small fluctuations
We parametrize the fluctuations about the equilibrium configuration, by perturbing the embedding according to
xi = xi (t, u),
yi = ycl,i (u) + yi (t, u),
i = 0 + i (t, u),
(3.1)
where the index i refers to the three strings forming the junction. Note that we have kept the gauge
choice (2.6) unperturbed by using worldsheet reparametrization invariance. We then calculate
the NambuGoto action for this ansatz and we expand it in powers of the fluctuations. The
zeroth-order term gives just the classical action and the first-order vanishes thanks to the classical
equations of motion.2 The resulting expansion for the quadratic fluctuations for the i = 1, 2 string
2 A careful treatment of the first-order terms gives rise to boundary terms. Demanding that they vanish turns out to
be equivalent to the zero force condition at the junction point. This can be shown quite straightforward for junctions in
a flat spacetime before a gauge choice is made and reproduces the conditions found, for instance, in [10]. However, in
curved spaces after the gauge choice u = is made, one obtains straightforwardly only the y-component of the zero
force condition. Switching to the gauge y = one obtains the u-component as well.
275
is written as [13]
(i)

1/2
hfx Fi
gfy
fx
h
2
dt du
x
xi 2 +
yi 2
y 2
1/2 i
3/2
1/2 i
2gf
y
2Fi
2Fi
2Fi

1/2
1/2
hf Fi
fy0 Fi
f
1
2
2
2
2
+

i +
g+
fy i2 ,
1/2 i
1/2
2gfy
4fy2
2Fi
4Fi
S2 =
1
2
(3.2)
where we have used, the condition (2.9) and all functions and their -derivatives are again evaluated at = 0 . Writing down the EulerLagrange equations for this action, using independence
of the various functions from t and setting
xi (t, u) = xi (u)eit ,
i = 1, 2, 3,
(3.3)
we find that for the strings 1 and 2 the linearized equations for the three types of fluctuations read

1/2
fx d
d
2 hfx Fi
+
xi = 0,
du F 1/2 du
gfy
i

gfy d
d
2 h
+
yi = 0,
1/2
du F 3/2 du
F
i
i

1/2
1/2
fy0 Fi
1
d
f d
2 hf Fi
2
2
f
(3.4)
+
i = 0.
y
1/2
du F 1/2 du
gfy
2fy2
2F
i
For the straight string the action for the quadratic fluctuations is

fy
1
fx
f
h
fx h
(3)
dt du y3 2 + x3 2 + 3 2 y32
x 2
S2 =
4
g
g
g
g
gfy 3

2 g 2
f h 2
3 +
3 ,
gfy
4 g
from which we obtain the equations

d fx
hfx
x3 = 0,
x
+ 2
du
g 3
gfy

h
d fy
y3 + 2 y3 = 0,
du
g
g

d f
hf
1 2g
3 = 3 .
3 + 2
du
g
gfy
2 g
(3.5)
(3.6)
Note that for the straight string the equations can be obtained by formally letting Fi g and
fy0 0 in (3.4).
Hence determining the stability of the string configurations of interest has been reduced to a
eigenvalue problem of the general SturmLiouville type

d
d
p(u; u0 )
r(u; u0 ) (u) = 2 q(u; u0 )(u), umin u0 u < , (3.7)
du
du
where the functions p(u; u0 ), q(u; u0 ) and r(u; u0 ) are read off from (3.4) and (3.6) and depend
parametrically on ui through the function Fi in (2.12), which in turn is determined by (2.16)
276
in terms of the junction point u0 and the NS and RR charges of the strings. Our aim is to find
the range of values of u0 for which 2 is negative. In fact the extremal such value of u0 (the
minimum as it turns out) will be decided by determining the zero mode = 0, which is an easier
problem. Note also that, although it may seem so, the equations above for the three strings and
for given type of fluctuations are not decoupled. The matching conditions at the junction point
couple them in an essential manner. This equation cannot be solved analytically in general, so
that sometimes it is convenient to transform the SturmLiouville into a Schrdinger equation and
use known approximating analytical methods that boil down to simple numerical problems.
3.2. Boundary and matching conditions
To fully specify our eigenvalue problem, we must impose boundary conditions on the fluctuations at the UV limit u , at the IR limit u = umin , as well as matching conditions at the
junction point at u = u0 . The boundary condition at the UV are chosen to be
(u) = 0,
as u ,
(3.8)
where is any of the fluctuations for the strings 1 and 2. These represent the fact that we keep
the dyons at fixed points at the boundary. In the far IR at u = umin we simply demand, for the
string 3 that extends also in there, finiteness of the solution and its u-derivative. Otherwise, all
perturbations might be driven away from small values.
We demand that the x fluctuations as well as their variations in the quadratic actions should
be equal at the junction point u = u0 , so that the latter does not break. Moreover, we demand that
the total boundary term resulting in deriving the classical equation of motions from the quadratic
(i)
actions S2 vanishes. These two requirements give rise to
x1 = x2 = x3 ,
Tp,q cos 1 x1
at du = u0 ,
+ Tp ,q cos 2 x2 Tm,n x3 = 0,
at u = u0 ,
(3.9)
where we have used (2.16) to simplify the second condition.

Identical conditions with those in (3.9) hold for the angular fluctuations as well
1 = 2 = 3 ,
Tp,q cos 1 1
at u = u0 ,
+ Tp ,q cos 2 2 Tm,n 3 = 0,
at u = u0 .
(3.10)
The appropriate boundary and matching conditions for the y fluctuations should be found
first in a coordinate system in which the classical solution does not change. This is simply given
by [13]
u = u + u(t, u),
u(t, u) =
y(t, u)
,
ycl (u)
(3.11)
and is easily checked that the classical solution is not perturbed at all. This redefinition does not
affect the x- and -fluctuations since they have trivial classical support and we keep only linear,
in fluctuations, terms. At the junction point the u fluctuations are continuous and this leads to a
discontinuity for the y fluctuations due to (3.11) for the strings 1 and 2. As before, we demand
that the total boundary term resulting in deriving the classical equation of motions from the
(i)
quadratic actions S2 vanishes, keeping in mind that the variations of the yi s in their quadratic
actions should be equal at the junction point. With the use of (2.16) we have the following
277
conditions for the y fluctuations:

y1 cot 1 + y2 cot 2 = 0,
3
Tp,q cos
1 y1
at u = u0 ,
+ Tp ,q cos 2 y2 Tm,n y3 = 0,

3
at u = u0 .
(3.12)
3.3. Zero modes

As we have noted we can obtain the boundary of stability in parametric space by studying the
zero mode problem. Hence, we examine closely the three distinct types of fluctuations.
3.3.1. Transverse zero modes
For the transverse fluctuations the zero mode solution obeying (3.8) for the strings 1 and 2 is

gfy
du
xi = ai Ii (u), Ii (u) =
(3.13)
, i = 1, 2,
fx fy fyi
u
where the ai s are multiplicative constants. The zero mode solution for the straight string is
u0
x3 = a3
du
u
fx
+ const.
(3.14)
When we specialize this expression in our examples we will see that either x3 and/or x3 diverges when u umin , so we choose a3 = 0 in order for the solution and its derivative to remain
finite. Thus x3 = const. and the zero mode solution becomes irrelevant for the stability analysis.
From (3.9) and the zero force condition, one finds a linear homogeneous system of algebraic
equations for a1 and a2 , whose determinant should vanish for non-zero solutions to exist. In this
way we find the following condition
sin 1 I1 (u0 ) + sin 2 I2 (u0 ) = 0.
(3.15)
Since Ii (u) > 0 this condition cannot be satisfied and therefore the transverse zero modes do
not exist, rendering the corresponding fluctuating modes as stable. Our finding is similar to the
conclusion that the transverse fluctuations of the single string fluctuation corresponding to the
potential of heavy quarkantiquark pairs are also stable [13].
3.3.2. Longitudinal zero modes
We next turn to the longitudinal fluctuations. The zero mode solution obeying (3.8) for the
strings 1 and 2 is

yi = bi Ji (u),
Ji (u) =
u

gfy
du
,
(fy fyi )3/2
i = 1, 2,
(3.16)
where the bi s are multiplicative constants. The zero mode solution for the straight string is
u0
y3 = b3
du
u
fy
+ const.
(3.17)
278
(a)
(b)
Fig. 2. Plots of L(u0 ) and E(L) for cases where there is an instability of the longitudinal modes. This covers the case of
non-extremal D3-brane, multicenter D3-branes on a sphere for the 0 = 0 trajectory and the Rindler space. The various
types of lines correspond to stable (solid dark), metastable (solid gray) and unstable (dashed gray) configurations, with
the stability determined by the analysis of Section 5.
As before, when we specialize this expression in our examples we will see that either y3 and/or
y3 diverges when u umin , so we choose b3 = 0 in order for the solution and its derivative to
remain finite. Therefore the zero mode fluctuation becomes irrelevant for the stability analysis.
Similarly to before, from (3.12) and the zero force condition one finds the following relation
cos 1 J1 (u0 ) = cos 2 J2 (u0 ).
(3.18)
This is an equation for u0 in terms of the string coupling and the strings charges, which can be
solved numerically. We will denote its solution, whenever it exists, by u0c . In general we will
have expansions of the form
(0)
u0c (gs ) = u0c +

n=1
n
u(0)
n gs ,
()
u0c (gs ) = u0c +
n
u()
n gs .
(3.19)
n=1
In the most important case of a quarkmonopole interaction, S-duality invariance of the result
dictates that u0c (gs ) = u0c (gs1 ) and therefore the two perturbative expansions above contain the
(0)
()
(0)
()
same information. Namely, u0c = u0c and all coefficients are equal, i.e., un = un . Actually,
(0)
then the critical value u0c does not change significantly from its leading order values u0c , the
reason being that the small and large gs results actually coincide and therefore there is not much
(0)
room for big changes in between. We show below that the leading term u0c can be computed
by (3.24) which is an approximative version of (3.18). In the examples below the leading order
result for u0c from (3.24) and that computed for gs 1 using (3.18) differ by two to ten percent.
A natural question is whether or not the value of u0c coincides with that corresponding to
the maximum value of L(u0 ), labelled by u0m if that exists, below which the energetically less
favorable branch in the EL diagram starts (a typical case is presented in Fig. 2). This might
have been expected since in [13,14] it was shown quite generally that an instability of longitudinal fluctuations of a single string used to compute the heavy quarkantiquark potential, occurs
precisely at the solutions of the equation L (u0 ) = 0 (which is identical to E (u0 ) = 0). In our
case following similar manipulations as in [13] and using (2.16) at the equilibrium point where
279
the i s are constant, in order to find the derivatives of ui with respect to u0 , we obtain

fy0
(sin 1 K1 + sin 2 K2 ),
L (u0 ) =
fy0
1
E (u0 ) =
f sin 1 Tp,q (sin 1 K1 + sin 2 K2 ),
2 y0
(3.20)
which are indeed proportional to each other and where we have defined3

gfy
1
Ki = du u
, i = 1, 2.
fy
(fy fyi )1/2
(3.21)
u0
The value of u0c for which (3.18) is satisfied is different from u0 = u0m for which L (u0 ) =
E (u0 ) = 0. In fact, in the examples we will explicitly work out, it occurs at smaller values,
which implies that part of the upper branch of the solution is still perturbatively stable although
energetically not favorable. However, the disconnected configuration becomes energetically favorable for all positive values of the energy. Thus part of the upper branch which is perturbatively
stable is in fact metastable. The fact that u0c = u0m can be demonstrated quite generally in the
small coupling constant limit for the most significant case, namely for the quarkmonopole system. We may show that

fy0
gfy
1

du u
+ O(gs ),
L (u0 ) =
(3.22)

fy
(fy fy0 )1/2
fy0
u0
with an analogous expression for E (u0 ). From this we compute the leading term u0m in an
expansion of the values u0m for which the maximum energy and length occur analogous to (3.19).
Using (2.22), (2.23) and the general identity
g0
Ji (u0 ) = 2Ki (u0 ) + 2
(3.23)
,
fy0 cos i
(0)
proved by partial integration, we may show that the zero mode condition (3.18) becomes
g0

fy0
du
u0
g
= O(gs ),
fy
(3.24)
(0)
from which one computes the leading order value of u0c in the expansion (3.19), namely u0c .
(0)
(0)
Coming from different conditions we have that generically u0m = u0c and this explicitly demonstrates that the value for u0m (gs ) found by requiring maximum length and energy is different than
the value of u0c (gs ) found by requiring the existence of a zero mode for the longitudinal fluctuations which signals perturbative instability of the system.
We note that the condition (3.24) can be directly obtained from the boundary conditions (3.12)
specialized to the zero mode solution for the longitudinal fluctuations. In the derivation one
3 In cases, such as in Rindler space, where the upper limit of integration is finite, i.e., u
max = , then there appears an
additional term in the expressions below given by Fi /fy computed at u = and also we replace infinity by in the
upper limit of integration.
280
should be careful and absorb a factor of 1/gs into the overall amplitude of the string 2. In this
way it becomes apparent that although in the small coupling limit the string 2 becomes stiff and
approximately straight, its ultra small in strength fluctuations couple to those of the string 1.
Similar considerations can be given for the higher modes as well. The above comments explain
also why although in the limit gs 0 (or gs ) the turning point u1 u0 (or u2 u0 ) and
therefore the corresponding equation for the fluctuations (3.4) becomes the same as the equation
that appears in the case of the single string for the quarkantiquark system [13], the conditions
for the existence of the zero mode are different in the two cases. They are so due to the different
boundary conditions.
Finally, we comment on whether or not the heavy dyondyon potentials as computed within
supergravity using the AdS/CFT correspondence violate some general principles regarding the
strength and sign of the corresponding force. For instance, for the quarkantiquark potential it
turns out that the force should always be attractive with strength an ever non-increasing function
of the separation distance [17]. This is a concavity condition for the quarkantiquark potential
and implies that the upper branch in the EL plot should be disregarded, in full agreement with
the results of [13,14] employing perturbative stability. In our case, using (3.20), we can show that
dE Tp,q sin 1
fy0 ,
=
dL
2

fy0
d 2 E Tp,q sin 1

,
=
4
dL2
fy0 L (u0 )
(3.25)
which can be written in a symmetric way under an exchange of the strings 1 and 2 using the
y-component of the non-force condition. The expressions (3.25) for the dyondyon potential are
analogous to those found in [16] for the quarkantiquark potential. Hence, the force is always attractive, but for the upper branch in the EL plot in which L (u0 ) > 0, has an increasing strength.
On the other hand we have found that part of the upper branch is perturbatively stable, so that
requiring concavity seems to be a stricter condition than perturbative stability. However, concavity is not a condition that gauge theory analysis of Wilson loops for heavy quarkmonopole
pairs requires. One can derive using reflection-positivity, that the energy of this pair is larger or
equal than the average of the energies of the quarkantiquark and monopoleantimonopole systems.4 This condition is rather trivially satisfied in our case, provided we use the physical lower
branches of the two pairs.
3.3.3. Angular zero modes
For the angular zero modes we cannot explicitly write down the solution for the zero mode
due to the presence of the restoring force term in the corresponding SturmLiouville equation in
the third lines of (3.4) and (3.6). Therefore, in this case we have to work out explicitly the result
in the various examples we will present. As it was shown in previous work [13,14] one can use
approximate analytic methods in which the Schrdinger description of the differential equation
(3.7) plays an important role.
4. Examples of classical string junction solutions
In this section, we study first the behavior of the dyondyon potentials emerging in Wilsonloop calculations for non-extremal and multicenter D3-brane backgrounds. Since we will work
with the NambuGoto action, we need mention in the expressions below only the metric and not
4 We thank C. Bachas for providing this information.
281
the self-dual five-form which is the only other non-trivial field present in our backgrounds. At
the end we work out the example of the Rindler space which captures the behavior of black holes
in various dimensions near the horizon.
4.1. Non-extremal D3-branes
We start by considering a background describing a stack of N non-extremal D3-branes [18].
In the field-theory limit the metric reads

u2
4
u2
2
2
ds 2 = 2 1 4 dt 2 + d x 32 + R 2 4
(4.1)
du
+
d
5 ,
R
u
u 4
where the horizon is located at u = and the corresponding Hawking temperature is T = R

2.
5
This metric is just the direct product of AdS5 -Schwarzschild with S and it is dual to N = 4 SYM
at finite temperature. For the calculations that follow, it is convenient to switch to dimensionless
variables by rescaling all quantities using the parameter . Setting u u and u0 u0 and
introducing dimensionless length and energy parameters by
1
L,
E,
2
(4.2)
we see that all dependence on drops out so that we may set 1 in what follows. The
functions in (2.2) depend only on u (reflecting the fact that all values of are equivalent) and are
given by

fx (u) = u4 1 /R 4 ,
g(u) = 1,
fy (u) = u4 1 /R 4 ,
f (u) =
u4 1
,
u2
h(u) =
u4
.
1
(4.3)
u4
In this case umin = uroot = 1, coinciding with the location of the horizon. For the dyondyon
potential we find that the separation length is given in terms of the junction point u0 as
L=R

Li = R 2

u41
1
u0
u4i 1
3u30

F1
du
(u4 1)(u4 u41 )

+ (1 2) = L1 + L2 ,

3 1 1 7 1 u4i
, , , , 4, 4 ,
4 2 2 4 u0 u0
i = 1, 2,
(4.4)
i = 1, 2.
(4.5)
whereas the energy is given by

E = Tp,q E1 + Tp ,q E2 + Tm,n (u0 1),

4
u 1
Ei = du
1 (u0 1)
u4 u4i
u0

1 1 1 3 1 u4
= u0 F1 , , , , 4 , i4 + 1,
4 2 2 4 u0 u0
282
In the above F1 (a, b1 , b2 , c, z1 , z2 ) is the Appell hypergeometric function and u0 ui 1. From

(2.16) one finds that the turning points of the two strings are given by

1/4
ui = u40 u40 1 cos2 i
(4.6)
, i = 1, 2,
where the angles i are given by (2.18), in terms of the strings charges. From (3.20) we have in
this case
L (u0 ) = R 2
4u30
(sin 1 K1 + sin 2 K2 ),
u40 1

3
1
7 1 1 11 1 u4i
3 1 1 7 1 u4i
Ki =
F1 , , , , 4 , 4
F1 , , , , 4 , 4 .
4 2 2 4 u0 u0
4 2 2 4 u0 u0
28u70
12u30
(4.7)
The function L(u0 ) has a single global maximum for any value of the string coupling. Its location
u0m depends on the string coupling and we obtain the following expansion for the quark
monopole system

u0m 1.177 0.037gs + O gs2 .

(4.8)
For L > Lmax , only the disconnected solution exists. For L < Lmax , Eq. (4.4) has two solutions
for u0 , corresponding to a short and a long string and, accordingly, E is a double-valued function
of L. The behavior described above is shown in the plots of Fig. 2 and we see that it is very similar
to the behavior of the L(u0 ) and E = E(L) for the heavy quarkantiquark system [19]. However,
as we have pointed out and we will see later in detail the upper branch, although energetically
less favorable than the lower one, it is not unstable in its entirety. We note that the classical string
junction in this case has also been examined in [20] where the behavior of Fig. 2 was also noted.
4.2. Multicenter D3-branes on a sphere
We now proceed to the case of multicenter D3-brane distributions. These distributions [21,22]
constricted as extremal limit of rotating D3-brane solutions [23,24] have been used in several
studies for the Coulomb branch of N = 4 SYM within the AdS/CFT correspondence, starting
with the works of [16,25]. Here, we will concentrate on the particularly interesting cases of
uniform distributions of D3-branes on a three-sphere.
The field-theory limit of the metric for N D3-branes uniformly distributed over a 3-sphere
with radii r0 reads

u2 r02 cos2
ds 2 = H 1/2 dt 2 + d x 32 + H 1/2
du2
u2 r02

+ H 1/2 u2 r02 cos2 d 2 + u2 cos2 d32 + u2 r02 sin2 d12 ,
(4.9)
where
H=
R4
.
u2 (u2 r02 cos2 )
(4.10)
It is convenient to switch to dimensionless variables by rescaling all quantities using the parameter r0 . Setting u r0 u and u0 r0 u0 and introducing dimensionless length and energy
283
parameters by
r0
1
L,
E
E,
(4.11)
r0
2
we see that all dependence on r0 drops out so that we may set r0 1 in what follows, we write
the functions in (2.2) as
L
u2 cos2
,
fy (u, ) = u2 u2 cos2 /R 4 ,
2
u 1

fx (u, ) = u2 u2 cos2 /R 4 ,
g(u, ) =
u2 cos2
.
(4.12)
u2 1
Therefore the conditions (2.9) are satisfied only for 0 = 0 and 0 = /2. We examine these two
cases in turn.
f (u, ) = u2 cos2 ,
h(u, ) =
4.2.1. The trajectory corresponding to 0 = 0

In this case umin = uroot = 1 and the integrals for the dimensionless length and energy read

du
2
2

+ (1 2)
L = R u1 u1 1
2 1)(u2 u2 )(u2 + u2 1)
u
(u
u0
1
1

u1 k
= R 2 2 1 1 , k1 2 , k1 F(1 , k1 ) + (1 2)
(4.13)
u1 1
and
E = Tp,q E1 + Tp ,q E2 + Tm,n (u0 1),

u2 1
Ei = du u
1
(u0 1),
(u2 u21 )(u2 + u21 1)
u0

= 2u2i 1 ki 2 K(ki ) F(i , ki ) E(ki ) E(i , ki )

(u20 u2i )(u20 + u2i 1)
+
+ 1, i = 1, 2,
u20 1
(4.14)
where F(, k), E(, k) and (, , k) are the incomplete elliptic integrals of the first, second and
third kind respectively, while K(k), E(k) and (, k) are the corresponding complete ones and

ui
ki =
,
ki = 1 ki2 ,
2u2i 1

2
2
2u2i 1
1 u0 ui
1
i = sin
(4.15)
=
sin
,
.
i
u20 1
u20 + u2i 1
From (2.16) one finds that

1 + i
, i = 1 + 4u20 u20 1 sin2 i ,
ui =
2
i = 1, 2,
(4.16)
284
where the angles i are given by (2.18). The function L(u0 ) has a single global maximum depending on the string coupling gs . The situation is similar to the quarkantiquark potential in this
background [16] and to the case of string junctions on black D3-branes in the present paper. The
general behavior is shown in the plots of Fig. 2.
4.2.2. The trajectory corresponding to 0 = /2
In this case umin = 1, but uroot = 0 and the integrals for the dimensionless length and energy
read

du

+ (1 2)
L = R 2 u21
4
2
4
u0 u (u 1)(u u1 )

R 2 (1 , 12 , k1 ) F(1 , k1 )
=
(4.17)
+ (1 2) ,
u1
2
and

E = Tp,q E1 + Tp ,q E2 + Tm,n u20 1,

u0
u2
du u
du u

1
Ei =
2
u 1
u2 1
u4 u4i
u0
1

(u20 u2i )(u20 + u2i )

ui
,
= 2ui E(i , ki ) E(ki ) + F(i , ki )
u20 1
2
where now

ki =
u2i + 1
i = sin
2u2i
1
k =
u20 u2i
u20 1

,

ui = u0 sin i , i = 1, 2,
(4.18)
1 k2,
i = sin

.
u20 + u2i
2u2i
(4.19)
(4.20)
where i angles are given by (2.18). In this case L(u0 ) is a monotonously decreasing function
which approaches infinity as u0 1 and zero as u0 and hence no maximal length exists,
Eq. (4.17) has a single solution for u0 given the length L and consequently E is a single-valued
function of L. This behavior is shown in Fig. 3. This confining behavior is similar to what was
found for the quarkantiquark system in [16].
4.3. Rindler space
Our next example is based on the Rindler space, which is a portion of flat Minkowski space.
In this space an observer experiences the Unruh effect, perceiving the inertial vacuum as a state
populated by a thermal distribution of particles at a temperature T = /2 , where is the surface
gravity. Black hole solutions near the horizon behave like the Rindler space, so that the study
(a)
285
(b)
Fig. 3. Plots of L(u0 ) and E(L) for cases where there might be angular instabilities. Note the appearance of a confining potential for cases where there are angular instabilities. This covers, among our examples, the case of multicenter
D3-branes on a sphere for the 0 = /2 trajectory. The meaning of the colors is as in Fig. 2.
of string in this background serves more general purposes. In [14,26] the string configuration
corresponding to an open string with its two endpoints located at the same radius u = (instead
of u ) we studied. Moreover, we have proven that this problem is exactly equivalent to the
classical-mechanical problem of the shape of a soap film suspended between two circular rings
and we have also performed a small fluctuations analysis around this classical configuration.
The metric for Rindler space has the form
ds 2 = 2 u2 dt 2 + dy 2 + du2 + ,
(4.21)
where u is the radial direction with the Rindler horizon corresponding to u = 0 and y is a generic
spatial direction. It is natural to extend the computation of [14] to the case of sting junctions with
the end points of strings 1 and 2 at u = and the third straight string stretching to u = 0. For
convenience we pass to dimensionless units
2
(4.22)
E
4 2
and then we use the formalism of Section 2 which readily applies to this setup. The length and
the energy of the string are determined by Eqs. (2.13) and (2.14) (without the subtraction term
for the energy and the upper limit in the integration changed to 1) which for our case read
u u,
u0 u0 ,
1
du
L=
u0
L L,
u1
u2 u21
+ (1 2) = u1 cosh
1
u0
cosh1
u1
u1

+ (1 2),
(4.23)
and
E = Tp,q E1 + Tp ,q E2 + Tm,n u20 ,
1

Ei =
u0
du u2
u2 u2i

1 u2i + u2i cosh1

1
u0
u0 u20 u2i u2i cosh1
.
ui
ui
(4.24)
286

ui = u0 sin i ,
i = 1, 2,
(4.25)
where the angles i are given by (2.18). The function L(u0 ) has a single global maximum. Its
location depends on the string coupling gs . For small values of the latter we get the expansion

u0m 0.552 0.450gs + 0.479gs2 0.100gs3 + O gs4 .

(4.26)
The situation is similar to that depicted in the plots of Fig. 2.
5. Examples of stability analysis
In this section we apply the stability analysis developed in Section 3 to the string solutions
reviewed in Section 4. We find that for the non-extremal D3-brane, the sphere with 0 = 0 and
the Rindler space we have a longitudinal instability corresponding to the upper branch of the
energy curve, for the sphere with 0 = /2 we have angular instabilities towards the IR, even
though the potential has a single branch.
5.1. The conformal case
For completeness we consider first the conformal case, corresponding to the 0 or r0 0
limit of any of the above solutions. The classical string junction solution has been constructed
in [12]. Since there is no -dependence, the fluctuations are equivalent to the transverse xfluctuations and therefore stable. On the other hand for the longitudinal fluctuations we have to
solve (3.18). From (2.16) one finds that

ui = u0 sin i , i = 1, 2,
(5.1)
where the angles i are given by (2.18). Moreover we compute

1
3 3 7 u4
Ji (u0 ) = 3 2 F1 , , , i4 .
4 2 4 u0
3u0
(5.2)
Using the above, we may check that (3.18) has no solution for any value of string coupling gs ,
thus we verify that indeed string junctions are stable in the conformal case.
5.2. Non-extremal D3-branes
We proceed next to the case of non-extremal D3-branes, where we recall that the potential energy is a double-valued function of the separation length. As before, since there is no
-dependence, the fluctuations are equivalent to the x-fluctuations and stable. For the longitudinal fluctuations we have to solve (3.18), where

1
3 1 3 7 1 u4
Ji (u0 ) = 3 F1 , , , , 4 , i4 , i = 1, 2.
(5.3)
4 2 2 4 u0 u0
3u0
Eq. (3.18) has a solution for any value of string coupling gs . This value u0c is smaller than the
value u0m in which the maximum of the length occurs. In the limit of small coupling constant
for the quarkmonopole system we have shown that we can approximate (3.18) by the condition
(3.24) which in our case becomes

7 1
3
3
F
,
1,
,
= O(gs ).
2 1
4
4 u40
2
287
(5.4)
Its solution gives the leading-order result for the critical value in which instability occurs

u0c 1.117 0.166gs + O gs2 ,
(5.5)
where we have included the first correction as well. This expansion is different than the expansion
of u0m in (4.8). Thus the whole upper branch is not entirely unstable, a behavior not expected
from energetic considerations. Let us compare the maximum length of the screened interaction
for the cases of the quarkantiquark and the quarkmonopole interactions. As gs varies from
qm
small to large values the maximum length is Lmax (0.425 0.005)R 2 , whereas for the quark
qq
qm
antiquark pair Lmax 0.869R 2 2Lmax . We see that the quarkmonopole pair is screened twice
as strong as the quarkantiquark pair. This is a general feature in all of the examples we encounter
in this paper.
5.3. Multicenter D3-branes on a sphere
The results of the stability analysis for the two orientations for the sphere distribution are
presented below.
5.3.1. The trajectory corresponding to 0 = 0
Angular fluctuations are stable due to the fact that, as in [13], we may transform, for each
string separately, the problem into a Schrdinger equation with a positive potential for all values
of ui . For the longitudinal fluctuations we have to solve (3.18), where

Ji (u0 ) =
du
u0
u u2 1
,
(u2 (u2 1) u2i (u2i 1))3/2
i = 1, 2.
(5.6)
This has a solution for all values of the string coupling. In the limit of small string coupling for
the quarkmonopole system the condition (3.24) reads explicitly
2u20
2u20 1

u0 + 1
u0
ln
= O(gs ),
2
u0 1
(5.7)
giving the leading-order result for the critical value in which instability occurs
u0c 1.084 + O(gs ).
(5.8)
Thus, part of the upper branch of the solution is perturbatively stable as in the case of the
string junction based on the non-extremal D3-brane solution. As before, we compare the maximum length of the screened interaction for the cases of the quarkantiquark and the quark
monopole interactions. As gs varies from small to large values the maximum length in the
qm
quarkmonopole interaction is Lmax (0.425 0.050)R 2 , whereas for the quarkantiquark pair
qq
qm
Lmax 1.002R 2 2.3Lmax .
288
5.3.2. The trajectory corresponding to 0 = /2

Longitudinal fluctuations are stable since an explicit check of (3.18) shows that it has no solution for any value of string coupling gs . Alternatively, we may transform the problem into a
Schrdinger problem as in [13] with an everywhere positive potential. For the angular fluctuations the potential is just constant
V = 1,
i = 1, 2.
(5.9)
The change of variables for the two string is

1
F(i , ki ), i = 1, 2,
ui 2
where i and ki are given by

2u20 sin i
u20 sin i + 1
1
i = sin
,
,
ki =
2
u2 + u0 sin i
2u20 sin i
zi (u) =
(5.10)
i = 1, 2.
The zero mode of the Schrdinger equation for the i = 1, 2 strings reads

i (zi ) = ci sin zi , zi [0, i ] as u (, u0 ] , i = zi (u0 ),
(5.11)
(5.12)
where ci are multiplicative constants. The straight string 3 plays a rle in the boundary conditions, the reason being that the mass term is present in the equation for the angular fluctuations
even for the zero mode. For the straight string we have two linear independent solutions sin z and
cos z, where the appropriate change of variable is
1
sin1 .
(5.13)
2
u
We examine in detail the case where only sin z is used and comment on the general linear combination later. From the matching conditions in (3.10) one finds the following condition
z=
sin 1 sin 1 cos 2 + sin 2 sin 2 cos 1
sin(1 + 2 )
sin 1 sin 2 = 0.
u0 (u20 1)
(5.14)
For the quarkmonopole case we may use that 1 + 2 = /2, to simplify it a bit. Furthermore,
in the small string coupling limit (5.14), becomes

1
1
+
tan
(5.15)
u0 u20 1 = O(gs ),
2u0 2 u2 1
0
which is a more explicit and easy to handle transcendental equation. Hence, the solution for the
leading order critical values for which instability of the angular perturbations occur is
u0c 1.378 + O(gs ).
(5.16)
The confining behavior turns out to be unstable since (5.14) has a solution for any value of
the string coupling gs . As gs varies from small to large values, the maximum length of the
qm
quarkmonopole interaction is Lmax (0.560 0.040)R 2 , whereas for the quarkantiquark pair
qq
qm
Lmax 1.700R 2 3Lmax .
Let us mention that, had we used the general solution of the Scrdinger equation, that is
cos(z + ), we would have obtained similar results for any phase with the exception of = 0 for
which case no instability occurs. Hence the above considerations and results are quite generic.
289
Finally let us note that, in order keep the discussion at a reasonable length, we refrained
from explicitly presenting the details for the other important uniform distribution of D3-branes,
namely, that on a disc. We mention the end result: For 0 = /2 the solution is stable against
all types of small fluctuations, while for 0 = 0 it is stable against longitudinal fluctuations and
unstable against angular ones after a certain separation length. This behavior reveals the expected
screening behavior which moreover is practically independent of the orientation of the string and
similar to that for the case of the quarkantiquark potential studied in full detail in [13].
5.4. Rindler space
Finally, we consider the case of the Rindler space. Fluctuations of the single string solution
corresponding to the heavy quarkantiquark potential have been considered in [14,26]. Turning
to our case, we see from the form of the metric only the longitudinal fluctuations are relevant for
our discussion, so that we have to solve (3.18), where

1
1
u0
1 1
1 u0
Ji (u0 ) =
(5.17)
cosh

+
cosh
.
ui
ui
u20 u2i
1 u2i
Eq. (3.18) has a solution for all values of the string coupling gs and we have the same behavior
as in y fluctuations in the non-extremal D3-branes. For small values of the string coupling latter
we get the expansion

u0c 0.368 0.214gs + 0.218gs2 0.058gs3 + O gs4 .

(5.18)
We see that it is different from the expansion of u0m in (4.26) corresponding to the maximum of
the length and the energy. Hence, the behavior is similar to that represented in Fig. 3.
6. Discussion
In this paper we have constructed string junctions on general backgrounds and used them to
compute the interaction energy of heavy dyons at strong coupling within the AdS/CFT correspondence. The most important case is that of the quarkmonopole pair for which we paid particular
attention. Then we turned into the perturbative stability analysis of these solutions which is important to distinguish the physically relevant regions in which the potentials can be trusted. We
formulated and further investigated general conditions for the existence of instabilities. Starting
with the conformal case we indeed verified that there are no instabilities at the conformal point
of N = 4 SYM, as expected. Nevertheless, we emphasize here that in the -deformed conformal
solution of [27], as well as for all backgrounds that asymptote this, there is an instability for
certain angular fluctuations. The reason is identical to the case of a single string for the quark
antiquark potential that was found in [14] (see Section 4.3.1) and occurs for values of the real
deformation parameter (in the standard notation) larger than a certain critical value (we remark
that the classical string junction of the -deformed background of the disc D3-brane distribution
has been considered in [28] and should suffer from this instability).
We found that part of the branches of the potential energy versus length that could be disregarded on energetic considerations can be perturbatively stable. The reason is traced to the fact
that the matching conditions at the junction point allow for interaction of all strings in the junction
that enhances its stability. This is unlike the findings in [13,14] for the quarkantiquark potential according to which the perturbative instability renders energetically unfavorable branches
290
as unstable. We pointed out that this result is not in conflict with general considerations on the
quarkmonopole potential. For instance, there is no concavity condition as is in the case of the
quarkantiquark potential. We found that the confining behavior of the dyon-potential in the
Coulomb branch of the N = 4 SYM with vacuum expectation values distributed uniformly on
a sphere, is unstable. This is similar to the result of [13] for the seemingly confining behavior
of the quarkantiquark potential for the same distribution. It is quite remarkable that demanding
perturbative stability of the string configurations renders as physically unacceptable the confining
behavior, which indeed is not expected from gauge theory considerations.
Our general formalism was based on a diagonal class of metrics (2.1). It can be extended to
supergravity backgrounds in which off-diagonal metric elements appear and in which additional
background form fields couple in the WessZumino part of the string actions in the junction, in
analogy with [14] for the quarkantiquark case. In particular, this will cover, within the AdS/CFT
correspondence, interactions of moving dyons in hot quarkgluon plasmas. We expect that also
in this case the associated energy will have two branches with the upper one being only partly
unstable. The behavior found in [29], namely that the maximum length behaves for high enough
velocity as Lmax 0.743(1 v 2 )1/4 R 2 , is expected to persist with the overall numerical coefficient replaced by, roughly, its half.
It will be interested to extend our analysis to junctions involving more than three strings where
we believe similar phenomena with the triple-junctions considered here will be found. A more interesting generalization is to consider systems of multiple external particles at finite temperature
and examine to what extend they prefer, under small perturbations to split, with raising temperature, into smaller systems. This possibility was argued for, based on energy considerations, in
[30] who examined in detail the interaction energy for the system of a quark, a monopole and a
dyon.
Finally, it has been proposed that F - and D-superstrings could have been produced at the early
Universe, then expanded at a cosmic size and that their network properties by forming junctions
make them distinguishable [31] from gauge theory cosmic strings (see, for instance, [32]). The
no-force condition for a stationary junction to form gives rise to a locally flat space time [33],
much-like the case of single cosmic strings [34]. Small fluctuations of cosmic string junctions on
flat spacetime have been analyzed [35] in closed analogy with corresponding work on superstring
junctions (see, for instance, [32]). In view of the above remarks it will be very interested to
study the formation and stability of cosmic size string junctions in cosmological backgrounds,
by adapting the results and techniques of the present paper, to such cases.
Acknowledgements
We would like to thank S. Avramis for useful discussions. We acknowledge support provided through the European Communitys program Constituents, Fundamental Forces and
Symmetries of the Universe with contract MRTN-CT-2004-005104, the INTAS contract 03-516346 Strings, branes and higher-spin gauge fields, the Greek Ministry of Education programs
PYAOPA with contract 89194. In addition, K. Siampos acknowledges support provided by
the Greek State Scholarship Foundation (IKY).
Appendix A. An analog from classical mechanics
Longitudinal string fluctuations exhibit a behavior which is not expected from the energetics
of the configurations. However, this behavior can be found in the classical mechanics problem
291
of determining the shape of a thin soap film stretched between two rings appropriately, modified
to serve our purposes. We briefly review first the solution of the unmodified classical problem
which is a catenary and can be found in standard textbooks [36]. In the notation of the present
paper and of the appendix of [13] the cylindrically symmetric solution reads
r(z) = u0 cosh
z
,
u0
L
L
z ,
2
2
(A.1)
in terms of the integration constant parameter u0 . The potential energy and length of the soap
film are given by the following expressions
1
L = 2u0 cosh1 ,
u0

1
E = 2
1 u20 + u20 cosh1
,
u0
0 u0 1.
(A.2)
Hence, E(L) is a double-valued function of L with the energetically favorable branch having
lower energy for a given distance, corresponding to a shallow catenary, whereas the upper branch
corresponds to a deep catenary. There is also the Goldschmidt solution, with two disconnected
soaps in the two rings with energy E = 2 . Thus we have the same qualitative behavior as in the
classical solution of the non-extremal D3-branes, depicted in Fig. 2. The value of u0 at the turning
point is u0m 0.552 leading to the maximal length of Lm 1.325. The Goldschmidt solution
becomes energetically favorable at a higher value of u0 0.826, corresponding to a smaller value
of L 1.055. The perturbative small fluctuation analysis has been performed in [37] and in [13].
The small fluctuations stability analysis of this solution gives the expected result that the upper,
energetically unfavorable, branch is perturbatively unstable.
We would like to add an extra term to the quadratic fluctuations (see Eq. (A.1) in Appendix A
of [13]) that, to leading order, would not affect the energy of the solutions and then check whether
part of the upper branch becomes perturbatively stable due to this modification of the action. In
order to fulfill the above requirement imagine that we first let the soap film assume its shape
given by (A.1) and then we turn on the perturbation which we take it to be quadratic in the normal
fluctuations of the soap film surface. The normal fluctuations decouple from the two tangential
perturbations which moreover are trivial [13]. This will not affect the classical solution and its
energy, but will certainly modify the stability analysis. There are several such terms that we may
add. Here we choose to present the two most natural ones.
In the first we add a term to the potential that acts only at the minimum of the catenary at
z=0
u0
(A.3)
()2 (z), k 0,
2
where the factor u0 is introduced for convenience. For the normal perturbations, we define
u = z/u0 , we separate variables according to
V=
(t, u, ) = (u)eit eim ,
m = 0, 1, 2, . . . ,
and eventually we end up with the SturmLiouville equation

2
d 2
2
+ m + (u) = 2 cosh2 u,
2 +
du
cosh2 u
(A.4)
= u0 ,
(A.5)
292
subject to the following boundary conditions

L

= 0,
2u0
(A.6)
representing fixed endpoints. Eq. (A.5) cannot be easily solved for non-vanishing values of ,
but we can obtain useful information for our stability problem by considering the zero-mode
problem having = 0. The transformation x = tanh u turns (A.5) to an associated Legendre
equation with the general solution given by a linear combination of P1m (tanh u) (they vanish
identically unless m = 0, 1) and Qm
1 (tanh u), with different coefficients to the regions left and
right of z = 0. However, we know from the work in [13] that in the absence of the extra term
(for = 0), only for m = 0 a zero mode exists for the value u0 0.552 we mentioned above.
Since the modification is by an repulsive -function term the possibility of having zero mode
solutions in the presence of it for |m| 1, is definitely excluded. Hence we concentrate to the
case with m = 0. The continuity of the solution at z = 0, the discontinuity of its derivative, as
read off from (A.5), and the boundary conditions (A.6), give a linear homogeneous system for the
four independent coefficients which in order to have non-trivial solution should have a vanishing
determinant resulting to the condition

1+a
P1 (a) = 2Q1 (a) a + 2 = a ln
(A.7)
, a = 1 u20 .
1a
This equation has a solution for all 0, which varies monotonously from u0c 0.552 to 0, as
[0, ). Importantly, the product u0c and therefore the term (A.3) we have added, remain
finite. For small and large values of we have the behaviors

u0c 0.552 0.160 + O 2

(A.8)
and

u0c = 2e/21 + O e .
(A.9)
An alternative to (A.3) potential term is

ku0
(A.10)
()2 , k 0,
2
where the factor u0 is, as before, introduced for convenience. In the SturmLiouville equation (A.5) we replace (u) u0 k. The same transformation x = tanh u gives an associated
Legendre equation with the general solution given by a linear combination of P1m (tanh u) and

= m2 + u0 k. For identical reasons as before we concentrate to the case

Qm
1 (tanh u), where m
with m = 0 and in fact refining the argument we should have that m
< 1 in order to find a normalizable zero mode. Imposing the boundary condition (A.6) and demanding that the determinant
of the resulting linear homogeneous system for two independent coefficients has non-trivial solutions, we obtain the condition

u k
u k
u k
u k
P1 0 (a)Q1 0 (a) = P1 0 (a)Q1 0 (a), a = 1 u20 .
(A.11)
V=
This can only be solved numerically and as before the product u0c k remains finite. For small and
large values of k we have the behaviors

u0c 0.552 0.102k 4 + O k 8

(A.12)
293
and

1
2
(A.13)
3 + O 1/k 5 .
k k
In both of the above examples we see that the critical value u0c can get arbitrarily small,
which implies that the entire upper, energetically less favorable branch corresponding to the
deep catenary, can become perturbatively stable.
u0c =
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]
O. Aharony, J. Sonnenschein, S. Yankielowicz, Nucl. Phys. B 474 (1996) 309, hep-th/9603009.

J.H. Schwarz, Nucl. Phys. B (Proc. Suppl.) 55 (1997) 1, hep-th/9607201.
K. Dasgupta, S. Mukhi, Phys. Lett. B 423 (1998) 261, hep-th/9711094.
A. Sen, JHEP 9803 (1998) 005, hep-th/9711130.
S.J. Rey, J.T. Yee, Nucl. Phys. B 526 (1998) 229, hep-th/9711202.
M. Krogh, S. Lee, Nucl. Phys. B 516 (1998) 241, hep-th/9712050.
Y. Matsuo, K. Okuyama, Phys. Lett. B 426 (1998) 294, hep-th/9712070.
O. Bergman, Nucl. Phys. B 525 (1998) 104, hep-th/9712211.
M.R. Gaberdiel, B. Zwiebach, Nucl. Phys. B 518 (1998) 151, hep-th/9709013.
C.G. Callan, L. Thorlacius, Nucl. Phys. B 534 (1998) 121, hep-th/9803097.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J.M. Maldacena, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/9803131.
J.A. Minahan, Adv. Theor. Math. Phys. 2 (1998) 559, hep-th/9803111.
S.D. Avramis, K. Sfetsos, K. Siampos, Nucl. Phys. B 769 (2007) 44, hep-th/0612139.
S.D. Avramis, K. Sfetsos, K. Siampos, Nucl. Phys. B 793 (2008) 1, arXiv: 0706.2655 [nucl-th].
J.M. Maldacena, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002;
S.J. Rey, J.T. Yee, Eur. Phys. J. C 22 (2001) 379, hep-th/9803001.
A. Brandhuber, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 851, hep-th/9906201.
B. Baumgartner, H. Grosse, A. Martin, Nucl. Phys. B 254 (1985) 528;
C. Bachas, Phys. Rev. D 33 (1986) 2723.
G.T. Horowitz, A. Strominger, Nucl. Phys. B 360 (1991) 197.
S.J. Rey, S. Theisen, J.T. Yee, Nucl. Phys. B 527 (1998) 171, hep-th/9803135;
A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Phys. Lett. B 434 (1998) 36, hep-th/9803137;
A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, JHEP 9806 (1998) 001, hep-th/9803263.
D.K. Park, Nucl. Phys. B 618 (2001) 157, hep-th/0105039.
P. Kraus, F. Larsen, S.P. Trivedi, JHEP 9903 (1999) 003, hep-th/9811120.
K. Sfetsos, JHEP 9901 (1999) 015, hep-th/9811167.
M. Cvetic, D. Youm, Nucl. Phys. B 477 (1996) 449, hep-th/9605051.
J.G. Russo, K. Sfetsos, Adv. Theor. Math. Phys. 3 (1999) 131, hep-th/9901056.
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, JHEP 0007 (2000) 038, hep-th/9906194.
D. Berenstein, H.J. Chung, Aspects of open strings in Rindler space, arXiv: 0705.3110 [hep-th].
O. Lunin, J.M. Maldacena, JHEP 0505 (2005) 033, hep-th/0502086.
C. Ahn, Phys. Lett. B 641 (2006) 481, hep-th/0605012.
H. Liu, K. Rajagopal, U.A. Wiedemann, Phys. Rev. Lett. 98 (2007) 182301, hep-ph/0607062.
U.H. Danielsson, A.P. Polychronakos, Phys. Lett. B 434 (1998) 294, hep-th/9804141.
J. Polchinski, AIP Conf. Proc. 743 (2005) 331, Int. J. Mod. Phys. A 20 (2005) 3413, hep-th/0410082;
E.J. Copeland, R.C. Myers, J. Polchinski, JHEP 0406 (2004) 013, hep-th/0312067.
A. Vilenkin, E. Shellard, Cosmic Strings and Other Gravitational Defects, Cambridge Univ. Press, 1994;
M.B. Hindmarsh, T.W.B. Kibble, Rept. Prog. Phys. 58 (1995) 477, hep-ph/9411342.
R. Brandenberger, H. Firouzjahi, J. Karouby, Lensing and CMB anisotropies by cosmic strings at a junction, arXiv:
0710.1636 [hep-th].
A. Vilenkin, Phys. Rev. D 23 (1981) 852.
294
[35] E.J. Copeland, T.W.B. Kibble, D.A. Steer, Phys. Rev. Lett. 97 (2006) 021602, hep-th/0601153;
E.J. Copeland, T.W.B. Kibble, D.A. Steer, Phys. Rev. D 75 (2007) 065024, hep-th/0611243.
[36] C. Caratheodory, Calculus of Variations and Partial Differential Equations of the First Order, Amer. Math. Soc.,
1999;
G.A. Bliss, Calculus of Variations, Open Court, Chicago, IL, 1925;
R. Courant, H. Robbins, What is Mathematics? Oxford Univ. Press, Oxford, 1941;
G.B. Arfken, H.J. Weber, Mathematical Methods for Physicists, 6th edition, ElsevierAcademic Press, 2005.
[37] L. Durand, Am. J. Phys. 49 (1981) 334.

Higher order couplings from heterotic orbifold theory

Kang-Sin Choi a, , Tatsuo Kobayashi b
a Physikalisches Institut, Universitt Bonn, Nussalle 12, D-53115 Bonn, Germany
b Department of Physics, Kyoto University, Kyoto 606-8502, Japan
Received 20 December 2007; accepted 9 January 2008

Abstract
We calculate couplings of arbitrary order from correlation functions among twisted strings, using conformal field theory. Twisted strings arise in heterotic string compactified on orbifolds yielding matter fields in
the low energy limit. We calculate completely the classical and the quantum amplitude including normalization, up to a contribution from Khler potential. The classical action has saddle points which are interpreted
as worldsheet instantons described by metastable untwisted strings, formed by twisted strings distributed at
certain fixed points. This understanding generalizes the area rule, in the case that the locations of twisted
strings do not form a polygon, and provides a general rule for calculating these kinds of instanton corrections. An interpretation of couplings involving linearly combined states is given, which commonly appear
in non-prime order orbifolds. The quantum part of the amplitude is given by ratios of gamma functions with
order one arguments.
PACS: 11.25.Mj; 11.25.Wx; 11.30.Fs; 11.30.Hv
Keywords: Yukawa coupling; Higher order coupling; Heterotic string; Orbifold compactification
1. Introduction
Superstring theory is a promising candidate for unified theory including gravity. Heterotic
orbifold construction is one of interesting constructions for four-dimensional string models [1,2].
(See also for recent works Refs. [3,4] and for review [5].) One can solve the equation of motion
E-mail addresses: kschoi@th.physik.uni-bonn.de (K.-S. Choi), kobayash@gauge.scphys.kyoto-u.ac.jp

(T. Kobayashi).
doi:10.1016/j.nuclphysb.2008.01.016
296
K.-S. Choi, T. Kobayashi / Nuclear Physics B 797 (2008) 295321
of the string on the orbifold background, and the geometrical picture is clear in heterotic orbifold
models. Thus, several aspects can be computed and can be understood from the geometrical
viewpoint.
Heterotic orbifold models have modes localized at fixed points, that is, twisted strings. 3-point
couplings as well as 4-point couplings of these localized modes have been computed analytically
[610], and the size of Yukawa coupling Y is obtained as Y eA , where A denotes naively the
area of the triangle corresponding to three fixed points of twisted strings. This will be clarified
more and generalized in this paper. This aspect is quite interesting from the phenomenological
viewpoint. One can obtain suppressed Yukawa couplings when twisted strings are localized far
away from each other. That is, one could explain the hierarchy of quark and lepton masses as
well as their mixing angles when they are localized at different places.
We have to study selection rules for allowed couplings in order to examine whether realistic
fermion masses and mixing angels can be realized from string theory. The space group selection
rules [7,11] constrain allowed Yukawa couplings rather strongly. For example, on prime order
orbifolds off-diagonal Yukawa couplings are not allowed, and we cannot obtain realistic mixing
angles by using only 3-point couplings with the minimum number of Higgs fields.1 On non-prime
order orbifolds, off-diagonal Yukawa couplings are allowed [11] and possibilities for leading to
realistic quark and lepton masses and mixing angles have been studied [13]. However, to realize
fermion masses and mixing angles in string theory is still a challenging issue.
In this paper, we study generic higher order couplings than renormalizable couplings in heterotic orbifold models. Higher dimensional operators become effective
Yukawa couplings after
symmetry breaking. Suppose that we have a coupling of type Ff H i i in the superpotential of
the effective field theory, where F and f are chiral matter fields corresponding to quarks and leptons, H denotes electroweak Higgs superfields and i correspond to several heavy modes. When
all scalar components of the superfields i develop their vacuum expectation values (VEVs), this
higher dimensional operator becomes a Yukawa coupling among chiral fermions F and f and
the electroweak Higgs fields H . Thus, there is a possibility for deriving quark/lepton masses and
mixing angles through this type of symmetry breaking, but by use of not only 3-point couplings.
Indeed, such possibility has been examined in explicit models [24]. Therefore, it is important
to study the selection rules of allowed higher order couplings and compute the magnitude of
allowed couplings.2 When i correspond to localized modes on orbifold fixed points, the above
effective Yukawa coupling may correspond to a 3-point coupling on a CalabiYau manifold,
where orbifold singularities are smoothed by the VEVs of i . Thus, calculations of higher order couplings on the orbifold are also important from the viewpoint of calculations of 3-point
couplings on the CalabiYau manifold around the orbifold limit.
We compute magnitudes of L-point couplings. Generic aspects of L-point couplings in heterotic orbifold models have been obtained in [8]. Here we apply it to concrete heterotic orbifold
models. Similar calculation has been carried out for generic L-point couplings in intersecting
D-brane models [17].3
The paper is organized as follows. In Section 2, we give a brief review on heterotic orbifold
models in order to fix our notation. Then, we study the selection rule due to discrete R-symmetry
1 In non-factorizable orbifold models, off-diagonal Yukawa couplings are allowed, but it is still difficult to derive
realistic Yukawa matrices [12].
2 It would also be useful to study non-Abelian flavor symmetries [14,15] and accidental global symmetries [16] in
string models.
3 See for 3-point couplings in intersecting D-brane models [1821].
297
and the space group. In Section 3, we compute classical contributions of L-point couplings. Their
quantum parts are calculated in Section 4. In Section 5, we consider normalization of correlation
functions. Section 6 is devoted to conclusion and discussion. In Appendix A, we give useful
formulae for hypergeometric functions and their multivariable generalizations.
2. Setup
2.1. Twisted strings and their vertex operators
The heterotic string theory consists of 10D right-moving superstring and 26D left-moving
bosonic string. For the common ten (bosonic) dimensions, we consider the background with
our 4D spacetime and 6D orbifold. The other 16D left-moving bosonic string correspond to
a gauge part. A 6D orbifold is a division of a 6D torus T 6 by a twist , while T 6 is obtained
as R6 /, where is a 6D lattice. The twist must be an automorphism of the lattice , and
its eigenvalues are diag(e2i1 , e2i2 , e2i3 ) in the complex basis Zi (i = 1, 2, 3). We mainly
concentrate ourselves to the case that T 6 is factorized as T 2 T 2 T 2 . To preserve 4D N = 1
supersymmetry (SUSY), they must satisfy the following condition,
1 + 2 + 3 = integer,
where i is not integer for i = 1, 2, 3.
The twisted string is a closed string up to orbifold identification

Z e2i z, e2i z = k Z(z, z ) + v, v ,
(1)
(2)
where is the above lattice (in the complex basis) defining the orbifold. It makes sense to
restrict the phase to be 1 ki 1. Its zero mode satisfies the same condition, and it is called
a fixed point on the orbifold. The fixed point can be represented by the corresponding space
group element, ( k , v). Note that the fixed point ( k , v) is equivalent to ( k , v + (1 k )).
They belong to the same conjugacy class. The sector with k = 0 corresponds to the so-called
untwisted sector.
The local operator called the twist operator k (z) takes into account the nontrivial boundary
condition (2) by inducing a branch point at z with the order k/N on the world-sheet, but the
theory remains local by moding out by orbifold projection. The ground state corresponding to
the twisted string on the fixed point ( k , v) is generated from the untwisted ground state |0 by
the twist field |k = k (0, 0)|0. These twist fields have the operator product expansions (OPEs)
Z(z)k (0, 0) zk/N 1 k (0, 0),
z)k (0, 0) z k/N k (0, 0),
Z(
(3)
(4)
which are understood as the most singular parts in the mode expansion for 0 Nk 1. For
the other region 1 Nk 0, we have the corresponding relations by replacing k with N k.
Their conformal weights for the holomorphic and antiAlso we have similar expressions for Z.
holomorphic parts are

1 k
k
hk = hNk =
(5)
1
,
2N
N
thus inducing a shift of zero point energy.
298
Each k -twisted sector has several ground states, that is, twist fields corresponding to several
fixed points under the k twist. When we specify the fixed point f , we denote f,k . Also we use
the notation ( k ,v) , where ( k , v) denotes the space group element corresponding to the fixed
point f under the k twist.
On non-prime order orbifolds, fixed points under higher twist k (k > 1) are not always fixed
under or twist fields ( k ,v) are not always eigenstates of the twist . To make eigenstates, we
have to take the following linear combinations [11,22],

1
( )
( k ,v) ( k ,v) + ( k ,v) + 2 ( k , 2 v) + + k1 ( k , k1 v) ,
(6)
k
where = e2i/m with integer to be determined by gauge quantum numbers and internal
momenta. This linear combination may include twist fields corresponding to fixed points, which
belong to the same conjugacy class.
We consider the covariant quantization with the explicit conformal and superconformal
ghosts. It is convenient to bosonize right-moving fermionic string and write bosonized degrees
of freedom by H t (z). In the bosonized formulation, untwisted massless modes have momenta pt
for t = 1, . . . , 5, which are quantized on the SO(10) weight lattice. The spacetime boson and
fermion correspond to SO(10) vector and spinor, respectively. The compact space corresponds to
SO(6). The twisted sector Tk has shifted SO(6) momenta, ri = pi + ki , which are often called
H -momenta.
A bosonic massless state has the corresponding vertex operator,
V1 = e
3

mi Zi
Ni
N
I I
( )
m i Z i i eirt Ht eiP X eikX ( k ,v) ,
(7)
i=1
naturally in the (1)-picture, where is the bosonized ghost, P I X I corresponds to the gauge
part and kX corresponds to 4D part. Here, mi Zi and m i Z i denote oscillators for the left-mover,
and Ni and N i are oscillator numbers, which these massless modes include. Similarly, we can
write massless modes corresponding to spacetime fermions as
V 1 = e
12
3

mi Zi
Ni
(f )
N
I I
( )
m i Z i i eirt Ht eiP X eikX ( k ,v) ,
(8)
i=1
in the ( 12 )-picture. We understand that the H fields contain the four-dimensional spin field. The
(f )
H -momenta for spacetime fermion and boson, ri and ri in the same supersymmetric multiplet
are related to each other by
(f )
ri = ri
+ (1, 1, 1, 1, 1)/2,
(9)
that is, (1, 1, 1, 1, 1)/2 corresponds to the H -momentum of unbroken 4D spacetime SUSY
charge. To each vertex operator, we have to include overall normalization
Ni 1/2
N i
3 1/2

i
i
2
2
,
gc
(10)

(mi 1)!

(m
i 1)!
i=1
from the state-operator mapping or the unitarity relation [23]. The closed string coupling gc is
expressed in terms of ten-dimensional gauge and gravitational couplings
gc =
1/2 gYM
=
.
4
2
(11)
299
Thus, including one more field suppresses the corresponding coupling by one inverse mass dimension O( 1/2 ) as we expect. We have omitted the two-cocycles, which determine the overall
sign [24].
We calculate a correlation function among L twisted matter fields including two spacetime
fermions on the ZN orbifold, along the lines [79,17]. It yields a higher order coupling in the
zero momentum limit ki 0. Since the background has the superconformal ghost charge 2, the
correlation function is of the form,

L
L

dzi :ccV
1 (z2 , z 2 )::ccV
1 (z3 , z 3 ): :V0 (zi , z i ): , (12)
1 (z1 , z 1 )::ccV
lim e
ki 0
i=4
i=4
such that the total ghost charge vanishes. Here is the worldsheet cosmological constant and
we also have three bosonic ghost fields c(
z) and c(z). We take radial ordering implicitly, which
reflects the ordering property from noncommutative space group.
In order to make the total superconformal ghost charge vanishing, we need vertex operators
V0 in the 0-picture. We can obtain V0 by operating the following picture changing operator on
V1 [25],

v
i + e2iriv H Z i ,
Q = e e2iri H Z
(13)
where r1v = (1, 0, 0), r2v = (0, 1, 0) and r3v = (0, 0, 1) for the components corresponding to the
6D compact space. Thus we have

3

m Ni m N i

1/2 irt Ht iP I X I ikX
i Zi
i Z i
e
e
V0 = ( /2) e
ik k
,v
i=1
3

i=1
2iriv H

i + e2iriv H Z i (i)
Z
3

(j ) ,
(14)
j =1,j
=i
up to the same normalization (10). Here, (j ) is the j th component of ( k ,v) . Containing no

derivatives, the higher order coupling is defined in the zero momentum limit k 0. Thus the
first term is not relevant. The only change is that some components of twist fields are replaced
by excited twist fields , and the normalization factors ( /2)1/2 , in accord with the number of
oscillators in (10). In the next subsection, we see this change does not modify the calculation in
the case that all the twist fields are simply , not excited twist fields.
2.2. Selection rules
Here we briefly summarize the selection rules [3,7,11,26,27]. The vertex operator consists of
( )
several parts, the 4D part eikX , the gauge part eiP X , the 6D twist field ( k ,v) , the 6D left-moving
oscillators Zi and the bosonized fermion eirH , as explained in the previous subsection. Each
part has its own selection rule for allowed couplings. The
selection rules of the 4D part and the
gauge
part
are
simple,
that
is,
the
4D
total
momentum
k and the total momentum of the gauge

part P should be conserved. The latter rule is nothing but the requirement of gauge invariance.
The other parts lead to non-trivial selection rules. In this subsection, we study the selection rule
( )
from the H -momenta and oscillators, as well as the selection rule from the 6D twist fields ( k ,v) .
The total H -momentum should be conserved like the 4D momentum and the gauge momentum P . For example, for 3-point couplings V1 V1/2 V1/2 , they should satisfy the following
300
condition,

ri = 1.
(15)
Here we take a summation over the H -momentum for the scalar components, using the fact that
the H -momentum of fermionic component differs by 1/2.
Another important symmetry is the twist symmetry of oscillators. We consider the following
twist of oscillators,
Zi e2ii Zi ,
i e2ii Z
i,
Z
Z i e2ii Z i ,
Z i e2ii Z i ,
(16)
(17)
without summation over i = 1, 2, 3. Allowed 3-point couplings V1 V1/2 V1/2 should be invariant under the above ZN twist.
However, for generic L-point couplings we have to carry out picture changing, and the pic i
ture changing operator Q includes non-vanishing H -momenta and right-moving oscillators Z
and Zi . Thus, the definition of H -momentum depends on the choice of the picture. However,
the R-charges, which are defined as [3]4
Ri ri + Ni N i ,
(18)
are invariant under picture-changing. Here we do not distinguish oscillator numbers for the leftmovers and right-movers, because they have the same phase under ZN twist. Indeed, physical
states with 1 picture have vanishing oscillator number for the right-movers, while the oscillator
number for the left-movers can be non-vanishing. Thus, hereafter Ni and N i denote the oscillator
number for the left-movers, because we study the physical states with 1 picture from now. For
simplicity, we use the notation Ni = Ni N i . Now, the selection rule due to R-symmetry is
written as

Ri = 1 mod Ni ,
(19)
where Ni is the minimum integer satisfying Ni = 1/ i , where i = i + m with any integer m.
For example, for Z6 -II orbifold, we have i = (1, 2, 3)/6, and Ni = (6, 3, 2).
Whereas the twist operator k itself does not transform under the twist of oscillators, the
excited twist operator k transforms like the oscillator in (17), since it is nothing but the product
of an oscillator and a twist operator, from the transformational point of view. The modified H momentum has a compensating property and the resulting amplitude is invariant under the twist
of oscillators, as it must be because the picture changing operator is invariant. However the OPE
Zk is of a similar form of a twisted operator
Z(z)k (0, 0) zk/N 1 k (0, 0),
z)k (0, 0) z k/N k (0, 0),
Z(
(20)
which is readily extracted from the OPE ZZk . It has a branch structure like z2k/N2 , but
the part zk/N1 is carried by k from its definition to leave (20). Thus the amplitude including
excited twisted operators is the same as one including only twisted operators, up to the overall
normalization.
4 See also [28] and references therein.
301
We have the space group selection rule. Here, we study the selection rule for twist fields,

phases should satisfy = 1.5 Next, we study the space
group selection rule. Now, let us consider L-point couplings of twisted states corresponding to
( ki , vi ) (i = 1, . . . , L). Their couplings are allowed if the product of space group is the identity,
i.e.,
( )
( k ,v) . First of all, the product of ZN
L

ki , vi = (1, 0).
(21)
i=1
Since space group elements do not commute, nor do vertex operators, the ordering of vertex
operators in the coupling is important. We have to take into account the fact that ( k , v) is equivalent to ( k , v + (1 k )). Thus, the condition for allowed couplings is that the product of
space group elements must be the identity up to such equivalence.
The
space group selection rule

includes the point group selection rule, which requires ki = 1, i.e., ki = 0 (mod N ) for the
ZN orbifold. The rules for the linearly combined states are discussed in detail in Ref. [27].
3. The classical contribution
Here we consider the 6D ZN orbifolds, which can be factorized as three 2D ZN orbifolds, and
we concentrate ourselves on the calculation of correlation functions on the 2D ZN orbifolds. The
following analysis can be extended to other cases, where the 6D ZN orbifold is not factorizable
or the 6D ZN orbifold includes 4D non-factorizable orbifold.
The nontrivial part is the correlation function among L twist operators
Z k1 k2 kL .
(22)
This can be calculated independently of the remaining components of vertex operators, because
they commute. Some of them should be excited twisted states k for the total ghost charge being
2. Indeed we noted above that the amplitude is the same with a number of factors (2/ )1/2 .
From the point group selection rule we have
L

ki
i=1
M,
(23)
where M must be an integer. For the moment we assume

M = L 2,
(24)
and we will relax this condition later. This choice is the most convenient one because of two
reasons. First, by doubling trick we can relate the corresponding amplitude with that of open
strings [17], where this is the closedness condition for L-sided polygon. This gives rise to the
generalized SchwarzChristoffel transformation and the area rule that we will show below. Another reason is that we can obtain the closed form of integration, in terms of a multivariable
hypergeometric function [29].
If we use path integral formulation, the correlation function (22) is divided as

Z = Zqu
(25)
exp(Scl ),
{Zcl }
5 That is automatic for physical states when the other selection rules are satisfied [26,27].
302
according to the classical value and the quantum fluctuation around it, i.e., Z = Zcl + Zqu .
The corresponding tree-level Feynman diagram is a sphere with a number of vertex operators inserted. Every object in string theory, including the vertex operators and the correlation
function (22) can be separated into holomorphic and antiholomorphic part. Considering one of
them, holomorphicity restricts many things in a very simple form. Using the compactification
C {} = S 2 we can cover all the coordinate on the sphere by holomorphic coordinate z except infinity. To take care of infinity we introduce another patch z = 1/u where z becomes
u = 0. Considering a holomorphic solution from
u Z = z2 Z,
(26)
if we need the LHS well-behaved at u = 0, in the RHS Z should drop faster than z2 as z goes
to infinity.
3.1. L-point coupling
We begin with calculating the classical contribution first. We see that the classical solution is
the completely factorized part for each inserted operator. Therefore, from the OPEs (3) and (4),
the classical solutions are obtained by the holomorphicity and the desired singular structures as
cl (z) =
Z
Zcl (z) = a(z),
L2

bl l (z).
(27)
l=2
Here we define the basis of (L 2) functions

(z) =
L

(z zi )ki /N 1 ,
i=1
l (z) =
L

(z z i )ki /N
L2

(z z j ),
l = 2, . . . , L 2.
(28)
j =2,j
=i
i=1
For the symmetric orbifold they are complete since Z cl (z) = (Zcl (z)) and Z cl (z) =
cl (z)) . From (26), the whole part (27) should behave as z2 as z . Using (24), we
(Z
can see (z) does. For l (z), we admitted an additional degree of freedom, i.e., changing the
power of z zj singularity by integer, since it does not modify the branch cut structure. Hence,
we have many free parameters bl .
To determine the coefficients a, bl , we should consider the global monodromy condition. The
relation (2) does not take into account the global phase if we transport branch cuts from more
than one fields. Taking a contour C encircling more than one point gives the relation between
overall coefficients in (27) and net translation v in the target space

= v.
C Z = dz Z + d z Z
(29)
C
This expression makes sense only if there is no additional phase. The quantum part does not
carry any amplitude C Zqu = 0. Hence, Eq. (29) shows a purely classical contribution. Upon
integration relating these branch cuts, the Pochhammer loop [30] is a clever way to encompass
the two branch points nontrivially without a phase. For each branch cut the contour goes in and
out exactly once through the cuts, as depicted in Fig. 1. Its effect is to encircle the fixed points:
303
Fig. 1. Pochhammer loop. Whatever direction the branch cut we have, the nontrivial loop goes in and out respectively
exactly once through each cut, enclosing two branch points.
f1 clockwise, f2 counterclockwise, f1 counterclockwise and then f2 clockwise. In terms of

space group elements (, v1 = (1 )(f1 + v1 )) and (, v2 = (1 )(f2 + v2 )) with v1,2 ,
we obtain

(, v2 )(, v1 )1 (, v2 )1 (, v1 ) = 1, 1 1 v2 + (1 )1 v1

= 1, 1 1 (1 )(f2 f1 + v) ,
(30)
where v = v2 v1 . The encircling is not necessarily once, i.e.,
= in general, in which we
cannot draw branch cuts. The net effect is a pure translation. It turns out that every contour is
generated by the basis of Pochhammer loops Ci encircling ith and (i + 1)th points.
Taking Pochhammer loops Ci around the vertices zi and zi+1 , from (30) we obtain

cl (z)
Ci Zcl = dz Zcl (z) + d z Z
Ci
Ci

1 e2iki+1 /N (fi+1 fi + v)

ki
ki+1
= 4ei(ki ki+1 )/N sin
sin
(fi+1 fi + v).
N
N
= 1e
2iki /N
(31)
We have (L 2) vectors fi+1 fi + v and L angles (with the constraint (24)), which completely
specify L-sided polygon.
Later we can express the solution in terms of the following integrals

1
i
Wl dz (z),
(32)
Wl d z i (z),
Cl
Cl
and

Wli = 1 e2iki /N 1 e2iki+1 /N Fli ,
i = 1, . . . , L 2,
(33)
i = 2, . . . , L 2.
(34)
with
Fl1
zl+1
=
(z) dz,
zl
Fli
zl+1
=
i (z) d z ,
zl
Note that Fli and Wli form (L 2) (L 2) matrices. In Appendix A, they are expressed in
terms of multi-valued hypergeometric functions [17,29]. With SL(2, C), we can set z1 , zL1 , zL
to be 0, 1, respectively and the others to the cross-ratios of xi .
304
Plugging (27), the solution is expressed as

ci Fli = fl+1 fl + v,
where we defined
ci =
L2

l = 1, . . . , L 2,
c1 a, ci bi ,
(35)
and by inverting them we obtain
l

(fl+1 fl + v) F 1 i ,
(36)
l=1
where the inverse is taken with respect to the matrix basis with indices l, i. Plugging into the
classical action, we obtain the final solution

1
2

Scl (x2 , . . . , xL2 ) =
(37)
bi bj Ij ,
|a| I +
4
i,j
where

2
d 2 z (z) ,
I (x2 , . . . , xL2 ) =
C
Ij (x2 , . . . , xL2 ) =

d 2 z i (z) j (z) .
(38)
We can expand this action by products of holomorphic and antiholomorphic functions, with careful choices of contours. It is nothing but the relation between open and closed string amplitudes
before integration over xi s [31],

L1
i

kl
i
(1) 1 exp 2i
Fi F1
I (x2 , . . . , xL2 ) =
N
i=2
l=2

j
j 1
L2

kl
j +i+1
+
(1)
(39)
Fi Fj ,
1 exp 2i
N
j =2 i=0
l=i+1
where
0
F0
dz
L1

(z xj )(1kj /N) ,
Fi = Fi1 ,
i = 1, . . . , L 2,
j =1
FL1
dz
1
L1

(z xj )(1kj /N) .
(40)
j =1
Plugging in (37), we obtain the classical action Scl (x2 , . . . , xL2 ). It is a function of L 3
complex variables xi which will be integrated out in the final amplitude. Later, we will integrate
this with variables x2 , . . . , xL2 over the entire complex plane. Among these, using the saddle
point approximation by adjusting xi s or equivalently ratios Fi+1 /Fi , i = 2, . . . , L 2, we find a
minimum
Fi+1 fi+2 fi+1
=
.
Fi
fi+1 fi
(41)
305
Note that in general the integrals Fi are complex and we coordinated the fixed points as complex
vectors on a given 2D torus and orbifold.
Inserting these into (36), we have ci = bi = 0 for all i > 1, except a = c1
= 0. The solution
is nothing but the generalized SchwarzChristoffel transformation [29], whose original version
maps the upper half plane into inside an L-polygon
Zcl (z) = a(z).
(42)
Namely, the points xi are mapped to vertex zi and the turning around angle is given by ki /N .
In this case, we obtain the instanton contribution is exponential of the polygon area

1
Zcl,min exp
(43)
(area
of
the
polygon)
.
2
This is valid under the assumption (24), i.e., forming a polygon, but in general case we have
more fundamental interpretation shortly.
Of course, there are other minima with the same value, where a and all bi vanish except
j
j
one, say bk
= 0. This corresponds to Fi+1 /Fi = (fi+2 fi+1 )/(fi+1 fi ) and the Schwarz
cl (z) = b k (z).
Christoffel transformation corresponds to Z
k
In forming the area from the SchwarzChristoffel transformation, the ordering is important. If
we just exchange two fields, we cannot satisfy the space group selection rule in general, and the
polygon becomes self-crossing, where the area rule is not applicable. In the correlation function
we take the radial ordering. In the superpotential of effective field theory, we do not see the
ordering, since the integration over all zi completely symmetrize the amplitude.
3.2. Four-point correlation function
The four-point correlation function provides a good example of calculation of the classical
part. In this case, the functions Fi are well-known hypergeometric functions, which are solutions
of second order linear differential equation. It is known [32] that if any three of the solutions
have the common domain of existence, there be a linear relation among them. In our case, we can
express all of Fi in terms of, say, F1 and F2 . They are shown in (A.3) and (A.4) of Appendix A.
Plugging these to (39) the holomorphic part is obtained as

F1 F2 + c22 |F2 |2 ,
I (x) = c11 |F1 |2 + c12 F1 F2 + c12
(44)
where
sin(k1 /N) sin((k2 + k3 )/N)
sin(k3 /N) sin((k1 + k2 )/N)
,
c22 =
,
sin(k4 /N)
sin(k4 /N)

sin((k1 + k2 )/N) sin((k2 + k3 )/N)
.
c12 = eik2 /N sin(k2 /N) +
sin(k4 /N)
The coefficient c12 reduces to
c11 =
c12 = eik2 /N sin(k1 /N) sin(k3 /N),
(45)
(46)
eik2 /N
in c12 is the relative phase of

only for the polygon case, using (24). The prefactor
(complex numbers) v32 and v21 , where vij = fi fj + v with v . From (36) we have the
coefficients
v32 F1 + v21 F2
v32 F1 v21 F2
a=
(47)
,
b=
.

F 1 F 2 F2 F 1
F1 F2 F2 F1
306
We can obtain the antiholomorphic action I (x)

and integral Fi (x)
from I (x) and Fi (x) respectively, by substituting ki /N 1 ki /N and x x.
With these we obtain the classical
action (37). The action does not have manifest duality symmetry.
We define the following modulus
F2
,
F1
(48)
which is in the Z2 case the modular parameter of two-torus, made by connecting two Riemann
sheets with two branch cuts [6,7]. As expected from (41), the minimum of I (x) is obtained at
=
v32
.
v21
(49)
Thus we have a = v21 , b = 0 and the minimum of classical action is obtained as

1

(50)
c11 |v21 |2 + c12 v21 v32
+ c12
v21 v32 + c22 |v32 |2 .
2
For the case of polygon, the classical action Scl,min reduces to

1
|v14 |2 sin(k1 /N) sin(k4 /N) |v32 |2 sin(k2 /N) sin(k3 /N)
Scl,min =
2
2
sin((k1 + k4 )/N)
2
sin((k2 + k3 )/N)

2
2
|v43 | sin(k3 /N) sin(k4 /N) |v21 | sin(k1 /N) sin(k2 /N)
1
. (51)
=
2
2
sin((k3 + k4 )/N)
2
sin((k1 + k2 )/N)
Scl,min =
This is the area of the quadrilateral formed by vertices at the fixed points f1 , f2 , f3 and f4 .
In the case with k1 + k4 = N and/or k2 + k3 = N , this expression is not well-defined. Without
loss of generality, the case with k1 + k4 = N and k2 + k3 = N leads to k1 = k3 = N k, k2 =
k4 = k, by use of (24), and such a case has been calculated in [7]. Here we have to come back
to (44), and the result agrees.6 The case with k1 + k4 = N or k2 + k3 = N leads to k1 = N k,
k2 = k, k3 = N l, k4 = l, and such a case has been calculated in [9].
From four-point amplitude, we can obtain three point amplitude by taking x to, say, . In
this case the fixed points f3 and f4 become coalescent and the classical action Scl,min reduces to
Scl,min =
1 |v32 |2 sin(k2 /N) sin(k3 /N)

.
2 2
sin((k2 + k3 )/N)
(52)
Note that this action depends on the choice of contour picture. Here, we chose one encircling
two fixed points f3 and f2 . We do not need worry about whether v is actually compatible to
factorization [7]. The Pochhammer loops in which v32 and v14 belong are independent.
In the special case with k1 = k2 = k3 = k4 = N/2, we have c11 = c22 = 0, c12 = i and
F1 = F1 , F2 = F2 yielding
Scl =
v32
1
2
2v21
|v21 v32 |,
=

2
i
2
(53)
which is again interpreted as twice the area of the rectangle, in unit of 2 if v21 is orthogonal
to v32 . This is the case of order 2 subsector ((N/2)th twisted sector) in even order orbifold. Note
that this action is not the minimum action, since in this case the classical action is not the function
6 To compare between our results and [7], we have to replace our modulus by e i(k/N 1) .
307
Fig. 2. (a) On Z4 orbifold, an order 4 couplings: two at a vertex, and the other two at the opposite vertex. (b) Massless
strings are localized. (c) Quantum effect grows twisted strings to form a untwisted string, which is a intermediate state
described by instanton. The shaded area, where v21 is the diagonal, is the minimal area swept.
of . In this case the area rule interpretation is somehow ambiguous. We will study more detail
in the following subsection.
We have considered L-point couplings only for L = 2, 3, 4 as examples. However, we will
study that higher order L-couplings reduce to a combination of lower order L -couplings with
L < L by the discussion of field coalesce in Section 3.4. In addition, since the number of fixed
points on T 2 /ZN orbifolds is limited, we can expect that most of higher order L-couplings
can be written as combinations of L-point couplings only with L = 2, 3, 4. We will study this
expectation in separated papers [27,33], by examining concrete orbifold models.
3.3. Non-polygon case: The meaning of area
Until now, we assumed the polygon condition (24) is satisfied. However, in general, the following relation
L

ki
i=1
L 2,
(54)
is possible. In the inequality case, the holomorphic part of the classical solution (27) decays
faster than z2 , whereas the antiholomorphic part decays not faster than z2 . The only sensible
treatment is to make antiholomorphic part vanishing.
For example, in the Z4 orbifold, the coupling of four first twisted sector fields,
(,0) (,e1 ) (,0) (,e1 ) ,
shown in Fig. 2(a), satisfies the space group selection rule. Note that (,0) (,0) (,e1 ) (,e1 ) does
not satisfy the space group selection rule, and the ordering is important. All of twist fields are of
order four, which cannot satisfy the relation (24). The classical solution is obtained as
Zcl = a(z z1 )3/4 (z z2 )3/4 (z z3 )3/4 (z z4 )3/4 ,
cl = 0.
Z
From global monodromy condition, we have

a=
v21
,
F1
F2
1
= (1 + i),
F1
2
and the classical action is given with c11 = c22 = 1, c12 = 32 (1 + i), yielding
I = 2|F1 |2 .
(55)
308
Fig. 3. Naive area rule is the special case of the swept-are rule when the twisted strings form a polygon.
Thus we have the classical contribution

Scl =
1
2|v21 |2 ,
2
which is interpreted as the area of the square whose diagonal is v21 .

