Nucl - Phys.B v.710

Nuclear Physics B 710 (2005) 357
Supersymmetric standard model spectra

from RCFT orientifolds
T.P.T. Dijkstra, L.R. Huiszoon, A.N. Schellekens
NIKHEF Theory Group, Kruislaan 409, 1098 SJ Amsterdam, The Netherlands
Received 16 December 2004; accepted 22 December 2004
Available online 12 January 2005
Abstract
We present supersymmetric, tadpole-free d = 4, N = 1 orientifold vacua with a three family chiral
fermion spectrum that is identical to that of the standard model. Starting with all simple current
orientifolds of all Gepner models we perform a systematic search for such spectra. We consider
several variations of the standard four-stack intersecting brane realization of the standard model, with
all quarks and leptons realized as bifundamentals and perturbatively exact baryon and lepton number
symmetries, and with a U (1)Y vector boson that does not acquire a mass from GreenSchwarz terms.
The number of supersymmetric Higgs pairs H1 + H2 is left free. In order to cancel all tadpoles,
we allow a hidden gauge group, which must be chirally decoupled from the standard model. We
also allow for non-chiral mirror-pairs of quarks and leptons, non-chiral exotics and (possibly chiral)
hidden, standard model singlet matter, as well as a massless BL vector boson. All of these less
desirable features are absent in some cases, although not simultaneously. In particular, we found cases
with massless ChanPaton gauge bosons generating nothing more than SU(3) SU(2) U (1). We
obtain almost 180 000 rationally distinct solutions (not counting hidden sector degrees of freedom),
and present distributions of various quantities. We analyse the tree level gauge couplings, and find a
large range of values, remarkably centered around the unification point.
2004 Elsevier B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Mj; 11.25.Wx; 12.60.-i
Keywords: String theory; Orientifolds of Gepner models; Standard model; Supersymmetry; RCFT
E-mail addresses: tdykstra@nikhef.nl (T.P.T. Dijkstra), lennaert_janita@hotmail.com (L.R. Huiszoon),

t58@nikhef.nl (A.N. Schellekens).
0550-3213/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2004.12.032
T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357
1. Introduction
String theory is hoped to be a consistent theory of quantum gravity, with the special
feature that it strongly constrains the matter it can couple to. Although direct experimental
tests of new predictions seem out of reach for the moment, it can at least be tested theoretically by verifying its internal consistency, in particular, with regard to gravity, and by
checking that the limited set of matter it can couple to includes the standard model. We
may be lucky enough that the way the standard model is embedded in string theory implies
predictions for future experiments, but it may also happen that using all known experimental constraints we are still left with more than one, or even a huge number of possibilities.
But at present it is still a serious challenge to find even one string vacuum (which may
actually be a metastable, approximate ground state) that is a credible standard model candidate. To find such a vacuum requires an in-depth analysis of known candidates based on
robust criteria derived from experiment. However, even within the known classes of vacua,
large areas have remained unexplored so far. As longs as that is the case, nature would
have to be very kind to us to allow us to find the ground state on which our universe is
based. In this paper we want to make a modest step towards broadening the set of accessible vacua, by means of a systematic exploration of orientifolds of Gepner models. It turns
out that this class is very rich, and includes an abundance of standard model-like spectra far
beyond anything that has come out of string theory so far. Some early, and partial results
were reported in [1].
Since 1984 there have been several attempts to search for standard model like string
vacua. Two general classes may be distinguished: heterotic string constructions (either
(2, 2) or (0, 2)), with the standard model particles realized as closed strings, and type-I
(orientifold, intersecting brane) constructions, characterized by open string realizations of
the standard model. Other possibilities certainly exist (e.g., M-theory compactifications),
and since new ways of obtaining the standard model are discovered about every five to ten
years, it might be an illusion to think that we are close to a complete picture.
To explore the two classes mentioned above, we have essentially three methods at our
disposal: free CFT constructions (free bosons and/or fermions and orbifolds thereof), geometric compactification (in particular CalabiYau compactifications) and interacting CFT
constructions. All three have been applied successfully to heterotic string construction. The
earliest example [2] of a three-family model was based on the TianYau CalabiYau manifold with Euler number 6 and a non-trivial homotopy group 1 (as far as we know still the
only known manifold of this kind), used for Wilson line symmetry breaking. About a year
later, the first orbifold examples were found [3,4] (see also [57] for later developments). In
[8] a three-family model was constructed using tensor products of N = 2 minimal models,
corresponding to a point in the moduli space of the TianYau manifold. This model has an
E6 gauge group. By reducing the (2, 2) world-sheet symmetry to (0, 2) a large number of
related models was constructed in [9] with E6 or SO(10) gauge symmetry. The symmetry
in the bosonic sector can be reduced further to obtain many models with a gauge group
SU(3) SU(2) U (1) [Other factors] and three families of quarks and leptons [10].
Another class of three family models was obtained using free fermions in [1113], and
extended in [14]. By means of a modified CalabiYau construction yielding (0, 2) spectra,
several more three-family models were obtained in [15]. All the foregoing models lead
to level 1 realizations of the standard model gauge groups SU(3) and SU(2), and hence
their spectra contain fractionally charged states, or they have a non-standard normalization
for the standard model U (1) generator (or both) [1618]. Unified models with standard
model charge quantization can be constructed by using higher level KacMoody algebras,
and indeed three family models were obtained [19]. Yet another approach to getting the
standard model, based on strongly coupled heterotic strings, was presented in [20]. All of
these heterotic three family models are supersymmetric. Constructing non-supersymmetric
heterotic strings is much easier, but those theories are not stable, in general.
The open string road towards the standard model has been considerably more difficult.
Until about 1995 this option was rarely taken seriously, with a few exceptions [21,22].
This changed after the discovery of D-branes [23] and the observation in [24] that the
problematic relation between the unification scale and the Planck scale could be avoided
in open string theories. The construction of realistic theories is complicated by several new
features, on top of those of the closed type-II theory to which the orientifold procedure
is applied: boundary and crosscap states and the requirement of tadpole cancellation. The
first four-dimensional chiral, supersymmetric theory was constructed in [25], but it was still
far from realistic. In order to get standard model-like spectra, in several later papers (e.g.,
[2634]) the requirement of supersymmetry was relaxed, and only part of the tadpoles
were cancelled, namely the RR tadpoles needed for consistency. The resulting theories
are not stable, but otherwise consistent. The first semi-realistic supersymmetric theory that
satisfies all tadpole conditions was presented in [35] and [36,37]. However, these spectra
contains chiral exotics, in addition to the standard model representations. In [38,39] a three
family supersymmetric PatiSalam (SU(4) SU(2) SU(2)) model was constructed, but
the two SU(2)s do not emerge directly from string theory but from a diagonal subgroup
of ChanPaton groups using a brane recombination mechanism. In [40] supersymmetric
PatiSalam models were found that emerge directly from the CP-groups, but with additional chiral exotic matter, and in [41] supersymmetric SU(5) GUT models with chiral
exotics were presented. Most of the foregoing constructions are based on orientifolds of
toroidal orbifolds [42,43], except for [31] which uses the quintic CalabiYau manifold. In
[44] standard model like spectra where obtained using branes at singularities. The first investigation of string spectra from open strings of orientifolds of Gepner models was done
in [45], in 6 dimensions. The first analysis in four dimensions was done by [46]. These
authors did not find chiral spectra, but got a first glimpse of the vast landscape of solutions.
Further work on this kind of construction, including some chiral spectra, was presented in
[4652]. In [1], the precursor of the present paper, the first examples were found of supersymmetric standard model spectra, with the standard model group appearing directly as
a ChanPaton group and without chiral exotics. Shortly thereafter the first such example
was found in the context of orientifolds of toroidal orbifolds [53]. In [54,55] the first semirealistic examples were found with flux compactifications and partially stabilized moduli,
although with chiral exotics. For recent results on chiral fermions from Gepner orientifolds
see [52]. For recent reviews with more references consult for example [5659].
Recently there has been a lot of work on (non)-supersymmetric vacua with moduli
stabilized by three-form fluxes on a CalabiYau manifold. Although this is an extremely
interesting development, we will not consider it here, simply because the methods we use
do not allow this at present.
For a variety of reasons we will require the string spectrum to be supersymmetric.

The first reason is phenomenological. Although we do not commit ourselves to a supersymmetry breaking mechanism or scale, the most obvious scenario is the standard one:
supersymmetry breaking at a few TeV, induced by gaugino condensation in a hidden sector
(which exists in most of our models), with supersymmetry playing the role of protecting
the gauge hierarchy. Indeed, such a hierarchy inevitably exist in these models, since six
dimensions are compactified on a CalabiYau manifold at a rational point in moduli space,
and hence there is no reason to expect some of the compactified dimensions to be extremely
large. There are two other reasons why we want to keep supersymmetry unbroken. First
of all, we can then be certain that the four-dimensional strings we construct are stable and
consistent. But the most important reason is a practical one. Spacetime supersymmetry
has the effect of extending the world-sheet chiral algebra, thereby organizing the fields
into a smaller number of primaries. This is what makes our computations manageable in
practice.
The use of rational conformal field theory (RCFT) in these constructions has wellknown advantages and disadvantages. The advantage of the algebraic approach is that we
can explore a large class of models with uniform methods. But clearly the disadvantage
is that one ends up in a special point in moduli space, both with regard to the CalabiYau
manifold, as well as the choice of branes wrapping it. It is not reasonable to expect such a
ground state to be exactly the standard model, because many observable quantities, such as
quark and lepton masses and gauge couplings will depend on the moduli, which have been
fixed at a specific value. Clearly we should focus on those features that do not depend on
the moduli. The primary feature to consider is then of course the chiral spectrum.
Apart from supersymmetry breaking there are several other important issues that we do
not consider here, such as standard model symmetry breaking, Yukawa couplings, Majorana masses for the neutrinos, etc. We see our results therefore mainly as a first exploration
of some interesting regions in the huge landscape of possible models. Once a promising
type of model has been identified, one may try to explore it in more detail, either by CFT
perturbations in the neighborhood of the special point, or by constructing the corresponding
CalabiYau and the set of branes on it, using the RCFT results as a guideline. Even within
the context of RCFT one could push these models further and compute certain couplings,
but unfortunately the required CFT techniques are not yet available for all couplings. For
example, three-point couplings between open strings are in principle computable in RCFT,
but to develop this formalism to include non-trivial modular invariant partition functions
and simple current fixed points would still require a substantial amount of work. Gauge
couplings, on the other hand, are easily computable, and we do so in all cases.
Since we end up with a very large set of solutions, our results should give a reasonable idea of what kind of spectra one may expect, and one can perform some statistical
analysis on this set, somewhat similar in spirit (though with a quite different philosophy)
to the approach presented in [60]. In addition, despite the inherent limitations of the algebraic approach, one may explore the effect of brane moduli as well as some CalabiYau
moduli. The former, since for a given CY-manifold we typically find a large number of
spectra, which can be interpreted in terms of branes in different discrete positions; the latter, because some distinct Gepner points may lie on the same CalabiYau moduli space.
Especially the brane position moduli seem to be probed rather effectively by rational points,
in certain cases.
1.1. Brane configurations considered
In this paper we consider boundary states in a rational type IIB CFT. The relevant physical open string quantities are annulus and Mbius coefficients. For the sake of clarity, it is
convenient to present the models in terms of an intersecting brane picture, although such a
picture is not really used in our construction. This picture would be appropriate in a large
volume limit and for type IIA string, the mirror of what we consider here. The geometric
interpretation of the construction considered here is presumably in terms of magnetized D3
and
branes [26]. In the following, by the intersection of two branes a and b we mean
D7
i i (/2)| , where Ai
A
0
ab
ab is an annulus coefficient and is the character restricted to
i
massless states. By the chiral intersection we mean the same quantity restricted to chiral
states.
We consider here a specific type of intersecting brane models, based on a four-stack
configuration with a baryon brane, a weak brane (or left brane), a right brane and a lepton
brane, labeled a, b, c, d, respectively [28]. These are the minimal brane configurations with
baryon and lepton number conservation and all quarks and leptons realized as bifundamentals. The ChanPaton gauge groups associated with these branes contain the standard
model gauge group. In addition we allow hidden branes, with gauge groups with respect
to which all standard model particles are neutral. These branes were introduced to cancel
massless tadpoles, but their gauge groups may play a useful phenomenological role, in
particular, for gaugino condensation.
The a and d branes are required to be complex, and carry ChanPaton group U (3)a and
U (1)d , respectively. The former contains the standard model gauge group SU(3). The overall phase factor of U (3)a corresponds to baryon number, and the U (1)d to lepton number.
In the standard model these symmetries are not gauged, and anomalous. In string theory
these anomalies are canceled by GreenSchwarz terms, involving a bilinear coupling of the
bary-photon and lepto-photon to massless two-form fields. These couplings produce a
mass for the linear combination B + L of these U (1) bosons, breaking the corresponding
combination of baryon and lepton number. Nevertheless the broken symmetry still prevents the appearance of dangerous baryon and lepton number violating couplings at least
perturbatively [28].
The b and c branes may be real or complex. In the standard four-stack realization of
the standard model the first family emerges as follows if they are both complex, with CPgroups U (2)b and U (1)c , respectively:
(u, d):
[a, b] or [a, b ],
uc :
dc:
(e , ):
e+ :
[a , c],
[a , c ],
[b, d] or [b , d],
[c, d ],
c:
[c , d ].
Here [x, y] denote strings beginning on brane x and ending on brane y, and x is the brane
conjugate to x. The Y -charge of the standard model is given by the linear combination
1
1
1
Y = Qa Qc Qd .
6
2
2
The overall phase symmetry in U (2)b is always anomalous with respect to the a and the
d branes and the corresponding gauge boson acquires a mass; the surviving gauge group
is SU(2)W . Note that B + L and U (2)b have independent anomalies with respect to the
standard model, so that there is no linear combination of these phases symmetries that
remains unbroken.
The standard model weak gauge group can also be constructed out of real branes on
top of an orientifold plane, yielding Sp(2). Since the spectrum is real with respect to the
c-brane, we may allow the group O(2)c here instead of U (1)c , with Qc replaced by the
(properly normalized) O(2) generator in the definition of the charge. Since O(2) branes
differ from Sp(2) branes only by a Mbius sign, we decided to allow the latter as well.
Strictly speaking this is a departure from our philosophy of looking only for the simplest
standard model realizations, but these models are as easy to look for as O(2) models, and
have the interesting feature of yielding leftright symmetric models with an SU(2)L
SU(2)R gauge group. Such gauge groups appear as part of the symmetry-breaking chain
of the PatiSalam model, and indeed in some examples there are related spectra with the d
brane on top of the a brane, yielding precisely the PatiSalam model. Then we end up with
following six types of models:
Type 0:
Type 1:
Type 2:
Type 3:
U (3) Sp(2) U (1) U (1),

U (3) U (2) U (1) U (1),
U (3) Sp(2) O(2) U (1),
U (3) U (2) O(2) U (1),
Type 4: U (3) Sp(2) Sp(2) U (1),