Similarly, for the coupling
(,0) 2 ,e1 (,e1 ) ,
we obtain the classical action,
Scl =
1 1
|v21 |2 .
2 2
In these cases, we have a different area rule: The area is not that surrounded by fixed points,
but one as follows. The classical solution describes a local minimum of the action, which is the
instanton of worldsheet nature, suppressed by . The selection rule tells us that these twisted
strings can potentially make an untwisted string, which is energetically not possible for massless
strings, completely localized at certain fixed points, as in Fig. 2(b). However they can oscillate
and grow to be larger sizes, and above a certain threshold, they can form an untwisted string as
in Fig 2(c). Noting that the instanton describes tunneling between vacua which is energetically
forbidden, we can understand that forming untwisted string corresponds to such tunneling.
Still we can have the hint from the modified area. It is the sweeping area for localized twisted
strings to grow to become a untwisted string. For the polygon case, i.e., that satisfying the condition (24), this interpretation is still valid, since still the area swept by twisted strings at each
vertex makes the polygon area.
On the other hand, in the case not satisfying the condition (24), we lost the interpretation
of the mapping Z being a generalized SchwarzChristoffel transformation, since in the target
space the fixed points fail to make a polygon.
The subleading correction is generated by identical fixed points on the orbifold, but more
separated in the covering space, i.e., f + (1 k ). Since they are identical points, they satisfy the selection rule. With this interpretation, we can understand the classical solution which
makes more than one untwisted strings possible. For example, for (,0) ( 2 ,e1 ) (,e1 ) , the coupling (,0) ( 2 ,3e1 ) (,3e1 ) , where each fixed point belongs to the same conjugacy class as the
previous one, satisfies the space group selection rule, but corresponds to a large v21 and a large
instanton action.
309
3.4. Fields coalescent at the same fixed point

In most cases, some of the fields sit at the same fixed point f . The corresponding correlation
function might be obtained by taking the limit zj zi in the correlation function,

f,k1 (zi ) f,k2 (zj ) .
(56)
Note that this limit is not always well-defined. It is because twist fields do not commute. By
conformal symmetry, the OPE has the generic form,
f,ki (zi )f,kj (zj ) cij (zi zj )ij f,ki +kj (zj ).
Equating the conformal weights of the both sides, we have

ki kj
N
(ki + kj N ),
N
ij = hki +kj hki hkj =
k
ki
(1 N
)(1 Nj ) (ki + kj N ).
(57)
(58)
Because of the nontrivial branch cut, this relation is asymmetric under the exchange of two twist
fields. This property is also reflected in the space group elements, which do not commute, either.
We can define an invariant block of twist fields, which correspond to the identity of the space
group (1, 0). Thus, these invariant blocks commute and satisfy

(59)
ij = 0 mod 1.
j
In the case where two points zi and zi+1 are in the successive order, and we can merge two
twists without ruining the radial ordering. They are neighboring points as polygon vertices. Then,
from (57) the two twists reduce to a single twist with the summed order. Also this implies the
classical solution becomes
(z zi )ki /N 1 (z zi+1 )ki+1 /N 1 (z zi )(ki +ki+1 )/N2 ,
(60)
where, again, the even integral power is not relevant to branch structure, so that we can make
ki /N + ki+1 /N 1 lie in [1, 1) for instance. This means, not all of L vertices form the Lpolygon, but effectively one with the lesser vertices. (Recall that the positions of vertices are
given as singularities in the classical solution.) In fact this is the familiar case when we obtain
a three point function from the four point function by setting two of the points coalesce. In the
latter limit, the polygon is triangle. We can see this in terms of space group elements. Neglecting
gauge group, which is not involved in the classical amplitude, we cannot distinguish the product

(, v)2 and 2 , v + v ,
(61)
when two identical twisted fields sit at the same point, and are put on the neighboring points in
the correlation function.
There is the case where ki + ki+1 is integer. In this case there is no singularity in the classical
solution and such vertex does not contribute to the area rule. In the extreme case where k order-k
twisted fields sit at the same points, the coefficients of higher order couplings are not suppressed.
Thus all we need to consider is the other nontrivial couplings. Fortunately, not all of them survive:
From R-symmetry invariance, the correlation function is further constrained.
However, from the radial ordering, there is a case in which the exchange of two branch cut
fails to give well-defined radial ordering. This is not possible for two fields which are not the
310
Fig. 4. The rule for worldsheet instanton sum.
neighboring fields in the correlation function, since the space group and the mapping of classical
solution is not commutative.
For example, in couplings among twist fields corresponding to two (, 0)s, two (, e1 )s and
two (, e1 + e2 )s on the Z3 orbifold, there are two following combinations possible, satisfying
the space group selection rule
(,0) (,0) (,e1 ) (,e1 ) (,e1 +e2 ) (,e1 +e2 ) ,
(62)
(,0) (,e1 ) (,e1 +e2 ) (,0) (,e1 ) (,e1 +e2 ) .
(63)
In the first case, the solution behaves like

Z = a(z z1 )2/3 (z z2 )2/3 (z z3 )2/3 (z z4 )2/3 (z z5 )2/3 (z z6 )2/3 .
(64)
The two points, zi and zi+1 for i = 1, 3, 5, correspond to the same fixed points on the target
space, i.e., Z(z1 ) = Z(z2 ) = f1 , Z(z3 ) = Z(z4 ) = f2 and Z(z5 ) = Z(z6 ) = f3 . Thus, we take
the limit zi+1 zi for i = 1, 3, 5. In such a limit, the solution behaves like
Z = a(z z1 )4/3 (z z3 )4/3 (z z5 )4/3 .
(65)
Thus, in the space group point of view, it is indistinguishable from the coupling among the second
twisted couplings
( 2 ,0) ( 2 ,e1 +e2 ) ( 2 ,e2 ) ,
(66)
which gives only the area of triangle. On the other hand, the latter coupling (63) gives twice the
area of the triangle. The classical solution maps from the different points to the same points. For
example the holomorphic part behaves
Z = a(z z1 )2/3 (z z2 )2/3 (z z3 )2/3 (z z4 )2/3 (z z5 )2/3 (z z6 )2/3 , (67)
and the two points, zi and zi+3 for i = 1, 2, 3, correspond to the same fixed points on the target
space, i.e., Z(z1 ) = Z(z4 ) = f1 , Z(z2 ) = Z(z5 ) = f2 and Z(z3 ) = Z(z6 ) = f3 . Otherwise the
ordering is ruined. It is also understood that two different and independent sets of twisted fields
can sweep the triangle, which means the instanton corrections from the complete polygons are
additive in the action. This case is depicted in Fig. 4.
311
We summarize how to calculate the classical part of L-point couplings among twist fields ki
(i = 1, . . . , L). First, we classify all of possible ordering of these twist fields, which satisfy the
space group selection rule. For each ordering of twist fields allowed by the space group selection
rule, we consider the following procedure. We combine two or more twist fields sitting at the
same fixed point to a single twist field like (57) and Fig. 3 if possible, that is, they satisfy (59).
When their total twist is just (2n) twist with integer n, correlation function reduces to much
simpler form. The resultant correlation function can be written as product of invariant blocks,
which satisfy the space group selection rule like Fig. 3. Each block includes smaller number L
of twist fields.7 Then, for each block, we calculate classical contributions, eScl , i.e., instanton
actions
corresponding to the minimum action and larger ones, and take their summation, i.e.,

i,{vi+1,i } exp(Scl ). Next, we take a production of classical contributions corresponding to each
block, i.e.,

(68)
exp(Scl ) .
invariant blocks
i,{vi+1,i }
Finally, we sum over all of possible ordering to obtain the total coupling, i.e,

exp(Scl ) .
possible ordering invariant blocks
(69)
i,{vi+1,i }
Because of the ordering, the correlation function does not possess the worldsheet duality like
in the VirasoroShapiro amplitude [34]. In the vanishing momentum limit we do not distinguish
the channel, thus the effective coupling like Yukawa coupling does not distinguish the order.
However even in this case, the above two cases are distinguishable, since this contribution is the
worldsheet effect, suppressed by .
3.5. Linearly combined states
In higher order twisted sector of a non-prime order orbifold, there are states formed by linear
combinations, as in (6), due to the orbifold projection. The linear combination of states can be
more precisely defined by that of vertex operators. It follows that the corresponding classical
solution consists of linear combination of individual solutions before combination, and also the
relative weights are inherited. For example the classical solution involving (6) contains the factor

1
(z z1 )1+k/N + (z z2 )1+k/N + + k1 (z zk )1+k/N ,
k
(70)
where zi is mapped to the fixed points (, i1 v).

For such couplings, we observe two points:
1. The selection rule for a linearly combined state is derived from that for each term.
2. A part of the classical solution, which does not satisfy the selection rule vanish.
The proof for the first is given in Ref. [27], and we easily see the latter is the case. We can show
that such term not satisfying the classical solution always contain the same element more than
7 In most of cases, L may be equal to L = 2, 3, 4. We will study this point in concrete models [33].
312
once. Setting these to the other(s) corresponding to shrinking the area to zero, if we do not want
to change the angles corresponding other vertices.
For instance, in the second twisted sector of Z4 orbifold, we have a coupling including linearly
combined states
1
1
( 2 ,0) (( 2 ,e1 ) ( 2 ,e2 ) ) ( 2 ,e1 +e2 ) (( 2 ,e1 ) ( 2 ,e2 ) ).
2
2
(71)
The total coupling is given by summation over each of the four correlation functions. However,
the nonvanishing contributions come from the only ones satisfying the selection rule. One that
does not satisfy the rule, e.g., ( 2 ,0) ( 2 ,e1 ) ( 2 ,e1 +e2 ) ( 2 ,e1 ) , does not contribute.
To sum up, for the linearly combined state, we can treat a linearly combined state as a complete
physical field, the classical amplitude has only contribution from the parts satisfying the space
group selection rule.
4. The quantum amplitude
Now we determine the quantum part of the amplitude (25),
Zqu = k1 kL qu .
(72)
We use the stress tensor method [7,9,17], in which (72) can be indirectly calculated, relying on
only the holomorphicity. All the information can be read from the Greens function

12 z Zw Z ki
.
g(z, w; zi )
(73)
Zqu
Its form we know from the holomorphic part of the OPE (3),

(z zi )(z zj ) k
=i,j (w zk )

g(z, w) = (z) (w)
aij
+
A(w)
,
(z w)2
(74)
i<j

where (w) = (w zi )ki /N . The condition i < j avoids overcounting, although otherwise
the formula would be more symmetric. The prefactor (z) (w), from (28), gives the desired
pole structure. The normalization is determined by requiring the z w limit of g(z, w) to be 1,
and the condition

aij = 1,
(75)
i<j
leads the desired conformal weight hk , as the coefficient of each double pole in z w. To have
no residue in z zi , we require the condition for aij ,
L

j =1
aij = 1
ki
,
N
(76)
where we use the fake number aij = aj i for the case i > j . Summing over i, this condition
implies (75). We have less condition for many aij , reflecting the freedom of our choice of A. Of
course, the physics does not depend on the choice aij . As z w, we have the OPE

T (z) ki
1 ki k j
1
Zqu
2
N N (z zi )(z zj )
i<j

2
1
A
1
1
aij
+
+
.
+
2
z zi
z zj
i (z zi )
313
(77)
i<j
We extract the residue (z zk ) for fixed k, and take z zk limit.

In the limit w z, the Greens function becomes energy-momentum (stress) tensor

1
1
,
T (z) lim z Zw Z
wz
2
(z w)2
(78)
where the last term arises from the normal ordering. Sandwiching the OPE in the correlation
function
hk kk (zk ) zk kk (zk )
+
,
T (z)kk (zk )
(79)
z zk
(z zk )2
with the fixed kk (zk ) and its conformal weight hk given in Eq. (5), we can completely calculate
h up to normalization
the holomorphic part Zqu

h
zk Zqu
T (z) ki qu
hk
h
zk ln Zqu =
(80)
= lim (z zk )
.
h
h
zzk
Zqu
Zqu
(z zk )2
We obtain
h
zk ln Zqu
= zk

i<j

kj
ki
A
1
+
1
.
aij 1
N
N
zk zj
(z
i
=k k zi )
(81)
At this stage, using SL(2, C) symmetry, we set z1 0, z2 x, zL1 1, zL . Then

zL dependent terms vanish since zL , but there are corrections to this from A, which are
also dependent on zL . We now have L unknown for L 1 constraint, and remaining freedom
corresponds to redefinition of A.
Similarly we define the antiholomorphic Greens function
h(z, w) =
L2

Bij i (z)j (w).
(82)
i,j =2
To obtain A and Bij , we apply the global monodromy condition for the quantum part

dz g(z, w) + d z h(z, w) = 0.
Cl
(83)
Cl
It is convenient to call the first term in the RHS (74) as gs (z, w). If all the zi are different, the
matrix Wil has the inverse. Thus we can multiply it to eliminate A and Bij , so that we obtain

(w)A =
L2

l
W 1 1
l
L2

j =1
j (w)Bij =

dy gs (y, w),
Cl
L2

W 1
l

dy gs (y, w).
i
Cl
(84)
314
Since we set zL , the only relevant terms are ones containing the factor aiL (z zL ). We
divide them by (w) and take the limit w . Then, in the sense of (79), we read off the
residue of (z zk ) in the integrand

gs (z, w)
lim (z zk ) lim
zzk
w (w)

1
1
kk
zk (z)
(85)
aij
+
,
1
zk zj
N zk zj
j
=k
i<j
where we have replaced aiL with aij by the relation (76). Therefore we have no dependence on
the specific choice of gs (z, w), or aij , as it should be.
Integration around the contour Cl gives

l
h
W 1 1 zk Wl1
=
zk ln Zqu
l
+ zk ln

(1 Nk )
(zk zj )
j
=k
(1 Ni )(1 Nj )
(zi zj )

(86)
i
=j
The first term in the RHS is contained in the derivative of det W det Wlk . By the chain rule, we
obtain
zk ln(det W ) =
L

L
L3

l
1 l
j
W 1 i zk Wli +
W
W .
j zk l
(87)
j =1,j
=k l=1
l=1
For the second term in the RHS of (86), in the same way, we find the singular structure as z zk ,

L2
j
=k l=1

ki
l
j
W 1 j zk Wl = zk ln
(zj zi ) N .
(88)
j
=k
Thus we obtain

h
zk ln Zqu
= zk (det W )
ln
L3

(zi zj )
i,j =1,i<j
L

(1 Ni )(1 Nj )
(zi zj )

(89)
i<j =1
up to antiholomorphic
action. In fact, there are additional factors, since we have used only the part

of the terms l (W 1 )l1 zk Wl1 in (87). The remaining terms completely vanish if we calculate
the antiholomorphic part in the same way. The similar analysis is carried out for Bij , by dividing
h Z antih , we arrive
by i (w). Combining it with the holomorphic part Zqu = Zqu
qu
Zqu = (det W )1
L2

i,j =1,i<j
(zi z j )
L

ki
kj
ki kj
N
(zi zj )(1 N )(1 N ) (zi z j ) N
(90)
i,j =1,i<j
up to the normalization, to be determined in the following section. The various branch cuts reflect the noncommutativity of the vertex operators under permutation, due to orbifold phase. As
well as the classical part, the quantum part has no world-sheet duality, either. Nonetheless the
complete amplitude will be single-valued, completed with other parts of vertex operators.
315
Fig. 5. Factorization of 2(L 1)-point amplitude to [2(L 1) k]-point and k-point amplitudes. Unitarity tells us the
intermediate states have mass poles, whose residues are the product of the factor amplitudes. The correct normalization
does not depend on a specific factorization and choice of intermediate state.
5. Factorization and normalization

To normalize the amplitude, we choose the reference of normalization

k
k
k (z, z )Nk (0, 0) = 1 |z|2 N (1 N ) , z 0.
(91)
Equivalently, two twist fields of the opposite twists coalescent at the same point become the
identity operator. In general we cannot make use of the normalization (91), since for each twist
field we need one with the opposite twist. Thus we use a doubling trick.
First we calculate 2(L 1) point function with twists

S(1; 2; . . . ; 2L 2) = k1 (z1 )Nk1 (w1 ) kL1 (zL1 )NkL1 (wL1 ) .
(92)
By setting wL1 zL1 we obtain the product of a (2L 3)-point function and a 3-point
function. From the unitarity, the intermediate state gives an on-shell pole, with the lowest state
being massless gauge bosons,
d 4 k S(1; 2; . . . ; 2L 4; j )S(j ; 2L 3; 2L 2)
.
S(1; 2; . . . ; 2L 2) i
(93)
(2)4
k 2 + i
This factorization is schematically drawn in Fig. 5.
Concentrating only twist fields, the intermediate gauge boson does not contain a twist field.
The former contains only 2L 4 twist operators and the latter has the form as in (91). Therefore
we have
k1 Nk1 kL1 NkL1 = k1 Nk1 kL2 NkL2 kL1 NkL1 .
(94)
We have already known the general solution. For this we have the same action, including the

classical and the quantum parts, except the factor det W . In this, each term l Wlil contains the
common factor

L2

ki+1
ki
sin
.
4ei(ki+1 ki )/N sin
(95)
N
N
l=1
Thus there is a discrepancy between the two by the factor

kl
det W(l)
kl1
= 4ei(kl kl1 )/N sin
sin
,
det W(l1)
N
N
(96)
where the subscript in the determinant indicates that these are for l and l 1 point correlation
functions, respectively. This is so, since the normalization of 2(L 1) point function have no
316
special dependence of a specific ki . In effect, we replace det W in (90) with det F , which is
defined in (34). This is a desirable result, since any choice of factorization should give the same
result, not depending on the special set of contour choice, which is reflected in sin i .
We cannot go from the four-point function into the product of two two-point functions, since
the point z cannot be arbitrary close to all of 0, 1, , already fixed by SL(2, C), at the same time.
For this, we do the Poisson resummation [7]. Since it involves a lattice transformation into its
dual lattice, as a result we have the overall factor of the lattice volume.
Finally we set zL , zL+1 , . . . , z2L2 to infinity to obtain the OPE

k1 kL1 L1 k L1 Nk Nk1 NkL1 , (97)
k1 k2L2
1
( k ,v)P,sel
where the selection rules should be satisfied. The RHS is nothing but the product of two identical
(L 1)th order couplings. Comparing the coefficients, we obtain the coupling of order L
YL-point (k1 , k2 , . . . , kL )

1
= lim det F(2L2) (k1 , k2 , . . . , kL1 , N k1 , N k2 , . . . , N kL1 ) 2 ,
all wi
(98)
where
ki
kL
=L1
.
N
N
L1
(99)
j =1
For the rest of vertex operator components, we have

c(z
1 )c(z1 )c(z
2 )c(z2 )c(z
3 )c(z3 ) = |z1 z2 |2 |z1 z3 |2 |z2 z3 |2 ,

1
1
e (z1 )e 2 (z2 )e 2 (z3 ) = (z1 z 2 )1/2 (z1 z 3 )1/2 (z2 z 3 )1/4 ,

(z2 ) (z3 ) = (z2 z 3 )1 ,

L

|zi zj | ki kj ,
eiki X (zi , z i ) =

iPi X
(zi ) =
e

eiri H (zi ) =
i,j =1,i<j
L

(zi zj ) Pi Pj /2 ,
i,j =1,i<j
L

(zi z j ) ri rj /2 .
(100)
i,j =1,i<j
Multiplying these, and using the massless condition, the branch cuts disappear in the overall
amplitude. Including universal geometric factor of sphere 8gc2 1 , we have the overall factor
1
L2 (L2) 2 (L4)
e
,
2(4)L+3 g10
(101)
up to contributions from picture changing and/or oscillator excitations (10). For the four point
coupling, we have volume factor suppression from the compactification.
As the simple example, we can extract the information for three point correlation function.
By SL(2, C) symmetry, we can fix three arbitrary given points completely, thus the correlation
317
function contains no information. To obtain it, we need four point correlation function with twists

l (1, 1), Nl (, ) ,
k (0, 0), Nk (x, x),
(102)
where we consider only one two-torus. From (32), we have
det F (x, x)
= F11 F22 F12 F21

k l
k l
l
k
k
l
=B 1 ,
F 1 , ;1 + ;1 z F
, 1 ; 1; z
N N
N N
N
N
N
N
(k l, z z exchanged).
(103)
Let us assume k + l < N . In the limit z, z , using the relations (A.5) and (A.6) in
Appendix A, the most dominant part has the coefficient
ik

2 (1 k+l
e
eil 1
N )
+
,
det F
(104)
2 (1 Nk ) 2 (1 Nl ) sin l
sin k
N
N
where we used the relation (a) (1 a) = / sin a . Thus, taking the entire compact dimension, the Yukawa coupling is obtained as
Yfk,l
a ,fb ,fc
= gYM,4D

d/2

kj
1/2 (1 N
(2A,j )
j =1
eikj
sin
lj
N
eilj
sin
kj
N
) (1
(1
1/2
kj +lj
N
lj
N)
exp(Scl )
(105)
with the classical action given in (52). For the case kj + lj > N , we can obtain the corresponding
amplitude by replacing kj , lj by N kj , N lj . It is notable that, in the quantum amplitude,
there is no contribution from the geometry, since it is canceled by the same dependence in the
four-dimensional gauge coupling arising from the dimensional reduction.
We can obtain similar expressions for the higher order couplings, expressed in terms of multivalued hypergeometric functions. One can be convinced that the asymptotic form of generalized
hypergeometric function in the above limit is the ratios of gamma functions [35] with arguments
of O(0.1). Thus we expect that the quantum parts of the higher order couplings are roughly of
order one. Thus we see that the size of higher order coupling is dominated by the classical part.
6. Conclusions
We have calculated couplings of arbitrary order, among untwisted and twisted fields of heterotic string on orbifolds, using conformal field theory. In the low energy limit, they correspond
to the matter superpotentials. They are given by the zero external momentum limit of radially
ordered correlation functions. The specification of orbifold and the shift vector determines the
possible couplings. This provides us with lessons for constructing low-energy effective field theory, in particular for vacuum configurations of compactified string theory and realistic quark and
lepton masses.
The higher order couplings are complicated due to two things. The first is the technical difficulties dealing with arbitrary number of twist fields and even some excited twist fields from
the picture changing. In the calculation of higher order coupling, the latter has just an effect of
318
changing the normalization, not changing the transformation property and the branch cut structure. The other difficulty arises from linearly combined states, which appear in higher twisted
sectors of non-prime order orbifolds. They make the interpretation of the classical and the quantum somewhat tricky.
The selection rules (mainly studied in Ref. [27]) from the location of fields and the Rsymmetry can be possible origins of discrete quantum numbers in effective field theory. They
also provide the understanding on discrete flavor symmetries. Because of the restrictive form
of selection rules, in particular in the heterotic string models, only limited number of couplings
are possible, since there are limited number of orbifolds and fixed points. This will be classified
elsewhere [33].
The classical part is an instanton amplitude of world-sheet nature. Its size is exponentially
suppressed by the effective area swept by the twisted strings to form untwisted strings. The latter
is energetically not allowed, metastable intermediate state. This generalizes the naive area rule
even if the twisted strings do not form a polygon. Decomposing the locations of twisted fields, we
can obtain handy rule for calculating the size. For couplings involving linearly combined states,
the only contribution comes from the terms satisfying the space group selection rule. The others
vanish individually.
We have also calculated the quantum amplitude with the complete normalization, up to the
Khler normalization. In this there is no contribution from the geometric distribution of the fields,
since the contributions from the normalization and that from the dimensional reduction cancel.
Besides the 1/2 suppressions from the closed string couplings and oscillator normalizations,
the coefficient is given by ratios of products of gamma function, whose argument is of O(0.1),
thus we expect a factor of O(1) from the quantum amplitude. The dominant amplitude is classical one, which is exponentially suppressed. Thus it is easy to generate hierarchy of Yukawa
couplings. However, from the top-down approach, it is not easy to locate the desired fields at the
desired positions.
Acknowledgements
We are grateful to Steve Abel, Massimo Bianchi, Jihn E. Kim, Hans-Peter Nilles, Saul RamosSanchez, Michael Ratz, Robert Richter and Patrick K.S. Vaudrevange for discussion. K.S.C.
thanks to Yukawa Institute for hospitality, where part of the work is done. K.S.C. is supported in
part by the European Union 6th framework program MRTN-CT-2004-503069 Quest for unification, MRTN-CT-2004-005104 ForcesUniverse, MRTN-CT-2006-035863 UniverseNet and
SFB-Transregio 33 The Dark Univerese by Deutsche Forschungsgemeinschaft (DFG). T.K. is
supported in part by the Grand-in-Aid for Scientific Research #1754025 and the Grant-in-Aid for
the 21st Century COE The Center for Diversity and Universality in Physics from the Ministry
of Education, Culture, Sports, Science and Technology of Japan.
Appendix A. Useful formulae
The four point amplitude is described by the standard hypergeometric functions. The following relations
F (a, b; c; z) =
(c) (c a b)
F (a, b; a + b + 1 c; 1 z)
(c a) (c b)
(c) (a + b c)
+
(1 z)cab
(a) (b)
F (c a, c b; 1 + c a b; 1 z),
(A.1)
F (a, b; c; z) = (1 z)cab F (c a, c b; c; z),
319
(A.2)
are useful.
The classical action contains the following functions,

k4 k 1
k4
k2 k1 K4
k4 /N
F0 (1 x) = (1)
B
,
F
,1 , +
;x ,
N N
N
N N
N

k 1 k2
k1
k3 k1 k2
,
F
,1 ; + ;x ,
F1 (x) = (1)(k2 +k3 )/N x 1+(k1 +k2 )/N B
N N
N
N N
N

k 2 k3
,
F2 (1 x) = (1)1+k3 /N (1 x)1+(k2 +k3 )/N B
N N

k3
k1 k2 k3
F
,1 ; + ;1 x ,
N
N N
N

k3 k4
k2 k3 k4
k4
,
F
,1 , + ;x ,
F3 (x) = B
N N
N
N N
N
F1 (x)
= F1 (ki N ki ; x),
F2 (1 x)
= F2 (ki N ki ; 1 x),
(A.3)
where B(a, b) = (a) (b)/ (a + b) and it is the Euler

beta function. For F0 and F3 we have
used above relations (A.1), (A.2), as well as (24). For ki = 2N , the above functions become
as

sin(k3 /N) i(k3 +k4 )/N sin (k2 + k3 )/N ik2 /N
F0 (1 x) =
F1 (x) + F2 (1 x) ,
e
e
sin(k4 /N)
sin(k1 /N)

sin(k1 /N) i(k2 +k3 )/N
sin((k1 + k2 )/N) ik2 /N
F2 (1 x) ,
e
e
F3 (x) =
F1 (x) +
sin(k4 /N)
sin(k1 /N)
= (1)(k2 +k3 )/N x 1(k1 +k2 )/N (1 x)
1(k2 +k3 )/N
F1 (x)

k1
k2
k3 k4 1
B 1 ,1
F3 (x),
B
,
N
N
N N
= (1)1(k3 +k4 )/N (1 x)
1(k2 +k3 )/N x 1(k1 +k2 )/N
F2 (1 x)

k2
k3
k1 k4 1
B 1 ,1
F0 (1 x).
B
,
N
N
N N
(A.4)
For factorization of the amplitude, we need the asymptotic behaviors of hypergeometric functions in the z limit
(c) (b a) a
(c) (a b) b
z + eib
z ,
(b) (c a)
(a) (c b)
(c) (b a) a (c) (a b) b
F (a, b; c; 1 z)
z +
z .
(b) (c a)
(a) (c b)
F (a, b; c; z) eia
(A.5)
(A.6)
For the higher order amplitude than four, we need a multivariable generalization of hypergeometric function, called Lauricella D function [36]. It is defined as
320
FD (a, b1 , . . . , br ; c; x1 , . . . , xr )

(a)m1 ++mr (b1 )m1 (br )mr m1
x1 xrmr ,
(c)m1 ++mr m1 ! mr !
m1 =0
(A.7)
mr =0
with |xi | < 1 for all i. Here (a)n is the Pochhammer symbol meaning
(a + n)
= a(a + 1) (a + n 1),
(a)
(1)n
, n < 0.
(a)n =
(1 a)n
(a)n =
n 0,
(A.8)
We can express the integration as [17]

Fi1 = eiki /N (xi xi+1 )1+(ki +ki+1 )/N
N1

(xi xj )1+kj /N B(ki /N, ki+1 /N)
j =1,j
=i,i+1
(N3)
(ki /N, 1 k1 /N, . . . , 1 kL1 /N ; ki /N + kki +1 /N ; xi,1 , . . . , xi,L1 ),
FD
1
i(kN /N +1)
B(kN /N, k1 /N)
F0 = e
(N3)
FD
(kN /N, 1 k2 /N, . . . , 1 kN 2 /N ; k1 /N
+ kN /N ; 1 x2 , . . . , 1 xN2 ),
where xij =
xi xi+1
xi xj
(A.9)
is the conformally invariant cross ratio.
References
[1] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[2] L.E. Ibez, H.-P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25;
L.E. Ibez, J.E. Kim, H.-P. Nilles, F. Quevedo, Phys. Lett. B 191 (1987) 282;
L.E. Ibez, J. Mas, H.P. Nilles, F. Quevedo, Nucl. Phys. B 301 (1988) 157;
A. Font, L.E. Ibez, F. Quevedo, A. Sierra, Nucl. Phys. B 331 (1990) 421;
A. Font, L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 210 (1988) 101;
A. Font, L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 213 (1988) 564, Erratum;
A. Font, L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 213 (1988) 274;
J.A. Casas, C. Munoz, Phys. Lett. B 209 (1988) 214;
J.A. Casas, C. Munoz, Phys. Lett. B 214 (1988) 63;
D. Bailin, A. Love, S. Thomas, Phys. Lett. B 194 (1987) 385;
Y. Katsuki, Y. Kawamura, T. Kobayashi, N. Ohtsubo, Y. Ono, K. Tanioka, Nucl. Phys. B 341 (1990) 611.
[3] T. Kobayashi, S. Raby, R.J. Zhang, Phys. Lett. B 593 (2004) 262, hep-ph/0403065;
T. Kobayashi, S. Raby, R.J. Zhang, Nucl. Phys. B 704 (2005) 3, hep-ph/0409098.
[4] S. Forste, H.P. Nilles, P.K.S. Vaudrevange, A. Wingerter, Phys. Rev. D 70 (2004) 106008;
W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, Nucl. Phys. B 712 (2005) 139;
W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, Phys. Rev. Lett. 96 (2006) 121602, hep-th/0606187;
S. Forste, H.P. Nilles, A. Wingerter, Phys. Rev. D 72 (2005) 026001;
S. Forste, H.P. Nilles, A. Wingerter, Phys. Rev. D 73 (2006) 066011;
K.S. Choi, S. Groot Nibbelink, M. Trapletti, JHEP 0412 (2004) 063;
H.P. Nilles, S. Ramos-Sanchez, P.K.S. Vaudrevange, A. Wingerter, JHEP 0604 (2006) 050;
J.E. Kim, B. Kyae, hep-th/0608085;
J.E. Kim, B. Kyae, Nucl. Phys. B 770 (2007) 47, hep-th/0608086;
O. Lebedev, H.P. Nilles, S. Raby, S. Ramos-Sanchez, M. Ratz, P.K.S. Vaudrevange, A. Wingerter, Phys. Lett. B 645
(2007) 88, hep-th/0611095.
321
[5] K.S. Choi, J.E. Kim, Quarks and Leptons from Orbifolded Superstring, Lecture Notes in Physics, vol. 696, Springer,
2006.
[6] S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465.
[7] L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 282 (1987) 13.
[8] J.J. Atick, L.J. Dixon, P.A. Griffin, D. Nemeschansky, Nucl. Phys. B 298 (1988) 1.
[9] T.T. Burwick, R.K. Kaiser, H.F. Mller, Nucl. Phys. B 355 (1991) 689;
J. Erler, D. Jungnickel, M. Spalinski, S. Stieberger, Nucl. Phys. B 397 (1993) 379.
[10] T. Kobayashi, O. Lebedev, Phys. Lett. B 566 (2003) 164;
T. Kobayashi, O. Lebedev, Phys. Lett. B 565 (2003) 193.
[11] T. Kobayashi, N. Ohtsubo, Int. J. Mod. Phys. A 9 (1994) 87.
[12] S. Forste, T. Kobayashi, H. Ohki, K.j. Takahashi, JHEP 0703 (2007) 011, hep-th/0612044.
[13] P. Ko, T. Kobayashi, J.h. Park, Phys. Lett. B 598 (2004) 263;
P. Ko, T. Kobayashi, J.h. Park, Phys. Rev. D 71 (2005) 095010, hep-ph/0503029.
[14] T. Kobayashi, H.P. Nilles, F. Ploger, S. Raby, M. Ratz, Nucl. Phys. B 768 (2007) 135, hep-ph/0611020.
[15] P. Ko, T. Kobayashi, J.h. Park, S. Raby, arXiv: 0704.2807 [hep-ph].
[16] K.S. Choi, I.W. Kim, J.E. Kim, JHEP 0703 (2007) 116, hep-ph/0612107.
[17] S.A. Abel, A.W. Owen, Nucl. Phys. B 682 (2004) 183.
[18] D. Cremades, L.E. Ibez, F. Marchesano, JHEP 0307 (2003) 038.
[19] M. Cvetic, I. Papadimitriou, Phys. Rev. D 68 (2003) 046001.
[20] S.A. Abel, A.W. Owen, Nucl. Phys. B 663 (2003) 197.
[21] I.R. Klebanov, E. Witten, Nucl. Phys. B 664 (2003) 3, hep-th/0304079.
[22] T. Kobayashi, N. Ohtsubo, Phys. Lett. B 245 (1990) 441.
[23] J. Polchinski, String Theory, vols. 1, 2, Cambridge Univ. Press, 1998.
[24] P. Goddard, D.I. Olive, Int. J. Mod. Phys. A 1 (1986) 303;
K.S. Choi, Nucl. Phys. B 708 (2005) 194, hep-th/0405195.
[25] D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 271 (1986) 93.
[26] O. Lebedev, H.P. Nilles, S. Raby, S. Ramos-Sanchez, M. Ratz, P.K.S. Vaudrevange, A. Wingerter, arXiv: 0708.2691
[hep-th].
[27] K.-S. Choi, T. Kobayashi, H.-P. Nilles, S. Ramos-Sanchez, P.K.S. Vaudrevange, in preparation.
[28] T. Araki, K.S. Choi, T. Kobayashi, J. Kubo, H. Ohki, arXiv: 0705.3075 [hep-ph].
[29] E.M. Opdam, Multivariable hypergeometric functions, in: Proceedings of the European Congress of Mathematics
2000, in: Progress in Mathematics, vol. I, Birkhuser, 2001.
[30] L. Pochhammer, Math. Ann. 35 (1890) 495.
[31] H. Kawai, D.C. Lewellen, S.H.H. Tye, Nucl. Phys. B 269 (1986) 1.
[32] E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, fourth ed., Cambridge Univ. Press, 1932.
[33] K.S. Choi, T. Kobayashi, in preparation.
[34] M.A. Virasoro, Phys. Rev. 177 (1969) 2309;
J.A. Shapiro, Phys. Rev. 179 (1969) 1345;
G. Veneziano, Nucl. Phys. B 74 (1974) 365.
[35] See, for example C. Ferreira, J. Lopez, J. Comput. Appl. Math. 151 (2003) 235.
[36] H. Exton, Multiple Hypergeometric Functions and Applications, Ellis Horwood, 1976.

Precision predictions for Z -production

at the CERN LHC: QCD matrix elements,
parton showers, and joint resummation
Benjamin Fuks a,b , Michael Klasen b, , Fabienne Ledroit b ,
Qiang Li b,c , Julien Morel b
a Physikalisches Institut, Albert-Ludwigs-Universitt Freiburg, Hermann-Herder-Strae 3,
D-79106 Freiburg i.Br., Germany

b Laboratoire de Physique Subatomique et de Cosmologie, Universit Joseph Fourier/CNRS-IN2P3/INPG,
53 Avenue des Martyrs, F-38026 Grenoble, France

c Institut fr Theoretische Physik, Universitt Karlsruhe, Postfach 6980, D-76128 Karlsruhe, Germany
Received 5 November 2007; received in revised form 8 January 2008; accepted 14 January 2008
Abstract
We improve the theoretical predictions for the production of extra neutral gauge bosons at hadron colliders by implementing the Z -bosons in the MC@NLO generator and by computing their differential and
total cross sections in joint pT and threshold resummation. The two improved predictions are found to be
in excellent agreement with each other for mass spectra, pT spectra, and total cross sections, while the
PYTHIA parton and power shower predictions usually employed for experimental analyses show significant shortcomings both in normalization and shape. The theoretical uncertainties from scale and parton
density variations and non-perturbative effects are found to be +6/3%, +/5%, and less than +3/6%,
respectively, and thus under good control. The implementation of our improved predictions in terms of the
new MC@NLO generator or resummed K factors in the analysis chains of the Tevatron and LHC experiments should be straightforward and lead to more precise determinations or limits of the Z -boson masses
and/or couplings.
PACS: 12.10.Dm; 12.38.Cy; 13.85.Qk
Keywords: Grand unified theories; QCD resummation; DrellYan type processes
E-mail address: klasen@lpsc.in2p3.fr (M. Klasen).

doi:10.1016/j.nuclphysb.2008.01.017
B. Fuks et al. / Nuclear Physics B 797 (2008) 322339
323
1. Introduction
Despite its impressive phenomenological success, the Standard Model (SM) of particle
physics is widely believed to suffer from a variety of conceptual deficiencies. In particular, it
provides no fundamental motivation why the strong and electroweak interactions should be described by three different gauge groups, i.e. SU(3), SU(2), and U (1). Grand Unified Theories
(GUTs) allow for a unification of these groups within a simple Lie group such as SU(5), SO(10),
or E6 . Depending on the rank of the unifying group, one or several extra neutral gauge (Z )bosons appear when the unification group is broken to the SM at higher scales, exhibiting the
existence of additional U (1) symmetries [1]. Anomaly cancellations and gauge invariance of
quark and lepton Yukawa couplings impose a number of restrictions on these additional symmetries. Viable families of models, that are consistent with constraints coming from the CERN LEP
collider, include those based on the BL and SO(10) symmetries and, provided that fermion
mass generation is not restricted to the SM Higgs mechanism and new charged fermions are
allowed, also those inspired by E6 [2].
If the extra Z -bosons couple to quarks and leptons with approximately SM strength and if
their mass is not too large, they will be produced at current and future hadron colliders and can
be easily detected through their leptonic decay channels. The search for these particles occupies therefore an important place in the experimental programs of the Fermilab Tevatron and the
CERN LHC. For example, the CDF Collaboration has searched the Tevatron Run II data for Z bosons in the e+ e decay channel, using the di-electron invariant mass and angular distributions
and setting lower mass limits of 650 to 900 GeV for a large variety of models [3]. Similar constraints come from electroweak precision fits and di-fermion production at LEP2 [4]. Within the
ATLAS Collaboration, the discovery reach in Z e+ e decays has recently been analyzed [5]
for the four classes of models defined in Ref. [2]. The CMS Collaboration has claimed a discovery reach of masses between 3.4 and 4.3 TeV for the Z + decay channel and an
integrated luminosity of 100 fb1 [6].
The currently available simulations for the LHC experiments rely completely on the PYTHIA
Monte Carlo (MC) generator [7], which is based on leading order (LO) QCD matrix elements, parton showers, and the Lund string hadronization model. It includes the full interference
structure of new Z -bosons with DrellYan photon and SM Z-boson exchange, and the abovementioned phenomenologically viable models can easily be implemented [8]. The description
of the transverse-momentum (pT ) spectrum of the produced vector-boson can be improved by
matching the parton shower to the hard emission of an extra parton [9], but the overall normalization of the theoretical cross section remains subject to large higher-order corrections and scale
uncertainties. The CDF Collaboration has therefore chosen to renormalize the generated Monte
Carlo events with a correction (K) factor in each invariant mass bin to the next-to-next-to-leading
order (NNLO) cross section [10]. However, this procedure does not lead to a correct description
of the transverse-momentum spectrum. Note also that in principle the LO cross section no longer
factorizes at NNLO, i.e. the K-factors for DrellYan and Z production need no longer to be
equal [2].
Here, we report on the implementation of extra Z -bosons in the next-to-leading order (NLO)
Monte Carlo generator MC@NLO [11], allowing to match the complete NLO matrix elements
with the parton shower and cluster hadronization model of the Monte Carlo generator HERWIG [12]. Since the LO cross section still factorizes completely at NLO, this requires the
implementation of the Z -boson mass, decay width, propagator, and couplings in MC@NLO,
together with the full interference with DrellYan photon and SM Z-boson exchanges. As was
324
the case for PYTHIA, the emission of one additional hard parton has previously been matched to
the HERWIG parton shower, albeit for SM vector-boson production only [13]. As an alternative,
the CKKW formalism [14] for the matching of hard real emissions has been implemented in both
PYTHIA and HERWIG for SM vector-boson production with the result that the three different
matching formalisms were found to vary systematically in a significant way and to depend in
addition strongly on the matching scale [15]. Similar results were obtained more recently in [16]
for W -boson plus jet production, where in addition various implementations of the MLM [17]
prescription for matching multiparton final states to parton showers have been compared to
the methods discussed above. Here, we perform a similar study of systematic uncertainties by
comparing the invariant-mass and transverse-momentum distributions as predicted with our implementation of Z -bosons in MC@NLO with the PYTHIA parton-shower and matrix-element
corrections. We also confront these MC predictions with a new computation of Z -boson production in the framework of joint resummation [18]. In addition, we compare the dependence of
the various predictions on the unphysical renormalization and factorization scales as well as on
the employed parton densities. The impact of hadronization corrections, dominant electroweak
corrections, and non-perturbative effects is also studied.
The remainder of this work is organized as follows: in Section 2 we first describe the PYTHIA
framework for Z -boson production and the associated matching of parton showers with the
emission of an additional hard parton. We then discuss our implementation of Z -boson production in the MC@NLO generator and recall its matching procedure of NLO matrix elements
with the HERWIG mechanism of parton showers. We also present briefly our calculation of Z boson production using joint resummation. In Section 3, we define our choice of electroweak SM
parameters and define the parameters of our exemplary Z model. We also discuss the various
corrections that we apply to the production cross section, i.e. dominant electroweak corrections,
next-to-leading order QCD matrix elements, parton showers, and hadronization. We then compare the numerical impact of these corrections and study the remaining theoretical uncertainties,
coming from the choice of renormalization and factorization scales, the parameterization of parton densities, and non-perturbative effects. Our conclusions are given in Section 4.
2. Theoretical setup
In 1984, Green and Schwarz showed that ten-dimensional string theories with E8 E8 or
SO(32) gauge symmetry are anomaly-free and thus potentially finite [19]. Among these two
gauge groups, only the former contains chiral fermions as they exist in the SM. After compactification, it leads to the E6 symmetry as an effective GUT group, that can be broken further
to [20]
E6 SO(10) U (1) SU(5) U (1) U (1) .
(1)
While the Z -bosons corresponding to the additional U (1) and U (1) symmetries can in general mix with each other,
Z ( ) = Z cos + Z sin
and Z ( ) = Z sin Z cos ,
(2)
90
in the
we consider in this work only a TeV-scale Z -boson as an exemplary case, i.e. =
convention of Ref. [21], and assume the Z Z to acquire its mass at considerably higher
scales, as it is naturally the case in the hierarchy of symmetry breaking of Eq. (1). We will
furthermore assume that SO(10) breaks down to SU(5) U (1) at the same scale where SU(5)
breaks down to the SM group SU(3)C SU(2)L U (1)Y with gauge couplings gs , g, and g .
325
The U (1) coupling

5
5
5
e

g =
g tan W =
g =
3
3
3 1 sin2 W
(3)

is then directly related
to the coupling g of the weak hypercharge U (1)Y by the usual grouptheoretical factor 5/3. Note that as the SO(10)-breaking scale increases
from the SU(5)
breaking scale to the E6 or Planck scale, g starts to deviate from 5/3g to roughly 2/3
times this value [22]. Using Eq. (3), we can thus express g in terms of a group-theoretical
factor, the SU(2)L gauge coupling g or alternatively the electromagnetic fine structure constant
= e2 /(4), and the squared sine of the angle W , describing the mixing of the neutral W 0 and B-bosons to the massless photon and the SM Z-boson of mass mZ after the breaking of
SU(2)L U (1)Y to U (1)em. . While the photon is protected from further mixing by the exact
electromagnetic symmetry, the SM Z-boson can in general mix with the additional Z -boson
with an angle . Its squared tangent
tan2 =
m2Z m21
(4)
m22 m2Z
depends then on the eigenvalues of the Z Z mass matrix m1,2 , which in turn depend on
the vacuum expectation values v10 and vSM of the SO(10)- and SM-breaking Higgs fields.
This mixing can in general induce a coupling of the Z -boson to the SM W -bosons, even
though they belong to different gauge groups. However, the ratio v10 /vSM is usually large and
m1 mZ m2 mZ , so that Z Z mixing will be neglected in the following. The DELPHI
Collaboration has constrained to be smaller than a few mrad (1.7 mrad for mZ = 440 GeV in
the -model) [23].
2.1. Z production in PYTHIA
In PYTHIA 6.403 [7], the production of extra neutral gauge bosons in hadron collisions has
been implemented including the full interference structure with SM photon and Z-boson exchanges. The Lagrangian describing the interaction of the extra neutral gauge boson Z with
fermions f ,
L=
g
f (vf af 5 )f Z
4 cos W
(5)
has been expressed in terms of generalized vector (vf ) and axial-vector couplings (af ), which
depend in general on the unifying gauge group and the Higgs representations employed to break
this group down to the SM gauge group. For the additional U (1) symmetry, that we use as our
standard example, these couplings are given in Table 1 for the down- and up-type quarks as well
as for the charged leptons and neutrinos. We will be interested in the DrellYan like production
Table 1
Vector and axial-vector couplings of down- and up-type quarks as well as charged leptons and neutrinos in the E6 inspired Z model in the convention of PYTHIA. sW is the sine of the electroweak mixing angle W
vd
2 6sW /3
ad
6sW /3
vu
0
au
6sW /3
vl
2 6sW /3
al
6sW /3
6sW /2
6sW /2
326
of electronpositron pairs
q q ( , Z, Z ) e e+ ,
(6)
i.e. the relevant couplings are primarily those of the five light quarkflavours q = u, d, s, c, b (with
masses mq much smaller than the partonic centre-of-mass energy s ) and positrons/electrons e
and to a lesser extent those of the other fermions, which contribute to the total decay width
appearing in the Z propagator.
2.2. Matching of parton showers with LO matrix elements in PYTHIA
The only part of the DrellYan like process in Eq. (6) that is sensitive to QCD corrections
is the quarkantiquark (and beyond the LO the quarkgluon) initial state. It can give rise to
an initial-state parton shower, that is modeled in PYTHIA by starting with the hard scattering
partons and then successively reconstructing the preceding branchings in a falling sequence of
space-like virtualities Q2 . They range from a maximal value Qmax = mZ () , that is of the order
of the hard scattering scale, to a cut-off scale Q0 = 1 GeV, that is close to a typical hadron mass.
not uniquely defined. It can in particular be increased to the
The scale Qmax is, however,
hadronic centre-of-mass energy S, so that the parton shower (PS) populates the full phase
space. However, it must then be matched to the QCD matrix element (ME) describing the emission of one extra hard parton
q q ( , Z, Z )g
(7)
at O(s ). The emission rate for the final (and normally hardest) q qg emission must therefore be corrected by a factor [9]

2
2
2
(d/dt)ME
i= ,Z,Z [t + u + 2mi s] + interference terms
=
,
Rq q(
,Z,Z )g (s, t) =
4
2
(d/dt)PS
i= ,Z,Z [s + mi ] + interference terms
(8)
which always lies between one-half and one. Here, s, t and u refer to the usual Mandelstam
variables of the process in Eq. (7), and m = 0. At the same order, the crossed process
qg ( , Z, Z )q
can occur, in which case the correction factor is

2
2
2
(d/dt)ME
i= ,Z,Z [s + u + 2mi t] + interference terms
RqgV q (s, t) =
=
.
2 2
4
(d/dt)PS
i= ,Z,Z [(s mi ) + mi ] + interference terms
(9)
(10)
Since this factor always lies between one and three, the g q q splitting must be preweighted
by a factor three in order to correctly reproduce the s-channel graph qg q ( , Z, Z )q.
Besides the dominant QCD radiation, we will also briefly investigate the effect of the relatively
suppressed QED radiation (e.g. q q ) as well as the corrections induced by the hadronization
of the additional partons.
In addition to the transverse momentum pT generated by hard emission and/or the initial-state
parton shower, an average intrinsic transverse-momentum
kT can be assigned to the shower
initiator in order to take into account the transverse motion of quarks inside the original hadron.
As the shower does not evolve below Q0 = 1 GeV, this same value is retained in PYTHIA
as the default value for
kT . However, just like Qmax , Q0
kT is not uniquely defined and
327
furthermore closely related to the non-perturbative regime of QCD. We therefore set

kT = 0 in
the following.
PYTHIA allows in principle also for the participation of multiple parton pairs in hadronic
collisions. We do, however, not make use of this possibility, as it has little numerical effect and
its description in perturbative QCD remains controversial.
2.3. Implementation of Z -bosons in MC@NLO
In MC@NLO [11], the implementation of SM Z-boson interactions with fermions f is based
on the Lagrangian [24]
g
(11)
f (af + bf 5 )f Z .
cos W
For photon interactions, af = ef sin W cos W , where ef is the fractional fermion charge, and
bf = 0. The vector coupling af and the axial vector coupling bf are related to those defined
in PYTHIA (see Eq. (5)) by af vf /4 and bf af /4. They are combined to form the
coefficients Af = af2 + bf2 and Bf = 2af bf , that appear in the squared matrix elements

|Mi |2 q q or qg , Z e e+ + X
2
eq
1
1
1
A ,B
1,0
Ti |1,0
+ 4
Ti | l l
= e 4 Ci
4
2
2
2
4
4
M
sin W cos W (M mZ ) + (Z mZ )2 Aq ,Bq

M 2 m2Z
2eq
1
al ,bl
2 2
Ti |
(12)
,
M sin W cos2 W (M 2 m2Z )2 + (Z mZ )2 aq ,bq
that have been averaged/summed over initial/final spins and colours. For the LO DrellYan (DY)
process in Eq. (6), the colour factor is CDY = NC /NC2 = 1/3 and
2

2

A ,B
TDY |Alq ,Blq = 8 Al Aq tDY
(13)
+ u2DY Bl Bq tDY
u2DY ,
where M 2 = sDY is the invariant mass of the lepton pair and sDY , tDY , and uDY refer to the
Mandelstam variables of the DY-process in Eq. (6). The NLO QCD-corrections to the DY-process
have been implemented in MC@NLO using the same convention [24].
One major new and technical aspect of our work is the implementation of Z -boson interactions and exchanges in the framework described above. To this end, we have defined the mass
and width of the Z -boson as well as its couplings in the convention of PYTHIA. The squared
matrix element has also been modified,

|Mi |2 q q or qg , Z, Z e e+ + X
2
eq
1
1
1 4
Ti |1,0 +
Ti |Al ,Bl
= e Ci
4
M 4 1,0 sin4 W cos4 W (M 2 m2Z )2 + (Z mZ )2 Aq ,Bq
1
1
Al ,Bl
+ 4
T
|

i
sin W cos4 W (M 2 m2Z )2 + (Z mZ )2 Aq ,Bq
M 2 m2Z
2eq
1
Ti |al ,bl
M 2 sin2 W cos2 W (M 2 m2Z )2 + (Z mZ )2 aq ,bq
M 2 m2Z
2eq
1
al ,bl
T
|

i
M 2 sin2 W cos2 W (M 2 m2Z )2 + (Z mZ )2 aq ,bq
328
(M 2 m2Z )(M 2 m2Z ) + Z mZ Z mZ
+2
sin4 W cos4 W [(M 2 m2Z )2 + (Z mZ )2 ] [(M 2 m2Z )2 + (Z mZ )2 ]

al a +bl b ,al b +al bl
Ti |aq al +bq lb ,aql b +a
(14)
.
b
q
q
It includes now the squared Z -boson exchange as well as its interferences with the photon and
SM Z-boson exchanges. Note that it is not sufficient to change only the Z-boson mass in the
existing MC@NLO implementation, but that it is also necessary to change the Z-boson width as
well as the couplings. As a consequence, all observables change: the mass spectrum due to the
modified width, the forward-backward asymmetries due to the modified couplings, etc. Furthermore, even if one changes all of these parameters, one still has Z interference, but no ZZ
interference. We have checked numerically that these modifications induce large differences already in the total cross section.
2.4. Matching of parton showers with NLO matrix elements in MC@NLO
In HERWIG [12], parton showers are generated by a coherent branching algorithm with parton splittings i j k, whose energy fractions zj = Ej /Ei are distributed according to the LO
DGLAP splitting functions. Phase space is restricted to an angular ordered region, which automatically takes infrared singularities into account, i.e. at each branching, the angle between
the two emitted partons is smaller than that of the previous branching. The emission angles
j k = (pj . pk )/(Ej Ek ) 12 j2k are distributed according to Sudakov form factors, that sum virtual corrections and unresolved real emissions and normalize the branching distributions to give
the probabilistic interpretation needed for a MC simulation. For initial-state radiation, the parton
shower follows, of course, a backward evolution. It is terminated when j k < Q20 /Ei2 , where
the space-like cut-off scale Q0 is set to 2.5 GeV by default. Below this scale, a non-perturbative
stage is imposed. In particular, a splitting of non-valence partons is enforced to allow for a smooth
transition to the valence partons inside the outer hadron. Although HERWIG also allows for the
simulation of a soft underlying event, we have not made use of this possibility. Since the parton
shower is supposed to describe only the soft/collinear region, any initial-state emission outside
this region, i.e. where > z2 , is suppressed.
As was the case for PYTHIA, hard matrix element corrections for SM vector boson production
have been implemented in HERWIG [13], where radiation in the region > z2 can now be
allowed. It is then distributed according to the matrix element describing the emission of an
additional parton. In the soft region < z2 already populated by the parton shower, the emission
of the hardest (largest pT ) parton generated so far is reweighted in order to avoid double counting.
Note that this need not be the first emission, since angular ordering does not necessarily imply
ordering in transverse momentum. Note also that the normalization of the total cross section is
still accurate to LO only.
An NLO accuracy of the total cross section has been achieved in MC@NLO [11] through
an implementation of NLO cross sections, matched to the HERWIG parton shower. Instead of
the LO matrix element implemented in the standard version of HERWIG, two separate samples
of Born-like or standard (S) and hard emission (H ) events are generated, that can have weight
(S,H )
= 1 and are statistically distributed according to the positive-definite standard and hard
wi
contributions to the NLO cross section JS,H . These are made separately finite by adding and
subtracting the NLO part of the expanded Sudakov form factor and are explicitly defined in
tot (S,H )
(JH + JS )/Ntot .
Ref. [11]. The total cross section is then given by tot = N
i=1 wi
329
2.5. Joint transverse-momentum and threshold resummation for Z -bosons

The LO matrix element predictions for Z -production are affected by fixed order (F.O.) QCD
corrections, that are logarithmically enhanced, when the Z -boson is produced close to the partonic threshold, i.e. z = M 2 /s 1, or when its transverse momentum is small, i.e. pT 0.
These corrections must then be resummed (res) to all orders, which is most easily achieved in
Mellin (N ) and impact parameter (b) space, and matched to the F.O. prediction by subtracting
from their sum the perturbatively expanded (exp) resummed prediction, i.e.

dN N
b db
d2 (F.O.)
d2
=
+
J0 (bpT )
2
2
2
2
2i
2
dM dpT
dM dpT
C
0
2 (res)

d
d2 (exp)
(N, b)
(N, b) .
dM 2 dpT2
dM 2 dpT2
(15)
In this way, a uniform precision is obtained, and the large theoretical (renormalization and factorization scale) uncertainties are reduced. The NLO cross section for DrellYan processes (with
= M 2 /S)
1
1
d2 (F.O.)
=
dxa
dxb fa/ ha (xa ; F )fb/ hb (xb ; F )
dM 2 dpT2
a,b
/xa

2

s (R ) (1)
(0)
pT (1 z) ab +
ab (z) + O s2
(16)
is well known [24], and the necessary modifications for implementing Z -bosons in the O(si )
partonic cross sections
dt
1
|Mi |2
(17)
2s
8s
have already been discussed above.
Since the pT - [25] and threshold-enhanced contributions [26] are both due to soft-gluon emission in the initial state, they may be resummed at the same time [18,27,28]. The logarithms are
then organized by the function
(i)
ab =
N)
= b +
(b,
N
N
1 + b/
with b bMeE /2 and N N eE ,
(18)
whose form with = 1/4 is constrained by the requirement that the leading and next-to-leading
logarithms in b and N are correctly reproduced in the limits b and N , respectively.
The choice of Eq. (18) with = 1/4 avoids the introduction of sizeable subleading terms into
perturbative expansions of the resummed cross section at a given order in s , which are not
present in fixed-order calculations [29]. The resummed cross section

d2 (res)
(N, b) =
fa/ ha (N + 1; F )fb/ hb (N + 1; F ) c(0)
c exp Gc (N, b; s , R )
2
2
dM dpT
a,b,c

s (R ) n (n)
ca cb
(19)
Habcc (N; R , F )
+
n=1
330
can then be factorized into a regular part with

(1)
(1)
(1)
(1)
Habcc (N ; R , F ) = ca cb
Hcc (R ) + ca Cc/b
Cc/a (N )
(N ) + cb
M2

(1)
(1)
+ ca c/b
c/a (N ) ln 2 ,
(N ) + cb
F
(20)
where in the DrellYan resummation scheme

(1)
Hcc (R ) 0,
(1)
Cq/g (N) =
(1)
Cq/q (N ) =
2
2 8
+
,
3N (N + 1)
3
and
1
,
2(N + 1)(N + 2)
(21)
and a perturbatively calculable eikonal factor

Gc (N, b; s , R ) = gc(1) () ln + gc(2) (; R ),
(22)
which depends through the functions

(1)
Ac 2 + ln(1 2)
and
(23)
0

A(1)
2 + ln(1 2)
c 1 1 2
gc(2) (; R ) =
ln
(1
2)
+
2
1 2
03
(1)

(2)
Ac
2
Bc(1) (N )
M 2 Ac
+
+ ln(1 2) +
ln 2 2
ln(1 2)
0
1 2
0
R
0
(24)
gc(1) () =
(1)
on the logarithm = 0 /s (R ) ln . The anomalous dimensions c/a (N ) are the N -moments

of the O(s ) AltarelliParisi splitting functions. Up to next-to-leading logarithmic order, the
(1,2)
coefficients needed in gc
are

67 2
5
(1)
(2)
TR NF , and
Aq = CF , Aq = CF CA
36 12
9
3
(1)
Bq(1) (N ) = CF + 2q/q (N ).
(25)
2
The usual coefficients of the QCD -function are

1
1
(26)
(11CA 4TR Nf ) and 1 =
17CA2 10TR CA Nf 6CF TR Nf ,
12
24
the number of effectively massless quark flavours is Nf , and CF = 4/3, CA = 3, and TR = 1/2
are the usual QCD colour factors. Re-expanding the resummed cross section leads to
0 =
d2 (exp)
(N, b)
dM 2 dpT2

=
fa/ ha (N + 1; F )fb/ hb (N + 1; F ) c(0)
c
a,b,c

s (R ) n (n)
(n)
abcc (N, b) + Habcc (N ; R , F )
ca cb
+
n=1
(27)
331
with the coefficient

(1)
(1)

(1)
(1)
(1)
2
abc
ln + Bc ca cb
+ ca c/b
c/a (N ) ln ,
c (N, b) = 2 Ac ca cb
(N ) + cb
(28)
which matches precisely with the NLO cross section.
3. Numerical comparison
In this section, we compare the three different theoretical approaches to Z -boson production
at the LHC discussed above, i.e. the matching of LO matrix elements with parton showers as
implemented in PYTHIA, the matching of NLO matrix elements with parton showers in our
modified version of MC@NLO, and the matching of NLO matrix elements with our improved
formalism of joint resummation at next-to-leading logarithmic (NLL) order. For a comparison
of joint resummation with transverse-momentum and threshold resummation in DrellYan like
processes we refer the reader to Ref. [18].
We first fix our choice of parton densities and of the strong, electroweak, and extra gaugeboson parameters and demonstrate that the three theoretical predictions coincide at the LO
partonic level for both the mass- and the pT -spectrum. Next, we show the impact of the dominant
electroweak corrections by running the fixed electromagnetic coupling from zero momentum
transfer to the mass of the Z -boson. We then analyze separately the different levels of improvement made possible in the three theoretical approaches, before comparing the three best
versions directly with each other. Finally, we discuss the impact of the remaining theoretical
uncertainties, coming from variations of the renormalization and factorization scales, the parton
densities and non-perturbative contributions, on the theoretical predictions.
3.1. Choice of strong, electroweak, and extra gauge-boson parameters
We will make predictions for pp collisions at the LHC at a centre-of-mass energy of S =

14 TeV, choosing the CTEQ6L (LO) and CTEQ6M (NLO MS) [30] sets as our default for
the parton densities in the protons for LO and NLO/NLL calculations, respectively. The strong
n =5
coupling constant s (R ) is always computed with two-loop accuracy, f = 226 MeV, and
MS
identifying the renormalization scale R (as well as the factorization scale F ) with the invariant
mass of the lepton pair M. As is customary, the theoretical uncertainty is estimated by varying
the scales by a factor of two around the central value.
For the electroweak mass, width, and coupling parameters, we use the values of the 2002 Review of the Particle Data Group [31], i.e. mZ = 91.188 GeV, Z = 2.4952 GeV, = 1/137.04,
and sin2 W = 0.23113, which are (still) used as default in the Z analysis of the ATLAS Collaboration. The only value that has changed in 2006 is sin2 W = 0.23122 [32], but the numerical
impact of this change remains visibly small. Using these parameters, a Z -boson mass of 1 TeV,
running the fine-structure constant to (1 TeV) = 1/124.43, and including the NLO QCD correction factor 1 + s (R )/ for Z -decays into quarks, we compute the total width of the Z -boson
in the -model within PYTHIA (see Section 2.1) to be = 12.04 GeV. Its branching ratio
to electronpositron pairs, representing the signal, is found to be 5.98%. These values are then
passed as parameters to our modified MC@NLO and new resummation programs.
In order to set a common theoretical basis, we show in Fig. 1 the invariant mass-spectrum of
the electronpositron pair in LO QCD, i.e. without any parton shower, matrix element correction,
332
Fig. 1. The LO mass-spectrum for Z production with fixed in PYTHIA (triangles), MC@NLO (stars) and at parton
level (full line), compared to the SM DrellYan background in MC@NLO (crosses).
resummation, or hadronization effects. The PYTHIA (triangles), MC@NLO (stars), and parton
level (full line) mass spectra, shown in the mass range of 900 to 1200 GeV around the mass peak
of the Z -boson at 1 TeV, coincide perfectly. For comparison, we also show the differential cross
section for the SM DrellYan process in MC@NLO (crosses), which coincides with the corresponding LO and PYTHIA predictions and represents the dominant (irreducible) background to
the Z -boson signal. Far from the resonance region, the two mass spectra would, of course, coincide. The transverse momentum of the lepton pair (pT ) is exactly zero in PYTHIA and at parton
level, while the forced splitting of non-valence partons in MC@NLO induces a distribution that
extends to non-zero values of pT even when the parton shower is switched off. Since this distribution is unphysical, we do not show it here. When integrated over pT , the MC@NLO total cross
section coincides, however, with PYTHIAs and the one at parton level.
The dominant electroweak corrections can be resummed by running the fixed value of the
fine-structure constant = 1/137.04 in Fig. 1 to the value = 1/124.43 at the mass of the Z boson in the MS-scheme. The ratio 2 (M)/ 2 (0) then induces an increase of about 22% in the
cross sections shown in Fig. 1. In the following, we will always use a running value of .
3.2. Numerical results with PYTHIA
In Fig. 2, we show the three different ways of improving on the parton-level predictions
that are implemented in PYTHIA, where one can either add QCD parton showers in the soft
and collinear regions (circles), corresponding to the leading-logarithmic (LL) approximation,
or in the full phase space (triangles), which however overestimates the cross section, so that it
must be renormalized to the matrix element describing the emission of an additional hard parton
(squares). By definition, neither the total cross section nor the mass spectrum are changed and
remain accurate to LO only, as can be seen from the fact that all histograms on the left-hand
333
Fig. 2. Mass (left) and transverse-momentum spectra (right) with PYTHIA with soft/collinear QCD parton showers
(circles), QCD parton showers populating the full phase space (triangles), and after adding LO matrix element corrections
(squares). The mass spectra have been normalized to the LO QCD prediction.
side of Fig. 2 coincide, after normalization to the LO prediction and within the statistical error
bars, with unity. The right-hand side of Fig. 2 demonstrates that the zero transverse momentum
of the Z -boson is smeared by the parton showers. It then peaks around 3 GeV and extends up to
700 GeV (see insert) and even beyond, if the available phase space is opened to the full hadronic
centre-of-mass energy as proposed by the power shower prescription. Note that strictly speaking the LL approximation is, of course, no longer valid there. Normalizing the power shower to
the correct QCD matrix element describing one real emission brings the integrated cross section
back into agreement with the LO prediction, as the increase of the cross section at large pT is
compensated by a reduction at small pT . These predictions may be improved further in PYTHIA
by adding QED parton showers, which are, however, formally suppressed by a factor of /s ,
so that their influence would not be visible in Fig. 2. The emitted additional partons may furthermore hadronize, which is modeled in PYTHIA with the Lund string model, but we have checked
again that the shapes of the distributions in Fig. 2 would not change significantly.
3.3. Numerical results with MC@NLO
With MC@NLO it is possible to correct not only the transverse-momentum shape, but also
the normalization of the total cross section. In our implementation, this is now also possible for
Z production at the LHC. This is demonstrated in Fig. 3, where we show on the left-hand side
that the NLO correction factor
K
dNLO
1.26
dLO
(29)
at the resonance (M = 1 TeV) is quite significant and depends also (although only slightly in
the considered mass range) on the invariant mass of the lepton pair. By definition, the normalization remains again unchanged by the HERWIG parton shower, which affects, however,
strongly the pT -spectrum (right). While the fixed-order prediction diverges as pT 0, its logarithmic singularity is effectively resummed by the parton shower and leads to a smooth turnover
334
Fig. 3. Mass (left) and transverse-momentum spectra (right) after matching the NLO QCD corrections to the HERWIG
QCD parton shower (triangles). The mass spectra have been normalized to the LO QCD prediction.
with a maximum at around 8 GeV, i.e. at a value that is considerably larger than in the case of
PYTHIA. Note that the parton shower in MC@NLO replaces part of the NLO contribution, so
that switching it off (not shown) does not lead to a fully correct NLO prediction. As was already the case in PYTHIA, we have checked that adding QED parton showers or hadronization,
which is modeled in HERWIG with the cluster model, does not lead to visible changes in the
distributions of Fig. 3.
3.4. Numerical results with joint resummation
Theoretically, the most precise predictions are obtained if logarithmically enhanced soft and
collinear parton emission is analytically resummed, both close to threshold and close to pT 0,
as described in Section 2.5 above. As can be seen from Fig. 4 (left), the NLO K-factor (dashed)
is increased further by the resummed contributions, in particular in pure threshold resummation
(dotted), which is only approximately
described by joint resummation, as we are still relatively
far from the production threshold at S = 14 TeV. Only for larger values of M would the joint
and threshold resummed predictions coincide. As the additional increase in K is only of the order
of a few percent, the prediction is nicely stabilized. We have checked that the K-factors of 1.26
at NLO and 1.28 at NNLO [10] for DrellYan lepton pairs with mass 1 TeV coincide precisely
with our NLO and threshold resummed results. Fig. 4 (right) demonstrates that the true NLO pT distribution (dashed) diverges as pT 0, as it must, and does indeed not exhibit the unphysical
maximum of MC@NLO without parton showers around 10 GeV. Resummation leads again to a
smooth turnover with a maximum at around 8 GeV. Note that the jointly resummed prediction
(full) follows the one with pT -resummation (dotted) very closely, as is to be expected from the
definition of the variable and the form of the function Gc (see above).
3.5. Comparison of numerical results and theoretical uncertainties
We now confront the best versions of the three different theoretical approaches with each
other, superimposing in Fig. 5 the LO matrix element corrected predictions obtained with
PYTHIAs power shower (circles), the NLO matrix element corrected predictions obtained
335
Fig. 4. Mass (left) and transverse-momentum spectra (right) in NLO QCD (dashed) and after resumming threshold and
pT logarithms (dotted) or both at the same time (full line). The resummed cross sections have been matched to those at
NLO, and the mass spectra have been normalized to the LO QCD prediction.
Fig. 5. Mass (left) and transverse-momentum spectra (right) in PYTHIA with LO matrix elements matched to QCD
parton showers (circles), in MC@NLO with NLO matrix elements matched to the HERWIG QCD parton shower (stars),
and after matching the NLO QCD corrections to joint resummation (full line). The mass spectra have been normalized
to the LO QCD prediction, and the renormalization and factorization scale uncertainties in the resummed predictions are
indicated as shaded bands.
with MC@NLOs parton shower (stars), and the jointly resummed prediction matched to the
NLO matrix elements (full line). The correction factors for the mass spectra (left), which have
been normalized to the LO QCD prediction, show only a very weak mass dependence. We have
multiplied the PYTHIA mass spectrum by hand with a global K-factor of 1.26. Otherwise, within
the statistical error bars the PYTHIA K-factor would just be unity, since the normalization of the
total cross section is changed neither by the parton shower nor by the LO matrix element correction, that serves to bring the power shower back into agreement with the LO QCD prediction.
336
Fig. 6. Dependence of the total Z -boson production cross section at the LHC on the common factorization/renormalization scale F,R in LO QCD (full), NLO QCD (dashed), and after matching the NLO QCD corrections to
joint resummation (dotted), LO matrix elements to the PYTHIA parton shower (dot-dashed), and NLO matrix elements
to the HERWIG parton shower (long dot-dashed).
The MC@NLO K-factor agrees almost perfectly with the one of joint resummation, since we
saw in Fig. 4 (left) that threshold and joint resummation lead only to a very modest increase of
the NLO K-factor. The theoretical uncertainty induced in the resummed prediction through the
simultaneous variation of the renormalization and factorization scale by a factor of two around
the central scale M is also considerably smaller than the K-factor, indicating a nice stabilization of the theoretical prediction. The pT -spectra (right) are, for all three improvements, no
longer divergent. The PYTHIA prediction rises and falls rather steeply around its maximum at
3 GeV, whereas the MC@NLO and resummed predictions rise and fall more slowly around the
peak at 8 GeV, which has furthermore a slightly smaller cross section. The agreement between
MC@NLO and joint resummation is impressive, in particular for a scale choice of M/2 (upper
end of the shaded band) at low pT and M at intermediate pT . We can therefore conclude that our
implementation of Z production in MC@NLO reaches almost the same level of precision as our
joint resummation calculation, but offers the additional advantage of an easy implementation in
the analysis chains of the LHC experiments.
The scale uncertainty of the total cross section (integrated over all transverse momenta and
over the invariant mass in the range from 900 to 1200 GeV) is shown in Fig. 6. The LO QCD
prediction (full) agrees with the PYTHIA prediction (dot-dashed) at the same order, as the total cross section and its scale dependence is not modified by the parton shower. Since s (R )
does not enter the calculation at this order, the full scale dependence is in fact due to the factorization scale. The LO scale dependence does, however, not give a reliable estimate of the
theoretical error, since the NLO cross section (dashed) is considerably larger. At NLO, the factorization scale dependence is reduced as expected, but s (R ) makes its appearance, so that an
additional renormalization scale dependence is introduced. The total NLO scale dependence is
337
Fig. 7. Mass (left) and transverse-momentum spectra (right) after matching the NLO QCD corrections to joint resummation with CTEQ6M (full) and MRST 2004 NLO [34] (dashed) parton densities. The mass spectra have been normalized
to the LO QCD prediction using CTEQ6L and MRST 2001 LO [35] parton densities, respectively. The shaded bands
indicate the deviations added in quadrature induced by the up and down variations along the 20 independent directions
that span the 90% confidence level of the data sets entering the CTEQ6 global fit [36].
reduced to +6/3% (vertical lines), once the leading and next-to-leading logarithms (NLL)
are resummed (dotted). The MC@NLO prediction (long dot-dashed) agrees with the one at
NLO (dashed) for the central scale, but has a weaker scale dependence due to the resummation of leading logarithms in the parton shower [33]. Since both the type (pT vs. joint) and order
(LL vs. NLL) of the resummed logarithms differ between MC@NLO and joint resummation, the
two scale variations need not (and do not) coincide.
We estimate in Fig. 7 the theoretical uncertainty coming from different parameterizations of
parton densities. Since the invariant mass M of the lepton pair is correlated with the momentum
fractions xa,b of the partons in the external protons, the normalized mass spectra (left) are indicative of the different shapes of the quark and gluon densities in the CTEQ6M (full) and MRST
2004 NLO [34] (dashed) parameterizations. The latter also influence the transverse-momentum
spectra, which are slightly harder for MRST 2004 NLO than for CTEQ6M. The shaded bands
show the uncertainty induced by variations along the 20 independent directions that span the
90% confidence level of the data sets entering the CTEQ6 global fit [36]. It remains modest, i.e.
about +/5%, and is thus similar to the scale uncertainty of +6/3%.
The uncertainty at low transverse momenta coming from non-perturbative effects in the PDFs
is usually parameterized with a Gaussian form factor describing the intrinsic transverse momentum of partons in the proton. We show in Fig. 8 the effects coming from three different choices
of the form factor, i.e. those of LadinskyYuan (LY-G) [37], BrockLandryNadolskyYuan
(BLNY) [38], and KonychevNadolsky (KN) [39], on the quantity
=
d (res.+NP) (R = F = M) d (res.) (R = F = M)
.
d (res.) (R = F = M)
It is obvious that these non-perturbative contributions are under good control, as their effect is
always less than +3/6% for pT > 5 GeV and thus considerably smaller than the scale and PDF
uncertainties.
338
Fig. 8. Variation (in percent) of the transverse-momentum spectrum after matching the NLO QCD corrections to joint
resummation with CTEQ6M parton densities for three different choices of a non-perturbative form factor.
4. Conclusions
In summary, we have improved the theoretical predictions for the production of extra neutral
gauge bosons at hadron colliders by implementing the Z bosons in the MC@NLO generator
and by computing their differential and total cross sections in joint pT and threshold resummation. The two improved predictions were found to be in excellent agreement with each other for
mass spectra, pT spectra, and total cross sections, while the PYTHIA parton and power shower
predictions usually employed for experimental analyses show significant shortcomings both in
normalization and shape. The theoretical uncertainties from scale and parton density variations
and non-perturbative effects were found to be +6/3%, +/5%, and less than +3/6%, respectively, and thus under good control. The implementation of our improved predictions in
terms of the new MC@NLO generator or resummed K factors in the analysis chains of the Tevatron and LHC experiments should be straightforward and lead to more precise determinations
or limits of the Z -boson masses and/or couplings. While we have shown numerical results for
Z -bosons associated with the U (1) gauge symmetry, our calculations are completely general
and easily applicable to different grand unification, extra-dimensional, or other models. Our modified MC@NLO program has been endorsed by the original authors and can be obtained, like the
joint resummation program, at http://lpsc.in2p3.fr/klasen/software.
Acknowledgements
This work has been supported by the CNRS/IN2P3 with a pilot grant for the LHC-TheoryFrance initiative and by two Ph.D. fellowships of the French ministry for education and research.
We thank B. Clment for finalizing several figures, D. de Florian for allowing us to use his parton
density routines, and S. Frixione for useful comments on the manuscript.
339
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
A. Leike, Phys. Rep. 317 (1999) 143.

M. Carena, A. Daleo, B.A. Dobrescu, T.M.P. Tait, Phys. Rev. D 70 (2004) 093009.
A. Abulencia, et al., CDF Collaboration, Phys. Rev. Lett. 96 (2006) 211801.
LEP electroweak working group, http://lepewwg.web.cern.ch.
F. Ledroit, J. Morel, B. Trocm, ATL-PHYS-PUB-2006-024, July 2006.
R. Cousins, J. Mumford, V. Valuev, CMS NOTE 2005/002, February 2005.
T. Sjstrand, S. Mrenna, P. Skands, JHEP 0605 (2006) 026.
C. Ciobanu, T. Junk, G. Veramendi, J. Lee, G. De Lentdecker, K. McFarland, K. Maeshima, FERMILAB-FN0773-E, July 2005.
G. Miu, T. Sjstrand, Phys. Lett. B 449 (1999) 313.
R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 359 (1991) 343;
R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 644 (2002) 403, Erratum.
S. Frixione, B.R. Webber, JHEP 0206 (2002) 029.
G. Corcella, et al., JHEP 0101 (2001) 010.
G. Corcella, M.H. Seymour, Nucl. Phys. B 565 (2000) 227.
S. Catani, F. Krauss, R. Kuhn, B.R. Webber, JHEP 0111 (2001) 063.
S. Mrenna, P. Richardson, JHEP 0405 (2004) 040.
J. Alwall, et al., Eur. Phys. J. C 53 (2008) 473.
F. Caravaglios, M.L. Mangano, M. Moretti, R. Pittau, Nucl. Phys. B 539 (1999) 215.
G. Bozzi, B. Fuks, M. Klasen, Nucl. Phys. B 794 (2008) 46.
M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
J.L. Rosner, Phys. Rev. D 35 (1987) 2244.
R.W. Robinett, J.L. Rosner, Phys. Rev. D 25 (1982) 3036;
R.W. Robinett, J.L. Rosner, Phys. Rev. D 27 (1983) 679, Erratum.
P. Abreu, et al., DELPHI Collaboration, Phys. Lett. B 485 (2000) 45.
P. Aurenche, J. Lindfors, Nucl. Phys. B 185 (1981) 274.
G. Bozzi, B. Fuks, M. Klasen, Phys. Rev. D 74 (2006) 015001.
G. Bozzi, B. Fuks, M. Klasen, Nucl. Phys. B 777 (2007) 157.
H.N. Li, Phys. Lett. B 454 (1999) 328.
E. Laenen, G. Sterman, W. Vogelsang, Phys. Rev. D 63 (2001) 114018.
A. Kulesza, G. Sterman, W. Vogelsang, Phys. Rev. D 66 (2002) 014011.
J. Pumplin, D.R. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.K. Tung, JHEP 0207 (2002) 012.
K. Hagiwara, et al., Particle Data Group, Phys. Rev. D 66 (2002) 010001.
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
S. Frixione, M.L. Mangano, JHEP 0405 (2004) 056.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 604 (2004) 61.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216.
W.K. Tung, H.L. Lai, A. Belyaev, J. Pumplin, D. Stump, C.P. Yuan, JHEP 0702 (2007) 053.
G.A. Ladinsky, C.P. Yuan, Phys. Rev. D 50 (1994) 4239.
F. Landry, R. Brock, P.M. Nadolsky, C.P. Yuan, Phys. Rev. D 67 (2003) 073016.
A.V. Konychev, P.M. Nadolsky, Phys. Lett. B 633 (2006) 710.

M2-brane probe dynamics and toric duality

Seok Kim a , Sangmin Lee b , Sungjay Lee b , Jaemo Park c,d,
a School of Physics, Korea Institute for Advanced Study, Seoul 130-722, Republic of Korea
b School of Physics and Astronomy, Seoul National University, Seoul 151-747, Republic of Korea
c Department of Physics, Postech, Pohang 790-784, Republic of Korea
d Postech Center for Theoretical Physics (PCTP), Pohang 790-784, Republic of Korea
Received 12 October 2007; accepted 11 January 2008

Abstract
We study the dynamics of a single M2-brane probing toric CalabiYau four-fold singularity in the context
of the recently proposed M-theory crystal model of AdS4 /CFT3 dual pairs. We obtain an effective Abelian
gauge theory in which the charges of the matter fields are given by the intersection numbers between loops
and faces in the crystal. We argue that the probe theory captures certain aspects of the CFT3 even though
the true M2-brane CFT is unlikely to be a usual gauge theory. In particular, the moduli space of vacua of
the gauge theory coincides precisely with the CalabiYau singularity. Toric duality, partial resolution, and
a possibility of new RG flows are also discussed.
PACS: 11.25.Yb
Keywords: AdS/CFT; M-theory; Toric geometry
1. Introduction
Throughout the development of the AdS/CFT correspondence [1] for the past decade, the
AdS4 /CFT3 cases in M-theory has remained much less understood than the AdS5 /CFT4 cases
in IIB string theory. The lack of a perturbative description of M-theory and its M2/M5-brane
world-volume theories is a main culprit.
* Corresponding author at: Department of Physics, Postech, Pohang 790-784, Republic of Korea. Fax: +82 54 279
3099.
E-mail address: jaemo@postech.ac.kr (J. Park).
doi:10.1016/j.nuclphysb.2008.01.012
S. Kim et al. / Nuclear Physics B 797 (2008) 340370
341
When supergravity serves as a good approximation to the full theory, M-theory on AdS4 is
no more difficult than IIB string theory on AdS5 . The difficulty still remains on the CFT side.
While the CFT4 can be derived by quantizing open strings on the D3-branes probing a conical
singularity, no systematic method is available to derive the world-volume theory of M2-branes
near a singularity. In early works on AdS4 /CFT3 , proposals for the CFT3 were made based on
global symmetries and analogy with quiver gauge theories on D-branes; see, for example [26].
None of the proposals was systematic enough to be applied to less symmetric geometries such
as the infinite families Y p,q [79].
For M2-branes probing toric CalabiYau four-fold (CY4 ) cones, a first step toward constructing the CFT3 was taken in [10,11]. It is an M-theory generalization of the brane-tiling model
[1216], which has proven extremely successful in studying the CY3 /CFT4 counterpart. As the
information on the CFT3 is encoded in a three-dimensional periodic lattice, the model was named
the M-theory crystal model. Following [11], we will distinguish the D3/CY3 model from the
M2/CY4 model by referring to the former as the tiling model and the latter as the crystal model.
By construction, the tiling/crystal models share many features, for example,
Tiling/Crystal
Edges (bonds)
Vertices (atoms)
Faces (tiles)
CFT4 /CFT3
matter fields
super-potential terms
gauge groups
This is by now a well established dictionary in the tiling model. As for the crystal model, the
first two items were verified in [10,11], but the third one posed a puzzle. The graph of the tiling
model splits the unit cell (T 2 ) into disjoint faces (tiles), and the D5/NS5-brane interpretation
[13,17] of the model explains why gauge groups are assigned to the faces. In contrast, the graph
of the crystal model does not divide the unit cell (T 3 ) into disjoint cells. At first sight, the crystal
model does not seem to assign any role to faces and cells.
There is another related problem. In the crystal model, the fundamental fields of the CFT3 are
interpreted as open M2-brane discs ending on M5-branes, and the gauge-invariant operators as
closed spherical M2-branes formed by gluing the M2-discs. It requires a new algebraic structure
beyond the usual matrix product. So we are tempted to conclude that the CFT3 cannot be written
down until the full non-Abelian M5-brane theory is constructed, which is a notoriously difficult
subject on its own.
If the non-Abelian nature of M-branes is the major obstacle to further progress, we may ask
whether the problem can be simplified by considering a single M2-brane probe. In the D3/CY3
case, the Abelian gauge theory for a single D3-brane probe proved useful in the early stages of the
development [1820] as well as in the discovery of the tiling model. In general, the probe theory
does not exhibit the dynamics of the full non-Abelian theory. For example, the super-potential
often vanish in the Abelian theory while it plays an essential role in the non-Abelian theory. On
the other hand, the probe theory does encode some important features of the full theory such as
the multiplicity of matter super-fields and the moduli space of vacua which coincide with the
CY3 singularity.
Bearing in mind both the usefulness and the limitation of the D3-brane probe theory, in the
present paper, we explore the probe theory of an M2-brane in the context of the crystal model.
The crystal model maps the M2-brane probe to an M5-brane. The information on the toric
CY4 geometry is reflected in the shape of the M5-brane in the internal space. We take the
world-volume theory of the M5-brane to be a free CFT with a single tensor multiplet of (2, 0)
342
supersymmetry in six dimension. Once we compactify the M5-brane theory along the internal
space, we obtain an effective Abelian gauge theory in (2 + 1)-dimensions with N = 2 supersymmetry. As the open M2-brane discs are charged under the self-dual tensor field on the M5-brane,
they indeed become charged matter fields in the probe theory. Moreover, the gauge groups can
be attributed to faces of the crystal as in the dictionary above. The charges of the matter fields
are then given by the intersection numbers between loops and faces in the crystal.
Some cautionary remarks are in order. There is no reason to expect that the M2-brane theory
is a gauge theory. In fact, the crystal model points strongly to the contrary. The Abelian gauge
theory under discussion should be regarded, at best, as an approximate description of the true
theory, to the same extent as the M5-brane theory is an Abelian tensor gauge theory. Physical
consequences derived from the probe theory may or may not survive in the full non-Abelian
theory. In either case, we find it reasonable to expect the probe theory to be a useful guide toward
the construction of the non-Abelian theory.
Partial resolution, which we study in Section 6, is a good example to illustrate the point. We
will show that the M2-brane probe theory admits partial Higgsing which translates to partial
resolution of the CY4 singularity. An important result here is that this theory does not allow all
possible resolution that a D3-brane probe does. Suppose a dual explanation of this difference can
be found in terms of the difference between IIB supergravity and M supergravity (a question we
hope to address in a future work). Then, it would impose a non-trivial constraint on any future
candidates for the non-Abelian CFT3 . Our discussion in Section 7 on possible RG flows could
serve similar purpose.
This paper is organized as follows. We begin with a brief review of the crystal model [10,11]
in Section 2. In Section 3, we explain how to identify the U (1) gauge groups and compute the
charges from the crystal model. We also explain how the charge assignment rule can be derived
from the world-volume theory of a single M5-brane. In Section 4, we show that the moduli space
of vacua coincides with the toric CY4 , thereby confirming the validity of the theory. In Section 5,
we show that two different gauge theories can correspond to the same CY4 in a way similar to
the toric duality of [20]. Toric duality of non-Abelian CFT4 is known to be a Seiberg duality. The
meaning of toric duality of non-Abelian CFT3 is an interesting open question. In Section 6,
we study partial Higgsing of a theory to obtain another one, thereby partially resolving the CY4
singularity, following similar consideration of the Abelian CFT4 [20]. In Section 7, we note that
the Abelian theory allows a massive deformation analogous to the KlebanovWitten flow [21]
from an orbifold CFT to the conifold CFT. We conclude with some discussions in Section 8.
2. M-theory crystal model: A brief review
We give a brief review of the M-theory crystal model [10,11]. The crystal model relates a toric
CY4 to a three-dimensional periodic graph (a crystal) which encodes informations on the CFT3 .
As in [11,15,20], the maps between a crystal and the corresponding toric diagram will be called
the forward and inverse algorithms; see Fig. 1.
The toric diagram forms a convex polyhedron in Z3 Z4 . The reduction from Z4 to Z3 is
a consequence of the CY condition. See, for example [22,23] for more information on toric
geometry in this context.
The crystal model follows from a T-duality of M-theory. We take the T-duality transformation
along T 3 T 4 in alignment with the projection Z4 Z3 . This corresponds to x 6,7,8 directions
in Table 1. By T-duality, we mean the element t in the SL(2, Z) SL(3, Z) duality group which
acts as t : C(3) + i gT 3 1/ . The stack of N M2-branes turns into a stack of N M5-
343
Fig. 1. Crystal vs. Toric diagram.
Table 1
The brane configuration for the CFT3 . Away from the origin of R3 (345), the special Lagrangian manifold is locally a
product of a 2-plane in R3 (345) and a 1-cycle in T 3 (678)
0
1
2
3
4
5
6
7
8
9
11
M5
M5
branes wrapping the dual T 3 . We call them the T -branes. The degenerating circle fibers turn into
another M5-brane extended along the (2 + 1)d world-volume and a non-trivial 3-manifold in
R3 T 3 . We call it the -brane. Preservation of supersymmetry requires that the -brane wrap
a special Lagrangian submanifold of R3 T 3 = (C )3 , and that it is locally a plane in R3 and a
1-cycle in T 3 . The result is summarized in Table 1.
The crystal graph is the intersection locus between the T -branes and the -brane projected
onto the T 3 (the alga projection). It was shown in [10] that the matter fields of the CFT3 are the
M2-discs whose boundaries encircle the bonds of the crystal, hence the first item of the dictionary
we discussed in the introduction. The derivation of the M2-disc picture made use of the fact that
the T -branes and the -brane can be merged into a single, smooth, brane configuration. This
observation will play a crucial role when we later discuss the gauge groups. It was also argued
in [10,11] that the atoms (vertices) of the crystal give the super-potential terms. As an evidence,
it was shown to reproduce the BPS spectrum of meson operators of the CFT3 .
In principle, the forward algorithm consists of two simple steps: (a) reading off the CFT3
(b) showing that its moduli space of vacua gives the toric CY4 . These steps were not directly verified in [10,11] since the information on the gauge group was missing. Instead, it was shown that
the fast forward algorithm [12,13] of the tiling model can be applied without any modification.
The fast forward algorithm is based on the concept of perfect matchings. A perfect matching is
a subset of bonds of the crystal, such that every atom of the crystal is an end-point of precisely
one such bond. The bonds in each perfect matching carry an orientation. We choose to orient the
arrows to go from a white atom to a black one. Perfect matchings have several nice properties.
In particular,
(1) Each perfect matching can be located in the toric diagram. The relative coordinate in the
toric diagram between two perfect matchings p and p is given by the homology charge of
(p p ) regarded as a one-cycle in T 3 .
344
Fig. 2. Inverse algorithm for C4 , reproduced from [11].
Fig. 3. Untwisting from alga to amoeba, reproduced from [11].
(2) The perfect matchings solve the Abelian version of the F-term condition for the chiral fields
Xi associated to the bonds, if we set

p Xi ,p
,
Xi =
where Xi , p
equals 1 if p contains the bond Xi and 0 otherwise.
The Abelian gauge theory of the present paper fills the gap by showing that the slow forward
algorithm also works. The perfect matchings will continue to be a useful device.
The inverse algorithm is not as well established as the forward algorithm. Roughly speaking,
one draws a 1-cycle in the T 3 for each edge of the toric diagram and then merge the 1-cycles
to make up the crystal. For simple examples with high degree of symmetry, the procedure is
unambiguous as in Fig. 2. Even in the general case, the result can be checked by using the
forward algorithm. It is certainly desirable to understand the inverse algorithm better by finding
an explicit expression for the special Lagrangian manifold .
It is useful to consider the projection of the crystal onto (the amoeba projection). We can
think of the amoeba as a 3-manifold with some defects. As we shrink the 2-fans along the radial
direction, the points at infinity form a locally one-dimensional defect. The 1-cycles paired
with the 2-fans are localized along the defects. Globally, the defect is isomorphic to the dual
toric diagram. When the toric diagram has no internal points, the amoeba has the topology of a
three-sphere apart from the defect.
It is possible to obtain the amoeba from the alga and vice versa using the untwisting procedure.
The untwisting flips the orientation of the space transverse to the bond of the crystal as shown in
Fig. 3. We apply the untwisting map to the alga of C4 , we obtain the amoeba depicted in Fig. 4.
Note that the dual toric diagram is a tetrahedron as expected. The 1-cycles are localized along the
345
Fig. 4. Dual toric diagram and amoeba diagram for C4 , reproduced from [11]. We represent a three-sphere as the union
of two balls with the surfaces identified. The dotted lines on the balls denote the defects.
Fig. 5. Toric diagrams of the CY4 considered in this paper.
dual toric diagram as they should be. Applying the untwisting map to C(Q1,1,1 ) yields a similar
result, with the dual toric diagram being a cube.
346
Fig. 6. Charges as intersection numbers between the matter loops and the charge faces.
We close this section with Fig. 5 which contains the toric diagrams of CY4 we consider in this
paper. The corresponding crystals will be shown in later sections.
3. Abelian gauge theory
3.1. Charge assignment rule
The main feature of the Abelian gauge theory associated to the crystal model is that the U (1)
gauge groups are attributed to faces of the crystal. Recall that a matter field in the crystal model
is represented by an M2-brane disc whose boundary loop is localized along a bond of the crystal.
Then, the charges of the matter fields are simply the intersection number between the matter
loops and the charge faces. Fig. 6 gives a pictorial description of the charge assignment rule.
In the figure, we fixed the orientation of the matter loops by imagining arrows from white atoms
to black ones and applying the right-hand rule. This ensures that the M2-discs surrounding an
atom can form an orientable two-sphere.
Although this is quite reminiscent of the charge assignment in the tiling model, we should
emphasize the difference between the two models. In the D5/NS5-brane picture of the tiling
model, it is very clear how the disjoint tiles of D5-branes give rise to gauge groups for the matter
fields that are open strings connecting the tiles. The role of faces in the crystal model is not very
clear at first sight. In addition, the number of independent gauge groups is not so easy to count
as in the tiling model.
We will shortly explain the M-theory origin of the charge assignment as well as how to count
the number of gauge groups. Before doing so, we give a concrete example of the charge assignment in Fig. 7. It has two independent charges. This example is especially easy to visualize
because the faces lie on planes, which is not true in general.
3.2. M-theory origin
As we reviewed in Section 2, the crystal model comes from two types of M5-branes: the T brane and the -brane. An important step, which was used already in identifying the matter fields
as open M2-branes, is that the two M5-branes can merge into a single M5-brane of a complicated
topology. Let us call it M. This M5-brane carries on its world-volume the self-dual two-form
field B. The Abelian gauge fields of the crystal model come from the KaluzaKlein reduction of
the B-field along the internal manifold M.
Open M2 branes ending on an M5-brane couples to the self-dual two-form field B as

SM2 B.
(3.1)
D
347
Fig. 7. Two independent faces for C(Q1,1,1 ).
When the M5-brane is extended over R1,2 M, we can use harmonic one-forms on M to
decompose B as
B(x, y) = Aa (x) a (y) + ,
(3.2)
where the omitted terms are required to ensure the self-duality condition dB = dB.
An M2-disc whose boundary is a 1-cycle Ci in M become a particle in (2 + 1)-dimension
with a coupling to the gauge field,

Si Qia Aa .
(3.3)
The charge is easy to read off

Qia = a = (Ci , Sa ).
(3.4)
Ci
In the last step, we replaced the integral of a over Ci by the intersection number between Ci
and the two-cycle Sa which is Poincar dual to wa .
It remains to show that faces of the crystal define 2-cycles of M. When the T -brane and
-brane merge to form M, the bonds of the crystal serve as the throat connecting the two
components. So, the faces should be carried over to the side. Recall that the transition from
the crystal to the amoeba involves the untwisting process. It means that the tangent plane to the
face go through a 180 rotation as it moves from an atom to its neighbor along a bond.
In the amoeba diagram, the bonds of the crystal and the dual toric diagram are linked nontrivially. So, it is impossible for the extended face to form a compact 2-cycle of M. The only
alternative is that it runs off to infinity to form a non-compact 2-cycle. An example of such a
2-cycle is depicted in Fig. 8. In the figure, the face emanates from the bonds (A1 B2 A2 B1 )
where Ai , Bi are define in Fig. 7. It is twisted in accordance with the untwisting process, and it
runs off to the dual toric diagram at infinity. The red segments of the dual toric diagram denote
the asymptotic boundary of the face.
We have checked for all examples considered in this paper that any face of the crystal can be
extended to the amoeba diagram and sent to asymptotic infinity. We take it as a strong evidence
that it works for any crystal, but we have not found a general proof yet.
348
Fig. 8. A 2-cycle in the amoeba diagram of C(Q1,1,1 ). The union of the two solid cubes with the surfaces identified
represent a three-sphere. The edges of the cube (green line) denote the dual toric diagram at asymptotic infinity. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 9. C4 .
Fig. 10. D3 .
3.3. Examples
To illustrate the charge assignment rule, we present a few more examples in the following set
of figures, Fig. 9 through 14, along with tables summarizing the charges. The faces are not drawn
in the figures explicitly, but the charge table can be used to figure out the faces for each charge.
An astute reader may have noticed that the charge matrix of C(Q1,1,1 ) and that of dP3 C
are the same except that C(Q1,1,1 ) does not have the 1,2 fields that dP3 C has. Similarly, D3
and C3 /(Z2 Z2 ) have almost identical charge matrices. We will discuss their implications in
Section 7.
Fig. 11. C2 /Z2 C2 .
Fig. 12. (C2 /Z2 )2 .
Fig. 13. C3 /(Z2 Z2 ) C.
Fig. 14. dP3 C.
349
350
4. Moduli space of vacua

Given a toric CY4 and the associated crystal, it is natural to expect that the Abelian theory
defined in the previous section has a moduli space of vacua, MV , that coincides with the CY4 .
We now show that it is indeed the case. First, we present a few illustrative examples to familiarize
the reader with the Abelian theory and its MV . Then we give a general proof that dim(MV ) = 4
in parallel with a similar proof in the tiling model. Finally, we make a comparison with the
D-brane gauge theory in the N = 4 cases.
4.1. Examples
1. C(Q1,1,1 ) We have eight gauge invariant variables zij k = Ai Bj Ck (i, j, k = 1, 2). Label
them as
z0 = A1 B1 C1 ,
z1 = A1 B1 C2 ,
z2 = A1 B2 C1 ,
z3 = A1 B2 C2 ,
z4 = A2 B1 C1 ,
z5 = A2 B1 C2 ,
z6 = A2 B2 C1 ,
z7 = A2 B2 C2 .
(4.1)
They satisfy the quadratic relations

z0 z7 = z1 z6 = z2 z5 = z3 z4 ,
z0 z3 = z2 z1 ,
z7 z4 = z5 z6 ,
z0 z5 = z1 z4 ,
z7 z2 = z6 z3 ,
z0 z6 = z4 z2 ,
z7 z1 = z3 z5 .
(4.2)
C(Q1,1,1 ).
As noted earlier in [2,5], this is precisely the algebraic definition of

2. C2 /Z2 C2
In addition to the neutral fields, , X1 , X2 , we have four gauge invariant variables: w1 =
Y1 Y2 , w2 = Y3 Y4 , u1 = Y1 Y4 , u2 = Y2 Y3 . The F-term conditions demands that
= X1 X 2 ,
w1 = w2 ( w).
C2
(4.3)
part and the algebraic relation, u1 u2 =

describes
We note that (X1 , X2 ) span the
C2 /Z2 .
3. (C2 /Z2 )2
The X fields and Y fields decouple from each other. The F-term condition demands that
X1 X2 = X3 X4 ( w). The gauge-invariant coordinates u1 = X1 X4 and u2 = X2 X3 satisfy
u1 u2 = w 2 , which gives C2 /Z2 . Similarly, the Y fields produce another factor of C2 /Z2 .
4. C3 /(Z2 Z2 ) C
There are seven gauge invariant coordinates, 1 , 2 , w1 A1 B1 C1 , w2 A2 B2 C2 , z1
A1 A2 , z2 B1 B2 , z3 C1 C2 , and two F-term conditions,
1 = 2 ( ),
w2 ,
w1 = w2 ( w).
(4.4)
Clearly, parameterizes the C factor and the algebraic relation, z1 z2 z3

describes the
orbifold C3 /(Z2 Z2 ).
5. dP3 C
The gauge-invariant coordinates are 1 , 2 and zij k = Ai Bj Ck . The F-term condition gives
two linear constraints:
= w2 ,
1 = 2 ( ),
z111 = z222 ( w).
(4.5)
351
Label the remaining six zij k as

s1 = A2 B1 C1 ,
s2 = A1 B2 C2 ,
t1 = A1 B2 C1 ,
t2 = A2 B1 C2 ,
u2 = A1 B1 C2 ,
u2 = A2 B2 C1 .
(4.6)
The gauge-invariant coordinates are subject to the quadratic relations,

w 2 = s1 s2 = t1 t2 = u1 u2 ,
ws1 = t2 u2 ,
ws2 = t1 u1 ,
wt1 = u2 s2 ,
wt2 = u1 s1 ,
wu1 = s2 t2 ,
wu2 = s1 t1 .
(4.7)
This coincides with the known algebraic definition of dP3 . As a check, we note that the
character function of dP3 [24],
Z=

1 + 4t + t 2
= 1 + 7t + 19t 2 + O t 3 ,
3
(1 t)
requires that there be seven monomials of degree one and nineteen independent monomials
of degree two. It is easy to see that we have the correct number of monomials.
deg 1:
w, si , ti , ui ,
deg 2:
w 2 ; wsi , wti , wui ; si2 , ti2 , u2i ;

s1 t2 , s2 t1 , t1 u2 , t2 u1 , u1 s2 , u2 s1 .
(4.8)
All other monomials of degree two are redundant due to the quadratic relations.
4.2. Dimension counting
As a warm-up exercise, we first review how to count the dimension of MV in the tiling model.
Recall the well-known relations in the tiling model:
1. (gauge groups) = (faces) f .
2. (bi-fundamentals) = (edges) e.
3. (super-potential terms) = (vertices) v.
To count dim(MV ) of the Abelian theory, we note that
1. (F-term conditions) w = v 2.
The F-term conditions equate all the super-potential terms, leading to v 1 equations. One
of them turns out to be redundant.
2. (D-term conditions) q = f 1.
The D-term conditions come from the U (1) gauge groups. All matter fields are neutral under
the diagonal U (1), since they are all bi-fundamentals.
The dimension of MV is the number of matter fields minus the number of constraints:
dim(MV ) = e w q. Combining the topological condition, f e + v = 0, with the relations
352
mentioned above, we find

dim(MV ) = e (v 2) (f 1) = 3,
(4.9)
in agreement with the fact that MV coincides with the CY3 .

A similar argument can be made for the Abelian theory of the crystal model. The following
relations continue to hold.
1.
2.
3.
4.
(bi-fundamentals) = (edges) e.
(super-potential terms) = (vertices) v.
(F-term conditions) w = v 2.
dim(MV ) = e w q, where q is the number of D-term conditions.
Now, we introduce as many faces as we need to partition the T 3 into several cells, so that the
topological condition c + f e + v = 0 can be used. Here, c and f are the number of cells
and faces, respectively. Note that, for a given crystal, f c = e v is a topological invariant that
does not depend on the details of the partition.
It is easy to see that q depends only on the difference f c. When a collection of faces
surround a cell, the total charge vanishes identically, because the matter loops either penetrate
the faces twice with opposite orientations or do not penetrate them at all. Therefore, we can
remove faces while keeping f c fixed until we arrive at c = 1. Let f be the number of faces in
this configuration. By construction, f 1 = f c.
The remaining f faces come in two distinct types. The first kind of faces live in the interior
of the cell, so that they can be removed without changing c = 1. All of them give independent
U (1) charges. The second kind of faces participate in forming the walls (2-cycles) surrounding
the unit cell. The overall charge corresponding to the sum of all faces on the same wall vanishes,
again because the loops either penetrate the faces twice or do not penetrate them at all. Since
three independent walls are needed to make up a three-torus, the number of independent gauge
groups gets reduced by three. Therefore, we have q = f 3 = f c 2, which in turn implies
dim(MV ) = e (v 2) (f c 2) = 4.
(4.10)
This is consistent with the fact that MV coincides with the CY4 .
4.3. Comparison with D-brane gauge theory
It is widely believed that the world-volume theory of M2-branes in flat spacetime can be
obtained by taking the strongly coupled, infrared limit of the N = 8 super-YangMills theory
living on D2-branes. The attempt to understand the M2-brane theory from the infrared limit of
the D2-brane theory was extended to the N = 4, orbifold examples in [25]. We will now review
the result of [25] in the Abelian context and compare it with our new model.
The M2-branes probe the product of two singular ALE (asymptotically locally euclidean)
spaces. Regarding the U (1)-fiber in one of the ALE spaces as the M-theory circle, the theories
can be considered as gauge theories on the D2-brane probe in type IIA theory with several D6
branes (and possibly O6 planes), transverse to R3 . To be concrete, we focus on the C2 /Zn
C2 /Zk examples. Compactifying on a circle in C2 /Zk , we get k D6-branes, transverse to R3 , with
the D2-brane probe. The D2-brane and all its mirror images under the Zn orbifold action defines
a U (1)n gauge theory. Each of the n vector multiplets has three scalars which represent the
position in the transverse R3 . The hyper-multiplets X are bi-fundamentals connecting adjacent
353
U (1)-factors of the gauge group [26]. In addition, there are other hyper-multiplets arising from
the open string modes connecting the D2- and D6-branes, which are charged under only one of
the U (1) gauge groups.
There are two types of geometric moduli parameters we can add to the theory. Each D-term
equation for the X fields allows a FayetIliopoulos (FI) parameter. Since all X fields are neutral
under the diagonal U (1)D , we have (n 1) independent FI parameters i . They correspond to the
moduli of the first ALE space, C2 /Zn . One can also introduce the k mass parameters, mi , for the
fields, which originate from the relative distance between D2- and D6-branes. The total mass
i mi can be eliminated by shifting the origin of the Coulomb branch. The remaining (k 1)
mass parameters can be understood as the moduli of the second ALE space, C2 /Zk .
In the N = 2 language, the N = 4 D-term equations describing the Higgs branch of X fields
are given by
1 2 2 2 1 2 2 2
D
X

i,i+1 Xi+1,i Xi1,i + Xi,i1 = i ,
1
2
1
2
Xi,i+1
Xi+1,i
Xi1,i
Xi,i1
= iF ,
(4.11)
where i runs over {1, . . . , n1}. The real parameter D and complex parameter F together form
a triplet of N = 4 FI parameters. The crystal model has only N = 2 manifest supersymmetry,
and the meaning of F is not clear at present. The equations for the fields are
k1

1 2 2 2
= 0,
k1

j =1
j =1

j1 j2 = 0.
(4.12)
Note that the equations for are decoupled from those for X as well as the FI parameters.
Let us now examine MV with both the mass and the FI parameters turned on. Since the fields
become massive, their Higgs branch is lifted. Turning on the FI terms leads to breaking gauge
symmetry down to the diagonal U (1)D under which only the fields are charged. Therefore,
MV is the product of the Higgs branch, MH , for the X fields described by the D-term Eq. (4.11)
and the Coulomb branch, MC , for the U (1)D gauge theory with k massive hyper-multiplets .
As is well known, the D-terms Eq. (4.11) coincide with Kronheimers hyper-Khler quotient
construction [27] of the first ALE space, C2 /Zn , with the singularities resolved by the FI parameters. The Higgs branch, MH , does not receive any quantum corrections, and remains to be
C2 /Zn even in the strong coupling regime.
The classical Coulomb branch for the U (1)D gauge theory is R3 S 1 . As in the N = 8 case,
the periodic scalar is the Hodge dual of the U (1)D gauge field, and the radius of the circle goes to
infinity in the IR limit. The quantum correction changes the Coulomb branch significantly [28],
so that it describes the second ALE space C2 /Zk with the singularities resolved by the mass
parameters. To summarize, the moduli space of vacua of the full theory is MV = MH MC =
C2 /Zn C2 /Zk , resolved by the FI and mass parameters.
If we now choose the M-theory circle in C2 /Zn , instead of C2 /Zk , then we have n D6-branes
V , of this dual theory should
with the D2-brane probe. Since the moduli space of vacua, M
capture the same orbifold geometry, it must be that
C,
MH = M
H.
MC = M
It is easy to confirm this from the gauge theory. In the dual theory, the first ALE space C2 /Zn is
described by the Coulomb branch for U (1) gauge theory with n massive hyper-multiplets and
the second C2 /Zk by the Higgs branch for k hyper-multiplets Y . The duality also exchanges the
354
Fig. 15. Toric diagram and crystal for (C2 /Zn )2 .
mass parameters and the FI parameters. The key observation of [25] is that this is an example
of the mirror symmetry of three-dimensional gauge theories [29], and that M-theory explains
why mirror symmetry holds in this case.
We now turn to our crystal model. For simplicity, we restrict our attention to the k = n cases.
The toric diagram and the crystal for C2 /Zn C2 /Zn are depicted in Fig. 15. The gauge group
of the Abelian model is factorized as (U (1)n /U (1)D )X (U (1)n /U (1)D )Y . The matter fields
1
2
, Xi+1,i
) are charged only under the first factor (U (1)n /U (1)D )X , as
(Xi,i+1
Qi
Qi+1
1
2
Xi,i+1
Xi+1,i
+
(i = 1, . . . , n)
1
2
with all other charges vanishing. The matter fields (Yj,j
+1 , Yj +1,j ) are charged only under the
second factor (U (1)n /U (1)D )Y , in the same way as the X fields. The super-potential of the
crystal model is
W=
n

1
2
1
2
1
2
1
2
Xi,i+1
,
Xi+1,i
Yi,i+1
Yi+1,i
Xi1,i
Xi,i1
Yi,i+1
Yi+1,i
(4.13)
i=1
from which we find the F-term conditions,

1
2
1
2
Xi,i+1
Xi+1,i
Xi1,i
Xi,i1
= 0,
1
2
1
2
Yj,j
+1 Yj +1,j Yj 1,j Y j,j 1 = 0.
(4.14)
Comparing the field contents, gauge groups and F-term conditions with those of the D2-brane
gauge theories, we note that MV of the crystal model can be regarded as the product of the
355
H , of the dual description.1

Higgs branch, MH , of the first description and the Higgs branch, M
In other words,
H.
MV (crystal) = MH M
(4.15)
Unlike in the D2-brane gauge theory, where the distinction between the Coulomb and Higgs
branches is inevitable, the crystal model treats the two factors on an equal footing.
Note that, although the moduli space of vacua factorizes, the super-potential (4.13) has products of X fields and Y fields. It would be interesting to understand how the super-potential (4.13)
arises from the dynamics of the D2-brane gauge theory.
5. Toric duality
It was observed in [11] that C(T 1,1 ) C admits two different crystals (see Fig. 16). The result
was based on the fast forward algorithm. Let us compare the Abelian gauge theories of the two
crystals.
The field theory corresponding to crystal (a) has one independent gauge group. The charge
assignment for the matter fields is as follows:
A1
Q +
A2
+
B1
B2
C
0
(5.1)
The super-potential vanishes as usual since it has only two atoms. The gauge invariant variables
parameterizing the moduli space of vacua are
z1 = A1 B1 ,
z2 = A1 B2 ,
z3 = A2 B1 ,
z4 = A2 B2 ,
w=C
(5.2)
C(T 1,1 ) C.
which satisfy the relation z1 z3 = z2 z4 . The moduli space is clearly

For crystal (b), all matter fields are neutral and the super-potential takes the form
W = 1 A1 C1 2 A1 C1 + 2 B2 C2 1 B2 C2 ,
(5.3)
from which we derive the F-term conditions:

A1 C1 = B2 C2 ,
1 = 2 .
(5.4)
Fig. 16. Two crystals for C(T 1,1 ) C, reproduced from [11].
1 Similar Higgs branches also appear in the hyper-Khler quotient description of the d = 3, N = 3 CFT discussed
in [3032].
356
The moduli space is again C(T 1,1 ) C.

Thus we see that the two crystals give two field theories describing the same toric singularity.
The same kind of degeneracy of four-dimensional Abelian gauge theories was discovered in [20],
and was named toric duality. It was shown later that toric duality is realized in the non-Abelian
quiver gauge theories as Seiberg duality.
To investigate the fascinating possibility of non-Abelian toric duality in M-theory, we will
have to construct the non-Abelian theory first, which is beyond the scope of this paper. In the
rest of this section, we will show that, at least in the Abelian context, toric duality of the crystal
model is similar to that of the tiling model in many respects.
5.1. GLSM charge matrix
In the discussion of toric duality in the tiling model [20], the gauged linear sigma model
(GLSM) description of the toric singularity plays a crucial role. The matter fields of the GLSM
are precisely the perfect matchings p . The charge matrix Q of the GLSM can be divided into
two parts, Q = (QF , QD )T , where QF is related to the F-term conditions of the (Abelian) field
theory while QD originates form the D-term conditions. Let us briefly review how to construct
QF and QD from the field theory. See [20] for details.
As we discussed in Section 2, the perfect matchings solve the F-term conditions automatically
via

Xi =
(5.5)
p Xi ,p
,
where Xi , p
equals 1 if p contains the bond Xi and 0 otherwise. In general, the number
of perfect matchings {p } can be larger than the number of fields {Xi }. Following [20], as an
intermediate step, we can choose a minimal set of independent fields {X m } after solving the
F-term conditions. Define a matrix T as
Tm = X m , p
.
Then, QF is determined as the cokernel of the T ,

Tm QF m
= 0.
(5.6)
(5.7)
By construction, all fields Xi are neutral under QF . They are only charged under the Abelian
gauge groups of the field theory, which are related to QD . Concretely, QD assigns charges
(QD )a to the perfect matchings p such that the charges of the Abelian gauge theory Qai
is reproduced through (5.5). As a consistency check, let us count the dimension of MV of the
GLSM. As discussed in Section 4, the number of independent matter fields {X m } is e v + 2
and the number of relevant U (1) gauge groups is q = e v 2. Suppose that there are p perfect
matchings. The charge matrix Q is then a (p (e v + 2) + (e v 2)) p = (p 4) p
matrix, which implies that the dimension of the moduli space of vacua of the GLSM is four as
expected.
The GLSM can be used to study blow-up of the toric singularity. The blow-ups are controlled
by the FI parameters associated to the gauge groups. In the tiling/crystal models, since the GLSM
charge matrix is constructed from the field theory underlying the tiling/crystal, not all FI parameters are allowed in the GLSM. The QD part of the GLSM charge matrix can inherit the FI
parameters from the gauge theory, but the QF part is not allowed to have the FI parameters.
357
5.2. C(T 1,1 ) C revisited

Let us construct the GLSM charge matrices of the two crystals for C(T 1,1 )C. For crystal (a),
all matter fields are independent since there is no super-potential. In terms of perfect matchings
(Fig. 17(a)), they are written as
A1 = p1 ,
B2 = p3 ,
B1 = p4 ,
A2 = p6 ,
C = p0 ,
(5.8)
which implies that the matrix T is an identity matrix and that QF = 0. Therefore, the GLSM
charge matrix consists of only QD , which is identical to the charge of the Abelian gauge theory:
Q = ( 1 1 1 1 0; ) ,
(5.9)
where we included the FI parameter of the field theory.

For crystal (b), the independent matter fields are chosen as {1 , C1 , A1 , B2 } after solving the
F-term conditions (5.4). These matter fields are written in terms of the perfect matching as
1 = p0 ,
C 1 = p 1 p3 ,
which leads to
0
1
T =
0
0
0
1
0
1
0
0
1
0
0
0
1
1
A1 = p4 p6 ,
1
1
C1
0
A1
0
B2
p1
0
1
0
0
p3
0
1
0
1
B2 = p3 p6 ,
p4
0
0
1
0
p6
0
0
1
1
p0
1
0
.
0
0
(5.10)
(5.11)
The charge matrix QF (cokernel of T ) is

QF = ( 1 1
1 1 0 ) .
(5.12)
The charge matrix QD is absent since all the matter fields in the theory are neutral. So, the GLSM
charge matrix becomes
Q = (1
1 1 0; 0 ) ,
(5.13)
where the last entry means that the FI term is absent.

We therefore conclude that the GLSM charge matrices of the two theories are identical, but
their physical origin is different. One comes from the D-term conditions, and the other from
the F-term conditions. This phenomenon, called FD ambiguity in [20], lies at the heart of toric
duality. A crucial difference is that we can have an FI term in crystal (a), but not in crystal (b). In
the next section, we will see what difference they make when we study possible blow-ups of the
singularity.
Fig. 17. Perfect matchings of C(T 1,1 ) C and SPP C.
358
5.3. Another example

We consider another illustrative example of toric duality, namely, SPP C. We have again
two different crystals depicted in Fig. 18. For crystal (a), the field theory has seven matter fields
whose charges under the relevant U (1) gauge group are assigned as
1
Q 0
2
0
A1
0
B1
+
B2
C1
C2
+
(5.14)
The super-potential W takes the form

W = 1 A1 B1 C1 2 A1 B1 C1 + 2 B2 C2 2 B2 C2 ,
(5.15)
which gives the F-term conditions

A1 B1 C1 = B2 C2 ,
1 = 2 .
(5.16)
After solving these F-term conditions, we choose the independent matter fields to be {1 , C1 , A1 ,
B1 , C2 }. They are written in terms of perfect matchings (Fig. 17(b)) as
1 = p0 ,
C 1 = p 1 p3 ,
from which we find
0 0
1 1
T =
0 0
0 0
1 0
0
0
1
0
1
0
0
1
0
0
0
0
0
1
1
0
0
0
1
0
A1 = p4 p6 ,
1
0
0
0
0
B1 = p7 p9 ,

and QF =
p1
1
1
C 2 = p 1 p4 p7 ,
p3 p4
1 0
1 1
p6
0
1
p7
1
0
p9
1
0
(5.17)
p0
0
0

.
(5.18)
The gauge charges of matter fields are reproduced if we choose QD to be
QD = ( 0
1 0 0 1 0 0 ) .
Altogether, the GLSM charge matrix Q for the crystal (a) is given by

1 1 0 0 1 1 0 ; 0
Q = 1 1 1 1 0 0 0 ; 0 .
0 1 0 0 1 0 0 ;
Fig. 18. Two crystals for SPP C.
(5.19)
(5.20)
359
For crystal (b), the charges of eight matter fields for the Abelian theory are
X1
Q1 0
Q2
X2
0
+
X3
0
0
X4
X5
X6
+
+
X7
0
0
X8
+
0
(5.21)
We can also read off the super-potential from the four atoms in the crystal,
W = X 1 X 2 X 3 + X 4 X 5 X 6 X 7 X 8 X3 X 4 X 6 X 1 X 2 X 5 X 7 X 8 ,
(5.22)
which gives the F-term conditions

X5 X7 X8 = X 3 ,
X1 X2 = X 4 X6 .
(5.23)
With gauge invariant monomials

z1 = X5 X8 ,
z4 = X3 ,
z2 = X2 X4 X8 ,
z5 = X7 ,
z3 = X1 X5 X6 ,
w = X1 X2 (= X4 X6 ),
(5.24)
the moduli space of vacua can be described as

z1 w 2 = z2 z3 ,
z1 z5 = z4 ,
(5.25)
which is an algebraic description of SPP C singularity. In terms of perfect matchings, the

matter fields are written by
X 1 = p 1 p3 ,
X 2 = p 7 p9 ,
X 4 = p 3 p9 ,
X 5 = p4 ,
X 3 = p 4 p6 p0 ,
X 6 = p 1 p7 ,
X 7 = p0 ,
X 8 = p6 .
(5.26)
By the method introduced in the previous subsection, we can construct the GLSM charge matrix Q
p1 p3 p4 p6 p7 p9 p0
0 1 1
0 ;0
1 1 0
Q=
(5.27)
,
1 1 1 1
0
0
0 ; 1
0 1 0
0
1
0
0 ; 2
which is the same as (5.20) except that the second row in this case comes from a D-term rather
than an F-term. These two crystals are therefore toric dual.
6. Partial resolution
As we have seen in previous section, the crystal model encodes the geometry of the CY
singularity via Khler quotient (GLSM). We can take Khler deformations of the singularity by introducing the blow-up parameters a to the moment maps. The moment maps with
generic parameters would resolve the singularity completely. For special values of the parameters, some residual singularities may survive. This procedure is known as partial resolutions.
Partial resolutions in the context of AdS5 /CFT4 has been extensively discussed, for example,
in [19]. Let us recall how D-branes know about the smoothing. When a D3-brane probes the
CY singularity, the closed string modes in the twisted sector couple to open string modes as the
FI terms on the world-volume gauge theory. Turning on the FI terms induces vevs for some of the
charged matter fields, which leads to (partial) gauge symmetry breaking via Higgs mechanism.
360
Fig. 19. (a) Perfect matchings of C3 /(Z2 Z2 ) C projected onto the (x, y)-plane. All perfect matchings except p0 lie
on the plane. (b) Perfect matchings of SPP C after partial resolution, which is identical to Fig. 17(b).
Keeping light fields only, one obtains the low energy effective theory which describe the D3brane sitting at the residual singularity.
Partial resolution in the tiling model was studied most systematically in [33]. In the tiling
diagram, partial resolutions results in taking off the edges corresponding to matter fields which
acquire the vacuum expectation value. It results in another tiling diagram consistent with the
remaining daughter singularity. The most efficient method of determining which edge to
remove is to utilize the amoeba projection and the perfect matchings [33]. Here, we will
use the older approach of [20] based on the GLSM of perfect matchings, since it shows the
connection to the Abelian gauge theory more clearly. Let us briefly summarize the procedure.
We begin by choosing a vertex of the toric diagram we would like to delete. The vertex should
be at a corner of the toric diagram; otherwise, the remaining toric diagram will not be convex.
Then we consider the moment map (D-term) equation of the GLSM with all allowed FI parameters included. The M-theory origin of the FI parameters is not clear, but we proceed with the
assumption that they exist. We look for a solution to the moment map equation with a non-zero
vev for the perfect matching sitting on the vertex to be deleted. The relation between matter
fields {Xi } of the gauge theory and the perfect matchings p of the GLSM then determines
which bonds of the crystal should be removed by the partial resolution. Let us work out several
illustrative examples in detail.
6.1. C3 /(Z2 Z2 ) C SPP C
We first consider the field theory for the crystal C3 /(Z2 Z2 ) C. See Fig. 7, where the
matter fields with their charges and the super-potential are given. The perfect matchings p of
this crystal are shown in Fig. 19.
In terms of the perfect matchings p , the matter fields are written as
A1 = p 4 p 5 p 6 ,
B1 = p7 p8 p9 ,
C 1 = p 1 p2 p3 ,
1 = p0 ,
A2 = p2 p5 p8 ,
B2 = p3 p6 p9 ,
C 2 = p 1 p4 p7 ,
2 = p0 .
Following the procedure of Section 5.1, we obtain the GLSM charge matrix,
(6.1)

Q C3 /(Z2 Z2 ) C
p 1 p2 p3
0 1
1
1 1 0
= 1
0 1
1 1 0
1 1
1
0
0 1
p4
0
0
1
1
0
0
p5
0
0
0
1
0
0
p6
0
0
1
0
1
0
p7
1
1
0
0
0
1
p8
0
1
0
0
0
0
p9
1
0
0
0
0
0
p0
0
0
0
0
0
0
361
;0
;0
;0 .
;0
; 1
; 2
(6.2)
Compared with the GLSM charge matrix of d = 4, N = 1 gauge theory on a D3-brane probing
C3 /(Z2 Z2 ), the crucial difference is that our present model can have at most two FI parameters, whereas the D3-brane theory has three. In what follows, we will show that due to the lack
of FI parameters, the M2-brane theory cannot capture all possible partial resolutions allowed in
the D3-brane theory.
The toric diagram of C3 /(Z2 Z2 ) C in Fig. 19 suggests that we can remove a point, say p5 ,
to obtain a partially resolved singularity SPP C. Let us check whether such a partial resolution
can be realized in the GLSM and the Abelian theory with a suitable choice of the FI terms.
In the GLSM, we should solve the corresponding moment map equation,
Q
y = (0, 0, 0, 0, 1 , 2 )T ,
(6.3)
where y denotes |p |2 . Since the fields of the GLSM are the perfect matchings, we know what
to expect of the partial resolution. We want to remove p5 from the toric diagram, so it should get
a vev. On the other hand, the perfect matchings unaffected by the resolution, {p1 , p3 , p7 , p9 },
should not get vevs. The unique solution satisfying these requirements is
y = (0, 1 , 0, 0, 1 , 0, 0, 1 , 0, 0)T ,
(6.4)
with 2 = 0. We find that, in addition to p5 , p2 and p8 also get vevs. To summarize, we obtain
the toric diagram for SPP C by removing p2 , p5 , p8 ; see Fig. 19(a), (b).
Let us translate these results into the field theory language. The D-term equations with the FI
terms are2
|C1 |2 |C2 |2 |A1 |2 + |A2 |2 = 1 ,
|B1 |2 |B2 |2 |C1 |2 + |C2 |2 = 2 .
(6.5)
When we turn on FI term 1 only, the point we choose on the moduli space of vacua is represented
by the vev
A2 = 1 1/2
(6.6)
with the others vanishing, since the matter field A2 is written as products only of resolved perfect
matchings {p2 , p5 , p8 }. After integrating out massive modes, the low energy theory is governed
by the super-potential
W = 1 (A1 B1 C1 B2 C2 ) 2 (A1 B1 C1 B2 C2 ),
(6.7)
2 We use the same notation for the FI parameters of the GLSM and those of the Abelian gauge theory, but in general
their numerical values are not the same.
362
with one relevant U (1) gauge symmetry under which the matter fields are charged as
A1
Q 0
B1
+
C1
1
0
B2
C2
+
2
0
(6.8)
This theory is nothing but the gauge theory for crystal (a) in Fig. 18. Note that this crystal can
be obtained simply by eliminating the bond A2 in the crystal for C3 /(Z2 Z2 ) C; compare
Figs. 13 and 18(a).
6.2. SPP C C(T 1,1 ) C or C2 /Z2 C2
We now ask whether we can further resolve SPP C to reach the daughter singularities
C(T 1,1 ) C or C2 /Z2 C2 . Recall that in Section 5.3, we analyzed the field theories for the
two crystals for SPP C that are toric dual to each other. We analyze both of them in turn.
For crystal (a) in Fig. 18, let us summarize the field theory information here for convenience.
We can express the matter fields in terms of perfect matchings as
A 1 = p 4 p6 ,
B2 = p3 p6 p9 ,
B1 = p7 p9 ,
C 1 = p 1 p3 ,
C 2 = p 1 p4 p7 ,
and the GLSM charge matrix is given by
p1 p3 p4 p6
0
1 1 0
Q(SPP C) =
1 1 1 1
0 1 0
0
1 = p0 ,
2 = p0 ,
p7 p9
1 1
0
0
1
0
(6.9)
p0
0
0
0
;0
.
;0
;
(6.10)
In the GLSM, we begin again by solving the moment map equation,

Q
y = (0, 0, )T .
(6.11)
To obtain C(T 1,1 ) C, we require that p9 get a vev. It is easy to find a solution,
y = (0, 0, 0, 0, , , 0)T ,
(6.12)
C(T 1,1 ) C
implying that p7 also gets a vev. This leads to a toric diagram for
with the surviving
perfect matchings {p0 , p1 , p3 , p4 , p6 }. On the gauge theory side, the D-term equation is
|B1 |2 |B2 |2 |C1 |2 + |C2 |2 = .
(6.13)
From (6.9), we see that B1 = p7 p9 is the only field to acquire a vev,

B1 = 1/2 .
(6.14)
The low energy theory becomes the gauge theory for crystal (b) in Fig. 16. As a check, note that
this can be obtained from crystal (a) in Fig. 18 by taking off the bond B1 .3
Next, we try to obtain C2 /Z2 Z2 by giving a vev to p6 . However, a solution to (6.11) with
a non-zero vev for p6 does not exist. All other solutions of (6.11) correspond to the complete
resolution of the singularity or the partial resolution to C(T 1,1 ) C. Thus we see that unlike
the D3-brane theory, the M2-brane theory probing C3 /(Z2 Z2 ) C, does not have room for
enough FI parameters to allow partial resolution to all conceivable daughter singularities.
3 To avoid confusion, note that the coordinate axes are oriented differently in the two figures.
363
We now turn to study partial resolutions of crystal (b) in Fig. 18. We again summarize the
field theory information here for convenience. The GLSM charge matrix is
p1 p3 p4 p6 p7 p9 p0
0 1 1
0 ;0
1 1 0
Q(SPP
C) =
(6.15)
.
1 1 1 1
0
0
0 ; 1
0 1 0
0
1
0
0 ; 2
Compared to (6.10), we have an additional FI parameter. We will see that it opens up new possibilities for partial resolutions. The D-term equations become
|X3 |2 |X4 |2 |X5 |2 + |X6 |2 + |X8 |2 = 1 ,
|X1 |2 + |X2 |2 |X4 |2 + |X6 |2 = 2 .
(6.16)
The super-potential for this theory takes the form

W = X 1 X 2 X 3 + X 4 X 5 X 6 X 7 X 8 X3 X 4 X 6 X 1 X 2 X 5 X 7 X 8 .
(6.17)
In terms of perfect matchings, the matter fields can be written as

X 1 = p 1 p3 ,
X 2 = p 7 p9 ,
X 4 = p 3 p9 ,
X 5 = p4 ,
X 3 = p 4 p6 p0 ,
X 6 = p 1 p7 ,
X 7 = p0 ,
X 8 = p6 .
Solving the moment map equations with the GLSM charge matrix (6.15), we find two solutions.
The first one is
y = (0, 1 , 0, 0, 0, 2 , 0)T ,
(6.18)
with 1 = 2 (= 0). Since p3 , p9 are resolved, only the matter field X4 acquires a vev
X4 = (1 )1/2 .
(6.19)
Replacing X4 by its vev in (6.17), we find that X3 and X6 get F-term masses. In fact, integrating
out the massive modes make the super-potential vanish completely. The remaining theory with
one gauge group and five fields {X1 , X2 , X5 , X7 , X8 } is identical to that of the crystal C(T 1,1 )
C in Fig. 16(a), once we relabel the fields by
X 1 A1 ,
X2 B2 ,
X5 B1 ,
X7 C,
X 8 A2 .
(6.20)
The charges Q in (5.1) is related to Q1 and Q2 in (5.21) by Q = Q1 Q2 .

Pictorially, elimination of the bond corresponding to X4 produces a bi-valent atom which
can be regarded as a mass term. Integrating out the massive modes translates into shrinking the
bi-valent atom and the bonds attached to it. The remaining crystal coincides with Fig. 16(a).
The other solution of the moment map equation is given by
y = (0, 0, 0, 1 , 0, 0, 0)T ,
2 = 0.
(6.21)
In the toric diagram, removing p6 leads to the daughter singularity C2 /Z2 C2 . We can check
it at the level of the field theory. From (6.18), we see that X8 is the only field to acquire a vev
through p6 .
X8 = 1 1/2 .
(6.22)
364
Fig. 20. Partial resolutions and toric dualities.
The low energy theory around this vacuum is now governed by the super-potential read off from
(6.17)
W = X 1 X 2 X 3 + X4 X 5 X 6 X 7 X 3 X 4 X 6 X 1 X 2 X 5 X 7 .
(6.23)
The charges under the remaining gauge group are given by

X1
Q
X2
+
X3
0
X4
X5
0
X6
+
X7
0
(6.24)
This agrees precisely with the field theory for the crystal C2 /Z2 C2 summarized in Fig. 11, if
we relabel the fields by
X3 ,
X6 Y3 ,
X1 Y2 ,
X5 X 1 ,
X2 Y1 ,
X7 X 2 .
X4 Y4 ,
(6.25)
One can also verify that removing the bond X8 from the crystal (b) in Fig. 18 and changing the
basis for the unit cell of the crystal give the crystal in Fig. 11.
Fig. 20 summarizes all the results we have obtained so far. The solid arrows denote partial
resolutions and the dashed bi-directional arrows denote toric duality.
7. New RG flows?
In AdS5 /CFT4 , there are two famous holographic RG flows. One is the KlebanovWitten
(KW) flow [21] from the orbifold C2 /Z2 C2 to the conifold C(T 1,1 ). The other one is the
PilchWarner (PW) flow [34,35], N = 4 N = 1 , and orbifolds thereof. To our knowledge,
nearly all examples of RG flows in AdS4 /CFT3 discussed in the literature [3642] are analogous
to the PW flow rather than the KW flow.4 We will now use our crystal model to argue that new
RG flows of the KW type may exist.
4 An exception is Ref. [2], where it was argued that an RG flow of the KW type from (C2 /Z )2 to C(Q1,1,1 ) exist. It
2
does not seem to be allowed in our crystal model.
365
Fig. 21. KlebanovWitten flow in the tiling model.
7.1. KlebanovWitten flow revisited

We begin with a short review of the KW flow with emphasis on its tiling model interpretation.
The super-potential of the C2 /Z2 C orbifold is encoded in the tiling (a) in Fig. 21.
WUV = Tr(1 A1 B1 2 B1 A1 + 2 B2 A2 1 A2 B2 ).
(7.1)
The RG-flow (massive deformation) is triggered by the so-called twisted mass term,
WUV =

m 2
Tr 1 22 .
2
(7.2)
Integrating out 1 and 2 leaves the super-potential

WIR =
1
Tr(A1 B1 A2 B2 B1 A1 B2 A2 ).
m
(7.3)
Can we understand not only the end-points of the flow but the entire flow in the tiling model?
The mass term (7.2) causes a problem, because in the tiling model a matter field can appear in
a super-potential term at most once. We can circumvent this difficulty by considering a slightly
different UV theory which flows to the same theory in the IR. The new UV theory, described by
the tiling (b) in Fig. 21, has the super-potential,

W UV = Tr 1+ A1 B1 2 B1 A1 + 2+ B2 A2 1 A2 B2 Tr 1+ 1 2+ 2 .
(7.4)
Integrating out 1 and 2 leads to the same super-potential as (7.3) in the IR.
Although the new UV theory is not quite the same as the original mass-deformed theory, it is
useful for a few reasons. First, not only the theory itself, but also the process of integrating out
massive fields can be described in the tiling model. It corresponds to removing the mass terms
(bi-valent atoms) by shrinking the bonds attached to it; see Fig. 21(b), (c). Second, it can be
derived from the original, undeformed, theory in a simple and systematic way.
Recall that the inverse algorithm of the tiling model produces the tiling graph from the set
of intersecting 1-cycles (zig-zag paths). The transition from tiling (a) to (b) in Fig. 21 can be
interpreted as an extra twisting of the 1-cycles. See Fig. 22. Note that the twisting produces a
new vertex corresponding to the mass term, flips the colors of the vertices on the left half, and
exchanges the locations of A2 and B2 .
To conclude, given the tiling graph for the UV theory, the twisting interpretation offers an
intuitive and efficient way to find out the tiling graph for the IR theory.
366
Fig. 22. Twisting.
7.2. M-theory flows

We now study massive deformation of the Abelian gauge theory in the crystal model. We find
two examples which resemble the KW flow. The twisting picture we discussed in the previous
subsection again gives an intuitive picture of the flow. Whether these flows really exist in the
non-Abelian theory is a very interesting and important problem. Although a direct analysis of
the non-Abelian theory is not feasible at present, its AdS/CFT dual description on the supergravity side should be possible in a way similar to the analysis of the KW flow [43]. A detailed
study of the supergravity flows will be reported elsewhere [44].
Our first example is the flow from the orbifold C3 /(Z2 Z2 ) C to D3 . We recall from
Section 3 that the C3 /(Z2 Z2 ) C model has the charge assignment,
1
Q1 0
Q2 0
2
0
0
A1
+
A2
B1
B2
+
0
C1
0
+
C2
0
and the super-potential

W = 1 (A1 B1 C1 A2 B2 C2 ) 2 (A1 B1 C1 A2 B2 C2 ).
(7.5)
If we add the twisted mass term,

m 2
1 22
W =
(7.6)
2
and integrate out 1 and 2 , then the resulting theory has a vanishing super-potential, while the
charges of the remaining fields remain unchanged. This is nothing but the Abelian theory of D3 .
Let us check how the moduli space of vacua changes. Recall that the C3 /(Z2 Z2 ) C
orbifold has the algebraic description z1 z2 z3 = w 2 with an unconstrained variable ; see the
paragraph containing (4.4). After the RG flow, the variable disappear and the algebraic equation
is deformed to z1 z2 z3 = w1 w2 .
We can summarize the massive deformation in the crystal picture as we did for the KW flow;
see Fig. 23. Adding the twisted mass term translates to extra twistings on the 1,2 bonds. Unlike
in the KW flow, the lattice vector should be changed in order to keep the new crystal in a unit
cell.
The same story holds for our second example, dP3 C C(Q1,1,1 ). Recall from Section 3
that the dP3 C theory has the charge assignment,
1
Q1 0
Q2 0
2
0
0
A1
+
A2
+
B1
B2
C1
0
+
C2
0
+
367
Fig. 23. RG flow from C3 /(Z2 Z2 ) C to D3 .
Fig. 24. RG flow from dP3 C to C(Q1,1,1 ).
and the super-potential

W = 1 (A1 B1 C1 A2 B2 C2 ) 2 (A1 B1 C1 A2 B2 C2 ).
(7.7)
Adding the twisted mass term for 1 and 2 and integrating them out, we find a vanishing superpotential and the charge matrix of the C(Q1,1,1 ) theory. It is also straightforward to show that
the algebraic equations describing the moduli space of vacua are related; see examples 1 and 5
in Section 4.1. The flow is summarized in Fig. 24.
8. Discussion
We made an attempt to write down the world-volume theory of a single M2-brane probes
in the context of the crystal model. The resulting effective theory turned out to be an Abelian
gauge theory. In the crystal model, we identified the gauge groups and found how to read off
the charges of the matter fields. The probe theory was shown to have the moduli space of vacua
which coincides precisely with the CY4 associated to the crystal. We also discussed toric duality
and partial resolution of the probe theory. Finally, we found a hint for the existence of new RG
flows in M-theory.
One interesting observation is that the geometry probed by M2-brane looks different from that
probed by D-branes and fundamental strings. It is known that fundamental string can probe geometric as well as non-geometric phases while D-branes can probe only geometric phases [45,46].
368
The study of the Abelian gauge theory of M2-brane suggests that the phase structure is more restricted. In the Abelian gauge theory, this comes from the less available FI parameters. This also
makes the pattern of toric dualities different from that observed in D3-brane theories defined on
CY3 singularities. It is an interesting problem to figure out how such restrictions of moduli space
arise. In the Abelian gauge theory we study, we explored the mesonic branches of the underlying
theory and it would be an interesting problem to find the tools to explore the baryonic branches.
Recent works [4749] may be relevant.
Aside from the ambitious task of constructing the non-Abelian theory, there are a few directions that deserve further study. First, we hope to find an explicit description of the special
Lagrangian manifold and the merged world-volume M discussed in Sections 2 and 3, so that
we can complete the inverse algorithm and verify the behavior of the 2-cycles which give the
charges.
Exactly marginal deformations of the CFT3 is another important topic. In AdS5 /CFT4 , it is
known [50] that a generic toric CY3 with d vertices on the toric diagram admits (d 1) exactly
marginal deformations. One of them is the diagonal (complexified) gauge coupling. Another one
is the so-called -deformation, which attaches phases ei to the super-potential terms. The
other (d 3) deformations in the gauge theory are combinations of relative gauge couplings and
super-potential terms. They are interpreted as turning on BN S + iCRR along the (d 3) 2-cycles
of Y5 ; as shown in [14], H2 (Y5 , Z) = Zd3 , where Y5 is the base of the CY3 cone. In the crystal
model, there is no analog of the overall gauge coupling. Nor is there deformation due to a vev of
a field along homology cycles, because M-theory has only the 3-form field C and H3 (Y7 , Z) = 0
generically. So, the -deformation seems to be the only generic marginal deformation. On the
CFT side, it again attaches phases ei to the super-potential terms. On the supergravity side, it
corresponds to turning on the C-field along the T 2 T 3 orthogonal to the R-symmetry direction
[5153]. There is a slight puzzle here. In view of the D = 3, N = 2 supersymmetry, the parameter
is expected to be a complex number, but all the known supergravity solutions have only real
values of . It will be nice to resolve this puzzle. It will be also interesting to understand nongeneric deformations in theories with N = 4 or more symmetry using the crystal model and other
approaches such as [54].
We conclude the discussion with more speculative comments. Very recently [55], a large
class of CFT3 was constructed using Chern-Simons theory coupled to matter fields. It will be
interesting to find out whether there is any overlap between this construction and our crystal
model. Finally, the 2d dimer (tiling) model is known to have many deep connections with other
areas of physics such as the Ising model, mirror symmetry of CY3 and black-holes. See [56] and
references therein for more information. It is tempting to suspect that similar relation may exist
for the 3d dimer (crystal) model.
Acknowledgements
It is our pleasure to thank Hoil Kim and Ho-Ung Yee for useful discussions. Seok Kim is
supported in part by the KOSEF Grant R010-2003-000-10391-0. Sangmin Lee is supported by
the KOSEF Basic Research Program, Grant R01-2006-000-10965-0. Sungjay Lee is supported in
part by the Korea Research Foundation G rant R14-2003-012-01001-0. Jaemo Park is supported
by the Science Research Center Program of KOSEF through the Center for Quantum SpaceTime
(CQUeST) of Sogang University with the grant number R11-2005-021 and by the POSTECH
BSRI research fund 2007.
369
References
[1] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[2] K. Oh, R. Tatar, Three-dimensional SCFT from M2 branes at conifold singularities, JHEP 9902 (1999) 025, hepth/9810244.
[3] C. Ahn, H. Kim, Branes at C4 / singularity from toric geometry, JHEP 9904 (1999) 012, hep-th/9903181.
[4] G. DallAgata, N = 2 conformal field theories from M2-branes at conifold singularities, Phys. Lett. B 460 (1999)
79, hep-th/9904198.
[5] D. Fabbri, P. Fr, L. Gualtieri, C. Reina, A. Tomasiello, A. Zaffaroni, A. Zampa, 3D superconformal theories
from Sasakian seven-manifolds: New nontrivial evidences for AdS4 /CFT3 , Nucl. Phys. B 577 (2000) 547, hepth/9907219.
[6] C. Ahn, N = 2 SCFT and M theory on AdS4 Q(1, 1, 1), Phys. Lett. B 466 (1999) 171, hep-th/9908162.
[7] J.P. Gauntlett, D. Martelli, J. Sparks, D. Waldram, SasakiEinstein metrics on S 2 S 3 , Adv. Theor. Math. Phys. 8
(2004) 711, hep-th/0403002.
[8] J.P. Gauntlett, D. Martelli, J.F. Sparks, D. Waldram, A new infinite class of SasakiEinstein manifolds, Adv. Theor.
Math. Phys. 8 (2006) 987, hep-th/0403038.
[9] M. Cvetic, H. Lu, D.N. Page, C.N. Pope, New EinsteinSasaki spaces in five and higher dimensions, Phys. Rev.
Lett. 95 (2005) 071101, hep-th/0504225.
[10] S. Lee, Superconformal field theories from crystal lattices, Phys. Rev. D 75 (2007) 101901, hep-th/0610204.
[11] S. Lee, S. Lee, J. Park, Toric AdS4 /CFT3 duals and M-theory crystals, JHEP 0705 (2007) 004, hep-th/0702120.
[12] A. Hanany, K.D. Kennaway, Dimer models and toric diagrams, hep-th/0503149.
[13] S. Franco, A. Hanany, K.D. Kennaway, D. Vegh, B. Wecht, Brane dimers and quiver gauge theories, JHEP 0601
(2006) 096, hep-th/0504110.
[14] S. Franco, A. Hanany, D. Martelli, J. Sparks, D. Vegh, B. Wecht, Gauge theories from toric geometry and brane
tilings, JHEP 0601 (2006) 128, hep-th/0505211.
[15] A. Hanany, D. Vegh, Quivers, tilings, branes and rhombi, hep-th/0511063.
[16] B. Feng, Y.H. He, K.D. Kennaway, C. Vafa, Dimer models from mirror symmetry and quivering amoebae, hepth/0511287.
[17] Y. Imamura, Global symmetries and t Hooft anomalies in brane tilings, JHEP 0612 (2006) 041, hep-th/0609163.
[18] M.R. Douglas, B.R. Greene, D.R. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506 (1997) 84, hepth/9704151.
[19] D.R. Morrison, M.R. Plesser, Non-spherical horizons. I, Adv. Theor. Math. Phys. 3 (1999) 1, hep-th/9810201.
[20] B. Feng, A. Hanany, Y.H. He, D-brane gauge theories from toric singularities and toric duality, Nucl. Phys. B 595
(2001) 165, hep-th/0003085.
[21] I.R. Klebanov, E. Witten, Superconformal field theory on threebranes at a CalabiYau singularity, Nucl. Phys. B 536
(1998) 199, hep-th/9807080.
[22] D. Martelli, J. Sparks, S.T. Yau, The geometric dual of a-maximisation for toric SasakiEinstein manifolds, Commun. Math. Phys. 268 (2006) 39, hep-th/0503183.
[23] D. Martelli, J. Sparks, S.T. Yau, SasakiEinstein manifolds and volume minimisation, hep-th/0603021.
[24] S. Benvenuti, B. Feng, A. Hanany, Y.H. He, Counting BPS operators in gauge theories: Quivers, syzygies and
plethystics, hep-th/0608050.
[25] M. Porrati, A. Zaffaroni, M-theory origin of mirror symmetry in three-dimensional gauge theories, Nucl. Phys.
B 490 (1997) 107, hep-th/9611201.
[26] M.R. Douglas, G.W. Moore, D-branes, Quivers, and ALE instantons, hep-th/9603167.
[27] P.B. Kronheimer, The Construction of ALE spaces as hyperKhler quotients, J. Diff. Geom. 29 (1989) 665.
[28] N. Seiberg, E. Witten, Gauge dynamics and compactification to three dimensions, hep-th/9607163.
[29] K.A. Intriligator, N. Seiberg, Mirror symmetry in three-dimensional gauge theories, Phys. Lett. B 387 (1996) 513,
hep-th/9607207.
[30] M. Billo, D. Fabbri, P. Fre, P. Merlatti, A. Zaffaroni, Rings of short N = 3 superfields in three dimensions and
M-theory on AdS4 N (0, 1, 0), Class. Quantum Grav. 18 (2001) 1269, hep-th/0005219.
[31] K.M. Lee, H.U. Yee, New AdS4 X7 geometries with N = 6 in M theory, JHEP 0703 (2007) 012, hep-th/0605214.
[32] H.U. Yee, AdS/CFT with tri-Sasakian manifolds, hep-th/0612002.
[33] I. Garcia-Etxebarria, F. Saad, A.M. Uranga, Quiver gauge theories at resolved and deformed singularities using
dimers, JHEP 0606 (2006) 055, hep-th/0603108.
370
[34] A. Khavaev, K. Pilch, N.P. Warner, New vacua of gauged N = 8 supergravity in five dimensions, Phys. Lett. B 487
(2000) 14, hep-th/9812035.
[35] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from holography supersymmetry
and a c-theorem, Adv. Theor. Math. Phys. 3 (1999) 363, hep-th/9904017.
[36] C. Ahn, S.J. Rey, Three-dimensional CFTs and RG flow from squashing M2-brane horizon, Nucl. Phys. B 565
(2000) 210, hep-th/9908110.
[37] C. Ahn, S.J. Rey, More CFTs and RG flows from deforming M2/M5-brane horizon, Nucl. Phys. B 572 (2000) 188,
hep-th/9911199.
[38] C. Ahn, J. Paeng, Three-dimensional SCFTs, supersymmetric domain wall and renormalization group flow, Nucl.
Phys. B 595 (2001) 119, hep-th/0008065.
[39] R. Corrado, K. Pilch, N.P. Warner, An N = 2 supersymmetric membrane flow, Nucl. Phys. B 629 (2002) 74, hepth/0107220.
[40] C. Ahn, K. Woo, Domain wall and membrane flow from other gauged d = 4, N = 8 supergravity. I, Nucl. Phys.
B 634 (2002) 141, hep-th/0109010.
[41] C. Ahn, K. Woo, Domain wall from gauged d = 4, N = 8 supergravity. II, JHEP 0311 (2003) 014, hep-th/0209128.
[42] D. Nemeschansky, N.P. Warner, A family of M-theory flows with four supersymmetries, hep-th/0403006.
[43] N. Halmagyi, K. Pilch, C. Romelsberger, N.P. Warner, The complex geometry of holographic flows of quiver gauge
theories, JHEP 0609 (2006) 063, hep-th/0406147.
[44] S. Kim, S. Lee, S. Lee, J. Park, in preparation.
[45] M. Douglas, B. Greene, D. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506 (1997) 84, hep-th/
9704151.
[46] E. Witten, Phase transitions in M-theory and F-theory, Nucl. Phys. B 471 (1996) 195, hep-th/9603150.
[47] A. Butti, D. Forcella, A. Zaffaroni, Counting BPS baryonic operators in CFTs with SasakiEinstein duals, hep-th/
0611229.
[48] D. Forcella, A. Hanany, A. Zaffaroni, Baryonic generating functions, hep-th/0701236.
[49] A. Butti, D. Forcella, A. Hanany, D. Vegh, A. Zaffaroni, Counting chiral operators in quiver gauge theories, arXiv:
0705.2771 [hep-th].
[50] Y. Imamura, H. Isono, K. Kimura, M. Yamazaki, Exactly marginal deformations of quiver gauge theories as seen
from brane tilings, hep-th/0702049.
[51] O. Lunin, J.M. Maldacena, Deforming field theories with U (1) U (1) global symmetry and their gravity duals,
JHEP 0505 (2005) 033, hep-th/0502086.
[52] C. Ahn, J.F. Vazquez-Poritz, Marginal deformations with U (1)3 global symmetry, JHEP 0507 (2005) 032, hepth/0505168.
[53] J.P. Gauntlett, S. Lee, T. Mateos, D. Waldram, Marginal deformations of field theories with AdS4 duals, JHEP 0508
(2005) 030, hep-th/0505207.
[54] B. Kol, On conformal deformations, JHEP 0209 (2002) 046, hep-th/0205141.
[55] D. Gaiotto, X. Yin, Notes on superconformal ChernSimons-matter theories, arXiv: 0704.3740 [hep-th].
[56] R. Dijkgraaf, D. Orlando, S. Reffert, Dimer models, free fermions and super quantum mechanics, arXiv: 0705.1645
[hep-th].
NUCLEAR PHYSICS B
Journal devoted to the experimental and theoretical study of the fundamental
constituents of matter and their interactions
Editorial Board
Supervisory Editors: G. ALTARELLI, Geneva, Switzerland; W. BARTEL, Hamburg, Germany; C. BECCHI, Genova,
Italy; R.H. DIJKGRAAF, Amsterdam, The Netherlands; D. KUTASOV, Chicago, IL, USA; L. MAIANI, Rome, Italy;
H. MURAYAMA, Berkeley, CA, USA; T. NAKADA, Geneva, Switzerland; N. NEKRASOV, Bures-sur-Yvette, France;
H. OOGURI, CalTech, Pasadena, CA, USA; H. SALEUR, USC, Los Angeles, CA, USA; A. SCHWIMMER, Rehovot,
Israel
Associate Editors for Particle Physics: S.J. Brodsky, SLAC; M. Dine, UC, Santa Cruz; G. Dvali, New York; G.W.
Gibbons, Cambridge, UK; D. Gross, St. Barbara; S.S. Gubser, Princeton; L. Ibanez, Madrid; R.L. Jaffe, MIT; K. Kajantie,
Helsinki; M. Luescher, CERN; E. Rabinovici, Jerusalem; L. Randall, Princeton; A. Sen, Bombay; G. Steigman, Ohio
State; G. t Hooft, Utrecht; Y. Totsuka, KEK; P. van Baal, Leiden; S. Weinberg, Texas; J. Zinn-Justin, Saclay
Associate Editors for Field Theory and Statistical Systems: B. Duplantier, Saclay; A.W.W. Ludwig, Santa Barbara;
S. Lukyanov, Rutgers, Piscataway; T. Miwa, Kyoto; G. Mussardo, Trieste; M. Oshikawa, Tokyo; G. Parisi, Frascati;
A.M. Polyakov, Princeton; R. Sasaki, Kyoto; M. Zirnbauer, Cologne
Aims and Scope
Nuclear Physics B focuses on the domain of high energy physics and quantum field theory, and includes three main sections, i.e. particle physics, field theory and statistical systems and physical mathematics. The emphasis is on original
research papers. Conference Proceedings on the topics covered by Nuclear Physics B are published in the (separately
available) journal Nuclear Physics B-Proceedings Supplements.
Abstracted/Indexed in:
Chemical Abstracts/Current Contents: Physical, Chemical & Earth Sciences. Also covered in the abstract and citation
database SCOPUS . Full text available on ScienceDirect .
Subscription Information 2008
Volumes 788805 are scheduled for publication in 54 issues. Publication frequency: three times a month. Subscription
prices are available upon request from the Publisher or from the Regional Sales Office nearest you or from this journals
website (http://www.elsevier.com/locate/nuclphysb). A combined subscription to Nuclear Physics A ISSN 0375-9474
(volumes 798814), Nuclear Physics B ISSN 0550-3213 (volumes 788805) and Nuclear Physics B Proceedings Supplements ISSN 0920-5632 (volumes 174185) is available at a reduced rate. Subscriptions are accepted on a prepaid
basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery
outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months
of the date of dispatch.
Orders, claims, and journal enquiries: please contact the Customer Service Department at the Regional Sales Office
nearest you:
Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive, Orlando, FL 32887-4800, USA; phone: (+1)
(877) 8397126 [toll free number for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
3631354; e-mail: usjcs@elsevier.com
Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE Amsterdam, The Netherlands; phone:
(+31) (20) 4853757; fax: (+31) (20) 4853432; e-mail: nlinfo-f@elsevier.com
Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku,
Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519;
phone: (+65) 63490222; fax: (+65) 67331510; e-mail: asiainfo@elsevier.com
Advertising information
Europe, USA, Canada and ROW: Advertising orders and enquiries can be sent to: Miss Katrina Barton, phone: (+44)
(0) 20 7611 4117; fax: (+44) (0) 20 7611 4463; e-mail: commercialsales@elsevier.com.
South America: Advertising orders and enquiries can be sent to: Mr Tino DeCarlo, The Advertising Department,
Elsevier Inc., 360 Park Avenue South, New York, NY 10010-1710, USA; phone: (+1) (212) 633 3815; fax: (+1) (212)
633 3820; e-mail: t.decarlo@elsevier.com.
US mailing notice Nuclear Physics B (ISSN 0550-3213) is published weekly by Elsevier B.V., P.O. Box 211, 1000
AE Amsterdam, The Netherlands. The annual institutional subscription price in the USA is US$ 13612 (valid in North,
Central and South America), including air speed delivery. Periodical postage paid at Rahway NJ and additional mailing
offices. USA Postmaster: Send change of address: Nuclear Physics B, Elsevier, 6277 Sea Harbor Drive, Orlando, FL
32887-4800. Airfreight and mailing in the USA by Mercury International Limited, 365, Blair Road, Avenel, NJ 07001.
Nuclear Physics B 797 (2008) iii

Publishers Note
We are pleased to announce a new Editor to Nuclear Physics B: Nikita Nekrasov.
Nikita Nekrasovs research covers a broad range of questions in modern mathematical physics,
motivated by the high energy physics and the search for the quantum theory of gravity. He has
obtained an exact solution of the low-energy effective theory with N = 2 super-symmetry in four
dimensions (SeibergWitten theory), using direct instanton calculations, found new and unexpected instanton, monopole and fluxon solutions of noncommutative gauge theories, constructed
gravity and string theory duals of the renormalization group flows of various four-dimensional
theories, found the string theory realization of the quantum space foam, related many-body interacting systems and gauge theories.
N. Nekrasov
IHS, 35, Route de Chartres,
F-91440 Bures-sur-Yvette, France
Email: nikitanpb@gmail.com
Graduated in 1994 from Moscow Physical-Technical Institute
Got PhD. in 1996 from Princeton University
Harvard Society of Fellows 1996-1999
IHS permanent professor since 2000.
The publisher welcomes the new Editor to the Board, and wishes him every success and
enjoyment in his new role.
0550-3213/2008 Published by Elsevier B.V.

doi:10.1016/S0550-3213(08)00188-0
Nuclear Physics B 797 [PM] (2008) 371394

The generalized curvature and Christoffel symbols

for a higher spin potential in AdSd+1 space
Ruben Manvelyan a,b , Werner Rhl a,
a Department of Physics, Erwin Schrdinger Strae, Technical University of Kaiserslautern, Postfach 3049,
67653 Kaiserslautern, Germany

b Yerevan Physics Institute, Alikhanian Br. Str. 2, 0036 Yerevan, Armenia
Received 22 August 2007; accepted 7 October 2007

Available online 17 October 2007
Abstract
The generalized curvature tensor and Christoffel symbols are determined in AdSd+1 background by a
modified ansatz of the de WitFreedman type by imposing gauge invariance. The resulting set of recurrence
relations and difference equations is solved. The Riemann curvature tensor is derived by antisymmetrization. All results are presented as finite power series in the inverse AdS radius and are unique. The fourth
order, which is complete for fields up to spin five, is calculated explicitly. Higher orders can be obtained
with the same method.
PACS: 11.10.-z
Keywords: Higher spin fields; Curvature; Riemann tensor; Affine structures; Cristoffel symbols
0. Introduction
In this paper we present a recursive procedure to construct generalized Christoffel symbols
and the curvature for higher spin (HS) gauge potential in AdSd+1 background. The linearized HS
gauge fields geometry on flat space was constructed by B. de Wit and D.Z. Freedman in the seminal article [1] and for spin s = 3 in [2]. The authors presented there a very elegant geometrical
hierarchy of generalized Christoffel symbols defined from the gauge transformation properties
E-mail addresses: manvel@physik.uni-kl.de (R. Manvelyan), ruehl@physik.uni-kl.de (W. Rhl).

doi:10.1016/j.nuclphysb.2007.10.012
372
R. Manvelyan, W. Rhl / Nuclear Physics B 797 [PM] (2008) 371394
in a flat background. This construction is still very important and interesting for many reasons.
First of all this hierarchy is a full geometrical partner of Fronsdals linearized dynamics [3] for
HS gauge fields in the case of flat space and puts this theory on the same level as linearized gravity. From another hand this hierarchy should be intimately connected with the generalized Weyl
tensors introduced in [4] and with the recent investigations of the so-called unfolded formulation
for higher spin theories (see [5] and references therein) which is pretending today to describe the
nonlinear interaction for HS fields in a general background. The AdS background for HS fields
was always necessary to construct a consistent HS interaction with gravity [6] but became especially important after discovery of the AdS4 /CFT 3 correspondence of the critical O(N) sigma
model and four-dimensional HS gauge theory in anti-de Sitter space [7]. This proposal increased
the interest in the general problems of the quantum interacting HS theories in AdS space.
Our interest in the de WitFreedman construction in AdS background originates from our last
results for the trace anomaly of the conformal scalar in the external HS gauge field in the bulk
of AdS4 obtained in [8] and [9]. Our main task is to express and explain our one-loop quantum
results in terms of geometrical objects like the generalized Weyl tensor and the generalized Euler
density. This we hope to materialize in the next publication applying the results obtained in this
article as a first but sufficiently complicated and nontrivial technical step. So here we present only
the unique construction for the HS curvature and Christoffel symbols in AdSd+1 background.
The aim is to construct the generalized curvature and Christoffel symbols linearized in the
HS field from an ansatz presenting a 1/L2 polynomial expansion with L the AdS radius, whose
leading term corresponding to L = is the de WitFreedman expression [1]. The terms of
(l)
higher order k in 1/L2 contain numerical coefficients Ar1 r2 r3 , r1 + r2 + r3 = k which we intend to
determine by imposing gauge invariance. This necessitates the derivation of a system of equations
which is linear in the As. This system is based on a matrix of triangular shape with blocks of
difference operators of first order on the diagonal. Solving this system of equations is the main
part of our analysis. At the order k = 2 the essential problems of this issue all occur and we solved
them. So we are convinced that our method works to any order up to the highest, k = [s/2], where
s is the spin of the HS field.
In Section 1 we review and reformulate in our compressed notations the flat space definitions
of [1] and correct the important formula obtained in [10] deriving the corresponding normalization factor in Appendix A. In Section 2 we develop the Lie algebra of covariant derivatives
on symmetric tensor fields in AdS space obtaining important commutation formulas. These formulas we apply in Section 3 to calculate the gauge variation of the modified de WitFreedman
type ansatz in all orders of the AdS curvature and to derive the matrix and the inhomogeneity
of the system of equations. In Section 4 we develop the concept of obstruction and present a
full solution for the recursion relations to first and second order (Appendix B) in the polynomial expansion in powers of the AdS inverse radius squared. In Section 5 we derive the Riemann
curvature tensor from the de WitFreedman expression by applying an antisymmetrizer which
renders the Riemann tensor the appropriate symmetry of a Young diagram with two rows of
equal length s. This method has been presented in Section 1 for the flat space. The whole article
is written in a mathematical language which represents symmetric tensor fields as homogeneous
polynomials in vectors of the tangent space of degree equal the spin of the field. This elementary language was always used in quantum mechanics (e.g., for electromagnetic multipoles) and
in the last thirty years commonly used in conformal field theory for spaces of arbitrary dimensions.
373
1. The basic definitions on flat space

In analogy with the electromagnetic potential A of spin one which possesses the field tensor
F as a gauge invariant curvature, we consider higher spin potentials h(s) as symmetric tensors
of rank s and their gauge invariant curvature R of tensor rank 2s. This curvature is the simplest
gauge invariant object that is linear in the potential, is obtained from the potential by acting on
it with a differential operator, but does not vanish if the free field equations are applied. These
equations are
h(s)
1 2 ...s4 = 0,
(1.1)
which is the postulate of vanishing double trace, and a second order field equation. To formulate
it we introduce the concise notation of homogeneous functions for symmetric tensors, namely
we contract these tensors with vectors from the tangent space denoted a, b and similarly
s

(s)
i
h (z; a) =
(1.2)
a
h(s)
1 2 ...s (z).
i=1
Then Fronsdals [3] wave equation is

1
2 (a)(a ) + (a)2 2a hs (z; a) = 0.
2
(1.3)
Potentials h(s) admit gauge transformations that leave the wave equations invariant. The gauge
functions (s1) (z; a) are assumed to be traceless
2a (s1) (z; a) = 0,
(1.4)
and the (classical) gauge transformation is

h(s) (z; a) h(s) (z; a) + (a) (s1) (z; a).
(1.5)
In the sequel we will not refer to any wave equation.