Type 5: U (3) U (2) Sp(2) U (1).
The complete spectrum of these theories can contain massless vector bosons from three
sources: the standard model part of the open sector, as listed above, additional hidden
branes, and the closed sector. The latter gauge bosons are nearly inevitable, but do not have
minimal couplings to the quarks and leptons. The gauge bosons from the hidden sector
may be absent altogether, but in any case do not couple to the standard model particles. If
we ignore these two kinds of vector bosons, the gauge group is quite close to that of the
standard model. As a Lie-algebra, it is a non-abelian extension of the standard model only
for types 4 and 5; in all other cases we get the standard model with at most one additional
non-anomalous U (1) factor, B L.
The gauge bosons coupling to the non-anomalous symmetries Y and B L can acquire
a mass from GreenSchwarz type two-point couplings to two-form fields, provided that
these couplings do not generate a contribution to the anomaly. We find that in most cases
such a mass contribution is absent, and that Y is more likely to acquire a mass than B L.
The latter statement is based only on the models presented in [1], where masslessness of Y
was only used as an a posteriori check. In the present paper brane stacks yielding non-zero
Y -mass were eliminated before attempting to solve the tadpole equations, but no condition
was imposed on the B L mass. We found a massive B L photon in about 3% of the
type-0 models, and for none of the type-1 models (of which we found only very few). There
are even models with a massive B L and no extra branes at all. These models (22 in total)
have precisely the standard model gauge group from the open sector, but there are still 18
additional RamondRamond vector bosons from the closed sector.
We expect masslessness of the Y-boson, in addition to lepton and quark chirality and
supersymmetry, to be among the features of these models that are unaffected by generic
perturbations around the rational point. The potential origin of a Y-boson mass would be
the generation of a two-point coupling to one of the RR two-forms away from the rational point. However, such a two-point coupling would very likely generate an anomalous
contribution in combination with a three-point coupling of the same RR-two form to two
gauge bosons.
The same logic applies to the B L gauge boson. If it is massless at the string level,
it should acquire a mass trough a fundamental or dynamical Higgs mechanism, just as the
Z and W bosons. Candidates for the required Higgses are the sneutrinos or two standard
model/hidden sector bifundamentals (one ending on the c-brane and one on the d-brane),
bound by a hidden sector gauge group.
Most other features cannot be expected to survive generic perturbations. In particular,
this concerns the massless particles that are non-chiral with respect to GS , namely the
mirrors, rank-2 tensors, leptoquarks and standard model/hidden sector bifundamentals. It
seems plausible that for many of them masslessness is an artifact of being in a rational point
in moduli space. They will get a mass when the moduli are changed, and one can investigate this by computing the coupling of the closed string moduli to the massless fermions.
However, some of the massless particles correspond to brane position moduli, and hence
to flat directions in the superpotential. One of the problems we have in investigating this
and other more detailed phenomenological issues is that we have to find a way to do a
meaningful analysis for a huge number of solutions.
There are many other brane configurations that would yield standard-model-like spectra.
For example, another attractive possibility might be to start with a U (5) stack (realizing
an SU(5) GUT) plus other branes or the reduction of such a stack to U (3) U (2). In
such models some quarks and leptons emerge as anti-symmetric tensors and baryon and
lepton number are not symmetries. Models of this type were studied in [41], but so far
with the disappointing result that there are additional chiral symmetric tensors of SU(5).
In principle, one could search for models of this kind in exactly the same way, just as
one could search for PatiSalam type models. It goes without saying that if we were to
relax some constraints and allow, for example, chiral exotics or diagonal embeddings of
standard model factors in the CP group, then the number of solution would almost certainly
explode. Nevertheless, such options, although unattractive, are not necessarily ruled out
experimentally.
1.2. Chirality
Let us explain more precisely what we mean by getting the standard model from
string theory, since this issue tends to cause confusion. Quite generally, one might ac-
10
cept any string spectrum if the gauge group GS emerging directly from open or closed
strings contains GSM = SU(3) SU(2) U (1), and if the fermion representation of GS
reduces to three times (3, 2, 16 ) + (3 , 1, 13 ) + (3 , 1, 23 ) + (1, 2, 12 ) + (1, 1, 1), written
in terms of left-handed Weyl fermions. These should be the only fermions that are chiral
with respect SU(3) SU(2) U (1). If at this stage there were additional chiral particles
one could still imagine mechanisms that give them a sufficiently large mass after standard model symmetry breaking and SU(3) confinement, but if that were true we simply
need more experimental input to go ahead. There may be additional massless fermions
that are non-chiral with respect to SU(3) SU(2) U (1) and their may be additional
fermions, chiral with respect to GS , that become non-chiral after the reduction from GS to
SU(3) SU(2) U (1).
The group GS is the complete CP-group from the open sector times any gauge group
generated by closed sector vector bosons. The physical realization of the group theoretical
reduction mentioned above can take many forms, such as mass generation for U (1)s
by GreenSchwarz anomaly cancelling terms, confinement or breaking by a fundamental
or dynamical Higgs effect. Furthermore, part of GS may remain unbroken, if the corresponding gauge bosons do not couple to quarks and leptons. We did not commit ourselves
here to a particular reduction mechanism. In the majority of the spectra we consider, GS
is embedded in GSM as GS = GSM GHidden (as a Lie-algebra), except in types 4 and 5,
where the U (1)c group is non-trivially extended to Sp(2). In the latter cases an additional
Higgs-like mechanism would be required to arrive at the standard model. Just as is the case
with the supersymmetry breaking and the standard model Higgs mechanism, one could
impose additional constraints on the results in order for a particular mechanism to be realized, but such constraints are less robust and more model-dependent than the requirement
of chirality, our main guiding principle.
The possibilities for GS -chiral particles that are GSM -non-chiral are the following:
(1) Right-handed neutrinos. These particles are singlets (and hence not chiral) w.r.t. GSM
but are chiral with respect to lepton number, which is broken. In our case, there are
always three of them. This is an inevitable consequence of tadpole cancellation, which
cancels the cubic anomalies in U (1)d , plus the fact that we only allow bifundamentals
of the (a, b, c, d) branes to be chiral.
(2) Higgsinos. In the MSSM the fermionic partners of the Higgs are non-chiral with respect to GSM , but in models of types 1, 3 and 5 there is a possibility for the Higgses to
be chiral with respect to U (2)b . This gauge symmetry breaks to SU(2) in the first step,
but its initial presence can still forbid the generation of large masses for the Higgs.
This is a desirable feature, as it may give a mechanism for getting light Higgs bosons.
(3) Mirror quarks and leptons, which are chiral with respect to U (2)b . These particles
can appear for the same reason as the Higgsinos, but are less desirable. For example,
the U (3) U (2) combinations (3, 2) + (3 , 2) is chiral, but becomes non-chiral when
U (2)b is reduced to SU(2). We have allowed such particles in principle, but (just as
the chiral Higgsinos), they occur only rarely.
(4) GSM singlets, which are chiral with respect to the hidden gauge group. Such particles
couple only the SM-matter gravitationally, and hence are acceptable as dark mat-
11
ter, if not too abundant. Furthermore, they may acquire a mass and/or be confined by
GHidden dynamics.
Unwanted chiral matter within the standard model sector can be avoided by the selection
of a, b, c and d branes, and chiral matter from open strings stretching between the SM
branes and the hidden branes can be avoided by appropriate selection of the latter. One
could in principle also forbid chiral rank-2 tensors within GHidden by an a priori constraint,
but it is very hard to forbid chiral bifundamentals within GHidden , except by constructing
all solutions and eliminating them a posteriori. Since constructing all solutions is nearly
impossible in most cases, we decided to allow GSM -singlets that are chiral with respect to
GHidden . Nevertheless, they occur in only a small fraction (about 12.5%) of our solutions.
This is largely due to the fact that our search is biased in favor of few additional branes, and
it is harder to make chiral spectra with fewer branes, and impossible with a single brane.
1.3. Scope of the search
In this paper we consider all modular invariant partition functions (MIPFs) that are symmetric simple current modifications of the charge conjugation invariant of all 168 minimal
N = 2 tensor products. The precise number of such MIPFs is determined as follows. Generically, it is just a matter of applying the procedure of [61,62] and restricting to symmetric
bi-homomorphisms X (defined more precisely in [62] and in the next chapter). However,
if there are identical N = 2 factors in the tensor product, there will be equivalences among
these MIPFs, and we remove equivalent ones. Furthermore, for small values of k (the level of the minimal model), especially k = 2, generically distinct simple current invariants
are in fact identical, and these are also removed from the set. We then end up with 5403
distinct MIPFs. They can be characterized in part by the resulting Hodge numbers h11 and
h21 , and by the number of gauge singlets in the heterotic string spectrum. These numbers
can be compared to tables of such spectra produced about fifteen years ago [9,63]. Unfortunately, a complete comparison is difficult, because the old results are either no longer
available, and certainly not in electronic form, or the search was not fully exhaustive, or the
symmetry of the MIPFs was not specified. But to the extent that a comparison is possible
the results seem to agree.
In a few cases there appear to be further equivalences, or at least some MIPFs may
correspond to distinct rational points on the same CalabiYau space. In total we found 873
different combinations of Hodge numbers, and 1829 Hodge numbers plus singlets (i.e.,
gauge singlets in the heterotic spectrum.)
For each of these MIPFs we consider all orientifolds allowed by the general formula
of [64]. These orientifolds are subject to three equivalence relations, one originating from
permutations of identical factors, and two as part of the general formalism. These equivalences are removed, and we then end up with a total of 49322 a priori distinct orientifolds.
Our results indicate that indeed they are generically distinct.
For each MIPF and orientifold we consider the complete set of boundaries. This means
that the number of boundaries is equal to the number of Ishibashi states of the MIPF.
The MIPF may be of any type: pure fusion rule automorphisms, extensions of the chiral
algebra, or combinations thereof. The beauty of the formalism of [64] is that it works inde-
12
pendently of such details. It is well-known that in the case of extensions one can distinguish
boundaries that respect the extended symmetry and boundaries that do not. A complete set
of boundaries contains both kinds. The CFT we start with is itself an extension of the
minimal model tensor product CFT (by alignment currents and the spacetime supersymmetry current). Those extensions are respected by all our solutions, by construction. If the
MIPF extends the chiral algebra further, one could work directly in the extended CFT and
only consider boundaries that respect the extension. One would then find a subset of our
solutions. Alternatively, one could also start with less symmetry, and treat, for example,
spacetime supersymmetry as a bulk extension. This would allow, in principle, supersymmetry breaking boundaries. In practice this is quite hard, because the number of primary
fields increases dramatically. Undoubtedly, so will the number solutions.
Our goal was to complete this analysis for all MIPFs, in order to arrive at a picture
that is as complete as possible. Unfortunately, the analysis could not be completed in all
cases. Two tensor combinations had too many primaries to finish the computation of chiral
intersections in a reasonable amount of time. In five others the number of candidate (a, b, c,
d) branes was so large that we decided to restrict ourselves to types 0 and 1, of which there
are fewer. For a given MIPF, orientifold and type, the number of four-stack candidates was
more than a million in some cases.
The main computational stumbling block are the tadpole equations. For every valid set
of (a, b, c, d) branes, the number of variables is equal to the number of boundaries that do
not have a chiral intersection with a, b, c and d. This number can become as large as several
hundreds, for a few tens of tadpole equations. Obviously, the time needed to evaluate this
completely grows exponentially with the surplus of variables over equations. This means
that beyond a certain number of variables it is impossible to decide conclusively that there
are no solutions. In those cases we did perform a systematic search for solutions with 0,1
and 2 hidden branes, and 3 if the number of variables was less than 100, 4 if the number
was less than 400. Furthermore, in the simpler cases we attempted solving the equations in
general. Of the 5403 MIPFs, 2 were not analysed at all, and 20 only for types 0 and 1, and
in 495 cases the tadpole equations were not fully analysed (153 of these did have solutions,
however).
There are several possible ways to count solutions as distinct. On the one hand, one
could regard solutions as identical if they are connected to each other by continuous, nonsingular variations of open or closed string moduli. In an RCFT approach this is hard to do,
since we cannot vary the moduli in a continuous way. The other extreme would be to count
all distinct massless spectra, including the hidden sector gauge groups and representations.
This is also not possible in our case, since we did not do a systematic search through
all hidden sector gauge groups. We have chosen an intermediate criterium: solutions are
regarded as distinct if they are of different type or have a different massless (chiral and
non-chiral) standard model spectrum. Furthermore, we treat different dilaton couplings for
the a, b, c, d branes and the O-plane as a distinction, and the absence or presence of a
hidden sector. By contrast, in [1] hidden sector distinctions were also counted. The number
of solutions for one of the MIPFs of the tensor product (6, 6, 6, 6) quoted in that paper
(more than 6000) reduces to 820 with our present way of counting.
13
1.4. Contents
This paper is organized as follows. In the next chapter, we review the ingredients of the
algebraic orientifold construction. Section 3 contains a general discussion of the massless
spectrum. In Section 4 we discuss tadpole and anomaly cancellation. Section 5 contains
our results. Due to the huge number of solutions it is impossible and pointless to present
detailed spectra. Therefore, we only give distributions of several quantities of interest, such
as the number of Higgs scalars. We also analyse the values of gauge coupling ratios at
the string scale. In Section 6 we present one example in more detail, a model without
any additional branes that in several respects is the simplest we encountered. In Section 7
we formulate some conclusions. An essential computational technique is to organize the
Ishibashi and boundary labels into simple current orbits, which leads to a dramatic speedup of the calculations. This is explained in Appendix A.
2. Algebraic model building
Our starting point is four-dimensional type-II string obtained by tensoring NSR fermions with a combination of N = 2 minimal models with total central charge 9. In a covariant
description, the NSR part of the theory is built out of four fermions with Minkowski
metric, and a set of superconformal ghosts. The type-II theory is modular invariant and
has two world-sheet supersymmetries needed for consistency. We assume it to be symmetric in left- and right-moving modes and have an extended chiral algebra leading to two
spacetime supersymmetries. To this theory we apply the orientifold procedure. Since our
approach is based on unitary rational CFT, it is convenient to use this description not only
for the minimal N = 2 factors, but also for the NSR part of the theory. To do so we use
a bosonic description of the latter (see [65] and references therein). This is convenient
because model-independent complications due to the GSO projection, spin-statistics and
superconformal ghosts are automatically taken care of. This implies that we are formally
constructing bosonic open strings. To obtain the spectrum we mimic the procedure used
in the covariant approach, namely fix a ghost charge to select the physical states. This
translates into a consistent truncation of the bosonic string characters. In the case of closed
strings, this procedure can be shown to map modular invariant bosonic strings to modular
invariant fermionic strings. In the case of open strings, it leads to fermionic open strings
satisfying all the integrality conditions on torus and Klein bottle as well as Annulus and
Mbius strip amplitudes, and that have the correct spin-statistics for all physical states,
and the proper symmetrization for RamondRamond states. We emphasize that this is only
used here as a bookkeeping device, and that we are not trying to conjecture a relation
between fermionic and bosonic strings.
Our starting point is a class of bosonic string theories with chiral algebra
E8 D5 Aint ,
(1)
where E8 and D5 are level 1 affine Lie algebras models. In this paper we can take the
model-dependent factor Aint to be of 4d Gepner type
Aint =
r

k=1
Ak ,
cint = 9,
14
and Ak is the N = 2 minimal model at level k. The E8 factor has no influence on the
massless spectrum. The only role of this factor is to cancel the conformal anomaly c8 +
c5 + cint + cbos + cghosts = 8 + 5 + 9 + 4 26 = 0, where cbos is the contribution of
the uncompactified bosons. The D5 factor describes the lightcone NSR fermions plus the
longitudinal NSR fermions and superconformal ghosts. The truncation that we implement
at the end of the day basically amounts to removing the contribution of the latter. For the
construction of the type-II string we follow the procedure explained in [9] (see also [66])
for heterotic strings, the only difference being that the fermionic truncation is applied to
both the left- and the right-moving sector. The tensor product (1) is first extended by means
of alignment currents (even combinations of the world-sheet supercurrents of the factors),
needed to maintain N = 1 world-sheet supersymmetry. To get theories with N = 2 space
time supersymmetry we must extend the algebra (1) further by the simple current group
generated by the currents (0, S, S, S, S, . . .), where S is a spinor representation of D5 or
a Ramond ground state of each minimal model that is a simple current (in D5 and each
of the minimal models there are two choices, but which one we take is irrelevant as long
as the same choice is made in both chiral sectors. For the minimal models we choose
(l, q, s) = (0, 1, 1), in the usual notation).
Because all aforementioned chiral algebra extensions are of simple current type, all
chiral data like the spectrum of primaries {i}, the conformal weights hi , the modular matrix
Sij and the fixed point resolution matrices S J can be expressed in terms of the chiral data
of the original unextended tensor product.
In this paper use a leftright symmetric extension in order to be able to apply the boundary/crosscap state formalism of [64]. There is a second, asymmetric choice obtained by
using instead of (0, S, S, S, S, . . .) the simple current (0, C, S, S, S, . . .) in of the chiral
sector. These are called type IIB (symmetric) and type IIA (asymmetric) extensions, respectively.
After applying these extension one obtains 168 [67] four-dimensional type-IIB theories.
Most of these have N = 2 spacetime supersymmetry, which will be broken to N = 1
by the orientifold procedure. Five of the 168 theories have N = 4 supersymmetry, and
can be ignored for further purposes, as they will never yield chiral N = 1 open strings.
We treat all these theories as non-supersymmetric CFTs. From the world-sheet point of
view, the world-sheet supercurrents are fields with conformal weight 3/2 which are not in
the chiral algebra (although their even powers are), and the spacetime supercurrents are
extended chiral algebra currents with conformal weight 1, which are treated just as any
other extension. There is one reason why conformal weight 1 currents are special, and that
is that the sub-algebra they generate is an affine Lie-algebra or U (1) factor. In this case
they extend D5 to E6 (or E7 in the N = 4 theories). In this way we obtain and RCFT
with Nprim primaries, whose ground states are in some representation of E6 , of the general
form m0 (1) m+ (27) m (27 ). For the N = 2 theories, Nprim varies between 260 and
108612.
These CFTs are our starting point and their chiral algebras are left unbroken in the rest
of the procedure. We will refer to it as the susy chiral algebra. As was discussed in [66],
one could consider the possibility of starting with a smaller chiral algebra, and allow for the
possibility that, for example, spacetime supersymmetry is present in the bulk, but broken
on some of the branes. While it is possible in principle to investigate this in our formalism,
15
the practical problem is that the chiral algebra becomes smaller, and hence the number of
primaries much larger.
Among the Nprim primaries there is almost always a subset Nsim that are simple currents.
This number ranges from 2187 (equal to Nprim ) for the tensor product (1)9 to just 1. Under
fusion, they form a discrete group G. These simple currents are used to build symmetric
modular invariant partition functions.
For a simple current MIPF one has to specify the following data:
A group H that consists of simple currents of A. All currents J in H must satisfy the
condition that the product of their conformal
weight hJ and order NJ is integer. In
general H is a product of cyclic factors H = ZN . The generator of the ZN will
be denoted as J ;
A symmetric matrix X that obeys
2X = QJ (J ) mod 1,
= ,
X = hJ ,
plus a further constraint N X Z for all , .
Here Q is the simple current monodromy charge, QJ (a) = h(a) + h(J ) h(J a), where
h is the conformal weight. When in the following we write X(J, J ) for arbitrary simple
currents in H we mean

n m X
X(J, J ) =

for J =
and J = Jm .
The resulting value of Zij is the number of currents L H such that
n
J
j = Li,
QM (i) + X(M, L) = 0 mod 1
for all M H. These MIPFs can be further chiral algebra extensions of the susy chiral algebra, fusion rule automorphisms or combinations thereof. The formalism of [64]
is insensitive to the distinction among these various types. With the exception of certain
pathological cases, this set of MIPFs is the most general one where the combinations (i, j )
of left and right representations that occur are linked by simple currents, i.e., i = Jj c for
some J cH . As the c indicates, we build simple currents starting from the charge
conjugation invariant. One could also start from the diagonal invariant, but there is no general formula available for the boundary and crosscap coefficients in that case. In many
(though not all) cases the diagonal invariant is itself a simple current automorphism of
the charge conjugation invariant, and hence is already included. It is well known that additional, exceptional MIPFs exist for the Gepner models (see [68,69]), including the
famous three-generation one for the tensor product (1, 16, 16, 16) [8], but again there is
no boundary/crosscap formalism available for these cases (although the boundary coefficient are known for the SU(2) exceptional invariants [70]).
The next step is to specify the orientifold data, which consist of
16
A Klein bottle current K. This can be any simple current of A that is local with all
order two currents in H, i.e., obeys
QI (K) = 0 mod 1 I H, I 2 = 0;
(2)
A set of phases K (J ) for all J H that satisfy

K (J )K (J ) = K (J J )e2iX(J,J ) ,
J, J H,
with K (J ) = ei(hKL hK ) (K, L), with (K, L) = 1.

Note that the phases K (J ) satisfy the same multiplication rule independent of the Klein
bottle current K, and that the solutions depend on K because of the second requirement.
The signs can be chosen freely provided the multiplication rule for holds. It is easy
to see that this implies that there is a freedom of choosing one sign for each independent
even factor in the simple current subgroup cH . All these choices yield valid orientifolds,
but they are some equivalences, which will be discussed below.
This data defines a (bosonic) orientifold with spectrum encoded in the total one-loop
partition function
1
(T + K + A + M),
(3)
2
where we distinguish between the four topologically distinct surfaces with vanishing Euler
number. These contributions can be expanded in (bi)linears of (hatted) characters in the
usual way [71]:

T =
(4)
Zij i j ,
K=
K i i ,
ij
A=
Na Nb A
ab i ,
M=
Na M i a i .
(5)
ab
The following objects are introduced:

The labels a, b that appear in the open string sector of the partition function are a
short-hand notation for the boundary labels. In full glory these labels are H-orbits [a]
of a chiral sector a with a possible degeneracy label a which is a (discrete group)
character of the a certain subgroup of the stabilizer, called the central stabilizer Ca
(see [64] and below). We write this as [a, a ];
The nonnegative integers Na := N[a,a ] are the CP-factors. These numbers must be
such that the total partition function is free of divergences. This will be reviewed in
Section 4.1.
The Klein bottle, annulus and Mbius coefficients factorize as
K =
i
,m
S i m U(m,J ) gJ,J
U(m,J )
m,J,J
S0m
Ai[a,a ][b,b ] =
i
M[a,
a]
,m
S i m R[a,a ](m,J ) gJ,J
R[b,b ](m,J )
S0m
m,J,J
,m
P i m R[a,a ](m,J ) gJ,J
U(m,J )
S0m
m,J,J
17
(6)
(7)
where P = T ST 2 S T [72] and S, T are the usual modular matrices. In these expressions the sums run over all Ishibashi labels. These labels are pairs (m, J ) that obey
m = J m,
QK (m) + X(M, J ) = 0 mod 1
(8)
for all M H. We note that we consider boundaries and crosscaps of trivial automorphism
type [73], which means that we require that the susy chiral algebra is preserved (and not
just preserved up to automorphism) in closed string scattering from one of these defects.
This implies that the closed strings that can couple to these defects must be C-diagonal. The
Ishibashi labels (8) are one-to-one to such closed string sectors. We have also introduced
The Ishibashi metric g ,m
,m
gJ,J
=
Sm0
K (J )J ,J c
SmK
for all J, J H. Here indicates the choice of Klein bottle current as well as the
phases K (J ) that define an orientifold;
The boundary reflection coefficients

|H|
J
R[a,a ](m,J ) =
(9)
(J )Sam
;
|Ca ||Sa | a
The crosscap reflection coefficients
1
U(m,J ) =
(K, L)PLK,m J,0 ,
|H|
(10)
LH
where S J is the fixed point resolution matrix S J , whose rows and columns are labelled
by fixed points a, m of J , implements a modular S-transformation on the torus with J
inserted. It is unitary and obeys [74]
J
SKi,j
= Fi (K, J )e2iQK (j ) SijJ .
(11)
The aforementioned central stabilizer is defined in terms of this quantity as

Ca = J Sa Fa (K, J )e2iX(K,J ) = 1 for all K Sa .
One can check that R is unitary. The reflection coefficients have an important physical
meaning, because they are (proportional to) the coupling of closed strings from Ishibashi
18
sector (m, J ) to D-brane [a, a ]. The oriented annulus amplitude therefore reads
or i
A [b,
b]
[a,a ]
:=
S i m R[a,a ](m,J ) R[b,

b ](m,J )
m,J,J
S0m
In the unoriented string, specified by the Klein bottle current K and the phases (J ), the
annulus amplitude is
i
Ai[a,a ][b,b ] = Aor [b, ]c [a,a ] ,
b
where [b, b ]c is the conjugate boundary label. Geometrically the pair of branes [b, b ]
and [b, b ]c are mapped to each other by the orientifold action R = R(K, ). In CFT
this image is encoded by the boundary conjugation matrix

1, [b, b ] = [a, a ]c ,
A0[a,a ][b,b ] =
0, otherwise.
Note that the various unoriented annuli can all be obtained from the unique oriented annulus by multiplication with the boundary conjugation matrix.
The physical meaning of U is as coupling constants between the Ishibashi sectors (m, J )
and the O-plane.
This formalism has been shown to lead to integer values for all open and closed string
particle multiplicities [75]. This results holds for all RCFTs, not just the minimal N = 2
models considered here. This universal validity gives additional confidence in its correctness, but the ultimate consistency check would be to demonstrate that all sewing constraints
are satisfied on all Riemann surfaces. This has been done for orientable surfaces [76], and
is underway for the non-orientable case.
2.1. Orientifold equivalences
For a general simple current MIPF the set of known orientifolds is parametrized by
a Klein bottle current K and a number of signs , one for each independent even factor
in the discrete group that defines the MIPF. The Klein bottle current can be any simple
current subject to the constraint (2), and there is no restriction on the signs. However, not
all these choices are inequivalent. The following equivalences exist between the choices
{K, } (here G is the full group of simple currents and H the subgroup used in the construction of the MIPF)

{K, } KJ 2 , , J G,

{K, } KL, , L H,

{K, } (K), () .
Here is the action induced by the permutation of identical minimal models (if any) on
the primary fields of the tensor product, and is the action induced on the signs . The
modified signs and can be worked out from the formula for the crosscap coefficients
(10) and the relation
PJ 2 a,b = J 2 (a)e2i[QJ (b)QJ (J a)] Pab ,
J (a) = ei[ha hJ a ] ,
19
but we will not present them explicitly here. The combined action of the three equivalences
organizes the various crosscap choices into equivalence classes, and we have taken into
account one representative from each class. The results seem to indicate that there are no
further equivalences: in general the number of (a, b, c, d) stacks of various types, as well
as the number of tadpole solutions is distinct.
Note that in each case the equivalence between orientifolds holds up to a certain permutation of the boundary labels. A subgroup of these transformation may fix the orientifold,
but lead to a residual equivalence of boundary choices for a given orientifold. We did not
attempt to remove this (and other) equivalence among boundaries, because it was much
easier to compare the resulting spectra and remove identical ones a posteriori.
3. Massless spectrum
Our prime interest will be the identification of the massless states in the partition function (3). Before doing this in the correct way, we have to perform a truncation to obtain the
spectrum of the superstring. This truncation is most easily described as follows. Note that
has spin 1 and that therefore the bosonic algebra A must contain a
the current (08 , S5 , S)
level one WZW factor that is larger than D5 . The only possibility is E6 . All primaries m of
A therefore decompose into primaries of E6 . Tachyonic states are always singlets of E6 .
Massless states are singlets, fundamentals 27, anti-fundamentals 27 or adjoints 78. The
truncation from the bosonic spectrum to a superstring spectrum is
left-movers:
1 ,
27 12 , 27 12 ,
78 V ,
right-movers: 1 ,
27 12 , 27 12 ,
78 V ,
where is a (complex) N = 1 chiral multiplet and V a N = 1 vector multiplet. Note that
E6 singlets are projected out. The 27 representation thus yields one real bosonic degree
of freedom, and one fermionic one. In heterotic strings a representation (27, R) (where
R is some gauge representation) is always accompanied by a (27, R ), and together they
form one N = 1 chiral multiplet, containing a complex boson and a Weyl fermion in the
representation R. In type-II closed strings, the combinations (27, 27) + (27 , 27 ) yields
one N = 2 vector multiplet, with four real bosonic and four real fermionic degrees of freedom. The combination (27, 27 ) + (27 , 27) yields one N = 2 hyper multiplet, with the
same number of degrees of freedom. Note that in principle one could switch the role of
the 27 and the 27 in the truncations for the right-movers with respect to the left-movers.
In covariant language, this corresponds to switching the ghost-charge assignment for the
fermions. This would map a IIA spectrum to a IIB spectrum and vice-versa. However, the
same interchange can also be achieved by going from a IIB to a IIA extension. In applications to orientifolds, it is clearly preferable to adopt a universal truncation rule (ghost
charge assignment) for left- and right-moving, as well as open string characters.
20
To every hi = 1 primary we can associate a Witten index
wi = m+
i mi
where m+
i (mi ) counts the number of 27 (27 ) in i. Note that
c
m
i c = mi wi = wi .
3.1. The oriented closed string spectrum

The torus contribution Z in (3) is the partition function of the parent theory of the
orientifold. After undoing the bosonic string map it describes type II string theory on some
CalabiYau 3-fold at the Gepner point. The ground state of the vacuum sector (00) is
projected out. At the first excited level it yields
V V = G + H,
the N = 2 gravity and universal hyper-multiplet. The other massless sectors yield the
model-dependent Abelian vector and hyper-multiplets. First note that modular invariance
implies
Zij = Zi c j c .
Complex sectors (ij ) + (i c j c ) contribute the following N = 2 multiplets:

+

vector multiplets,
Zij m+
i mj + mi mj
+
+
hyper-multiplets.
Zij mi mj + mi mj
Real sectors (ij ), i = i c , j = j c contribute
+
Zij m+
i mj
vector multiplets,
+
Zij m+
i mj
hyper-multiplets.
The total numbers h21 of vector multiplets and h11 of hyper-multiplets are
h21 =
1 + +

Zij mi mj + m
i mj ,
2
ij
h11 =
1
2
+
Zij m+
i mj + mi mj .
ij
The sum is over all fields, including conjugates, and the factor 1/2 corrects for double
counting; for real fields m+
i = mi . Note that
:= 2[h21 h11 ] =

ij
wi wj c Zij
21
is a topological invariant. The string theory we have constructed therefore has the same
massless spectrum as type IIB string theory on a CalabiYau manifold X3 with Hodge
numbers h11 and h21 1.
The spectrum of the other dual pair, (IIA/X3 , IIB/X 3 ), can be obtained by conjugating
the right-moving spacetime supercurrent, which results in the IIA extension. It is easy to
see that there are h21 hyper multiplets and h11 vector multiplets in this case.
3.2. The unoriented closed string spectrum
The first step in the orientifold procedure is the truncation of the type II spectrum to
states that are invariant under the involution = (K, ). The resulting massless d =
4, N = 1 spectrum can be obtained from 12 (Z + K) by simple counting arguments. The
vacuum sector gives the universal gravity multiplet and a chiral multiplet that contains the
dilaton. Off-diagonal sectors do not flow in the direct Klein bottle. Their contribution is
halved by the orientifold projection. Complex off-diagonal sectors (ij ) + (i c j c ) + (j i) +
(j c i c ) contribute the following N = 1 multiplets:

+

Zij m+
vector multiplets,
i mj + mi mj
+ +

+
Zij mi mj + mi mj + 2Zij m+
chiral multiplets.
i mj + mi mj
Real off-diagonal sectors (ij ) + (j i) contribute
+
Zij m+
i mj
vector multiplets,
+
3Zij m+
i mj
chiral multiplets.
Diagonal sectors are symmetrized or anti-symmetrized according to the Klein bottle coefficient. Complex diagonal sectors (ii) + (i c i c ) contribute

1
+

(Zii Ki ) m+
vector multiplets,
i mi + mi mi
2
+

1
+

+
(Zii + Ki ) m+
chiral multiplets.
i mi + mi mi + Zii mi mi + mi mi
2
Real diagonal sectors contribute
1
+
(Zii Ki )m+
vector multiplets,
i mi
2
1
+
+
(Zii + Ki )m+
chiral multiplets.
i mi + Zii mi mi
2
Define

+ +
1

h11 :=
(Zij ij Ki ) mi mj + mi mj ,
4
ij
1 Note that in [1] the Hodge numbers of the type IIA compactification were listed. In this paper we list them
for the type IIB compactification, the closed string theory to which we apply the orientifold procedure.
22
where we sum over all primaries. Then the total number of closed string Abelian vector
multiplets is h
11 and the total number of model-dependent closed string chiral multiplets is
+
h21 + h11 . In a geometrical setting the numbers h+

11 and h11 denote the number of harmonic
(1, 1)-forms that are anti-invariant or invariant under the orientifold action.
3.3. The oriented open string spectrum

The massless gauge bosons of the a U (Na ) gauge group come from the first excited
level of the vacuum sector. The annulus coefficient Ai a b counts states in the bifundamental
(Va , Vb ) representations of the spacetime gauge group. Note that
Ai
= Ai a b .
(12)
Let Ma+ b (Ma b ) denote the number of chiral (anti-chiral) multiplets that transform according to (Va , Vb ). These numbers are given by

i b
m
Ma b =
i A a ,
i
where we sum over all primaries. Due to (12) the spectrum obeys the d = 4 CPT relation Mb a = Ma b . The net chirality is measured by the anti-symmetric chiral intersection
matrix

wi Ai a b .
Ia b = Ma+ b Ma b =
i
3.4. The unoriented open string spectrum

The open string spectrum of the orientifold is encoded in 12 (A + M). The boundary
conjugation matrix A0 ab defines the orientifold image or conjugate a c of brane a. Complex pairs of branes a = a c give rise to unitary gauge groups. For real branes a = a c the
gauge group depends on the Mbius coefficient Ma0 . When Ma0 = 1/ + 1, the first excited
level of the vacuum is symmetrized/anti-symmetrized, signalling a Sp(Na )/SO(Na ) gauge
group. We can summarize this as

G=
U (Na )
SO(Na )
Sp(Na ).
a,complex
a,real
a,pseudo-real
The CP-factors Na are determined by tadpole cancellation (see Section 4.1). In our conventions Aiab counts states in (Va , Vb ). When we conjugate a brane label, the corresponding
vector must be conjugated. So Aia c b counts states in (Va , Vb ), etc. Note that2
c
Aiab = Aia c bc .
(13)
2 This follows from

,m
Ra(m,J
) = gJ J Ra c (m,J ) ,
c
which can be derived from Ai ab = Ai a b , unitarity of S and completeness.
23
For off-diagonal sectors a = b, the sectors (a, b) and (b, a) must be identified. The total
+
(Mab
) of chiral (anti-chiral) multiplets that transform according to (Va , Vb )
number Mab
is

i
Mab
=
m
i Aab ,
i
where we sum over all primaries. Due to (13) the spectrum obeys the d = 4 CPT relation
. The net chirality of fermions transforming as (Va , Vb ) is measured by

Mac bc = Mab

+
ab = Mab
Mab
=
wi Ai ab .
i
Note that the ordering of the indices is irrelevant. From (13) we easily derive
ab = a c bc ,
abc = a c b .
(14)
When b = a c the representations are adjoints (Adj) of U (Na ). Of course, adjoint matter
cannot give rise to chiral matter in d = 4, as can easily be seen from (14). In order to make
contact with geometry, we note that ab is the upper-half part of the geometric intersection
matrix and that the lower-half is given by a c bc . Diagonal sectors a = b are projected by
the Mbius strip to symmetric (S) or anti-symmetric (A) representations of the gauge group
G(Na ). Note that3
Mi
= M i ac .
(15)
In a self-explanatory notation, the number of chiral (anti-chiral) multiplets in these representations are

1 i
Ma,S
=
mi Aaa + Mai ,
2
i
=
Ma,A

1 i
mi Aaa Mai .
2
i
From (13) and (15) these multiplicities obey the CPT relations Ma,S
= Mac ,S and Ma,A
=
Ma c ,A . The net chirality is
a,S =

1 i
wi Aaa + Mai ,
2
i
a,A =

1 i
wi Aaa Mai .
2
i
3 We now also need

,m
U(m,J
) = gJ J U(m,J ) .
24
For real branes a, this index vanishes, as befits the symmetric and anti-symmetric representations of symplectic
andi orthogonal groups in d = 4. When we compare with a geometric
description,
i wi Aaa is the intersection between a brane and its image, whereas

i
i wi Ma is the intersection between a brane and the orientifold plane(s).
4. Tadpoles and anomalies

4.1. Tadpole cancellation
Non-vanishing one-point functions of massless scalars on the disk or crosscap may
cause several problems. If the scalar is physical particle in the spectrum, surviving the
orientifold projection, a tadpole indicates an instability in the vacuum. This manifests itself
as an infinity in the Euler number 0 diagrams. In this case the theory would be unstable, but
one might still hope that a stable vacuum exists. If the scalar is not a physical particle, the
presence of a tadpole renders the theory inconsistent, and this may manifest itself through
an uncanceled anomaly. If all Klein bottle coefficients are positive, all scalars from NSNS
sectors are physical, but since the RR sector always has a projection with opposite sign,
the RR scalars are unphysical.
In a supersymmetric theory the NSNS and RR sectors are linked, and canceling unphysical tadpoles is equivalent to cancelling all tadpoles. The condition for the cancellation of
all tadpoles is

(16)
Nb Rb(m,J ) = 4m U(m,J )
b
for all Ishibashi labels (m, J ) for which the sector (mmc ) + (mc m) in the torus (4) yields
massless spacetime scalars. Here 0 = 1 and 1 otherwise. Tadpole cancellation is a
condition on the CP-factors Na of the gauge groups.
There are two further constraints on the CP multiplicities. If two boundaries a and b are
conjugate, one must require that Na = Nb , and if the CP-group associated with label a is
symplectic Na must be even.
The dilaton couplings R0b are always positive (the Ishibashi label m = 0 is nondegenerate, so there is no need for the degeneracy label J ). Hence one can only satisfy
the dilaton tadpole condition if U0 < 0. The overall sign of the crosscap coefficients is a
free parameter, which must be fixed so that U0 < 0. Changing this sign changes the sign of
all Mbius coefficients.
4.2. Anomaly cancellation
The chiral gauge anomalies can be obtained from a formal polynomial that is proportional to

i
i
P(F ) =
wi
Aab cha (F ) chb (F ) +
Ma cha (2F )
i
ab
25
where
cha (F ) :=
1
Tra F n ,
n!
n
where the trace in Tra F n is taken over the fundamental representation of U (Na ), and
over the anti-fundamental in Tra c F n ; F is the field strength two-form. To obtain the cubic anomalies in four dimensions one restricts to polynomial to six-forms and applies the
descentmethod. The argument F can be expanded in a Lie-algebra basis with generators
T k as k F k T k . If F lies entirely within a real subalgebra all odd terms (in F ) in the
polynomial vanish, and in particular there are no four-dimensional anomalies.
Following [77], we can easily show that tadpole cancellation implies cancellation of
the purely cubic terms in the polynomial, but not the others (the purely cubic terms are
obtained by keeping only terms proportional to Tr F 3 , without using group-dependent factorizations of such terms (such factorizations of cubic traces exist only for U (1) and U (2)).
To do so, we use the fact that chiral characters or Witten indices transform to themselves
under modular transformations, as well as under transformations involving the P -matrix,
because they are independent of the modulus

wi = i
Sji wj ,
wi = i
Pji wj ,
j
substituting this into the cubic part of the polynomial, and using the expression for the
annulus and Mbius strip amplitudes (6) and (7) we get