In [10] the curvature in flat space was defined by
R(s)1 1 ,2 2 ,...,s s (z) =
s

(s)
gii gii gii gii 1 2 s h1 2 ...s (z),
(1.6)
i=1
which generalizes corresponding expressions for the electromagnetic case A = h(1) of spin one
and the gravitational case g = h(2) of spin two to arbitrary spin. It is obvious that R is antisymmetric under the exchange inside a single pair
(i , i ) (i , i ),
(1.7)
and symmetric under the exchange of two such pairs (i different from j )
(i , i ) (j , j ).
(1.8)
Moreover the curvature R is cyclic in the sense

R1 1 ,2 2 ,3 ...s + R1 2 ,1 2 ,3 ...s + R2 1 ,1 2 ,3 ...s = 0.
(1.9)
Further relations follow from the wave equations which we are not interested in this context.
374
The de WitFreedman curvature (s) of h(s) can be derived from R (s) (1.6) by [1]
1 (s)
R1 P (1 ),2 P (2 ),...,P (s ) (z),
s!
(s)
=
1 2 ,...,s ;1 2 ,...,s
(1.10)
P ()
where the summation is over all permutations P () of the labels . Thus is totally symmetric
in each of the s-tupels of labels. From now on we use the harmonic polynomial representation
(s) (z; a, b) =
s

(s)
a i bi 1 2 ...s ;1 2 ...s
i=1
(s)

a i bi R(s)1 1 ,2 2 ,...,s (z).
(1.11)
i=1
This relation can in fact be inverted so that R and contain the same information. This inversion formula has been written down first in [10] without proof (and with a wrong normalization
factor). Therefore we will prove it here and derive the normalization factor in Appendix A.
Consider the Young diagram with two rows, each of length s. The symmetrizer of either row
is denoted S and the antisymmetrizer of the s columns is denoted A. Then AS is the symmetrizer
of the diagram. It is idempotent,
(AS)2 = cs AS,
(1.12)
where we leave the normalization cs free. Denoting the s-fold derivative of h(s) in (1.6) by (s) ,
we have
R (s) = A (s) = AS (s) .
(1.13)
On the other hand was defined by

(s) = SR (s) .
(1.14)
It follows that
A (s) = ASR (s) = cs R (s) .
(1.15)
In Appendix A it is shown that with our normalizations of S and A the constant is

cs = 1,
(1.16)
so that the desired inversion formula is

R (s) = A (s) ,
(1.17)
or
R(s)1 1 ,...,s s (z) =
s

(s)
gii gii gii gii 1 2 ...s ;1 2 ...s (z).
(1.18)
i=1
This completes the proof.

Now we aim at a reformulation of the de WitFreedman curvature in homogeneous polynomial form (1.11)
(s) (z; a, b) =
375
s

(s)
a i bi bi a i 1 2 . . . s h1 2 ...s (z)
i=1

s

s
=
(1)sl
(a)l (b)sl h(s)
sl,l (z; a, b),
l
(1.19)
l=0
where s l, l are the degrees of homogeneity in a respectively b.

This can be simplified using
l (s)
hsl,l (z; a, b) = n1
l (ba ) h (z; a),
(s)
(1.20)
where
nl = s(s 1)(s 2) (s l + 1),
(1.21)
so that
(s)
(z; b, a) =
s

(1)l
l!
l=0
(a)l (b)sl (ba )l h(s) (z; a).
(1.22)
We use this expansion to prove gauge invariance of (s) . This proof is needed in the sequel.
Proving gauge invariance of R (s) is trivial.
In fact the gauge variation is
(s) (z; b, a) =
s

(1)l
l=0
l!
(a)l (b)sl (ba )l (a) (s1) (z; a),
(1.23)
where
(ba )l (a) (s1) (z; a) = (a)(ba )l (s1) (z; a) + l(b)(ba )l1 (s1) (z; a).
(1.24)
Inserting this into (1.23) and substituting l by l + 1 in the second term gives zero.
Thus we generalize our ansatz for a curvature on AdSd+1 as
(s) (z; b, a) =
L2k k (z; b, a),

(s)
(1.25)
k=0
including finitely many 1/L2 correction terms only and the leading term at k = 0
(s)
0 (z; b, a) =
s

(1)l
l=0
l!
(a)l (b)sl (ba )l h(s) (z; a),
(1.26)
which is obtained from (1.22) by replacing space derivatives by covariant derivatives and fixing
their order (another order could also do it). In an analogous fashion and following the seminal
article [3] we make an ansatz for generalized Christoffel symbols by a polynomial expression in
L2 and for all 1 m s 1

L2k k(m,s) (z; b, a),
(m,s) (z; b, a) =
(1.27)
(m,s)
(z; b, a) =
k0
m

l=0
(1)l
(a)l (b)ml (ba )l h(s) (z; a),
l!
(1.28)
376
(s,s) (z; b, a) = (s) (z; b, a).
(1.29)
(m,s)
We use as ansatz the following representation for k

(m,s)
k
mr
2 2r3

r1 ,r2 ,r3 l=2r1 +r2
including the case m = s
r
(1)l (l) 2 r1
(ab)r2 b2 3
Ar 1 r 2 r 3 a
l!
(a)l2r1 r2 (b)mlr2 2r3 (ba )l h(s) (z; a)

(ri 0, r1 + r2 + r3 = k).
(1.30)
The main motivation being that in AdS length scale L, a, b have scale +1 whereas has scale
1. So a power L2k in the denominator necessitates an invariant compensator in the numerator.
The covariant derivative cannot be used for this purpose.
The coefficients A in (1.30) must be determined by requiring gauge invariance with respect to
transformations (1.5). In the case of the curvature (s) the requirement of gauge invariance
(s) (z; b, a) = 0,
(1.31)
and in the case of the Christoffel symbols (i.e., for m < s) tracelessness with respect to b of the
gauge variation
Trb (m,s) (z; b, a) = 0
(1.32)
are postulated. In the flat case and for m = 2 the second order Fronsdal equation of motion for a
spin s gauge field results from (see [2,3])
Trb (2,s) (z; b, a) = 0,
(1.33)
which is gauge invariant (only in the flat case) due to

(1)m
(a)m+1 (ba )m (s1) ,
m!
(1)m
Trb (m,s) (z; b, a) =
(a)m+1 (b)m2 Tra (s1) (z; a) = 0.
(m 2)!
(m,s) (z; b, a) =
(1.34)
(1.35)
These requirements can be transformed into recursive equations of k(m,s) , k(s) . The derivation and solution of these is the topics of this article. Though the task seems different in both
cases, we can treat it in parallel a long time. We will display the mathematical apparatus for an
elegant treatment of these problems first.
2. Differential algebra on symmetric tensors
In this section we develop the Lie algebraic algorithm of covariant derivatives acting on symmetric higher spin fields on anti-de Sitter spaces. Application of gradients to these fields produce
tensor fields which are symmetric in two different sets of labels, we call them bisymmetric
tensors trs (z; a, b). These are homogeneous polynomials in a of degree r and in b of degree s.
The vectors a, b belong to the tangential space at z, T AdS(z). The linear space of such C fields
over AdS is denoted Trs (a, b) and their union is

(2.1)
Trs (a, b) = T (a, b).
r,s
377
On such space T (a, b) we can act with the differential operator algebra which in the sequel will
be our main tool.
The first class consists of a-gradients, b-gradients, a-divergences, b-divergences etc., e.g.,
a-gradient:
a-divergence:
(a)trs (z; a, b) Tr+1,s (a, b),

(a )trs (z; a, b) Tr1,s (a, b).
(2.2)
(2.3)
Another class of operators consists of purely tensorial operators such as

(ab )trs (z; a, b) Tr+1,s1 (a, b),
(2.4)
or the Euler operators (aa ) and (bb ).

On curved spaces the commutator of two gradients

1 = (a), (b) = 1 ,
(2.5)
is of great importance and, for AdS spaces of radius L, assumes the simple form1

1 2
b (ab ) a 2 (ba ) (ab) (bb ) (aa ) .
(2.6)
L2
In the same fashion we define

1
2 = (a), 1 = 2 a 2 (b) (ab)(a) ,
(2.7)
L

1
2 = (b), 1 = 2 b2 (a) (ab)(b)
(2.8)
L
so that by replacing a by b and vice versa, k and k exchange their roles. Define higher order
commutators by

(a), k = k+1 ,
(2.9)

(b), k = k+1 ,
(2.10)
1 =
then we can easily see that a repetitive structure arises
2 n
a
2n+1 =
1 ,
L2
2 n1
a
2n =
2 ,
L2
(2.11)
(2.12)
1 This definition of the commutator is in agreement with the following conventions for the Euclidean AdS
d+1 metric
and curvature
ds 2 = g (z) dz dz =
L2
dz dz ,
(z0 )2
g=
Ld+1
,
(z0 )d+1
[ , ]V = R V R V ,
1
1

R = 0 2 = 2 g (z) g (z) ,
(z )
L
d
d
d(d + 1)
.
R = 0 2 = 2 g (z), R =
(z )
L
L2
378

b2 n
1 ,
L2
2 n1
b
2 .
2n =
L2
2n+1 =
(2.13)
(2.14)
Next we expand
n

n
Cnk
(b)nk k ,
(b)n , (a) =
(2.15)
k=1
n

n
(b)n , 1 =
Dnk
(b)nk k+1 ,

(b)n , 2 =
k=1
n

(2.16)
n
Enk
(b)nk k+2 .
(2.17)
k=1
The summation starts with k = 1 in each case because by commutation the power of (b)n is
lowered at least by one. By Jacobis identity we obtain immediately for k > 0
n
n
n
Cnk
= Dnk
= Enk
.
(2.18)
A recursion relation is obtained from

(b)n+1 , (a) = (b) (b)n , (a) + (b), (a) (b)n ,
which is solved by

n
n
k,0 (1)k1 .
Cnk =
k
(2.19)
(2.20)
It follows finally that

(b)n , (a) =
[ n+1 ]
2 k1
n
n2k+1 b
(b)
1
2k 1
L2
k=1
[n]

2 k1
2

n
n2k b
2 .
(b)
2k
L2
(2.21)
k=1
(s)
This formula is the basis for the recursion relations for k .

3. The gauge variation of (m,s)
(m,s)
3.1. The gauge variation of 0
In order to derive equations from the gauge invariance postulate, according to which under
h(s) (z; a) h(s) (z; a) + (a) (s1) (z; a),
(m,s)
(3.1)
(m,s)
is invariant, we have to make sufficiently detailed studies of the gauge variations of 0

(m,s)
and k
, k 1. The first object gives the inhomogeneous terms in the linear system of equations, the second objects give the matrix (linear operator) acting on the unknown coefficients A.
379
In the sequel we will call a term level k, if its differential operators are canonically ordered
according to (a)r (b)s (ba )t and its coefficient is O(L2k ). The levels define an ordering
scheme for the curvature and its gauge variation. We assume m arbitrary s.
(m,s)
(1.28) and decompose its gauge variation into contributions of the difWe start from 0
ferent levels. From (1.28) we have
(m,s)
m

(1)l
l=0
l!
(a)l (b)ml (ba )l (a) (s1) (z; a),
(3.2)
which can, following the treatment of the flat case, be brought into the form
(m,s)
m

(1)l
l=0
l!

(a)l (b)ml , (a) (ba )l (s1) (z; a)
(1)m
(3.3)
(a)m+1 (ba )m (s1) (z; a).
m!
The last term vanishes for m = s or after taking a b-trace. We can neglect it for both the curvature
and the Christoffel symbols.
Using (2.21) the commutator can be evaluated as
2 k1

m

(1)l
ml
(m,s)
l b
=
(a)
(b)ml2k+1 1
0
l!
2k 1
L2
k>0 l=0

ml
ml2k
(b)
2 (ba )l (s1) .
(3.4)
2k
+
From (2.8) we recognize that the 2 term is not canonically ordered. Performing the ordering we
obtain

(b)ml2k 2 = L2 b2 (a)(b)ml2k (ab)(b)ml2k+1

b2
(b)ml2k , (a) .
(3.5)
2
L
The first term (1 term) in the square bracket of (3.4) contributes only to the level k but the
second term (2 term) gives rise to infinitely many terms of level larger than k.
Consider now the 1 term of (3.4). From (2.6) we obtain

1 (ba )l (s1) = L2 a 2 (ba )l+1 + (2l s + 1)(ab)(ba )l

l(s l)b2 (ba )l1 (s1) .
(3.6)
+
We cast the terms of level k resulting into the general form

L2k
m

r
(1)l (k,l) 2 r1
(ab)r2 b2 3 (a)l2r1 r2 +1
Qr1 r2 r3 a
l!
r r r
l=0
(b)
1 2 3
mlr2 2r3
(ba )l (s1) ,
(3.7)
and the ri are submitted to

3

i=1
ri = k.
(3.8)
380
The only nonvanishing coefficients Q follow from (3.5), (3.6)

ml1
ml
(k,l)
Q00,k =
(s l 1)
,
2k 1
2k

ml
ml
(k,l)
Q01,k1 =
(2l s + 1) +
,
2k 1
2k

ml+1
(k,l)
Q10,k1 =
l.
2k 1
(3.9)
(3.10)
(3.11)
In the sequel we will replace k by k1 , eventually.

Now we consider the commutator term in (3.5) applying (2.21) again. Introducing a new label
k2 we obtain the expansion

2 k1 +k2 1
m

(1)l+1
ml
b
(a)l
l!
2k1
L2
k1 ,k2 1 l=0

m l 2k1
(b)ml2k1 2k2 +1 1
2k2 1

m l 2k1
(b)ml2k1 2k2 2 (ba )l (s1) .

2k2
(3.12)
On the level k1 + k2 remains

L2(k1 +k2 )
(1)l
l
(b)
l!
mlr2 2r3

1 ,k2 ,l) a 2 r1 (ab)r2 b 2 r3 (a)l2r1 r2 +1
Q(k
r1 r 2 r 3
ri
(ba )l (s1) ,
where the only nonvanishing coefficients are
m l 1 m l 2k1 1
(k1 ,k2 ,l)
Q00,k1 +k2 =
(s l 1)
2k2 1
2k1

m l m l 2k1
+
,
2k1
2k2
and
(3.13)
(3.14)
(k1 ,k2 ,l)

Q01,k
1 +k2 1
m l m l 2k1
=
(2l s + 1)
2k1
2k2 1

m l m l 2k1
,
2k1
2k2
(3.15)
and finally
(k1 ,k2 ,l)

Q10,k
1 +k2 1
m l + 1 m l 2k1 + 1
=+
l.
2k2 1
2k1
(3.16)
We notice that all products of two binomial coefficients can be expressed by multinomial coefficients.
381
Continuing this procedure we obtain the contribution at level K by summing over all partitions
{k1 , k2 , k3 , . . . , kn } for which
ki 1,
ki = K.
(3.17)
i=1
For all such partitions the contribution is then

2K2
m

(1)l+n1
ml
b
(a)l
L
l!
2k1 , 2k2 , . . . , 2kn 1, m l 2K + 1

ki 1, ki =K l=0

ml
ml2K+1
ml2K
(b)
1
2
(b)
2k1 , 2k2 , . . . , 2kn , m l 2K
(ba )l (s1) ,
(3.18)
where we used multinomial coefficients. Neglecting the commutator in (3.18) we obtain in terms
(k ,k ,...,kn ,l)
for a fixed partition
of coefficients Qr1 1r2 r32
L2K
m

(1)l
l!
(k ,k ,...,kn ,l) 2 r1
Q(r11r2 r23 )
l=0
mlr2 2r3
(b)
r
(ab)r2 b2 3 (a)l2r1 r2 +1
(ba )l (s1) ,
(3.19)
where the only nonvanishing coefficients are for fixed l and fixed partition

ml1
(k1 ,k2 ,...,kn ,l)
= (1)n
(s l 1)
Q00,K
2k1 , 2k2 , . . . , 2kn1 , 2kn 1, m l 2K

ml
(3.20)
,
+
2k1 , 2k2 , . . . , 2kn , m l 2K

ml
(k1 ,k2 ,...,kn ,l)
Q01,K1
= (1)n+1
(2l s + 1)
2k1 , 2k2 , . . . , 2kn1 , 2kn 1, m l 2K + 1

ml
(3.21)
,
+
2k1 , 2k2 , . . . , 2kn , m l 2K

ml+1
(k1 ,k2 ,...,kn ,l)
= (1)n
l.
Q10,K1
(3.22)
2k1 , 2k2 , . . . , 2kn1 , 2kn 1, m l 2K + 2
3.2. The gauge variation of k(m,s) for k > 0
That part of (m,s) which is of level k is introduced by the ansatz (see (1.30))
L
2k
(m,s)
k
m

(1)l
l=0
l!
(b)
2 r1
r
A(l)
(ab)r2 b2 3 (a)l2r1 r2
r1 r2 r3 a
ri
mlr2 2r3
(ba )l h(s) ,
(3.23)

ri = k. The strategy is to derive the unknown cowhere the sum over the ri is restricted to
efficients A from the known coefficients Q by requiring (1.31), (1.32). The idea is to postulate
(1.31) for both cases, the curvature and the Christoffel symbols, tentatively, in order to derive a
382
system of difference equations for the A. Then one solves these if they are solvable. Otherwise
one determines the obstruction of solvability. Only for m < s the obstructions are effective as we
shall see (after an intricate analysis) and the weaker constraint (1.32) is applied.
The sum over l for given ri is restricted by
2r1 + r2 l m r2 2r3 ,
(3.24)
which implies
k [s/2].
(3.25)
The gauge variation is easily obtained in the form

(m,s)
L2k k
m

r
(1)l 2 r1
(ab)r2 b2 3
a
l!
l=0 ri

(l)
l2r1 r2 +1
Ar1 r2 r3 A(l+1)
(b)mlr2 2r3
r1 r2 r3 (a)

l2r1 r2
(b)mlr2 2r3 , (a) (ba )l (s1) .
+ A(l)
r1 r2 r3 (a)
(3.26)
(m,s)
Thus k
decomposes into a level k term and terms of level K = k + , > 0 by expansion
of the commutator into contributions of different ordered partitions {k2 , k3 , . . . , kn } of
= k 2 + k3 + + k n ,
ki > 0.
(3.27)
By the now standard analysis we get for the first commutator and the fixed partition {k2 }
2k2 2
m
r
(1)l (l) 2 r1
b
Ar1 r2 r3 a
(ab)r2 b2 3 (a)l2r1 r2
L2k
L
l!
ri
l=0

m l r2 2r3
(b)mlr2 2r3 2k2 +1 1

2k2 1

m l r2 2r3
mlr2 2r3 2k2
2 (ba )l (s1) .
(3.28)
(b)
2k2
After neglecting the new commutator generated by 2 and inserting (3.5), (3.6) and simplifying
we get for this partition {k2 } the expression with the As of level k added to an equation for the
level k + k2
m

(1)l
l=0
l!

L2K
r
r +k 1
a 2 1 (ab)r2 b2 3 2 (a)l2r1 r2 +1 (b)mlr2 2r3 2k2 +2
ri
m l r2 2r3 + 1
(l+1)
Ar1 ,r2 ,r3 1
(s l 1)
2k2 1

m l r2 2r3 + 2
(l)
Ar1 ,r2 ,r3 1
2k2

m l r2 2r3 + 1
m l r2 2r3 + 1
+ A(l)
(2l
s
+
1)
+
r1 ,r2 1,r3
2k2 1
2k2

m l r2 2r3 + 1
(l1)
Ar1 1,r2 ,r3
(3.29)
l (ba )l (s1) ,
2k2 1
where the ri are restricted to

ri = k + 1.
383
(3.30)
The recursion relations for the coefficients A at level one and two are determined and solved
in the subsequent section. For general partitions {k2 , k3 , . . . , kn } the relevant formulae are easily
obtained but too clumsy to be given here.
4. Setting up and solving the recursion relations
(m,s)
4.1. 1
and the concept of obstruction
It is natural to solve the condition of gauge invariance (3.1) level by level in increasing order.
The equations we have to solve have the following structure
(l)
(k,l)
A(l+1)
r1 r2 r3 Ar1 r2 r3 = Qr1 r2 r3 +
k1

k1 =1
1 ,kk1 ,l) +
Q(k
r1 r 2 r 3
Qr(k1 1r2,kr32 ,kk1 k2 ,l) +
k1 ,k2 1,k1 +k2 k1
)
q1 2 3 A(lr1
1 ,r2 2 ,r3 3
(l,l )
l ,1 ,2 ,3
(0 i ri , 1 1 + 2 + 3 k 1, r1 + r2 + r3 = k),
(4.1)
where the r.h.s. A-terms exist if k 2, result from formulas such as (3.29), and must be summed
over all partitions. In the simplest case k = 1 they do not arise. In that case we have
(l)
(1,l)
A(l+1)
r1 r2 r3 Ar1 r2 r3 = Qr1 r2 r3
(r1 + r2 + r3 = 1).
This case k = 1 is studied first.

In the sequel we use Pascals summation formula which for

al
[a l]n
(l+1)
(l)
A
A =
=
,
n
n!
implies
(4.3)
(l)

a+1l
=
+ constant.
n+1
(4.2)
(4.4)
We insert the expressions (3.9)(3.11) for k = 1 into (4.1)

1
(1,l)
Q001 = (m l 1)(2s m l 2),
2
1
(1,l)
(b) Q010 = (m l)(3l 2s + m + 1),
2
(1,l)
(c) Q100 = l(m l + 1).
(a)
(4.5)
(4.6)
(4.7)
In the case k = 1 there are no lower level A-terms.

These coefficients are nonzero in the intervals
(a) 0 l m 2,
(4.8)
(b) 0 l m 1,
(4.9)
384
(c) 1 l m.
(4.10)
From the ansatz (1.30), (3.23) which excludes negative powers of a 2 , (ab), b2 and a, b follows (see (3.25))
2r1 + r2 := lmin l lmax := m r2 2r3 .
(4.11)
Moreover from the multinomials in Q (3.20)(3.22) and from (4.11) we can deduce the support
of Q(k,l) in l
max{0, lmin 1} l lmax ,
(4.12)
which exactly reproduces (4.8)(4.10).

Boundary conditions of the difference equations are obtained from those equations that have
only one A on the left-hand side, namely
(a)
A(m2) = Q(1,m2) ,
(b)
(c)
(m1)
(m)
=Q
(1,m1)
=Q
(1,m)
=Q
(1,0)
,A
(2)
,A
(4.13)
(1)
=Q
(1,1)
(4.14)
(4.15)
so that we have to deal with two boundary conditions in cases (b) and (c), which, except under
special circumstances, is one too many. The generalization to arbitrary k is straightforward. Only
in the case
(r1 , r2 , r3 ) = (0, 0, k),
(4.16)
we have one, and in all other cases we have two boundary conditions. But two boundary condi(l)
tions are in general incompatible with first order difference equations. As usual we define Ar1 r2 r3
to be zero outside its support (4.11). The incompatibility can then be expressed by an obstruction r1 r2 r3 through

)
max )
Q(...,l
r1 r2 r3 = A(lr1max
(4.17)
r2 r3 +
r1 r2 r3 + A-terms of lower level as in (4.1),
all partitions
which, if the upper boundary condition is satisfied, vanishes identically.

For the cases with two boundary conditions we determine the semi-solution satisfying only
the lower boundary condition. For k = 1 this gives

ml
ml
(l)
(s m)
(proper solution),
(a) A001 =
(4.18)
3
2

ml+1
m+1
(l)
(s m)
(semi-solution),
(b) A010 = (l + s m)
(4.19)
2
2

l
l
(l)
m
(semi-solution),
(c) A100 = 2
(4.20)
3
2
and the obstruction is in either case

m+1
(b)
=
(m s),
2

m+2
(m)
(1,m)
.
(c) 100 = A100 + Q100 =
3
(m1)
010 = A010
(1,m1)
+ Q010
(4.21)
(4.22)
385
These obstruction functions enter into the gauge variation of which to the order O(L2 ) gives
(m)(s)
(m)(s)
(z; a, b) + 1
(z; a, b)

m
1
(1)
m1
2
2
2
(a)
L (a) m(m + 1)(m + 2)a (ba )m (s1) (z; a)
=
m!
6
m
(1) m(m + 1)
(a)m1 (ba )m1 (s1) (z; a).
+ (ab)(s m)
2(m 1)!
L2 0
(4.23)
From this result we conclude that for the curvature with s = m and for the level k = 1 we have
obtained a gauge invariant expression, since
(ba )s (s1) (z; a) = 0.
(4.24)
For the Christoffel symbols we have to make the b-trace of the gauge variation traceless

1 2 (m,s)
(z; a, b) + 1(m,s) (z; a, b)
2b L 0
2
(1)m
=
m(m + 1)(s m)(s m + 1)(a)m1 (ba )m2 (s1) (z; a).
2(m 2)!
(4.25)
This trace we can cancel only adding an additional b2 term

b2 m(m + 1)(s m)(s m + 1)
L1 h(s) =
d + 2m 3
m1
(1)l
(a)l1 (b)ml1 (ba )l1 h(s) (z; a).
2(l 1)!
(4.26)
l=1
The trace of the gauge variation of this term cancels exactly the r.h.s of (4.25).
(l)
Effectively this term leads to a d-dependent shift of A001 coefficients independent of l, which
amounts to a change in the (single) boundary condition for the difference equation for these
terms. The ansatz (1.30), (3.23) remains unchanged. In (3.29) we must, however, insert now the
modified A(l)
00k

m + 1 (s m)(s m + 1)
(l)
(l)
(l)
A001 A001 = A001
(4.27)
.
2
d + 2m 3
(l)
(l)
At the end we change the notation and write again A001 for A001 .
The case m = 2 is special because in this case all our sums degenerate and we have only the
(0)
(1)
(2)
boundary conditions for the remaining A001 , A010 , A100 coefficients. Instead of (4.23) we have
(1)
here (with B = A010 )

L2 0(2,s) (z; a, b) + (2 s)b2 + B(ab)(ba ) a 2 (ba )2 h(s) (z; a)

= (ab) (B + 3 2s)(b) + (B 3 + s)(a)(ba ) (s1) (z; a),
(4.28)
which can be made traceless if we choose

B =

1 2
s 6s + 6 .
s
(4.29)
386
So the correct second Christoffel symbol is

1
(2,s) (z; a, b) = (b)2 h(s) (z; a) (a) (b)(ba ) (a)(ba )2 h(s) (z; a)
2

(ab)
1
(s 2 6s + 6)(ba )
2 a 2 (ba )2 + (s 2)b2 +
s
L
h(s) (z; a).
(4.30)
Taking the trace we obtain the so-called gauge invariant Fronsdal operator

1
1
(2,s)
(s)
(z; a, b) = 2h (z; a) (a) (a ) (a)2a h(s) (z; a)
2b
2
2

1
2 a 2 2a + s 2 + s(d 5) 2(d 2) h(s) (z; a),
(4.31)
L
which is the AdS generalization of (1.3). This completes the discussion of the case k = 1.
4.2. The solutions for the cases k > 1
We return now to the general case of arbitrary k. After insertion of the solutions and semisolutions into (1.30), (3.23), we get for the gauge variation of the Christoffel symbol or the
curvature
k

(1)m 2k
2k
2n
(m,s)
L
L n
L (a)m+1 (ba )m (s1) (z; a)
=
m!
n=0

r1 ,r2 ,r3 ;r3 =k
r 1 r 2 r 3
r
(1)lmax 2 r1
(ab)r2 b2 3
a
(lmax )!
(a)m2k+1 (ba )lmax (s1) (z; a)

(lmax = m r2 2r3 ).
(4.32)
For the curvature the first term on the right and the first term in the sum for (r1 r2 r3 ) = (k, 0, 0)
vanish since for this one
lmax = m = s.
(4.33)
The obstruction functions r1 r2 r3 can be expanded as a finite polynomial of s m with coefficients no longer dependent on s

(s m)n r1 r2 r3 (n)
r 1 r 2 r 3 =
(4.34)
n0
(see (4.21), (4.22) for k = 1 and Appendix B for k = 2). In all the cases explicitly calculated we
obtained
r1 r2 r3 (0) = 0,
(4.35)
except for r1 = k which for the gauge invariance requirement is irrelevant as we have just seen.
We assume therefore that this is true for all k. This implies that the curvature
m
(s)
[2]

k=0
(s)
k ,
(4.36)
387
as expressed by (1.30), (3.23) after insertion of the A is gauge invariant. This closes the curvature
problem.
In the case of the Christoffel symbols m < s we must apply a b-trace to (4.32). In the natural
basis of functions this implies
r
(1)mr2 2r3 2 r1
1
(ab)r2 b2 3 (ba )mr2 2r3 (s1) (z; a)
2b
a
2 (m r2 2r3 )!
rrr
t
(1)mt2 2t3 2 t1
Mt11t22t3 3
a (ab)t2 b2 3
=
(m t2 2t3 2)!
t t t
1 2 3
(ba )mt2 2t3 2 (s1) (z; a),
(4.37)
where the sum extends over

t1 + t2 + t3 = k 1.
(4.38)
Only in three cases the matrix M has nonvanishing entries

(1)
t1 = r1 + 1, t2 = r2 2, t3 = r3 :
(4.39)
(2)
t1 = r1 , t2 = r2 , t3 = r3 1:
1
r r2 r3
Mr11+1,r
= r2 (r2 1),
2 2,r3
2

r r r
Mr11,r22 ,r3 3 1 = r3 2(m r3 1) + d + 1 ,
(4.40)
t1 = r1 , t2 = r2 1, t3 = r3 :
r r r3
Mr11,r22 1,r
3
(3)
= r2 (m s r2 2r3 ).
(4.41)
= 0.
This implies in particular that Mtk00
1 t2 t3
Though an obstruction 00k does not exist, we introduce now such function which is assumed
to be expandable in powers of s m in the same way as the obstructions, for the only purpose
to make the Christoffel symbols b-traceless, and insert it in (4.32). In this equation r3 = k is now
allowed. Due to (4.37) b-tracelessness of (4.32) amounts to
rrr
Mt11t22t3 3 r1 r2 r3 = 0.
(4.42)
ri
For k = 1 this implies

010
001
M000
010 + M000
001 = 0,
(4.43)
and for k = 2
011
002
M001
011 + M001
002 = 0,
020
020
M010
110
M100 110
011
+ M010
011
101
+ M100 101
(4.44)
= 0,
020
+ M100
020
(4.45)
= 0.
(4.46)
Equations of the type (4.43), (4.44) are understood as definitions of the new functions 00k . For
general k they are solved by
00k =
2k + s m 1
01,k1 .
k[2(m k) + d 1]
(4.47)
From this equation follows also that 00k is of first order in s m since 01,k1 is. For k = 1 we
used this equation already in (4.27) with 010 taken from (4.21). For k = 2 we calculate 002
to all orders in s m in Appendix B. Moreover there are for each k a set of 12 (k 1)(k + 2)
identities (for k = 2 these are (4.45), (4.46)) which must be fulfilled. From the results given in
388
Appendix B it is easy to see that to the first order in s m these identities are satisfied. We
assume therefore that all such identities are obeyed to all orders in s m.
The term with factor 00k which we want to add to (4.32) is a gauge variation (derivative)
which we ought to integrate to obtain a local linear functional acting on the gauge field h(s) . An
ansatz for this functional is
2 k m2k
(1)l

b
L2k Lk h(s) = 00k
(4.48)
(a)l (b)m2kl (ba )l h(s) (z; a).
2
l!
L
l=0
(m2k,s)
(3.2)
Then its gauge variation is analogous to the gauge variation of 0

L2k Lk (a) (s1)
2 k
b
(1)m2k
(a)m2k+1 (ba )m2k (s1) (z; a)
= 00k
(m 2k)!
L2

m2k
(1)l

l
m2kl
l (s1)
, (a) (ba )
(z; a) .
(a) (b)
+
l!
(4.49)
l=0
The first sum on the right-hand side is the object we want to integrate. The desired local linear
functional of h(s) stands on the left-hand side. But there are correction terms from the second
sum of the right-hand side which are higher order in L2 due to the commutator. We assume that
these correction terms are included already in the ansatz (1.30), (3.23) for the next level. Then
we have to take account only of the modification
(l)
(l)
A(l)
002 A002 = A002 + 002 ,
(4.50)
in equations such as (3.29).

The problem of the Christoffel symbols has therefore also a unique solution.
5. The Riemann tensor
As explained in Section 1 we can derive the Riemann tensor (linearized in the HS field) by
applying an antisymmetrizer to the de WitFreedman tensor. We introduce an antisymmetric
tensor A from the tangential tensor space (as a special case we could use A = a b a b )
and contract it s times with R. Then we get for the flat case (up to a normalization)
R (s) (z; A) =
(s)

i Ai i h(s)
1 ,2 ,...,s (z).
(5.1)
i=1
We introduce the shorthand

= A ,
(5.2)
so that (5.1) simplifies to

R (s) (z; A) = h(s) (z; ).
(5.3)
Here we always understand that vectorial differential operators contracted into h(s) act on the
position z.
389
The method of contracting A can be easily carried over to the AdS expressions (1.26), (1.30)
with m = s. We choose one b and combine it with all possible a and replace this pair by A .
(s)
This gives s! terms for each term from k . We divide by s! at the end. With the shorthands
= A ,
B
(5.4)
= A A ,
(5.5)
we obtain after some algebra

R (s) (z; A) =
[s/2]

k=0
(1)k k (s)
h (z; , , . . . , ; B, B, . . . , B ).

L2k
(5.6)
s2k
The function h(s) depending on arguments and B must be read as follows

h (z; , , . . . , ; B, B, . . . , B ) =

(s)
s2k
s2k

i=1
sk

j =s2k+1
B j j +k h(s)
1 2 ,...,s (z).
(5.7)
We recognize that R (s) depends on a simple sequence of numerical constants k only. They can
(l)
be expressed in a surprisingly simple fashion by the coefficients Ar1 r2 r3
sr
2 2r3
k =
r1 ,r2 ,r3 ;r1 +r2 +r3 =k l=2r1 +r2
A(l)
r1 r2 r3 .
(5.8)
(l)
Inserting the coefficients Ar1 r2 r3 at s = m from (4.18)(4.20) and Appendix B we get

0 = 1,

s +1
s +1
+3
,
1 = 4
4
3

s +1
s+1
s +1
+ 114
+ 45
.
2 = 70
7
6
5
(5.9)
(5.10)
(5.11)
It is remarkable that these coefficients k vanish when 2k > s as they should. Some details of
the arguments leading to (5.8)(5.11) are presented in Appendix C.
6. Conclusions
We derived de WitFreedman formulas for the curvature and the generalized Christoffel symbols both linearized in the higher spin field h(s) (z; a) on AdSd+1 space to the second order
(level) in the expansion in powers of the AdS inverse radius squared. These expressions are
complete for spin smaller or equal five. The Riemann curvature is then derived in the same order, it permits to obtain Ricci and Weyl tensors by elementary manipulations. The concepts and
arguments given here allow to formulate a computer algorithm by which higher orders necessary
for spin greater or equal six can be obtained. New obstacles are not expected.
Acknowledgements
This work is supported by the German Volkswagenstiftung. The work of R.M. was supported
in part by the INTAS grant #03-51-6346.
390
Appendix A. Derivation of cs
For simplicity we consider the flat space with dimension not smaller than two. Set i = 1,
i = 2 for all i in (1.6). Then instead of (1.18) with still undetermined cs we have by expansion

s

(s)
(s)
1
sl s
(1)
.
R12,12,12,...,12 = cs
(A.1)
l 1,1,...,1l ,2,2,...,2s ;1,1,...,1sl ,2,2,...,2s
l=0
In turn from (1.10) we have

(s)
1,1,...,1l ,2,2,...,2s ;1,1,...,1sl ,2,2,...,2s

(s)
= (s!)1
R1P (1 ),1P (2 ),...,1P (l );2P (l+1 ),2P (l+2 ),...,2P (s ) ,
(A.2)
P ()
where P (i ) denote the permutations of the i and over all these we sum in (A.2). But only those
permutations count by symmetry for which
P (i ) = 2 whenever 1 i l,
(A.3)
P (i ) = 1 whenever l + 1 1 s.
(A.4)
Counting the number of permutations that are relevant we obtain l!(s l)! so that
1
(s)
(s)
sl s
R12,12,...,12 .
11,1,...,1l ,2,2,...,2s ;1,1,...,1sl ,2,2,...,s = (1)
l
It follows that cs = 1 as claimed in (1.16).
(A.5)
Appendix B. Results for k = 2

For the curvature m = s the term {r1 r2 r3 } = {k00} and the first term of (4.32) drop out since
the sth derivative with respect to a of (s1) vanishes. The remaining obstructions reduce to
r1 r2 r3 (0) and are calculated for k = 2 and shown to be zero (see below). Assuming that this
holds also for all levels k, the derivation of the de WitFreedman curvature is completed. We
perform polynomial expansions of A and the integration constants c in powers of s m

(s m)n A(l)
A(l)
(B.1)
r1 r2 r3 =
r1 r2 r3 (n),
n0
cr 1 r 2 r 3 =
(s m)n cr1 r2 r3 (n).
(B.2)
n0
We use the notation of the inverse Pochhammer

x
n!.
[x]n =
n
Then we obtain for the all (four) cases {r1 r2 r3 } with exception of r1 = 2 and r2 = 2:
cr1 r2 r3 (0) = r1 r2 r3 (0) = 0.
(B.3)
(B.4)
(l)
A011
to all orders in s m but the other ones only to the first order,
We give the coefficients
remember m = (d + 2m 3)1
A(l)
011 (0) =
1
1
1
[m l + 1]6 + (2m 11)[m l + 1]5 + (m 3)[m l + 1]4 , (B.5)
12
24
8
1
1
1
(l)
A011 (1) = [m l + 1]5 + (m 4)[m l + 1]4 + [m + 1]2 [m l]3
3
4
12

1
+ m [m + 1]2 [m l]3 + (m 2)[m l 1]2 + c011 (1),
4
1
c011 (1) = 011 (1) = m [m + 1]4 ,
2
1
1
(l)
A011 (2) = [m l + 1]4 + [m + 1]2 [m l 1]2 ,
4
4
c011 (2) = 011 (2) = 0,
1
1
1
(l)
A101 (0) = [m l + 2]6 (m 2)[m l + 2]5 + [m + 2]3 [m l]3 ,
18
12
36
7
1
1
(l)
A101 (1) = [m l + 2]5 (3m 4)[m l + 2]4 + [m + 2]3 [m l]2
45
12
12

1
+ m [m + 1]2 2[m l]3 3(m 2)[m l]2 + c101 (1),
12
1
c101 (1) = 101 (1) = m m[m + 1]4 ,
12
1
1
(l)
A020 (0) = [m l + 2]6 (m 3)[m l + 2]5
8
4
1
+ (m 2)(m 3)[m l + 2]4 ,
8

1
1
1
(l)
A020 (1) = [m l + 2]5 (2m 5)[m l + 2]4 [m + 1]2 [m l]3
4
8
4
(m 3)[m l]2 + c020 (1),
1
c020 (1) = 020 (1) = [m + 1]4 ,
8
1
1
(l)
A110 (0) = [m l + 3]6 + (5m 8)[m l + 3]5
6
12
1
(m 1)(3m 4)[m l + 3]4
12

1
[m + 2]3 [m l + 1]3 (m 1)[m l + 1]2 ,
12
1
1
1
(l)
A110 (1) = [m l + 3]5 + (6m 5)[m l + 3]4 + [m + 1]2 [m l + 1]3
6
24
6
1
(4m 1)[m + 1]2 [m l + 1]2 + c110 (1),
12
1
c110 (1) = 110 (1) = (2m + 3)[m + 1]4 .
24
Whereas by definition
391
(B.6)
(B.7)
(B.8)
(B.9)
(B.10)
(B.11)
(B.12)
(B.13)
(B.14)
(B.15)
(B.16)
(B.17)
(B.18)
+1)
(m+1r2 2r3 )
Ar(l1max
(n) = r1 r2 r3 (n),
r2 r3 (n) = Ar1 r2 r3
(B.19)
1)
(2r1 +r2 1)
(n) = 0,
Ar(l1min
r2 r3 (n) = Ar1 r2 r3
(B.20)
and the latter equation determines cr1 r2 r3 (n), the equality of r1 r2 r3 and cr1 r2 r3 in all cases listed
here is a nontrivial result.
392
Now we study the exceptional cases {200} and {002}. There is no dependence on s in the first
case.
1
1
1
(l)
A200 = [l]6 (5m 6)[l]5 + m(m 2)[l]4 ,
(B.21)
18
30
8
1
200 =
(B.22)
(5m + 12)[m + 2]5 .
360
The second case gives to all orders in s m

1
5[m l + 1]6 3[m l + 1]5 + 15[m l]4 ,
(B.23)
360
1
(l)
[m l + 1]5 2[m l]4
A002 (1) =
12

+ m [m + 1]2 [m l 1]3 + 3[m l 2]2 ,
(B.24)

1
1
(l)
A002 (2) = [m l]4 + m [m + 1]2 [m l 1]3 + 6[m l 2]2 ,
(B.25)
8
12
1
(l)
A002 (3) = m [m + 1]2 [m l 2]2 .
(B.26)
4
There is no proper obstruction in this case. But the trace compensator 002 is from (4.47) and
(B.7), (B.9)
(l)
A002 (0) =
1
002 = (s m)(s m + 3)m m1 [m + 1]4 .
4
(B.27)
Appendix C. Derivation of the Riemann tensor

By the total symmetry of the tensor field h(s)
(ba )l h(s) (z; a) = [s]l h(s) (z; b, a),
(C.1)
where

s
,
[s]l = l!
l
(C.2)
it is impossible to contract a pair a, b in its arguments to the antisymmetric tensor A. On the

other hand the number of a (respectively b) in the arguments of h(s) is the same as the number
of b (respectively a) in the prefactors
r
2 r1
(ab)r2 b2 3 (a)l2r1 r2 (b)slr2 2r3 .
a
(C.3)
Therefore it is also impossible to contract a pair a, b from inside the prefactors (C.3) to an A.
Contracting an a from (a)l2r1 r2 with a b from h(s) yields

a b A = ,
(C.4)
and contracting a b from (b)slr2 2r3 with an a from h(s) we get

b a + A = + .
(C.5)
The combinatorical factor and sign is

(1)l
l!(s l)!
.
(2r1 + r2 )!(2r3 + r2 )!
(C.6)
393
After this contraction there are 2r1 + r2 bs and r2 + 2r3 as left as arguments in h(s) .
Next we make the contraction with (ab)r2 . We take one pair g a b and contract it with
b a from the arguments of h(s) . This yields

g A +A = B .
(C.7)
There is no sign factor and the combinatorical factor is
(2r1 + r2 )!(2r3 + r2 )!
.
(2r1 )!(2r3 )!
(C.8)
After these contractions the number of as left in h(s) is 2r3 and the number of bs is 2r1 .
Now we contract with as from (a 2 )r1 . We get
g a a b b = B .
(C.9)
The sign and combinatorical factor is

(1)r1 (2r1 )!.
In the same fashion we obtain for the contraction of
(C.10)
(b2 )r3
the sign and combinatorical factor
(1)r3 (2r3 )!.

Thus we finished the contractions.
We collect all sign and combinatorical factors and obtain

s
(1)l2r1 r2 l!(s l)!(1)r1 +r3 = s!(1)r1 +r2 +r3 = s!(1)k .
(1)l
l
(C.11)
(C.12)
In this way the proof of the results of Section 5 is almost completed.

The technical problem to perform the summation (5.8) is solved by using [11] and manipulating the summation label in such a fashion that the first term in each sum is one as it is in the
quoted formula.
References
[1] B. de Wit, D.Z. Freedman, Systematics of higher spin gauge fields, Phys. Rev. D 21 (1980) 358367.
[2] T. Damour, S. Deser, Geometry of spin 3 gauge theories, Ann. Poincar Phys. Theor. 47 (1987) 277;
T. Damour, S. Deser, Higher derivative interactions of higher spin gauge fields, Class. Quantum Grav. 4 (1987) L95.
[3] C. Fronsdal, Singletons and massless, integral spin fields on de Sitter space (elementary particles in a curved space
VII), Phys. Rev. D 20 (1979) 848;
C. Fronsdal, Massless fields with integer spin, Phys. Rev. D 18 (1978) 3624.
[4] V.E. Lopatin, M.A. Vasiliev, Free massless bosonic fields of arbitrary spin in d-dimensional de Sitter space, Mod.
Phys. Lett. A 3 (1988) 257.
[5] M.A. Vasiliev, Actions, charges and off-shell fields in the unfolded dynamics approach, Int. J. Geom. Meth. Mod.
Phys. 3 (2006) 37, hep-th/0504090;
X. Bekaert, S. Cnockaert, C. Iazeolla, M.A. Vasiliev, Nonlinear higher spin theories in various dimensions, hepth/0503128.
[6] E.S. Fradkin, M.A. Vasiliev, On the gravitational interaction of massless higher spin fields, Phys. Lett. B 189 (1987)
89;
E.S. Fradkin, M.A. Vasiliev, Cubic interaction in extended theories of massless higher spin fields, Nucl. Phys. B 291
(1987) 141;
M.A. Vasiliev, Higher-spin gauge theories in four, three and two dimensions, Int. J. Mod. Phys. D 5 (1996) 763,
hep-th/9611024;
M.A. Vasiliev, Nonlinear equations for symmetric massless higher spin fields in (A)dSd , hep-th/0304049.
394
[7] I.R. Klebanov, A.M. Polyakov, AdS dual of the critical O(N ) vector model, Phys. Lett. B 550 (2002) 213, hepth/0210114.
[8] R. Manvelyan, W. Rhl, The structure of the trace anomaly of higher spin conformal currents in the bulk of AdS4 ,
Nucl. Phys. B 751 (2006) 285, hep-th/0602067.
[9] R. Manvelyan, W. Rhl, The quantum one loop trace anomaly of the higher spin conformal conserved currents in
the bulk of AdS4 , Nucl. Phys. B 733 (2006) 104, hep-th/0506185.
[10] F.A. Berends, G.J.H. Burgers, H. Van Dam, Explicit construction of conserved currents for massless fields of arbitrary spin, Nucl. Phys. B 271 (1986) 429441.
[11] I.S. Gradshteyn, I.M. Ryzhik, Tables of Integrals, Series, and Products, Academic Press, New York, 1965,
Eq. 0.151.1.

Two 3-branes in RandallSundrum setup

and current acceleration of the universe
Anzhong Wang a, , Rong-Gen Cai b , N.O. Santos c,d,e
a CASPER, Department of Physics, Baylor University, Waco, TX 76798-7316, USA
b Institute of Theoretical Physics, Chinese Academy of Sciences, PO Box 2735, Beijing 100080, China
c School of Mathematical Sciences, Queen Mary, University of London, London E1 4NS, UK
d LERMA/CNRS-FRE 2460, Universit Pierre et Marie Curie, ERGA, Bote 142, 4 Place Jussieu,
75005 Paris Cedex 05, France

e Laboratrio Nacional de Computao Cientfica, 25651-070 Petrpolis RJ, Brazil
Received 8 January 2007; accepted 12 November 2007

Available online 19 November 2007
Abstract
Five-dimensional spacetimes of two orbifold 3-branes are studied, by assuming that the two 3-branes are
spatially homogeneous, isotropic, and independent of time, following the so-called bulk-based approach.
The most general form of the metric is obtained, and the corresponding field equations are divided into three
groups, one is valid on each of the two 3-branes, and the third is valid in the bulk. The Einstein tensor on the
3-branes is expressed in terms of the discontinuities of the first-order derivatives of the metric coefficients.
Thus, once the metric is known in the bulk, the distribution of the Einstein tensor on the two 3-branes is
uniquely determined. As applications, we consider two different cases, one is in which the bulk is locally
AdS5 , and the other is where it is vacuum. In some cases, it is shown that the universe is first decelerating
and then accelerating. The global structure of the bulk as well as the 3-branes is also studied, and found that
in some cases the solutions may represent the collision of two orbifold 3-branes. The applications of the
formulas to the studies of the cyclic universe and the cosmological constant problem are also pointed out.
PACS: 03.50.+h; 11.10.Kk; 98.80.Cq; 97.60.-s
Keywords: Orbifold branes; Accelerating Universe; Dark energy
E-mail addresses: anzhong_wang@baylor.edu (A. Wang), cairg@itp.ac.cn (R.-G. Cai), nos@cbpf.br,

santos@ccr.jussieu.fr.
doi:10.1016/j.nuclphysb.2007.11.009
396
A. Wang et al. / Nuclear Physics B 797 [PM] (2008) 395430
1. Introduction
Superstring and M-theory all suggest that we may live in a world that has more than three
spatial dimensions. Because only three of these are presently observable, one has to explain why
the others are hidden from detection. One such explanation is the so-called KaluzaKlein (KK)
compactification, according to which the size of the extra dimensions is very small (often taken
to be on the order of the Planck length). As a consequence, modes that have momentum in the
directions of the extra dimensions are excited at currently inaccessible energies.
Recently, the braneworld scenarios [1,2] has dramatically changed this point of view and,
in the process, received a great deal of attention. At present, there are a number of proposed
models (see, for example, [3] and references therein). In particular, Arkani-Hamed et al. (ADD)
[1] pointed out that the extra dimensions need not necessarily be small and may even be in
the scale of millimeters [4]. This model assumes that Standard Model fields are confined to a
three (spatial) dimensional surface (a 3-brane) living in a larger dimensional bulk, while the
gravitational field propagates in the whole bulk. Additional fields may live only on the brane
or in the whole bulk, provided that their current undetectability is consistent with experimental
bounds. One of the most attractive features of this model is that it may potentially resolve the
long standing hierarchy problem, namely the large difference in magnitudes between the Planck
and electroweak scales.
An alternative approach was proposed by Randall and Sundrum (RS) [2], which will be referred to as RS1 model. One of the most attractive features of this model is that it will soon be
explored by LHC [5]. For critical reviews of the model and some open issues, we refer readers
to [6]. The spacetime in this model is five-dimensional, with the extra dimension being compactified on an S 1 /Z2 orbifold, that is, the extra dimension is periodic, [, ], and its points
with (x , ) are identified with the ones (x , ). In such a setting, the spacetime necessarily
contains two 3-branes, located, respectively, at the fixed points = 0, and = . The brane at
= 0 is usually called hidden (or Planck) brane, and the one at = is called visible (or TeV)
brane. The corresponding 5D spacetime is locally anti-de Sitter (AdS5 ), and described by the
metric,
ds52 = e2krc || dx dx rc2 d 2 ,
(1.1)
for which the 4D spacetimes of the two 3-branes are Poincar invariant, where , = 0, 1, 2, 3,
and k is a constant and of the order of M, where M is the five-dimensional mass scale. denotes the 4D Minkowski metric, and rc the radius of the extra dimension. To obtain the relationship between M and the four-dimensional Planck scale Mpl , let us consider the gravitational
perturbations, given by

ds52 = e2krc || + h x dx dx rc2 d 2 ,
(1.2)
we find that
Sg(5) =
M3
16
M3
=
16

dx 4 d

4
d x

g (5) R (5)
d rc e2krc ||
g (4) R (4) ,
(1.3)
397
where g (4) + h . Then, comparing Eq. (1.3) with

Sg(4)
2
Mpl
16
d x
g (4) R (4) ,
(1.4)
we find that

M3
(1.5)
1 e2krc .
k
Thus, in the RS1 model, for large krc the Planck Mpl is weakly dependent on rc and Mpl M.
Therefore, in contrast to the ADD model, the RS1 model predicts that M and Mpl are in the
same order, M Mpl 1016 TeV. The resolution of the hierarchy problem comes from the
warped factor e2krc on the visible brane. To show this explicitly, let us consider a scalar field H
confined on the brane,

2
vis
=
d 4 x gvis gvis H H |H |2 m20
Seff.
2
Mpl
=
2

d 4 x g H H |H |2 e2krc m20 ,
(1.6)
where H ekrc H , and
gvis g5 (x, ) = e2krc g .
(1.7)
Therefore, the mass measured by g is given by

m = ekrc m0 .
(1.8)
1016
Then, for krc 11 and m0

TeV, we find m TeV.
When extra dimensions exist, a critical ingredient is the stability of the extra dimensions.
Goldberger and Wise showed that values of krc 11 are indeed natural and can be provided by
a stable configuration [7].
In this paper, we study cosmological models in the RS1 setup. In the original model, only
gravity propagated in the bulk and the standard model fields were confined to the TeV brane. As
a results, the bulk is necessarily a generalized Schwarzschildanti-de Sitter space [8]. However,
soon it was realized that much richer phenomena can be obtained, if some of the standard model
fields are allowed to propagate in the bulk [5]. In order to incorporate all these situations, we
shall allow the bulk to be filled with any matter field(s).
In addition, much of the work in brane cosmology has taken the so-called brane-based
approach [3,9]. In this paper, we shall follow the bulk-based approach [10], initiated by the
authors of [8]. In [10] spacetimes with one 3-brane were studied systematically by the bulkbased approach, and in particular the most general form of the bulk metric was found. In this
paper, we shall generalize those studies to the case of two 3-branes. As shown explicitly in
Section 2, their general form of the bulk metric is a particular case of ours for two 3-branes.
The rest of the paper is organized as follows: In Section 2, starting with the assumption that
the two orbifold 3-branes are spatially homogeneous, isotropic, and independent of time, and
that the extra dimension has S 1 /Z2 symmetry, we derive the most general form of the metric.
It is very important to note that the orbifold symmetry manifests itself explicitly only in certain
coordinate system. In this paper, we take the point of view that this particular coordinate system
398
is the one in which the two 3-branes are all at rest. With this in mind, using gauge freedom we first
map the two 3-branes into fixed points, in which the proper distance between the two 3-branes in
general is time-dependent. Then, using distribution theory we develop a general formula for two
3-branes with orbifold symmetry. In particular, we divide the field equations into three different
groups, given, respectively, by Eqs. (2.25), (2.26) and (2.27), where Eq. (2.25) holds in the bulk,
and Eqs. (2.26) and (2.27) hold on each of the two 3-branes. The Einstein tensor on the two
3-branes is given in terms of the discontinuities of the metric coefficients. Thus, once the metric is
(0)
(c)
known in the bulk, GAB and GAB are known, and Eqs. (2.26) and (2.27) will uniquely determine
the distribution of the matter fields on the two 3-branes. In Section 3, as applications of our
general formulas, we consider two 3-branes in an AdS5 bulk, by imposing the conditions that the
cosmological constant on the visible brane vanishes and the equation of state of its matter field
is given by pc = wc , with w being a constant. From this example we can see that different cutpaste operations result in different models, even although the bulks are all locally anti-de Sitter.
The global structure of the bulk and the 3-branes is also studied, and found that in some cases
the geodesically complete spacetime represents infinite number of 3-branes. In Section 4, similar
considerations are carried out, but now the bulk is vacuum. After re-deriving the general vacuum
solutions of the five-dimensional bulk [10], we apply our general formulas to these solutions for
two 3-branes, and whereby obtain the conditions under which the universe undergoes a current
acceleration. The study of global structure of these solutions shows that some solutions may
represent the collision of two orbifold 3-branes. Section 5 contains our main conclusions and
remarks.
It should be noted that the problem has been studied so intensively, there is inevitably overlap
between previous works and what we present here. However, as far as we know, it is the first time
to present such a general treatment of two orbifold 3-branes in arbitrary bulks and systematically
study the global structures of the bulk for all the solutions with/without the bulk cosmological
constant.
2. Gauge freedom and gauge choice for two homogeneous and isotropic 3-branes
In this paper we consider spacetimes that are five-dimensional and contain two 3-branes.
Since we shall apply such spacetimes to cosmology, we further assume that the two 3-branes
are spatially homogeneous, isotropic, and independent of time. The fifth dimension is periodic
and has a reflection (orbifold) symmetry with respect to each of the two 3-branes. Then, one can
see that the space between the two 3-branes represents half of the periodic space along the fifth
dimension.
To start with, let us first consider the conditions that the space on the two 3-branes is spatially
homogeneous, isotropic, and independent of time. It is not difficult to show that such a space
must have a constant curvature at any point of the two 3-branes, and its metric must take the
form [11],
dk2 ab dx a dx b =
dr 2
+ r 2 d 2 ,
1 kr 2
(2.1)
where d 2 d 2 + sin2 d 2 , and a, b = 1, 2, 3. The constant k represents the curvature of

the 3-space, and can be positive, negative or zero. Without loss of generality, we shall choose
coordinates such that k = 0, 1. Then, one can see that the most general metric for the fivedimensional spacetime must take the form,

ds52 = gAB dx A dx B = gij dx i dx j A2 x i dk2 ,
(2.2)
399
Fig. 1. The trajectories of two orbifold 3-branes: (a) In the (x 0 , x 4 )-plane. (b) In the (t, y)-plane, in which the two
3-branes are at rest, but the proper distance between then in general is time-dependent.
where i, j = 0, 4, A, B = 0, 1, 2, 3, 4, and dk2 is given by Eq. (2.1). In such coordinates, the

two 3-branes are located on the hypersurfaces

1 x i = 0,
(2.3)
2 x i = 0,
as shown by Fig. 1(a). Clearly, the metric (2.2) is invariant under the coordinate transformation,

i
x = f i xj .
(2.4)
As mentioned in the last section, the orbifold symmetry manifests itself explicitly only in coordinates in which the two 3-branes are all at rest. Therefore, before imposing such a symmetry,
using one degree of the freedom of Eq. (2.4), we first map the two 3-branes to the fixed points
y = 0 and y = yc , for example, by setting
y=
1 + 2
yc ,
(2.5)
where yc , , and are arbitrary constants [cf. Fig. 1(b)]. Without loss of generality, in this paper we assume yc > 0 and consider the brane located in this surface as our TeV brane. Clearly, by
properly choosing these constants, the above coordinate transformation is non-singular.1 Then,
we can use the other degree of freedom of Eq. (2.4) to set g04 = 0, so that the metric can be cast
in the form,
ds52 = N 2 (t, y) dt 2 A2 (t, y) dk2 B 2 (t, y) dy 2 .
(2.6)
It should be noted that in [10] the authors used the two degrees of freedom of Eq. (2.4) to
set N(t, y) = B(t, y) and gty = 0 for the case of one 3-brane. In such chosen (t, y)-coordinates,
the 3-brane in general is located on the hypersurface y = (t). Then, using the remaining gauge
freedom, t = t + [(t + y ) (t y )]/2, and y = y + [(t + y ) + (t y )]/2, one can
bring the 3-brane to the fixed point y = 0, while still keep the metric in the same form in terms
1 A possible singular point of the transformation (2.5) is the place where the two branes collide.
400
of t and y . For details, we refer readers to [10]. Clearly, this is possible only for the case of one
3-brane. For the case of two 3-branes, in general we have N (t, y) = B(t, y), as shown above.
The non-vanishing components of the Einstein tensor for the metric (2.6) are given by

A,y A,y
B,y
N 2 A,yy
A,t A,t
B,t
N2
G00 = 3 2
+
+3
+
+ 3k 2 ,
A
A
A
B
A A
B
B
A

A,ty
A,t N,y
B,t A,y
G04 = 3
,
A
A N
B A

N,yy
A,y A,y
B,y
N,y
A,y
B,y
A,yy
A2
+
+
2
+
2
Gab = 2 ab 2
A
N
A
A
B
N
A
B
B

2
A
B,tt
A,t A,t
B,t
N,t
A,t
B,t
A,tt
2 ab 2
+
+
+2
2
+
kab ,
A
B
A A
B
N
A
B
N

A,y A,y
N,y
B 2 A,tt
A,t A,t
N,t
B2
+
+3
+
3k 2 ,
G44 = 3 2
(2.7)
A
A A
N
A
A
N
N
A
where F,y F /y, etc.
The reflection symmetry with respect to the two branes can be obtained by the replacement
y |y|,
(2.8)
so that the most general metric for two 3-branes finally takes the form,

ds52 = N 2 t, |y| dt 2 A2 t, |y| dk2 B 2 t, |y| dy 2 ,
(2.9)
where
...
y + 2yc , 2yc y yc ,
y,
yc y 0,
|y| y,
0 y yc ,
2yc y, yc y 2yc ,
...

(2nyc y)H y (2n 1)yc 1 H (y 2nyc )
=
n=

(y 2nyc )H (y 2nyc ) 1 H y (2n + 1)yc ,
(2.10)
n=
as shown in Fig. 2, where H (x) denotes the Heavside function, defined by,

1, x > 0,
H (x) =
0, x < 0.
Hence, we obtain

d|y|
H y (2n 1)yc 1 H (y 2nyc )
=
dy
n=

n=

H (y 2nyc ) 1 H y (2n + 1)yc ,
(2.11)
401
Fig. 2. The function |y| defined by Eq. (2.10).

d 2 |y|
=
2
(y
2ny
)
2
y (2n + 1)yc ,
c
2
dy
n=
n=
(2.12)
where (x) denotes the Dirac function, with

dH (x)
= (x),
dx

f (x)(x) dx = f (0),
(2.13)
for any given test function f (x).

However, the orbifold symmetry first restricts y to the range, yc y yc , and then identifies
the points (x , y) with the ones (x , y), so that y finally takes its values only in the range,
0 y yc . Then, Eq. (2.12) yields
d|y|
= 1,
dy
d 2 |y|
= 2(y) 2(y yc ),
dy 2
(2.14)
for y [0, yc ].
It should be noted that, although the two 3-branes are all at rest in the above chosen gauge,
which will be referred to as the canonical gauge, the proper distance between them in general is
time-dependent, and given by
yc
D(t) =
B(t, y) dy.
(2.15)
The moment where D(t) = 0 is the one when the two 3-branes collide.
We also note that the expressions given by Eq. (2.7) for the Einstein tensor hold even for
spacetimes with orbifold symmetry, but now in the sense of distribution. To see this, let us consider a given function F (t, |y|), for which we find
F (t, |y|) F (t, y)
=
,
t
t
F (t, |y|) F (t, y)
=
,
y
y
402
2 F (t, |y|) 2 F (t, y)

=
,
ty
ty
2 F (t, |y|) 2 F (t, y)
=
,
t 2
t 2
2 F (t, |y|) 2 F (t, y)
=
+ [F,y ]0 (y) + [F,y ]c (y yc ),
y 2
y 2
(2.16)
for y [0, yc ], where

F (t, y)
F (t, y)
F (t, y)
[F,y ]0 lim
,
lim
=2
y
y
y y=0+
y0+
y0

F (t, 2yc y)
F (t, y)
F (t, y)
.
lim
= 2
[F,y ]c lim
y
y
y y=yc
yyc+
yyc
(2.17)
Note that in writing Eq. (2.17) we had assumed that the derivatives of F (t, |y|) exist from each
side of the limits. It can also be shown that

m
H m (x) = H (x),
1 H (x) = 1 H (x),

1
H (x) 1 H (x) = 0,
H (x)(x) = (x),
2
+

1
F (x)H (x) + F (x) 1 H (x) (x) = F + (0) + F (0) (x),

(2.18)
2
in the sense of distributions, where m is an integer, and
F + (0) lim F + (x),
x0+
F (0) lim F (x).
(2.19)
x0
Inserting Eqs. (2.16)(2.19) into Eq. (2.7), we find that the Einstein tensor corresponding to the
metric (2.9) can be written as
(D)
(0)
(c)
GAB = GAB + GAB (y) + GAB (y yc ),
(2.20)
(D)
for y [0, yc ], where GAB denotes the Einstein tensor calculated in the region y (0, yc ) and
given by Eq. (2.7), and

A20 [A,y ]0 [N,y ]0

N 2 [A,y ]0 0 0
(0)
a b
GAB = 2 2
(2.21)
+
B 3 02
A B ,
ab A
A0
N0
B0
B0 A0

[A,y ]c [N,y ]c
A2c
N 2 [A,y ]c 0 0
(c)
a b
+
B 3 c2
,
ab A
GAB = 2 2
(2.22)
Ac
Nc
Bc
Bc Ac A B
with
F0 (t) F (t, 0),
Fc (t) F (t, yc ),
(2.23)
where F {N, A, B}. Assuming that the energymomentum tensor (EMT) takes the form,
(D)
TAB = TAB +
1
1
(0)
(c)
T (y) +
T (y yc ),
B0 (t) AB
Bc (t) AB
(2.24)

(D)
(0)
403
(c)
where TAB denotes the bulk EMT, and TAB and TAB the EMTs of the two 3-branes, the Einstein
equations take the form,
(D)
(D)
GAB = 5 TAB + gAB , y (0, yc ),

1
(0)
(0)
(y = 0),
+ 0 g
5 T
G(0)
=
B0 (t)

1
(c)
(c)
G(c)
(y = yc ),
+ c g
5 T
=
Bc (t)
(2.25)
(2.26)
(2.27)
where
(0)
(t) g (t, 0),
g
(c)
g
(t) g (t, yc ).
(2.28)
The constants , 0 and c denote, respectively, the bulk and the brane cosmological constants.
Assuming that the EMTs of the two 3-branes are given by perfect fluids,
(0)
(0)
(0)
T
= (0 + p0 )u(0)
u p0 g ,
(c)
(c)
(c)
= (c + pc )u(c)
T
u pc g ,
(2.29)
where
0
u(0)
= N 0 ,
0
u(c)
= N c ,
we find that Eqs. (2.26) and (2.27) reduce to

[A,y ]0 [N,y ]0
1
+
2
+ 0 = 5 p0 ,
B0
A0
N0
3 [A,y ]0
+ 0 = 5 0 ,
B0 A0
and
[A,y ]c [N,y ]c
1
+
2
+ c = 5 pc ,
Bc
Ac
Nc
3 [A,y ]c
+ c = 5 c .
Bc Ac
(2.30)
(2.31)
(2.32)
This completes our general description of two homogeneous and isotropic 3-branes with orbifold symmetry, in which the bulk can be filled with any kind of matter fields. In the following,
we consider some specific solutions.
3. Dynamical two 3-branes in AdS5
In this section, we first consider the global properties of AdS5 , and then apply the formulas
developed in the last section to study two 3-branes in the background of AdS5 .
3.1. AdS5 space in horospheric coordinates
AdS5 can be realized as a hyperboloid embedded in a flat six-dimensional space with two
timelike coordinates,
AB Y A Y B = 2
(A, B = 0, 1, . . . , 5),
(3.1)
404
where AB = diag{+1, 1, 1, 1, 1, +1}, and (/6)1/2 , where denotes the cosmological constant of the five-dimensional bulk. In fact, introducing the Einstein universe (EU)
coordinates, {tc , , , , }, via the relations,
Y 0 = cos tc sec ,
Y 1 = sin sin cos tan ,
Y 2 = sin sin sin tan ,
Y 3 = sin cos tan ,
Y 4 = cos tan ,
Y 5 = sin tc sec ,
(3.2)
we find that

ds52 = AB dY A dY B hyperboloid
2

1
dtc d 2 sin 2 d32
2
( cos )
1
ds 2 ,
( cos )2 5(E)
(3.3)
where d32 [ d 2 + sin 2 d 2 + sin 2 sin 2 d 2 ] is the line element on the three-sphere S 3 .
The range of the EU coordinates that covers the whole AdS5 space is tc , 0 /2,
0 , , and 0 2 . The timelike hypersurface = /2 represents the spatial infinite. The spacelike hypersurfaces tc = are identified, which represent closed timelike curves
(CTCs). The topology of AdS5 is S 1 (time) R 4 (space). To avoid CTCs for AdS5 , one may simply unwrap the timelike coordinate tc and extend it to the range, < tc < , so the resulting
space has the topology R 1 (time) R 4 (space), which will be referred to as CAdS5 . Since the
spatial infinity is timelike, neither CAdS5 nor AdS5 has Cauchy surface. In the rest of this paper,
for the sake of simplicity, when we talk about AdS5 , we always means CAdS5 .
It is interesting to note that CAdS5 covers only half of the static Einstein universe given by
2
in Eq. (3.3) with the same range for tc , , and , but now 0 [cf. Fig. 3].
ds5(E)
To study RS1 model, it is found convenient to introduce the horospheric coordinates,
{t, x 1 , x 2 , x 3 , z}, via the relations,
Y 0 = t/z,
Y i = x i /z,

Y 5 + Y 4 = 1/(z),

Y 5 Y 4 = z + 2 x2 t 2 /z,
(3.4)
where < t, x i , z < (i = 1, 2, 3). Reversing the sign of z in the above expressions reverses
the signs of all Y A s, which corresponds to the antipodal map on the AdS5 hyperboloid [12].
Thus, in order to cover all of AdS5 space, at least two horospheric charts, one with positive z and
the other with negative z are needed. Vanishing z corresponds to the spacelike infinity of AdS5
( = /2). In terms of the horospheric coordinates, the AdS5 metric takes the conformally-flat
form,
405
Fig. 3. The AdS5 space seen on the Einstein cylinder. The coordinates , and are suppressed. Thus, each point
represents a three-sphere. The lines ab and cd are identified, which represent CTCs. Unwrapping tc coordinate and then
extending it to the range, tc (, ), one can avoid CTCs. The dotted periodic lines represent timelike geodesics, and
the 45 straight lines the null geodesics. The vertical line = /2 represents the spatial infinity, and a timelike geodesic
never reaches it.
ds52 =

1 2
dt dz2 d02 .
(z)2
(3.5)
As Gibbons first noticed [12], although the distance alone the spacelike geodesics of constant t
and x i diverges as |z| , the spacetime described by the metric (3.5) is not geodesically
complete with respect to timelike and null geodesics. Thus, to obtain a geodesically complete
AdS5 space, one needs to extend the region covered by metric (3.5). Since the metric has Poincar
invariance on the hypersurfaces z = constant, we may suppress the coordinates x i , and only
consider the extension of the (1 + 1)-dimensional AdS2 space,
ds22 =

1 2
dt dz2 .
2
(z)
(3.6)
Indeed, understanding the causal structure of AdSn (n 2) reduces to that of AdS2 [13]. Clearly,
2 =
metric (3.6) is conformally related to the (1 + 1)-dimensional Minkowski spacetime, dsM
2
dt 2 dz2 . Then, the corresponding Penrose diagram is given by Fig. 4. Points on each side of the
vertical line AB are related by the anti-podal map, z z. The dashed curve EF is a timelike
geodesics, which reaches the null boundary AC within a finite proper time. Thus, extension
beyond it is needed, so does across the other three null boundaries. However, these boundaries
are Cauchy horizons. For example, events beyond the boundaries AC and AD are not uniquely
determined by initial data imposed on the line CD. As a result, such extension is not unique.
As noticed by Gibbons, the coordinate z jumps from infinitely large positive values to infinitely
large negative values as one crosses the horizons [12]. Therefore, one possible extension is given
by Fig. 5, which covers the whole plane with infinite lattice of diamonds [13].
406
Fig. 4. The Penrose diagram of AdS2 space defined by metric (3.6). The vertical curves are z = z0 and z = zc , respectively, with zc > z0 > 0.
3.2. RS1 model

To obtain the RS1 solution, we first make the coordinate transformation,
z = ey ,
so that metric (3.5) takes the form,

ds52 = e2y dt 2 d02 dy 2 .
(3.7)
(3.8)
Then, applying the replacement (2.8) in the above metric coefficients, we obtain the RS metric (1.1) with y = rc . Comparing it with the general form of the metric (2.9), we find

N t, |y| = A t, |y| = e|y| ,

B t, |y| = 1,
k = 0.
(3.9)
Substituting Eq. (3.9) into Eqs. (2.31) and (2.32), we obtain
1
(6 0 ),
5
1
c = pc = (6 + c ).
5
0 = p0 =
(3.10)
RS chose
0 = c = 6,
(3.11)
so that no matter fields appear on the two branes. In Fig. 4, the two 3-branes are represented by
the two solid vertical curves with z0 1 and zc 1 eyc , respectively. The bulk is the region
between these two curves, the shaded region. As shown in the last sub-section, the spacetime is
not geodesically complete, and extension is needed. One possible extension is that given by Fig. 5
but with two vertical curves, one is z = z0 and the other is z = zc , where the bulk is the region
between the two curves. Clearly, in this extension we have infinite number of RS1 universes.
407
Fig. 5. The Penrose diagram of the extended AdS2 space. The vertical curves, both solid and dashed, represent hypersurfaces of constant z > 0.
3.3. Dynamical 3-branes in AdS5

To find new solutions, we first make the coordinate transformation,

t = F t + f (y ) + G t f (y ) ,

z = F t + f (y ) G t f (y ) ,
(3.12)
so that metric (3.5) takes the form,

ds52 =

1
4F G 2
dt f 2 (y) dy 2 2
d02 ,
2
2
(F G)
(F G)2
(3.13)
where F , G and f are arbitrary functions, and a prime denotes ordinary differentiation with
respect to their indicated arguments. Note that in writing Eq. (3.13), we had dropped the primes
from t and y . Replacing y by |y| in the metric coefficients, the resulting metric finally takes the
form of Eq. (2.9), but now with

2 F G
N t, |y| =
,
(F G)

2 F G
B t, |y| =
f |y| ,
(F G)

1
A t, |y| =
(3.14)
,
(F G)
408
where

F = F t + f |y| ,

G = G t f |y| ,
and |y| is given by Eq. (2.10). On the hypersurface y = 0, the metric reduces to

ds52 y=0 = d2 a02 () d02 ,
with
d
(3.15)
(3.16)

2 F0 G0
dt,
(F0 G0 )

F0 F t + f (0) ,
1
a0 ()
,
(F0 G0 )

G0 G t f (0) ,
1
b0 ()
.
(F0 + G0 )
From the above expressions it can be shown that

F
a0 b0 a 0
=
+
,

b02 a02
F G y=0

G
a0 b0 a 0
=
b2 a 2
F G
y=0
(3.17)
(3.18)
from which we find that

1/2
0 2
b0
+ H02
=
,
2
a0
b0
a 0
H0 ,
0 = 1,
a0
(3.19)
where a 0 da0 /d. Inserting the above expressions into Eq. (3.18) we find that, in order for F0
and G0 to be real, 0 must be chosen such that,

1, F0 > 0, G0 > 0,
0 =
(3.20)
+1, F0 < 0, G0 < 0.
Substituting Eqs. (3.14)(3.20) into Eq. (2.31) we obtain
1/2
1
= 5 0 f0 (0 + 0 ),
6
1
H0
= 5 0 f0 (0 + p0 ),
(H02 + 2 )1/2 2
H02 + 2
where 0 0 /5 , and f0 f (0)/|f (0)|.