,m
P(F )cubic = i
(S0m )1 wm
R(m,J )a gJ,J

m,J,J

3
3
3
.
R(m,J )b Na Trb F + Nb Tra F + U(m,J ) 8 Tra F
Now we interchange the summed indices a and b in the second term, use the fact that
,m
,m
gJ,J
= gJ ,J , and substitute for U(m,J ) the right-hand side of (16) (note that only terms
with m = 0 contribute, because the vacuum sector is non-chiral in four dimensions), and
we see immediately that P(F )cubic = 0.
For non-abelian factors in the gauge group this implies simply the usual cancellation of
cubic anomalies. For U (1)-factors that get chiral contributions only from vectors, it also
has the expected consequence: the total number of vectors and conjugate vectors must be
the same in each such factor. The situation is a bit more interesting as soon as chiral tensors
contribute. For example, if we assign charge 1 to the (anti)-fundamental representations
of a U (1) factor, then symmetric and anti-symmetric tensors have charge 2. The anomaly
cancellation implied by tadpole cancellation has nothing to do with the third power of
these charges, but is an extrapolation of U (N ) anomaly cancellation to N = 1. Hence
we get a contribution proportional to 1 for vectors, (N 4) = 3 for anti-symmetric
tensors and (N + 4) = 5 for symmetric ones (note furthermore that anti-symmetric
U (1) tensors do not even exist; in a massless anti-symmetric sector the first state in
the spectrum is a massive symmetric tensor). This is not a problem, because precisely for
26
U (1) factors the cubic traces factorizes into lower order traces, which are not cancelled by
tadpole cancellation anyway, but are removed by the GreenSchwarz mechanism.
In this paper we only encounter vector/anti-vector anomaly cancellation within the standard model gauge groups because we have required that all chiral particles attached to the
(a, b, c, d) branes should be bifundamentals. This implies, in particular, the presence of
three left-handed anti-neutrinos to cancel the c and d-brane anomalies. It might be worth
considering to drop the restriction to bifundamentals for the c and d branes and achieve
anomaly cancellation by means of tensors, but we will not pursue that possibility here.
In those cases where the b-brane group is U (2), tadpole cancellation imposes the constraint that the numbers of vectors and anti-vectors of U (2) should be equal. This constraint
is discussed in [28], and lead in that context to a quantization of the number of families in
multiples of three. This is not the case here, because the Higgs also makes contributions to
the U (2) anomaly. In the hidden sector more interesting examples of anomaly cancellation
are possible, because chiral tensors may (and indeed do) contribute. These anomaly cancellations are of course a useful check on our results, but they are still far less restrictive
than in six dimensions, where we have performed such checks on the complete solution to
the tadpole solutions a few years ago.
The surviving part of the anomaly polynomial,

wi
Aiab Tra F Trb F 2 + Tra F 2 Trb F
P(F )rest =
i
ab
must be cancelled by a generalized GreenSchwarz mechanism. This mechanism involves

coupling between RR p-forms C(p) and the U (Na ) field strength Fa . For the oriented open
string these couplings are

a
a
(17)
Sa =
C(2)
Na Fa + C(0)
Tra F 2 ,
a
M4
where we decided to expand the RR fields in the highly degenerate basis spanned by the
complete set of boundary labels a. Here Fa is the U (1)a field strength and Tra F 2 is a
trace over the vector representation of U (Na ). In a geometric language the branes wrap
homology classes a and we can choose a basis a of 3-forms such that a b = ab .

The RR forms are KK reductions along this basis. We have CY a b = Iab , the chiral
intersection matrix of branes a and b. Poincare duality in this basis then reads

a
b
=
Iab dC(0)
.
dC(2)
b
From the couplings (17) we can easily read off the contribution to the U (1)a SU(Nb )2
mixed anomaly,
Na Iab .
For unoriented strings the sectors a and a c are identified and the couplings are encoded in

a

ac
a
ac
Fa +
C(0)
Tr Fa2 ,
(Sa + Sa c ) =
Na C(2)
C(2)
+ C(0)
a
M4
M4
27
where we sum over pairs (a, a c ). The GS-diagram is proportional

Na (Iab + Iabc Ia c b Ia c b ),
which has the correct form to cancel the mixed chiral U (1)a SU(Nb )2 anomaly.
4.3. Massless U (1)s
A nonzero coupling C(2) F generates an effective mass for the U (1) gauge field. A linear
combination

i Fi
i
is massless if and only if

i

ic
= 0.
i Ni C(2)
C(2)
(18)
This equation can have nontrivial solutions because the basis a that we are using is
a } in a complete, nonhighly non-degenerate. It is natural to expand this brane-basis {C(2)
(m,J )
degenerate basis {C(2) } that is one-to-one to the Ishibashi labels (m, J )

(m,J )
a
=
Ra(m,J ) C(2) ,
C(2)
(19)
(m,J ),wm =0
where R are the boundary reflection coefficients. This equation has a well-known geometrical analogue, namely as an expansion of the basis a in a homology basis. In a
geometric orientifold the tadpole conditions are written in a convenient homology
basis i
for H 3 (CY). The RR 2-form in the brane basis can then be expanded as C a = i ai C i
where Ci are the reductions of the ten-dimensional RR 5-form along the i and ai are the
wrapping numbers or charges of the brane. From the CFT tadpole condition it follows that
the Ishibashi labels are the natural basis at the Gepner point and that the reflection coefficients are the wrapping numbers. This is the motivation for (19). Then (18) becomes

i Ni [Ra(m,J ) Ra c (m,J ) ] = 0
i
for all Ishibashi labels (m, J ).
5. Results
In this section we will present the results of our search. We list some statistics and characteristics of the Gepner models and MIPFs we scanned in Tables 2, 3 and 4. Furthermore,
we display plots of distributions of standard model particle multiplicities, the Higgs and
some characteristics of the various models like ratios of gauge couplings, the number of
hidden branes and features of the hidden gauge group. We will also present a small investigation into varying the number of chiral families.
28
Table 1
List of standard model representations that can appear, and their labelling
U (3)a
U (2)b
V
V
V
V
0
0
0
0
V
V
0
0
V
V
0
0
9
10
0
0
V
V
11
Nr.
1
2
3
4
5
6
7
8
Sp(2)b

V

V
U (1)c
U (1)d
Massless particle
0
0
0
0
V
V
V
V
(u, d)
(u, d)
uc
dc
(, e )
(, e )
c
e+
V
V
0
0
H1
H2
(3, 1, 13 )1,1
(3, 1, 23 )1,1
0
0
V
V
0
0
V
V
SO(2)c
Sp(2)c

V

V
12
13
Adj
(8, 1, 0)0,0 + (1, 1, 0)0,0
14
(3 , 1, 13 )2,0
15
(6, 1, 13 )2,0
16
Adj
(1, 3, 0)0,0 + (1, 1, 0)0,0
17
(1, 1, 0)0,0
18
(1, 3, 0)0,0
19
Adj
(1, 1, 0)0,0
20
(1, 1, 1)0,0
21
(1, 1, 1)0,0
22
23
24
0
0
0
0
0
0
0
0
0
Adj
A
S
(1, 1, 0)0,0
(1, 1, 1)0,2
25
(3, 1, 16 )1,0
26
(1, 2, 0)0,0
27
(1, 1, 12 )0,0
28
(1, 1, 12 )0,1
29
(1, 1, 0)0,0
5.1. The numbers

In total we found 179520 distinct standard model spectra with solutions to all tadpole
equations. There are few cases where the same spectra are obtained for different MIPFs.
In some cases these MIPFs have the same Hodge numbers and are also otherwise indistinguishable. This occurs, for example, for the tensor product (2, 2, 2, 6, 6) and presumably
indicates an unresolved redundancy. In other cases, the Hodge numbers are different, although, surprisingly, the open sector is identical. The total number of such equivalent
29
Table 2
Summary results for all Gepner models
Nr.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
Tensor
Prim
(1, 5, 41, 1804)

(1, 5, 42, 922)
(1, 5, 43, 628)
(1, 5, 44, 481)
(1, 5, 46, 334)
(1, 5, 47, 292)
(1, 5, 49, 236)
(1, 5, 52, 187)
(1, 5, 54, 166)
(1, 5, 58, 138)
(1, 5, 61, 124)
(1, 5, 68, 103)
(1, 5, 76, 89)
(1, 5, 82, 82)
(1, 6, 23, 598)
(1, 6, 24, 310)
(1, 6, 25, 214)
(1, 6, 26, 166)
(1, 6, 28, 118)
(1, 6, 30, 94)
(1, 6, 31, 86)
(1, 6, 34, 70)
(1, 6, 38, 58)
(1, 6, 40, 54)
(1, 6, 46, 46)
(1, 7, 17, 340)
(1, 7, 18, 178)
(1, 7, 19, 124)
(1, 7, 20, 97)
(1, 7, 22, 70)
(1, 7, 25, 52)
(1, 7, 28, 43)
(1, 7, 34, 34)
(1, 8, 14, 238)
(1, 8, 16, 88)
(1, 8, 18, 58)
(1, 8, 22, 38)
(1, 8, 28, 28)
(1, 9, 12, 229)
(1, 9, 13, 108)
(1, 9, 20, 31)
(1, 10, 11, 154)
(1, 10, 12, 82)
(1, 10, 13, 58)
(1, 10, 14, 46)
(1, 10, 16, 34)
(1, 10, 18, 28)
(1, 10, 19, 26)
28539
29772
31482
9399
35442
37800
4575
12690
48258
6156
64449
20748
2835
108612
12600
13650
14742
15876
18270
20664
2464
26460
8260
4018
46536
9396
10224
11880
4116
14760
21060
7128
33264
8082
10218
12690
2034
28410
2875
1015
6160
7704
8970
10332
11748
14994
4698
2280
S.C.
1
2
3
1
6
7
1
3
14
2
21
7
1
42
2
4
6
8
12
16
2
24
8
4
48
3
6
9
3
18
27
9
54
4
12
20
4
60
1
1
11
6
12
18
24
36
12
6
MIPF
Intersect.
1, 0, 0
2, ,
2, ,
1, 0, 0
4, 1, 0
2(1), 1, 0
1, 0, 0
2(1), 1, 0
4, 1, 0
2, 0, 0
4(2), 3, 0
2(1), 1, 0
1, 0, 0
8(1), 3, 0
2, 0, 0
6(1), 1, 0
4(2), 2, 0
10(7), 8, 0
12(9), 10, 2
14(12), 12, 1
2, 0, 0
20(16), 18, 4
10(7), 8, 1
6(3), 3, 0
28(24), 26, 5
2, 0, 0
4(1), 1, 0
6(2), 4, 0
2, 0, 0
12(6), 8, 0
10(2), 7, 0
6, 2, 0
16(8), 10, 2
6(3), 3, 1
12(7), 7, 0
12(7), 9, 1
6(1), 1, 0
20(8), 14, 2
1, 0, 0
1, 0, 0
2, 0, 0
4, 1, 0
12(4), 9, 0
12(4), 9, 0
20(12), 17, 3
36(18), 29, 5
12(3), 6, 2
4, 0, 0
0
12
1128
0
144
54
0
81044
234
0
1744
0
18260
212000
4939262
1289765
2912087
0
1539069
435815
87494
3687098
0
29672
2513
0
38210
7523
19
62992
3447270
70962
240804
2888
152706
0
0
0
372
277382
19649
328229
198765
19344
0
Sol.
SM spectra
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
217
67
297
221
0
0
2466
1587
36
18
0
0
67723
16490
0
0
0
0
0
0
0
0
0
0
0
0
0
0
70
13
16
16
0
0
4
2
0
0
1288
188
0
0
0
0
0
0
0
0
0
0
0
0
7389
3687
503
173
32
14
0
0
(continued on next page)
30
Table 2 (continued)
Nr.
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
Tensor
(1, 10, 22, 22)
(1, 11, 11, 76)
(1, 12, 12, 40)
(1, 12, 13, 33)
(1, 12, 19, 19)
(1, 13, 13, 28)
(1, 13, 18, 18)
(1, 14, 14, 22)
(1, 16, 16, 16)
(2, 3, 19, 418)
(2, 3, 20, 218)
(2, 3, 22, 118)
(2, 3, 23, 98)
(2, 3, 26, 68)
(2, 3, 28, 58)
(2, 3, 34, 43)
(2, 3, 38, 38)
(2, 4, 11, 154)
(2, 4, 12, 82)
(2, 4, 13, 58)
(2, 4, 14, 46)
(2, 4, 16, 34)
(2, 4, 18, 28)
(2, 4, 19, 26)
(2, 4, 22, 22)
(2, 5, 8, 138)
(2, 5, 10, 40)
(2, 5, 12, 26)
(2, 6, 7, 70)
(2, 6, 8, 38)
(2, 6, 10, 22)
(2, 6, 14, 14)
(2, 7, 7, 34)
(2, 7, 10, 16)
(2, 8, 8, 18)
(2, 8, 10, 13)
(2, 10, 10, 10)
(3, 3, 9, 108)
(3, 3, 10, 58)
(3, 3, 12, 33)
(3, 3, 13, 28)
(3, 3, 18, 18)
(3, 4, 6, 118)
(3, 4, 7, 43)
(3, 4, 8, 28)
(3, 4, 10, 18)
(3, 4, 13, 13)
(3, 5, 5, 68)
(3, 6, 6, 18)
Prim
26532
9828
12138
595
10500
19845
3220
5336
33210
6320
6972
8232
9120
3036
13340
1232
22920
3540
4160
4830
5340
7140
2320
1100
12060
2862
1230
6006
3024
3780
5544
9744
4032
2040
6480
630
12000
2900
3280
1700
6300
9200
2100
1584
3280
540
4410
2394
1064
S.C.
MIPF
Intersect.
Sol.
72
13
28
1
21
45
10
16
108
2
4
8
10
4
20
2
40
4
8
12
16
24
8
4
48
4
4
28
8
16
32
64
18
12
40
4
96
5
10
5
25
50
4
3
20
4
15
7
8
50(20), 39, 10
2, 1, 0
10(2), 8, 1
1, 0, 0
4, 0, 0
10, 5, 0
4, 1, 0
14(1), 10, 2
26(5), 17, 5
2, 0, 0
6(4), 4, 0
5(4), 4, 1
4(1), 1, 0
6(4), 4, 0
12(8), 9, 4
2, 0, 0
10(8), 9, 2
6(4), 4, 0
30(26), 26, 12
12(7), 9, 1
27(24), 25, 13
60(48), 54, 23
30(23), 25, 15
6(2), 3, 0
54(39), 51, 25
6(3), 4, 0
6(1), 3, 0
12, 7, 3
5(1), 3, 0
27(18), 22, 7
59(20), 53, 27
87(20), 71, 30
6, 0, 0
12, 6, 1
44(3), 32, 16
6, 3, 0
92(7), 71, 34
2, 0, 0
4, 1, 0
2, 0, 0
6, 2, 0
12, 4, 0
6(3), 3, 0
2, 0, 0
12(1), 7, 0
6, 0, 0
4, 0, 0
2, 0, 0
10, 1, 0
499730
96
22942
0
0
1854
20
17112
125476
0
315210
180431
1204
18416
142742
0
151792
2304160
5133558
468648
1918601
3700006
745644
23872
3403934
191424
5502
34744
28234
323662
361546
758636
0
10504
1019592
1320
850844
0
124
0
118
1492
176520
0
3348
0
0
0
88
15055
0
8
0
0
0
0
63
9204
0
0
6
0
0
482
0
1044
0
598
146
17411
218598
801
0
423560
0
0
2426
0
6313
18964
62856
0
4
222006
0
137472
0
0
0
0
0
0
0
0
0
0
0
0
SM spectra
2780
0
2
0
0
0
0
23
494
0
0
4
0
0
176
0
216
0
294
69
5607
45055
360
0
43532
0
0
431
0
1368
3624
5424
0
1
17311
0
9878
0
0
0
0
0
0
0
0
0
0
0
0
31
Table 2 (continued)
Nr.
Tensor
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
(3, 8, 8, 8)
(4, 4, 5, 40)
(4, 4, 6, 22)
(4, 4, 7, 16)
(4, 4, 8, 13)
(4, 4, 10, 10)
(4, 5, 5, 19)
(4, 6, 6, 10)
(4, 7, 7, 7)
(5, 5, 5, 12)
(6, 6, 6, 6)
(1, 1, 2, 11, 154)
(1, 1, 2, 12, 82)
(1, 1, 2, 13, 58)
(1, 1, 2, 14, 46)
(1, 1, 2, 16, 34)
(1, 1, 2, 18, 28)
(1, 1, 2, 19, 26)
(1, 1, 2, 22, 22)
(1, 1, 3, 6, 118)
(1, 1, 3, 7, 43)
(1, 1, 3, 8, 28)
(1, 1, 3, 10, 18)
(1, 1, 3, 13, 13)
(1, 1, 4, 5, 40)
(1, 1, 4, 6, 22)
(1, 1, 4, 7, 16)
(1, 1, 4, 8, 13)
(1, 1, 4, 10, 10)
(1, 1, 5, 5, 19)
(1, 1, 6, 6, 10)
(1, 1, 7, 7, 7)
(1, 2, 2, 5, 40)
(1, 2, 2, 6, 22)
(1, 2, 2, 7, 16)
(1, 2, 2, 8, 13)
(1, 2, 2, 10, 10)
(1, 2, 3, 3, 58)
(1, 2, 3, 4, 18)
(1, 2, 4, 4, 10)
(1, 2, 4, 6, 6)
(1, 3, 3, 3, 13)
(1, 3, 3, 4, 8)
(1, 4, 4, 4, 4)
(2, 2, 2, 3, 18)
(2, 2, 2, 4, 10)
(2, 2, 2, 6, 6)
(2, 2, 3, 3, 8)
(2, 2, 4, 4, 4)
Prim
9200
2322
3150
3888
1218
7200
1890
1540
5184
6615
9632
2124
2496
2898
3204
4284
1392
660
7236
1260
1584
2010
324
4410
1431
1890
2484
819
4320
1890
924
5184
492
1512
816
252
3288
920
160
2250
420
1400
260
4266
1040
1520
3024
720
1500
S.C.
MIPF
100
12
24
36
12
72
7
16
27
49
128
6
12
18
24
36
12
6
72
6
9
30
6
45
18
36
54
18
108
21
24
81
4
32
12
4
96
10
4
72
16
25
10
216
32
64
128
20
48
18, 11, 2
10, 6, 0
44(9), 33, 15
26(1), 17, 6
10(2), 7, 0
110(3), 74, 15
2, 0, 0
54, 22, 3
6, 0, 0
5, 0, 0
76, 44, 10
4, 0, 0
12, 0, 0
10, 0, 0
10, 2, 0
30, 8, 0
12, 0, 0
4, 0, 0
26, 7, 0
4, 0, 0
5, 0, 0
8, 1, 0
4, 0, 0
10, 1, 0
10, 0, 0
30, 0, 0
40, 1, 0
10, 0, 0
96, 6, 0
4, 0, 0
20, 1, 0
19, 2, 0
6, 0, 0
31, 7, 0
12, 0, 0
6, 0, 0
120, 22, 1
4, 0, 0
6, 0, 0
118, 19, 4
27, 0, 0
4, 0, 0
4, 0, 0
112, 11, 2
27, 2, 0
230, 60, 7
305, 101, 6
12, 0, 0
180, 5, 0
Intersect.
5406
56000
465222
93764
3682
999730
0
6874
0
0
174232
0
0
0
788
260
0
0
3864
0
0
2
0
54
0
0
2
0
652
0
16
8
0
3434
0
0
7566
0
0
10920
0
0
0
5552
1168
48876
111080
0
4640
Sol.
SM spectra
96
4
0
0
51448
8737
2590
59
0
0
277752
9983
0
0
64
6
0
0
0
0
70864
1310
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
72
2
0
0
0
0
730
99
0
0
0
0
0
0
172
6
0
0
4832
92
10304
95
0
0
0
0
32
Table 2 (continued)
Nr.
Tensor
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
(3, 3, 3, 3, 3)
(1, 1, 1, 1, 5, 40)
(1, 1, 1, 1, 6, 22)
(1, 1, 1, 1, 7, 16)
(1, 1, 1, 1, 8, 13)
(1, 1, 1, 1, 10, 10)
(1, 1, 1, 2, 3, 18)
(1, 1, 1, 2, 4, 10)
(1, 1, 1, 2, 6, 6)
(1, 1, 1, 3, 3, 8)
(1, 1, 1, 4, 4, 4)
(1, 1, 2, 2, 2, 10)
(1, 1, 2, 2, 4, 4)
(1, 2, 2, 2, 2, 4)
(2, 2, 2, 2, 2, 2)
(1, 1, 1, 1, 1, 2, 10)
(1, 1, 1, 1, 1, 4, 4)
(1, 1, 1, 1, 2, 2, 4)
(1, 1, 1, 2, 2, 2, 2)
(1, 1, 1, 1, 1, 1, 1, 4)
(1, 1, 1, 1, 1, 1, 2, 2)
(1, 1, 1, 1, 1, 1, 1, 1, 1)
Prim
4000
972
1134
1944
756
2592
96
1350
252
240
2673
912
900
440
2944
810
1944
540
264
2187
324
2187
S.C.
125
27
54
81
27
162
6
108
24
15
324
96
72
64
512
162
486
108
96
729
162
2187
MIPF
8, 0, 0
8, 0, 0
16, 0, 0
34, 1, 1
8, 0, 0
58, 0, 0
4, 0, 0
72, 0, 0
10, 0, 0
4, 0, 0
142, 0, 0
52, 0, 0
110, 0, 0
138, 0, 0
1031, 10, 0
34, 0, 0
156, 0, 0
48, 0, 0
46, 0, 0
124, 0, 0
24, 0, 0
152, 0, 0
Intersect.
0
0
0
6
0
0
0
0
0
0
0
0
0
0
448
0
0
0
0
0
0
0
Sol.
SM spectra
0
0
0
3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
spectra is, however, very small, and if we remove these equivalences there are 179119
left. In the rest of the paper we use the former set as the basis of our analysis.
In Table 2 we list for each Gepner model the search results. The second column contains
the values of the factors k in the tensor product. Models for which we only searched for
solutions of type 0 and 1 are denoted by a dagger . The third column specifies the number
of primaries, the fourth the number of simple currents and the fifth gives information on the
modular-invariant partition functions. In this column, the first entry is the total number of
symmetric simple current MIPFs. This is computed after removing MIPFs that are related
to each other by permutations of the identical factors of the tensor product. Generically, all
MIPFs related to different simple current subgroups and different matrices X are distinct.
However, in special cases generically distinct simple current MIPFs coincide (the simples
example is SU(2) level 2, where the generic A and D invariants coincide). In the table we
list the number after removing such coincidences. Between parentheses we indicate for
how many of these MIPFs the tadpole conditions were not completely solved for all standard model brane stack configurations; however, even in those cases we searched for all
solutions with at most three extra branes (or at most two if the number of candidate branes
was larger than 400). The next entry in column four is the number of MIPFs for which at
least one standard model brane configuration was found, and the last entry in column 4 is
the number of MIPFs for which a solution to the tadpole equations was obtained. Column
five gives the total number of standard model configurations, summed over all MIPFs, column six gives the total number of configurations with solutions to the tadpole conditions,
33
Table 3
All MIPFs with solutions
Tensor
(h11 , h21 , S)
Boundaries
(1, 6, 28, 118)
(24, 84, 429)

(75, 75, 565)
(24, 84, 425)
(29, 125, 557)
(14, 98, 451)
(43, 91, 509)
(29, 53, 353)
(29, 53, 351)
(19, 163, 649)
(9, 129, 525)
(27, 123, 569)
(19, 67, 377)
(59, 59, 449)
(23, 95, 437)
(29, 77, 457)
(53, 53, 443)
(38, 62, 377)
(17, 95, 419)
(29, 29, 251)
(8, 68, 315)
(17, 41, 259)
(35, 59, 369)
(13, 97, 405)
(29, 65, 355)
(20, 32, 241)
(20, 32, 249)
(38, 38, 311)
(10, 46, 253)
(23, 59, 343)
(7, 127, 467)
(17, 89, 387)
(7, 55, 263)
(19, 67, 323)
(19, 67, 343)
(22, 58, 321)
(13, 37, 243)
(25, 49, 303)
(27, 51, 319)
(41, 41, 323)
(25, 85, 423)
(23, 23, 225)
(31, 31, 273)
(11, 101, 401)
(16, 64, 325)
(8, 44, 227)
(21, 57, 301)
(20, 32, 243)
(41, 77, 463)
2400
6090
1980
3312
1656
4464
3420
1520
2968
1484
3944
2968
7888
2136
3696
3252
2538
1894
2316
728
1540
2000
1088
1666
1920
1920
2544
576
1044
1088
1504
1148
1804
1876
2256
1632
2256
2256
2948
948
952
1352
1230
2196
1512
2196
1512
2160
(1, 6, 30, 94)

(1, 6, 34, 70)
(1, 6, 38, 58)

(1, 6, 46, 46)
(1, 7, 34, 34)

(1, 8, 14, 238)
(1, 8, 18, 58)
(1, 8, 28, 28)
(1, 10, 14, 46)
(1, 10, 16, 34)
(1, 10, 18, 28)

(1, 10, 22, 22)
(1, 12, 12, 40)

(1, 14, 14, 22)
(1, 16, 16, 16)
(2, 3, 22, 118)
Nr.
6
2
6
10
9
2
11
6
12
10
2
13
4
10
2
4
2
2
12
8
11
2
20
22
25
29
7
6
1
20
22
19
34
35
30
29
44
31
36
7
9
13
12
13
14
16
10
4
Types
(6, 0, 0, 0, 50, 0, 0)
(2, 0, 0, 2, 15, 0, 0)
(6, 0, 0, 0, 221, 0, 0)
(10, 0, 3, 2, 0, 0, 0)
(9, 0, 0, 0, 1557, 0, 0)
(2, 0, 0, 2, 11, 0, 0)
(11, 0, 0, 0, 6, 0, 0)
(6, 0, 0, 0, 18, 0, 0)
(12, 0, 87, 0, 11, 0, 0)
(10, 0, 0, 0, 16294, 0, 0)
(2, 0, 4, 20, 37, 0, 0)
(13, 0, 6, 1, 29, 0, 0)
(4, 0, 0, 0, 1, 0, 0)
(10, 0, 0, 2, 0, 0, 0)
(2, 0, 7, 0, 4, 0, 0)
(4, 0, 0, 0, 16, 0, 0)
(2, 0, 0, 0, 2, 0, 0)
(2, 0, 110, 20, 54, 0, 0)
(12, 0, 4, 0, 0, 0, 0)
(8, 0, 0, 0, 3447, 0, 0)
(11, 0, 123, 0, 92, 0, 0)
(2, 0, 0, 16, 6, 0, 0)
(20, 0, 8, 0, 16, 0, 0)
(22, 0, 20, 42, 44, 10, 0)
(25, 0, 6, 0, 0, 0, 0)
(29, 0, 9, 0, 12, 0, 0)
(7, 0, 0, 6, 0, 0, 0)
(6, 0, 0, 0, 12, 0, 0)
(1, 0, 0, 0, 2, 0, 0)
(20, 0, 1, 0, 4, 0, 0)
(22, 0, 139, 153, 130, 0, 0)
(19, 0, 1227, 0, 886, 0, 0)
(34, 0, 0, 3, 0, 0, 0)
(35, 0, 0, 39, 0, 0, 0)
(30, 0, 0, 2, 0, 0, 0)
(29, 0, 8, 0, 14, 0, 0)
(44, 0, 0, 22, 0, 0, 0)
(31, 0, 31, 20, 3, 0, 0)
(36, 0, 0, 6, 2, 0, 0)
(7, 0, 1, 0, 1, 0, 0)
(9, 0, 13, 0, 8, 0, 0)
(13, 0, 0, 2, 0, 0, 0)
(12, 0, 219, 0, 35, 0, 15)
(13, 0, 2, 2, 0, 0, 0)
(14, 0, 88, 0, 88, 0, 0)
(16, 0, 0, 40, 0, 0, 0)
(10, 0, 5, 0, 0, 0, 0)
(4, 0, 0, 0, 4, 0, 0)
34
Table 3 (continued)
Tensor
(h11 , h21 , S)
Boundaries
(2, 3, 28, 58)
(17, 101, 463)

(17, 101, 463)
(39, 87, 475)
(39, 87, 475)
(11, 131, 533)
(23, 71, 389)
(58, 34, 371)
(33, 39, 325)
(33, 39, 325)
(39, 33, 325)
(39, 33, 325)
(42, 48, 357)
(42, 48, 357)
(48, 42, 357)
(48, 42, 357)
(51, 57, 447)
(51, 57, 447)
(45, 45, 350)
(37, 61, 413)
(19, 67, 339)
(20, 56, 308)
(21, 57, 321)
(26, 62, 343)
(29, 65, 407)
(56, 20, 308)
(62, 26, 343)
(20, 44, 299)
(34, 58, 367)
(25, 37, 287)
(28, 40, 309)
(35, 47, 329)
(37, 25, 287)
(16, 100, 440)
(13, 73, 339)
(13, 73, 339)
(15, 69, 339)
(15, 69, 339)
(21, 75, 421)
(21, 75, 421)
(12, 54, 297)
(12, 54, 297)
(24, 66, 353)
(24, 66, 353)
(20, 56, 306)
(20, 56, 306)
(26, 62, 339)
(26, 62, 339)
(15, 45, 285)
(15, 45, 285)
1408
1472
2552
2668
1200
2616
2480
1438
1460
1678
1724
1798
1856
2048
2110
2180
2230
2048
1540
864
1152
1152
1440
1440
2640
2796
864
1728
1152
1440
1440
1896
1000
800
820
962
1036
1180
1250
770
796
1360
1394
1040
1120
1144
1232
914
976
(2, 3, 38, 38)

(2, 4, 12, 82)
(2, 4, 13, 58)

(2, 4, 14, 46)
(2, 4, 16, 34)
Nr.
6
8
1
2
7
8
12
28
11
27
15
29
9
19
4
13
6
8
1
24
25
23
18
1
21
7
26
16
8
10
15
19
12
38
44
43
11
47
19
41
45
15
20
36
10
48
17
42
13
Types
(6, 0, 0, 0, 80, 0, 0)
(8, 0, 0, 0, 86, 0, 0)
(1, 0, 0, 0, 1, 7, 0)
(2, 0, 0, 0, 2, 0, 0)
(7, 0, 0, 0, 215, 0, 0)
(8, 0, 0, 1, 0, 0, 0)
(12, 0, 0, 0, 4, 0, 0)
(28, 0, 9, 0, 71, 0, 0)
(11, 0, 8, 0, 6, 0, 0)
(27, 0, 16, 0, 32, 0, 0)
(15, 0, 2, 0, 4, 0, 0)
(29, 0, 0, 0, 19, 0, 0)
(9, 0, 0, 0, 12, 0, 0)
(19, 0, 5, 0, 1, 0, 1)
(4, 0, 17, 0, 53, 0, 1)
(13, 0, 0, 0, 12, 0, 17)
(6, 0, 0, 0, 2, 0, 0)
(8, 0, 0, 0, 2, 0, 0)
(1, 0, 0, 0, 5, 64, 0)
(24, 0, 0, 0, 13, 0, 0)
(25, 0, 173, 0, 359, 0, 0)
(23, 0, 0, 0, 841, 0, 0)
(18, 0, 21, 0, 165, 4, 0)
(1, 0, 24, 0, 26, 24, 0)
(21, 0, 0, 0, 9, 0, 0)
(7, 0, 1, 0, 0, 0, 0)
(26, 0, 793, 0, 1151, 0, 0)
(16, 0, 6, 0, 116, 0, 8)
(8, 0, 318, 0, 887, 0, 0)
(10, 0, 6, 0, 266, 0, 3)
(15, 0, 3, 5, 11, 0, 0)
(19, 0, 3, 0, 1, 0, 0)
(12, 0, 0, 0, 817, 0, 0)
(38, 0, 0, 0, 19, 0, 0)
(44, 0, 1, 0, 182, 0, 0)
(43, 0, 28, 0, 1999, 0, 0)
(11, 0, 15, 0, 2086, 0, 0)
(47, 0, 84, 0, 861, 76, 0)
(19, 0, 0, 0, 184, 50, 0)
(41, 0, 2964, 0, 4549, 0, 0)
(45, 0, 1248, 0, 1808, 0, 0)
(15, 0, 26, 0, 209, 352, 24)
(20, 0, 17, 0, 70, 6, 0)
(36, 0, 1768, 0, 2253, 0, 0)
(10, 0, 504, 0, 876, 0, 0)
(48, 0, 80, 18, 412, 68, 0)
(17, 0, 4, 0, 261, 14, 0)
(42, 0, 5345, 0, 7519, 0, 0)
(13, 0, 1281, 0, 1855, 0, 0)
35
Table 3 (continued)
Tensor
(2, 4, 18, 28)
(2, 4, 22, 22)
(2, 5, 12, 26)
(h11 , h21 , S)
Boundaries
(23, 53, 307)

(23, 53, 307)
(22, 46, 309)
(22, 46, 309)
(31, 55, 330)
(31, 55, 330)
(26, 44, 293)
(26, 44, 293)
(44, 26, 293)
(44, 26, 293)
(25, 37, 279)
(24, 30, 261)
(24, 30, 261)
(30, 24, 261)
(30, 24, 261)
(36, 42, 327)
(42, 36, 327)
(31, 31, 270)
(31, 31, 270)
(39, 39, 313)
(39, 39, 313)
(11, 131, 536)
(7, 103, 399)
(14, 98, 501)
(9, 81, 344)
(10, 82, 361)
(13, 85, 367)
(16, 88, 393)
(8, 68, 329)
(14, 74, 341)
(10, 58, 309)
(13, 61, 335)
(21, 69, 344)
(20, 56, 313)
(16, 40, 261)
(19, 43, 279)
(28, 52, 341)
(52, 28, 341)
(20, 32, 261)
(32, 44, 313)
(23, 23, 247)
(31, 31, 268)
(31, 31, 279)
(32, 32, 273)
(32, 32, 273)
(39, 39, 311)
(8, 80, 341)
(23, 59, 327)
(23, 59, 327)
1120
1202
1300
1400
1768
1904
854
976
1024
1166
736
686
724
830
940
1060
1276
896
1024
1036
1184
864
648
1080
864
864
1080
1296
648
1080
864
1080
1728
1788
1416
1476
2160
3888
1668
2412
1356
1968
2460
2040
2592
2100
528
780
858
Nr.
49
18
46
16
14
1
27
9
11
4
18
28
19
29
15
21
17
22
8
10
1
12
43
1
44
42
22
20
45
18
11
13
16
48
46
49
2
35
17
30
21
47
19
28
50
31
8
1
2
Types
(49, 0, 115, 10, 266, 4, 2)
(18, 0, 274, 2, 803, 0, 1)
(46, 0, 112, 10, 511, 24, 0)
(16, 0, 455, 0, 931, 0, 39)
(14, 0, 0, 14, 0, 0, 0)
(1, 0, 12, 2, 23, 0, 0)
(27, 0, 0, 0, 7, 0, 0)
(9, 0, 0, 0, 15, 0, 0)
(11, 0, 8, 0, 16, 0, 0)
(4, 0, 3, 0, 8, 0, 0)
(18, 0, 0, 0, 12, 0, 0)
(28, 0, 34, 0, 72, 0, 0)
(19, 0, 31, 0, 27, 0, 0)
(29, 0, 33, 0, 44, 0, 0)
(15, 0, 1, 0, 4, 0, 0)
(21, 0, 0, 0, 1, 0, 0)
(17, 0, 0, 0, 0, 0, 1)
(22, 0, 0, 0, 0, 0, 0)
(8, 0, 2, 0, 8, 0, 0)
(10, 0, 0, 0, 6, 0, 0)
(1, 0, 0, 0, 22, 0, 0)
(12, 0, 0, 0, 1292, 0, 0)
(43, 0, 0, 0, 71, 0, 0)
(1, 38, 255, 154, 1123, 158, 0)
(44, 0, 1113, 0, 1977, 0, 0)
(42, 0, 0, 0, 404, 0, 0)
(22, 0, 169, 0, 1011, 168, 0)
(20, 0, 88, 68, 582, 0, 61)
(45, 0, 5089, 0, 7025, 0, 0)
(18, 0, 497, 154, 1704, 0, 0)
(11, 0, 3162, 0, 7156, 0, 0)
(13, 16, 2235, 231, 2695, 81, 152)
(16, 0, 147, 150, 147, 0, 0)
(48, 0, 138, 0, 85, 0, 0)
(46, 0, 120, 0, 251, 0, 3)
(49, 0, 274, 0, 173, 0, 1)
(2, 0, 0, 5, 22, 56, 0)
(35, 0, 0, 0, 2, 0, 0)
(17, 0, 383, 1, 319, 1, 0)
(30, 0, 0, 1, 8, 0, 0)
(21, 0, 0, 0, 4, 0, 0)
(47, 0, 43, 0, 73, 0, 0)
(19, 0, 0, 0, 1, 4, 0)
(28, 0, 27, 0, 65, 0, 0)
(50, 0, 1, 0, 0, 0, 0)
(31, 0, 0, 0, 8, 0, 3)
(8, 0, 0, 0, 32, 0, 0)
(1, 0, 60, 0, 252, 82, 0)
(2, 0, 0, 0, 5, 0, 0)
36
Table 3 (continued)
Tensor
(h11 , h21 , S)
Boundaries
(2, 6, 8, 38)
(14, 50, 289)