Similarly, on the TeV brane y = yc , introducing the quantities,

2 Fc Gc
d
dt,
(Fc Gc )

Gc G t f |yc | ,
Fc F t + f |yc | ,
1
1
ac ( )
,
bc ( )
,
(Fc Gc )
(Fc + Gc )
(3.21)
(3.22)
(3.23)
we find
bc
1/2
c 2
+ Hc2
,
ac
a c
Hc ,
c = 1.
ac
bc2
409
(3.24)
Eq. (2.32) can be written as

1/2
1
= 5 c fc (c + c ),
6
1
H c
= 5 c fc (c + pc ),
2
(Hc2 + 2 )1/2
Hc2 + 2
(3.25)
(3.26)
where H c dHc /d , fc = f (yc )/|f (yc )|, c c /5 , and

1, Fc > 0, Gc > 0,
c =
+1, Fc < 0, Gc < 0.
(3.27)
Note that, in order to have c 0 for any given c , we must choose c and fc such that
c fc = 1,
(3.28)
a choice that we shall assume in the following discussions. To study the above system of differential equations further, let us consider the case where
c = 0,
pc = wc ,
(3.29)
with w being an arbitrary constant.

3.3.1. w > 1
In this case, Eqs. (3.25) and (3.26) have the general solution,

1
ac ( ) = C0 sin 3(1 + w) 3(1+w) ,
6
c ( ) =
,
5 sin[3(1 + w) ]
(3.30)
where C0 is a positive constant. From the above expressions we can see that in this case the
universe starts to expand from a big-bang like singularity at = 0, until the moment max
/[6(1 + w)]. From that moment on the universe starts to contract until the moment f
/[2(1 + w)], where the whole universe is crushed into a spacetime singularity.
For 1 < w < 2/3 the universe experiences an accelerating period. In fact, from Eq. (3.30)
we find that
a c ( ) = C0 2
(2 + 3w) + sin2 [3(1 + w) ]

6w+5
sin 3(1+w) [3(1 + w) ]
(3.31)
from which we can see that the universe is accelerating during the time 0 < < 0 for
w < 2/3, where

1
Arcsin 3|w| 2 .
0
(3.32)
3(1 |w|)
410
Fig. 6. The Penrose diagram for the case w > 1 of the solutions (3.30) in the (t, z)-plane of the horospheric coordinates,
defined by Eq. (3.4). The evolution of the visible 3-brane, located on the hypersurface y = yc , is denoted by the vertical
curve GP H , and the evolution of the invisible 3-brane, located at y = 0, is given by the dotted curve GP H . The
five-dimensional bulk is the region between the two vertical curves. The spacetimes of the 3-branes are singular at the
points G and H , but not the bulk.
To study the global properties of the solutions, we first notice that

1
,
bc

1
1
1
=
sin (1 + w) 3(1+w) ,
zc (t) = Fc (t) Gc (t) =
ac
C0
tc (t) = Fc (t) + Gc (t) =
(3.33)
as we can see from Eqs. (3.12) and (3.23). Then, we find

dzc
= c cos 3(1 + w) .
dtc
From Eqs. (3.33) and (3.34) we obtain
= 0,
,
= max ,
c tc = 0,
, = .
f
(3.34)
(3.35)
In Fig. 6 the vertical line GP H represents the visible 3-brane located at y = yc . For c = 1,
the 3-brane starts to expand from the initial moment = 0, denoted by the point G in the figure,
which is singular and corresponds to (tc , zc )G = (, ) in the (t, z)-plane [cf. Fig. 4]. The
expansion is continuous until the moment = max , where (tc , zc )P = (0, zcmin ), with zcmin
1/(C0 ). Afterwards, the universe starts to collapse and a spacetime singularity of the 3-brane
is finally formed at the moment f , where (tc , zc )H = (, ). The proper time of this whole
process is finite and given by f . It should be noted that, unlike the RS1 solution, the extension
beyond the two points G and H are not needed because now they are spacetime singularities of
the 3-brane. It is also interesting to note that the five-dimensional bulk is not singular at all even
at the points G and H .
411
For c = +1, the evolution of the visible 3-brane follows the same curve GP H , but now the
whole process is inverse in the (t, z)-plane. That is, the visible 3-brane starts to expand from the
singular point H until its maximal point P , and then collapses into a spacetime singularity at
the point G.
Repeating the above analysis for the invisible 3-brane at y = 0, it is not difficult to see that its
trajectory in the (t, z)-plane of the horospherical coordinates is the dotted curve GP H in Fig. 6.
So, the whole bulk is the region between the two curves GP H and GP H .
3.3.2. w = 1
In this case, we have
ac ( ) = C0 eHc ,
1/2
6 2
(3.36)
H + 2
,
5 c
where now Hc is a constant. Clearly, in the present case the universe is a de Sitter space, which
is expanding exponentially. Inserting Eq. (3.36) into Eq. (3.24) and then integrating the resulting
expression, we find
c ( ) = pc =
bc ( ) = c
C0 Hc
e Hc .
(Hc2 + 2 )1/2
(3.37)
Thus, we obtain
Hc
zc
= c 2
,
tc
(Hc + 2 )1/2
1 Hc , ,
zc =
e
=
0,
+ ,
C0
, ,
2
2
1/2
(H + )
c tc = c
(3.38)
eHc =
0,
+ .
C0 Hc
Fig. 7 shows the trajectory of the visible 3-brane in the plane of the horospheric coordinates t and
z for c = 1. The 3-brane starts from the moment = or (tc , zc ) = (, ), denoted by
the point B in the figure, and ends at = , where (tc , zc ) = (0, 0), as shown by the vertical
solid curve BO.
It should be noted that the spacetime described by the metric
ds42 = d 2 C02 e2Hc d02
(3.39)
is not geodesically complete. As a matter of fact, it covers only half of the hyperboloid [14],
V 2 + W 2 + X 2 + Y 2 + Z 2 = Hc2 ,
(3.40)
V + W = Hc1 eHc > 0.
(3.41)
with
To extend the spacetime to the other half of the hyperboloid, we introduce the coordinates and
via the relations,

= Hc1 ln Hc (V + W ) ,
1
cosh(Hc ) sin ,
r=
(3.42)
2
C0 Hc (V + W )
412
Fig. 7. The Penrose diagram for the case w = 1 and c = 1 of the solution (3.36) in the (t, z)-plane of the horospheric
coordinates, defined by Eq. (3.4). The evolution of the visible 3-brane, located on the hypersurface y = yc , is denoted by
the vertical curve O BO, and the trajectory of the invisible brane, located at y = 0, is given by the dotted curve O BO.
The five-dimensional bulk is the region between the two vertical curves. From the point of view of observers who live
on the brane, these trajectories are geodesically complete and no extension is needed. But, from the point of view of
observers in the five-dimensional bulk, the five-dimensional spacetime is not geodesically complete, and extension is
needed. A possible extension is given by Fig. 8.
where
V +W =

1
sinh(Hc ) + cosh(Hc ) cos .
Hc
Then, we find that the metric (3.39) becomes

ds42 = d 2 Hc2 cosh2 (Hc ) d 2 + sin2 d02 ,
(3.43)
(3.44)
which covers the whole hyperboloid for < < , 0 , , and 0 2 .

On the other hand, from Eqs. (3.38) and (3.42) we find
1
,
C0 Hc (V + W )
c (Hc2 + 2 )1/2
.
tc =
C0 Hc2 (V + W )
zc =
(3.45)
Thus, across the surface V + W = 0, both zc and tc change their signs. Then, the extended region
must be the one given by the vertical dashed curve O B in Fig. 7. Therefore, in the present case
the geodesically complete spacetime of the visible 3-brane, described by metric (3.44), is that of
the segment O BO.
However, as shown at beginning of this section, the spacetime described by metric (3.5) is not
geodesically complete, and one possible extension is that given by Fig. 5. Clearly, in the present
413
Fig. 8. The Penrose diagram of the extended spacetime for w = 1 and c = 1, given by Eq. (3.36). The vertical curves,
both solid and dashed, represent the history of a visible 3-brane. Each segment, consisting of a solid and a dashed part,
is geodesically complete for observers living on the visible brane. The trajectory of the invisible 3-brane is represented
by similar vertical curves, and the five-dimensional bulk is the region between the two curves, as shown in Fig. 7.
case, the five-dimensional bulk described by the solutions of Eqs. (3.14) and (3.36) is also not
geodesically complete. A possible extension is given by Fig. 8. It is interesting to note that the
segment O BO is geodesically complete from the point of view of observers who live on the
visible 3-brane, but this is no longer true from the point of view of the five-dimensional bulk.
For the latter, the geodesically complete bulk is that given by Fig. 8, in which we have infinite
number of geodesically complete visible 3-branes.
When c = +1, the global properties of the corresponding solution can be obtained from the
case of c = 1 by the replacement tc tc .
Similarly, it can be shown that the trajectory of the invisible brane located at y = 0 is similar to
that of the visible brane, and is given by the dotted curve O BO in Fig. 7. The five-dimensional
bulk is the region between the two vertical curves.
3.3.3. w < 1
In this case, we have
C0
,
sin[3(|w| 1) ]1/(3(|w|1))
6
,
c ( ) =
5 sin[3(|w| 1) ]
ac ( ) =
(3.46)
from which we can see that the universe is singular at = 0, where the radius of the universe
is infinitely large ac (0) = . As time increases, the radius of the universe decreases until the
414
moment = min when it reaches its minimum ac (min ) = ac0 , where
min
(3.47)
.
6(|w| 1)
After this moment, the radius of the universe starts to increase until the moment = 2min , at
which it becomes infinitely large, and the spacetime becomes singular. Note that in this case the
acceleration is always positive,
a c ( ) = C0 2
3(|w| 1) + cos2 [3(|w| 1) ]
.
(3.48)
6|w|5
sin 3(|w|1) [3(|w| 1) ]
On the other hand, we have

dzc
= c cos 3 |w| 1 ,
dtc

1
1
sin 3 |w| 1 3(|w|1)
zc (t) =
C0

0,
= 0,
= (C 0 )1 , = min ,
(3.49)
0,
= 2min .
Thus, in the (tc , zc )-plane of the horospherical coordinates, for the case c = 1 the visible
3-brane starts to collapse from the singular point (tc , zc ) = (0, 0) until the point zc (min ) =
(C0 )1 . Afterward, it starts to expand up to the moment = 2min , where (tc , zc ) = (tc0 , 0),
at which a spacetime singularity is developed [cf. Fig. 9]. Since the 3-brane is singular at the
points O and P , the spacetime of the visible 3-brane is not extendible beyond these two points.
The global properties for the case of c = +1 can be obtained from the one of c = 1 by the
replacement tc tc .
Similarly, it can be shown that the trajectory of the invisible 3-brane is denoted by the dotted
curve OP in Fig. 9.
4. Two 3-branes in vacuum bulk with k = 0
When the bulk is vacuum and the spatial sectors of the two 3-branes are flat, k = 0, the
corresponding solutions of the five-dimensional bulk are well-known [10,15], and they can be
divided into three different classes, which will be referred to, respectively, as Class IA, Class IB,
and Class II solutions. In the following we consider them separately.
4.1. Class IA solutions
This class of solutions, after the replacement of Eq. (2.8), is given by

F G
N 2 t, |y| = 2/3 ,
G

A2 t, |y| = G2/3 ,

F G 2
B 2 t, |y| = 2/3 f |y| ,
(4.1)
G
where F and G are given by Eq. (3.15), with f (|y|) being arbitrary function of |y| only. As first
noticed by Taub [15], the spacetime outside the 3-branes in this case is flat. In fact, introducing
415
Fig. 9. The Penrose diagram of the extended spacetime for w < 1 and c = 1 of the solutions (3.46) in the (t, z)-plane
of the horospheric coordinates. The trajectory of the visible (invisible) 3-brane is denoted by the solid (dotted) curve
OP . The spacetimes of the visible and invisible 3-branes are singular at the points O and P .
the coordinates T and X i (i = 1, 2, 3, 4) via the relations,

T X 1 = G1/3 t f (y) ,
t+f
(y)

2 2 2
F (x) dx + G1/3 t f (y) x 2 + x 3 + x 4 ,
T + X1 = 3

X i = G1/3 t f (y) x i (i = 2, 3, 4),
(4.2)
the metric outside the 3-branes will take the Minkowski form, ds52 = AB dX A dX B . From the
above expressions we find that
T R = 3G
2
1/3

t f (y)
2
t+f
(y)
F (x) dx,
(4.3)
where R 2 (X 1 )2 + (X 2 )2 + (X 3 )2 + (X 4 )2 .
Introducing and ac ( ) via the relations,

Fc Gc 1/2
d =
dt,
2/3
Gc
1/3
ac ( ) = Gc (t),
(4.4)
we find that Eq. (2.32) yields,

1
H c
= 5 c (c + pc ),
Hc
2
1
Hc = 5 c (c + c ),
6
(4.5)
416
where, as previously, Hc a c /ac , H c dHc /d , and c f (|yc |)/|f (|yc |)|. It is remarkable
to note that Fc (t) does not appear explicitly in Eq. (4.5).
When c = 0 and pc = wc , Eq. (4.5) has the solution,
1
ac ( ) = 3(1+w) ,
(4.6)
from which we find that

6w+5
3w + 2 3(1+w)
,
2
9(1 + w)
2c
c ( ) =
.
5 (1 + w)
a c ( ) =
(4.7)
Clearly, to have c non-negative, we much choose c such that

1, w > 1,
c =
+1, w < 1.
(4.8)
From Eq. (4.7) we can also see that the universe is accelerating for w < 2/3.
It should be noted that Eqs. (4.6) and (4.7) hold only for w = 1. When w = 1 from
Eq. (4.5) we find that
ac ( ) = a0 eHc ,
c ( ) = pc ( ) =
6c Hc
5
(w = 1),
(4.9)
where a0 and Hc are positive constants. Obviously, the spacetime of the visible 3-branes is always
acceleratingly expanding.
On the other hand, from Eq. (4.3) we find that the visible 3-brane in the five-dimensional
Minkowski coordinates T and X i is a hyperboloid,
T R
2

y=yc
= 3G
1/3

t f (yc )
t+f
(yc )
F (x) dx.
(4.10)
Similarly, the invisible 3-brane located at y = 0 is also a hyperboloid,

T R
2

y=0
= 3G
1/3

t f (0)
t+f
(0)
F (x) dx.
(4.11)
The two 3-branes will collide at the moment td , where td is a solution of the equation,

G(t f (yc ))
G(t f (0))
1/3
t+f (yc )
= t+f (0)
F (x) dx
(4.12)
F (x) dx
The bulk is the region between these two hyperboloids. However, since it is flat, no radiation,
neither gravitational nor non-gravitational, from the two 3-branes is leaked into the bulk, although
there may be energy exchange between the two 3-branes at the moment when they collide [cf.
Fig. 10].
417
Fig. 10. The two 3-branes in the Minkowski coordinates T and X i (i = 1, 2, 3, 4), defined by Eq. (4.2), at a given time t .
Each of the two branes is a hyperboloid. In some cases the two branes collide, provided that Eq. (4.12) has real solutions.
4.2. Class IB solutions

In this case, the solutions are given by

F G
N 2 t, |y| = 2/3 ,
F

2
A t, |y| = F 2/3 ,

F G 2
B 2 t, |y| = 2/3 f |y| .
F
Introducing and ac ( ) via the relations,

1/2
Fc Gc
d =
dt,
2/3
Fc
1/3
ac ( ) = Fc (t),
(4.13)
(4.14)
we find that Eq. (2.32) yields,

1
H c
= 5 c (c + pc ),
Hc
2
1
Hc = 5 c (c + c ).
(4.15)
6
Comparing Eq. (4.15) with Eq. (4.5) we can see that we can obtain one from the other by simply
replacing c by c . Thus, the physics of the solutions in the present case can be deduced from
the ones given in the last case. Therefore, in the following we shall not consider this case any
further.
418
Fig. 11. The Penrose diagram of the solutions of Eq. (4.18) in the (T , Y )-plane, defined by Eq. (4.17). The five-dimensional spacetime is singular at T = 0.
4.3. Class II solutions

In this case, the corresponding solutions are given by

F G
N 2 t, |y| =
,
(F + G)2/3

A2 t, |y| = (F + G)2/3 ,

2
F G
f |y| .
B 2 t, |y| =
(F + G)2/3
(4.16)
Before studying the dynamics of the 3-branes, let us first consider the global structure of the
spacetime described by the above solutions without 3-branes, that is, the solutions of Eq. (4.16)
with |y| being replaced by y where we temporarily extend the range of y to y (, ). Then,
by the coordinate transformation,
T F + G,
Y F G,
(4.17)
the corresponding metric takes the form,

1 2
dT dY 2 T 2/3 d02 .
(4.18)
2/3
4T
Clearly, the 5D spacetime is singular at T = 0, and the corresponding Penrose diagram is given
by Fig. 11.
Introducing , ac ( ) and bc ( ) via the relations,
1/2

Fc Gc
dt,
d =
(Fc + Gc )2/3
ds52 =
ac ( ) = (Fc + Gc )1/3 ,
bc ( ) = (Fc Gc )1/3 ,
(4.19)
from Eq. (2.32) we find,

1
H c + 2(Hc2 2 )
= 5 c0 c (c + pc ),
2
1
= 5 c0 c (c + c ),
6
(4.20)
419
where c0 = 1, and

Hc2
4
4
9ac
1/2
,
0a3
bc ( ) = c 2 c .
bc
(4.21)
To study this case further, in the rest of this sub-section, we also restrict ourselves to the case
where c = 0 and pc = wc . Then, from Eq. (4.20) we find that
a c2 4(1 + 3w)
= 0.
ac
9ac3
a c + (2 + 3w)
(4.22)
To solve this equation, it is found convenient to distinguish the following cases.

4.3.1. w > 13
In this case, introducing the function X( ) via the relation
1
ac ( ) = X 3(1+w) ,
(4.23)
where is an arbitrary constant and different from the one introduced in Eq. (2.5), Eq. (4.22)
reduces to
3w1
4(1 + w)(1 + 3w) 3(1+w)
X
X
= 0.
4
3
Choosing the constant as

4(1 + w)(1 + 3w) 1/4

,
=

3
(4.24)
(4.25)
we find that Eq. (4.24) finally takes the form

X X = 0,
(4.26)
where
1 3w
.
3(1 + w)
It can be shown that Eq. (4.26) allows the first integral,

X 2n + X1
,
X =
n
(4.27)
(4.28)
where X1 is an integration constant, and

n
1 + 3w
1
=
,
2
3(1 + w)
(4.29)
from which we find that for 1/3 < w < we always have 0 < n 1. To have an expanding
universe, in the following we shall choose the + sign in Eq. (4.28).
420
Fig. 12. The Penrose diagram of the Class II solutions in the (T , Y )-plane for w > 1. The solid (dotted) curves OAP
(OA P ) and OBP (OB P ) are the trajectories of the visible (invisible) 3-branes, respectively, for c0 = +1 and c0 = 1.
The spacetimes of both visible and invisible 3-branes are singular at the point O.
When w = 13 , Eq. (4.28) can be integrated explicitly and gives,

1/4
4
,
( + 0 )2 02
ac ( ) =
3
+ 0
a c ( ) =
,
3[( + 0 )2 02 ]3/4
( + 0 )2 + 202
a c ( ) =
,
12[( + 0 )2 02 ]7/4

1
3c0 c 0
w=
,
c ( ) =
3
5 [( + 0 )2 02 ]
(4.30)
where 0 is a constant. Clearly, in this case the universe starts to expand from the singular point
= 0, and the acceleration of the expansion is always negative.
On the other hand, from Eqs. (4.17), (4.19) and (4.30) we find that

3c0 0 4 3/2
d
Yc ( ) =
,
2
3
[( + 0 )2 02 ]1/4

3/2

3/4
4
( + 0 )2 02
.
Tc ( ) =
(4.31)
3
Thus, we have Tc (0) = Yc (0) = 0, and
Tc ( ) 3/2 ,
Yc ( ) 1/2 ,
(4.32)
as . Then, one can see that the trajectory of the visible 3-brane in the (T , Y )-plane is
given by the solid curve OAP for c0 = +1 and OBP for c0 = 1, respectively, in Fig. 12.
Similarly, it can be shown that the trajectory of the invisible 3-brane in the (T , Y )-plane is
given by the dotted curve OA P for c0 = +1 and OB P for c0 = 1 in Fig. 12. Then, the fivedimensional bulk is the region between the solid and dotted curves in each case. The spacetimes
of both visible and invisible 3-branes are singular at the initial moment = 0.
421
When w = 13 , using the integral,

dx
x 2n + a 2n

2n
x
1 1
1
x
= nF
, ;1 + ;
,
a
2 2n
2n
a
we find that Eq. (4.28) has the solution,

2n
2n
1/2

X0
1
X
1 1
,
, ;1 + ;
=
XF
2 2n
2n
X0
n
1/2n
where X0 = X1
(4.33)
(4.34)
, and F (a, b; c; z) denotes the ordinary hypergeometric function with
F (a, b; c; 0) = 1.
(4.35)
Then, from Eqs. (4.23), (4.34) and (4.35), we find

ac ( = 0) = 0.
(4.36)
On the other hand, using the relation,

F (a, b; c; z) =
we find
XF
1
(c)(b a)
(z)a F a, 1 c + a; 1 b + a;
(b)(c a)
z

(c)(a b)
1
+
(z)b F b, 1 c + b; 1 a + b;
,
(a)(c b)
z

2n
X0n 1n
1
X
1 1
,
, ;1 + ;
X

2 2n
2n
X0
1n
(4.37)
(4.38)
as X . Therefore, we must have

ac ( = ) = .
(4.39)
From the above expressions, we also find that

a c ( ) =
1
6w+5
X 1+w + (3w + 2)X02n X 3(1+w) .

3(1 + w)(1 + 3w)
(4.40)
Thus, in the present case the universe is always decelerating. It can also be shown that
c ( ) =
0
0
,
X
6c0 c X0n
3(1 + w)(1 + 3w)52
(4.41)
From Eqs. (4.17), (4.19) and (4.23), on the other hand, we find that

2n
1
1
3
5 + 6w
X
,
;
;
,
Yc ( ) = c0 3 X 1+w F
2 2(1 + 3w) 2(1 + 3w)
X0
1
Tc ( ) = ac3 ( ) = 3 X 1+w .
Thus, we have Tc (0) = Yc (0) = 0, and
(4.42)
422

1
Tc ( ) X 1+w ,

23w
3X0n
0 3
3(1+w)
+ C0 ,
X
Yc ( ) c
2 3w
as (or X ), where

1/(1+w)
3X0
3
3w 2

.
C0
2(1 + 3w)
2(1 + 3w)
2(1 + 3w) 1/2
(4.43)
(4.44)
Since 1/(1 + w) > (2 3w)/[3(1 + w)] for w > 1/3, we find that the trajectory of the visible
3-brane in the (T , Y )-plane is also given by the curves OAP and OBP in Fig. 12, respectively,
for c0 = 1 and c0 = 1. The trajectory of the invisible 3-brane at y = 0 is similar to the visible
3-brane, as indicated by the dotted vertical curves.
4.3.2. w = 13
In this case, Eq. (4.22) has the solution,
ac ( ) = a0 1/2
(w = 1/3),
(4.45)
where a0 is a positive constant. Then, the corresponding energy density is given by

c0
,
3c0 c
16 1/2
0
c
.
1 4
5
9a0
c ( ) =
(4.46)
It is interesting to note that choosing a0 = 3/4 we have c = 0, and the spacetime on the
3-brane becomes vacuum.
To study the global structure of the 3-branes in this case, we first notice that
Tc ( ) = a03 3/2 ,
Yc ( ) = Yc0 3/2 ,

16 1/2
Yc0 c0 a03 1 4
.
9a0
(4.47)
Since |Yc0 | < a03 , it can be shown that the corresponding Penrose diagram for the visible and
invisible 3-branes is also given by Fig. 12.
4.3.3. 1 < w < 13
In this case, choosing the constant as that given by Eq. (4.25), we find that Eq. (4.24)
becomes
X + X = 0,
(4.48)
where is still given by Eq. (4.27). It can be shown that Eq. (4.48) has the first integral,

X 2m + X1
X =
(4.49)
,
m
423
where
m
1 + 3w
1
=
> 0.
2
3(1 + w)
(4.50)
When w = 23 , Eq. (4.49) has the solution,

1/2
4
(w = 2/3),
ac ( ) = a0 + 2
3a0
(4.51)
where a0 is a positive constant. Then, we find that

a c ( ) = a0
+
[ ( +
2
3a02
,
4
)]1/2
3a02
4
,
9a03 [ ( + 4 2 )]3/2
3a0

1/2
0
6c c
4
c ( ) =
,
+ 2
5
3a0
a c ( ) =
(4.52)
from which we can see that the universe starts to expand from the singular point = 0. In this
case the acceleration of the expansion is always negative.
On the other hand, it can be shown that

3/2
4
Tc ( ) = a03 + 2
,
3a0

2
0 3 2
Yc ( ) = c a0 + 2 .
(4.53)
a0
Then, we find that Tc ( ) |Yc ( )| +, as . Hence, the trajectories of the visible and
invisible 3-branes in this case are also given by Fig. 12 in the (T , Y )-plane.
When w = 23 , Eq. (4.49) has the solution,

2m
1 1 + m 1 + 3m
X
1+m
1+m
X
(4.54)
F
,
;
;
= ,
2 2m
2m
X0
m
where X1 X02m . Then, it can be shown that

0,
= 0,
X( ) =
, ,
and that
c ( ) = c0 c
12X1
2
5 (1 |w|)(3|w| 1)
1/2
(4.55)
1
,
X
6|w|5

2(3|w|1)
X 3(1|w|)
a c ( ) =
X 3(1|w|) X1 3|w| 2 .
3(1 |w|)(3|w| 1)
(4.56)
Therefore, when 1 < w < 2/3, there exists a moment, f , after which a c ( ) becomes positive, that is, the universe turns to its accelerating phase from a decelerating one, where f is the
424
root of the equation,

X
2(3|w|1)
3(1|w|)

X1 3|w| 2 = 0.
(4.57)
On the other hand, it can be shown that

2m
3 1/2
23w
1 3w 2
9w
X
0 3 X1
3(1+w)
Yc ( ) = c
F
X
,
;
;
,
2 3w
2 2(1 + 3w) 2(1 + 3w)
X0
1
Tc ( ) = 3 X 1+w ,
(4.58)
from which we find that Tc (0) = 0 = Yc (0), and

1
Tc ( ) X 1+w ,
2
c0 Yc ( ) X 3(1+w) ,
(4.59)
as X (or ). Following the arguments given above it can be shown that in this case
the trajectories of the two 3-branes are given by Fig. 12, too.
4.3.4. w = 1
In this case, introducing the function X( ) via the relation
ac = eX ,
(4.60)
we find that Eq. (4.22) can be written as

8
X + e4X = 0,
(4.61)
9
which has the first integral,
2 4X
X =
(4.62)
e
+ X0 ,
3
where X0 is an integration constant. When X0 = 0 it can be shown that the corresponding
solu
tion represents a vacuum 3-brane (c = 0), as that given by Eq. (4.45) with a0 = 3/4. When
X0 < 0 there are no real solutions. When X0 > 0, Eq. (4.62) has the solution,

4
1/2
.
ac ( ) = a0 sinh
(4.63)
3a02
Then, we find that
c ( ) = pc =
a c ( ) =
4c0 c
,
5 a02
4
9a03 sinh3/2 (
4
)
3a02
sinh

4
1 ,
3a02
(4.64)
which shows that at the beginning the universe is decelerating, but after the moment = f it
starts to accelerate, where f is given by
f =
3a02
ln(1 + 2 ).
4
(4.65)
425
Fig. 13. The Penrose diagram of the solutions of Eq. (4.63) in the (T , Y )-plane for w = 1. The solid (dotted) curves
OA and OB are the trajectories of the visible (invisible) 3-branes, respectively, for c0 = +1 and c0 = 1.
It is interesting to note that in the present case we have c = pc = Const.

On the other hand, from Eqs. (4.17), (4.19) and (4.63) we find that

4
d ,
Yc ( ) = 2c0 a0 sinh3/2
3a02

4
3
3/2
.
Tc ( ) = a0 sinh
3a02
(4.66)
Clearly, in this case we have Tc (0) = 0 = Yc (0), and Tc ( ) |Yc ( )| , as . Then,

the trajectory of the visible 3-brane in the (T , Y )-plane is given by the solid curve OA for
c0 = +1 and OB for c0 = 1 in Fig. 13. Similarly, the trajectory of the invisible 3-brane is
given by the dotted curves OA and OB, respectively, for c0 = +1 and c0 = 1. The bulk is the
region between the solid and dotted curves.
4.3.5. w < 1
In this case, choosing the constant as that given by Eq. (4.25), Eq. (4.24) becomes
X X = 0,
but now with
1 + 3|w|
.
3(|w| 1)
Eq. (4.67) has the first integral,

X 2l + X0
,
X =
l
(4.67)
(4.68)
(4.69)
where
3|w| 1
.
(4.70)
3(|w| 1)
To have an expanding universe, now we must choose the sign in Eq. (4.69). Then, Eq. (4.69)
has the solution,

2l
2l
1/2
X0
1 1
1
X
XF
(4.71)
(s ),
, ;1 + ;
=
2 2l
2l
X0
l
l
426
where X1 X02l , and

l 1l
1
1
1
s
X0 1 +

2l
2 2l
From Eqs. (4.37) and (4.71) we find

0,
= s ,
X( ) =
, = 0,
(4.72)
(4.73)
and
, = ,
s
0,
= 0.
On the other hand, we also have

1/2
1
12X1
,
c ( ) = c0 c 2
X
5 (|w| 1)(3|w| 1)
ac ( ) = X
1
3(|w|1)
(4.74)
6|w|5
2(3|w|1)

X 3(|w|1)
a c ( ) =
X1 X 3(|w|1) ,
3(|w| 1)(3|w| 1)
from which we find

< 0, < f ,
a c ( ) =
> 0, > f ,
(4.75)
(4.76)
where f is defined by,

3(|w|1)
X(f ) = X12(3|w|1) .
(4.77)
Thus, in this case the universe also has an accelerating phase, although at the beginning it is
decelerating. From Eq. (4.75), on the other hand, we find that
, = ,
s
( ) =
(4.78)
0,
= 0,
that is, in these models a big rip singularity develops at s . In this case, it can also be shown that

1
c0 3 |w|1
6|w| 5
3|w| + 2

Yc = 1/2 X0
2(3|w| 1)
2(3|w| 1)

2l
1
1
3
6|w| 5
X
+ c0 3 X |w|1 F
,
;
;
,
2 2(1 3|w|) 2(3|w| 1)
X0
1
Tc = 3 X |w|1 .
(4.79)
Thus, as 0 (or X ), we have Tc ( ), Yc ( ) 0, and as X 0 (or s ) we have

1
c0 Yc ( ) 3 X |w|1 +,
1
Tc ( ) 3 X |w|1 +,

Tc ( ) Yc ( ) 0.
(4.80)
Then, the trajectory of the visible 3-branes is given by the solid curve OA for c0 = +1 and OB
for c0 = 1 in Fig. 14. Similarly, it can be shown that the trajectory of the invisible 3-branes
427
Fig. 14. The Penrose diagram of the solutions of Eq. (4.71) in the (T , Y )-plane for w < 1. The solid (dotted) curves
OA and OB are the trajectories of the visible (invisible) 3-branes, respectively, for c0 = +1 and c0 = 1. At the point
O the five-dimensional bulk is singular, but the spacetimes of the two 3-branes are not, while at the points A and B the
spacetimes of the two 3-branes have big rip singularities, but the bulk is regular.
is given by the dotted curve OA for c0 = +1 and OB for c0 = 1. The spacetime of the fivedimensional bulk is the region between the solid and dotted curves. It is interesting to note that
the bulk is singular at the point O (T = 0), but the spacetimes of the two 3-branes are not. On the
other hand, at the points A and B the spacetimes of the two 3-branes have big rip singularities,
but the bulk is regular there.
5. Conclusions and remarks
In this paper, we have systematically studied cosmological models in the RandallSundrum
setting of two 3-branes [2]. By assuming that the two orbifold 3-branes are spatially homogeneous, isotropic, and independent of time, we have derived the most general form of metric with
orbifold symmetry, given by Eqs. (2.9) and (2.10). Then, using distribution theory, we have developed the general formulas for two orbifold 3-branes. In particular, the field equations have
been divided into three sets, one holds in between the two 3-branes, given by Eqs. (2.7) and
(2.25), one holds on each of the two 3-branes, given, respectively, by Eqs. (2.21), (2.26), (2.22),
and (2.27).
Then, in Section 3 we have applied our general formulas to the case where the bulk is an
AdS5 space. Assuming that the cosmological constant on the visible brane vanishes and that the
equation of state of the matter field takes the form, pc = wc , with w being a constant, we have
been able to find the explicit expression for the expansion factor ac ( ), where denotes the
proper time of the visible brane. Although the curvature of the three spatial space of the brane
is zero, it has been shown that for w > 1 the universe starts to expand from a big-bang like
singularity at = 0, until the moment = /[6(1 + w)], where (/6)1/2 , and is the
cosmological constant of the bulk. From that moment on the universe starts to contract until the
moment = /[2(1 + w)], where the whole universe is crushed into a spacetime singularity.
For 1 < w < 2/3 the universe experiences an accelerating period. The global structure of
the bulk and the embedding of the 3-branes in the bulk have also been studied, and found that in
some cases the geodesically complete spacetimes contain infinite number of 3-branes [cf. Fig. 8].
In Section 4, similar considerations have been carried out for the case where the bulk is
vacuum. It has been shown that the vacuum solutions can be classified into three different
families, Class IA, Class IB, and Class II, given, respectively, by Eqs. (4.1), (4.13) and (4.16).
For Classes IA and IB solutions, it has been found that the universe is always accelerating for
428
w < 2/3, while for Class II solutions with w < 2/3 the universe is first decelerating and then
accelerating. The study of the global structure of the bulk as well as the 3-branes have shown that
some solutions may represent the collision of two orbifold 3-branes [cf. Fig. 10].
It would be very interesting to look for solutions where the bulk is filled with other matter
fields. In particular, look for solutions where acceleration of the universe is realized without dark
energy, similar to the one-brane case [3].
Another interesting application of our general formulas is to cyclic universe scenario [16] to
study the collision of two orbifold 3-branes and the cosmological constant problem [17].
Acknowledgements
One of the author (A.W.) would like to thank the Center of Astrophysics, Zhejiang University
of Technology, for hospitality, and Baylor University for Summer 2006 sabbatical leave. R.G.C.
was supported by a grant from Chinese Academy of Sciences, and grants from NSFC, China
(Nos. 10325525 and 90403029).
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
[2] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[3] V.A. Rubakov, Phys. Usp. 44 (2001) 871;
S. Frste, Fortschr. Phys. 50 (2002) 221;
C.P. Burgess, et al., JHEP 0201 (2002) 014;
E. Papantonopoulos, Lect. Notes Phys. 592 (2002) 458;
R. Maartens, Living Rev. Rel. 7 (2004);
P. Brax, C. van de Bruck, A.C. Davis, Rep. Prog. Phys. 67 (2004) 2183, hep-th/0404011;
U. Gnther, A. Zhuk, Phenomenology of brane-world cosmological models, gr-qc/0410130;
P. Brax, C. van de Bruck, A.C. Davis, Brane world cosmology, Rep. Prog. Phys. 67 (2004) 2183, hep-th/0404011;
V. Sahni, Cosmological surprises from braneworld models of dark energy, astro-ph/0502032;
R. Durrer, Braneworlds, hep-th/0507006;
D. Langlois, Is our universe brany?, hep-th/0509231;
A. Lue, Phenomenology of DvaliGabadadzePorrati cosmologies, Phys. Rep. 423 (2006) 1, astro-ph/0510068;
D. Wands, Brane-world cosmology, gr-qc/0601078;
R. Maartens, Dark energy from brane-world gravity, astro-ph/0602415.
[4] C.D. Hoyle, et al., Phys. Rev. Lett. 86 (2001) 1418;
J. Chiaverini, et al., Phys. Rev. Lett. 90 (2003) 151101;
J.C. Long, et al., Nature (London) 421 (2003) 922;
C.D. Hoyle, et al., Phys. Rev. D 70 (2004) 042004, hep-ph/0405262.
[5] A. Pomarol, Phys. Lett. B 486 (2000) 153;
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Lett. B 473 (2000) 43;
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Rev. Lett. 84 (2000) 2080;
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Rev. D 63 (2001) 075004;
E. Pree, M. Sher, Phys. Rev. D 73 (2006) 095006, and references therein.
[6] T.G. Rizzo, Pedagogical introduction to extra dimensions, hep-ph/0409309;
C. Cski, TASI lectures on extra dimensions and branes, hep-th/0404096.
[7] W.D. Goldberger, M.B. Wise, TASI lectures on extra dimensions and branes, Phys. Rev. Lett. 83 (1999) 4922;
W.D. Goldberger, M.B. Wise, Phys. Rev. D 60 (1999) 107505;
C. Charmousis, R. Gregory, V.A. Rubakov, Phys. Rev. D 62 (2000) 067505;
C. Csaki, M. Graesser, L. Randall, J. Terning, Phys. Rev. D 62 (2000) 045015;
B. Grinstein, D.R. Nolte, W. Skiba, Phys. Rev. D 63 (2001) 105016;
U. Gen, M. Sasaki, Prog. Theor. Phys. 105 (2001) 591;
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
429
G.D. Kribs, Physics of the radion in the RandallSundrum scenario, eConf. C 010630 (2001) P317, hep-ph/0110242;
J.E. Kim, B. Kyae, H.M. Lee, Phys. Rev. D 66 (2002) 106004;
S. Bae, P. Ko, H.S. Lee, J. Lee, Radion phenomenology in the RandallSundrum scenario, hep-ph/0103187;
A. Mazumdar, A. Prez-Lorenzana, Phys. Lett. B 508 (2001) 340;
Z. Chacko, P.J. Fox, Phys. Rev. D 64 (2001) 024015;
P. Bintruy, C. Deffayet, D. Langlois, Nucl. Phys. B 615 (2001) 219;
S. Kanno, J. Soda, Phys. Rev. D 66 (2002) 083506;
S. Nojiri, S.D. Odintsov, A. Sugamoto, Mod. Phys. Lett. A 17 (2002) 1269;
P. Brax, C. van de Bruck, A.C. Davis, C.S. Rhodes, Phys. Lett. B 531 (2002) 135;
S. Nasri, P.J. Silva, G.D. Starkman, M. Trodden, Phys. Rev. D 66 (2002) 045029;
B. Grzadkowski, J.F. Gunion, Phys. Rev. D 68 (2003) 055002;
S. Mukohyama, A. Coley, Phys. Rev. D 69 (2004) 064029;
A. Mazumdar, R.N. Mohapatra, A. Prez-Lorenzana, JCAP 0406 (2004) 004;
M. Toharia, Mod. Phys. Lett. A 19 (2004) 37;
P.R. Ashcroft, C. van de Bruck, A.-C. Davis, Phys. Rev. D 71 (2005) 023508;
C. Charmousis, U. Ellwanger, JHEP 0402 (2004) 058;
G.L. Alberghi, D. Bombardelli, R. Casadio, A. Tronconi, Phys. Rev. D 72 (2005) 025005;
G.L. Alberghi, A. Tronconi, Phys. Rev. D 73 (2006) 027702;
D. Konikowska, M. Olechowski, M.G. Schmidt, Phys. Rev. D 73 (2006) 105018;
E.E. Boos, Y.S. Mikhailov, M.N. Smolyakov, I.P. Volobuev, Mod. Phys. Lett. A 21 (2006) 1431;
A.S. Mikhailov, Y.S. Mikhailov, M.N. Smolyakov, I.P. Volobuev, hep-th/0602143, and references therein.
P. Kraus, JHEP 9912 (1999) 011;
H.A. Chamblin, M.J. Perry, H. Reall, JHEP 9909 (1999) 014;
D. Ida, JHEP 0009 (2000) 014;
L. Anchordoqui, C. Nunez, K. Olsen, JHEP 0010 (2000) 050;
H.A. Chamblin, H. Reall, Nucl. Phys. B 562 (1999) 133;
S. Mukohyama, T. Shiromizu, K.-I. Maeda, Phys. Rev. D 62 (2000) 024028;
S.W. Hawking, T. Hertog, H.S. Reall, Phys. Rev. D 62 (2000) 043501;
S. Gubser, Phys. Rev. D 63 (2001) 084017;
Y. Wu, M.F.A. da Silva, N.O. Santos, A. Wang, Phys. Rev. D 68 (2003) 084012.
C. Cski, M. Graesser, C. Kolda, J. Terning, Phys. Lett. B 462 (1999) 34;
T. Nihei, Phys. Lett. B 465 (1999) 81;
J. Cline, C. Grojean, G. Servant, Phys. Rev. Lett. 83 (1999) 4245;
N. Kaloper, Phys. Rev. D 60 (1999) 123506;
D.J. Chung, K. Freese, Phys. Rev. D 61 (2000) 023511;
T. Shiromizu, K.-I. Maeda, M. Sasaki, Phys. Rev. D 62 (2000) 024012;
R. Maartens, Phys. Rev. D 62 (2000) 084023;
P. Bintruy, C. Deffayet, U. Ellwanger, D. Langlois, Nucl. Phys. B 477 (2000) 285;
L. Mersini, Mod. Phys. Lett. A 16 (2001) 1583;
H.-Y. Liu, P.S. Wesson, Astrophys. J. 562 (2001) 1;
A.N. Aliev, A.E. Gumrukcuoglu, Class. Quantum Grav. 21 (2004) 5081;
C. de Rham, S. Webster, Phys. Rev. D 71 (2005) 124025;
C. de Rham, S. Webster, Phys. Rev. D 72 (2005) 064013;
C. de Rham, S. Fujii, T. Shiromizu, H. Yoshino, Phys. Rev. D 72 (2005) 123522;
T. Shiromizu, S. Fujii, C. de Rham, H. Yoshino, Phys. Rev. D 73 (2006) 087301.
P. Bowcock, C. Charmousis, R. Gregory, Class. Quantum Grav. 17 (2000) 4745.
J.A. Wolf, Spaces of Constant Curvature, fifth ed., Publish or Perish, Inc., Wilmington, Delaware, USA, 1984.
G.W. Gibbons, Nucl. Phys. B 394 (1993) 3.
M. Cvetic, S. Griffies, H.H. Soleng, Phys. Rev. D 48 (1993) 2613;
S. Griffies, Field theoretic and spacetime aspects of supersymmetric walls and the spacetime aspects of nonsupersymmetric vacuum bubbles, PhD dissertation, University of Pennsylvania, 1993.
S.W. Hawking, G.F.R. Ellis, The Large Scale Structure of SpaceTime, Cambridge Univ. Press, Cambridge, 1973.
A.H. Taub, Ann. Math. 53 (1951) 472, Reprinted in Gen. Relativ. Gravit. 36 (2004) 2699;
A.H. Taub, Phys. Rev. 103 (1956) 454;
430
J. Ipser, P. Sikivie, Phys. Rev. D 30 (1984) 712.

[16] J. Khoury, B.A. Ovrut, P.J. Steinhardt, N. Turok, Phys. Rev. D 64 (2001) 123522;
J. Khoury, B.A. Ovrut, N. Seiberg, P.J. Steinhardt, N. Turok, Phys. Rev. D 65 (2002) 086007;
P.J. Steinhardt, N. Turok, Phys. Rev. D 65 (2002) 126003;
P.J. Steinhardt, N. Turok, Science 296 (2002) 1436;
N. Jones, H. Stoica, S.H.H. Tye, JHEP 0207 (2002) 051;
J. Khoury, P.J. Steinhardt, N. Turok, Phys. Rev. Lett. 92 (2004) 031302;
J. Khoury, A briefing on the ekpyrotic/cyclic universe, astro-ph/0401579;
N. Turok, P.J. Steinhardt, Beyond inflation: A cyclic universe scenario, hep-th/0403020;
A.J. Tolley, N. Turok, P.J. Steinhardt, Phys. Rev. D 69 (2004) 106005;
Y.-S. Piao, Phys. Rev. D 70 (2004) 101302;
N. Turok, M. Perry, P.J. Steinhardt, Phys. Rev. D 70 (2004) 106004;
N. Turok, M. Perry, P.J. Steinhardt, Phys. Rev. D 71 (2005) 029901, Erratum;
Y.-I. Takamizu, K.-I. Maeda, Phys. Rev. D 70 (2004) 123514;
Y.-I. Takamizu, K.-I. Maeda, Phys. Rev. D 73 (2006) 103508;
G.W. Gibbons, H. Lu, C.N. Pope, Phys. Rev. Lett. 94 (2005) 131602;
T.J. Battefeld, S.P. Patil, R.H. Brandenberger, Phys. Rev. D 73 (2006) 086002;
A.J. Tolley, Phys. Rev. D 73 (2006) 123522;
P.J. Steinhardt, N. Turok, Science 312 (2006) 1180, and references therein.
[17] S. Weinberg, Rev. Mod. Phys. 61 (1989) 1;
S.M. Carroll, The cosmological constant, astro-ph/0004075;
T. Padmanabhan, Phys. Rep. 380 (2003) 235;
S. Nobbenhuis, Categorizing different approaches to the cosmological constant problem, gr-qc/0411093;
J. Polchiski, The cosmological constant and the string landscape, hep-th/0603249, and references therein.

One-loop corrections to the mass

of self-dual semi-local planar topological solitons
A. Alonso Izquierdo a , W. Garcia Fuertes b ,
M. de la Torre Mayado c , J. Mateos Guilarte d,
a Departamento de Matematica Aplicada, Universidad de Salamanca, Spain
b Departamento de Fisica, Universidad de Oviedo, Spain
c Departamento de Fisica, Universidad de Salamanca, Spain
d Departamento de Fisica and IUFFyM, Universidad de Salamanca, Spain
Received 10 September 2007; received in revised form 16 November 2007; accepted 19 November 2007
Available online 28 November 2007
Abstract
A formula is derived that allows the computation of one-loop mass shifts for self-dual semilocal topological solitons. These extended objects, which in three spatial dimensions are called semi-local strings,
arise in a generalized Abelian Higgs model with a doublet of complex Higgs fields. Having a mixture of
global, SU(2), and local (gauge), U (1), symmetries, this weird system may seem bizarre, but it is in fact the
bosonic sector of electro-weak theory when the weak mixing angle is 2 . The procedure for computing the
semi-classical mass shifts is based on canonical quantization and heat kernel/zeta function regularization
methods.
PACS: 03.70.+k; 11.15.Kc; 11.15.Ex
Keywords: Semi-local topological solitons; Heat-kernel/zeta function regularization; One-loop shifts to soliton masses
1. Introduction
The purpose of this paper is to address the computation and analysis of one-loop shifts to the
masses of semi-local planar topological solitons arising in a natural generalization of the Abelian
E-mail address: guilarte@usal.es (J.M. Guilarte).

doi:10.1016/j.nuclphysb.2007.11.023
432
A. Alonso Izquierdo et al. / Nuclear Physics B 797 [PM] (2008) 431463
Higgs model. Seen in (3 + 1)-dimensional spacetime these solitons become semi-local strings
whereas their masses give the string tensions, see [1,2]. At the critical point that marks the phase
transition between type I and type II superconductivity, the set of semi-local self-dual topological
solitons is an interesting 4l-dimensional moduli space, where l is the number of quanta of the
magnetic flux, see [3,4]. Recently, superconducting semi-local (non-self-dual) strings with very
intriguing properties have been discovered in this model [5].
Computations of one-loop mass corrections will be performed using the heat kernel/zeta function regularization method. The high-temperature asymptotic expansion of the heat function, see
[69], is a powerful tool that was applied for the first time to the calculation of kink mass shifts
in [10] (the N = 1 SUSY case) and [11] (the non-SUSY case). One-loop corrections to N = 2
supersymmetric self-dual NielsenOlesen vortices were computed in a similar approach by Vassilevich and Rebhanvan NieuwenhuizenWimmer in Refs. [12] and [13]. In the second paper,
the authors also showed that the central charge of the SUSY algebra is modified in one-loop order
in such a way that the Bogomolny bound is saturated at the semi-classical level. Sometime later,
we calculated the one-loop mass shift for N = 0 (non-SUSY) self-dual NO vortices carrying a
quantum of magnetic flux in Ref. [15]. Mass shifts have been given for spherically symmetric
self-dual vortices (when several solitons of a quantum of flux have coinciding centers) in [16] up
to four magnetic flux quanta. Cruder approximations were also provided for the mass shift of two
separated self-dual NO vorticeseach of them with a quantum of magnetic fluxas a function
of the inter-center distance.
In Ref. [17], we studied the one-loop correction to the energy of a degenerate manifold of
kinks that arise in a very interesting family of models with two real scalar fields in (1 + 1)
dimensions. These field theoretical systems are obtained through dimensional reductionplus
a reality conditionof an N = 1 supersymmetric WessZumino model with two chiral superfields, see [1820]. A comparison between the mass shifts of these composite kinks and the
correction to the mass of the ordinary 24 kink was offered in [21]. In this paper, we address
a similar, but more difficult, situation in (2 + 1) dimensions, comparing one-loop mass corrections of the topological solitons that arise in two planar Abelian gauge systems; one with two
complex scalar fields and the other with a single complex scalar field. The methodology used to
accomplish this task is explained in detail in Ref. [22], where a complete list of references can
be found.
Our paper is organized as follows: in Section 2 we describe the model and develop perturbation theory around one of the vacua. A one-loop renormalization is also performed. Section 3
is devoted to summarizing the structure of the moduli space of self-dual topological solitons.
As a novelty, we also apply a variation of the de VegaShaposnik method [23] to find numerical solutions for spherically symmetric topological solitons. In Section 4, we explain how to
obtain one-loop mass shifts in terms of generalized zeta functions of the second-order differential operators ruling the small fluctuations of the bosonic and ghost fields, and in Section 5
the high-temperature expansion of the heat kernel is used to give the final formula for one-loop
mass shifts of semi-local self-dual topological solitons after application of Mellins transforms.
We present our results in Section 6 by means of Mathematica calculations of the coefficients of
the asymptotic series giving the heat functions. Finally, in Appendix A we offer Tables A.1A.5
where these coefficients are shown.
433
2. The planar semi-local Abelian Higgs model

2.1. The model
The semi-local Abelian Higgs model [1] describes the minimal coupling between an U (1)gauge field A and a doublet of complex scalar fields in a phase where the Higgs mechanism
takes place. The term semi-local refers to the fact that while the global symmetry of this system
is SU(2) U (1) only the U (1) factor is gauged. Defining non-dimensional spacetime variables,
1
x , and fields, v, A vA , from the vacuum expectation value of the Higgs
x ev
field v and the U (1)-gauge coupling constant, e, the action for the semi-local Abelian Higgs
model in (2 + 1) dimensions reads:

2
1
1
2
v
3

d x F F + (D ) D
x 1 ,
S=
e
4
2
8
where the covariant derivative is defined as D = ( x iA ). The physical spectrum differs
from that of the standard AHM model in that, along with the massive vector boson and Higgs
scalar, there is a complex Goldstone field. The parameter 2 = e2 measures the ratio between
the square of the masses of the Higgs, M 2 = v 2 , and vector particles, m2 = e2 v 2 . Here, is
the Higgs field self-coupling. We choose a system of units where c = 1, but h has dimensions of
length mass.
2.2. Feynman rules in the Feynmant Hooft R gauge
In order to obtain the Feynman rules, we expand the action around a classical vacuum. We
shall profit from the SU(2) global invariance and we choose a vacuum V with real upper and
vanishing lower components. The shift of the fields

1 + H (x ) + iG(x )
x =
2(x )
corresponds to the identification of H (x ), G(x ), and (x ), respectively as Higgs, real, and
complex Goldstone fields. The choice of the Feynmant Hooft R gauge
R(A , G) = A + G
requires a FaddeevPopov determinant to restore unitarity, which amounts to introducing a complex ghost field . All this together allows us to write the action in the form

1
v
3
d x A g (2 + 1) A +
S=
e
2
1
1
2
1
+ G G G2 + H H H 2 +
2
2
2
2

2 2
H H + G2 + A H G GH + H A A
2

2 1

2 2
+ iA + A A ||2
H + G2 + G2 + H 2 A A
8
2

2 2 2
|| || + H 2 + G2 + 2H ,
2
434
Table 1
Propagators
Particle
Field
Higgs
H (x)
Propagator
Real Goldstone
G(x)
Complex Goldstone
(x)
Ghost
(x)
Vector boson
A (x)
Diagram
ieh
v(k 2 2 +i)
ieh
v(k 2 1+i)
ieh
v(k 2 +i)
ieh
v(k 2 1+i)
ieh g
v(k 2 1+i)
Table 2
Third- and fourth-order vertices
Vertex
Weight
Vertex
Weight
Vertex
Weight
Vertex
Weight
3i 2 hve
3i 2 hve
i(k + q ) hve
2ig hve
i 2 hve
3i 2 hve
(k q ) hve
2i hve g
2i hve g
i 2 hve
2i 2 hve
i 2 hve
i hve
2i hve g
i 2 hve
i 2 hve
in order to deduce the Feynman rules for the propagators and vertices shown in Tables 1 and 2.
The propagator of the complex Goldstone boson plus two trivalent and four tetravalent vertices
accounting for the interactions of Goldstoneanti-Goldstone pairs must be added to the Feynman
rules of the Abelian Higgs model in the Feynmant Hooft gauge. Nevertheless, in order to
guarantee that all physical quantities will be ultraviolet finite there is still a remaining piece to be
added: the action for the counter-terms. We shall compute this up to one-loop order in the sequel.
2.3. One-loop mass renormalization counter-terms
All ultraviolet divergences come from the integral:

I c2 =
d 3k
i
.
(2)3 k 2 c2 + i
We also note that even if there are massless particles propagating in (2 + 1) dimensions, I (0) is
infrared convergent. The renormalization of the semi-local AHM requires the computation of the
following one-loop graphs:
435
1. Higgs boson tadpole:

+
+
+
+
2

= 2i + 1 I (1) i 2 I (0) + finite part.
2. Higgs boson self-energy:
+
= 2i 2 + 1 I (1) i 2 I (0) + finite part.

3. Real Goldstone boson self-energy:
+
= 2i 2 + 1 I (1) i 2 I (0) + finite part.

4. Complex Goldstone boson self-energy:
= 4i 2 + 1 I (1) 2i 2 I (0) + finite part.

5. Vector boson self-energy: The potentially divergent part is
+
+
+

= 2i I (1) + I (0) g + finite part.
In (2 + 1) dimensions the graphs above are the only ultraviolet divergent diagrams for
any number of loops in the diagrams, not only in one-loop order, and the theory is superrenormalizable. Thus, in a minimal subtraction scheme we get rid off all ultraviolet divergences
arising in the vacuum sector of the model by adding the counter-terms

h 2
2 + 1 I (1) + 2 I (0) 1 x 1 x + 2 x 2 x 1 ,
2

LA
I (1) + I (0) A x A x ,
c.t. = h
LSc.t. =
which gives rise to the vertices shown in Table 3. In our renormalization scheme, finite renormalizations have been adjusted in such a way that, on one hand, the divergence due to the tadpole
graph is exactly canceled when the theory is fine-tuned to the so called self-dual limit 2 = 1
and, on the other hand, the global SU(2) symmetry remains unbroken up to one-loop order.
3. Semi-local vortices
3.1. Finite energy solutions
Apart from the ground states built around the S 3 classical homogeneous solutions, the semilocal AHM has room for other quantum states arising from the extended classical configurations,
436
Table 3
Counter-term vertices
Diagram
Weight
i[2( 2 + 1)I (1) + 2 I (0)]
i[2( 2 + 1)I (1) + 2 I (0)]
i[2( 2 + 1)I (1) + 2 I (0)]
i[4( 2 + 1)I (1) + 2 2 I (0)]
2i[I (1) + I (0)]g
which are stable owing to topological reasons. These configurations are the topological solitons
that appear at tree level as non-homogeneous solutions of the static field equations:
i Fij =

i
Dj (Dj ) ,
2
Di Di =

2
1 ,
4
such that their energy,

2
1
1
2
2
1 ,
E = d x Fij Fij + (Di ) Di +
4
2
8
is finite. The topological character of these solutions is peculiar in that although the scalar vacuum manifold is S 3 , and thus simply connected, the configuration space C

C = (
x ) Maps R2 , C2 , Ai (
x ) Maps R2 , T R2 /E(, Ai ) < +
is the union of Z disconnected sectors. Finite energy configurations show the asymptotic behavior

S 1 = 1,
Di |S
1 = (i iAi ) 1 = 0,
S
1 is the circle that bounds the plane at infinity. Parametrizing S 1 by the polar angle
where S
V for = 0, the
= arctan xx21 , and, modulo the global SU(2)-symmetry, choosing |S
1 =
1
1
boundary conditions on the covariant derivatives provide a map, S S1 , between the sphere
at the infinity spatial and the fiber S11 at the north pole of the base S 2 of the S 3 vacuum Hopf
bundle:

il

1 |S
1 |S
1
1 + i2 |S 1
ce
=
=
, l Z.
|S
1 =
0
2 |S
3 |S
1
1 + i4 |S 1
Continuous maps between one-dimensional spheres are classified according to the first homotopy group and, because the temporal evolution is continuous, 0 (C) = 1 (S11 ) = Z, the zero
homotopy group of C is non-trivial. Thus, C = lZ Cl is the union of disconnected sectors characterized by an integer number l.
Moreover, the boundary condition for the vector field Ai |S

1 = i i |S 1 is tantamount
to:
437

2
1
Ai (
x ) S 1 = 1 (
x)
(
x ) S 1 2 (
x)
(
x ) S1
xi
xi

4
3
+ 3 (
x)
(
x ) S 1 4 (
x)
(
x ) S1 .
xi
xi

il
The SU(2)-orbit of lV = e is:
0
i
il i( +l)

1
e sin
e
e 1
ei2 cos
sin
GlV =
=
,
ei2 cos ei1 sin
ei(2 l) cos
0
where 1 , 2 [0, 2] and [0, 2 ] are the Hopf coordinates of the S 3 sphere. Therefore,

2
cos (1 + l) sin2 + sin2 (1 + l) sin2
x ) S 1 = l
Ai (
xi

2
+l
cos (2 l) cos2 + sin2 (2 l) cos2
xi
=l
xi
and the topological (winding) number l has a direct physical interpretation in terms of the magnetic flux carried by the planar soliton:

g=

A1 (
x ) dx 1 + A2 (
x ) dx 2 = l

1
S

2
dx 1 +
dx 2 = l d = 2l.
x1
x2
0
3.2. Self-dual semi-local topological solitons

We shall restrict ourselves to the critical point between type I and type II superconductivity:
2 = 1. The energy can be arranged in a Bogomolny splitting:

2
2
d x
1
1
E=
(D1 iD2 ) (D1 iD2 ) + F12 1
+ |g|.
2
2
2
One immediately realizes that the solutions of the first-order equations

1
F12 1 = 0
D1 iD2 = 0,
(1)
2
are absolute minima of the energy, and are hence stable, in each topological sector that has a classical mass proportional to the magnetic flux. Because the first-order equations can be obtained
from the self-duality equations of Euclidean 4D gauge theory through dimensional reduction, the
vortex solutions of (1) are called self-dual at the limit 2 = 1.
We follow [3] to summarize the properties and existence of the so-called semi-local self-dual
topological solitons: the solutions of (1) for non-negative l (plus sign in the first-order equations).1 The equation on the left in (1) is tantamount to:

z ) = iz log a (z, z ), a = 1, 2, (z, z ) = 1 (z, z ) ,
A(z,
(2)
2 (z, z )
1 It is trivial to obtain the solutions for negative l if the solutions for positive l are known; simply, complex conjugation
gives the solution of Eq. (1) with the other signs.
438
where the complex notation for coordinates and fields is:

z = x1 + ix2 ,
1
z = (1 i2 ),
2
From (2), one sees that:

2 (z, z )
=0
z log
1 (z, z )
(z) =
1
A = (A1 iA2 ).
2
2 (z, z ) Qm (z)
=
1 (z, z )
Pl (z)
where Qm and Pl are polynomials of respective degree m and l in z, such that (z) is locally analytic. The behavior of the Higgs field at infinity (up to global SU(2) transformations) compatible
with finite energy requires that m < l and Pl (z) be monic:
Pl (z) = zl + pl1 zl1 + + p1 z + p0 ,
Ql1 (z) = ql1 zl1 + + q1 z + q0 .
Therefore, the moduli space of semi-local topological solitons depends on 2l complex (4l real)
parameters: (pa , qa ), a = 0, 1, . . . , l 1. Varying the values of (pa , qa ) one varies the l zeroes
of 1 and the l 1 zeroes of 2 in such a way that the locations of the zeroes of the two Higgs
fields plus the scale and orientation of 2 parametrize the SSTS moduli space. Eq. (1) on the
right becomes

u(z, z ) + 1 eu(z,z) = log |Pl |2 (z) + |Ql1 |2 (z) ,
(3)
where

u(z, z ) = (z, z ) + log 1 + ||2 (z) ,
(z, z ) = log 1 1 (z, z ).
By functional analysis techniques, it is possible to demonstrate the existence and uniqueness of

a solution of (3) compatible with finite energy boundary conditions [3,4]. Therefore, the finite
energy solutions of (1) with a magnetic flux 2l take the form

(l)
1 (z, z )
1
1
Pl (z)

=
e 2 u(z,z) ,
(l)
2
2
(z)
Q
l1
2 (z, z )
|Pl | (z) + |Ql1 | (z)
where u(z, z ) is a solution of (3). In the l = 1 case we have that:

(l=1)

(z, z ; q0 )
1
1
1
z z0
=
e 2 u(|zz0 |;q0 ) .
(l=1)
q
2
2
0
2
(z, z ; q0 )
|z z0 | + |q0 |
Setting the parameter q0 to zero, we find the embedded NielsenOlesen vortex centered at z0 :
zz0
(l=1)
(z, z ; 0)
1
1
= |zz0 | e 2 u(|zz0 |;0) ,
(l=1)
0
2
(z, z ; 0)

4|q0 |2
,
log |z z0 |2 + |q0 |2 =
(|z z0 |2 + |q0 |2 )2
however, tends to zero for very large |q0 | and the solution of (3) becomes: u(|z z0 |;
|q0 | 1) 0. Therefore,

(l=1)
(z, z ; |q0 | 1)
1
1
z z0
=
,
(l=1)
q0
2
(z, z ; |q0 | 1)
|z z0 |2 + |q0 |2
is precisely the field profile of the lump centered at z0 with radius |q0 | and topological charge 1
in the planar CP1 model. The Higgs fields spreads over the vacuum manifold S 3 for very large
439
|q0 | whereas the ANO profiles are found for very small |q0 |. Self-dual semi-local topological
solitons interpolate between self-dual ANO vortices and CP1 -lumps when |q0 | varies between 0
and .
3.3. Self-dual semi-local topological solitons with spherical symmetry
Among the topological solitons, the simplest ones are those in which all the roots of both
Pl (z) = (z z0 )l and Qm (z) = qm (z z0 )m are at the same point. Let us focus on this case
byconsidering the spherically symmetric ansatz around this point chosen at the origin. If r =
+ x12 + x22 , for configurations of the form

(x1 , x2 ) =
f (r)eil
|h(r)|ei(+m)
(r)
Ai (x1 , x2 ) = lij 2 xj ,
r

,
f (r) =
h(r) =
r (lm)
r 2(lm)
+ |qm
qm
r 2(lm) + |qm |2
|2
e 2 u(r;qm ) ,
e 2 u(r;qm ) ,
the first-order PDEs reduce to a system of non-linear ordinary differential equations:

2

1
1 d
= f 2 (r) + h(r) 1 ,
r dr
2l

df
l
= f (r) 1 (r) ,
dr
r

l
m
d|h|
= |h|(r)
(r) ,
dr
r
l
(4)
(5)
(6)
to be solved together with the boundary conditions

lim f (r) = 1,
f (0) = 0,
lim h(r) = 0,
r

h(0) = |h0 |m,0 ,
lim (r) = 1,
(0) = 0
(7)
required by energy finiteness plus regularity at the center of the vortex. Let us recall that the
boundary conditions at infinity also require that mbe < l. The magnetic field and the energy
density of the spherically symmetric vortices in terms of the field profiles f (r), (r) are:
l d
,
2r dr

2 2

1 1
2
h(r) 2 2 + l f (r) 1 (r) 2
+
1
1
f
(r)
E(r) =
8 l2
r2
2

l 2 |h(r)|2 m
+
.
(r)
l
r2
B(r) =
3.4. Semi-local strings with a quantum of magnetic flux

We now go on to the most elementary solutions that carry a quantum of magnetic flux, or,
l = 1 = m + 1. We follow the procedure developed in [23] to solve the non-linear ODE system
(4)(6) with boundary conditions (7). First, we consider small values of r and in the first-order
440
differential equations we test the power series

f (r) f1 r + f2 r 2 + f3 r 3 + f4 r 4 + ,
(8)
(r) 1 r + 2 r + 3 r + 4 r + ,
(9)
h(r) h0 + h1 r + h2 r + h3 r + h4 r + ,
2
(10)
where fj and j , j = 1, 2, 3, . . . , are real, whereas hj , j = 0, 1, 2, . . . , are complex coefficients.

The coupled first-order ODEs are solved at this limit by (8)(10) if

f1
f1
|h0 |2 1 r 3 +
|h0 |2 1 2|h0 |2 1 + 4f12 r 5 + ,
8
128

2

1
1
1
2
2
(r) 1 |h0 | r
|h0 | |h0 |2 1 + f12 r 4
4
32
8

1
1 2
|h0 |2 |h0 |2 1 3|h0 |2 2 +

f1 5|h0 |2 4 r 6 + ,
768
192

h0
h
0
|h0 |2 1 r 2 +
|h0 |2 1 2|h0 |2 1 + 4f12 r 4 + .
h(r) h0 +
8
128
We stress that
1
q0
e 2 u(0;q0 ) ,
h0 =
|q0 |
f (r) f1 r +
is determined by the behavior of the solution at the origin. After that, only a free parameter, f1 , is
left in the exact solution near the origin. Second, a numerical scheme is implemented by setting a
boundary condition at a non-singular point of the ODE system, which is obtained from the power
series for a small value of r (r = 0.001 in our case). This scheme prompts a shooting procedure
by varying f1 , where the correct asymptotic behavior of the solutions is obtained by setting an
optimal value for f1 for a given value of h0 . Finally, the first-order ODE system is solved for
large r by means of a power series in 1r :

f (r) =
f j r j ,
(r) =
j r j ,
h(r) =
hj r j ,
j =0
with the result that
j =0
j =1

1 2 3 1 4
|h1 |2 2

r + 2 h + h
r 4
f (r) 1
2
8

1 2
1 2
5 1 4

+ h
h
32 + 5 h
r 6
16

1 2
1 2 35 1 4
35 1 6

+ h
h +
h
1152 + 158 h
r 8 + ,
4
128
2
2 4
2
2
2
(r) 1 h1 r 2 + h1 8 + h1 r 4 + h1 192 + 24 h1 h1 r 6
2
2
4 6
+ h1 9216 + 1120 h1 48 h1 + h1 r 8 + ,

1 3

h(r) |h|1 r 1 |h | r 3 + h1 3 2 + 3 h1 2 r 5
2
8

1 3
1 4
1 2
5
+ h 32 + 5 h h r 7 + .
16
441
Again, only one free parameter, h1 , is left. The value of h1 is fixed by demanding continuity of
the solution at intermediate distances (r = 15 in our case) obtained by gluing the short-r and
large-r approximations. In particular, this has the important implication that

|h0 | = 0 h1 = 0,
linking the null value of |h0 |, which gives the embedded ANO vortex, with the null value of
the constant |h1 | setting the behavior of the solution for very large r. Another important remark
is that the large r behavior of self-dual semi-local defects differs from the large r behavior of
self-dual ANO vortices that decay exponentially.
Fig. 1 shows the results obtained with this procedure for several values of h0 . Note that h0 =
1
q0
2 u(0;q0 ) , u(0; 0) = , u(0; ) = 0, h0 = 0 for the ANO vortices, and h0 = 1 for the
|q0 | e
CP1 -lumps. It is observed in the graphics that the field profiles reach their vacuum values at
distances of the order of r = 15. Consequently, almost identical numerical solutions would be
generated by sewing the numerical and the asymptotic solutions together at r greater than 15.
4. One-loop correction to the masses of semi-local self-dual topological solitons
4.1. Casimir energies
The starting point for computing the quantum corrections to the soliton mass is the evaluation
of the zero-point oscillations of the fields around both the soliton and the vacuum. The measurement of the zero-point soliton energy with respect to the vacuum energy shows a strong analogy
with the Casimir effect, where the plates are substituted by the soliton profile.
Let us first consider small fluctuations around the semi-local soliton by writing:
(
x ) = S(
x ) + S(x0 , x),
Aj (
x ) = Vj (
x ) + aj (x0 , x).
x ) we respectively denote the scalar and vector fields of the topological soliton
By S(
x ) and Vk (
solution, whereas S(x0 , x) and aj (x0 , x) calibrate the small deviations of the bosonic fields
with respect to the classical solution. Of course, not all fluctuations are physically relevant, and
we should avoid pure gauge deformations. In order to do so, we impose the Weyl/background
gauge condition:

i
A0 (x0 , x) = 0,
j aj (x0 , x) + S (
x )S(x0 , x) S (x0 , x)S(
x ) = 0.
2
Under these constraints, the ground state energy in the topological sectors up to O( 3 ) order
(one-loop) is:

v2
T
d 2x
H (2) =
(11)
+ T K + K G ,
2
x0 x0
where the bosonic fluctuations are arranged in the column vector
a (x , x)
1
a2 (x0 , x)
S 1 (x0 , x)
1
(x0 , x) = 2
,
S1 (x0 , x)
S 1 (x0 , x)
2
S22 (x0 , x)
442
(b)
(c)
(a) Functions f (r), h(r) and (r) and (b), (c) energy density for h0 = 0.1.
(a)
(b)
Fig. 1. Field profiles and energy densities for semi-local topological defects.
(c)
(a)
(b)
(c)
(a)
(b)
(c)
(a)
Fig. 1. (continued)
443
444
and (
x ) corresponds to ghost fluctuations. K and K G are second-order differential operators.
K is matrix-valued and fairly complicated
21 S12
K =
2 S 1
1 1
21 S 2
2
21 S21
and
KG
21 S12
21 S11
21 S22
22 S12
22 S11
22 S22
22 S12
2Vk k
S11 S21 + S12 S22
22 S11
22 S22
22 S21
2Vk k
S11 S22 + S12 S21
S11 S21 + S12 S22
S11 S22 + S12 S21
S11 S22 S12 S21
S11 S21 + S12 S22
2Vk k
21 S21
S11 S22 S12 S21
S11 S21 + S12 S22
2Vk k
22 S21
is the scalar differential operator:
K G = k k + |S1 |2 + |S2 |2 .
In all these formulas, the notation is

1

S1 (
x ) + iS12 (
x)
S1 (
x)
=
S(
x) =
x)
S2 (
S21 (
x ) + iS22 (
x)
a = S a + ab V S b , and the differand analogously for S. The covariant derivatives are j SM
j M
j M
ential operators in the diagonal of K take the form
A = k k + |S1 |2 + |S2 |2 ,
B = k k +

1
3|S1 |2 + |S2 |2 + 2Vk Vk 1 ,
2

1
|S1 |2 + 3|S2 |2 + 2Vk Vk 1 .
2
The energy due to the zero point fluctuations around the vacuum must be subtracted. As is
fairly obvious, the Hamiltonian for the vacuum fluctuations has exactly the same form as (11) for
the vacuum values: S1 = 1, S2 = 0, V1 = V2 = 0. We shall denote by K0 and K0G the operators
K and K G in the vacuum background.
When all the fluctuation modes are unoccupied, we formally obtain the contribution to the
semi-local topological soliton ground state energy as a difference between two super-traces of
differential operators acting on column vectors of L2 (R2 ) functions. The first super-trace comes
from (the star stresses the fact that zero modes do not contribute to the trace) oscillations of the
fields around the soliton,
C = k k +

1 h m

1
1
hm
Tr K 2 Tr K G 2 =
STr K 2 ,
2
2
and the second one accounts for vacuum fluctuations of the fields:

1
1
1

h m
Tr K02 Tr K0G 2 = STr K02 .
E0 =
2
ETS =
Therefore,
1
h m 1
STr K 2 STr K02 ,
(12)
2
formally measures the semi-local topological soliton Casimir energy.
Before proceeding, a further explanation of how we choose small fluctuations belonging to
L2 (R2 ) deserves a pause. Following the conventional QFT approach, we put the system in a
very large but finite two-dimensional box and impose periodic boundary conditions on (x0 , x)
C
= ETS E0 =
MTS
445
and (
x ). In Appendices IIIV of Ref. [22] it is shown that taking the infinite area limit at
the end leaves no remnants for kinks and self-dual vortices. Thus, this procedure uses invisible
boundary conditions in such a way that, at intermediate stages, one works on circles or genus one
Riemann Surfaces. Nevertheless, the rapid (exponential) decay of the Higgs field to its vacuum
value in these cases ensures that, starting from large lengths or areas, there will be a very small
dependence of the results on the size. Except for the embedded ANO vortices, all the Higgs
fields of the semi-local self-dual topological solitons decay to their vacuum values as r1k for
some positive k. Therefore, we expect a more significant dependence on the size of the box for
these less concentrated solitons.
We might try topological boundary conditions like those used in Ref. [14] for the bosonic
and supersymmetric kink. In this planar gauge theoretical setting, the analogous form of the
anti-periodic boundary conditions would be:
(x1 , x2 ) = e
PmL
P0
[Vk (
x ) dxk ]
(x1 + mL, x2 + mL).
The line integral is along a path starting at the point P0 = (x1 , x2 ), ending at PmL = (x1 +
mL, x2 + mL), and passing through regions far away from the vortex core. These twisted boundary conditions are also invisible in the sense that no boundary is introduced. Rather, the fact
that we are dealing with a non-trivial line bundle of first Chern class equal to one over a genus
one Riemann surface is taken into account. In a purely bosonic framework, however, the twisted
boundary conditions, like periodic boundary conditions, will not leave any mark at the infinite
area limit.
4.2. Counter-term energies
The Casimir energy of the previous section is of order h , and this is also the order of the
counter-terms found in Section 2. Thus, the one-loop semi-local topological soliton mass also
receives contributions from the counter-terms for scalar and vector fields. At the self-dual limit
2 = 1, these contributions are

hm

S
=
4I (1) + I (0)
d 2 x 1 |S1 |2 |S2 |2 ,
Mc.t.
2

A
d 2 x Vk Vk .
Mc.t. = h m I (1) + I (0)
We now reshuffle the sum of these two quantities into two pieces, respectively proportional to
I (1) and I (0):
I(1)
Mc.t.
=
where
hm
I (1) (1) (S, Vk ),

2
I(0)
Mc.t. =
h m
I (0) (0) (S, Vk ),
2
(13)

1
(S, Vk ) = 4 d x 1 |S1 | |S2 | Vk Vk ,
2

(0) (S, Vk ) = d 2 x 1 |S1 |2 |S2 |2 2Vk Vk .
(1)
The total contribution to the one-loop mass shifts from the mass renormalization counter-terms
is:
I(1)
I(0)
Mc.t. = Mc.t.
+ Mc.t. .
446
4.3. Deformation of the first-order equations

It follows from the classical degeneracy of vortices that there are some static deformations of
a topological solution that do not cost energy at the tree level: those that give rise to a new selfdual soliton. Therefore, the dimension of the kernel of the operator K ruling the bosonic small
fluctuations around the topological soliton should be equal to the dimension of the moduli space
of self-dual semi-local soliton solutions. To check this, we proceed as in the previous section
by expanding the bosonic fields around the solution, but this time we consider only static small
fluctuations:
x ) = Vj (
x ) + aj (
x ).
Aj (
(
x ) = S(
x ) + S(
x ),
The modified fields are still solutions of the first-order equations if (14) and (15) are satisfied:
F12 =

1
1
2
2 a1 + 1 a2 +

1
S S + S S = 0
2
(14)
and
D1 + iD2 = 0
(1 + i2 )S i(a1 + ia2 )S = 0,
(15)
where k = k iVk . Also as in the previous section, in order to avoid pure gauge fluctuations
we set the static background gauge:

i
S S S S = 0.
2
Thus, the tangent space to the moduli space of self-dual vortices is the kernel of the first-order
deformation operator D:
1
S11
S12
S21
S22
2

S12
S11
S22
S21
2
1
2
S1 S1 2 + V1 1 V2
0
0
.
D=
S2
1
S1
1 + V2 2 + V1
0
0
1
S2 S22
0
0
2 + V1 1 V2
j aj (
x) +
S22
S21
1 + V2
2 + V1
= D D,
and it is also possible to prove that the Kernel of D is empty

One easily checks that K
[24]. Thus, K and D have the same kernel and the dimension of that kernel is the dimension of
the classical moduli space: 4l.
4.4. Zeta-function regularization
The zero-point energies for the topological soliton and the vacuum are formally super-traces:
the differences of traces of differential operators. Such traces are divergent quantities. The standard procedure for dealing with this delicate point is understanding these traces as the generalized
zeta functions of the corresponding differential operators; see [6,9,12]. Given a differential operator A acting on an L2 space of functions, the corresponding generalized zeta function is:

s
A (s) =
n , s C,
Spec A
447
where n are the eigenvalues of A. We shall regularize the zero-point energies in the form:

h 2 s
K (s) K G (s) ,
ETS (s) =
2
2 m

h 2 s
K0 (s) K G (s) ,
E0 (s) =
2
0
2 m
i.e., by assigning to the meromorphic generalized zeta functions the value obtained by analytic
continuation at the point s of the s-complex plane. The physical limit is:

C
C
MTS
= lim MTS
(s) = lim ETS (s) E0 (s) .
s 12
s 12
Because K0 and K0G are free Schrdinger operators, their zeta functions are well known [11,15]:
m2 L2 [s 1] m2 L2
1
(s)
2
(s 1)(s)
m2 L2 [s 1]
,
K G (s) =
0
4
(s)
K0 (s) =
where (s) and [s, a] are the complete and incomplete Euler Gamma functions, respectively.
The contribution from the mass counter-terms also involves divergent quantities proportional
to the integrals I (1) and I (0). To regularize these integrals, we apply the residue theorem to
integrate I (c) in the complex k0 -plane. On a square of area m2 L2 each integral becomes a infinite
sum over discrete momenta:

d 2k
1
1 1
1
1

I (c) =
=
.
2
2
(2) k k + c 2 m2 L2

+c
k
k
2

kZ
Accordingly, these integrals are the generalized zeta functions of the Euclidean KleinGordon
and Laplacian operators evaluated at s = 12 . In this way, we also define the mass renormalization
correction as a meromorphic function in the complex s-plane:
2 s

h
R
+1 (s) (1) (S, Vk ) + (s) (0) (S, Vk ) ,
(s) =
MTS
2
2
2L m
and take the physical limit:
R
R
MTS
= lim MTS
(s).
s 12
While +1 is exactly K0G , and its zeta function has been written above, calculation of (s)
is a bit tricky. From the partition function for the Laplacian, via the Mellin transform, we have:
1
m2 L2
(s) =
lim
4 0 (s)
1
d
0
s2

+
s2
or
(s) =

1
m2 L2
lim
[s 1, ] + [s 1, ] .
4 0 s1 (s)
448
Because [s 1, ]
=
(s) =
s1
s1
s
s
and [s 1, ]
= [s 1], we obtain:
m2 L2
1
4
(s 1)(s)
if Re s < 1.
We finally find:

1 ( 12 )
1
1
1
=
= ,
I (1) =
+1
2
8 ( 12 )
4
2m2 L2

1
1
1
I (0) =
= .