(21, 57, 333)
(26, 50, 313)
(28, 52, 331)
(22, 34, 265)
(25, 37, 271)
(33, 33, 275)
(14, 62, 299)
(15, 63, 307)
(16, 64, 331)
(10, 46, 253)
(16, 40, 249)
(19, 43, 255)
(22, 46, 275)
(22, 46, 319)
(13, 25, 215)
(15, 27, 239)
(17, 29, 223)
(18, 30, 223)
(20, 32, 241)
(20, 32, 261)
(22, 34, 253)
(26, 38, 283)
(29, 17, 223)
(32, 20, 241)
(32, 20, 261)
(38, 26, 275)
(21, 21, 235)
(25, 25, 243)
(25, 25, 243)
(28, 28, 253)
(28, 28, 253)
(31, 31, 259)
(33, 33, 273)
(8, 80, 319)
(10, 82, 355)
(11, 83, 389)
(9, 57, 273)
(10, 58, 271)
(14, 62, 283)
(14, 62, 371)
(13, 49, 277)
(13, 49, 277)
(8, 32, 215)
(9, 33, 233)
(10, 34, 251)
(11, 35, 217)
(13, 37, 233)
(13, 37, 253)
720
1080
1080
1200
720
1080
1440
864
864
864
432
864
864
864
1152
864
864
1008
1080
1080
1008
1152
1080
1152
1512
1440
1512
864
1008
1152
1008
1440
1440
1728
768
768
768
768
768
768
1032
1032
1032
768
768
768
768
768
1032
(2, 6, 10, 22)
(2, 6, 14, 14)
Nr.
9
15
12
16
25
1
13
56
15
58
13
22
53
20
38
55
21
30
52
54
29
32
18
39
16
4
1
51
36
33
35
6
40
23
64
73
14
60
22
19
17
71
80
67
21
62
65
66
63
Types
(9, 0, 0, 0, 530, 0, 0)
(15, 0, 26, 0, 14, 0, 0)
(12, 0, 35, 0, 33, 0, 0)
(16, 0, 1, 0, 74, 0, 18)
(25, 0, 0, 0, 447, 0, 0)
(1, 0, 47, 0, 42, 0, 0)
(13, 0, 0, 0, 7, 0, 0)
(56, 0, 150, 0, 120, 0, 0)
(15, 0, 26, 0, 34, 0, 0)
(58, 0, 3, 0, 325, 0, 0)
(13, 0, 0, 0, 1774, 0, 0)
(22, 0, 76, 0, 297, 0, 6)
(53, 0, 26, 0, 36, 0, 0)
(20, 0, 0, 0, 61, 0, 0)
(38, 0, 0, 0, 8, 0, 0)
(55, 0, 0, 0, 2, 0, 0)
(21, 0, 20, 0, 18, 0, 0)
(30, 0, 5, 0, 0, 0, 0)
(52, 0, 1, 0, 7, 0, 0)
(54, 0, 5, 0, 59, 0, 0)
(29, 0, 30, 0, 60, 0, 0)
(32, 0, 10, 0, 55, 0, 0)
(18, 0, 1, 0, 2, 0, 0)
(39, 0, 0, 0, 11, 0, 0)
(16, 0, 1, 0, 5, 0, 0)
(4, 0, 1, 0, 1, 0, 0)
(1, 0, 0, 0, 3, 0, 0)
(51, 0, 25, 0, 10, 0, 0)
(36, 0, 0, 0, 1, 0, 0)
(33, 0, 0, 0, 21, 0, 0)
(35, 0, 0, 0, 16, 0, 0)
(6, 0, 0, 0, 17, 0, 0)
(40, 0, 0, 0, 7, 0, 0)
(23, 0, 0, 0, 1, 0, 0)
(64, 0, 44, 0, 73, 0, 1)
(73, 0, 109, 0, 385, 0, 0)
(14, 0, 7, 0, 34, 0, 0)
(60, 0, 204, 0, 806, 0, 18)
(22, 0, 123, 0, 422, 0, 4)
(19, 0, 4, 0, 99, 0, 0)
(17, 0, 77, 0, 21, 0, 0)
(71, 0, 56, 0, 68, 0, 0)
(80, 0, 56, 0, 68, 0, 0)
(67, 0, 12, 0, 52, 0, 0)
(21, 0, 154, 0, 197, 0, 0)
(62, 0, 105, 0, 0, 0, 0)
(65, 0, 24, 0, 23, 0, 0)
(66, 0, 7, 0, 37, 0, 0)
(63, 0, 49, 0, 75, 0, 0)
37
Table 3 (continued)
Tensor
(2, 7, 10, 16)

(2, 8, 8, 18)
(2, 10, 10, 10)
(h11 , h21 , S)
Boundaries
(15, 39, 241)

(15, 39, 253)
(16, 40, 263)
(17, 41, 265)
(18, 42, 263)
(20, 44, 255)
(21, 45, 273)
(18, 30, 221)
(21, 33, 243)
(26, 38, 271)
(26, 38, 271)
(23, 23, 225)
(23, 23, 229)
(25, 25, 265)
(34, 34, 347)
(31, 31, 271)
(7, 79, 325)
(6, 72, 325)
(12, 66, 315)
(12, 66, 315)
(6, 48, 249)
(6, 48, 249)
(13, 49, 249)
(13, 49, 249)
(13, 49, 259)
(13, 49, 259)
(21, 57, 313)
(21, 57, 313)
(16, 46, 257)
(16, 46, 257)
(28, 34, 257)
(28, 34, 257)
(3, 105, 381)
(3, 105, 381)
(7, 91, 371)
(7, 91, 371)
(3, 69, 277)
(3, 69, 277)
(7, 67, 279)
(7, 67, 279)
(9, 63, 265)
(11, 59, 319)
(11, 59, 319)
(7, 43, 231)
(7, 43, 231)
(9, 45, 243)
(9, 45, 243)
(13, 49, 251)
(13, 49, 251)
768
768
1032
1272
1272
1536
1536
1032
1272
1272
1272
1536
1032
1272
1272
544
468
528
648
702
414
456
504
576
756
864
588
672
576
654
1134
1296
320
328
640
656
320
328
640
656
984
824
832
824
832
824
832
1112
1120
Nr.
69
68
61
28
25
82
23
72
32
81
86
74
16
18
1
9
34
9
37
17
33
35
28
8
36
14
10
1
11
15
16
2
44
57
54
11
46
56
52
63
26
13
55
64
51
65
53
66
18
Types
(69, 0, 39, 0, 33, 0, 0)
(68, 0, 36, 0, 8, 0, 0)
(61, 0, 40, 0, 77, 0, 0)
(28, 0, 19, 0, 13, 0, 0)
(25, 0, 0, 0, 7, 0, 0)
(82, 0, 1, 0, 2, 0, 0)
(23, 0, 8, 0, 14, 0, 0)
(72, 0, 18, 0, 389, 0, 0)
(32, 0, 18, 0, 9, 0, 0)
(81, 0, 0, 0, 49, 0, 0)
(86, 0, 0, 0, 49, 0, 0)
(74, 0, 8, 0, 8, 0, 0)
(16, 0, 16, 0, 20, 0, 0)
(18, 0, 0, 0, 34, 0, 0)
(1, 0, 0, 0, 12, 0, 8)
(9, 0, 0, 0, 1, 0, 0)
(34, 0, 0, 0, 86, 0, 0)
(9, 0, 0, 0, 140, 0, 0)
(37, 0, 53, 0, 379, 52, 0)
(17, 0, 10, 0, 84, 4, 0)
(33, 0, 2103, 0, 2548, 0, 0)
(35, 0, 345, 0, 344, 0, 0)
(28, 0, 1194, 0, 3295, 0, 0)
(8, 0, 649, 0, 1683, 0, 0)
(36, 0, 394, 8, 781, 114, 0)
(14, 0, 562, 1, 739, 0, 6)
(10, 0, 55, 0, 126, 0, 0)
(1, 0, 0, 0, 187, 8, 0)
(11, 0, 101, 0, 113, 0, 0)
(15, 0, 127, 0, 203, 0, 0)
(16, 0, 0, 0, 11, 4, 0)
(2, 0, 22, 0, 10, 0, 0)
(44, 0, 0, 0, 67, 0, 0)
(57, 0, 0, 0, 377, 0, 0)
(54, 0, 58, 0, 62, 0, 0)
(11, 0, 37, 0, 169, 0, 0)
(46, 0, 225, 0, 764, 0, 0)
(56, 0, 771, 0, 1060, 0, 0)
(52, 0, 100, 0, 148, 0, 3)
(63, 0, 85, 11, 475, 0, 0)
(26, 0, 1, 0, 1, 0, 0)
(13, 0, 12, 0, 0, 0, 0)
(55, 0, 18, 0, 0, 0, 0)
(64, 0, 21, 0, 49, 0, 0)
(51, 0, 94, 0, 23, 0, 0)
(65, 0, 265, 0, 651, 0, 1)
(53, 0, 98, 14, 85, 0, 0)
(66, 0, 22, 1, 76, 0, 0)
(18, 0, 65, 4, 72, 2, 0)
38
Table 3 (continued)
Tensor
(3, 8, 8, 8)
(4, 4, 6, 22)
(4, 4, 7, 16)
(4, 4, 10, 10)
(h11 , h21 , S)
Boundaries
(15, 51, 271)

(9, 33, 231)
(9, 33, 231)
(17, 41, 247)
(17, 41, 247)
(9, 27, 193)
(13, 25, 207)
(17, 29, 215)
(17, 29, 215)
(19, 31, 231)
(19, 31, 231)
(19, 31, 235)
(19, 31, 235)
(27, 39, 299)
(27, 39, 299)
(35, 23, 243)
(35, 23, 243)
(11, 47, 283)
(13, 25, 213)
(9, 57, 289)
(13, 61, 289)
(6, 42, 223)
(12, 48, 255)
(12, 48, 256)
(9, 33, 211)
(13, 37, 224)
(17, 41, 243)
(19, 43, 261)
(19, 43, 276)
(21, 45, 283)
(16, 28, 219)
(18, 30, 221)
(24, 36, 261)
(31, 31, 283)
(11, 59, 283)
(14, 50, 261)
(15, 39, 233)
(17, 41, 283)
(20, 32, 285)
(23, 23, 217)
(4, 70, 279)
(9, 69, 319)
(8, 62, 263)
(10, 64, 299)
(5, 53, 232)
(9, 57, 249)
(9, 57, 252)
(4, 46, 219)
(6, 48, 223)
1312
640
656
1112
1120
984
640
824
832
1648
1664
1112
1120
1112
1120
2224
2240
880
1120
546
330
354
510
408
438
600
570
534
840
420
348
402
954
588
288
432
720
864
1008
1008
380
308
430
406
440
512
640
332
296
Nr.
16
50
62
67
21
25
48
12
49
19
59
15
24
14
1
22
2
9
12
37
12
30
11
28
8
10
1
14
15
35
13
34
43
17
14
15
13
5
18
12
76
67
22
16
77
74
21
14
66
Types
(16, 0, 1, 0, 3, 0, 0)
(50, 0, 438, 0, 421, 0, 0)
(62, 0, 160, 0, 188, 0, 0)
(67, 0, 0, 0, 10, 0, 0)
(21, 0, 9, 0, 11, 0, 0)
(25, 0, 1, 0, 1, 0, 0)
(48, 0, 1, 0, 0, 0, 0)
(12, 0, 138, 0, 172, 0, 0)
(49, 0, 138, 0, 516, 0, 0)
(19, 0, 0, 0, 2, 0, 0)
(59, 0, 0, 0, 15, 0, 0)
(15, 0, 72, 0, 56, 0, 0)
(24, 0, 10, 0, 19, 0, 0)
(14, 0, 0, 0, 32, 0, 0)
(1, 0, 6, 0, 62, 0, 0)
(22, 0, 0, 0, 5, 0, 0)
(2, 0, 0, 0, 1, 0, 0)
(9, 0, 0, 0, 0, 0, 0)
(12, 0, 0, 0, 3, 0, 0)
(37, 0, 0, 0, 7, 0, 0)
(12, 0, 0, 0, 544, 0, 0)
(30, 0, 1147, 0, 2200, 0, 0)
(11, 0, 0, 0, 497, 110, 0)
(28, 0, 0, 0, 78, 0, 0)
(8, 0, 612, 0, 621, 0, 0)
(10, 0, 312, 2, 366, 0, 0)
(1, 0, 30, 21, 105, 4, 0)
(14, 0, 54, 0, 160, 0, 0)
(15, 0, 0, 2, 0, 0, 0)
(35, 0, 2, 0, 3, 0, 0)
(13, 0, 289, 0, 212, 0, 0)
(34, 0, 182, 0, 485, 0, 0)
(43, 0, 0, 0, 0, 0, 3)
(17, 0, 10, 0, 8, 0, 0)
(14, 0, 0, 0, 4, 0, 0)
(15, 0, 5, 0, 1, 0, 0)
(13, 0, 0, 3, 0, 0, 0)
(5, 0, 0, 0, 22, 0, 0)
(18, 0, 0, 0, 21, 0, 0)
(12, 0, 0, 3, 0, 0, 0)
(76, 0, 0, 0, 16, 0, 0)
(67, 0, 0, 0, 1, 0, 0)
(22, 0, 0, 14, 67, 0, 0)
(16, 0, 56, 0, 34, 0, 0)
(77, 0, 1070, 0, 1264, 0, 0)
(74, 0, 218, 0, 273, 0, 0)
(21, 0, 94, 32, 73, 4, 0)
(14, 0, 1324, 0, 1409, 0, 0)
(66, 0, 712, 0, 285, 0, 0)
39
Table 3 (continued)
Tensor
(4, 6, 6, 10)
(6, 6, 6, 6)
(1, 2, 2, 10, 10)

(1, 2, 4, 4, 10)
(1, 4, 4, 4, 4)
(2, 2, 2, 4, 10)
(2, 2, 2, 6, 6)
(1, 1, 1, 1, 7, 16)
(h11 , h21 , S)
Boundaries
Nr.
Types
(7, 43, 215)

(13, 49, 247)
(10, 40, 219)
(10, 40, 223)
(15, 39, 321)
(14, 20, 195)
(14, 38, 229)
(12, 24, 197)
(24, 24, 225)
(5, 69, 267)
(3, 59, 223)
(3, 43, 207)
(7, 47, 215)
(5, 37, 203)
(9, 41, 211)
(11, 27, 199)
(13, 29, 203)
(15, 23, 199)
(17, 25, 203)
(5, 29, 235)
(5, 53, 248)
(5, 41, 212)
(6, 30, 196)
(11, 35, 213)
(4, 40, 213)
(13, 13, 200)
(9, 45, 251)
(9, 45, 251)
(13, 25, 219)
(15, 27, 245)
(21, 9, 211)
(21, 9, 211)
(21, 9, 211)
(3, 51, 235)
(9, 33, 223)
(9, 33, 223)
(17, 17, 215)
(17, 17, 215)
(17, 17, 215)
(23, 23, 217)
260
448
550
744
1056
888
416
320
448
400
368
400
736
368
800
736
736
1136
1136
224
112
106
106
160
246
240
208
320
320
416
272
272
416
288
384
384
576
576
576
72
79
80
81
86
68
101
41
50
21
7
37
58
55
38
18
56
47
68
69
19
76
75
65
67
58
89
202
87
101
213
79
109
224
106
149
189
135
157
285
32
(79, 0, 1237, 0, 587, 0, 0)