2
2
2
2m L
4
Contrary to the kink cases, which are one-dimensional problems, a finite answer is obtained
in the regularized integrals via the associated zeta functions. The reason is that in this twodimensional problem the physical limit s = 12 is not a pole of the zeta functions and only finite
renormalizations will be necessary.
5. The semi-local topological soliton heat kernel and generalized zeta function
Control of K (s) and K G (s) is much more difficult. A convenient way for dealing with the
zeta functions of differential operators acting on infinite dimensional spaces is by means of heat
kernel techniques. In this section, we shall develop this method, applied to our soliton operators.
5.1. The heat kernel of a differential operator. Seeley densities
The heat equation kernel of an N N matrix differential operator of the general form
x )k + V (
x)
K = K0 + Qk (
is the solution of the K-heat equation

x , y; ) = 0,
I + K KK (
with initial condition: KK (

x , y; 0) = I (2) (
x y). We are particularly interested in the diagonal
x = y heat-kernel, because the Mellin transform of the partition function

Tr e K = tr d 2 x KK (
x , x; )
R2
gives the generalized zeta function.

To find the kernel, one writes [8]
x , y; ) = CK (
x , y; )KK0 (
x , y; ),
KK (
where K0 is the operator K very distant from the origin, where Qk (
x ) and V (
x ) take their
x , y; ) satisfies the N N -matrix transfer equation
asymptotic constant vacuum values. CK (

1
x k yk
x , y; ) = 0
I+
k I Qk I + Qk k + V CK (
2
x , y; 0) = I at infinite temperature.
and is the unit matrix CK (
449
Solving the transfer equation as an inverse-temperature power series expansion,

CK (
x , y; ) =
cn (
x , y; K) n ,
n=0
the PDE equation becomes tantamount to the recurrence relation between the densities
cn (
x , y; K):

1
nI + (xk yk ) k I Qk cn (
x , y; K) = [I Qk k V ]cn1 (
x , y; K), n 1
2
to be started from: c0 (
x , y; K) = I. While it is easy to find the first diagonal density, c1 (
x , x; K) =
V (
x ), the determination of higher-order densities becomes more and more involved. To make
the problem more tractable, we introduce the following notation:
(1 ,2 )
1 +2 [cn ]AB (
x , y; K)
,
y
x
x11 x22
CnAB (
x ) = lim
[cn ]AB (
x , x; K) =(0,0) CnAB (
x ).
Thus, at the y x limit the recurrence relations between densities and partial derivatives of
densities can be written in compact form:
AB
(
x)
(k + 1 + 2 + 1)(1 ,2 ) Ck+1
= (1 +2,2 ) CkAB (
x ) + (1 ,2 +2) CkAB (
x)

N
1
2

1 2 r+t QAD
1 (1 r+1,2 t) DB
Ck (
x)
r
t
x1r x2t
d=1 r=0 t=0

r+t QAD
2 (1 r,2 t+1) DB
+
C
(
x
)
k
x1r x2t

2
N 1 1
1 1 2 r+t QAD
1
1 (1 1r,2 t) DB
1
Ck+1 (
x)
+
2
t
r
x1r x2t
d=1 r=0 t=0

(1 t,2 1r)
1
N 2 1
2 1 1 r+t QAD
1
DB
2
2
Ck+1
(
x)
2
t
r
x1t x2r
d=1 r=0 t=0
2
1 r+t AD
N

1 2 V
d=1 r=0 t=0
to be solved starting from

c0 (
x , x; K) = I
x1t x2r
(,) C AB (
x ) = 0,
0
(0,0) C AA (
x ) = 1,
0
(1 t,2 r)
CkDB (
x)
if = 0, and/or = 0,
A = 1, 2, . . . , N.
5.2. The Mellin transform of the asymptotic expansion

We must now deal with the cases N = 6 and N = 1, respectively for the operators K and K G .
A good approximation to the generalized zeta functions of both operators is obtained from the
450
Mellin transform [9]

1
K (s) =
d s1 Tr eK ,
(s)
K G (s) =
1
(s)
d s1 Tr eK ,
G
applied to the high-temperature expansion of the partition functions

4
6

I
O
1 n
K
cn (K) AA +
cn (K) AA ,
=
e
Tr e
4
n=0
Tr eK =
G
where

cnI (K) AA =
A=1
A=5

n cn K G ,
n=0
d 2 x e [cn ]AA (
x , x; K),
x , x; K),
cnO (K) AA = d 2 x [cn ]AA (

G
= d 2 x [cn ] x, x; K G .

cn K

A = 1, 2, 3, 4,
A = 5, 6,
The factor e , which appears in front of [cn (K)]AA for A = 1, 2, 3, 4, obeys the fact that the
corresponding modes in have one unit of mass, while the modes for A = 5, 6 are massless.
The generalized zeta functions are thus divided as sums of meromorphichigh-temperature
regimeand entireto be neglected, low temperature regimefunctions of s:
4 1

1
d s+n2 cnI (K) AA e
K (s) =
4(s)
n=0 A=1 0
6 1

s+n2
cnO (K) AA
A=5 0
4

A=5
1
cnO (K) AA
4(s)(s + n 1)
1
n=0 0
n=0
d Tr eK
[s + n 1, 1]
cnI (K) AA
4(s)
1
K G (s) =
(s)

1
n=0 A=1
6

1
+
(s)
s+n2
cn K

+
1
BK (s),
(s)
1
+
(s)
d Tr eK

[s + n 1, 1]
1
cn K G
+
B G (s),
4(s)
(s) K
where [s + n 1, 1] are incomplete Euler Gamma functions. We shall neglect the entire parts
B(K) and B(K G ) and keep a finite number of terms, N0 , in future use of these generalized zeta
functions for the regularization of ultraviolet divergences.
451
5.3. The high-temperature one-loop semi-local vortex mass shift formula

The contribution of the c1 coefficients to the semi-local topological soliton Casimir energy is
2 s
4
I

[s, 1]
h
(1)C
c1 (K) AA c1 K G
MTS (s) =
2
2
4(s)
m
A=1

6

O
1
+
c1 (K) AA
,
4s(s)
A=5
but the first Seeley coefficients due to bosonic and ghost fluctuations, respectively, give:
4

c1I (K) AA =
A=1
6

c1O (K) AA
A=5

c1 K G =
Therefore,
(1)C

d 2 x 5 5|S1 |2 3|S2 |2 2Vk Vk ,

d 2 x 1 |S1 |2 3|S2 |2 2Vk Vk ,

d 2 x 1 |S1 |2 |S2 |2 .

h m
[1/2, 1]
(1) (S, Vk ) + 2 d 2 x |S2 |2
16
(1/2)

2
(0) (S, Vk ) 2 d 2 x |S2 |2
.
(1/2)
MTS (1/2) =
On the other hand, the contribution to the one-loop semi-local string tension shift of the mass
renormalization counter-terms is2 :

hm
[1/2, 1]
2
R
MTS
(1/2) =
(1) (S, Vk )
(0) (S, Vk ) ,
16
(1/2)
(1/2)
which almost cancels the contribution to the semi-local string Casimir energy density of the c1
coefficients. We finally obtain the high-temperature one-loop semi-local topological soliton mass
shift formula:
N 4

0

G
3
hm
MTS =
cn (K) AA cn K
n ,1
2
16
n=2
A=1

6

[cn (K)]AA
+
+ 4l 8
n 32
A=5

1
hm
(16)
d 2 x |S2 |2 (x1 , x2 ) , 1 2 .
2
8
In this final formula (16) there are four types of terms:
2 [1/2, 1] instead of [1/2] is used in this formula to be consistent with the approximation in M (1)C ( 1 ).
TS
2
452
First, polynomial expressions in incomplete Gamma functions times the heat-kernel expansion coefficients for Ksaving only the four first diagonal contributions due to massive
bosonic particlesand K G coming from fermionic massive particles. All of them start
from the second-order coefficients.
Second, polynomial expressions in 1 3 , including the last two diagonal heat kernel coeffin 2
cients that collect the contribution of massless Goldstone particles. The starting coefficients
are also of second order.
Third, a factor proportional to 4l, taking into account the subtraction of the 4l zero modes.
An extra piece proportional to the norm of the second Higgs field due to the imperfect
cancelation of the contribution of first-order Seeley coefficients by mass renormalization
counter-terms.
Finally, let us mention that by cutting the expansion at a finite number, N0 , we admit an
errorbesides the rejected entire partswhich is a priori proportional to [N0 12 , 1] 1 1 ,
N0 2
for N0 large. Nevertheless, as we shall see in Section 6, once all the calculations have been done
the degree of convergence of the results is very good, meaning that the proportionality coefficient
in that error is very small. The reliability of the method is therefore quite high.
6. Mathematica calculations
6.1. Seeley densities for spherically symmetric semi-local vortices
We shall apply (16) to spherically symmetric vortices. The heat kernel local coefficients, however, depend on successive derivatives of the solution. This dependence can increase the error in
the estimation of these local coefficients because we are handling an interpolating polynomial
as the numerically generated solution, and the successive derivation with respect to r of such
a polynomial introduces inaccuracies. Indeed, this operation is plugged into the algorithm that
generates the local coefficients in order to speed up this process. It is thus of crucial importance
to use the first-order differential equations (4)(6) in order to eliminate the derivatives of the solution and to write the local coefficients as expressions that depend only on the fields. Recalling
the form of the spherically symmetric solutions,
S11 (x1 , x2 ) = f (r) cos ,
S12 (x1 , x2 ) = f (r) sin ,
S21 (x1 , x2 ) = h(r),

S22 (x1 , x2 ) = 0,
(r)
(r)
V1 (x1 , x2 ) =
sin ,
V2 (x1 , x2 ) =
cos ,
r
r
use of the first-order equations shows that:
S11
x1
S21
x2
S12
x1
V1
x1
S12 f (r)
f (r)
=
1 (r) cos2 ,
1 (r) sin2 ,
r
x2
r
S11
h(r)
f (r)
=
=
(r) cos ,
(r) cos sin ,
r
x2
r
S21
f (r)
h(r)
=
=
(r) sin cos ,
(r) sin ,
r
x1
r

cos 2 2f (r)(r) f 2 (r) + h2 (r) 1
=
+
,
2
r
2
=

V1
(r) sin2 2
f (r) + h2 (r) 1 ,
= cos 2 2 +
x2
2
r

V2
(r) cos2 2
f (r) + h2 (r) 1 ,
= cos 2 2
x1
2
r

V2
cos 2 2f (r)(r) f 2 (r) + h2 (r) 1
=
+
.
x2
2
r
2
Bearing this in mind, we solve the recurrence relations to find:
2(r)2
5f (r)2 3h(r)2 ,
r2

1
tr c2I (r) =
4(r)4 + 27r 4 f (r)4 8r 2 (r) 1 + 14f (r)2 + h(r)2
12r 4

+ 8(r)2 2 3r 2 + 9r 2 f (r)2 + 3r 2 h(r)2

+ f (r)2 56r 2 64r 4 + 34r 4 h(r)2 + r 4 37 32h(r)2 + 7h(r)4 ,

1
tr c3I (r) =
4(r)6 4r 2 (r)3 14 + 35f (r)2 36h(r)2
6
120r

+ 4(r)4 20 + 9r 2 + 32r 2 f (r)2 + 26r 2 h(r)2

2r 2 (r) 57r 2 f (r)4 + f (r)2 32 + 331r 2 75r 2 h(r)2

4 1 + h(r)2 16 9r 2 + r 2 h(r)2

+ (r)2 256 144r 2 117r 4 + 99r 4 f (r)4 16r 2 h(r)2 + 94r 4 h(r)2

61r 4 h(r)4 + 2r 2 f (r)2 56 + 183r 2 + 19r 2 h(r)2

+ r 4 16 + 151r 2 29r 2 f (r)6 + 32 135r 2 h(r)2

+ 16 + 23r 2 h(r)4 + r 2 h(r)6

+ f (r)4 20 + 199r 2 57r 2 h(r)2 + f (r)2 392 321r 2

+ 2 68 + 111r 2 h(r)2 27r 2 h(r)4 ,
tr c1I (r) = 5
2(r)2
f (r)2 3h(r)2 ,
r2

1
tr c2O (r) =
4(r)4 r 4 f (r)4 + 8r 2 (r) 1 + 2f (r)2 h(r)2
4
12r

8(r)2 2 + r 2 + r 2 f (r)2 5r 2 h(r)2 + 2r 2 f (r)2 4 + 9r 2 h(r)2

+ r 4 1 8h(r)2 + 19h(r)4 ,
tr c1O (r) = 1
tr c3O (r) =

1
4(r)6 4r 2 (r)3 14 + 9f (r)2 + 84h(r)2
120r 6

4(r)4 20 + 3r 2 + 2r 2 f (r)2 + 4h(r)2

+ (r)2 256 + 48r 2 3r 4 + 45r 4 f (r)4 + 2r 2 40 + 89r 2 h(r)2

115r 4 h(r)4 + 2r 2 f (r)2 8 + 5r 2 35r 2 h(r)2

2r 2 (r) 53r 2 f (r)4 + 4 1 + h(r)2 16 3r 2 + 7r 2 h(r)2

f (r)2 32 + 17r 2 + 47r 2 h(r)2
453
454

+ r 4 16 + 3r 2 + 3r 2 f (r)6 32 + 19r 2 h(r)2 + 16 + 23r 2 h(r)4

+ 33r 2 h(r)6 + f (r)4 52 r 2 + 39r 2 h(r)2

+ f (r)2 24 5r 2 + 72 + 22r 2 h(r)2 + 69r 2 h(r)4 ,
c1G (r) = 1 f (r)2 h(r)2 ,

1
c2G (r) = 2 3r 2 + 2r 2 f (r)4 5r 2 + 4(r)2 h(r)2 + 2r 2 h(r)4
6r

+ f (r)2 4 5r 2 + 8(r) 4(r)2 + 4r 2 h(r)2 ,

1
10r 4 + 4r 4 f (r)6 + 23r 4 8r 2 (r) + 16 1 + r 2 (r)2 + 32(r)3
c3G (r) =
4
60r
+ 16(r)4 h(r)2 + r 2 17r 2 + 8(r) 16(r)2 h(r)4 + 4r 4 h(r)6

+ r 2 f (r)4 24 17r 2 + 40(r) 16(r)2 + 12r 2 h(r)2

+ f (r)2 32(r)3 + 16(r)4 + 8r 2 (r) 5 + 6h(r)2

+ 16(r)2 1 + r 2 2r 2 h(r)2

+ r 2 24 + 23r 2 2 10 + 17r 2 h(r)2 + 12r 2 h(r)4 ,
etcetera, by means of a computing program implemented on a PC with Mathematica.
6.1.1. One-loop mass shift for the mass of the ANO vortex
Denoting simply

tr cnI
= 2

dr r tr cnI (r),
tr cnO
= 2

dr r tr cnO (r),
cnG
= 2
dr rcnG (r),
0
by plugging the NO vortex solution in these expressions, i.e., the case h0 = 0.0, embedded in
this model, we find Table 4. Because the NO vortex solutions have been generated numerically,
integration over the whole plane of the Seeley densities can also only be performed numerically.
Therefore, we are forced to put a cut-off into the area and replace the infinite plane by a discus
of radius R, which in the calculations above was chosen to be R = 10 000 (compare with the
profiles of Fig. 1). Use of these numbers in formula (16) provides us with Table 5, where the
one-loop mass shifts in h m units of semi-local self-dual NO vortices are shown up to sixth order
in the asymptotic formula.
Table 4
The sixth lowest Seeley coefficients for NielsenOlesen self-dual vortices
h0 = 0.0
n
tr cnI
tr cnO
cnG
1
2
3
4
5
6
41.4469
30.3736
12.9447
4.22603
1.05059
0.20900
91.8429
0.96286
0.0592415
0.001512548
0.000758663
0.00023912
12.599
2.61518
0.32005
0.0230445
0.0013023
0.0000698185
455
Table 5
Convergence of the one-loop mass shift for self-dual
semi-local NO vortices in units of h m
N0
MV (N0 )
l=1
2
3
4
5
6
1.61536
1.66862
1.67809
1.67966
1.67989
Our result for the one-loop mass shift of semi-local self-dual NO vortices is:
MV (N0 = 6) = 1.67989h m.
(17)
The ratio between the mass shifts of self-dual NO vortices in the semi-local and normal Abelian
Higgs model is:
MVSLAHM
MVAHM
1.67989
= 1.53486,
1.09449
see [15]. Similar relations exist between ratios of kink mass shifts in the ()42 model and the
BNRT model; a two-field system that depends on a real positive parameter :
BNRTM
MK
M
MK
BNRTM
MK
M
MK
0.693943
= 1.47299 if = 0.99,
0.471113
0.698445
= 1.48254 if = 1.01,
0.471113
see [17].
Several comments are in order:
1. The coefficients tr cnI , for n = 2, 3, . . . , 6, are identical, within numerical precision, to the
same coefficients for self-dual NO vortices in the Abelian Higgs model. Therefore, the difference of the vortex mass shift arising in the semi-local Abelian Higgs model is due to the
contribution of the tr cnO coefficients and to the double number of zero modes.
2. tr c1I and tr c1O are very large negative numbers. In the case at hand, h0 = 0, their contribution cancels against the energy induced by mass renormalization counter-terms. A similar
behavior of h0 = 0 self-dual semi-local topological solitons would mean that rather than decreasing their mass should grow as the result of one-loop fluctuations, a possibility that we
shall study below.
6.2. Seeley coefficients for semi-local topological solitons: One quantum of magnetic flux and
various values of h0
To accomplish this goal, we offer Tables A.1A.5 in Appendix A with similar numerical
calculations on discuses of radius R = 102 , R = 103 , R = 104 , R = 105 , and R = 106 for the
values h0 = 0, h0 = 0.1, h0 = 0.3, h0 = 0.6, and h0 = 0.9. Inspection of these tables raises some
points that merit comment:
456
1. We see from the tables for the h0 = 0.0 case that although the first-order coefficients grow
spectacularly with the radius, the highest coefficients are quite stable against growing areas
of integration, suggesting good behaviora finite value very close to the value obtained at
R = 104 of the one-loop correction in the infinite area limit.
2. Things start to be different when we consider the h0 = 0.1 case. First, the positive and negative contributions of the first-order Seeley coefficients are not completely canceled out by
mass renormalization counter-terms. Second, of special importance is the departure of the
values of tr cnO (K) from the same numbers for the NO vortex, whereas tr cnI (K) and cn (K G )
change slowly with h0 . Moreover, tr c1O (K) for h0 = 0.1, besides not being exactly canceled,
is negative and large. This effect prompts a increasingly less negative one-loop correction,
eventually changing sign, with larger and larger R. This behavior differs completely from
that of embedded NO vortices and suggests a different fate for semi-local topological solitons
when quantum fluctuations enter the game.
3. The numbers offered in the tables for h0 = 0.3 follow a similar pattern to those of the
h0 = 0.1 semi-local topological soliton. There are, however, some quantitative differences.
tr cnI (K), n = 1, h0 = 0.3, differs from the same number of the NO vortex more than the
h0 = 0.1 topological soliton. Thus, these numbers grow with h0 and R. tr c1I (K) decreases
with respect to the same number for the NO vortex and with R. There is a change of sign
at R = 103 if h0 = 0.1 and at R = 104 if h0 = 0.3. The numbers tr c1O (K) behave as in the
h0 = 0.1 case but depart more from NO values. If h0 = 0.1 the ghost coefficients cn (K G )
are slightly larger than these numbers for h0 = 0, but they start to decrease (rapidly with R)
when h0 = 0.3.
4. The pattern observed for the evolution of the Seeley coefficients when the area grows in
the cases h0 = 0.1 and h0 = 0.3 is reinforced for broader semi-local topological defects
with h0 = 0.6 and h0 = 0.9. The unbalanced first-order coefficients tr c1O , obtained by mass
renormalizations, rapidly tend toward huge negative values.
6.2.1. Mass shift of semi-local topological solitons
The consequence is as follows: whereas the one-loop mass shift of embedded ANO vortices
is always negative and varies extremely slowly as the area increases towards more negative values, one-loop mass shifts of genuine semi-local topological solitons with |h0 | > 0 become less
negative, and even positive, for larger areas, as is shown in Table 6.
The classical degeneracy in energy between semi-local topological defects seems to be broken
by one-loop fluctuations, the embedded ANO vortices becoming the ground states in the topological sector of one quantum of magnetic flux. It is remarkable how strong this effect becomes
for large |h0 |.
Table 6
One-loop mass shifts for semi-local topological solitons: Five values of h0 , five values of R, and fixed N0 = 6
R
MV (N0 = 6, R)
h0 = 0.0
MV (N0 = 6, R)
h0 = 0.1
MV (N0 = 6, R)
h0 = 0.3
MV (N0 = 6, R)
h0 = 0.6
MV (N0 = 6, R)
h0 = 0.9
102
103
104
105
106
1.67955
1.67971
1.67989
1.68005
1.68026
1.61672
1.58311
1.55133
1.51957
1.48779
1.05000
0.626167
0.252586
0.12086
0.49433
2.10142
4.5485
6.41655
8.5741
10.7203
24.6066
42.7747
60.9433
79.1116
97.2798
(a)
457
(b)
(c)
Fig. 2. Functions A(r), B(r) and C(r) for (a) h0 = 0.0, (b) h0 = 0.3 and (c) h0 = 0.9.
These numerical results find support in the following qualitative arguments based in the analysis of the potentials arising in the matrix Schrdinger operators governing small fluctuations
around semi-local self-dual topological solitons. The more pertinent, diagonal, operators in K
and K G are:
A = K G = +f 2 (r) + h2 (r) = +A(r),

2 (r)
1
B = + 3f 2 (r) + h2 (r) + 2 2 1 = +B(r),
2
r

2
(r)
1
2
2
C = + f (r) + 3h (r) + 2 2 1 = +C(r).
2
r
A glance at Fig. 2 reveals:
1. For ANO vortices, h0 = 0.0 and h(r) = 0, there are attractive potential wells with the bottom at the origin. Thus, the correction must be negative. Also, the field profiles reach their
vacuum values exponentially and the corrections are quite stable with respect to the area of
the normalization box.
2
2
2. For h0 = 0.1, h(r) is not zero, see also Fig. 1 and take into account that: r(r)
2 r0 r . We
observe that the corresponding terms push the wells upwards from the bottom, making them
less attractive. This explains why the mass shifts are less negative.
3. For h0 = 0.3, h(r) is big enough to globally produce a change from attractive to repulsive
potential forces provided that the area of the normalization box is sufficiently large. There is
consequently a change in the sign of the mass correction.
458
4. For h0 = 0.6 and h0 = 0.9, potential barriers dominate starting from relatively small areas
of the normalization box. The field profiles go to their vacuum values very slowly and the
barriers become very wide, explaining the strong dependence on the size of the normalization
box.
We conclude that the classical degeneracy is broken by one-loop fluctuations. Even if the
classical topological bound is saturated by all the solitons in the moduli space of solutions of the
first-order equations, quantum effects can distinguish between the different energy densities of
these extended objects. The two extremes are the ANO vortices, h0 = 0, where the energy density
is concentrated around the zero of the Higgs field, and the CP1 -lumps, with energy densities
uniformly distributed over the whole spatial plane. A similar effect has been observed before in
the moduli space of degenerated two-component kinks analyzed in Ref. [17].
6.2.2. Infrared divergences: Quantum fate of semi-local topological solitons
The dependence on the area of the normalization box is due to the slow decay (nonexponential) to their vacuum values of genuine semi-local topological solitons as compared with
ANO vortices. Plugging the asymptotic form of the spherically symmetric topological soliton
solutions obtained in Section 3.4 into the Seeley densities, we find the following behavior at
infinity in terms of the parameter |h1 | (which sets the large r behavior of the solutions):

2 4 2 4
1
4
1 h1 + 3 12 h1 h1 + O 5 ,
2r tr c1I (r)
r
r
r
r

4
1
4
2
2
4
1 + h1 + 3 4 h1 + h1 + O 5 ,
2r tr c1O (r)
r
r
r
r

8
1
2
2rc1G (r) 3 h1 + O 5 ,
r r
r

1 2 1 4

2
1
2
I
1 2

+O 5 ,
h + 3 1 + 4 h h
2r tr c2 (r)
r r
r
r

2
1
2
2
4
h1 2 +
1 + 4 h1 h1 + O 5 ,
2r tr c2O (r)
3
r r
r
r

64
1
2
2
4
h1 +
768 h1 80 h1 + O 9 ,
2rc2G (r)
r 3r 5
r7
r

2
1
2
2
2
4
h1 +
3 h1 h1 + O 5 ,
2r tr c3I (r)
r 3r
3r 3
r

2

2
2
h1 2 + h1 4 + O 1 ,
h 1 2 +
4
2r tr c3O (r)
r
3r
3r 3
r5

384 1 2
1
h +O 8 ,
2rc3G (r)
7
r 5r
r

12 1 2 1 4
1
h1 2 +
h
h
+
O
,
2r tr c4I (r)
3
r 6r
5
6r
r5

1
2
2 4
2r tr c4O (r) h1 + 3 4 h1 h1 + O 5 ,
r 6r
6r
r

832
43
h1 4
h1 6 +
h1 8 + h1 10 + O 1 ,
2rc4G (r)
r 21r 9
30r 9
48r 9
768r 9
r 11
459

1 2
1
1 2 1 4
h
h
,
h +
2
+
O
3
r 30r
30r
r5

1
1 2
1 2 1 4
h
h
h +
2r tr c5O (r)
4
+
O
+
,
3
r 30r
30r
r5

2048 1 2 416 1 4
43 1 6
1 1 8
1 1 10
h
h +
h
h +
h
2rc5G (r) 9
r r
105
189
540
864
138240

1
+ O 11 ,
r

1 2
1 1 2
1
1 1 4
I

+O 5 ,
h +
h
h
2r tr c6 (r)
3
r 180r
12
15r 7
r

1
1
h1 2 h1 4 + O
h1 2 +
.
2r tr c6O (r)
r 180r
4
45r 3
r5
2r tr c5I (r)
The key observation is the appearance of infrared logarithmic divergences in the Seeley coefficients tr cnI (K) and tr cnO (K) for all n. The ghost coefficients cnG (K G ), however, are infrared
convergent. The combination of the signs that we have seen in the previous subsections and the
large r behavior show that one-loop mass shifts of semi-local topological solitons tend to +
at the infinite area limit. Semi-local topological defects grow infinitely massive due to the infrared effects of one-loop fluctuations. This phenomenon seems to be amazingly close to the
non-existence of Goldstone bosons in (1 + 1) dimensions.
There is a very important exception: for ANO vortices, |h1 | = 0 and only the first-order coefficients are infrared divergent. However, the contribution of these coefficients is totally canceled
by mass renormalization counter-terms. Our results suggest that only the ANO vortices between
all the semi-local topological solitons survive one-loop quantum fluctuations. It would be very
interesting to try a more analytic approach to this problem in order to fully elucidate this delicate
issue.
Acknowledgements
This work has been partially financed by the Spanish Ministerio de Educacion y Ciencia and
the Junta de Castilla y Leon under grants: FIS2006-09417 and VA013C05.
Appendix A
Semi-local strings with h0 = 0.0. Embedded NO vortices.
Table A.1
Seeley coefficients: Semi-local strings with h0 = 0.0
h0 = 0.0, R = 102
n
1
2
3
4
5
6
tr cnI
16.3087
30.3548
12.9428
4.2259
1.050558
0.209003
tr cnO
33.9956
0.959353
0.0596236
0.00149819
0.000757146
0.000239188
h0 = 0.0, R = 103
cnG
12.5761
2.61136
0.319668
0.0230172
0.00122871
0.0000697495
n
1
2
3
4
5
6
tr cnI
12.5691
30.3641
12.9437
4.22596
1.05058
0.209003
tr cnO
cnG
62.9193
12.5875
0.960952
2.61327
0.0594326
0.319859
0.00151183
0.0230309
0.0000747904
0.00122947
0.000239154
0.000069784
460

h0 = 0.0, R = 104
n
1
2
3
4
5
6
tr cnI
41.4469
30.3736
12.9447
4.22603
1.05059
0.20900
tr cnO
91.8429
0.96286
0.0592415
0.001512548
0.000758663
0.00023912
h0 = 0.0, R = 105
cnG
12.599
2.61518
0.32005
0.0230445
0.0013023
0.0000698185
tr cnI
70.3247
30.3829
12.9456
4.2261
1.05059
0.209003
n
1
2
3
4
5
6
tr cnO
120.767
0.964771
0.0590504
0.00153913
0.000759469
0.000239086
cnG
12.6105
2.61709
0.320242
0.0210582
0.00123099
0.00002698506
h0 = 0.0, R = 106
n
1
2
3
4
5
6
tr cnI
99.2028
30.3920
12.9466
4.22617
1.05058
0.209003
tr cnO
149.690
0.9666682
0.0588602
0.00155181
0.0007600532
0.00023874
cnG
12.6220
2.61901
0.320432
0.02300754
0.00123155
0.0000695093
Semi-local strings with h0 = 0.1.

Table A.2
Seeley coefficients: Semi-local strings
h0 = 0.1, R = 102
n
1
2
3
4
5
6
tr cnI
19.0369
31.2832
13.2000
4.28364
1.06171
0.210813
tr cnO
35.1867
1.69082
0.277417
0.567246
0.0102584
0.00159792
h0 = 0.1, R = 103
cnG
12.8061
2.62117
0.31637
0.0225427
0.00119161
0.0000667653
n
1
2
3
4
5
6
h0 = 0.1, R = 104
n
1
2
3
4
5
6
tr cnI
34.3878
32.5602
13.5154
4.35407
1.0753
0.213057
tr cnO
94.099
2.58966
0.536179
0.124453
0.0236316
0.00383473
tr cnI
87.8124
33.8375
13.8312
4.42449
1.0889
0.215306
tr cnO
153.011
3.48881
0.784926
0.192179
0.030047
0.00607261
tr cnO
64.6429
2.14009
0.406805
0.090590
0.01694523
0.00271637
cnG
13.0898
2.66844
0.321364
0.0228803
0.00121037
0.000067618
h0 = 0.1, R = 105
cnG
13.3734
2.71571
0.326091
0.023218
0.00122913
0.0000684707
h0 = 0.1, R = 106
n
1
2
3
4
5
6
tr cnI
7.67553
31.9215
13.3577
4.31885
1.0685
0.211935
cnG
13.9407
2.81037
0.335521
0.00238892
0.00126624
0.0000689434
n
1
2
3
4
5
6
tr cnI
61.1001
33.1989
13.6732
4.38928
1.0821
0.214679
tr cnO
123.555
3.03924
0.665552
0.158316
0.0303191
0.0049531
cnG
13.6571
2.76299
0.330819
0.0235856
0.00124789
0.0000693256
461

Table A.3
h0 = 0.3, R = 102
n
1
2
3
4
5
6
tr cnI
27.8909
39.9546
15.2219
4.77643
1.15832
0.226592
tr cnO
48.6659
7.69438
2.2614
0.540926
0.10716
0.0176963
h0 = 0.3, R = 103
cnG
11.7815
2.22326
0.247514
0.0159336
0.000764659
0.0000395987
n
1
2
3
4
5
6
h0 = 0.3, R = 104
n
1
2
3
4
5
6
tr cnI
22.9489
43.4678
17.7429
5.43346
1.29128
0.248822
tr cnO
124.360
15.4073
4.96241
1.20654
0.24083
0.0399494
tr cnI
2.4708
40.2110
16.4824
5.10495
1.22480
0.237707
tr cnO
86.5124
11.5509
3.61196
0.873748
0.173998
0.0288234
cnG
10.8818
2.07325
0.232514
0.0148622
0.000705138
0.0000368937
h0 = 0.3, R = 105
cnG
9.98213
1.92331
0.21752
0.0137912
0.000645635
0.0000341891
n
1
2
3
4
5
6
tr cnI
48.3715
46.7241
19.0034
5.76197
1.35776
0.259938
tr cnO
162.207
19.2687
6.31286
1.53933
0.30766
0.0510755
cnG
9.08246
1.77336
0.202525
0.0127201
0.000586148
0.0000314752
h0 = 0.3, R = 106
n
1
2
3
4
5
6
tr cnI
73.7864
49.9803
20.2636
6.09059
1.42424
0.271056
tr cnO
200.055
23.1201
7.6633
1.87212
0.374493
0.0622017
cnG
8.18278
1.62347
0.187535
0.011658
0.00052486
0.0000284767

Table A.4
h0 = 0.6, R = 102
n
1
2
3
4
5
6
tr cnI
107.840
72.5374
26.5220
7.45753
1.67277
0.309679
tr cnO
113.330
41.052
12.8641
3.13219
0.615441
0.101414
h0 = 0.6, R = 103
cnG
13.3675
1.73758
0.135265
0.00625589
0.000232254
8.66613 106
n
1
2
3
4
5
6
h0 = 0.6, R = 104
n
1
2
3
4
5
6
tr cnI
157.661
122.770
42.8451
11.50061
2.48061
0.44232
tr cnO
265.885
89.8409
28.9709
7.17046
1.42244
0.235944
tr cnI
132.751
97.6535
34.6838
9.48185
2.0767
0.376957
tr cnO
189.607
65.4471
20.9178
5.15141
1.01896
0.168681
cnG
14.4510
1.91781
0.153289
0.0075433
0.000303781
0.0000119181
h0 = 0.6, R = 105
cnG
9.98213
1.92331
0.21752
0.0137912
0.000645635
0.0000151754
n
1
2
3
4
5
6
tr cnI
182.571
147.886
51.0065
13.5303
2.88453
0.511506
tr cnO
342.162
114.235
37.0239
9.18952
1.82592
0.303206
cnG
16.6184
2.27905
0.189413
0.0101236
0.000447104
0.0000184335
462

h0 = 0.6, R = 106
tr cnI
207.481
173.003
59.1167
15.5545
3.28844
0.578786
n
1
2
3
4
5
6
tr cnO
418.440
138.629
45.0769
11.2086
2.2294
0.370467
cnG
17.7021
2.45961
0.207444
0.0114354
0.000517096
0.0000217957

Table A.5
h0 = 0.9, R = 102
n
1
2
3
4
5
6
tr cnI
638.887
326.423
109.814
27.9842
5.71851
0.97486
tr cnO
601.992
290.227
94.6224
23.3075
4.59264
0.755329
h0 = 0.9, R = 103
cnG
17.3563
1.24111
0.0897853
0.00588282
0.000325247
0.0000152661
n
1
2
3
4
5
6
h0 = 0.9, R = 104
n
1
2
3
4
5
6
tr cnI
1472.190
746.482
246.919
62.0179
12.5121
2.10653
tr cnO
1459.110
700.081
230.135
57.2676
11.3800
1.88677
tr cnI
1055.500
536.449
178.399
45.0018
9.11549
1.54073
tr cnO
1030.530
495.158
162.381
40.2880
7.798641
1.32206
cnG
25.0050
2.5130
0.216979
0.014969
0.00083022
0.0000282526
h0 = 0.9, R = 105
cnG
32.6701
3.79044
0.344728
0.0240939
0.00133716
0.0000612957
n
1
2
3
4
5
6
tr cnI
1888.870
956.515
315.440
79.0340
15.2088
2.67232
tr cnO
1887.670
905.004
297.889
74.2472
14.7736
2.45247
cnG
40.3354
5.06804
0.472483
0.0332125
0.000184412
0.0000843383
h0 = 0.9, R = 106
n
1
2
3
4
5
6
tr cnI
2305.550
1166.550
383.961
96.05501
19.3054
3.23313
tr cnO
2316.230
1109.930
365.644
91.2268
18.1672
3.01818
cnG
48.007
6.34561
0.60024
0.0423323
0.0423323
0.000107346
References
[1]
[2]
[3]
[4]
[5]
T. Vachaspati, A. Achucarro, Phys. Rev. D 44 (1991) 3067.

T. Vachaspati, A. Achucarro, Phys. Rep. 327 (2000) 347.
G.W. Gibbons, M.E. Ortiz, F. Ruiz Ruiz, T.M. Samols, Nucl. Phys. B 385 (1992) 127.
A. Jaffe, C. Taubes, Vortices and Monopoles, Birkhauser, Boston, 1980.
P. Forgacs, S. Reuillon, M.S. Volkov, Phys. Rev. Lett. 96 (2006) 041601;
P. Forgacs, S. Reuillon, M.S. Volkov, Nucl. Phys. B 751 (2006) 390.
[6] K. Kirsten, Spectral Functions in Mathematics and Physics, Chapman and Hall/CRC, New York, 2002.
[7] D.V. Vassilevich, Phys. Rep. C 388 (2003) 279360.
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
463
M. Stone, Ann. Phys. 155 (1984) 56.

P.B. Gilkey, Invariance Theory, the Heat Equation and the AtiyahSinger Index Theorem, Publish or Perish, 1984.
M. Bordag, A. Goldhaber, P. van Nieuwenhuizen, D. Vassilevich, Phys. Rev. D 66 (2002) 125014.
A. Alonso Izquierdo, W. Garcia Fuertes, M.A. Gonzalez Leon, J. Mateos Guilarte, Nucl. Phys. B 635 (2002) 525.
D. Vassilevich, Phys. Rev. D 68 (2003) 045005.
A. Rebhan, P. van Nieuwenhuizen, R. Wimmer, Nucl. Phys. B 679 (2004) 382394.
H. Nastase, M. Stephanov, A. Rebhan, P. van Nieuwenhuizen, Nucl. Phys. B 542 (1999) 471514.
A. Alonso Izquierdo, W. Garcia Fuertes, J. Mateos Guilarte, M. de la Torre Mayado, Phys. Rev. D 70 (2004)
061702(R).
A. Alonso Izquierdo, W. Garcia Fuertes, J. Mateos Guilarte, M. de la Torre Mayado, Phys. Rev. D 71 (2005) 125010.
A. Alonso Izquierdo, W. Garcia Fuertes, M.A. Gonzalez Leon, J. Mateos Guilarte, Nucl. Phys. B 681 (2004) 163.
D. Bazeia, M. Dos Santos, R. Ribeiro, Phys. Lett. A 208 (1995) 8488.
D. Bazeia, J.R. Nascimento, R. Ribeiro, D. Toledo, J. Phys. A 30 (1997) 81578166.
M.A. Shifman, M.B. Voloshin, Phys. Rev. D 57 (1998) 25902598.
A. Alonso Izquierdo, J.M. Guilarte, M.T. Mayado, W.G. Fuertes, J. Phys. A 39 (2006) 64636472.
A. Alonso Izquierdo, W.G. Fuertes, M.A.G. Leon, J.M. Guilarte, J.M.M. Castaeda, M.T. Mayado, Lectures on the
mass of topological solitons, hep-th/0611180.
H.J. de Vega, F.A. Schaposnik, Phys. Rev. D 14 (1976) 1100.
W. Garcia Fuertes, J. Mateos Guilarte, Eur. Phys. J. C 9 (1999) 167.

On the unitarity of gauged non-compact WZNW strings

Jonas Bjrnsson , Stephen Hwang,
Department of Physics, Karlstad University, SE-651 88 Karlstad, Sweden
Received 15 October 2007; accepted 28 November 2007
Abstract
In this paper we investigate the unitarity of gauged non-compact WZNW strings, i.e., string theories formulated as G/H WZNW models, where G is a non-compact group. These models represent string theories
on non-trivial curved spacetimes with one time-like component. We will prove that for the class of models
connected to Hermitian symmetric spaces, and a natural set of discrete highest weight representations, the
BRST formulation, in which the gauging is defined through a BRST condition, yields unitarity. Unitarity
requires the level times the Dynkin index to be an integer, as well as integer valued highest weights w.r.t.
the compact subalgebra. We will also show that the BRST formulation is not equivalent to the conventional GKO coset formulation, defined by imposing a highest weight condition w.r.t. H . The latter leads to
non-unitary physical string states. This is, to our knowledge, the first example of a fundamental difference
between the two formulations.
PACS: 02.20.Tw; 11.25.Hf; 11.25.Pm; 11.40.Ex
Keywords: WZNW-model; Non-compact backgrounds; No ghost theorem; Hermitian symmetric spaces; CFT
1. Introduction
String theory on non-trivial non-compact spaces is a subject that has attracted much attention
over the years. Originally, [18], the interest was driven by the fact that more general noncompact backgrounds than Minkowski space was most likely to be inevitable, if one considered a
theory that included gravity. In more recent years the insight into the non-perturbative properties
of string theory leading to M-theory, the non-compact backgrounds associated with anti-de Sitter
E-mail addresses: jonas.bjornsson@kau.se (J. Bjrnsson), stephen.hwang@kau.se (S. Hwang).

doi:10.1016/j.nuclphysb.2007.11.027
J. Bjrnsson, S. Hwang / Nuclear Physics B 797 [PM] (2008) 464498
465
(AdS) spaces became of central interest [9]. In spite of the rich development in this respect the
fact remains that surprisingly little is known about properties beyond low-energy solutions to
string theory for these backgrounds.
The first non-trivial model that was studied was string theory formulated as a WZNW model
based on the group SL(2, R) (or SU(1, 1), SO(2, 1)) [28] and [1012]. In Ref. [13] more
general classes of models were studied. These classes were based on the coset G/H where
H = H Z(H ), with Z(H ) being the one-dimensional center of H , is the maximal compact
subgroup of G. G/H is then associated with a non-compact Hermitian symmetric space. All
such possible spaces have been classified (see [14] and references therein). There are four large
classes1 : G = SU(p, q) with p, q 1, SO(p, 2) with p 3, SO (2n) with n 2 and Sp(2n, R)
with n 1, and, in addition, two cases associated with real forms of the exceptional groups
E6 (E6|14 ) and E7 (E7|25 ). The corresponding string theories studied in [13] are then represented as WZNW models associated with the cosets G/H . Time-like field components in this
model are represented by current modes associated with Z(H ). Z(H ) is topologically a circle,
S 1 , so that to get more realistic string models one should consider the infinite covering so that
S 1 R. For the purpose considered here, this is not particularly important, as the question of
unitarity does not depend on this.
In this work we will re-examine these models for all the above groups except the simplest
cases of rank one. In part this was motivated by the discovery of non-unitary states in the physical state space, contrary to the claim in [13]. Therefore, the conventional coset construction of
Goddard, Kent and Olive [15], the so-called GKO construction, is by itself inadequate in eliminating negatively normed states for the discrete highest weight representations considered.
This breakdown leads us to investigate the alternative formulation of the coset model based on
using the BRST symmetry, as initially proposed by Karabali and Schnitzer [16]. The BRST formulation will be shown to yield a unitary string theory for the same type of representations. This
is, to our knowledge, the first time the BRST and GKO coset constructions give fundamentally
different results. For coset constructions using compact groups the situation is different. There
the two constructions yield the same result [17], at least as far as unitarity is concerned.
We will also show that unitarity requires the level of the affine Lie algebra times the Dynkin
index of the embedding to be an integer and the highest weight components in the compact
directions to be integer valued. The Dynkin index is one in all cases except Sp(n, R), where it
is two. The quantization of the level is somewhat unexpected. In the simplest case, SL(2, R),
unitarity did not impose any restriction on the level. Furthermore, for the present case, as we
divide out all compact directions except one, one would intuitively believe that the level is not
quantized. A condition of integer level has, however, been noted previously in the context of
gauged WZNW models of non-compact type [18].
The string theories, that we will prove to be unitary, will have negative levels with an absolute
value that is larger than the critical one, i.e., the dual Coxeter number g . This will lead to
values of the conformal anomaly, c, that will be larger than the difference of the dimensions
of G and H . The requirement of integer (or half integer) values of the level, k, will lead to an
upper limit of c, which occurs for k = g 1, or k = g 12 for the case Sp(2n, R). In
addition, we will derive a condition k 2gh in all cases except for Sp(2n, R). This limit is, in
general, larger than 26, so that one may achieve criticality without other conformal field theories,
1 Due to isomorphisms between different Lie algebras, we only need to treat the cases p 5 for SO(p, 2), n 5 for
SO (2n).
466
provided the value 26 is an allowed value. The explicit expression of the conformal anomaly, for
h being simple, is
c = c g ch =
dh
kdg
k + gg
+ gh
(1.1)
where dg is the dimension of Lie algebra and = Ih g k. Ih g is the Dynkin index of the
embedding. We will give a list of models and the minimum and maximum values of c. One
simple example where we have criticality, is the case G = SU(2, 1), where H = SU(2). Here
g = 3 and the maximum allowed value is for k = 4. This gives precisely c = 26. In this
example the net number of dimensions is 8 3 = 5, i.e., the model represents a unitary string
theory on a five-dimensional spacetime. There is only one additional case where c = 26. This
occurs for SU(5, 1) at k = 13 where d = 11.
One would like to study other non-compact backgrounds, e.g., bosonic AdS spaces. Such
backgrounds may be studied using the WZNW models based on the coset SO(p, 2)/SO(p, 1) [6].
These models differ in an important way from the ones discussed above, namely in that time
is no longer embedded as a central element in the compact subgroup, but in a more complicated way. Although this case cannot be treated as a straightforward generalization of known
techniques, we believe that our treatment here may be of use for this case.
Let us introduce some notations and basic definitions. We will follow the conventions used
in [19]. Denote by g and h the Lie algebras corresponding to the non-compact group G and its
maximal compact subgroup H , which admits a Hermitian symmetric space of the form G/H .
Let gC and hC denote the corresponding complex Lie algebras. We will always take the rank rg
of g to be greater than one. One knows that h has a one-dimensional center, thus one can split h
as h u(1). We choose hC such that it is a normal embedding in gC , i.e., using the CartanWeyl
basis, the Cartan elements of hC , as well as generators corresponding to positive/negative roots,
are all in the corresponding decomposition of gC .
Denote by all roots, +/ the positive/negative roots, s the simple roots, c the compact
+
non-compact roots and +
roots, +
c = c the compact positive roots, n the
n the positive
non-compact roots. We take the long roots to have length 2. Let and define the coroot by
= 2(, )1 . Let (i) + denote the simple roots. When we need to distinguish between

different root systems, we denote by g and h the roots in gC and h C , respectively. We
fix the basis of the root space such that the highest root is non-compact. In addition, one can
g
choose the basis of roots such that if c then the first component is zero and the other
rg 1 components are, in general, non-zero. This, furthermore, yields an isomorphism between

g
c and h . gg denotes the dual Coxeter number of g. It is well known that there is a unique
non-compact simple root and if +
n then the coefficient of the non-compact simple root is
always one in a simple root decomposition of . Dynkin diagrams and relations between positive
non-compact roots of the Lie-algebras are presented in Appendix A in Figs. 114 following [20].
The CartanWeyl basis of gC is

H i , H j = 0,
i
H , E = i E ,
E ,E
= e, E
+ +,0
rg

i=1
i H i ,
(1.2)
467
where e, = 0 if + , i are components in the Dynkin basis, {(i) }, i = 1, . . . , rg , of the

weight space and i = ( , (i) ). H i , i = 2, . . . , rg , is a Cartan subalgebra of h C . The central
rg
(1)i H i , where we have used ( (i) , (i) ) = 2. With this
element in hC is given by H i=1
normalization we have

H, E = E
(1.3)
for +
n and zero otherwise. The CartanWeyl basis can be extended to the affine Lie algebra g C ,
i
j
Hm , Hn = mkG(g)ij m+n,0 ,
i
Hm , En = i Em+n
,
rg

2
+
i
i Hm+n +
mkm+n,0 ,
Em , En = e, Em+n + +,0
(1.4)
(, )
i=1
(i)
(j )
(g)
,
) is the metric on the weight space with inverse Gij = ((i) , (j ) )
where G(g)ij = (
and k is the level. The affine extension of the central element H will be denoted by Hn . We
denote by the affine roots and by |0; a highest weight state of a g C module with weight
= (, k, 0). It satisfies
g
J+ |0; = 0,
(1.5)
H0i |0; = i |0; ,
(1.6)
g
J+
g
J
i , E ,
} for m > 0, n 0 for + . We also define
= {Hmi , En , Em
= {Hm
where
n
+
C
Em } for m > 0, n 0 for . The irreducible highest weight g module that is defined by
g
acting with J on the highest weight state is denoted by H .

g
g
The generators are defined to have the Hermite conjugation properties (J+ ) = J w.r.t. g ,
where the minus sign appears for n and the plus sign otherwise. The generators of the
Cartan subalgebra are Hermitian. The norm of |0; is defined to be one. Norms of other states
g
g
g
|s J |s are then defined iteratively by s |s = s|(J ) J |s .
A weight is said to be dominant if (,

)
0, s and, in this paper, antidominant if (,
+ )
< 0, s . Here = 12 + and is the affine extension of this,
= (, g , 0). Subscripts g and h are used to distinguish the different .
The dominant and
antidominant weights are said to be integral if they have integer components. Integral dominant highest weight representations are often called integrable. Dominant affine weights require
k (, ) 0 and antidominant affine weights require k + g < 0 and k < (, ) 1.
The explicit form of the determinant of inner products, the ShapovalovKacKazhdan determinant [21,22], for a weight in a representation is
P (n)

1
det S []
=C
(, + ) n(, )
2
n=1 +
P (n(,0,m))

1
m k + g + (, + ) n(, )
2
n=1 m=1

r P (n(0,0,1))
m k + g g
,
n=1 m=1
(1.7)
468
is the degeneration of
where = (, k, 0), is the highest weight of the representation, P()
C is a constant. Using the above expression, yields an important propstates of weight ,
erty of antidominant weights, namely that the corresponding highest weight Verma modules are
irreducible. This follows since the determinant is then always non-zero.
The paper is organized as follows. First we consider the GKO coset construction and show the
problems that appear in this case. In Section 3 we prove unitarity based on the BRST approach
of Karabali and Schnitzer. In Section 4 we study the explicit form of the BRST invariant states.
The last section is devoted to discussions.
2. The GKO coset construction
In this section we will study the GKO coset construction of the G/H WZNW model. The
space of states in this construction is defined by the coset conditions
h
J+ |; = 0,
H0i |; = i |; ,
i = 2, . . . , rg ,
(2.1)
h
}, m > 0, n 0 for + , i = 2, . . . , r , and H i ,

where |; Hg , J+ = {Hmi , En , Em
g
c
0
i = 2, . . . , rg , span the Cartan subalgebra of h C . The physical string state space is defined as
the state space that, in addition to the coset conditions (2.1), also satisfies the standard Virasoro
conditions Ln | = (L0 1)| = 0, n > 0, where the Virasoro generators are constructed in
the standard fashion for coset models.
The problem with unitarity that arises for the GKO construction of these string theories
k (2). In this example,
k (2, 1) and h = su
are best illustrated in a simple example, g = su
(1)
(2)
(1)
(2)
= { , , ( + )} and c = { (2) }. Let |0; 1 , 2 be a highest weight
This means that 1 < 1, 2 < 1,
state in a g C module with antidominant highest weight .
1
2
k < ( + ) and k < 3. Consider the state

S; 1 2n, 2 + n E (1) n 0; 1 , 2 .
0
This state is easily seen to be a highest weight state of h C . The weight w.r.t. H02 is 2 + n,
so that for n 2 the weight is not antidominant. Taking 2 to be an integer, we then have
that the h C Verma module over |S; 1 2n, 2 + n is reducible. Thus, in this Verma module,
there exists further highest weight states. These states are orthogonal to any state that belongs to
a h C -module. In particular, it has zero norm. They are, however, not null-states as the original
Verma module is irreducible. Consequently, these states couple to states that are outside any h C
module, i.e., the g C Verma module is not fully reducible w.r.t. h C which shows that the treatment
of [13] is not valid (cf. Lemma 3).
One may show that the resulting string theories are non-unitary by considering the following
states

( (1) + (2) ) (1) n
(1) n+1 (2)
0; 1 , 2 .
E0
+ C1 E0
E0
| = E0
(2.2)
469
For C1 = 1/2 this is a highest weight state of h C . In addition, this state satisfies the Virasoro
conditions.2 The norm of this state is

n1

2 n 1 1
2
n!
1
1 + j ,
2
(2.3)
j =0
which is negative when n (2 + 1). The examples constructed above show that, even for the
finite dimensional case, the GKO coset construction fails. These problems are not specific to the
k (2, 1). They will appear for the other algebras in the class of models we consider
algebra g = su
here with the exception of
slk (2, R), since in this case h is absent.
One may also consider other discrete representations. A natural choice is the class of unitary
representations for Hermitian symmetric spaces considered in [23] and [20]. Then the irreducible
gC -module will be unitary and we avoid the problems encountered above. These representations
+
+
have highest weights such that (, ) 0 for +
c and (, + ) < 0 for \ c .
One of the first problems that arises when one tries to generalize to the affine case is that these
conditions cannot be straightforwardly taken to the corresponding affine conditions. Imposing
+
+ )

< 0 for all +
(,
) 0 for +
c and (,
c \ c leads to an inconsistency, as the
first condition implies k 0 whereas the second k < 0.
If one considers the simple state
|0; ,
E1
(2.4)
where is the highest root, then it is straightforward to see that these states are highest weight
states w.r.t. h C and, in addition, are annihilated by L1 . Thus it is a physical states provided
we can satisfy the
condition. The requirement of positive norm of the states gives
rmass-shell
g
(, (i) )i + k < 0. From (, + ) < 0 we see that k < 0. Thus, it
the condition i=1
follows that the condition of affine antidominant highest weights w.r.t. the non-compact roots,
i.e., (, + ) < 0, is the only condition of the two above that might be possible.
k (2, 1). However, even
One may make a more extensive analysis of some simple case, e.g., su
for such a comparatively simple case the analysis is very involved. Based on an assumption of
the form of the determinant of the inner product matrix of states in the coset, for some critical
weights and mode numbers, which follows from the knowledge of how h is embedded in g , one
finds that the number of negatively normed states does not stay fixed within a family of states.
Thus, the same type of phenomena that occurred above for purely antidominant highest weights
is likely to occur also in this case, implying that unitarity is broken.
3. The BRST approach
We now consider the alternative formulation of the coset construction that was proposed
in [16]. This approach uses the BRST symmetry to define the coset space. For compact algebras this was shown to yield the same result as the conventional GKO coset formulation using
highest weight conditions [17]. In order to construct a nilpotent BRST charge one starts with
the g WZNW model at level k and supplement it with an auxiliary sector, which is a h WZNW
2 The mass-shell condition is fulfilled by adjusting the highest weight and, if required, adding an internal conformal
field theory.
470
model of level = 2gh , where = Ih g k. Ih g is the Dynkin index of the embedding
( (gC ), (gC ))
,
( (h C ), (h C ))
Ih g =
(3.1)
where () is the highest root in each algebra. We will denote by h = h and the corresponding
k
h
current modes by H ni and E n where i = 2, . . . , rg and c . We define H to be the (irre
ducible) state space over a highest weight state with weight .
From the commutators of the subalgebra, see Eq. (1.4), one can determine a BRST-charge

En + E n :
:ci,n Hni + H ni : +
:cn
Q1 =
nZ
nZ,c
rg

i=2 c m,nZ
1
i :ci,m cn bmn
:

+
cn bmn : + +,0
e, :cm
,c m,nZ
rg

i
i :cm
cn bmn : ,
(3.2)
i=2
where : : denotes normal ordering and we have introduced the bc-ghosts with the non-zero
brackets

j
cm,i , bn = m+n,0 i j ,

cm , bn = m+n,0 +,0 .
(3.3)
It is conventional to define the following ghost vacuum
i
|0 b,c = bp |0 b,c = 0,
bm
cn,i |0 b,c = cq |0 b,c = 0,

+
(3.4)

for m 0; n > 0;
and p 0 and q 0;
and p > 0 and q > 0. The state space
spanned the bc-ghosts by acting with bc-creation operators is denoted by Hbc . The hermiticity
i and (c ) = c
, (cn ) = cn
, (bni ) = bn
properties are defined to be (bn ) = bn
n,i
n,i .
The coset construction is now formulated through the BRST condition, so that states in the
coset space satisfy
Q1 |S = 0,
b0i |S = 0,
i = 2, . . . , rg .
(3.5)
States satisfying these equations and that are non-trivial in the Q1 cohomology, i.e., non-exact,
are true states in the coset model. Now Eq. (3.5) does not represent physical states in our case,
since the string theory, that is represented by this WZNW model and possibly some unitary
conformal field theory coupled to it, is defined by including the Virasoro conditions. Thus, we
define the full BRST operator

Lgn + Lnh + Ln n,0 n
m:m n Pm+n :
Q = Q1 +
nZ
m,nZ

i
n:m cn
bm+n : + n:m cn,i bm+n
: .
m,nZ
(3.6)
471
Ln originates from some unitary CFT and (, P) are the usual conformal ghosts. Note that one
could have defined another BRST operator by defining

Q2 =
(3.7)
:(Ln n,0 )n :
m:m n Pm+n :,
nZ
m,nZ
g/h
where Ln = Ln + Ln , Lg/h Lg Lh . Then the full BRST charge, Q , could be defined
as Q Q1 + Q2 , since Q1 and Q2 commute. As these two charges, Q and Q , share the same
constraint surface there exists a canonical transformation which connects the two theories at the
classical level. Although this does not automatically imply equivalence at the quantum level, we
will show that they will lead to physical state spaces which are isomorphic. Here the P-ghost
state space HP is defined as for the bc-ghosts, with a vacuum state
Pm |0 ,P = 0,
n |0 ,P = 0,
(3.8)
for m 0 and n > 0. The corresponding state space is denoted by HP . The full ghost vacuum
is the product of the two separate ghost parts, |0 ghost = |0 b,c |0 ,P . We denote the product
space by Hghost = Hbc HP . We denote by H ghost the subspace of states satisfying b0i | = 0,
i = 2, . . . , rg , and P0 | = 0.
From the BRST operators one can extract a few interesting BRST exact quantities

H0tot,i Q, b0i = Q , b0i
(3.9)
i :bm
cm
:,
= H0i + H0i +
,m

(h ) j
1
i
i
: Hm+n
Gij bm :
H m+n
2( + gh )
mZ

(, )

Em+n Em+n bm :
:
+
2

h ,tot
Ln
Q ,

= Lhn
gh
+ Lnh + Ln ,
(3.10)
Ltot
n [Q, Pn ]
= [Q , Pn ] + Lnh ,tot

= Lgn + Lnh + Ln + Ln n,0 ,

gh
(3.11)
where

(g)
(, )
1
j
i

:Gij Hm Hnm : +
:Em Enm : ,
=
2(k + gg )
2
mZ

(, )
(h )
1

j
i

:G
Lhn =
H
H
:
+
E
:
:E
m nm
m
nm ,
ij
2( + gh )
2
c
mZ

(h )
(, )
1
i j

h
:Gij Hm Hnm : +
:Em Enm : ,
Ln =
2( + gh )
2
Lgn
mZ
472

gh
Ln =

i
cm,i : +
m:bnm
cm : .
m:bnm
(3.12)
mZ
In the above expressions we have, for simplicity, assumed that h is simple, which is not the
case for g = su(p, q). The above expressions are easily modified for this case. The resulting
expressions are sums of two terms corresponding to a decomposition in terms of two simple
subalgebras.
Using Q as our BRST charge, the physical state space of the string theory is defined by the
conditions
Q| = 0,
b0i | = 0,
i = 2, . . . , rg ,
P0 | = 0.
(3.13)
h
g
Q
We denote by H the sub-space of states of H H HlCFT
Hghost satisfying the above

represents some unitary CFT. If we replace Q
equations and being Q non-exact. Here HlCFT

Q
Q

by Q we denote the corresponding state space by H . States in H have to satisfy

H0tot,i | = 0,
(3.14)
Ltot
0 | = 0,
(3.15)
which follows directly from Eq. (3.13) by taking the commutator of Q with
and P, respectively. The cohomology defined by the above equations (3.13)(3.15) is often called a relative
cohomology.
The representations that we will focus on in this work are antidominant highest weight representations for g C . We believe that these are relevant for the string theories that we consider
here. This belief is mainly motivated by the fact that, as we will show, that the corresponding
string theories are unitary. The other class of natural discrete representations are the ones that
are unitary for the finite dimensional case [20,23]. For these representations the situation is far
more complicated since the corresponding affine state spaces have a complex structure, which at
present has not been worked out.
For the h -sector, the class of representations that are natural is found by studying the requirement that there should exist conventional BRST invariant ground-states. Such states are of the
form
b0i

|0; ,
|0; |0;
|0 ghost .
(3.16)
Using Eq. (3.14) we have

0 = H0tot,i |0; ,
= i + i + 2hi |0; , ,
i = 2, . . . , rg .
(3.17)
Thus, i + i + 2hi = 0, i = 2, . . . , rg , so that if we choose to be antidominant we must

require to be a dominant h C weight. As we will see, the requirement of unitarity will, in the
generic case, single out dominant integral representations of the auxiliary h -sector as the only
possible ones. This implies that the auxiliary sector has representations that are unitary. This is in
contrast to the situation for the coset construction of a unitary CFT for the compact case, where
the auxiliary sector has antidominant weights and thus the representations are non-unitary. Note,
473
that one may straightforwardly show that one needs to have an antidominant component of in
the non-compact direction.
Q
For a general state to be in H we have to require Eq. (3.14), which implies i + i +

r
2hi jg=1 mj (j )i = 0 for some integers mj and i = 2, . . . , rg . As (j )i Z and hi = 1, i,
we have the following lemma.
Q
Lemma 1. If H
is non-trivial then (i + i ) Z, i = 2, . . . , rg .
4. Unitarity
In Ref. [24] a technique was introduced to analyze the relative cohomology. We will adapt
this technique to the present case. Define, therefore, the character
h Vir)
(g,
(, , )

rg

tot
tot,i
Ngh
Tr exp 2i L0 exp i
i H0 + iH0 (1)
,
(4.1)
i=2
Ngh is the ghost number of the state in question relative to the ghost vacuum. The trace is taken
h
g
over all states in H H HCFT
Hghost . Therefore, the character decomposes into separate
parts
h Vir)
(g,
(, , ) = e2i g (, , ) h (, ) CFT ( ) gh (, ) CFT gh ( ).
(4.2)
As is well known [24], the character defined in Eq. (4.1) gets only contributions from non-trivial
BRST invariant states. However, since the physical states not only satisfy the BRST condition,
but all the conditions in Eqs. (3.13), (3.14) and (3.15), we must instead consider the following
function

h Vir)
d B (g,
(, )

h Vir)

(di ) (g,
(, , )
i=2
rg

rg

i=2
rg

tot
di Tr exp 2i L0 exp i
i H0tot,i + iH0 (1)Ngh ,

i=2
(4.3)
h
g
H ghost . The - and -integrations are forwhere the trace is now taken over H H HlCFT

mal integrations to project onto the - and -independent term, d d e2ip eir = p,0 r,0 ,
h Vir) (, ) the generalized
which is required by Eqs. (3.14) and (3.15). We will denote B (g,
branching function.3 This definition was first introduced in [17] and extended in [26]. We also
3 The antidominant highest weight representations are not completely reducible w.r.t. to h
C , as was shown in Section 2.
Consequently, one cannot define a branching function in the conventional fashion. The definition used here coincides with
branching functions for integrable representations [26] and is, therefore, a natural generalization of this concept to the
present case.
474
define another function, which we will call a signature function:

rg

tot,i
(g,h Vir) (, , ) Tr exp 2i Ltot

exp i
i H0 + iH0 (1)Ngh . (4.4)
0
i=2
The prime on the trace indicates that the trace is taken with signs, i.e., a state with positive
(negative) norm contributes with a positive (negative) sign in the trace. We define a corresponding
coset signature function

h Vir)
(, )
d S (g,

rg

h Vir)

d
(di ) (g,
(, , )
i=2
rg

rg

tot
tot,i
Ngh
di Tr exp 2i L0 exp i
i H0 + iH0 (1)
.

i=2
i=2
(4.5)
Since the projection of the character onto states satisfying Eqs. (3.14) and (3.15) gives the total
Q
number of states in H for given weights and the same projection of the signature function

gives the difference between the number of positive and negative norm states in the same state
space, we have the following lemma.
Lemma 2. HQ is unitary if, and only if,

h V ir)
h V ir)
d B (g,
(, ) S (g,
(, ) = 0.
(4.6)
It should be remarked that for the string on a Minkowski background, the -integration enforcing the mass-shell condition, may be dropped, as the momentum squared is allowed to take any
value, in particular, any negative value allowing arbitrarily large grades. In the present situation
and taking, for simplicity, h to be simple, the momentum squared is replaced by the difference
(,+2g )
2(k+gg )
(,
+2
h )
2(+gh )
which, in general, does not take any value and, in particular not arbitrarily
large negative values. This is obvious from the mass-shell condition

+ 2h )
(, + 2g ) (,
+ N + l 1 = 0,
2(k + gg )
2( + gh )
(4.7)
where l 0 originates from some unitary CFT. We now state the main result of this paper.
Theorem 1. Let be an antidominant weight. Furthermore, denote by Nmax the largest grade
that is allowed by the mass-shell condition. Assume HQ to be non-trivial.

(i) Necessary and sufficient conditions for H
to be unitary are one of the following:
(a) Nmax = 0:
i , i = 2, . . . , rg , are integral and is dominant integral.
475
(b) 0 < Nmax < (h , )

+ 1:
is dominant.
i , i = 2, . . . , rg , are integral, is dominant integral and
(c) Nmax (h , )
+ 1:
is dominant integral.
i , i = 2, . . . , rg , are integral and
Q
(ii) H
Q
and H
are isomorphic.
Note that the cases (a)(c) are formulated in terms of existence of states up to a certain grade.
Q
The first case (a) occurs if H has only non-trivial states at zero grade, (b) occurs if there are

only physical states at levels less than the first null or non-unitary state in the tilde sector, while
(c) is the generic case. It is only in the case (c) that is required to be an integer.
Proof. We first prove that the stated conditions are sufficient for unitarity making use of
Lemma 2. We must, therefore, determine the characters and signature functions of the different sectors of states. The character for the g C module is straightforward to determine as the
Verma module is irreducible for our choice of representations. We have
g
C2 ()
g (q, , ) = q 2(k+gg )

exp i(, )
+
c

+
n

m=1

+
n

+
n
1
1 exp[i(, )]
1
1 exp[i] exp[i(, )]

1
1
r
m
m
g
(1 q )
1 q exp[i(, )]
c
1
1 q m exp[i] exp[i(,
)]

1
,
1 q m exp[i] exp[i(, )]
(4.8)
where C2 () = (, + 2) is the quadratic Casimir and is the highest weight w.r.t. the finite dimensional algebra, q = exp[2i ]. Furthermore, we have introduced the notation
rg i
i=2 (i) . As compact roots have no first component we suppress the notation for these
rg
1
roots, i.e., = for c . We have also defined (, ) 2( (1) , (1) ) 1 + i=2
i i .
Next, we determine the signature function. This is also straightforward for the g -sector, as
one can compute the signature function in the limit of large absolute values of the weights and
level. This follows since the ShapovalovKacKazhdan determinant formula (1.7) shows that the
determinant of inner products does not pass any zeros in taking this limit and, consequently, the
norms of the states do not change signs. In this limit, the affine Lie algebra will diagonalize and
we have one time-like direction corresponding to the compact center that is not in h . We find,
476
therefore,
g
C2 ()
g (q, , ) = q 2(k+gg )

exp i(, )
+
n

m=1
+
c
1
1 exp[i] exp[i(, )]

1
1
r
m
m
g
(1 + q )
1 + q exp[i(, )]
c
+
n
+
n
1
1 + exp[i(, )]
1

1
.
(4.9)
The character for the bc-ghosts is straightforward to compute and is given by

2

1 exp i(, )
gh (q, ) = exp i(, 2h )
+
c

2
m 2(rg 1)
m
1q
.
1 q exp i(, )
(4.10)
m=1
The signature function is also easy to determine. Each ghost pair will give two states of opposite
ghost numbers. Diagonalizing this pair of states, one finds one state of positive norm and one of
negative norm. Therefore, the signature function is

gh (q, ) = exp i(, 2h )
1 exp i(, ) 1 + exp i(, )

+
c

r 1
r 1
1 qm g 1 + qm g
m=1

m
m
1 q exp i(, ) 1 + q exp i(, ) .
(4.11)
In the same way one can determine the P-ghost character to be

CFT gh (q) =

2
1 qm
(4.12)
m=1
and the signature function
CFT gh

1 qm 1 + qm .
(q) =
m=1
(4.13)
477
Putting all the above pieces together we have the character
1 (, , ) q
g
C2 ()
)
2(k+gg

exp i(, ) + i(, 2h )

1 exp i(, )

+
c

1 qm
m=1

+
n
+
n

+
n
1
1 exp[i] exp[i(, )]
rg 1

1 q m exp i(, )
1

1
.
(4.14)
In the same way the corresponding signature function is found to be 1 (, , ) = 1 (, , ).
h
Let be the character of the h sector. If the signature function of this sector satisfies
h
h
h Vir) (, , ) =
= , then we have for the full character and signature function (g,
h
h
h Vir) (, , ) and by Lemma 2 we have unitarity. But =

(g,
occurs precisely when
h
is dominant integral, see Theorem 11.7(b) in Ref. [25]. Then
H is unitary, which is when
is integer and, by Lemma 1, i , i = 2, . . . , rg , is integral, or else HQ is empty. Thus, we have

h
h
proven the sufficient conditions in case (c). For case (a) we only need that = holds for
the terms corresponding to grade zero. Then it is sufficient that the finite dimensional algebra
has a unitary representation. This is the case when is dominant and integral, which implies by
Lemma 1, that i , i = 2, . . . , rg , is integral. Finally, for case (b) we have from the grade zero
case, (a), that is dominant integral and i integral, i = 2, . . . , rg . In addition, we need to see
h
under what conditions one has unitarity for H up to grade Nmax . We can always take to
be sufficiently large so that the ShapovalovKacKazhdan determinant, Eq. (1.7), has no zeros
corresponding to affine roots. Then all states in the state-space are unitary by case (a). As the
value of is decreasing, one finds a first zero corresponding to the simple root (h , 0, 1). But
+ 1 Nmax > 0, so that the restriction on Nmax will prevent this zero from appearing.
(h , )
Therefore, the determinant does not change sign in going from grade zero to Nmax , which proves
unitarity for case (b).
We now proceed to prove the necessary conditions for unitarity. Again we will use Lemma 2.
g
Recall that 1 (, , ) (, , ) gh (, ) CFT gh ( ) and that we have, using the above
explicit expressions for the characters and signature functions for the g- and ghost sectors,
1 = 1 . This implies that
478

0=

=
h Vir)

h Vir)
di (g,
(, , ) (g,
(, , )
h

h
di e2i 1 (, , ) (, ) (, ) .
(4.15)
We make a general expansion and write4

h
h
(q, ) (q, ) = q
h
C2 ()
2(+g )
h
ei(,)

n=0
Nn, q n ei(,) ,
(4.16)
where the coefficients, Nn, , are twice the number of negatively normed states at the specific
grade and weight corresponding to n and . The sum over is over the root lattice of h C ,
except for the q 0 term, where the sum is over the positive root lattice. We can make a similar
decomposition of 1
g
C2 ()
1 (, , ) = q 2(k+gg )

exp i(, ) + i(, 2h )
Mm, q m ei(,) ,
m=0
(4.17)
where the coefficients Mm, are known through an expansion of the explicit expression of the
characters Eqs. (4.8), (4.10) and (4.12). Inserting the two expansions into Eq. (4.15) yields

dq

i
l 0
di q
g
h
C2 ()
C ()
2 2
2(k+gg ) 2(+g )
h
MlCFT
ql

Mm, Nn, q m+n ei(,)i(,) = 0,
(4.18)
m,n=0 ,
where we have inserted a contribution from a unitary CFT. We now study these equations and
we will show the following lemma.
Lemma 3. If Eq. (4.18) implies N0, = 0 then the equations also imply Nn, = 0 for n =
1, . . . , Nmax , where Nmax denotes the largest positive integer for which
+ 2h )
(, + 2g ) (,
(Nmax 1),
2(k + gg )
2( + gh )
(4.19)
or else Nmax = 0.
Proof. The integration over q in Eq. (4.18) enforces the mass-shell condition Eq. (3.15). This
condition is of the general form
+ 2h )
(, + 2g ) (,
+ N + l 1 = 0,
2(k + gg )
2( + gh )
(4.20)
4 In this case and in other similar expressions below, we have, for simplicity, taken h to be simple. For h not simple,
h
C2 ()
is replaced by a sum of two similar terms.

2(+g )
h

Q
where N is the grade. If this equation has no solution, then H

our assumption. If
479
is empty, which is contrary to
+ 2h )
(, + 2g ) (,
>0
2(k + gg )
2( + gh )
(4.21)
then Eq. (4.20) implies N = 0 and the integration over q will only get a contribution from the q 0 term of the sum in Eq. (4.18). In this case the lemma is trivially true. If Nmax satisfying Eq. (4.19)
exists, then the integration over q may get contributions from any term q m+n , m + n Nmax ,
in the sum, since we must be able to solve the equation for any l 0. In this case, Eq. (4.18)
implies the equations

i
di

Mm, Nn, q m+n ei(,)i(,) = 0
(4.22)
m,n=0 ,
for any m + n Nmax . The right-hand side of this equation depends on the two parameters q and
exp(i). Thus, we can expand the equation w.r.t. these parameters. Focussing on the first two
powers of q we have the zeroth order term

di
M0, N0, ei(,)i(,) = 0,
(4.23)
i
and the first order term

di
[M0, N1, + M1, N0, ]ei(,)i(,) = 0.
i
(4.24)
By assumption, Eq. (4.23) implies N0, = 0. Inserting this into Eq. (4.24) yields that N1, satisfies exactly the same equation as N0, . Thus, N1, = 0. We easily see that this continues to
higher orders so that the q n -order equation gives Nn, = 0. This proves the lemma. 2
We now want to establish that N0, = 0 follows from Eq. (4.18), i.e., that Eq. (4.23) implies
N0, = 0. First we will consider the simplest cases when the rank of g is two. In this case there
are only two distinct cases, corresponding to the complex algebras A2 or B2 and we establish the
following result.
Lemma 4. Let g = su(2, 1) or sp(4, R). Then either H
is empty or Eq. (4.23) implies N0, = 0
which in turn implies that is dominant and integral and is integral.

for all ,
Proof. In these cases h = su(2). The character 1 is then

1 (, ) = ei(+2) 1 e2i
1
1
1 ei() 1 ei(+)
p

= ei(+2) 1 e2i
eip+i(p2r)
p=0 r=0
(4.25)
480
up to two factors exp(i ) and q () , which are unimportant in the following. We have, furthermore,
su(2)
su(2)
= ei
N0,m e2im .
(4.26)
m=0
Inserting these two expressions into Eq. (4.23) yields

p

d ei(++2)
1 e2i ei(p2r2m)
N0,m eip
0=
m,p=0
r=0
N0,m eip

d ei(++2+p2m)
ei(+p2m)
m,p=0
N0,m eip (++2+p2m,0

+p2m,0
).
(4.27)
m,p=0
Q
.