(80, 0, 15, 0, 5, 0, 0)
(81, 0, 0, 0, 59, 0, 0)
(86, 0, 0, 0, 3, 0, 0)
(68, 0, 0, 0, 35, 0, 0)
(101, 0, 0, 0, 1, 0, 0)
(41, 0, 0, 0, 1, 0, 1)
(50, 0, 0, 0, 1, 0, 0)
(21, 0, 0, 0, 3, 0, 0)
(7, 0, 4, 0, 14, 0, 0)
(37, 0, 142, 0, 226, 0, 0)
(58, 0, 30, 0, 30, 0, 0)
(55, 0, 0, 0, 6, 0, 0)
(38, 0, 63, 0, 100, 0, 0)
(18, 0, 2, 0, 18, 0, 2)
(56, 0, 0, 0, 6, 0, 0)
(47, 0, 0, 0, 4, 0, 0)
(68, 0, 2, 0, 2, 0, 0)
(69, 0, 0, 0, 1, 0, 0)
(19, 0, 2, 0, 0, 0, 0)
(76, 0, 1, 0, 1, 0, 0)
(75, 0, 1, 0, 1, 0, 0)
(65, 0, 25, 0, 40, 0, 0)
(67, 0, 0, 30, 0, 0, 0)
(58, 0, 0, 5, 0, 0, 0)
(89, 0, 0, 0, 1, 0, 0)
(202, 0, 0, 0, 2, 0, 0)
(87, 0, 19, 0, 37, 0, 0)
(101, 0, 14, 0, 14, 0, 0)
(213, 1, 0, 0, 0, 0, 0)
(79, 0, 0, 0, 1, 0, 0)
(109, 0, 0, 0, 1, 0, 0)
(224, 0, 0, 0, 1, 0, 0)
(106, 0, 31, 0, 24, 0, 0)
(149, 0, 0, 0, 1, 0, 0)
(189, 0, 0, 0, 1, 0, 0)
(135, 0, 0, 0, 2, 0, 0)
(157, 0, 0, 0, 2, 0, 0)
(285, 0, 0, 0, 2, 0, 0)
(32, 0, 0, 2, 0, 0, 0)
and the last column gives the total number of distinct (as defined in Section 1.3) standard
model spectra.
In total we found solutions to the tadpole equations for 44 of the 168 Gepner Models
and for 333 of the 5403 MIPFs. For 4079 MIPFs we did not even find any standard model
four stack configuration. In 649 cases there are four-stacks, but no solution to the tadpole
equations and in 342 cases we could not rule out the existence of solutions. The total
number of standard model configurations that exist is 45051902; of these 1635985 yield
40
Table 4
Distribution of chiral standard model types, distinguished by CP group and the U (2)b anomaly. In the Higgs
column, 2n implies that there are n supersymmetric Higgs pairs (2, 12 ) + (2, 12 ) that are chiral with respect
to U (2)b (the sign is just a convention)
Type
Quarks
Leptons
Higgs
Total
0
1
1
1
2
3
3
3
3
3
3
3
3
4
5
5
5
0
3
9
9
0
3
3
3
3
9
9
9
9
0
3
3
9
0
3
3
9
0
1
3
1
3
3
3
5
9
0
1
3
9
0
0
6
0
0
4
6
2
0
12
6
4
0
0
2
0
0
10564
32
1
22
49661
141
24
240
740
24
95
1
116
116304
2
1507
46
solutions. Many of the latter have the same standard model spectrum, which reduces the
total to 179520.
The 168 models are listed in a particular order, starting with the ones with the smallest
number of factors. More or less coincidentally this order corresponds rather well to the
degree of difficulty, in decreasing order. The number of (a, b, c, d) stacks is very small
at the bottom of the list, and increases to millions at the top, but only for particular tensor
combinations. However, despite the large number of candidates, the tensor combinations at
the top of the list yield very few solutions to the tadpole equations. Some tensor combinations with a large number of solutions have a recognizable feature in common: the values
of k + 2 typically takes values that factorize into powers of 2 and 3. However, there are
counterexamples in both directions.
In Table 3 we list the 333 modular invariants for which tadpole solutions were found.
Column two specifies the MIPF in terms of the Hodge numbers of the corresponding
type-IIB CalabiYau compactification4 and the number of singlets in a Heterotic compactification. Unfortunately, this does not always specify the MIPF uniquely. However, in
most cases they are distinguished by the number of boundaries, listed in column three.
Column 4 contains the (unique) label of the MIPF.5 In the last column we list the number
of different standard model spectra organized according to type. The first six entries refer
to types 0, . . . , 5 defined above, with a massless B L vector boson. The last entry is the
4 Hence h is the number of hyper multiplets (not including the universal one from the gravitational sector)
11
and h21 the number of vector multiplets of the closed type IIB string before orientifolding.
5 This label specifies a simple current subgroup and a matrix X that defines the modular invariant. It is difficult
to list all this information efficiently, but it is available from the authors on request.
41
number of type 0 spectra with a massive B L. In principle, this could have occurred for
type 1 as well, but no examples were found.
Table 4 lists the total number of solutions for each type, where we distinguish chiral
subtypes for types 1, 3, and 5. These subtypes are defined in terms of the contribution of
the quark doublets, the lepton doublets and the Higgses to the U (2)b anomaly. Since this
anomaly cancels, the subtypes are characterized by two independent parameters. It is easy
to see that the quark contribution must be an odd multiple of three (which we have chosen
to be positive), the lepton contribution must be odd, and the Higgs contribution even. If the
lepton contribution is larger than three the spectrum contains chiral mirror leptons, i.e.,
mirror pairs of leptons that are chiral with respect to the full CP group, but non-chiral with
respect to the standard model gauge group. For example, to get a lepton SU(2) anomaly
of 5 one must have the representation 4(1, 2 , 12 ) + (1, 2 , 12 ) (up to purely non-chiral
mirrors). Strictly speaking such models can perhaps not be described as just the chiral
standard model, but we admitted them as a curiosity. The same phenomenon is possible
for quarks as well, but we did not find any examples. Note that the last column shows twice
the number of chiral supersymmetric Higgs pairs (1, 2, 12 ) + (1, 2, 12 ), so that the number
of such Higgs pairs in the entire set can be 0, 1, 2, 3, 4 or 6. Interestingly only 0 and 3 were
found for type 1.
5.2. Features of found spectra
As discussed above there are several features we have considered as relevant to distinguish standard model spectra. We will now present which values these parameters take and
see if there is any notable structure.
Looking at these distributions one could be tempted to draw statistical conclusions.
Even if one would adopt this point of view, there are several reasons why one should be
careful. For one, our search algorithm was set up to maximize the number of different
solutions (see Section 1.3). Also there are relations between different distributions (for
example, if stack c is realized as SO(2)c , ec and c come from the same branes, hence
have the same mirror-distribution). Another bias to keep in mind is caused by the fact that
equations are much easier to solve for a small number of branes. This biases the search
towards fewer branes.
First we will look at the number of mirrors of the standard model particles. The total
chirality of the standard model particles is fixed to 3. We allowed, however, for additional
non-chiral pairs, the so-called mirrors. The plot for the number of mirrors is similar for
all standard model particle hence we show only the one (e, ) mirror pairs and the total
number of mirrors of standard model particles in a model (Fig. 1). It is interesting to see
that the distribution of number of mirrors is sharply peeked at zero mirrors. From the total
plot we see that 2018 models have no mirrors at all and that the distribution peeks at 4
mirrors.
The only bifundamentals coming from string states between the standard model branes
which should not be chiral are strings stretched between branes a and d. These particles
would be leptoquarks. In Fig. 2 we plot the number of non-chiral leptoquarks. The distribution of leptoquarks with equal sign lepton and baryon number is peaked at zero. This is
42
Fig. 1. Mirrors.
Fig. 2. Non-chiral leptoquarks.
the only non-chiral exotic that peaks at zero (apart from individual mirror distributions,
as noted above). The distribution of the opposite sign leptoquarks distribution peaks at 2.
Finally, one can have massless non-chiral representations coming from strings that have
both ends on the same SM brane (or one end on the conjugate brane). These states are
symmetric, anti-symmetric or adjoint tensors of the standard model gauge group. The distributions of non-chiral pairs of all these particles peak at a non-zero value, except for the
ones where in a majority of the models that particular state does not exist.6 In Fig. 3 we
plot the total number of these particles in a model and the distribution of adjoints of SU(3)
as an example.
5.3. Higgs
In supersymmetric extensions of the standard model the Higgs always comes in nonchiral (w.r.t. to GSM ) pairs so the Higgsinos will not generate an Y-anomaly. In Fig. 4 we
plot the number of Higgsino pairs, which is thus equal to half the number of standard model
6 Adjoints of SO(2) are counted as anti-symmetric tensors, adjoints of Sp(2) as symmetric tensors. Massless
anti-symmetric representations of U (1) actually have no massless states at all, and were not counted.
43
Fig. 3. Non-chiral tensor representations.
Higgs doublets in that model. As is clear from the picture, the number of Higgs pairs peaks
at three. The maximum number of Higgs pairs we found is 56. Note that there also models
with no Higgs. These models have an obvious deficiency, although it is conceivable that
some (composite-)particle from the hidden sector will play the role of the Higgs.
For types based on U (2)b (type 1, 3 and 5) there is a possibility for the Higgs of being
chiral with respect to U (2)b . This chirality was ignored in Fig. 3, but the chirality distribution is displayed in Table 4, as explained above.
5.4. Hidden branes
Our search algorithm was set up to maximize the different standard model spectra, not
the hidden brane degrees of freedom. Hence we present in all plots and tables the number
of solutions where we identified solutions with different hidden sectors, but which are
otherwise equivalent. As explained in Section 1.3 we did, however, construct all solutions
containing 0, 1 and 2 extra branes for all standard model brane stacks, three extra branes
for all stacks with less than 400 candidate hidden CP groups, and four extra branes for
less than 100 candidate hidden CP groups. In many other cases we attempted to extract
a solution from the tadpole equations without a limit on the number of branes. The latter
searches were stopped as soon a one solution was found and are limited by computer
time constraints. Therefore, they are not systematic. In Fig. 4 we show the total number of
solutions found for each hidden brane multiplicity. This plot is based on a total of 10526078
44
Fig. 4. Number of Higgs.
Fig. 5. Hidden sector of all solutions.
solutions. These solutions are different only in the sense that their CP multiplicities are
distinct. Undoubtedly there are still some equivalences in this set that were not taken into
account. The number of solutions with 0, 1 and 2 branes is 31215, 148324 and 1170556,
respectively. As one can see, the number of solutions grows very fast with the number of
hidden branes, but is cut off beyond three branes for the reasons explained above. One can
make a plausible guess what the picture might really look like, had we been able to push
the search for solutions much further. Most likely, it would continue growing dramatically
for quite a while. Since the number of candidate CP groups is typically a few hundred, the
45
distribution necessarily peaks well below that number, but this could easily be around ten
or twenty. It is clear that the total number of solutions with distinct standard model plus
hidden spectrum can be many orders of magnitude larger than those we found. In Fig. 5
we plot the total dimension of the total hidden gauge group. The biggest gauge group we
encountered has dimension 780. The biggest factors we found are SO(18), Sp(24) and
U (18).
5.5. Gauge couplings
A remarkable property of the standard model is that its gauge group fits naturally within
the gauge group SU(5). This explains two empirical facts: the observed quantization of
electric charge, and the observed convergence of the coupling constants at short distances
(although with present data, the latter success only survives if the additional assumption of
supersymmetry is made). Both of these nice features seem to be lost in the class of models
considered here.
Fractional charges are inevitable if there exist additional branes outside the standard
model. Strings stretching between the standard model and the hidden branes yield particles
(fundamental or QCD bound states) with half-integer electric charge. Although they may
be non-chiral (as they are in all models presented here) or even completely absent from the
massless spectrum, they exist inevitably as massive open string states. Interestingly, it is
not completely straightforward to realize group-theoretical unification in heterotic strings
either. In the standard realizations with KacMoody level 1 one ether gets SU(5) without
a massless Higgs boson to break it, or one gets SU(3) SU(2) U (1) with additional
(though not necessarily massless) fractional charges [18]. These problems can be avoided,
for example, by considering higher KacMoody levels, but it is difficult to argue that charge
quantization is a natural property of string theory.
Heterotic SU(5) and SO(10) models do explain the observed coupling constant convergence, but they make a troublesome prediction for the unification scale, which is off
by two to three orders of magnitude. It is on this point that open string models have an
advantage, simply because there is no such prediction [24]. But there is also no prediction
for the unification of the gauge couplings, because they emerge from dilaton couplings of
four, a priori unrelated, branes.
In [78] it is argued that a realization of a supersymmetric extension of the standard
model with intersecting branes naturally leads to a model where, respectively, branes a and
d and c and b wrap the same cycles. This leads then at the string scale to the following
relation between the three standard model coupling constants:
2 1
1
1
=
+
.
Y
3 s
w
(20)
That would mean that these models do not necessarily have full unification, but they do
reproduce a relation which is compatible with the SU(5) relation between the coupling
constants.
To do a proper calculation of gauge unification we would have to assume a unification
scale and evolve the couplings downward from that scale. Since we are in a rational point
of the moduli space, this scale is fixed to a value of order the string scale. The massless
46
spectrum we obtain contains, in general, several non-chiral particles that are presumably
artifacts of the rational points. If one assumes that all these particles remain massless until
the TeV scale, one should take them into account in the renormalization group flow. More
ambitiously, one could work out the masses and the gauge couplings as a function of the
moduli near the rational point, and take all of this into account. Here we will only address
a simpler question, namely if there is any evidence for relations among the couplings at the
string scale. If there is no relation, then clearly one might still get the correct low energy
gauge couplings by starting outside the unification point in the space of couplings, and
compensate this with exotic matter in the renormalization group flow, but the concept of
unification is then anyway lost.
The coupling constants can be computed as follows. Up to a universal factor, they are
given by
1
= R0a e
ga2
for CP-groups of orthogonal and symplectic type and
1
= (R0a + R0ac )e = 2R0a e
ga2
for unitary groups, with the conventional normalization Tr T a T b = 12 ab for the generators
in all three cases. This immediately gives the following expression for the ratio of weak
and string couplings:
g22
g32
R0a
,
R0b
where = 1 for spectra of types 1, 3 and 5, and = 12 for spectra of type 0, 2 and 4. The
canonically normalized U (1) generator of stack a is Ya = 1 1, for stack d it is Yd = 1 1,
6
and for stack c it is Yc = 1 1 for complex boundaries, and Yc = 12 3 for real ones. The
2
standard model U (1) factor Y is the given by
1
1
1
Ya Yc Yd
6
2
2
for types 0 and 1, and
1
1
Ya Yc Yd
6
2
for types 2, 3, 4 and 5. This leads to the following expression for the coupling constants
1
1 1
1 1
1 1
=
+
+
2
2
2
6 ga
2 gc
2 gd2
gY
for types 0 and 1, and
1
1 1
1
1 1
=
+ 2+
2
2
6 ga
2 gd2
gc
gY
47
Fig. 6. Gauge coupling constants at the string scale.
for types 2, 3, 4 and 5. Hence in both cases the expression for the couplings in terms of
reflection coefficients is the same:

1
1
R0a + R0c + R0d .
=
3
gY2
The fact that the same expression is obtained is due to the fact that an orthogonal or symplectic c-stack can be viewed as a limit of a unitary one, with the two conjugate branes
moved on top of the orientifold plane. From this expression we obtain the following formula for sin2 W = gY2 /(g22 + gY2 ):
sin2 W =
R0b
R0b +
1
6 R0a
+ 12 R0c + 12 R0d
In Fig. 6 we plot sin2 (w ) against the ratio s /w . In this plot the value for SU(5) unification are also indicated, as well as its renormalization group flow. The solid line denotes
the upper limit on sin2 (w ). Clearly the models we found occupy a substantial part of the
allowed region. It is amusing to see that this point falls neatly in the middle of the cloud
formed by our models. One aspect of the plot that attracts attention is a faint line7 roughly
7 This line is not very visible in the picture. A picture with higher resolution can be downloaded from
http://www.nikhef.nl/~t58/GaugeCouplings.ps.
48
in the middle of the cloud. It turns out to be described by relation (20). Approximately
10% of the models is on that line. Not surprisingly, these models are characterized by the
fact that the reflection coefficients for, respectively, branes a and d and branes b and c are
the same, which is precisely the RCFT equivalent of the condition that [78] used for unification. So although a portion of the models we found has a relation between the coupling
constants compatible with SU(5) unification, this is certainly not a generic feature.
5.6. Varying number of chiral families
For a limited set of models we repeated our search for the standard model spectrum
with a different number of families, in order to get a feeling what the constraint of having 3
chiral families means. Because more chiral particles implies more intersections one could
expect that getting a spectrum with more chiral families is harder. Indeed this tendency is
visible in plots Fig. 7. Because the number of standard model brane configurations to be
analyzed for 1 and 2 families was so big, we performed a search for 1 up to 9 families for
18 models and a search for 3 up to 9 models for a bigger set of 78 models.
But there is more structure. For example, odd types (the ones based on U (2)b ) are
much more abundant with even families. If one takes a closer look and distinguishes by
subtype, that is the way the U (2)b anomaly cancels, this becomes a bit clearer. Both states
coming from [b, x] and [b , x] can contribute to the total chirality of the SU(2)W doublets.
They do, however, contribute oppositely to the U (2) anomaly. This leaves the possibility to
cancel the anomaly between the different U (2) chiralities of a doublet. In the even family
number models solutions of the subtype where the anomaly cancels fully between the
different U (2) chiralities of a given doublet are approximately three orders of magnitude
more abundant than subtypes where the anomaly cancels partly between different doublets.
Apparently it is easy to have the intersection number of [b, x] equal to that of [b , x]. This
is also explains why there are so relatively few odd family number solutions with U (2)b ;
For the family number to add up to an odd number, one of the two intersection numbers
[b, x] and [b , x] has to be odd and the other even. The easy way of cancelling the anomaly
is not a possibility.
Under the assumption that more chiral intersections are harder to get, we can also understand why 4, 2, 3 and 5 are typically less abundant then types 0 and 1. The latter types
are based on U (1)c and have all weak singlets coming from their own intersection, with c
or cc . The former types, however, have a real c-brane and both lepton singlets (and both
quark singlets) come from the same intersection. This means that they have 2 times the
number of families less chiral intersections.
A similar effect does not occur if we exchange U (2)b for Sp(2)b , because as discussed
above, both U (2)b chiralities contribute to one doublet. Hence both types based on U (2)b
and Sp(2)b have the same number of chiral b-brane intersections.
Also in the other types there is a clear even/odd effect. But if one looks, for odd/even
family number separately, at the trend in the number of solutions for a certain type against
family number it is to good approximation an exponential drop.
One feature that deserves mentioning is the occurrence of type 1 models with a massive
B L for 1, 2 and 4 families, something we did not find for three families. These models
have as the gauge group that couples to the standard model particles precisely the standard
Fig. 7. Solutions for varying number of chiral families.
49
50
model. That this did not occur for three families is probably a matter of statistics. As we
have seen type 1 models with three families are already rare, and furthermore from the
search for 1, 2 and 4 family number models we know that a massive B L only happens
in a few percent of the type 1 solutions.
6. The simplest case

Here we will discuss in some detail one of the simplest cases where a solution was
found. It occurs for the tensor product (1, 1, 1, 1, 7, 16). This has 1944 primaries, 81 of
which are simple currents forming a discrete group Z3 Z3 Z3 Z3 . Since there are
only odd factors, there is just one orientifold choice for each MIPF.
After removing all permutations of the identical factors, this tensor product has 34 simple currents MIPFs. Only one of them has standard-model-like (a, b, c, d) branes. This
MIPF is defined by a simple current group that is isomorphic to Z3 Z3 Z3 , and is
generated by the three currents:

0, 0, 0, 0, (1, 1, 0), (7, 3, 0), 0 ,

0, 0, 0, 0, (1, 1, 0), (7, 3, 0), 0 ,

0, 0, (1, 1, 0), 0, 0, (7, 5, 0), (16, 16, 0) .
Our notation is as follows: between the square brackets we indicate the decomposition of
the tensor product representation into minimal model representations, with the first entry
in each square bracket representing the spacetime sector. The minimal model representations are indicated in the usual way as (l, q, s), except for the vacuum, which is denoted
as 0. Note that only orbit representatives are shown: the minimal model representations
are subject to field identification, and the entire tensor representation is actually just one
representative of a chiral algebra orbit. Hence other, but equivalent, expressions can be
given.
The matrix X yielding the MIPF under consideration is