If + is not an integer then by Lemma 1 there are no non-trivial states in H

integer, then Eq. (4.27) is non-trivial and we can simplify the equation to
0=
N0,m ei(2m2)
m=a
b
If + is an
N0,m ei(+2m)
,
(4.28)
m=0
]} and b = [ +2 ]. Here [ ] denotes the integer part. This can be

where a = max{0, [ +2+3
reduced to

N0,++2+m
(4.29)
=0
for max 0, ( + + 2) m < ,
+
N0,m = 0 for 0 m
.
N0,++1m
(4.30)
One can now use the representation theory for discrete representations for su(2). Eq. (4.29)
implies that only finite dimensional representations are possible. This follows since the sign
of infinite dimensional representations of su(2) is alternating at a finite weight, contradicting
Eq. (4.29). Thus, Z+ .
We now have two different possibilities to consider. Either all states have positive norms or
all have negative norms. The second case is impossible, since this implies
N0,m = 0,
N0,m = 0,
0 m ,
+ 1 m < .
(4.31)
Inserting + 1 < 0 into Eq. (4.29) implies N0, = 0, which is a contradiction. This proves the
assertion of the lemma for g = su(2, 1).
The case g = sp(4, R) is proved in a similar way. The compact root in this case is the short
root, so the character 1 is

1 (, ) = ei(+2) 1 e2i
1
1 ei+2i
1
1
i
i2i
1e
1e
up to factors exp(i ) and q () , which are unimportant in the following context.
(4.32)
481
Inserting this equation and Eq. (4.26) into Eq. (4.23) yields

1
i(n+m)
d ei(++2)
0=
1 e2i ei(2n2p2m)
N
e
0,m
i
1e
=
1
1 ei
m,n,p=0

m,n,p=0
+
2
N0,m
m=0
N0,m ei(n+m) [++2+2n2p2m,0

++2n2p2m,0
]
ei(2nm+(+)/2)
n=0
m= +
2 +1
1
=
1 e2i
N0,m

ei(2nm+(+)/21)
n=m +
2 +1
N0,m e
i(m(+)/21)
m= +
2 +1
2
N0,m e
i(m+(+)/2)
. (4.33)
m=0
From the first equality of this equation, one can see that if +
/ 2Z then there does not exist
Q
any non-trivial states in H . So, we take + 2Z. Eq. (4.33) then yields

N0, + +1+m N0, + m = 0,

2
N0,++2+m
= 0,
0 m +2 ,

max 0, ( + + 2) m < .
(4.34)
The last equation is the same as Eq. (4.27), therefore, Z+ . From Lemma 1, is integervalued. This concludes the proof for the case sp(4, R). 2
We now consider the general case and take g to have rank three or more. Let denote a noncompact root w.r.t. g. Let furthermore (i) be a long simple compact root. Assume + (i) is a
root but + 2 (i) and (i) are not roots. Then we have a subalgebra gC
1 with real form g1 ,
(i) , E , E (+ (i) ) , H i , H H j and H + (i) H + H i .
where gC
is
generated
by
E
j
1
This algebra is A2 , hence, the real form is g1 = su(2, 1). Assume is non-compact root and (i)
is a short simple root. Furthermore, assume + (i) and + 2 (i) are roots, but + 3 (i) and
C
(i) are not roots. Then we have a subalgebra gC
2 with real form g2 , where g2 is spanned
(i)
(i)
(i)
(i)
(i)
by E , E (+ ) , E (+2 ) , H i , H j H j , H (+ ) H + 2H i and H (+2 )
C2 , hence the real form is sp(4, R). We can now prove the
H + H i . This algebra is B2 =
following.
g
Lemma 5. Let g have rg 3 and (i) c s then there exist g1 g if (i) is long or g2 g
if (i) is short.
Proof. Consider first the case were (i) is a long simple root. We should prove that one can
always choose such that + (i) is a root but + 2 (i) and (i) are not roots. This
non-trivial fact follows straightforwardly by inspection of the explicit diagrams of non-compact
positive roots given in the appendix of Ref. [20], which for convenience is reproduced in Appendix A of this paper. Consider, e.g., the case g = so(2p, 2). The Dynkin diagram and the diagram
482
of non-compact positive roots is depicted in Figs. 3 and 4 respectively, where (1) is the unique
simple non-compact root. Choosing (i) = (3) we can take = (1) + (2) . We can infer from
the diagram that + (3) is a root but + 2 (3) and (3) are not roots, so the correct
subalgebra is g1 . This one can do for all simple long roots.
Consider now the case when (i) is a short root. One may in the same way study the explicit
diagrams for non-compact positive roots of the corresponding diagrams. For instance, consider
sp(2p, R) for which the Dynkin diagram and diagram of non-compact positive roots are depicted
in Figs. 9 and 10, respectively. Consider the compact short root (3) . From the graph it follows
that = (1) + 2 (2) . From the same graph it follows that + (3) and + 2 (3) are roots but
(3) and + 3 (3) are not, so the correct subalgebra is g2 . The other algebras may be treated
completely analogously. 2
We now continue the proof of the theorem and consider first the g1 algebra connected to the
simple root (i) and the corresponding auxiliary h 1 = su(2) algebra. The corresponding ghost
operators are given by c0 , b0 , c0,i and b0i . Introduce a grading

grad E0 = 0, (i) , , + (i) ,

(i)
grad E 0 = 0,

(i)
grad c0 = 0,

(i)
grad b0 = 0.
(i)
(i)
(4.35)
The rest of the operators have grad minus one and, furthermore, the vacuum in each sector has
grad zero. Then a generic state may be written as a sum of states of different values of grad,
|S = |N = 0 + |N = 1 + ,
(4.36)
H (i),ghost . We write the BRST charge

where |N = 0 is in Htot(g1 ) (i) H 1(g1 ) H 1(i) HlCFT
as
g
Q = Qsu(2) + Qrest ,
(4.37)
where Qsu(2) is the BRST charge for the su(2) subalgebra connected to (i) . The BRST condition yields
0 = Q|S
= Qsu(2) |N = 0 + Qrest |N = 0 + O(N = 1).
(4.38)
Qsu(2) |N
= 0 is a state in the N = 0 sector of states and Qrest |N = 0 = O(N = 1).

Now,
Thus, Qsu(2) |N = 0 = 0 and the problem reduces exactly to the one treated in Lemma 4. This
means that if there are non-trivial solutions, unitarity requires (i) to dominant, integral and
(i) to be integral. A short root (i) , for which we have a subalgebra g2 , is treated analogously,
the only difference being the grading. By Lemma 5 we can choose i = 2, . . . , rg , so that the
same conclusion applies to all components of and to the components i = 2, . . . , rg of . This
concludes case (a) of the theorem.
Q
If there are states in H up to level Nmax , then from Lemma 3 we have Nn, = 0 for n =

h
0, . . . , Nmax . This implies that H is unitary up to and including level Nmax . If Nmax <
(, )
+ 1 then there does not exist any new zeros in the determinant of the inner products,
483
h
Eq. (1.7), in H (cf. the discussion in the end of the first part of the proof). This together with
the requirement that is dominant and integral coming from states at grade zero, implies that
has to be dominant. This is case (b).
Finally, if Nmax (, )
+ 1, then we have
h n
|n E1
|0, ,
(4.39)
where n is the smallest number such that n (, )

+ 1. This state has either negative or
zero norm. Zero norm occurs if and only if n = (, )
+ 1. Therefore, Z+ since is
is dominant and integral, which proves the necessary conditions
dominant and integral. Thus,
in (c).

Let us prove the last assertion (ii), i.e., that HQ and HQ are isomorphic. Replacing Q with

Q does not change the expressions for the characters and signature functions, since the traces
are taken in the same state spaces. From Eqs. (3.9) and (3.11) it follows that H0tot,i and Ltot
0 are
also BRST exact w.r.t. Q . Therefore, the generalized branching and coset signature functions
will again only get contributions from non-trivial Q invariant states. Hence, the isomorphism
follows and, in particular, the conclusion that we will achieve unitarity still holds. This concludes
the proof of the theorem. 2
From the proof above we have, using the explicit expression for the character of h C for
integrable highest weights, the following
Corollary 1. The generalized branching function for the unitary cases in Theorem 1 is given by
(a)

B (g,h ) () =
dq
rg

(di )q
h
C2
1]
2(k+gg ) 2(+g )
h
g
C2
i=2

exp i( + + 2h , ) + i

1
exp[i]
exp[i(, )]
+
n

(1)sign(w) exp i w( + h ) h ,
wW (h )

Tr q L0 1 .
(4.40)
(b)
B
(g,h )

() =
dq
rg

i=2
(di )q
h
C2
1]
2(k+gg ) 2(+g )
h
g
C2

exp i( + + 2h , ) + i

1
exp[i]
exp[i(, )]
+
n
484

m=1 +
n
+
n

m=1
1
)]
1
1
(1 q m )

Tr q L0 1 .

(1)sign(w) exp i w( + h ) h ,
wW (h )
(4.41)
(c)
B (g,h ) () =
h
C2
C2
rg

[
) 2(+g ) 1]
2(k+gg
h
dq
(di )q
g
i=2

exp i( + + 2h , ) + i

1
exp[i]
exp[i(, )]
+
n

m=1 +
n

m=1
)]
+
n
1
1
(1 q m )
)]
(1)sign(w)
wW (h )

exp i w + h + + gh h ,
+
)(
+g
(,
h )+ 12 (,
h )

Tr q L0 1 ,
(4.42)
where L is the coroot lattice of the horizontal Lie algebra h C .

We remark here again that these expressions are valid for h simple, but may easily be generalized to the case where we have a sum of simple terms.
Let us end this section by proving one additional result. Using the coset construction one can
define a conformal field theory as the G/H coset, where H is the maximal compact subgroup
of G. We will prove that this CFT is unitary for integral dominant and antidominant weights, as
above.
We define the coset model by using the BRST charge Q1 defined as Q1 in Eq. (3.2) with
the only difference that one adds the contribution for the one-dimensional center. For the same
highest weight representations as considered above one easily derives the characters and signature functions, as the only difference comes from the extra u k (1) excitations. The character and
signature function of the g -sector is unchanged from above Eqs. (4.8) and (4.9). For the auxiliary
485
sector we now add the extra u k (1) field and get
(q, ) = q
h
u(1)
C2
C
22k ]
2(+g )
h
exp i(,
)

+
c
1
1 exp[i(, )]

1
1
(1 q m )rg
1 q m exp[i(, )]
c
m=1

(1)sign(w)

wW (h )
+
)(
+g
(,
h )+ 12 (,
h )
(4.43)
Finally, the bc-ghost character and signature functions are given analogously by Eqs. (4.10)
and (4.11), where rg 1 is replaced by rg . Putting all parts together we find that the complete
generalized branching function
h)
(q)
B (g,
=q
h
u(1)
C2
C
22k ]
2(k+gg ) 2(+g )
h
g
C2
d

+
n
rg

di exp i( + + 2h , ) + i( + )
i=2
1
1 exp[i] exp[i(, )]

m=1 +
n
(1)sign(w)
wW (h )
)]
)]

+
)(
+g
(,
h )+ 12 (,
h )
(4.44)
and coset signature functions are equal. The resulting coset space is, therefore, unitary. We have,
thus, proven the following theorem.
h
Theorem 2. Let | H H Hghost . The G/H coset conformal field theory defined by
Q1 | = 0,
b0i | = 0,
i = 1, . . . , rg ,
(4.45)
The generalized
is unitary for antidominant integral weights and dominant integral weights .
branching function is given by

h)
(q) = exp
B (g,
h
u(1)
C2
C2
C2
2(k + gg ) 2( + gh )
2k
486
rg

di exp i( + + 2h , ) + i( + )
i=2

+
n
1
1 exp[i] exp[i(, )]

m=1 +
n
wW (h )
1
1
1 q m exp[i] exp[i(, )] 1 q m exp[i] exp[i(, )]
(1)sign(w)

+
)(
+g
(,
h )+ 12 (,
h )
(4.46)
and the corresponding one for h not simple.

By expanding the nominators above and performing the integration, as was done in [26], one
may give a more explicit but involved expression for these functions.
5. BRST invariant states
In this section we will investigate the nature of the BRST invariant state space that we in
the preceding section have shown is unitary. We restrict ourselves to states which belong to the
relative BRST cohomology, that is, states which satisfy
b0i | = P0 | = 0,
i = 2, . . . , rg .
(5.1)
We decompose the BRST charge accordingly

0.
Q = Q + H0tot,i c0,i + Mi b0i + MP
(5.2)
1 Q1 H tot,1 c0,i
We first determine the states in the cohomology for the BRST charge Q
Mi b0i and then the states in the relative cohomology for BRST charge Q.
In [17] a technique was introduced to be able to analyze the cohomology. This technique
was an extension of techniques first used in the string context in [24]. A fact which makes the
present situation more difficult is that the h -sector has a highly reducible highest weight Verma
module. Null states make it impossible to define a homotopy operator in the way that was done
in [17]. We can, however, adapt the techniques to our case by focusing on the g-sector instead, as
this Verma module is irreducible. The embedding of highest weight h C modules is, as we have
discussed in Section 2, not trivial. This will imply, as we will see, that we will not, in general, be
able to reduce the BRST invariant states to the form that we would expect. In the following the
dependence on the conformal ghosts and P will be suppressed.
487
Let us introduce a gradation for a subset of the operators

grad En
= 1,
for n > 0; n = 0 and
/ +
c ,
i
grad Hn = 1, for n > 0, i = 2, . . . , rg ,

grad bn
= 1,
for n > 0; n = 0 and
/ +
c ,
i
grad bn = 1,
for n > 0, i = 2, . . . , rg ,

/ +
grad cn = 1, for n > 0; n = 0 and
c ,
i
grad cn = 1, for n > 0, i = 2, . . . , rg .
(5.3)
Let all other operators in g C , as well as the h -sector have zero grad. We also define a gradation
of states by defining the highest weight state in the g - and h -sectors as well as the ghost vacuum
defined in Eq. (3.4) to have zero grad. The grad of states in the h -sector as well as the ghost
sector is then fixed by applying the gradation in Eq. (5.3). The grad of states in the g -sector is
fixed by defining an ordering of operators such that all creation operators in h C are moved to
left of the rest of the creation operators and applying the gradation of the operators in Eq. (5.3).
1 = d0 + d1 , where the index denotes
The gradation splits the BRST charge into two terms, Q
that applying this to a state of grad N one gets terms with at most grad N or N 1, respectively.
The interesting operator is d0 , which has the form

i
cn En
+
cn,i Hn
+
c0 E0 .
d0 =
(5.4)
n>0,c
n>0
+
c
A state which has a grad N , which we always assume to be finite, has the general form
h ,1
|p, q = J
h ),1
h ,p (g,
J
. . . J
h ),r1
(g,
. . . J
1
|0; R |s b
. . . b |gh ,
(5.5)
where |s is an arbitrary state in the h -sector. We may without loss of generality assume these
states to have a definite ghost number as well as Ltot
0 eigenvalue. We have here introduced a
h )
(g,
h
simplified notation where J is a generic creation operator in h C , J
is a generic creation
q

C
operator which is not in h and b is a generic b-ghost creation operator. We now define a
homotopy operator 0 . This operator acts on the states with p > 0 as
0 |p, q =
p

1 h ,1
h ),1
h ),r1 i
h ,p (g,
(g,
h ,i
J . . . J . . . J J
. . . J
b
p+q
i=1
1
|0; R |s b
. . . b |gh .
(5.6)
In the sum above, capped terms are omitted. One can see that this homotopy operator satisfies
(0 d0 + d0 0 )|p, q, N = (1 p+q,0 )|p, q, N + O(N 1).
(5.7)
For states which are BRST invariant and has p > 0 one can directly see that the order N terms
are trivial
|N = d0 0 |N + O(N 1)
= Q 1 0 |N + O(N 1).
(5.8)
488
Therefore, the only states one needs to consider are linear combinations of states of the form
h ),1
(g,
|0, q = J
h ),r1
(g,
. . . J
1
|0; R |s b
. . . b |gh .
(5.9)
In the same way one can show that terms involving b excitations are not BRST-invariant if there
h )
(g,
does not exist any J

excitations. This implies that there cannot exist any c-ghost excitations,
leaving non-trivial states in the cohomology of the form
h ),1
(g,
|0, 0 = J
h ),r1
(g,
. . . J
|0; R |s |0 bc ,
(5.10)
1 on an arbitrary linear combination | of these

or linear combinations of these. Applying Q
states implies that it has to satisfy
h
h
J+ + J+ | = 0.
(5.11)
This is as far as we have been able to determine the form of the states in the coset space using
the BRST formulation. Notice that this form is not manifestly BRST invariant. In general, states
of this form can only be invariant for non-trivial |s . We believe, however, that one may be able
to go one step further. Our conjecture is that for integral dominant weights of h C it is possible
to reduce the non-trivial states to the following form
| = |hw; |0;
|0 bc .
(5.12)
We note here that

Here, |hw; is a highest weight state w.r.t. h C in g C with highest weight .
states of this form are trivially BRST invariant. They also appear to be of the form that one would
get through the conventional coset condition in the g-sector. However, there is a crucial difference. Applying H0i,tot to this state and requiring it to be zero to get non-trivial BRST invariant
states, yields that the weights satisfy
i + 2 i = 0,
i +
i = 2, . . . , rg .
(5.13)
is a dominant integral weight this condition implies that is antidominant and integral
Since
w.r.t. h C . Comparing with the conventional coset formulation, we thus see that the BRST formulation restricts the highest weights of h C that can appear in g C to be antidominant, whereas
the conventional coset construction does not. It is precisely this which makes the BRST formulation give unitarity and the conventional coset formulation fail to do so. The non-unitary highest
weight states that need to be projected out are done so in the BRST approach by being trivial
BRST invariant states.
We will give two arguments in support for our conjecture that all states in the cohomology
can be written as in Eq. (5.12). If we study the limit k and large absolute values of the
highest weights for g C - and h C -modules, then it is simple to construct the general solution to the
BRST condition. In this limit the compact and non-compact generators decouple from each other,
implying that linear combinations of states of the form Eq. (5.27) are highest weight states w.r.t.
to the h C currents in the g -sector. This in turn implies that the states are only BRST invariant if
|s is highest weight w.r.t. the auxiliary h C current modes leading to |s = |0; . Consequently,
in this limit our conjecture is true.
The second argument is based on an explicit example from su(2, 1) which shows how it would
work. Let us take a simple example from the horizontal part of the algebra. The relative BRST
charge for the horizontal part of the algebra is

(2)

(2)
(2)
(2)
(2)
(1)
Q = E0 + E 0 c0 + E0 + E 0 c0 .
489
(5.14)
Consider the state

( (1) + (2) ) 2
| = ks1 E0
|0; |s1 |0 gh
( (1) + (2) )
+ ks2 E0
E0 |0; |s2 |0 gh

(1) 2
+ ks2 E0
|0; |s2 |0 gh .
(1)
(5.15)
to be zero on this state yields

Requiring Q

ks1 |s1 = k1 0; 2 ,

2
(2)
ks2 |s2 = 2 k1 E 0; 2 + k2 0; 2 2 ,

1
(2) 2
0; 2 1 k2 E (2) 0; 2 2
k1 E
ks3 |s3 = 2 2
0
( + 1)
2 + 2

+ k3 0; 2 4 .
(5.16)
Thus, the states in the cohomology are

( (1) + (2) ) 2
2 ( (1) + (2) ) (1) (2)
|1 = k1 E0
2 E0
E0
E

(1) 2 (2) 2
1
E
+ 2 2
|0; 0; 2 |0 gh ,
E0
( + 1)

1
( (1) + (2) ) (1)
(2)
|2 = k2 E0
E0
2
|0; 0; 2 2 |0 gh ,
E0
+2

(1) 2
|3 = k3 E0
|0; 0; 2 4 |0 gh .
(5.17)
One can easily see that the first state is related to

( (1) + (2) ) 2
2
(2) ( (1) + (2) ) (1)
|1 = k1 E0
+ 2 E E0
E0

(2) 2 (1) 2
1
E0
+ 2 2
|0; 0; 2 |0 gh ,
E
( + 1)
(5.18)
by a BRST trivial term. The same is true for the second state. It can be mapped to the state

1
( (1) + (2) ) (1)
(2)

E
|2 = k2 E0
(5.19)
E0
+ 2
|0; 0; 2 2 |0 gh
+2 0
by a trivial term. This shows that our conjecture is correct for these states as |1 , |2 and |3
are highest weight states of A1 .
We now turn to the study of the BRST charge Q2 , corresponding to the conformal symmetry. The analysis of this cohomology is more or less standard combining the techniques of [24]
and [27]. We will for completeness outline the analysis. First one considers the relative space
which is annihilated by P0 and Ltot

0 and the corresponding BRST charge Q2 . We then proceed
490
by constructing a basis for the states in Eq. (5.10)

h )
, m |gh ,
Lm1 . . . Lmp Hn1 . . . Hnq Ps1 . . . Psr l (g,
(5.20)
h ) , m is a highest weight state of the Virasoro

where Hn generates the center of h and |l (g,
h ) and m refer to the eigenvalues of the
algebra and the u k (1)-algebra generated by Hn . l (g,
corresponding zero-modes. |gh is a ghost state with arbitrary -ghost excitations. A proof that
this is a basis can be found in [27]. Let us introduce a grading of operators
n > 0,
grad(Ln ) = 1,
n > 0,
grad(Pn ) = 1,
grad(n ) = 1, n > 0.
(5.21)
All other operators have grad zero. States will be graded correspondingly by first defining the
ground states in each sector to have zero grad and then applying Eq. (5.21). The BRST charge
= d V + d V , where
then splits into two terms, Q
0
1

Ln n
d0V =
(5.22)
n>0
V refer to the grading.
and indices of the operators d0V and d1
For p > 0 in Eq. (5.20) one defines a homotopy operator,
0V |p, r =
p
1
Lm1 . . . L mi . . . Lmp Hn1 . . . Hnq
p+r
i=1
h )
Pmi Ps1 . . . Psr l (g,
, m |gh ,
(5.23)
where capped operators are again omitted. Using this homotopy operator one can see that all
BRST invariant states with a highest grad N > 0 and p > 0 are BRST-trivial
|; p, r = Q 2 0V |; p, r + O(N 1).
(5.24)
Thus, we are left with the case p = 0. Applying the BRST operator we immediately find that in
order to get zero we must take r = 0. This implies in turn that we cannot have any -excitations.
Therefore, the only states one needs to study in more detail are linear combinations of states of
the form
h )
, m |0 ,P .
|q = Hn1 . . . Hnq l (g,
(5.25)
The BRST invariance of linear combinations of these states implies
Ln | = 0,
(L0 1)| = 0,
(5.26)
which are the standard Virasoro string conditions for physical states. For m = 0 it follows from
[13] that one does not have any non-trivial on-shell states with Hn excitations. For m = 0
h ) , m = 0 |0
one has a state of the form H1 |l (g,
,P which is physical. From the on-shell
h ) = 0 which is only possible in the free field limit, k . In

condition one will get that l (g,
this limit it is well known that this state decouples from the space of physical states.
Combining the result of the analysis of the two BRST charges we have the proposition.
491
h
g
Proposition 1. Let | H H Hghost and satisfy Eq. (3.13), such that it is a non-trivial
state in the relative cohomology. Then | can be taken to be of the form

h ),1
(g,
| = J
h ),r1
(g,
. . . J
|0; R |s |0 ghost
(5.27)
and satisfy the conditions

(Ln n,0 )| = 0, n 0,

En + E n | = 0, { c , n > 0} +
c , n=0 ,

i
Hn + H ni | = 0, {i = 2, . . . , rg , n > 0},
(5.30)
(Hn )| = 0,
(5.31)
n > 0.
(5.28)
(5.29)
Let us comment on the case that we in place of the two separate BRST conditions apply the
full BRST charge Q . As was proven in the previous section, the state spaces that we will get
are isomorphic. One may independently show that the non-trivial physical states defined by Q
are equivalent to the ones above using the analysis presented in this section. The argument is
quite straightforward. The gradation used in analyzing Q1 may just as well be applied to Q
since the leading term is the same in both cases. The analysis leads, therefore, to an identical result. The same reasoning can be applied to the second grading that was used to analyze
the conformal BRST charge. Hence, the result is identical to the one above, which was our
claim.
6. Concluding remarks
Our proof of unitarity using the generalized branching functions and the coset signature functions of the state space obscures to a large extent the explicit form of the physical states. This is
why we also performed an explicit analysis of the solutions. We were not, however, completely
successful in this respect as we could not fix the form of the physical states to the extent we
expected. We were only able to show that the non-trivial states have no ghost excitations and, in
addition, satisfy a highest weight condition w.r.t. the combined current modes of h C h C . The
explicit form of the states are given by Eq. (5.27) satisfying the conditions in Eqs. (5.28)(5.31)
and they are not manifestly BRST invariant. In addition, the equations involve the auxiliary h sector in a non-trivial fashion. Although this is not a problem as such, since this sector of states
is unitary, it is still unsatisfying.
The branching function given above should be combined with a corresponding piece of opposite chirality to produce a partition function. For consistent string theories one should be able
to construct modular invariant combinations of characters. We believe that just like for SL(2, R)
one has to introduce spectrally flowed sectors as was first done in [8]. We hope to come back to
this in the future.
The BRST-approach to the coset model gives, as we have stressed several times, a consistent truncation of highest weights of h C allowing only antidominant integral highest weights
to appear in the g C -module. This solves the problem of non-antidominant highest weights appearing in the conventional coset formulation. It is the fact that the auxiliary sector only has
dominant integral highest weights which enforces this restriction in the g C -module. We believe
492
that the rle the auxiliary sector plays in this connection may be crucial also when one introduces interactions for these string models. If the interaction picture is consistent, one must find
that a tensor product between two allowed representations yields another allowed representation. One indication that this could be true is that for the auxiliary h -sector tensor products
between integrable representations yield integrable representations [28]. Then BRST invariance
would enforce that only antidominant integral weights can occur for the g -sector, ensuring consistency.
We have also proven that the G/H WZNW model, where H is the maximal compact subgroup
of G, is a unitary conformal field theory. Showing that the G/H models are consistent is in
itself important and our result opens up new possibilities, as this large class of models has not
been studied previously. The extra U (1) which is divided out represents time in our string
formulation. It is perhaps not surprising that if the string theory is unitary then the conformal field
theory constructed by dividing out the time-like excitations is also unitary. In fact, the analysis of
Section 4 shows that the physical subspace of the string theory is a subspace of the G/H coset
space.
The demonstration of unitarity of G/H models is most likely an important result for another
reason. As mentioned in the introduction, we believe that this can be used to prove the unitarity of
bosonic strings on AdS spaces represented by the cosets SO(p, 2)/SO(p, 1). To prove unitarity
for these models one cannot directly proceed as we have done here. The reason is that the signature functions are not known for these cases. It is not possible to proceed as in Section 5 either.
In order to prove unitarity one needs a basis of states such that the non-unitarity is manifested in
a simple fashion. For the case we have treated here, it is manifested by the simple U (1) factor.
Still, we believe one can make progress towards a unitarity proof. We hope to report on this in
the near future.
Acknowledgements
We would like to thank I. Bars and J. Fuchs for stimulating discussions. S.H. is partially
supported by the Swedish Research Council under project No. 621-2005-3424.
Appendix A. Dynkin diagrams and non-compact positive roots of relevant real forms
Here the relevant Dynkin diagrams are presented (Figs. 114). In the diagrams (1) denotes
the simple non-compact root and the highest root. The diagrams which show the relations
between different non-compact positive roots are from Jakobsen [20].
Fig. 1. Dynkin diagram connected to su(p, q).
Fig. 2. Positive non-compact roots connected to su(p, q).
Fig. 3. Dynkin diagram connected to so(2p, 2).
Fig. 4. Positive non-compact roots connected to so(2p, 2).
Fig. 5. Dynkin diagram connected to so(2p 1, 2).
493
494
Fig. 6. Positive non-compact roots connected to so(2p 1, 2).
Fig. 7. Dynkin diagram connected to so (2p).
Fig. 8. Positive non-compact roots connected to so (2p).
Fig. 9. Dynkin diagram connected to sp(2p, R).
Fig. 10. Positive non-compact roots connected to sp(2p, R).
Fig. 11. Dynkin diagram connected to E6|14 .
Fig. 12. Positive non-compact roots connected to E6|14 .
495
496
Fig. 13. Dynkin diagram connected to E7|25 .
Fig. 14. Positive non-compact roots connected to E7|25 .
Appendix B. Tables of relevant conformal field theories

Below we give tables of relevant conformal field theories and the corresponding c-values
(Tables 16). cmax is the largest value, less than or equal to 26, which is possible from the allowed
values of k.
497
Table 1
kp(p 2 1)
su(p, 1)/su(p): c = kp(p+1)
k+p
k+p+1
p
cmax
kmax
cmin
2
3
4
5
6
7
8
9
10
11
12
26
21
23
26
24.8462
25.9804
25.6667
25.8272
25.9876
25.9826
25.9992
4
7
10
13
19
25
36
52
80
148
485
5
7
9
11
13
15
17
19
21
23
25
Table 2
2 +4p+3)
2 1)
3k
su(p, 2)/(su(p) su(2)): c = k(pk+p+2
k(p
k+p k+2
p
cmax
kmax
cmin
2
3
4
5
6
22
25.0545
25.9093
25.9052
25.9985
8
13
23
51
298
9
13
17
21
25
Table 3
2 +6p+8)
2 1)
8k
su(p, 3)/(su(p) su(3)): c = k(pk+p+3
k(p
k+p k+3
p
cmax
kmax
cmin
3
4
25.9722
25.9963
30
261
19
25
Table 4
k(p+1)(p+2)
kp(p1)
so(p, 2)/so(p)* : c = 2(k+2[p/2]+1)
2(k+2[p/2]1)
p
cmax
kmax
cmin
3
4
5
6
7
8
9
10
11
12
21.2500
22
24.7500
24.8000
25.6235
25.9322
25.9168
25.9377
25.9908
25.9977
5
8
11
16
22
16
46
31
135
448
7
9
11
13
15
17
19
21
23
25
* For p = 3 one uses that so(3, 2) sp(4, R) and that so(3) is isomorphic to su(2). For p = 4 one uses so(4, 2)
=
=
su(2, 2) and so(4)

= su(2) su(2).
498
Table 5
k(p2 1)
so (2p)/su(p): c = kp(2p1)
k+2p1 k+p
p
cmax
kmax
cmin
3
4
5
21.0000
24.8000
25.9358
7
16
58
7
13
21
Table 6
k(p2 1)
sp(2p, R)/su(p): c = kp(2p+1)
k+p+1 k+p
p
cmax
kmax
cmin
2
3
4
21.2500
25.5882
25.9241
5
10
36
7
13
21
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]
J. Balog, L. ORaifeartaigh, P. Forgacs, A. Wipf, Nucl. Phys. B 325 (1989) 225.

L.J. Dixon, M.E. Peskin, J.D. Lykken, Nucl. Phys. B 325 (1989) 329.
P.M.S. Petropoulos, Phys. Lett. B 236 (1990) 151.
S. Hwang, Nucl. Phys. B 354 (1991) 100.
M. Henningson, S. Hwang, Phys. Lett. B 258 (1991) 341.
I. Bars, D. Nemeschansky, Nucl. Phys. B 348 (1991) 89.
E.S. Fradkin, V.Y. Linetsky, Phys. Lett. B 277 (1992) 73.
M. Henningson, S. Hwang, P. Roberts, B. Sundborg, Phys. Lett. B 267 (1991) 350.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
J.M. Maldacena, H. Ooguri, J. Math. Phys. 42 (2001) 2929, hep-th/0001053.
J.M. Maldacena, H. Ooguri, Phys. Rev. D 65 (2002) 106006, hep-th/0111180.
J.M. Maldacena, H. Ooguri, J. Son, J. Math. Phys. 42 (2001) 2961, hep-th/0005183.
S. Hwang, Phys. Lett. B 435 (1998) 331, hep-th/9806049.
J. Helgason, Differential geometry, Lie Groups and Symmetric Spaces, Academic Press, 1978.
P. Goddard, A. Kent, D.I. Olive, Phys. Lett. B 152 (1985) 88.
D. Karabali, H.J. Schnitzer, Nucl. Phys. B 329 (1990) 649.
S. Hwang, H. Rhedin, Nucl. Phys. B 406 (1993) 165, hep-th/9305174.
I. Bars, K. Sfetsos, Phys. Lett. B 277 (1992) 269, hep-th/9111040.
J. Fuchs, C. Schweigert, Symmetries, Lie Algebras and Representations: A Graduate Course for Physicists, Cambridge Univ. Press, Cambridge, UK, 1997, 438 p.
H. Jakobsen, J. Funct. Anal. 52 (1983) 385.
N.N. Shapovalov, Funct. Anal. Appl. 6 (1972) 307.
V.G. Kac, D.A. Kazhdan, Adv. Math. 34 (1979) 97.
T. Enright, R. Howe, N. Wallach, Progr. Math. 40 (1983) 97143.
I.B. Frenkel, H. Garland, G.J. Zuckerman, Proc. Natl. Acad. Sci. 83 (1986) 8442.
V.G. Kac, Infinite Dimensional Lie Algebras, Cambridge Univ. Press, Cambridge, UK, 1990.
S. Hwang, H. Rhedin, Mod. Phys. Lett. A 10 (1995) 823, hep-th/9408087.
S. Hwang, Phys. Lett. B 276 (1992) 451, hep-th/9110039.
D. Gepner, E. Witten, Nucl. Phys. B 278 (1986) 493.

Sutherland models for complex reflection groups

N. Cramp a,b, , C.A.S. Young c
a International School for Advanced Studies, Via Beirut 2-4, 34014 Trieste, Italy
b Istituto Nazionale di Fisica Nucleare, Sezione di Trieste, Italy
c Department of Mathematical Sciences, University of Durham, South Road, Durham DH1 3LE, UK
Received 22 August 2007; received in revised form 12 November 2007; accepted 29 November 2007
Abstract
There are known to be integrable Sutherland models associated to every real root system, or, which
is almost equivalent, to every real reflection group. Real reflection groups are special cases of complex
reflection groups. In this paper we associate certain integrable Sutherland models to the classical family
of complex reflection groups. Internal degrees of freedom are introduced, defining dynamical spin chains,
and the freezing limit taken to obtain static chains of HaldaneShastry type. By considering the relation of
these models to the usual BCN case, we are led to systems with both real and complex reflection groups
as symmetries. We demonstrate their integrability by means of new Dunkl operators, associated to wreath
products of dihedral groups.
PACS: 02.30.Ik; 03.65.Fd; 75.10.Jm
Keywords: Integrable systems; Sutherland model; Complex reflection groups; Dunkl operators; Hecke algebra
1. Introduction
The Sutherland model [1] is an important and much-studied integrable quantum-mechanical
system. It describes N particles moving on a circle, whose pairwise interactions are determined
by a potential proportional to the inverse square of the chord-length separating the particles (as
in Fig. 1). The Hamiltonian, in the simplest case of identical spinless bosons, is
E-mail addresses: crampe@sissa.it (N. Cramp), charlesyoung@cantab.net (C.A.S. Young).

doi:10.1016/j.nuclphysb.2007.11.028
500
N. Cramp, C.A.S. Young / Nuclear Physics B 797 [PM] (2008) 499519
Fig. 1. Interactions between two particles ( and ) in the Sutherland models associated to the A (left) and D (right)
series of real reflection groups. The image of under reflection is drawn as .

1
1 2
+
.
2
1
2
2
xi
sin ( 2 (xi xj ))
N
H =
i=1
(1.1)
i=j
The model was first introduced in [2]. It, and the wider family of CalogeroSutherlandMoser
models [3] to which it belongs, have since appeared in areas physics apparently far removed from
the original condensed-matter context: see for example [4]. Operator methods were used to solve
the system in [5,6]; the Yangian symmetry of the model was derived in [7,8]. For recent reviews
and references to the extensive literature see [911].
The Hamiltonian
(1.1) is closely related to the Coxeter group AN1 , because the potential can

be written as sin2 ( 12 (x )), where = { i j : i = j } is the root system of AN1 .
Similar integrable models exist also for all other finite Coxeter groups [1217]. The classical
families BCN and DN describe systems with boundaries, via a kind of method of images: in the
case of DN , whose roots are i j , one has

N

1
1 2
1
+
+
,
H =
2
xi2
sin2 ( 12 (xi xj )) sin2 ( 12 (xi + xj ))
i=j
i=1
(1.2)
and the second term describes the interaction of particle i with the image of particle j in the
boundary (see Fig. 1). In the B and C cases, the extra roots i give an interaction between
the particles and the boundary.
These families, ABCD, of course exhaust the classical irreducible finite Coxeter groups. Finite Coxeter groups are finite real reflection groups [18]: that is, subgroups of the orthogonal
group generated by a finite number of elements s O(N) such that
s 2 = 1,
s has eigenvalue + 1 with multiplicity N 1.
(1.3)
But it is possible to weaken the first of these requirements and consider subgroups of U (N )
generated by finitely many s U (N ) obeying
s n(s) = 1,
s has eigenvalue + 1 with multiplicity N 1
(1.4)
for some n(s) N. By doing so one obtains complex reflection groups. The irreducible finite
complex reflection groups were classified in [19]. There is one classical three-parameter family
G(pr, p, N), p, r, N N, which includes the four classical families of real reflection groups as
special cases, and then 34 exceptional cases.
501
Fig. 2. Interactions between particles and for the Sutherland models associated to the classical complex reflection
group G(3, 1, N ) (left) and a wreath product of a dihedral group (right). The images of are drawn as .
In this paper our main goal is to construct Sutherland models for classical complex reflection
groups. At first sight it is perhaps not clear that one should expect this to work, because complex
reflection groups lack a great deal of the usual structure that comes with real reflection groups.
The notions of root system, length function and Coxeter graph are either absent or, at best, less
natural in the complex case [20]and the definition of the BCD Sutherland models sketched
above appears to rely explicitly on the root system data. Nevertheless, it turns out that there
do exist integrable models of Sutherland type associated to complex reflection groups in a very
natural fashion. The basic idea is sketched in Fig. 2each particle has a number of images, but
now these images are generated by rotations.
The plan of this paper is as follows. In Section 2 we begin with some algebraic preliminaries
on complex reflection groups, and then introduce a key tool, Dunkl operators [21], from which
we construct integrable Hamiltonians of Sutherland type. These models turn out to be members
of a class of CalogeroSutherland models first introduced and solved in [22,23]. In Section 4 we
introduce models with internal spin degrees of freedom, and in Section 5 static or frozen
chains in which the spins are in fact the only degrees of freedom.
In Section 6 we turn to models in which the set of images of each particle is generated by
a dihedral group, as illustrated in Fig. 2. As we shall see, these models possess both a complex
reflection group and a real reflection group as symmetries, embedded within a larger group which
will turn out to be a wreath product of a dihedral group. An important part of our construction
will be the introduction of new Dunkl operators, associated to such wreath products. We then go
on to introduce spin degrees of freedom and static chains with dihedral symmetry.
We conclude by noting some open questionsprimarily of solution and Hamiltonian
reductionconcerning the new models of this paper, and some broader reasons for investigating
integrable systems with complex reflection groups as symmetries.
2. Complex reflection algebras
2.1. Classical complex reflection groups
The complex reflection group G(m, 1, N) is generated by {a, e1 , e2 , . . . , eN1 }, subject to the
relations
ei2 = 1,
ei ei+1 ei = ei+1 ei ei+1 ,
a m = 1, ae1 ae1 = e1 ae1 a,
ei ej = e j ei
(|i j | > 2),
aej = ej a
(j > 1).
(2.1)
502
The ei generate a copy of the permutation group SN on N objects, a generates a copy of Zm =

Z/mZ, and the full group is a semidirect product G(m, 1, N) = (Zm )N SN . This structure
is sometimes referred to as the wreath product of Zm with SN , denoted Zm SN . It will be
convenient to write
Pij = Pj i = ei ei+1 ej 1 ei+1 ei
(2.2)
(i < j )
for the transposition i j (in particular ei = Pii+1 ) and

Q1 = a,
(2.3)
Qi = Pi1 Q1 Pi1
(i > 1).
(2.4)
In terms of these elements the defining relations imply, and can be recovered from,
Pij2 = 1, Pij Pj k = Pik Pij = Pj k Pik ,
Qim
= 1,
Pij Pkl = Pkl Pij ,
Pij Qi = Qj Pij ,
Pij Qk = Qk Pij ,
Q i Qj = Q j Qi
(i, j, k, l all distinct).
(2.5)
For any divisor p of m, the complex reflection group G(m, p, N) is the subgroup of
G(m, 1, N) generated by
p 1

a , a e1 a, e1 , e2 , . . . , eN1 .
(2.6)
The classical real reflection groups occur as the special cases AN1 = SN = G(1, 1, N), BCN =
G(2, 1, N) and DN = G(2, 2, N).1
2.2. Extended degenerate affine Hecke algebras
We will overload notation slightly by using G(m, 1, N) also to refer to the group algebra of
G(m, 1, N) over C. Let us define H (m, 1, N), C, to be the algebra generated by
{a, d, e1 , e2 , . . . , eN1 },
(2.7)
obeying (2.1) and the further relations

ej d = dej (j > 1),
de1 ae1 = e1 ae1 d,

s
s
de1 de1 + d
a e1 a = e1 de1 d +
a s e1 a s d.
ad = da,
sZm
(2.8)
(2.9)
sZm
In addition to the Qi and Pij of (2.2)(2.4), it is also useful to introduce d1 , . . . , dN , defined

recursively by
d1 = d,
di+1 = Pii+1 di Pii+1 +
(2.10)
Qis Pii+1 Qis
(i = 1, . . . , N 1).
(2.11)
sZm
It follows from (2.8)(2.9) that

[di , dj ] = 0,
[di , Qj ] = 0.
(2.12)
1 But note that, in what follows, the models we construct do not reduce, in the BC and D cases, to the standard
Sutherland Hamiltonians, la (1.2), for these groups. We return to this point in Section 6.
503
H (m, 1, N) can be regarded as an affinization of G(m, 1, N), with d in a very loose sense
a lowest root. Indeed, when m = 1 the sums above collapse and one recovers the relations
Pii+1 di = di+1 Pii+1 + of the degenerate affine Hecke algebra, first introduced in [25].
We may define also H (pr, p, N ), an affinization of G(pr, p, N), to be the subalgebra of
H (pr, 1, N) generated by {a p , d, a 1 e1 a, e1 , e2 , . . . , eN1 }. Note that the relation (2.9) does
not conflict with closure, because G(pr, p, N) does contain all the elements

s1 1 s1
a s e1 a s = a 1 e1 a
(2.13)
e1 a e1 a
.
Extended degenerate affine Hecke algebra associated with the BCN reflection groups appeared
previously in [33,37]. They differ from definition of H (2, 1, N) here though: in particular, the
relation [di , Qj ] = 0 does not hold there.
3. Dunkl operators and Hamiltonians
3.1. Realization of H (m, 1, N)
The next stage is to realize these abstract algebraic relations in a concrete physical model.
Consider a quantum-mechanical system of N particles on the unit circle. Let qi = exp(ixi ) be
the position operator of the ith particle and write the position-space wavefunction as
(3.1)
(q1 , q2 , . . . , qN ).
Let Pij be the operator which transposes the positions of particles i and j ,
Pij (. . . , qi , . . . , qj , . . .) = (. . . , qj , . . . , qi , . . .),
(3.2)
and Qi the operator which rotates particle i through ( m1 )th of a revolution,

2i
Qi (. . . , qi , . . .) = (. . . , qi , . . .), where = exp
.
m
(3.3)
It is easy to see that these Pij and Qi satisfy the defining relations (2.5) of G(m, 1, N). We also
have that
Qi qi = qi Qi ,
= 1
Qi ,
qi
qi
Pij
=
Pij .
qi
qj
Qi
Pij qi = qj Pij ,
(3.4)
(3.5)
The crucial step is the introduction of differential operators, called Dunkl operators [5,26,27],
that realize the algebraic relations of the di . The problem of finding such operators for complex
reflection groups was solved in [21]. Following that paper, with minor modifications that will
allow us to obtain a slightly more elegant Hamiltonian, we define

qi
di = qi
(3.6)
+
Qis Pij Qis
Qis Pij Qis
s
qi
qi qj
j =i sZm
j >i sZm
j <i sZm
j >i sZm

s qj
qi
= qi
+
Qis Pij Qis +
Q s Pij Qis .
s
qi
qi qj
qi s qj i
(3.7)
504
Theorem 3.1. These provide a realisation of H (m, 1, N).

This is essentially theorem (3.8) of [21], and the same strategy of proof works here. But when
we come to introduce new Dunkl operators for wreath products of dihedral groups, in Section 6,
it will be useful to have noted the following alternative
Proof. It is easy to verify that (2.11) holds: were it not for the final term on the right of (3.6),
the di would obey Pij di Pij = dj . The final term involves an ordering of the particles and is
responsible for the extra piece in (2.11).
It remains to show that d1 = d obeys (2.8) and (2.9). The first of these is straightforward to
check by direct computation.2 The second is nothing but the statement that the Dunkl operators
commute:
[d1 , d2 ] = 0,
(3.8)
which is really the key property. To prove it, first recall the Dunkl operators of the AN1 case:
qi
+ m
Pij m
Pij ,
Zi = qi
(3.9)
qi
q i qj
j =i
j >i
where we have chosen the coupling to be m. Observe that then

1 s
Qi Zi Qis .
di =
m
(3.10)
sZm
This motivates the definition of a family of projectors: we write

Ad(Q)(X) = Q 1 XQ
(3.11)
and define
ir =
1 sr s
Ad Qi .
m
(3.12)
sZm
These obey
id =
r,
r t = r,t r .
(3.13)
rZm
Also, for any A and B,

Q r A t B = r r A Q t B = r+t r A t B Q,
so
(3.14)
r+t ( r A)( t B) = ( r A)( t B).

0 AB =
r,tZm
It follows that

r A r B .
rA tB =

0
(3.15)
rZm
Armed with these facts we argue as follows. Given the result [5,26,27] that
[Zi , Zj ] = 0
(3.16)
2 We sketch the arguments, for the more involved case of dihedral groups to be considered in Section 6, in Appendix A.
we have in particular that

ir jt Zi , jt ir Zj .
0 = i0 j0 [Zi , Zj ] =
505
(3.17)
r,tZm
But one may compute, for all i = j ,
qi
ir jt Zi = r,0 t,0 qi

qi
s
s
s
s
+
Q Pih Qi
Qi Pih Qi
q s qh i
h{i,j
/
} i
h>i,h=j
sZm

qi
s
s
j >i s
s
+ r+t,0
(3.18)
rs
Q
P
Q
Q
P
Q
ij i
ij i .
i
qi s qj i
t,0
rs

sZm
The only terms in (3.17) which can survive in view of the s here are

it jt Zi , jt it Zj .
0 = i0 j0 Zi , i0 j0 Zj +
(3.19)
tZm ,t=0
The second term (with i > j , without loss of generality) is

qj
qi
s
s
s
ts
s
ts
Q
P
Q
,
1
Q
P
Q
ij
j
i

i
j
j
qi s qj i
qj s qi
tZm ,t=0 sZm
s Zm

qi
qi
s
s
s
ts
s
=
ts
Q
P
Q
,
Q
P
Q
ij i
ij i ,

qi s qj i
qi s qj i

tZm ,t=0 sZm
s Zm
(3.20)
and this vanishes. (In the sum over t , the summands cancel in conjugate pairs, t against t ,
except for the possible t = 1 term which is zero by itself.) But now since i0 j0 Zi = di
by (3.18), we have that indeed
0 = [di , dj ],
completing the proof.
(3.21)
2
3.2. Hamiltonians
It follows from the discussion above that the quantities
I (k) =
N
dik
(3.22)
i=1
form a commuting set. The I (k) are algebraically independent for k = 1, 2, . . . , N , and these give
N commuting conserved quantities of the model with Hamiltonian

N

s qi qj
2
H = I (2) =
(3.23)
+ Qis Pij Qis
2
qi
s
2
qi
(qi qj )
i<j sZm
i=1
N

i=1
qi
qi
2

i=j sZm

s qi qj
+ Qis Pij Qis
(qi s qj )2
(3.24)
506
Fig. 3. Partial picture of the model for N = 2. The particles are represented by full circles whereas their images by
rotations of 2/m are represented by empty circles.
which is therefore, by construction, integrable. After the change of coordinates qj = exp(ixj )

the Hamiltonian takes the form
H =
N

2
1

+
2
1
2
4
xi
sin ( 2 (xi xj +
i=j sZ
i=1
m
2s
m ))

+ Qis Pij Qis .
(3.25)
One sees that each particle xi interacts with every other particle xj both directly and via
its images under rotations. A sketch of the case G(3, 1, 2) is shown in Fig. 3. Note that the
Hamiltonian is not local, because of the final term which exchanges (and moves) particles. To
find local Hamiltonians it is useful to introduce spins, as follows.
4. Particles with spin
We now generalize the models above to particles with internal spin degrees of freedom,
s Cn . Let us introduce a map Q U (n) of order m (Qm = 1) and write Qi for Q acting on
the spin s i of the ith particle. Let also Pij be the operator which exchanges the spins of the ith
and j th particles. Pij and Qi then obey the same defining relations of G(m, 1, N) as Pij , Qi
in (2.5). The two copies of G(m, 1, N) commute.
The introduction of spins typically makes the restriction to the case of identical particles
much richer and more interesting. For bosons (fermions) the wavefunction should now be
(anti)symmetric under exchange of positions and spinsthat is, under
Pij Pij .
(4.1)
These generate the group SN of exchange symmetries. In the original AN -series Sutherland
models, on wavefunctions with definite exchange statistics it is possible [8,28] systematically
to eliminate Pij in favour of Pij in the Hamiltonian and higher conserved quantities, and so
obtain a purely local model with spinspin interactions.
We would like to do something similar in the present case. Here the exchange-symmetry
group is contained in several larger groups of discrete symmetries (involving the Qi and Qi ).
It is natural to pick one of these as a group of generalized exchange symmetries and demand
507
definite behaviour of the wavefunction under it. There are a number of possibilities: one could
for example demand that a full copy of G(m, 1, N), generated by e.g. Pij Pij and Qi Qi , be
promoted in this sense. But to do so would be overly restrictive on physical wavefunctions;
instead, it will suffice to demand invariance under
s
Qis Pij Qis Qs
i Pij Qi
(4.2)
for all i = j and for all s = 0, 1, . . . , m 1. These generate a copy of G(m, m, N).
(We focus for definiteness on +1-eigenstates of (4.2), but the more general case with arbitrary
eigenvalues ps {1} for each value of sin particular, p0 = 1, giving fermionscan be
treated very similarly.)
Let be the projector onto such states. To write explicitly, let g
Pg and g
Pg be the
maps representing abstract elements g G(m, m, N) on, respectively, spins and positions. Then
=
1
N !mN1
Pg Pg
(4.3)
gG(m,m,N )
and we consider wavefunctions such that

= .
(4.4)
(k)
Define Ispin to be the operator obtained by first moving all the Pij , Qi in I (n) , as defined in
(3.22), to the right of all positions xi and derivatives
Qi
Qi .3 It follows from the property

s
s
Qi Pij Qis = Qs
i Pij Qi ,
xi ,
and then replacing them Pij

Pij ,
(4.6)
of that
(k)
Ispin = I (k)
(4.7)
so that these operators agree on wavefunctions obeying (4.4).

Note next the following properties of the I (k) :

(k)

(k)
I , Qi = 0
I , Pij = 0,
(4.8)

k
which follow from the definition I (k) = N
i=1 di and the algebra (2.11)(2.12) of the Dunkl
operators. It is also trivially the case that

(k)

(k)
I , Qi = 0.
I , Pij = 0,
(4.9)
Consequently, for any monomial M({I (k) }) in the I (k) , M({I (k) }) obeys the same relations (4.6) as itself,
s

(k)
s
Qi Pij Qis M I (k) = Qs
(4.10)
,
i Pij Qi M I
3 The replacement map could also be defined, following [8], as the projection
G(m, 1, N )PP ,QQ1 G(m, 1, N )P ,Q G(m, 1, N )P ,Q ;
Then indeed Pij = (Pij Pij )Pij = Pij and Qi = (Qi Q1

i )Qi = Qi .
AB
B.
(4.5)
508
and thus

(k)
Ispin M I (p) = I (k) M I (p) .
(4.11)
(k)
Given now any string of Ispin s, not a priori assumed to commute, repeated use of this fact allows
(k)
one to replace each Ispin by I (k) , working from the inside out:

(k) ( )
(k)
Ispin Ispin = Ispin I ( ) = I (k) I ( ) = .
(4.12)
Having done so, the result
(k) ( )
= 0,
I ,I
(4.13)
may be used to reorder the I (k) at will, and the above procedure then reversed to return I (k)
(k)
Ispin . Thus, in particular, we have that the N independent evolution operators
Uk (t) = e
(k)
itIspin
k = 1, . . . , N
(4.14)
commute amongst themselves when acting on physical wavefunctions:

Uk (t)U (t ) = U (t )Uk (t).
(4.15)
Therefore the model described by the Hamiltonian

(2)
Hspin = Ispin
N

2
1

=
+
2
2 1
4
x
sin ( 2 (xi xj +
i
i=j sZ
i=1
m
2s
m ))

s
+ Qs
(4.16)
i Pij Qi
is integrable.
5. Static spin chain
In this section, we find static integrable spin models from the new integrable models introduced above. Indeed, it is well-known that from the AN Sutherland model it is possible to find
static spin chains [29,30], called usually HaldaneShastry models [31,32]. Using similar methods, we will prove that the following Hamiltonian

s qi qj
Q s Pij Qis
H =
(5.1)
(qi s qj )2 i
i=j sZm
is integrable for some particular values of the positions qi . First, we introduce the following
operators

s qj
qi
s
s
Q
P
Q
+
Q s Pij Qis .
di =
(5.2)
ij
i
qi s qj i
qi s qj i
j <i sZm
Note that
di = qi q i
j >i sZm
+ di . Since the relation [dj , dk ] = 0 is valid for any , we deduce that
[dj , dk ] = 0.
(5.3)
Similarly, from the relation [H, di ] = 0, it follows that, for any i,

s qj q
.
[H , di ] = qi pi ,
(ql s qj )2
j = sZm
(5.4)
509
Fig. 4. Position of the spins on a circle of diameter 1 for N = 5 and m = 3.
The commutators [H , di ] therefore vanish if and only if

qi qj (qi + s qj )
s
= 0, i = 1, 2, . . . , N,
(qi s qj )3
(5.5)
j =i sZm
and these conditions are fulfilled if

2ik
.
qk = exp
mN
(5.6)
Then, the Hamiltonian (5.1) with the particular values of the positions given by (5.6) is integrable
and may be written as follows
1
1
H =
(5.7)
Qks Pk Qks .
2
4
sin
(
(k
N
s))
mN
k= sZ
m
By the same procedure as in Section 4, we can obtain an integrable Hamiltonian acting on spins
(see Fig. 4)
1
1
s
H spin =
(5.8)
Qs
k Pk Qk .
2
4
sin
(
(k
N
s))
mN
k= sZ
m
6. Models with dihedral symmetry

In the previous sections we introduced Sutherland models based on the complex reflection
group G(m, 1, N). They reduce to the original Sutherland models, as in (1.1), in the special case
G(1, 1, N) = AN1 , but for other values of m they are new. In particular, although there is an
isomorphism of groups
G(2, 1, N)
= (Z2 )N SN ,
= BCN
(6.1)
as we noted above, our models certainly do not coincide with the usual BCN Sutherland models
in this case, because the Z2 generator is realized in different ways. In the G(m, 1, N) models,
recall,
Qi (. . . , qi , . . .) = (. . . , qi , . . .),
(6.2)
510
where = e2i/m , whereas in the BCDN case the action of the Z2 generator is [30],

Ki (. . . , qi , . . .) = . . . , qi1 , . . . .
(6.3)
In this section we show that it is possible to include both types of symmetry, rotation and reflection.
6.1. Dunkl operators for wreath products of dihedral groups
To take into account the new operators Ki , we must find the group W (m, N) generated by the
Qi , Ki and Pij , which will contain as subgroups both DN and G(m, 1, N). First note that Qi
and Ki satisfy, for each i,
Qim = 1,
Ki 2 = 1 and Ki Qi = (Qi )1 Ki ,
(6.4)
which are the defining relations of the dihedral group of order m, denoted Dihm . We deduce that
the group W (m, N) must be the wreath product
W (m, N) = Dihm SN = (Dihm )N SN .
(6.5)
A minimal set of generators for W (m, N) is

{a, b, e1 , e2 , . . . , eN1 },
(6.6)
obeying the relations

ei2 = 1,
ei ei+1 ei = ei+1 ei ei+1 ,
ei ej = e j ei
(|i j | > 2),
a m = 1,
ae1 ae1 = e1 ae1 a,
aej = ej a
(j > 1),
be1 be1 = e1 be1 b,
bej = ej b
(j > 1),
ba
= a 1 b,
b2
= 1,
(6.7)
and in terms of these Pij and Qi are again defined as in (2.2) and (2.4) while
K1 = b,
(6.8)
Ki = Pi1 K1 Pi1
(6.9)
(i > 1).
To construct integrable models, we must extend this algebra as explained in the previous
sections. Define H, (W (m, N )), , C, to be the algebra generated by {D, b, a, e1 , e2 , . . . ,
eN 1 }, obeying (6.7) and the further relations

a 2s ,
aD = Da,
bD = Db +
(6.10)
sZm

a s e1 a s e1 be1 = e1 be1 D +
a s e1 a s ,
D+
sZm
De1 ae1 = e1 ae1 D,
sZm

s
s
s
s
D e1 De1 +
a e1 a
a e1 a
= e1 De1 +
D,
sZm
(6.11)
ej D = Dej
(j > 1).
sZm
(6.12)
Defining as before
D1 = D,
Di+1 = Pii+1 Di Pii+1 +
Qis Pii+1 Qis
(i = 1, . . . , N 1)
511
(6.13)
sZm
it follows from (6.10)(6.12) that

[Di , Dj ] = 0,
[Di , Qj ] = 0.
(6.14)
The hard step, just in the case of G(m, 1, N), is to find a concrete realization of the Di satisfying these abstract relations.
Theorem 6.1. For any C, the differential operators
D i = qi

qi
qi
s
s
s
s
+
Q
P
Q
+
K
Q
P
Q
K
ij
i
ij
i
i
i
i
qi
qi s qj i
qi s qj1
j =i sZ
m
s qi s qi
j >i sZm

sZm
s q
Q 2s Ki
1 i
Qis Pij Qis
(6.15)
obey (6.10)(6.12) and (6.13).

We have verified this by direct computation, which is conceptually straightforward though
somewhat laborious. It is also possible to adapt the proof we used in the G(m, 1, N) case above,
as follows.
Proof. Some details of the verification of (6.10) are given in an Appendix A, but as in the
G(m, 1, N) case the important and difficult step is to show that
[Di , Dj ] = 0.
(6.16)
The Dunkl operators of the BCN case [33] may be written

Yi = qi

qi
qi
+ m
Pij +
K
P
K
i ij i
qi
q i qj
qi qj1
j =i
qi
qi q i
j >i
K m
1 i
Pij .
(6.17)
It follows from the result

[Yi , Yj ] = 0
(6.18)
that, with the projectors ir as defined in (3.12),

0 = i0 j0 [Yi , Yj ] =

ir jt Yi , ir jt Yj .
r,tZm
Now one can compute, for i = j ,
(6.19)
512
qi
ir jt Yi = r,0 t,0 qi
+
t,0
rs

h{i,j
/
}
sZm

qi
s
s
s
s
Q Pih Qi
Qi Pih Qi
qi s qh i
h>i,h=j
qi
Ki Qis Pih Qis Ki

s q 1
q
h
h{i,j
/
} i
sZm

s
qi
+
Q 2s Ki
1
s
s
qi q i

qi
+ r+t,0
rs
Q s Pij Qis
qi s qj i
sZm

qi
s
s
j >i s
s
+
K
Q
P
Q
K
Q
P
Q
i i
ij i i
ij i .
i
qi s qj1
+ t,0
rs
The s mean that the only terms that can possibly survive in (6.19) are

it jt Yi , it jt Yj .
0 = i0 j0 Yi , i0 j0 Yj +
(6.20)
(6.21)
tZm ,t=0
As before, the second term vanishes on closer inspection, and since i0 j0 Yi = Di by (6.20), we
have established (6.16) as required. 2
6.2. Integrable models
Setting = ( + )/2 and = ( )/2, we may rewrite the Dunkl operators (6.15) as
D = d +

m1
k= s=0
m1

s=0
s q qk
K Q s P k Q s K
s q qk 1

s q
s q
+
Q 2s K .
s q + 1 s q 1
(6.22)
This equivalent form is useful for computing Hamiltonians. As

in the previous sections,
explained
2 is integrable because it is
we know that the model described by the Hamiltonian H = N
D
=1
one of a set of N independent mutually-commuting conserved quantities, namely
J (k) =
N
D k .
(6.23)
=1
The explicit form of the Hamiltonian depends on the parity of m: for m odd,

s q qk
Hodd = H
+ K Q s P k Q s K
( s q qk 1)2
k= sZm

s q

s q
2s
2s
+
+ Q K
+ Q K
(1 + s q )2
(1 s q )2
sZm
where H is given by (3.23)while for m even
(6.24)
Heven = H

k= sZm
sZm
513

s q qk
+ K Q s P k Q s K
s
2
( q qk 1)

s q
+ Q 2s K .
s
2
(1 q )
(6.25)
In the case = (i.e. = 0), the boundary term in the Hamiltonian, for m odd, can be simplified
and becomes

s q
(6.26)
+ Q s K .
s
2
(1 q )
sZ
2m
After the change of coordinates q = exp(ix ), the Hamiltonians are

1

Hodd = H +
+ K Q s P k Q s K
1
2s
2
4
sin ( 2 (x + xk + m ))
k= sZm

/4
+
+ Q 2s K
1
2s
2
cos ( 2 (x + m ))
sZm

/4
2s
+ 2 1
K
+
Q
,

sin ( 2 (x + 2s
m ))

1

Heven = H +
+ K Q s P k Q s K
1
2s
2
4
sin ( 2 (x + xk + m ))
k= sZ
(6.27)
1

+
1
2
4
sin ( 2 (x +
sZ
m

+ Q 2s K ,

2s
m ))
(6.28)
where now H is given by (3.25).

6.3. Models on spins
As explained in Section 4, it is possible to construct models acting on spins using the suitable projectors on the wavefunctions. In addition to the map Q introduced at the beginning of
Section 4, we introduce now K U (n) such that
K 2 = 1,
KQ = Q1 K.
Such matrices certainly exist: for example

with ai = an+1i
Q = diag a1 , . . . , an
(6.29)
and
K = antidiag(1, . . . , 1).
(6.30)
In addition to the conditions (4.2) on the wavefunctions, we demand that the physical wave
functions be invariant under
Qi2s Ki Q2s
i Ki ,
i, s.
The explicit form of the projector is the product b where is defined by (4.3) and

1 2s 2s
b =
Q
Q
j
j (1 + Kj Kj ).
(2s)N
j
sZm
(6.31)
(6.32)
514
At this point, a supplementary difficulty appears in comparison to the previous case (the
same problem appears in the usual BCN case in comparison to the AN case) because we get
[J (k) , Qi ] = 0 and [J (k) , Pij ] = 0 but
(k)

J , Ki = 0.
(6.33)
Fortunately, we can show that this commutator vanishes when k is even and, in particular, for
k = 2 which corresponds to the Hamiltonian. Up to this restriction, we can use the same procedure to Section 4 with the projector b and deduce that the dynamical spin model described
by the Hamiltonian, for m even,

1

s
even
s
Hspin
= Hspin +
Q
P
Q
K
+
K

k

4
sin2 ( 12 (x + xk + 2s
m ))
k= sZ
m
1

+
2 1
4
sin ( 2 (x +
sZ
m

2s
m ))
+ Q2s
K
(6.34)
or, for m odd,

1

s
+ K Qs
P k Q K
1
2s
2
4
sin ( 2 (x + xk + m ))
k= sZm

/4
+
+ Q2s
K
1
2s
2
cos ( 2 (x + m ))
sZm

/4
2s
+ 2 1
K
+
Q

sin ( 2 (x + 2s
m ))
odd
= Hspin +
Hspin
(6.35)
is integrable.
6.4. Spin chain
As explained in Section 5, it is possible find an integrable static spin chain from a dynamical
one. Using this procedure,4 we can prove that the following Hamiltonian, for m odd,

s q qk
s q qk
s
s
s
s
H odd =
Q
P
Q
+
K
Q
P
Q
K
k

k

(qk s q )2
( s q qk 1)2
k= sZm

s q
s q
+
(6.36)
Q 2s K
(1 s q )2 (1 + s q )2
sZm
is integrable if, for = 1, . . . , N ,

qj (q + s qj ) qj ( s q qj + 1)
s 2
+
(q s qj )3
( s q qj 1)3
j =
sZm

s
s
2 1 q
2 1 + q
+
= 0.
(1 + s q )3
(1 s q )3
Similarly, for m even, we prove that the Hamiltonian
4 It is convenient to rescale the coupling constants of the boundary: and .
(6.37)
H even =

k= sZm
s q qk
s q qk
s
s
Q
P
Q
+
K Q s P k Q s K
k

(qk s q )2
( s q qk 1)2

sZm
s q
Q 2s K
(1 s q )2
(6.38)
is integrable if, for = 1, . . . , N ,

s
qj (q + s qj ) qj ( s q qj + 1)
s
2 1 + q
2
+
= 0.
(q s qj )3
( s q qj 1)3
(1 s q )3
sZm
515
(6.39)
j =
As discussed above, we can now replaced in the Hamiltonians H odd and H even the operators acting on the positions by the operators acting on spins, while preserved integrability. We finish this
section by discussing the different solutions of relations (6.37) and (6.39) in which the positions
are equidistant. The solutions depend on the value of the coupling constant and (or and ).
Let us define L to be the number of siteswhich may differ from N , the number of spinsand
let L = e2i/L . Different possible distributions of the coordinates qi , for m odd, are given in
Table 1.
The number in the column Figure corresponds to the labels of the figures below where the
particular case m = 3 is taken. In these figures, the black circles represent the positions of the
original spins whereas the white circles represent the images of these spins. The grey circles are
empty sites. Of course, we can recover the usual BCN cases studied in [30] when we put m = 1
Table 1
2
L
2N m
2N m + m
2(N + 1)m
1
4
qk
Figure
1
4
k 1
L 2
k 1
2
L
k
L
k
L
9
4
1
4
1
4
9
4
9
4
9
4
Fig. 5. Position of the sites for m = 3 and L = 2N m.
Fig. 6. Position of the sites for m = 3, L = 2N m + m and

k 12
qk = L
516
Fig. 7. Position of the sites for m = 3, L = 2N m + m and

k.
q k = L
Fig. 8. Position of the sites for m = 3 and L = 2(N + 1)m.
in the previous table. The case when m is even seems more complicated: we found no solution
for relation (6.39).
7. Conclusions and outlook
In this paper we considered two families of Sutherland models, in which each particle possesses a set of images determined by a cyclic or dihedral group. The former we were led to by the
desire to find models in which complex reflection groups act as symmetries; in the latter, the role
of the reflection group is played by a wreath product of a dihedral group. The Dunkl operators
were the key ingredient in demonstrating integrability. In the cyclic case these had been found
in [21], but in the dihedral case they have not appeared previously, to the authors knowledge.
We sought to emphasise the link between the models and complex reflection groups. In the
cyclic cases the models themselves are a special case of systems previously obtained by appropriate reduction of a matrix model [22] and of rational spin-Calogero models [23]; in the latter
case the equivalence may be seen by re-writing the sin2 potential as an infinite sum of inverse
squares. (Further generalizations of these models involving a twisted symmetry element were
found in [24].) In principle our models with dihedral symmetry could also be obtained by reductions of rational models involving parity in addition to translation symmetry, though this has not
been done explicitly.
There are a number of interesting open questions concerning these models. First, one should
be able to solve for the energy eigenstates exactly. This could be achieved by simultaneously
diagonalizing the Dunkl operators by means of (suitably generalized) Jack polynomialssee
e.g. [6]. One also strongly expects, looking at Fig. 2, that it should be possible to obtain all the
models here from the standard A-series Sutherland model via a suitable reduction procedure [9,
11,14], just as is true of the BCN case. This is usually related to folding of Dynkin diagrams (see
e.g. [16]); here we expect broader notion of folding will come into play, and the projectors used
in our proof of commutation the Dunkl operators (Theorem 3.1) seem rather suggestive.
We stress however that these models are of interest in their own right, regardless of their origin
via reduction. In particular they should possess some extended symmetry algebra, analogous to
the Yangian and reflection-algebra symmetries of, respectively, the A and BC Sutherland models,
but respecting the underlying complex reflection group. It is worth remarking here on intriguing
hints in the mathematics literature that, the lack of root systems and so on notwithstanding, cer-
517
tain complex reflection groups are actually closely analogous to real crystallographic ones (i.e.
Weyl groups) withvery loosely speakingthe role of Lie algebras being played by objects
called spetses [34]. These remain somewhat mysterious, and one can speculate that the machinery of integrable models (Hopf algebras, R-matrices and so on) might provide a helpful new
perspective, as it has in the past for Lie algebras and their representation theory.
Finally let us note a few more open questions. Do there exist Sutherland models for the exceptional complex reflection groups, perhaps via reduction as in [35]? The Sutherland model is the
trigonometric member of the CalogeroMoser family: can one generalize to elliptic potentials?
What models, presumably conformal field theories [36], are obtained in the limit of large N ?
Appendix A. Properties of Dunkl operators
Here we give some details of the argument that D1 = D obeys (6.10)(6.12). Consider the
relation in (6.10) involving b = K1 . For the terms at order , one finds first that

q1
q1
s
s
s
s
Q P1j Q1 +
K1 Q1 P1j Q1 K1
K1
q1 s qj 1
q1 s qj1
j =1 sZm

q1
q1
s
s
s
s
Q
P
Q
+
K
Q
P
Q
K
+
1j 1
1 1
1j 1 1 K1
q1 s qj 1
q1 s qj1
j =1 sZm

q11
q1
=
+
K1 Q1s P1j Q1s
1
sq
s q 1
q
j
1
j
1
j =1 sZm

1
q1
q1
+
+
Q1s P1j Q1s K1
1
1
q1 s qj
q1 s qj

=
K1 Q1s P1j Q1s + Q1s P1j Q1s K1
(A.1)
j =1 sZm
which then is precisely cancelled by the contribution from the other order- piece in (6.15). Note
of course that the algebra of Ki with qi , implicit the action (6.3) of Ki on wavefunctions, is
Ki qi = qi1 Ki and Ki q i = 1 Ki . Since also
qi
K1 q 1
= q1
Ki
q1
q1
(A.2)
(because qi1 = qi1 ) we have that K1 D1 = D1 K1 up to the terms involving and . It is

straightforward to verify that these give

Q12s
K1 D1 = D1 K1 +
(A.3)
sZm
as claimed. The other relation in (6.10), Q1 D1 = D1 Q1 is almost immediate.

Next consider (6.11): K2 = e1 be1 commutes term by term with the right-hand side of

a s e1 a s
D1 +
sZm

q1
q1
s
s
s
s
= q1
+
Q P1j Q1 +
K1 Q1 P1j Q1 K1
q1
q1 s qj 1
q1 s qj1
j =1 sZ
m
518

sZm
s qi
sq
s 1
i qi
Q12s K1 +

Qis Pij Qis
(A.4)
j >2 sZm
except when j = 2 in the sum: but there

q1
q1
s
s
s
s
K2
Q
P
Q
+
K
Q
P
Q
K
12 1
1 1
12 1 1
q1 s q2 1
q1 s q21
sZm

q1
q1
s
s
s
s
=
K2 Q1 P12 Q1 K2 +
K2 K1 Q1 P12 Q1 K1 K2 K2
q1 s q2
q1 s q21
sZm

q1
q1
s
s
s
s
=
K
Q
P
Q
K
+
Q
P
Q
(A.5)
K2
1 1 12 1
1
12 1
q1 s q2 1
q1 s q21
sZm
and, on renaming s
s, one sees that the two terms are merely exchanged. Thus indeed K2
commutes with D + sZm a s e1 a s . To show that Q2 D1 = D1 Q2 is straightforward.
References
[1] B. Sutherland, Exact results for a quantum many-body problem in one dimension II, Phys. Rev. A 5 (1972) 1372.
[2] B. Sutherland, Quantum many-body problem in one-dimension: Ground state, J. Math. Phys. 12 (1971) 246.
[3] F. Calogero, Solution of the three-body problem in one dimension, J. Math. Phys. 10 (1969) 2191;
F. Calogero, Ground state of a one-dimensional N -body problem, J. Math. Phys. 10 (1969) 2197;
F. Calogero, Solution of the one-dimensional N -body problem with quadratic and/or inversely quadratic pair potentials, J. Math. Phys. 12 (1971) 419;
J. Moser, Three integrable Hamiltonian systems connected with isospectral deformations, Adv. Math. 16 (1975)
197;
J. Moser, Integrable systems of non-linear evolution equations, in: J. Moser (Ed.), Dynamical Systems, Theory and
Applications, in: Lecture Notes in Physics, vol. 38, Springer-Verlag, 1975.
[4] A. Gorsky, N. Nekrasov, Hamiltonian systems of Calogero type and two-dimensional YangMills theory, Nucl.
Phys. B 414 (1994) 213, hep-th/9304047;
G.W. Gibbons, P.K. Townsend, Black holes and Calogero models, Phys. Lett. B 454 (1999) 187, hep-th/9812034;
E. DHoker, D.H. Phong, SeibergWitten theory and CalogeroMoser systems, Prog. Theor. Phys. Suppl. 135
(1999) 75, hep-th/9906027;
N. Dorey, An elliptic superpotential for softly broken N = 4 supersymmetric YangMills theory, JHEP 9907 (1999)
021, hep-th/9906011;
H.L. Verlinde, Superstrings on AdS(2) and superconformal matrix quantum mechanics, hep-th/0403024;
A. Agarwal, A.P. Polychronakos, BPS operators in N = 4 SYM: Calogero models and 2D fermions, JHEP 0608
(2006) 034, hep-th/0602049.
[5] L. Brink, T.H. Hansson, M.A. Vasiliev, Explicit solution to the N -body Calogero problem, Phys. Lett. B 286 (1992)
109.
[6] L. Lapointe, L. Vinet, Exact operator solution of the CalogeroSutherland model, Commun. Math. Phys. 178 (1996)
425, q-alg/9509003.
[7] F.D.M. Haldane, Z.N.C. Ha, J.C. Talstra, D. Bernard, V. Pasquier, Yangian symmetry of integrable quantum chains
with long range interactions and a new description of states in conformal field theory, Phys. Rev. Lett. 69 (1992)
2021.
[8] D. Bernard, M. Gaudin, F.D.M. Haldane, V. Pasquier, YangBaxter equation in spin chains with long range interactions, J. Phys. A 26 (1993) 5219, hep-th/9301084.
[9] F. Calogero, Classical Many Body Problems Amenable to Exact Treatments, Springer, 2001;
J.F. van Diejen, L. Vinet (Eds.), CalogeroMoserSutherland Models, Springer, 2000.
[10] P. Etingof, Lectures on CalogeroMoser Systems, math.QA/0606233.
[11] A.P. Polychronakos, Physics and mathematics of Calogero particles, J. Phys. A 39 (2006) 12793, hep-th/0607033.
[12] M.A. Olshanetsky, A.M. Perelomov, Quantum completely integrable systems connected with semi-simple Lie algebras, Lett. Math. Phys. 2 (1977) 7.
519
[13] D. Kazhdan, B. Kostant, S. Sternberg, Hamiltonian group actions and dynamical systems of Calogero type, Commun. Pure Appl. Math. 31 (1978) 481.
[14] M.A. Olshanetsky, A.M. Perelomov, Classical integrable finite dimensional systems related to Lie algebras, Phys.
Rep. 71 (1981) 313;
M.A. Olshanetsky, A.M. Perelomov, Quantum integrable systems related to Lie algebras, Phys. Rep. 94 (1983) 313.
[15] A.J. Bordner, E. Corrigan, R. Sasaki, Generalised CalogeroMoser models and universal Lax pair operators, Prog.
Theor. Phys. 102 (1999) 499, hep-th/9905011;
A.J. Bordner, E. Corrigan, R. Sasaki, CalogeroMoser models. I: A new formulation, Prog. Theor. Phys. 100 (1998)
1107, hep-th/9805106.
[16] A.J. Bordner, R. Sasaki, K. Takasaki, CalogeroMoser models. II: Symmetries and foldings, Prog. Theor. Phys. 101
(1999) 487, hep-th/9809068.
[17] J. Fernndez Nez, W. Garca Fuertes, A.M. Perelomov, Irreducible characters and ClebschGordan series for the
exceptional algebra E6: An approach through the quantum CalogeroSutherland model, math-ph/0406067;
J. Fernndez Nez, W. Garca Fuertes, A.M. Perelomov, Quantum trigonometric CalogeroSutherland model,
irreducible characters and ClebschGordan series for the exceptional algebra E7, math-ph/0507015.
[18] J.E. Humphreys, Reflection Groups and Coxeter Groups, Cambridge Univ. Press, 1990.
[19] G.C. Shephard, J.A. Todd, Finite unitary reflection groups, Canad. J. Math. 6 (1954) 274.
[20] M. Brou, G. Malle, R. Rouquier, On Complex Reflection Groups and Their Associated Braid Groups, Representations of Groups, vol. 16, Amer. Math. Soc., 1995, p. 1;
M. Brou, G. Malle, R. Rouquier, Complex reflection groups, braid groups, Hecke algebras, J. Reine Angew.
Math. 500 (1998) 127;
A.M. Cohen, Finite complex reflection groups, Ann. Sc. Ecole Norm. Sup. 9 (1976) 379.
[21] C.F. Dunkl, E.M. Opdam, Dunkl operators for complex reflection groups, Proc. London Math. Soc. 86 (2003) 70,
math.RT/0108185.
[22] A.P. Polychronakos, Generalized CalogeroSutherland systems from many-matrix models, Nucl. Phys. B 546
(1999) 495, hep-th/9806189.
[23] A.P. Polychronakos, Generalized Calogero models through reductions by discrete symmetries, Nucl. Phys. B 543
(1999) 485, hep-th/9810211.
[24] A.P. Polychronakos, CalogeroMoser models with noncommutative spin interactions, Phys. Rev. Lett. 89 (2002)
126403, hep-th/0112141.
[25] V.G. Drinfeld, Degenerate affine Hecke algebras and Yangians, Funct. Anal. Appl. 20 (1986) 62.
[26] C.F. Dunkl, Differential-difference operators associated to reflection groups, Trans. Am. Math. Soc. 311 (1989) 167.
[27] A.P. Polychronakos, Exchange operator formalism for integrable systems of particles, Phys. Rev. Lett. 69 (1992)
703.
[28] J.A. Minahan, A.P. Polychronakos, Integrable systems for particles with internal degree of freedom, Phys. Lett.
B 302 (1993) 265, hep-th/9206064.
[29] A.P. Polychronakos, Lattice integrable systems of HaldaneShastry type, Phys. Rev. Lett. 70 (1993) 2329.
[30] D. Bernard, V. Pasquier, D. Serban, Exact solution of long-range interacting spin chains with boundaries, hep-th/
9501044.
[31] F.D.M. Haldane, Exact JastrowGutzwiller resonating-valence-bond ground state of the spin-1/2 antiferromagnetic
Heisenberg chain with 1/r 2 exchange, Phys. Rev. Lett. 60 (1988) 635.
[32] B.S. Shastry, Exact solution of an S = 1/2 Heisenberg antiferromagnetic chain with long-ranged interactions, Phys.
Rev. Lett. 60 (1988) 639.
[33] V. Caudrelier, N. Crampe, Integrable N -particle Hamiltonians with Yangian or reflection algebra symmetry, J. Phys.
A 37 (2004) 6285, math-ph/0310028.
[34] M. Brou, G. Malle, J. Michel, Towards Spetses, Transf. Groups, vol. 4, Birkhuser, 1999, p. 1.
[35] A. Fring, C. Korff, Non-crystallographic reduction of generalized CalogeroMoser models, J. Phys. A 39 (2006)
1115, hep-th/0509152.
[36] M. Cadoni, P. Carta, D. Klemm, Large N limit of CalogeroMoser models and conformal field theories, Phys. Lett.
B 503 (2001) 205, hep-th/0011266.
[37] F. Finkel, D. Gmez-Ullate, A. Gonzlez-Lpez, M.A. Rodrguez, R. Zhdanov, On the Sutherland spin model of BN
type and its associated spin chain, Commun. Math. Phys. 233 (2003) 191, hep-th/0202080.

Induced cosmology on a regularized brane

in six-dimensional flux compactification
Eleftherios Papantonopoulos a , Antonios Papazoglou b,c, ,
Vassilios Zamarias a
a Department of Physics, National Technical University of Athens, Zografou Campus GR 157 73, Athens, Greece
b APC 1 , 10 rue Alice Domon et Lonie Duquet, 75205 Paris Cedex 13, France
c GReCO/IAP 2 , 98 bis Boulevard Arago, 75014 Paris, France
Received 28 September 2007; accepted 17 December 2007

Abstract
We consider a six-dimensional EinsteinMaxwell system compactified in an axisymmetric two-dimensional space with one capped regularized conical brane of codimension one. We study the cosmological
evolution which is induced on the regularized brane as it moves in between known static bulk and cap
solutions. Looking at the resulting Friedmann equation, we see that the brane cosmology at high energies is
dominated by a five-dimensional 2 energy density term. At low energies, we obtain a Friedmann equation
with a term linear to the energy density with, however, negative coefficient in the small four-brane radius
limit (i.e., with negative effective Newtons constant). We discuss ways out of this problem.
PACS: 04.50.-h; 11.25.Mj
Keywords: Extra dimensions; Cosmology with extra dimensions
* Corresponding author at: APC, 10 rue Alice Domon et Lonie Duquet, 75205 Paris cedex 13, France.
E-mail addresses: lpapa@central.ntua.gr (E. Papantonopoulos), papazogl@iap.fr (A. Papazoglou),

zamarias@central.ntua.gr (V. Zamarias).
1 UMR 7164 (CNRS, Universit Paris 7, CEA, Obervatoire de Paris).
2 UMR 7095 (CNRS, Universit Paris 6).
doi:10.1016/j.nuclphysb.2007.12.031
E. Papantonopoulos et al. / Nuclear Physics B 797 [PM] (2008) 520536
521
1. Introduction
Six-dimensional brane world models are rather interesting to be studied for a number of reasons. Firstly, it was in the framework of brane theories with two extra transverse dimensions
that the large (sub-millimeter) extra dimensions proposal for the resolution of the electroweak
hierarchy problem was provided [1]. This scenario makes the study of extra dimensional theories, and string theory in particular, relevant to low energy phenomenology (colliders as well as
astrophysical and cosmological observations) and is testable in the very near future. The different possibilities of realizing a brane world model in six dimensions must therefore be studied to
allow for a comparison with experiment.
Quite a different motivation has been the proposal to ameliorate the cosmological constant
problem, using codimension-2 branes (for a recent review on the subject see [2]). These branes
have the interesting property that their vacuum energy instead of curving their world-volume,
just introduces a deficit angle in the local geometry [3]. Models with this property which exhibit
no fine-tuning between the brane and bulk quantities have been known as self-tuning (for early
attempts to find similar models in five dimensions see [4]). Such self-tuning models with flux
compactification [5,6] have been extensively looked, but the flux quantization condition always
introduces a fine tuning [7], unless one allows for singularities more severe than conical [8]. Alternative sigma-model compactifications have been shown to satisfy the self-tuning requirements
[9]. However, the successful resolution of the cosmological constant problem would also require
that there are no fine-tuning between bulk parameters themselves. No such self-tuning model has
been found yet with all these properties.
A further motivation in studying such models with codimension-2 branes is that gravity on
them is purely understood. The introduction of matter (i.e., anything different from vacuum energy) on them, immediately introduces malicious non-conical singularities [10]. A way out of this
problem is to complicate the gravity dynamics by adding a GaussBonnet term in the bulk or an
induced curvature term on the brane, in which case the singularity structure of the theory is altered and non-trivial matter is allowed [11]. However, the components of the energymomentum
tensor of the brane and the bulk are tuned artificially and the brane matter is rather restricted [12].
Alternatively, one can regularize the codimension-2 branes by introducing some thickness and
then consider matter on them [13]. For example, one can mimic the brane by a six-dimensional
vortex (as e.g. in [14]), a procedure which becomes a rather difficult task if matter is added on it.
Another way of regularization was proposed recently, which consists merely of the reduction
of codimensionality of the brane. In this approach, the bulk around the codimension-2 brane is cut
close to the conical tip and it is replaced by a codimension-1 brane which is capped by appropriate
bulk sections [15] (see [16] for a similar regularization of cosmic strings in flat spacetime). This
regularization has been applied to flux compactification systems in six dimensions for unwarped
rugby-ball-like solutions in [15], for warped solutions with conical branes (with or without
supersymmetry) in [17] and for even more general warped solutions allowing non-conical branes
in [18]. Specific brane energymomentum tensor is required to build static solutions, involving a
brane field with Goldstone-like dynamics. It is interesting to note that in the non-supersymmetric
case, only quantized warpings are allowed [17].
In the present paper we will try to go a step further and consider an isotropic cosmological
fluid on the above-mentioned regularized branes. To have a cosmological evolution on the regularized branes, the brane world-volume should be expanding and in general the bulk space should
also evolve in time. Instead of tackling this problem in its full generality, which seems a formidable task, in the present work we will consider the motion of the regularized codimension-1
522
brane in the space between the bulk and the brane-cap which remains static (see e.g. [19]). In
this way, a cosmological evolution will be induced on the brane in a similar way as in the mirage
cosmology [20], but with the inclusion of the back-reaction of the brane energy density (i.e., the
brane is not considered merely a probe one). Since in the mirage cosmology, the four-dimensional
scale factor descends from the warp factor in the four-dimensional part of the bulk metric, we
will discuss the regularized brane in the case of warped bulk [17], rather than unwarped bulk
[15]. It is worth noting that the above procedure provided in five dimensions the most general
isotropic brane cosmological solutions [21].
Solving the Israel junction conditions, which play the rle of the equations of motion of the
codimension-1 brane, we find the Friedmann equation on the brane which at early times is dominated by an energy density term proportional to 2 , like in the RundallSundrum model in five
dimensions. To recover four-dimensional cosmology at late times, we split as usual the energy
momentum tensor to a part which is the contribution of the static vacuum brane and to a part of
additional matter. We find a regime which has a four-dimensional dependence on the energy density. However, in the interesting case where the brane moves close to its equilibrium point, which
in turn is close to the would-be conical singularity, the coefficient of the linear to the energy
density term is negative (i.e., we obtain negative effective Newtons constant). Thus, we cannot
recover the standard cosmology at late times. This seems to be the consequence of considering
the bulk sections static. It is possible, that this behaviour is due to a ghost mode appearing among
the perturbations of the system, after imposing the staticity of the bulk sections. Furthermore, the
above result points out that there is a difference between the six-dimensional brane cosmology
in comparison to the five-dimensional one. The study of brane cosmology in Einstein gravity
in five dimensions, can be made either in a gauge where the bulk is time-dependent and the
brane lies at a fixed position, or in a gauge where the bulk is static and the brane movement into
the bulk induces a cosmological evolution on it [22]. This, however, does not seem to hold in
six-dimensions anymore.
The paper is organized as follows. In Section 2 we review the static regularized brane solution
in a bulk of general warping. In Section 3 we derive the equations of motion of the codimension-1 brane and in Section 4 we study the induced cosmological evolution on the moving brane.
Finally in Section 5 we draw our conclusions.
2. Setup and static brane solutions
Before discussing the time-dependent scenario, let us remind ourselves of the static solution
which we will use in the following for the brane motion. The bulk theory that we will use is a sixdimensional EinsteinMaxwell system which in the presence of a positive cosmological constant
and a gauge flux, spontaneously compactifies the internal space [23]. The known axisymmetric
solutions have in general two codimension-2 singularities at the poles of a deformed sphere [24].
We will study the case where only one (e.g., the upper) codimension-2 brane is regularized by the
introduction of a ring-like brane at r = rc with an appropriate cap. The dynamics of the system
is given by the following action
4

M
1 2
v2
6
5
2
R i F d x + + (D )
S = d x g
2
4
2

d 4 x T ,
(1)
523
Fig. 1. The internal space where the upper codimension-2 singularity has been regularized with the introduction of a
ring-like codimension-1 brane. The parameters of the action and the solution are denoted in the appropriate part of the
internal space.
where M is the six-dimensional fundamental Planck mass, i are the bulk (i = 0) and cap (i = c)
cosmological constants, FMN the gauge field strength, T the tension of the lower codimension-2
brane, the 4-brane tension, the 4-brane Goldstone scalar field necessary for the regularization
and v the vev of the Higgs field from which the Goldstone field originates. For the coupling
between the Goldstone field and the bulk gauge field we use the notation D = ea ,
with a = AM X M the pullback of the gauge field on the ring-like brane and e its coupling to
the scalar field. In the above action we omitted the GibbonsHawking term. The configuration is
shown in more detail in Fig. 1.
The solution for the bulk and cap regions depends on a parameter which is a measure of the
warping of the space (for = 1 we obtain the unwarped case) and is given by [17]

ds62 = z2 dx dx + Ri2
Fr = ci Ri M 2 S

dr 2
+ ci2 f d 2 ,
f
1
,
z4
(2)
(3)
with Ri2 = M 4 /(2i ) and the following bulk functions

1
(1 )r + (1 + ) ,
(4)
2

5
1 8 1
1
1
2
31
+
f (r) =
(5)
z
,
5(1 )2
1 3 z3
1 3 z6
1 5
with S() = 35 3 1
3 . The range of the internal space coordinates is 1 r 1 and 0
< 2 . Taking into account that in the limit r 1, it is f 2(1 r)X with the constants X given by
z(r) =
X+ =
5 + 3 8 8 3
,
20(1 )(1 3 )
X =
3 + 5 8 8 5
,
20 4 (1 )(1 3 )
(6)
524
the cap is smooth at r = +1 as long as is cc = 1/X+ . Furthermore, the metric is continuous if

c0 R0 = cc Rc , which gives Rc = + R0 with + = X+ c0 .3 The conical singularity at r = 1 is
supported by a codimension-2 brane with tension
T = 2M 4 (1 c0 X )
(7)
while the parameters of the 4-brane , v are fixed by the radii R0 , Rc and the brane position rc
[17]. Furthermore, the gauge field is quantized as
2c0 R0 M 2 eY = N,
N Z,
(8)
(1 3 )
with Y = 3 3 (1) S and the brane scalar field has solution = n with n Z. The two quantum
numbers n, N are related through the junction conditions as

2
N
5(1 8 )
3
.
n=
(9)
2 (1 3 ) 8(1 5 )
Since the quantities n, N are integers, the above relation imposes a restriction to the values of the
admissible warpings , which implies that static solutions are consistent only for discrete values
of the warping .
3. Moving brane junction conditions
To study the cosmological evolution on the 4-brane we introduce an energymomentum tensor
(br)
of a perfect fluid on the brane. Then the total energymomentum tensor t
= (2/ + )
Sbr /+ (where Sbr is the brane action) will be given by

(br)
= diag(, P , P , P , P )
(10)
and a possible coupling of the brane matter to the bulk gauge field (consistent with the cosmological symmetries) by
Sbr /a = (l, L, L, L, L).
(11)
Splitting the above quantities to one part responsible for the static solution and another expressing
the presence of matter on the brane, we have
=+
2
v 2 (n eA+
)
2c02 R02 f (rc )
P =
P = +
l = lm ,
L = Lm ,
+ m 0 + m ,
2
v 2 (n eA+
)
2c02 R02 f (rc )

2
v 2 (n eA+
)
2c02 R02 f (rc )
(12)
+ Pm 0 + Pm ,
(13)
+ Pm ,
(14)
(15)
(16)
3 In this brief presentation of the background, we have taken = 1 in comparison with [17]. The physical quantities,
however, are -independent and depend only on + .
L = ev 2 n eA+
+ Lm
525
(17)
with all the other quantities vanishing and m , Pm , Pm , lm , Lm , L m the matter contributions.
In general, the inclusion of the above matter contributions will have the effect of giving some
time dependence both to the brane as well as the bulk solutions. Since the study of the full time
dependent problem is rather difficult, we will make in the present paper the approximation that
the bulk remains static and that the brane matter merely makes the brane to move between the
two static bulk sections, away from its equilibrium point r = rc . This is to be regarded as a first
step towards understanding the generic brane cosmological evolution.
To embed the brane in the static bulk, let us take the brane coordinates be = (, x i , ).
[The brane-time coordinate is not to be confused with the Goldstone field which will not
appear in our subsequent analysis.] Then the brane embedding X M in the bulk is taken for both
sections to be
Xi = x i ,
X r = R( ) and
X = ,
(18)
while for the time coordinate embedding we choose for the outer bulk section
0
X(out)
= ,
(19)
and for the inner cap section

0
X(in)
= T ( ).
(20)
M
N
The continuity of the induced metric
= gMN X X , apart from the relation c0 R0 =
cc Rc as in the static case, gives a relation of the time coordinate T in the upper cap region with
the brane time coordinate (dots are with respect to )

2
2 2
2
2 R R0
2 R0
T 1 + 2
(21)
= 1R
.
f z2
T f z2
Then the induced metric on the brane reads

2
ds(5)

2
2 R0
= z 1 R
d 2 + z2 d x2 + c02 R02 f d 2 .
f z2
2
(22)
The continuity of the gauge field, on the other hand, is guaranteed by the fact that its only
non-vanishing component is A and X is -independent.
Apart from the continuity conditions we have to take into account the junction conditions for
the derivatives of the metric and the gauge field, which read
1 (br)
t ,
M 4

Sbr
nM FNM X N = .
a
{K } =
(23)
(24)
We denote {H } = H in + H out the sum of the quantity H from each side of each brane. The
extrinsic curvatures are constructed using the normal to the brane nM which points inwards to
the corresponding part of the bulk each time (we use the conventions of [25]). The left-hand
sides of the above equations are computed in detail in Appendix A. Then the junction conditions
526
for the metric are the following: The ( ) component is

f
f
z
1
4
M 3 +
|T | = .
1

z
2f
+
R02
2
R0 1 R 2
(25)
fz
The (ij ) component is

R0 M 4
+ T 2
R
.

R
2 2z + f
2R
1 + |T |
z
2f
|T |
2 R02 )3/2
z2 f (1 R
fz

f
1 3
f z2 z
|T |
1
= P .
+ 2 3 +
z
2f
+
R0
(26)
The () component is
R0 M 4
+ T 2
R

.

f
z
R
2
R 5 +
1 + |T |
z
2f
|T |
2

2 R02 3/2
z2 f 1 R
fz

4f zz
1 3
+
|T |
1
= P .
+
R02
Finally, the () component of the gauge field junction condition is

c0 M 2 S f
1

|T | = L,
1
2
+
R
0
4
2
z 1R
2
(27)
(28)
fz
while its other components dictate that the couplings of the brane matter to the other bulk gauge
components vanish
l = L = 0.
(29)
Eq. (25) gives a Friedmann equation for the brane coordinate R, while Eq. (26) gives an
acceleration equation. On the other hand, (27) and (28) are constraint equations for the matter on
the brane. We will for the moment assume that the brane matter is such that P and L are given
by the latter equations. In a concrete model of matter on the brane the latter two equations will
introduce some additional constraint for its evolution. It is easy to see from (27) and (28), that in
the static limit P and L tend to zero.
4. The cosmological dynamics of the 4-brane
For discussing the dynamics of the moving brane, we must study the Friedmann (25) and the
acceleration (26) equations. Firstly, in order to bring the Friedmann equation to a more standard
form, we rewrite the metric in the form
2
= d 2 + a 2 ( ) d x2 + b2 ( ) d 2 ,
ds(5)
with a = z(R( )) and b = c0 R0 f (R( )). The brane proper time is given by

R2
2 0 .
2 = z2 1 R
f z2
(30)
(31)
527
From now on we will assume without loss of generality that > 0. It is evident from the above,
that cosmological evolution from mere motion of the brane in the static bulk is possible only
when there is warping in the bulk. In the unwarped version of this model [15], mirage cosmology
is impossible and some bulk time-dependence is compulsory.
The Hubble parameters for the two scale factors are given by
Ha
1 da
z
= 2
a d
z
(32)
,
R2
2 02
1R
fz
and
Hb
1 db
f
=

b d
2f z
(33)
2
2 R02
1R
fz
Then the ratio of Ha and Hb gives a precise relation between them for our particular model. It is
given by
Hb =
zf
Ha ,
2f z
(34)
and we notice that since in the model we study it is always f < 0, in the neighborhood of r = 1,
the two Hubble parameters have opposite sign. This means that if the four-dimensional space
expands, the internal space shrinks.
and the time embedding T as funcFrom (32) and (21), we can express the brane velocity R
tions of the Hubble parameter Ha as

2 AH 2 R 2
2
1 + +
H
z
a
a 0
=
| =
,
|
T
,
R
(35)
z 1 + AH 2 R 2
1 + AHa2 R02
a
with A =
z2
f z 2
evaluated on the brane.

and T from (35) back to (25), we obtain the Friedmann equation as a
Then substituting R
function of the Hubble parameter
1
M4
2 AH 2 R 2 = ,
C
1 + AHa2 R02
1 + +
(36)
a 0
R0
+

f
) is evaluated on the brane. We can solve the above equation for Ha and
where C = f (3 zz + 2f
obtain a more familiar form
Ha2 =
2 )2
2
M 8 C 2 (1 +
1 + +
1
1
2
.
+
4A
2 R2A
4M 8 C 2 A
2 2+
4R04 +
0
(37)
The above Friedmann has unconventional dependence on the energy density. In particular the
inverse square dependence is known to occur for motions in backgrounds of asymmetrical warping [26].
The equilibrium point rc of the system is found if we set Ha = 0, which gives the brane energy
density without matter
0 =
M 4 Cc
(1 + ),
R0 +
(38)
528
Fig. 2. The generic form of the effective Newtons constant Geff as a function of the brane position R( ). As the brane
approaches the equilibrium point rc , we always have Geff < 0. Additionally, Geff diverges as r 1 and at one point rd
in between.
where Cc is the value of C at rc and is the same appearing in (12). Let us note that there exists
a second root of (37) which is not compatible with (36). [This is because we squared twice
(36) to take (37).] The behaviour of the function C is that, as r 1 it limits to C and
monotonically decreases and limits to C as r 1. On the other hand A, is always
positive with A as r 1, and O(1) in the intermediate region. In the unwarped limit
1 the Friedmann equation becomes as expected trivial, i.e., Ha = 0.
Before studying various limits of the above equation, let us define the effective four-dimen(4)
sional matter energy density m by averaging over the azimuthal direction (we assume that m is
independent of )

2+
(4)
m = d g m =
(39)
R0 f m ,
X+
with similar definitions for the other 4-brane quantities.
Let us suppose now that initially 1 R( ) < rc < 1, with 1 rc 1. The goal is to
find how R( ) behaves. To recover a four-dimensional Friedmann equation at late times we
can assume that the brane energy density is small in comparison with the static case energy
(4)
(4)
density, i.e., m
0 , so we can expand (37) in powers of m and obtain the following fourdimensional form of the Friedmann equation
(4)2
8
(4)
,
+ (a) + O m
Geff m
3
where the effective Newtons constant is
4

C 1 + + 2
3X+ Cc (1 + )
1
.
Geff =
2 M 4 AC 2 f C 4 1
32 2 R02 +
+
c
Ha2 =
(40)
(41)
The quantity (a) depends on the parameters of the bulk and plays the rle of the mirage matter
induced on the brane from the bulk and it is given by

2
1 + +
(1 + )2 Cc2 C 2 1 + + 2
(a) =
(42)
+ 2
2
.
2 A C2
Cc 1 +
(1 + )2
4R02 +
The behaviour of Geff as a function of R( ) is given in Fig. 2 and has the following important
features: At the points where the geometry becomes conical (r 1) the effective Newtons
529
Fig. 3. The generic form of the mirage matter contribution as a function of the brane position R( ). At the static
equilibrium point rc it vanishes and at rd it diverges.
constant is negative and diverging. In between, there is a point rd , with

3(1 8 ) 1/5
,
zd =
8(1 3 )
(43)
which is a root of C and Geff diverges to +. At this point the matter energy density is bound
to vanish. It is important to note that even in the region where Geff is positive, there is always a
strong time variation of Geff , which for
1 dGeff
= Ha ,
Geff d
(44)
has > O(10), in contradiction with observations [27] which dictate that < 0.1. But even more
important is the fact that close to the static equilibrium point, where the cosmology is supposed
to mimic best the one of a codimension-2 brane, we get negative Newtons constant
Geff (R = rc ) =
3X+
2
2
8 R0 + (1 + )M 4 AC c f
< 0,
(45)
since we have that for any value of the parameters and for rc in the neighborhood of r = +1, it
is Cc < 0.
Let us also look at the mirage matter contribution , which is depicted in Fig. 3. As expected,
it vanishes for the static equilibrium point rc , it is finite at the boundaries r = 1 and diverges
at the root rd of C. Again there is no region in the brane position interval where the contribution
of is constant enough to resemble a cosmological constant contribution to the four-dimensional
Friedmann equation.
On the opposite limit, that the matter energy density is much larger than the static case energy
(4)
density, i.e., m
0 , we get the expected asymptotics
1
2 ,
(46)
4M 8 C 2 A m
which is a five-dimensional Friedmann law (with time-varying five-dimensional Newtons constant) at early times.
Taking under consideration the difficulties of the model to give a Friedmann equation with the
correct sign of Geff , we will not proceed with the presentation of the analysis of the acceleration
equation (26). With this equation, one finds even more difficulties towards obtaining a realistic
Ha2 =
530
four-dimensional evolution. For example, in the low energy limit, one gets a coefficient of the
linear energy density term, which is not related to the Geff , that we obtained from the Friedmann
equation, in the way it does in standard four-dimensional general relativity.
This pathological features of the low energy density limit, where the expansion does not appear to have an effective four-dimensional limit, can find a potential explanation when looking
at the energy continuity equation on the brane. Taking the covariant divergence of the Israel
junction condition (23), we can obtain the following equation
(4)
(4) (br)
= M 4 K .
t
(47)
Then using the Codazzi equation
N
(4)
K = GK n hK
N X ,
and the bulk Einstein equation, we arrive at the simple expression [25]

(B)
(br)
N
(4)
.
t
= TK hK
N n X

(48)
(49)
Because of the jump of the bulk energymomentum tensor across the 4-brane, the energy
momentum tensor on the 4-brane is not conserved. This is a usual feature of moving brane
cosmologies in asymmetrically warped backgrounds [28]. In more detail, the form of the above
equation for the particular model is given by
zf
S 2 Ha
d
H
=
+ 3( + P )Ha + ( + P )
,
a
d
2f z
z z7 C f
(50)
where in the right-hand side we have used the Friedmann equation (37). In a straightforward but
lengthy calculation, one can see that the latter equation can be derived from (25), (26), (27). The
problems in the four-dimensional limit, that we faced previously, can be traced to large energy
dissipation off the brane as well as a large work done during the contraction of the ring-brane.
5. Conclusions
In the present paper we made a first step towards the study of the cosmology of a codimension-2 brane, which is regularized by the method of lowering its codimensionality (cutting the
space close to the conical tip and replacing it by a ring brane with an appropriate cap). As a first
approximation, we assumed that the bulk and the cap remain static as the brane moves between
them. The motion of the brane then induces a cosmological evolution for the matter on the brane.
The junction conditions provide the Friedmann and acceleration equations on the brane.
We can see already from the Friedmann equation that we cannot recover standard cosmological evolution of the brane at low energies. The effective Newtons constant is negative in
the interesting limit that the brane approaches its equilibrium point, close to the would-be conical singularity. In other words, we obtain antigravity in this limit. Even away from this point,
i.e., when the brane moves away from its equilibrium point, the Newtons constant varies significantly in contradiction with standard cosmology. At one position of the internal space, the
Newtons constant even diverges and forces the matter energy density to vanish. Taking all the
above into account, we did not present the further analysis of the system by considering the
acceleration equation.
This result is not altered by supersymmetrizing the model [6]. The bulk and cap solutions are
different from the non-supersymmetric case due to the presence of the dilaton. However, the only
531
difference in the Friedmann is a redefinition of the quantities A and C. In the supersymmetric

case, these quantities read

z2
3z
f
Asusy = 4 2 and Csusy = f
(51)
+
.
2z
2f
fz
It is easy to see again that the effective Newtons constant is negative for the motion near the
would-be conical tip and that even away from that point, it is very strongly varying.
The reason for this unconventional cosmological evolution, is the use of the specific restricted
ansatz for the solution of the systems equation of motion. The staticity of the bulk was proved
to be an oversimplification. We imagine that this restriction on the system may result to the
appearance of some scalar mode in the perturbative analysis of [29], which for a certain region
of the brane motion (close to the pole of the internal manifold) is ghost-like. This mode may
then be responsible for the negative effective Newtons constant. The unrestricted perturbative
analysis in [29] resulted in a linearized four-dimensional Einstein equation for distances larger
than the compactification scale. The same happened in [15] in the unwarped model perturbation
analysis4 (see [30] for a related analysis with a brane induced gravity term), even though it would
not have mirage evolution, as we studied it here, because of the absence of a warp factor. The
appearance of the standard four-dimensional linearized dynamics, shows that they are realized
when the bulk is necessarily time-dependent.
Clearly, the next step should be the study of the system in a setup where the bulk is also
time-dependent. In that respect, brane cosmology in six dimensions seems to be different for
the one in five dimensions. In Einstein gravity in five dimensions, one can always work in a
gauge where the bulk is static and the brane acquires a cosmological evolution by moving into
the bulk. However, in six dimensions there are more degrees of freedom which make this gauge
choice not general. In the present paper, we have frozen these degrees of freedom hoping to find
consistent cosmological solutions, but it turned out that this does not give a viable cosmology.
There are several known time-dependent backgrounds which can be used to look for realistic
brane cosmologies [31]. We plan to address this issue in the near future.
Addendum
At the last stages of this work, we became aware of the work of [32], where a cosmological
evolution, similar in spirit with the present paper, is discussed. More precisely, the time evolution
of the modulus of a complex scalar field on the brane is studied, while keeping the bulk static.
Our results are in agreement with [32] and in many respects complementary. In particular, as in
our case, four-dimensional late time cosmology is not recovered.
Acknowledgements
We would like to thank C. Charmousis, J. Gray, D. Langlois and M. Minamitsuji for helpful
discussions. This work is co-funded by the European Social Fund and National Resources
(EPEAEK II)-PYTHAGORAS. E.P is partially supported by the European Union through the
Marie Curie Research and Training Network UniverseNet (MRTN-CT-2006-035863).
4 The question of stability of the regularized models was not discussed in [15,29] and it could be that these compactifications have modes with negative mass squared.
532
Appendix A. Computation of extrinsic curvatures

In this appendix we will calculate the extrinsic curvatures on the brane positions, which will
be used in the main text to evaluate the junction conditions. Let the brane position in the bulk be
M
N
X M ( ), from which we evaluate the induced metric on the brane as
= gMN X X .
Firstly, we should calculate the normal vector of the brane nM , which is orthogonal to all the
tangent vectors of the brane X M , that is X M nM = 0, and is normalized as nM nN g MN = 1.
Once the normal vector is computed, one can evaluate the projection tensor hMN = gMN
nM nN , which is related to the induced metric as hMN = X M X N and satisfies also
=
hMN X M X N due to the orthogonality of the normal to the tangent vectors. Afterwards, the
extrinsic curvature is given by KMN = hK

M hN K n (the covariant derivative computed with
gMN ) and with trace K = g MN KMN . The pullback of the extrinsic curvature on the brane is given
K
by K = KMN X M X N and has the property that K =
= K. The combination
which appears in the junction conditions is K = K K .

As is discussed in [17], the metric component grr in the radial coordinate direction need not
be continuous. This then introduces normal vectors nM for each side of the brane which are
not opposite (since they are normalized with different grr ). Nevertheless, the junction conditions
make perfect sense if the discontinuous grr and the non-opposite nM s are adequately used.
Let us first compute the quantities we are interested in for the outer bulk section. The normal
vector is
R0
nout
M =
f
1
2
2 R02
1R
fz
0,
1, 0).
(R,
(A.1)
[From now on we suppress the notation out which is to be understood for all subsequent
quantities.]
The projection tensor hMN = gMN nM nN has components
h00 =
1
2
2 R02
1R
fz
h0r = R
0,
hr0 = Rh
0
R02 0
h ,
f z2 0
hij = ji ,
2
hrr = R
R02 0
h ,
f z2 0
h = 1.
(A.2)
(A.3)
The Christoffel symbols for the bulk metric that we need are
r =
z
,
z
rrr =
f
,
2f
zz f
,
R02
1
r
= c02 ff .
(A.4)
(A.5)
[Also r = f /(2f ) but we do not need it. All other are zero.]
We can now compute the extrinsic curvatures KMN = hK

M hN K n . Note that since we have
expressed the normal vectors in the brane coordinates, the bulk derivative is understood to be
K
taken using the projection rule hK

M K = gMK X . The calculation gives

2
r +R
2 2 0 rrr
K00 = h00 nr R
00
0r

0 2
zz f
z
f
2
,
= h0 nr R 2 + R 2 +
z
2f
R0
4
R2
2 R0
0 K00 ,
K05 = R
K
=
R
K00 ,
55
f z2
f 2 z4
zz f
Kij = ijr nr = 2 nr ij ,
R0
1
r
K =
nr = c02 .ff nr .
2
The pullbacks of the extrinsic curvatures on the brane K = KMN X M X N are

2 2
zz f
z
f
2 R0
2
,
K00 = nr R 2 + R 2 +
K = 1 R
z
2f
f z2
R0
zz f
nr ij ,
R02
1
K = K = c02 ff nr .
2
Kij = Kij =
533
(A.6)
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
(A.12)
Now we can construct the hatted quantities K = K K . They read

R2
z
f
f z2
2 0
3
+
,
K = nr 2 1 R
z
2f
f z2
R0

nr ij
f z2 z
f
+ 2R
2 2z + f
K ij =
3
+
,
R
2
z
2f
z
2f
R02
2 R02 )
(1 R
fz

2
nr c0 R02 f
z
f
zz f
2
K =
.
R 4 2 + R 5 +
2
z
2f
R0
2 R02 )
z2 (1 R
fz
(A.13)
(A.14)
(A.15)
Let us now compute the quantities we are interested in for the inner cap section. The normal
vector is

R
1
+ R0
nin
(A.16)
=
,
0,
1,
0
.

M
2
f
T
R2 R0
2
1 + T 2 f z2
[From now on we suppress the notation in which is to be understood for all subsequent
quantities.]
The projection tensor hMN = gMN nM nN has components
h00 =
1
2
2 R 2 R0
1 +
T 2 f z2
2
h0r = +
hij = ji ,
R2
R
0 0
h ,
T f z2 0
h = 1.
hr0 =
R
h0 ,
T 0
2
hrr = +
2 R2
R
0 0
h ,
T 2 f z2 0
(A.17)
534
The Christoffel symbols for the bulk metric that we need are
r =
z
,
z
rrr =
f
,
2f
zz f
,
2 R 2
+
0
(A.18)
1 2
r
= c+
ff .
2
(A.19)
We can now compute the extrinsic curvatures as

.

2

0 2
R
1 R
r
0
r
K00 = h0 nr
00 + 2 20r rr
T T
T

.

2 z
2
R
1 R
f
zz f
,
2 + 2 2 +
= h00 nr
z
2f
T T
T
R+
2
K05 = +
R2
R
0
K00 ,
T f z2
4
K55 = +
2 R4
R
0
K00 ,
T 2 f 2 z4
zz f
n ,
2 R 2 r ij
+
0
1 2
r
nr = c+
ff nr .
K =
2
Kij = ijr nr =
(A.20)
(A.21)
(A.22)
(A.23)
The pullbacks of the extrinsic curvatures on the brane K = KMN X M X N are

2 2 2
2 R R0
K = T 2 1 +
K00
T 2 f z2

.
2 z
R
1 R
f
zz f
2
2 +
,
= T nr
2 2+
z
2f
T T
T 2
+ R0
zz f
n ,
2 R 2 r ij
+
0
1 2
ff nr .
K = K = c+
2
Kij = Kij =
(A.24)
(A.25)
(A.26)
Now we can construct the hatted quantities K = K K . They read

2 2 z
f
f z2 2
2 R R0
3 +
,
K = nr 2 2 T 1 + 2
z
2f
T f z2
+ R0

.
2 z
nr ij
R
1 R
f
Kij =
+2 2 2 +
2
z
2f
T T
T
2 R 2 R0 )
(1 +
2
2
T f z

z
f
f z2
,
2 2 3 +
z
2f
+ R0

.
2 z
nr c02 R02 f
R
f
1 R
zz f
K =
5
+
.
4
+
2
2 R2
z
2f
T T
T 2
+
2 R 2 R0 )
0
z2 (1 +
T 2 f z2
(A.27)
(A.28)
(A.29)
535
We can now write down the jumps of the hatted extrinsic curvatures, which participate in the
junction conditions

z2 f
R02
1
f
z
2
1
{K } = 3 +
(A.30)
1R
|T | ,
z
2f
R0
+
f z2

.

R0 ij
+ T 2 R 2R
2 2z + f
R
1 + |T |
{K ij } =
2
z
2f
|T |
2 R02 )3/2
f (1 R
fz

1 3
f z2 z
f
,
1
+ 2 3 +
(A.31)
|T |
z
2f
+
R0

.

c02 R03 f
z
f
2 R
2
R + T
{K } =
1 + |T |
R 5 +
2
R
z
2f
|T |
2 02 )3/2
z2 f (1 R
fz

1 3
4f zz
1
.
|T |
+
(A.32)
2
+
R0
Furthermore, the jump of the gauge field is given by

1
c0 M 2 S f
nr F r =
1
|T | .
2
+
2 R02
z4 1 R
(A.33)
fz
References
[1] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398;
N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Rev. D 59 (1999) 086004, hep-ph/9807344.
[2] K. Koyama, arXiv: 0706.1557 [astro-ph].
[3] J.W. Chen, M.A. Luty, E. Ponton, JHEP 0009 (2000) 012, hep-th/0003067.
[4] N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, Phys. Lett. B 480 (2000) 193, hep-th/0001197;
S. Kachru, M.B. Schulz, E. Silverstein, Phys. Rev. D 62 (2000) 045021, hep-th/0001206;
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;
C. Csaki, J. Erlich, C. Grojean, Nucl. Phys. B 604 (2001) 312, hep-th/0012143;
J.E. Kim, B. Kyae, H.M. Lee, Nucl. Phys. B 613 (2001) 306, hep-th/0101027.
[5] S.M. Carroll, M.M. Guica, hep-th/0302067;
I. Navarro, JCAP 0309 (2003) 004, hep-th/0302129;
Y. Aghababaie, C.P. Burgess, S.L. Parameswaran, F. Quevedo, Nucl. Phys. B 680 (2004) 389, hep-th/0304256.
[6] G.W. Gibbons, R. Guven, C.N. Pope, Phys. Lett. B 595 (2004) 498, hep-th/0307238;
C.P. Burgess, F. Quevedo, G. Tasinato, I. Zavala, JHEP 0411 (2004) 069, hep-th/0408109.
[7] I. Navarro, Class. Quantum Grav. 20 (2003) 3603, hep-th/0305014;
H.P. Nilles, A. Papazoglou, G. Tasinato, Nucl. Phys. B 677 (2004) 405, hep-th/0309042;
H.M. Lee, Phys. Lett. B 587 (2004) 117, hep-th/0309050.
[8] C.P. Burgess, AIP Conf. Proc. 743 (2005) 417, hep-th/0411140;
N. Kaloper, D. Kiley, JHEP 0603 (2006) 077, hep-th/0601110.
[9] A. Kehagias, Phys. Lett. B 600 (2004) 133, hep-th/0406025;
S. Randjbar-Daemi, V.A. Rubakov, JHEP 0410 (2004) 054, hep-th/0407176;
H.M. Lee, A. Papazoglou, Nucl. Phys. B 705 (2005) 152, hep-th/0407208.
[10] J.M. Cline, J. Descheneau, M. Giovannini, J. Vinet, JHEP 0306 (2003) 048, hep-th/0304147.
[11] P. Bostock, R. Gregory, I. Navarro, J. Santiago, Phys. Rev. Lett. 92 (2004) 221601, hep-th/0311074;
C. Charmousis, R. Zegers, Phys. Rev. D 72 (2005) 064005, hep-th/0502171.
536
[12] E. Papantonopoulos, A. Papazoglou, JCAP 0507 (2005) 004, hep-th/0501112;

G. Kofinas, Phys. Lett. B 633 (2006) 141, hep-th/0506035;
E. Papantonopoulos, A. Papazoglou, JHEP 0509 (2005) 012, hep-th/0507278.
[13] B. Carter, R.A. Battye, J.P. Uzan, Commun. Math. Phys. 235 (2003) 289, hep-th/0204042;
M. Kolanovic, M. Porrati, J.W. Rombouts, Phys. Rev. D 68 (2003) 064018, hep-th/0304148;
S. Kanno, J. Soda, JCAP 0407 (2004) 002, hep-th/0404207;
J. Vinet, J.M. Cline, Phys. Rev. D 70 (2004) 083514, hep-th/0406141;
I. Navarro, J. Santiago, JHEP 0502 (2005) 007, hep-th/0411250;
J. Vinet, J.M. Cline, Phys. Rev. D 71 (2005) 064011, hep-th/0501098.
[14] P. Peter, C. Ringeval, J.P. Uzan, Phys. Rev. D 71 (2005) 104018, hep-th/0301172.
[15] M. Peloso, L. Sorbo, G. Tasinato, Phys. Rev. D 73 (2006) 104025, hep-th/0603026.
[16] J.R.I. Gott, Astrophys. J. 288 (1985) 422.
[17] E. Papantonopoulos, A. Papazoglou, V. Zamarias, JHEP 0703 (2007) 002, hep-th/0611311.
[18] C.P. Burgess, D. Hoover, G. Tasinato, arXiv: 0705.3212 [hep-th].
[19] B. Cuadros-Melgar, E. Papantonopoulos, Phys. Rev. D 72 (2005) 064008, hep-th/0502169;
E. Papantonopoulos, gr-qc/0601011.
[20] A. Kehagias, E. Kiritsis, JHEP 9911 (1999) 022, hep-th/9910174.
[21] P. Kraus, JHEP 9912 (1999) 011, hep-th/9910149;
D. Ida, JHEP 0009 (2000) 014, gr-qc/9912002.
[22] S. Mukohyama, T. Shiromizu, K.i. Maeda, Phys. Rev. D 62 (2000) 024028, hep-th/9912287;
S. Mukohyama, T. Shiromizu, K.i. Maeda, Phys. Rev. D 63 (2001) 029901, Erratum.
[23] P.G.O. Freund, M.A. Rubin, Phys. Lett. B 97 (1980) 233;
S. Randjbar-Daemi, A. Salam, J.A. Strathdee, Nucl. Phys. B 214 (1983) 491.
[24] S. Mukohyama, Y. Sendouda, H. Yoshiguchi, S. Kinoshita, JCAP 0507 (2005) 013, hep-th/0506050;
H. Yoshiguchi, S. Mukohyama, Y. Sendouda, S. Kinoshita, JCAP 0603 (2006) 018, hep-th/0512212.
[25] H.A. Chamblin, H.S. Reall, Nucl. Phys. B 562 (1999) 133, hep-th/9903225.
[26] D. Ida, JHEP 0009 (2000) 014, gr-qc/9912002.
[27] O.G. Benvenuto, E. Garcia-Berro, J. Isern, Phys. Rev. D 69 (2004) 082002.
[28] R.A. Battye, B. Carter, A. Mennim, J.P. Uzan, Phys. Rev. D 64 (2001) 124007, hep-th/0105091;
D. Yamauchi, M. Sasaki, arXiv: 0705.2443 [gr-qc].
[29] T. Kobayashi, M. Minamitsuji, Phys. Rev. D 75 (2007) 104013, hep-th/0703029.
[30] N. Kaloper, D. Kiley, JHEP 0705 (2007) 045, hep-th/0703190.
[31] K.i. Maeda, H. Nishino, Phys. Lett. B 154 (1985) 358;
K.i. Maeda, H. Nishino, Phys. Lett. B 158 (1985) 381;
A.J. Tolley, C.P. Burgess, C. de Rham, D. Hoover, New J. Phys. 8 (2006) 324, hep-th/0608083;
B. Himmetoglu, M. Peloso, Nucl. Phys. B 773 (2007) 84, hep-th/0612140;
T. Kobayashi, M. Minamitsuji, arXiv: 0705.3500 [hep-th];
E.J. Copeland, O. Seto, arXiv: 0705.4169 [hep-th].
[32] M. Minamitsuji, D. Langlois, arXiv: 0707.1426 [hep-th].

Covariant particle statistics and intertwiners

of the -deformed Poincar algebra
C.A.S. Young , R. Zegers
Department of Mathematical Sciences, University of Durham, South road, Durham DH1 3LE, UK
Abstract
To speak about identical particlesbosons or fermionsin quantum field theories with -deformed
Poincar symmetry, one must have a -covariant notion of particle exchange. This means constructing intertwiners of the relevant representations of -Poincar. We show, in the simple case of spinless particles, that
intertwiners exist, and, supported by a perturbative calculation to third order in 1 , make a conjecture about
the existence and uniqueness of a certain preferred intertwiner defining particle exchange in -deformed
theories.
PACS: 11.10.Nx; 11.30.Cp; 02.20.Uw; 02.40.Gh
Keywords: Deformed Poincar algebra; -deformation; Particle statistics
1. Introduction and overview

The so-called -deformation of Minkowski space and the Poincar algebra is an important
and much-studied example of a deformation of classical spacetime and its symmetries. The fourdimensional -Poincar algebra was first obtained in [1], following earlier work on the twodimensional case [2], by a contraction of a simple quantum group which parallels the usual
InnWigner [3] contraction SO(1, 4) Poincar.
In terms of the coordinate functions x , -deformation preserves the usual commutativity of
the position coordinates but introduces a deformation in the commutators involving time [4,5]:
E-mail addresses: charlesyoung@cantab.net (C.A.S. Young), robin.zegers@durham.ac.uk (R. Zegers).

doi:10.1016/j.nuclphysb.2007.12.021
538
C.A.S. Young, R. Zegers / Nuclear Physics B 797 [PM] (2008) 537549
i j
x , x = 0,
1
(1.1)
x0, xi = xi ,
where is a deformation parameter with dimensions of length. A characteristic effect of deformation is a modification of the dispersion relation, which becomes1

E 2
2 sinh
P P.
(1.2)
2
This dispersion relation has appeared in other contexts, including doubly special relativity
[68]for discussion see [9,10]and, after exchanging E P , the worldsheet magnons of
the AdS/CFT correspondence [11].
The present work is a contribution to the project of formulating quantum field theories with deformed symmetries. There is an extensive literature on this topic (and on classical field theory):
see, e.g., [8,1218]. Two broad approaches exist: from one point of view, quantum field theories
are constructed, roughly speaking, by quantizing classical field theories; in this approach the
non-commutative algebra of coordinates assumes primacy [5,1922]. Here, on the other hand,
we follow the point of view of Weinberg in [23] that particles, thought of as irreducible representations of the spacetime symmetry algebra, are the fundamental building blocks, with quantum
fields and their field equations emerging later. The first thing one needs, then, is a knowledge of
the irreps of -Poincar, i.e., of single-particle states. Fortunately -Poincar and Poincar are
known to be isomorphic as algebras (not as Hopf algebras) so they share the same representations
[2426]. Next, to understand interacting theories, one must be able to take tensor products of representations, i.e., to build multi-particle states. The coalgebra structure of -Poincar (which is
genuinely distinct from that of Poincar) allows one to do just this.
In ordinary QFT there is a further key ingredientthe notion of identical, or indistinguishable, particles, with definite exchange statistics. This is most familiarly expressed in terms of the
algebra of creation operators which act on the Fock vacuum:
a (p)a (q) a (q)a (p) = 0
(1.3)
with + () for bosons (respectively, fermions). What this really says is that the space of twoparticle states is not some tensor product V V but rather the quotient of this by the operation
( ) which exchanges the factors:
: |p |q |q |p.
(1.4)
It is important that this definition is covariant, so that the notions of particle exchange and hence
identical particles are the same in all frames. This is true since commutes with action of the
(co-commutative) Poincar algebrathat is, it is an intertwiner of representations.
In this paper we ask: is there an analogue of in the -deformed case? The question is nontrivial because no universal R-matrix is known (although cf. [27,28]), so it is necessary to look for
intertwiners by hand.2 What one means by -deformed quantum field theory hinges on whether
1 This formula is expressed in terms of the original basis of -Poincar, whereas the Casimir (3.12) below is in the
bicrossproduct basis [39].
2 This is in contrast to another much-studied deformation, -deformation, in which the non-commutativity of the coordinates is specified by a constant matrix: [x , x ] = [2931]. -deformation does possess a universal R matrix and
indeed is a twist of the usual commutative case (see, e.g., [32]). Because twists are in a certain sense gauge transformations [33], all the relevant structuresincluding the algebra of creation operatorscan be obtained in a controlled way,
but there has been some debate about whether the resulting theory is physically distinct from the undeformed case [34].
539
suitable intertwiners exist: if there are no such intertwiners then there is no covariant notion of
identical particles, and the counting of multi-particle states will differ from the undeformed case;
cf. [35]. This may indeed be the case for some deformations of Poincar, but it is a radical
conceptual departure that, in our view, is physically unattractive and to be avoided if possible; it
is better to modify the minimum one can get away with. Our approach is thus closer in spirit to the
very interesting papers [36,37], with the distinction that we shall insist that our notion of particle
exchange is covariant, i.e., intertwines the action of the boosts Ni as well as the momentum
operators Pi .
In what follows, after briefly recalling the definition of the -Poincar Hopf algebra, we will
address the question above, restricting ourselves to the simplest possible case: the two-particle
sector of a theory of one species of spinless particle. We find that intertwiners do exist, and also
introduce a list of further conditions which we conjecture (based on a perturbative calculation to
third order in 1 ) specify the unique intertwiner naturally associated with -deformed particleexchange.
2. Intertwiners for the -Poincar Hopf algebra
We will work in the bicrossproduct basis, first presented in [39].3 In this basis the Lorentz
algebra is undeformed:
[Mi , Mj ] = ij k Mk ,
[Mi , Nj ] = ij k Nk ,
[Ni , Nj ] = ij k Mk ,
(2.1)
where Mi and Ni generate respectively rotations and boosts. The remaining algebraic relations,
involving the generators P0 and Pi of time- and space-translations, are
[Pi , P0 ] = 0,
[Pi , Pj ] = 0,
(2.2)
[Mi , P0 ] = 0,
[Mi , Pj ] = ij k Pk ,

2P0
1
1

[Ni , Pj ] = ij
1 e +
P P Pi Pj .
2
2
k
(2.3)
[Ni , P0 ] = Pi ,
(2.4)
The coalgebra is given by

Ni = Ni 1 + e
Mi = Mi 1 + 1 Mi ,
Pi = Pi 1 + e
P0 = P0 1 + 1 P0 ,
P0
P0
Ni +
1
ij k Pj Mk ,
Pi
(2.5)
(2.6)
and, to complete the Hopf algebra structure, the antipode map is

SMi = Mi ,
SP0 = P0 ,
SNi = e
SPi = e
P0
k
P0
k
Ni +
P0
1
ij k e Pj Mk ,
(2.7)
(2.8)
Pi .
These relations define the -Poincar Hopf algebra P . In the limit one recovers the
Poincar algebra, and the coproducts collapse to give the usual Leibnitz rule J = J 1 + 1 J .
P
3 One also often sees the modified bicrossproduct basis [8,35,36,38], which is defined by P (mod) = e 20 P (here) .
i
i
540
Single-particle states then, by definition, fall into irreducible representations of P . As usual

we can label states by their momentum eigenvalues:
P0 |p0 , p,
= p0 |p0 , p,
,
Pi |p0 , p,
= pi |p0 , p,
.
(2.9)
Here denotes any further quantum numbers needed to label the states (strictly, part of the
definition of single-particle states is that runs over a discrete set of indices only [23]). In this
paper we would like to focus on the simplest case, that of spinless particles; then, since any
purely internal degrees of freedom (commuting with P ) can be safely ignored, we may drop the
index from now on.
Let V denote the space of single-particle states, spanned by the |p0 , p.
Consider the infinitesimal boost (1 + n N ), where R is small and n n = 1. We compute its action on V in
the usual fashion, dropping terms of O( 2 ):
0 , p
Pi (1 + n N)|p

= (1 + n N )(1 n N )Pi (1 + n N )|p0 , p

= (1 + n N ) Pi + nj [Pi , Nj ] |p0 , p

2p
1

0
= pi ni
+

1e
p p + pi n p (1 + n N )|p0 , p
2
2
(2.10)
(2.11)
(2.12)
(2.13)
and similarly for P0 , from which we conclude that

2p0
1

(1 + n N )|p0 , p
=
p0 n p,
pi ni
1 e +
p p + pi n p
2
2
(2.14)
up to a possible normalization factor. In the same way, we may compute the action of infinitesimal
boosts on two-particle states in V V :
(1 Pi )(1 1 + n N )|r0 , r |s0 , s
(2.15)

= (1 1 + n N ) 1 Pi + nj [1 Pi , Nj ] |r0 , r |s0 , s
(2.16)

r0
2s0
r0

1
= si ni e
1 e +
s s + e si n s + ri n s ni rs
2
2

(1 1 + n N )|r0 , r |s0 , s,
(2.17)
and similarly for Pi 1, P0 1 and 1 P0 , yielding
(1 1 + n N )|r0 , r |s0 , s

2r

0
+
1e
r r + ri n r
=
r0 n r, ri ni
2
2

r0
r0
2s0

1
r0

s0 e ns , si ni e
+
1e
s s + e si n s
2
2
+ ri ns ni rs .
(2.18)
We choose to set the normalization factor to unity here.

As an aside, let us note that the antipode map is not used here, which is natural given that
there are no conjugate representations in sight. It is in fact true that the Hopf-algebraic notion of
541
the adjoint action of Ni on Pj agrees with the usual un-deformed one:

adNi Pj := SNj (1) Pi Nj (2) = [Pi , Nj ].
(2.19)
Consequently, and as one might expect, the total momentum p of an m-particle state transforms, under the action of a boost on the state, in the same fashion as does the vector P of
algebra generators under the adjoint action of the boost.4 But (2.19) holds only by virtue of the
specific form of the coproduct and antipode in P ; it is certainly not generic, and in particular adNj Pi = [Pi , Nj ]. Thus it is important to remember that the total momentum of a
boosted two-particle state is not equal to
ad11+ nN P
(2.23)
evaluated on the original state.

Our goal now is to find maps
:V V V V
(2.24)
that intertwine (i.e., commute with) the action of P :

J = J
for all J P ,
(2.25)
and that reduce in the limit to exchange of factors in the tensor product:
= : |r |s |s |r
(2.26)
(which is of course an intertwiner of classical Poincar, because the coproducts are cocommutative). As in [36], we will seek maps of the form

: |r0 , r |s0 , s
f0 (r, s), f(r, s)
g0 (r, s), g(r, s) .
(2.27)
This is clearly not the most general form conceivablewe could take a linear superposition
of basis states on the right-hand sidebut it will turn out to work. First let us demand that
commute with boosts. On the one hand we have, following the rule of (2.18),
(1 1 + n N ) |r0 , r |s0 , s

2f0
1

=
f0 n f, fi ni
1 e +
f f + fi n f
2
2

f0
f0
2g

1
g0 e n
g , gi ni e
1 e +
g g
2
2
(2.28)
4 To see this explicitly, note that ad

nN P evaluates to some power series in the generators P . Call this series A (P ).
m A(P ) = A(m P ), since is an algebra homomorphism. Now on a state | V m such that m P | = p | one
has
m m

P 1
+ n m N | = 1m + n m N m P + m P , m nN |
(2.20)
so the change in total momentum is given by

p | := m P , m n N | = m [P , n N ]|

= m ad nN P | = m A (P )| = A m P | = A (p)|
which establishes that p = A (p), i.e., that indeed p transforms like P .
(2.21)
(2.22)
542

f0
g ni f
e gi n g + fi n
g .
(2.29)
On the other hand, Taylor-expanding f and g to first order in , we have that

(1 1 + n N )|r0 , r |s0 , s

0 , (1 + n D)f
0 , (1 + n D)g
j
(1 + n D)g
j,
=
(1 + n D)f
where the differential operator Di is defined to be

2r
1

1
0
Di := ri
+
+ ri rj
1e
r r
r0
2
2
ri
rj

r0
r0
2s0

1
1 r0
e si
e
+ e si sj
1 e +
s s
s0
2
2
si
sj
1
1
+ rj si
r s .
sj
si
(2.30)
(2.31)
(2.32)
Comparing (2.29) and (2.31) one sees that commutes with the action of boosts if and only
if the following set of PDEs are satisfied:
Di f0 = fi ,

2f
1
1

0
f f + fi fj ,
+
1e
Di fj = ij
2
2
(2.33)
(2.34)
f0
Di g0 = gi e ,
(2.35)

f0
2g0
f0

1
1
1
1
Di gj = ij e
1 e +
g g + e gi gj + fj gi ij f g.
2
2
(2.36)
Of course we also require that commutes with spatial rotations. This should follow automatically from the equations above because every rotation generator can be written as a bracket of
boosts. It is a useful check to verify that this is so. First, one can compute
[Di , Dj ] = ri
rj
+ si
sj
rj
ri
sj
si
(2.37)
and recognize the right-hand side as the correct realization of the rotation generator ij k Mk on
functions of two momenta, given that the algebra and coalgebra of the Mi is entirely classical.
Second, it follows from (2.33)(2.36) that
[Di , Dj ]f0 = 0,
[Di , Dj ]fk = j k fi ik fj ,
(2.38)
[Di , Dj ]g0 = 0,
[Di , Dj ]gk = j k gi ik gj
(2.39)
so that indeed f0 and g0 are singlets of so(3) while f and g are vectors.
The final requirement for to be an intertwiner of P is that it should commute with the
translation generators P . In view of the coproduct (2.6), this is the case if and only if
f0 + g0 = r0 + s0 ,
fi + e
f0
gi = ri + e
(2.40)
r0
si ,
(2.41)
543
as previously noted in [36]. The crucial point now is that (2.33)(2.36) and (2.40)(2.41) are
a consistent set of equations. We have verified that, when g is eliminated in favour of f using
(2.40)(2.41), (2.35)(2.36) are equivalent to (2.33)(2.34).
The problem of finding intertwiners therefore reduces to finding solutions
f(r0 , r, s0 , s)
f0 (r0 , r, s0 , s),
(2.42)
of the partial differential equations (2.33)(2.34). In the following, we will adopt a hands-on
approach, and look for solutions order-by-order in 1 , which will lend insight into what extra
conditions must be given to specify a unique suitable solution.
3. Solution: Perturbative approach
The operators Di have well-defined expansions in inverse powers of ; we will assume that
f (r, s) and g(r, s) do too. The need to recover the correct classical intertwiner (2.26) in the
limit fixes the leading-order behaviour, and we write
1 (1)
1 (2)
1 (1)
1 (2)
fi = si + fi + 2 fi + .
f0 + 2 f0 + ,
At first order in 1 Eqs. (2.33)(2.34) are then

f0 = s0 +
(0) (1)
(1)
Di f0 = fj
(0) (1)
Di fj
(3.1)
+ r0 sj ,
(3.2)
(1)
(3.3)
= ij f0 ij (r0 s0 r s) + rj si ,
where
(0)
Di
= ri
r0
si
s0
r0
ri
s0
si
(3.4)
is the leading term in Di . f (1) must be quadratic in r, s and transform correctly under rotations.
This leaves only a few possibilities and, by inspection, one finds that there is a unique solution,
namely
(1)
f0 = r s,
fi(1) = ri s0 .
(3.5)
1
,
and in fact order-by-order as far as one

It is then possible to continue to second order in
wishes: at any given order m one will find a set of PDEs linear in f (m) , and with source terms,
all of degree m + 1 in the momenta, arising from the (by then known) action of D D (0) on f
up to order m . However, there is generally some freedom in this procedure, and this is first
seen at order 12 , where we have found that the general solution is
(2)
f0 = Ar0 (r r r0 r0 ) + Bs0 (s s s0 s0 ) + (1 C1 C2 )r0 r s + C1 s0 r r

+ C2 s0 r0 r0 (D1 + D2 )s0 r s + D1 r0 s s + D2 r0 s0 s0 ,
(2)
fi
= Ari (r r r0 r0 ) + Bsi (s s s0 s0 )
(3.6)

1
si (r r r0 r0 )
+ (1 C1 C2 )ri r s + (C1 + C2 )ri r0 s0 + C1
2

1
ri s s (1 + D1 )ri s0 s0
(1 + D1 + D2 )si r s + (D1 + D2 )si r0 s0 + D1 +
2
(3.7)
for arbitrary coefficients A, B, C1 , C2 , D1 , D2 .
544
We want to arrive at a unique intertwiner defining the notion of exchange of identical particles in -deformed theories, so we should use some other input to eliminate this freedom. One
obvious thing we can do, in the spirit of keeping as much of the usual structure as possible, is to
demand that 2 = 1. Given the definition of in (2.27), this means

f f (r, s), g(r, s) = r,
g f (r, s), g(r, s) = s,
(3.8)
which, combined with (2.40)(2.41), yields a new condition on f . Fortunately this condition is
satisfied to order 1 , where we have no freedom. At order 12 it turns out to produce
(2)
(2)
f0 (r, s) f0 (s, r) = s0 rr r0 s s,

(3.9)
1
1
(2)
(2)
fi (r, s) fi (s, r) = r0 r0 si s0 s0 ri + r sri r ssi + r0 s0 ri r0 s0 si ,
(3.10)
2
2
which reduces the number of free parameters in the solution above from 6 to 3 by forcing
A = B,
C1 = 1 + D 1 ,
C2 = D 2 .
(3.11)
There is a further important constraint on the form of . As in the usual undeformed case, the
full space V of spin zero single-particle states, spanned by the modes |p0 , p,
falls into a disjoint
union of irreps of P , labelled by their values of the mass Casimir

P0
P0 2
e P P
C = 2 sinh
(3.12)
P0 P0 P P = P P .
2
Thus far we have not been careful about labelling the irrep to which a given ket belongs, but it
now becomes important. States |r |s in the tensor product have three invariant labels, namely
C 1,
1C,
C
(3.13)
(giving, in the limit, the two masses r r and s s and the impact parameter (r +
s) (r + s) ). On physical grounds, it is reasonable to assume that identical particles, in the
sense we are seeking to define, must have equal masses, and that these masses are not altered
by the operation of exchanging them. That is to say, if we write C (p) for the value of the mass
Casimir evaluated on |p, i.e.,
C (p)|p = C |p,
(3.14)
then in (2.27) we should assume (or, in the case of f and g, demand) that
C (r) = C (s) = C (f ) = C (g).
(3.15)
(Note that of course commutes with C by construction and so automatically preserves the
third label in (3.13) above.)
Expanding order by order one finds that this condition (3.15) too is already satisfied to order 1
and produces a new constraint at order 12 :
1
1
1
f0(2) (r, s)s0 f(2) (r, s) s + (r s)2 r rs0 s0 (r s)(s s) r ss0 s0 = 0.
2
2
2
(3.16)
Somewhat remarkably, at least on the face of it, this constraint is compatible with the solution (3.6)(3.7) and relations (3.11). Equating the coefficients of the various terms in r, s gives
545
a large but highly redundant collection of equations, to which there is a single solution for the
remaining unknowns:
A = 0,
1
C1 = ,
2
C2 = 0.
(3.17)
Further, it may be verified that at this order the function g(r, s), determined by (2.40)(2.41),
also has the same value of the Casimir, i.e., that the final equality in (3.15) holds.
We have repeated the steps above at the next order, 13 , using M APLE to help with the rather
lengthy calculations. It turns out, even more strikingly than at second order, that the naively
overdetermined sets of equations which arise are in fact consistent and have exactly one solution.
This solution is
1
1
rs + 2 (s0 r r r0 s s + r0 r s + s0 r s)
2
1
+ 3 (r sr r + r ss s + 2r0 s0 r r 2r0 s0 r s 2r0 s0 s s) + , (3.18)
4
1
1
fi (r, s) = si + ri s0 + 2 (ri r s si r s + ri r0 s0 si r0 s0 ri s0 s0 )
2

1
2
+ 3 ri 2r0 rs r0 s s 2r0 s0 s0 + s0 r r 2s0 r s + s0 s0 s0
3
4

+ si (2r0 r s + r0 s s s0 r r) + ,
(3.19)
f0 (r, s) = s0 +
1
1
r s + 2 (r0 s s s0 r r r0 r s s0 r s)
2
1
3 (r sr r + r ss s + 2r0 s0 r r 2r0 s0 r s 2r0 s0 s s) + , (3.20)
4
1
1
gi (r, s) = ri si r0 + 2 (ri r s + si r s ri r0 s0 si r0 s0 + si r0 r0 )
2

1
2
+ 3 si 2s0 r s + s0 r r + 2s0 r0 r0 r0 s s 2r0 r s r0 r0 r0
3
4

+ ri (2s0 r s + s0 r r r0 s s) + .
(3.21)
g0 (r, s) = r0
To summarize, to order 13 these are the unique functions f and g such that the map in (2.27)
has the following properties:
(1)
(2)
(3)
(4)
is covariant, i.e., commutes with the action of P ;

reduces to the usual map flipping the factors in the limit;
squares to the identity;
preserves the masses of particles.
We will therefore take this to be the complete list of defining properties of the particle-exchange
map, and make the following
Conjecture. Such a map exists and is unique, to all orders in 1 .
546
We have no proof of this, but if it were false then the fact that it works in such an intricate way
at the first few orders would be a strange accident; in our view it is much more plausible that the
conjecture holds, to all orders, by virtue of some deeper argument that remains to be found.
4. Solution: Exact approach
In light of the discussion above, it is clearly important to solve Eqs. (2.33)(2.34) exactly, to
all orders in 1 . In this section we make a certain amount of progress in this direction, at least in
the simpler two-dimensional case, but unfortunately we will not be able to write solutions in a
form which gives us sufficient control over the condition (3) in the list above.
It follows from the condition C (f ) = C (r) in (3.15) that

2f0
f0
f0
1
1

1 e + f f =
C (r)e + e
(4.1)
2
2
2
which considerably simplifies the right-hand side of (2.34). There then exists a choice of new
variables
0 = e
f0
C (r)
,
2 2
(4.2)
1 f0
e fi ,
in which (2.33)(2.34) are linear and homogeneous:

i =
(4.3)
Di 0 = i ,
(4.4)
Di j = ij 0 .
(4.5)
This type of transformation has been seen before: it was shown in [24,25] that, by similar changes
of variables, all trace of the -deformation can be removed from the algebraic relations Poincar. (The deformation persists, of course, and can still be seen in the coalgebra, which
becomes very unwieldy.) And indeed (4.5) says nothing but that, under the action of the Lorentz
algebra on its parameters r and s, (r, s) transforms as a classical Lorentz four-vector:

g (r, s) = (g) (r, s)
(4.6)
where is the fundamental matrix representation.
We can make, similarly, a new choice of the independent variables, (r, s) (, )
C (r)
1 r0
,
i = e ri ,
2
2
s0
C (s)
1 s0
0 = e 1
,
i = e si ,
2 2
in terms of which the operators Di , (2.32), read
r0
0 = e 1
(4.7)
(4.8)
0
+ j i

i
.
C (r)
0
i
j
i
2 2
(4.9)
Were it not for the last two terms in the bracket and the overall coefficient of the latter, this expression would coincide exactly with the usual differential form of the undeformed two-particle
boost.
1
Di = i
0
+
0
i
0 + 1 +
547
An explicit solution of (4.4)(4.5) can be obtained in the two-dimensional case. In two dimensions there is only one boost operator, D, and its expression is given in (4.9) with i = j = 1.
The last two terms in the bracket cancel and we are left with a simpler expression which, up to
the overall coefficient of the bracket, looks a lot like the undeformed two-particle boost operator. A more convenient parameterization is then obtained by choosing a set of momentum space
coordinates adapted to the mass shell:
0 = m cosh ,
1 = m sinh ,
(4.10)
0 = m cosh ,
1 = m sinh .
(4.11)
We have chosen the same mass m for both sets of variables as we are considering identical
particles. The value of m is related to the common value of the Casimir by

C (r)
C (r)
m2 = 2 1 +
(4.12)
4 2
(this is most easily seen in the rest frame of, say, ). In terms of the rapidities, the differential
operator D reads
D=
m cosh + 1 +
C (r)
2 2
Eqs. (4.4)(4.5) thus reduce to

C (r)
1 +
= 0,
m cosh + 1 +
2 2
(4.13)
(4.14)
where we have set = 0 1 . These equations admit a set of solutions in closed form:

C (r)
1
= e F + 2 tanh
(4.15)
,
m1
tanh
2
2 2
parameterized by the arbitrary functions of one variable F . We have thus obtained the intertwiners in two dimensions exactly. However, applying the extra condition 2 = 1 results in
implicit equations for F that we have not been able to solve.
What appears to be needed is a reformulation of the definition (2.27) of the intertwiner
which includes the condition 2 = 1 in a natural way from the outset. We expect that this should
be possible by choosing a new parameterization of the space of two-particle states in which the
action of is simple, and we hope to return to this question in a future work.
Acknowledgements
C.Y. is grateful to the Leverhulme trust for financial support. R.Z. is supported by an EPSRC
postdoctoral fellowship.
References
[1] J. Lukierski, A. Nowicki, H. Ruegg, New quantum Poincar algebra and k deformed field theory, Phys. Lett. B 293
(1992) 344;
J. Lukierski, H. Ruegg, Quantum Poincar in any dimension, Phys. Lett. B 329 (1994) 189, hep-th/9310117.
[2] E. Celeghini, R. Giachetti, E. Sorace, M. Tarlini, J. Math. Phys. 31 (1990) 2548.
[3] E. Inn, E.P. Wigner, On the contraction of groups and their representations, Proc. Natl. Acad. Sci. 39 (1953) 510.
548
[4] S. Zakrzewski, Quantum Poincar group related to the -Poincar algebra, J. Phys. A: Math. Gen. 27 (1994) 2075
2082.
[5] M. Dimitrijevic, L. Jonke, L. Moller, E. Tsouchnika, J. Wess, M. Wohlgenannt, Field theory on -spacetime, Czech.
J. Phys. 54 (2004) 1243, hep-th/0407187;
M. Dimitrijevic, L. Jonke, L. Moller, E. Tsouchnika, J. Wess, M. Wohlgenannt, Deformed field theory on spacetime, Eur. Phys. J. C 31 (2003) 129, hep-th/0307149.
[6] G. Amelino-Camelia, Nature 418 (2002) 34, gr-qc/0207049.
[7] M. Daszkiewicz, K. Imilkowska, J. Kowalski-Glikman, S. Nowak, Scalar field theory on -Minkowski spacetime
and doubly special relativity, Int. J. Mod. Phys. A 20 (2005) 4925, hep-th/0410058.
[8] A. Agostini, G. Amelino-Camelia, M. Arzano, F. DAndrea, A cyclic integral on -Minkowski noncommutative
spacetime, Int. J. Mod. Phys. A 21 (2006) 3133;
A. Agostini, G. Amelino-Camelia, M. Arzano, A. Marciano, R.A. Tacchi, Generalizing the Noether theorem for
Hopf-algebra spacetime symmetries, Mod. Phys. Lett. A 22 (2007) 1779, hep-th/0607221;
A. Agostini, G. Amelino-Camelia, M. Arzano, F. DAndrea, Action functional for -Minkowski noncommutative
spacetime, hep-th/0407227;
A. Agostini, G. Amelino-Camelia, F. DAndrea, Hopf-algebra description of noncommutative-spacetime symmetries, Int. J. Mod. Phys. A 19 (2004) 5187, hep-th/0306013.
[9] J. Lukierski, A. Nowicki, Doubly special relativity versus -deformation of relativistic kinematics, Int. J. Mod.
Phys. A 18 (2003) 7, hep-th/0203065.
[10] J. Lukierski, Quantum deformations of Einsteins relativistic symmetries, AIP Conf. Proc. 861 (2006) 398, hepth/0604083.
[11] N. Beisert, V. Dippel, M. Staudacher, A novel long range spin chain and planar N = 4 super-YangMills, JHEP 0407
(2004) 075, hep-th/0405001;
M. Staudacher, The factorized S-matrix of CFT/AdS, JHEP 0505 (2005) 054, hep-th/0412188;
N. Beisert, The su(2|2) dynamic S-matrix, hep-th/0511082;
C. Gomez, R. Hernandez, Quantum deformed magnon kinematics, JHEP 0703 (2007) 108, hep-th/0701200;
C.A.S. Young, q-deformed supersymmetry and dynamic magnon representations, J. Phys. A 40 (2007) 9165, arXiv:
0704.2069 [hep-th].
[12] J. Lukierski, H. Ruegg, W.J. Zakrzewski, Classical quantum mechanics of free relativistic systems, Ann. Phys. 243
(1995) 90, hep-th/9312153.
[13] P. Kosinski, J. Lukierski, P. Maslanka, Local D = 4 field theory on -deformed Minkowski space, Phys. Rev. D 62
(2000) 025004, hep-th/9902037.
[14] H. Grosse, M. Wohlgenannt, On -deformation and UV/IR mixing, Nucl. Phys. B 748 (2006) 473, hep-th/0507030.
[15] S. Doplicher, K. Fredenhagen, J.E. Roberts, Spacetime quantization induced by classical gravity, Phys. Lett. B 331
(1994) 39;
S. Doplicher, K. Fredenhagen, J.E. Roberts, The quantum structure of spacetime at the Planck scale and quantum
fields, Commun. Math. Phys. 172 (1995) 187, hep-th/0303037.
[16] H.C. Kim, C. Rim, J.H. Yee, Blackbody radiation in -Minkowski spacetime, arXiv: 0705.4628 [hep-th].
[17] G. Amelino-Camelia, M. Arzano, Coproduct and star product in field theories on Lie-algebra non-commutative
spacetimes, Phys. Rev. D 65 (2002) 084044, hep-th/0105120.
[18] L. Frappat, A. Sciarrino, Phys. Lett. B 347 (1995) 28.
[19] L. Moller, A symmetry invariant integral on -deformed spacetime, JHEP 0512 (2005) 029, hep-th/0409128.
[20] P. Kosinski, P. Maslanka, J. Lukierski, A. Sitarz, Towards -deformed D = 4 relativistic field theory, Czech. J.
Phys. 48 (1998) 1407.
[21] S. Kresic-Juric, S. Meljanac, M. Stojic, Covariant realizations of -deformed space, Eur. Phys. J. C 51 (2007) 229,
hep-th/0702215;
S. Meljanac, M. Stojic, New realizations of Lie algebra -deformed Euclidean space, Eur. Phys. J. C 47 (2006) 531,
hep-th/0605133.
[22] M. Dimitrijevic, L. Moller, E. Tsouchnika, Derivatives, forms and vector fields on the -deformed Euclidean space,
J. Phys. A 37 (2004) 9749, hep-th/0404224.
[23] S. Weinberg, The Quantum Theory of Fields. Vol. 1: Foundations, Cambridge Univ. Press, 1995.
[24] P. Kosinski, J. Lukierski, P. Maslanka, J. Sobczyk, The classical basis for deformed Poincar (super)algebra and
the second deformed supersymmetric Casimir, Mod. Phys. Lett. A 10 (1995) 2599, hep-th/9412114.
[25] S. Giller, C. Gonera, M. Majewski, Deformation map for generalized -Poincar and -Weyl algebras, Acta Phys.
Pol. B 27 (1996) 2131.
[26] P. Kosinski, J. Lukierski, P. Maslanka, -deformed Wigner construction of relativistic wave functions and free fields
on -Minkowski space, Nucl. Phys. B (Proc. Suppl.) 102 (2001) 161, hep-th/0103127.
549
[27] J.G. Bu, H.C. Kim, Y. Lee, C.H. Vac, J.H. Yee, -deformed spacetime from twist, hep-th/0611175.
[28] S. Meljanac, A. Samsarov, M. Stojic, K.S. Gupta, -Minkowski spacetime and the star product realizations, arXiv:
0705.2471 [hep-th].
[29] R.J. Szabo, Quantum field theory on noncommutative spaces, Phys. Rep. 378 (2003) 207, hep-th/0109162.
[30] M.R. Douglas, N.A. Nekrasov, Noncommutative field theory, Rev. Mod. Phys. 73 (2001) 977, hep-th/0106048.
[31] C.S. Chu, P.M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hep-th/9812219;
N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[32] P. Aschieri, M. Dimitrijevic, F. Meyer, J. Wess, Noncommutative geometry and gravity, Class. Quantum Grav. 23
(2006) 1883, hep-th/0510059.
[33] S. Majid, Foundations of Quantum Group Theory, Cambridge Univ. Press, 1995.
[34] Y. Abe, Noncommutative quantization for noncommutative field theory, Int. J. Mod. Phys. A 22 (2007) 1181, hepth/0606183;
P.P. Kulish, Twists of quantum groups and noncommutative field theory, hep-th/0606056;
G. Fiore, J. Wess, On full twisted Poincar symmetry and QFT on MoyalWeyl spaces, Phys. Rev. D 75 (2007)
105022, hep-th/0701078;
G. Fiore, Can QFT on MoyalWeyl spaces look as on commutative ones?, arXiv: 0705.1120 [hep-th];
R. Oeckl, Untwisting noncommutative R d and the equivalence of quantum field theories, Nucl. Phys. B 581 (2000)
559, hep-th/0003018;
A.P. Balachandran, A. Pinzul, B.A. Qureshi, Twisted Poincar invariant quantum field theories, arXiv: 0708.1779
[hep-th];
J.G. Bu, H.C. Kim, Y. Lee, C.H. Vac, J.H. Yee, Noncommutative field theory from twisted Fock space, Phys. Rev.
D 73 (2006) 125001, hep-th/0603251;
M. Chaichian, A. Tureanu, G. Zet, Twist as a symmetry principle and the noncommutative gauge theory formulation,
Phys. Lett. B 651 (2007) 319, hep-th/0607179;
P. Aschieri, F. Lizzi, P. Vitale, Twisting all the way: From classical mechanics to quantum fields, arXiv: 0708.3002
[hep-th];
M. Riccardi, R.J. Szabo, Duality and braiding in twisted quantum field theory, arXiv: 0711.1525 [hep-th].
[35] M. Arzano, A. Marciano, Fock space, quantum fields and -Poincar symmetries, arXiv: 0707.1329 [hep-th].
[36] M. Daszkiewicz, J. Lukierski, M. Woronowicz, -deformed statistics and classical fourmomentum addition law,
hep-th/0703200.
[37] M. Daszkiewicz, J. Lukierski, M. Woronowicz, Towards quantum noncommutative -deformed field theory, arXiv:
0708.1561 [hep-th].
[38] A. Agostini, F. Lizzi, A. Zampini, Generalized Weyl systems and -Minkowski space, Mod. Phys. Lett. A 17 (2002)
2105, hep-th/0209174.
[39] S. Majid, H. Ruegg, Bicrossproduct structure of Poincar group and noncommutative geometry, Phys. Lett. B 334
(1994) 348, hep-th/9405107.

Nucl - Phys.B v.797 PDF

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Nucl - Phys.B v.797 PDF

Diunggah oleh

Hak Cipta:

Format Tersedia

Nuclear Physics B 797 (2008) 126

KaluzaKlein excitations of gauge bosons at the LHC

Received 27 July 2007; accepted 18 December 2007

E-mail address: moreau@th.u-psud.fr (G. Moreau).

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

1 The severe indirect bounds on M

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

(Z hadrons), are correctly reproduced with

For instance, if the coupling

2 24]. For this c

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

VKK 2  3(f2 1) VKK VKK

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

where mf is the mass of the fermion f , f =

3. Top and bottom quark pair production

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

RS 30 for a moderate luminosity and

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

(cos > 0) (cos < 0)

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

due to EW KK gauge bosons is expected to exceed the statistical uncertainties on ARS

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

4.2. Associated production with heavy quarks

gg/q q Q Q  VKK Q Q  QQ,

stand for the final state heavy

pTt,t  300 GeV,

pTb,b  100 GeV,

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

RS (pp t tb/t tb)

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

A. Djouadi et al. / Nuclear Physics B 797 (2008) 126

[14] A. Djouadi, G. Moreau, F. Richard, Nucl. Phys. B 773 (2007) 43.

Nuclear Physics B 797 (2008) 2777

Electroweak corrections to hadronic production

Received 3 August 2007; accepted 14 December 2007

E-mail address: schulze@particle.uni-karlsruhe.de (M. Schulze).

J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777

J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777

J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777

2. Definitions and conventions

production, and similarly for

J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777

The pT -distribution for the unpolarized partonic subprocess ab W k reads

For the 2 3 subprocess ab W k we define the following five independent invariants

and the four dependent invariants

The pT -distribution for this subprocess reads

= Nab d3 MabW k  FO,3 (3 ),

J.H. Khn et al. / Nuclear Physics B 797 (2008) 2777

(pT,q + pT, )2 , otherwise pT,j = pT,q . We note that, in

VKK 2 3(f2 1) VKK VKK

gg/q q Q Q VKK Q Q QQ,

pTt,t 300 GeV,

pTb,b 100 GeV,

= Nab d3 MabW k FO,3 (3 ),

(pT,q + pT, )2 , otherwise pT,j = pT,q . We note that, in