1 1 0 1
0 1 1 .
3 1 1 1
This yields a closed string spectrum corresponding to Hodge numbers (23, 23), and a heterotic spectrum with 217 singlets. There are 72 Ishibashi states and hence the same number
of boundary states. The total number of sets of (a, b, c, d) stacks is six, all of them of type 3.
Three sets saturate the dilaton tadpole condition, whereas the other three do not. We find
that for the former all other 17 tadpole conditions are also satisfied, whereas for the latter
three there is no possibility of adding branes in order to solve the tadpole conditions. The
three sets of boundaries yielding solutions can be described as follows:

set 1: a1 , ac1 b, bc (c) d, dc ,

set 2: a2 , ac2 b, bc (c) d, dc ,

set 3: a3 , ac3 b, bc (c) d, dc .
51
In other words, the b, c and d branes are always identical, and the complete set of boundaries we need is given by

a1 : 0, 0, 0, 0, 0, (1, 1, 0), (2, 2, 0) ,

ac1 : 0, 0, 0, 0, 0, (1, 1, 0), (2, 2, 0) ,

a2 : 0, 0, 0, 0, 0, (1, 1, 0), (8, 2, 0) 1 ,

ac2 : 0, 0, 0, 0, 0, (6, 4, 0), (8, 8, 0) 1 ,

ac3 : 0, 0, 0, 0, 0, (1, 1, 0), (8, 2, 0) 2 ,

ac3 : 0, 0, 0, 0, 0, (6, 4, 0), (8, 8, 0) 2 ,

b:
0, (1, 1, 0), (1, 1, 0), (1, 1, 0), (1, 1, 0), (6, 6, 0), 0 ,

bc : 0, (1, 1, 0), (1, 1, 0), (1, 1, 0), (1, 1, 0), 60, 0 ,

c:
0, 0, 0, 0, (1, 1, 0), (7, 7, 0), (16, 16, 0) ,

d:
0, 0, 0, 0, 0, (7, 7, 0), (16, 14, 0) ,

dc : 0, 0, 0, 0, 0, (7, 7, 0), (16, 14, 0) .
The three sets only differ in the choice of the U (3)-brane. The last two choices, a2 and
a3 and their boundary conjugates, are resolved fixed points of the supercurrent extension
of the tensor product, and they differ only in their fixed point degeneracy labels, indicated
here as 1 and 2. Not surprisingly, they yield the same spectrum. The spectra of sets 1 and 2
are distinct, and are as follows:
5(V , V , 0, 0)3 + 9(V , 0, V , 0)3 + 5(0, V , V , 0)3 + 3(0, V , 0, V )3
+ 3(0, 0, V , V )3 + 12(V , V , 0, 0)0 + 4(0, V , 0, V )0 + 8(Ad, 0, 0, 0)0
+ 4(A, 0, 0, 0)0 + 8(S, 0, 0, 0)0 + 4(0, A, 0, 0)0 + 3(0, 0, S, 0)0 + 4(0, 0, 0, A)0
and
3(V , V , 0, 0)3 + 9(V , 0, V , 0)3 + 5(0, V , V , 0)3 + 3(0, V , 0, V )3
+ 3(0, 0, V , V )3 + 4(V , V , 0, 0)0 + 4(0, V , 0, V )0 + 4(Ad, 0, 0, 0)0
+ 4(A, 0, 0, 0)0 + 6(S, 0, 0, 0)0 + 4(0, A, 0, 0)0 + 3(0, 0, S, 0)0 + 4(0, 0, 0, A)0 .
The notation is as follows: V denotes a vector, A anti-symmetric tensor, S a symmetric
tensor and Ad and adjoint representation. The denote complex conjugation, and the last
subscript denotes the net chirality of the representation. Our convention is to represent all
four-dimensional fermions as left-handed spinors. The precise meaning of, for example,
N (V , V , 0, 0)M is then 12 (N + M)(V , V , 0, 0) + 12 (N M)(V , V , 0, 0). Hence in
the first case, we have 3 chiral standard model SU(3) SU(2) U (1) multiplets (3, 2, 16 ),
plus a single mirror pair (3, 2, 16 ) + (3 , 2, 16 ). Note that the second factor is actually
U (2) instead of SU(2). The additional U (1) is anomalous, and the corresponding gauge
boson acquires a mass. The mirrors are non-chiral not just with respect to the standard
model group, but with respect to the full group (although this was not a priori required).
There is a second set of fully non-chiral mirrors of the same standard model representation
52
in both cases, six in the first and 2 in the second. These differ from the former ones by
having opposite U (1)b -charges. The tadpole equations satisfied by these two cases are in
fact identical.
Since these spectra only differ by the choice of the a-branes, the lepton and Higgs sectors
are identical. Both have 5 Higgs pairs (1, 2, 12 ) + (1, 2, 12 ), with the interesting feature
that 3 of them are chiral with respect to the U (2) group. Another noteworthy feature is the
absence of lepto-quarks.
Finally we present the closed spectrum. The only model-dependent feature concerns the
diagonal sector. It turns out that all relevant multiplicities are 1, and that all non-vanishing
Klein bottle coefficients are 1. The 23 closed string vector multiplets originate from 21 diagonal fields in the closed string partition function, and two off-diagonal ones. Of the first
21 only the symmetric components survive the Klein bottle projection, and the two offdiagonal ones are reduced to a single field, consisting of a symmetric and an anti-symmetric
component. Hence the projected closed string spectrum consists of a supergravity multiplet, one vector multiplet and 22 chiral multiplets originating from h11 and another 23
chiral multiplets originating from h21 .
7. Conclusions
We performed the first semi-systematic search for standard model spectra in open string
models. We presented 179520 models that have as their chiral spectrum just the standard
model. Most of the models do have non-chiral exotics and/or mirrors, for which we presented the abundances. The number of Higgs was left a free parameter. We found models
with as much as 56 Higgses. The distribution peaks at three pairs of Higgs. We also calculated the standard model gauge couplings and found (perhaps not surprisingly) no relation
that hinted at unification.
Although we are confident that within the Gepner models we constructed most realizations of the standard four-stack model, there is much room for improvement and
generalization.
More MIPFs of the Gepner models are known than the ones considered here. First
of all, there are diagonal invariants and their simple current modifications. It turns out
that in many cases (but not all) these can be obtained by simple currents, and hence are
already included. In other cases more work is needed, see, e.g., [79]. Furthermore, there
are exceptional SU(2) invariants, exceptional invariants of SU(2) tensor products [68], and
invariants related to interchanges of fixed points. Unfortunately, there is no boundary state
formalism available to deal with these cases. There also is no guarantee that we found
all orientifolds. Up to now, the orientifold degrees of freedom we use here (Klein bottle
currents and crosscap signs) has included everything known by other methods, but most
likely this is just a conjecture awaiting a counter example.
There are additional N = 2 coset CFTs and presumably even RCFTs that are not cosets.
Indeed, there is a much larger class of KazamiSuzuki [80] models. In principle, our
methods are applicable to these models as well, provided we know the exact spectrum.
In cases without field identification fixed points this can be computed straightforwardly,
but if there are fixed points the formalism has not been developed yet. Another class that
53
might be accessible with similar methods are the interchange orbifolds of identical minimal
or KazamiSuzuki models [81,82].
To get an idea how much is still missing, note that we encountered only 873 different
Hodge number pairs, while, for example, in [83] already more than 30 000 are presented.
For any accessible RCFT one may consider other brane constructions than the four stack
model, for example, a construction that would yield SU(5) unification at the string scale,
or broken versions of such a model. This would have the obvious advantage of restoring
gauge unification. One could also look for PatiSalam like models, and for models that
require a brane recombination mechanism to yield the standard model. As mentioned in
the introduction, examples of these types have been constructed already. In principle, there
is no reason why the standard model could not be realized in such a way, but of course
this opens a Pandoras box of possibilities. The number of solutions would explode even
more if we were to allow chiral exotics, a possibility that can never be rigorously ruled out
experimentally. But if indeed the quarks and leptons we know today provide misleading
information about the true chiral spectrum of nature, then the goal of finding the standard
model in string theory is essentially unachievable at present.
The obvious way to control this explosion of candidate standard models is to apply more
constraints. There are several criteria that might be used to reject some of the many models
on our list, such as absence of a Higgs candidate, the absence of plausible mechanisms to
break Sp(2)c , or to generate a B L gauge boson mass. However, it is hard to come up
with a criterium that is at the same time very effective and unquestionable.
One could also try to work out the Yukawa coupling of the standard model particles to
the Higgs sector. This requires open string three-point couplings, computable in principle
in RCFT, but not yet in practice. In cases with more than one supersymmetric Higgs pair,
there is the obvious problem of deciding which combination gets a vacuum expectation
value. This could be treated as a set of free parameters, and fitted to the observed couplings.
Doing that would requires a renormalization group evolution of those couplings, which in
its turn requires detailed knowledge of the full (chiral and non-chiral) light spectrum. Then
one still faces the problem of the moduli-dependence of the result. To some extent, this
may be studied by probing some directions in moduli space with a large number of rational
points, as we did for the gauge couplings. However, such an effort is probably premature.
Most likely an essential ingredient is missing for the understanding of Yukawa couplings,
and that is especially true for neutrino masses.
Fortunately, more experimental constraints can be expected in the coming years, especially from the LHC, and also from neutrino and astrophysics experiments. With more
experimental input we might come to the conclusion that another type of model should
be searched for, and with a different set of constraints. It is to be expected that the class
considered here will yield equally rich and abundant results in other cases.
Acknowledgements
We would like to thank Dieter Lst for suggesting a check of the results of [78]. We
would like to thank Jeff Templon, David Groep, and Davide Salomoni for assistance with
running jobs on the NIKHEF computer farm. The work of A.N.S. has been performed as
54
part of the program FP 52 of the Foundation for Fundamental Research of Matter (FOM),
and the work of T.P.T.D. and A.N.S. has been performed as part of the program FP 57 of
FOM. The work of A.N.S. has been partially supported by funding of the Spanish Ministerio de Ciencia y Tecnologa, Project BFM2002-03610. The work of L.R.H. was supported
in part by the Federal Office for Scientific, Technical and Cultural Affairs through the Interuniversity Attraction Poles ProgrammeBelgian Science Policy P5/27.
Appendix A. Boundary and Ishibashi orbits

To compute all annuli using the formalism of [64] it is sufficient to compute all reflection
coefficients, and use (6), summing over all Ishibashi labels to compute all annuli Ai ab .
Computationally this scales as Ni (NB )3 , where Ni is the number of primaries one needs
to consider (in our case only the massless ones) and NB the number of boundaries, or
equivalently, the number of Ishibashi labels. This sort of computation is analogous to the
computation of the fusion coefficients using the Verlinde formula, which, for N primaries,
scales as N 4 . In the latter case, such computations can be speeded up drastically if there is
a non-trivial simple current group G. This leads to relations among the matrix elements of
the matrix S of the form
SJ i,j = e2iQJ (j ) Sij .
(A.1)
Fusion computations can then be made more efficient for the following three reasons:
(1) Simple current charge conservation, i.e.,
Nabc = 0 if QJ (a) + QJ (b) + QJ (c) = 0 mod 1;
(2) Simple current relations among fusion entries, NKa,J b,Lc = Nabc if J L = K;
(3) The summation in the Verlinde formula can be restricted to orbit representatives.
In the absence of fixed points, i.e., if all orbit are equally long, this reduces the size of the
computation by a factor |G|4 to (NO )4 , where NO is the number of orbits.
In the present case the quantity of interest is Ai ab , for a subset of labels i. In order
to make use of similar results, we have to establish an action of simple currents on the
boundary labels a as well as an action on the Ishibashi labels in the internal summation.
First we define a boundary simple current. Consider the oriented annulus
Sim
Rma Rmb
,
Aia b =
S
0m
m
where for simplicity the boundary and Ishibashi labels a and m implicitly include the
degeneracy label. The labels i are primaries of the bulk CFT, and we can consider the case
that i is a simple current, J . Then one can show that for every a there is precisely one
b such that AJ a b = 1, and that for all other b this quantity vanishes. The proof goes as
follows:
SJ m

Rma Rmb
=
e2iQJ (m) Rma Rmb
.
AJ a b =
S
0m
m
m
55
For each a, the vectors Rma have norm 1. Hence this sum can at most be 1, and it is equal
to 1 if and only if
Rmb = e2iQJ (m) Rma .
(A.2)
Since the boundaries are independent, there can be at most one boundary b that satisfies
this, and since the set is complete, there must be precisely one (note that the matrix AJ has
to satisfy the completeness condition AJ AJ c = 1 and hence cannot have an entire row that
is zero). Hence it makes sense to define J a b as the action of J on a. Using (A.2) one
can easily show that
Ai,J a J b = Aia b .
This establishes the second of the three properties listed above, with K = 1. Since the label
i is restricted to massless fields a further generalization analogous to K = 1 is not really
useful, since the simple current action on i does not preserve its conformal weight.
A simple current J may yield a trivial action on all boundaries. It is easy to show that
J (K a) = (J K a) = K (J a), where J K denotes the fusion product of the CFT.
Therefore, if J and K both fix all boundaries, so does J K, and hence the set of simple
currents that fix all boundaries is a subgroup GX of G. The boundary simple current group
is defined as the discrete group G/GX . The elements of that group have distinct actions on
at least one of the boundaries.
In all cases covered by the simple current formalism of [64] there is a special boundary
originating from the CFT vacuum state, a = 0. One might conjecture a natural correspondence between the elements of the boundary simple current group and the set of boundaries
obtained by the action of G on the boundary 0. However, it turns out that this is not a one-toone correspondence: it may happen that J 0 = 0, but J a = a for some other a. Hence the
boundary simple current group may be larger than the set of simple current boundaries.
Relation (A.2) is the analog of (A.1) for the action on the boundary label. An action on
the Ishibashi label can be derived directly from (9) and (11):
R(I m,J )[a,a ] = Fm (I, J )e2iQI (a) R(m,J )[a,a ] ,
if the set of degeneracy labels J for m and Lm are the same. This is the case if and only if
L is local with respect to the simple current group H that defines the MIPF. The subgroup
of G with that property will be called the Ishibashi simple current group. The simple current twist Fm (I, J ) is a sign in all known cases, and this implies that its occurrence here
is irrelevant, since we are free to change all boundary and crosscap coefficients simultaneously by signs: Rma m Rma , Um m Um . This allows us to redefine the boundary
coefficients on each Ishibashi orbit in such a way that for the orbit representatives, m0 , we
get simply
R(I m0 ,J )[a,a ] = e2iQI (a) R(m0 ,J )[a,a ] .
Note that the crosscap coefficients Um,J vanish if J = 0, so these sign changes are irrelevant for crosscaps.
56
Now consider the first simplification listed above, charge conservation. The annulus
amplitude Ai ab can be expressed in terms of the fixed point fusion coefficients [75]
Sjm
J j
J
Jc
N ab =
Sma
Smb
S0m
m
as a linear combination of the form

KLi
(L, J, a, b) N J ab .
LH J cH
Here (L, J, a, b) are complex coefficients and K is the Klein bottle current. Using (11)
j
and the property Fa (K, N )Fa (K, M) = Fa (K, N M) it follows that (N J )ab = 0 unless
QM (a) + QM (b) = QM (j ) mod 1 for all simple currents M. For the annulus this implies
that Ai ab = 0 unless there is an L H so that QM (a) + QM (b) = QM (i) + QM (K) +
QM (L) mod 1.
Finally, consider the third simplification. The sum over m can be reduced to a sum over
Ishibashi orbit representatives provided that charge conservation is satisfied, i.e., QI (a) +
QI (b) = QI (i) + QI (K) + QI (L) mod 1, for all L H. Since Ishibashi currents I are
always local with respect to H, this condition is always satisfied.
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]
T.P.T. Dijkstra, L.R. Huiszoon, A.N. Schellekens, hep-th/0403196.

B.R. Greene, et al., Nucl. Phys. B 278 (1986) 667.
L.E. Ibanez, et al., Phys. Lett. B 191 (1987) 282.
D. Bailin, A. Love, S. Thomas, Phys. Lett. B 194 (1987) 385.
J.A. Casas, C. Munoz, Phys. Lett. B 209 (1988) 214.
J.A. Casas, C. Munoz, Phys. Lett. B 214 (1988) 63.
A. Font, et al., Nucl. Phys. B 331 (1990) 421.
D. Gepner, hep-th/9301089.
A.N. Schellekens, S. Yankielowicz, Nucl. Phys. B 330 (1990) 103.
A.N. Schellekens, unpublished, 1990.
A.E. Faraggi, D.V. Nanopoulos, K.J. Yuan, Nucl. Phys. B 335 (1990) 347.
A.E. Faraggi, Phys. Lett. B 278 (1992) 131.
A.E. Faraggi, Nucl. Phys. B 387 (1992) 239, hep-th/9208024.
S. Chaudhuri, G. Hockney, J. Lykken, Nucl. Phys. B 469 (1996) 357, hep-th/9510241.
S. Kachru, Phys. Lett. B 349 (1995) 76, hep-th/9501131.
X.G. Wen, E. Witten, Nucl. Phys. B 261 (1985) 651.
G.G. Athanasiu, et al., Phys. Lett. B 214 (1988) 55.
A.N. Schellekens, Phys. Lett. B 237 (1990) 363.
Z. Kakushadze, et al., Int. J. Mod. Phys. A 13 (1998) 2551, hep-th/9710149.
R. Donagi, et al., Adv. Theor. Math. Phys. 5 (2002) 93, hep-th/9912208.
A. Sagnotti, hep-th/0208020.
P. Horava, Nucl. Phys. B 327 (1989) 461.
J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724, hep-th/9510017.
E. Witten, Nucl. Phys. B 471 (1996) 135, hep-th/9602070.
C. Angelantonj, et al., Phys. Lett. B 385 (1996) 96, hep-th/9606169.
R. Blumenhagen, et al., JHEP 0010 (2000) 006, hep-th/0007024.
G. Aldazabal, et al., J. Math. Phys. 42 (2001) 3103, hep-th/0011073.
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
L.E. Ibanez, F. Marchesano, R. Rabadan, JHEP 0111 (2001) 002, hep-th/0105155.

R. Blumenhagen, et al., Nucl. Phys. B 616 (2001) 3, hep-th/0107138.
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 530 (2002) 202, hep-th/0108131.
R. Blumenhagen, et al., JHEP 0207 (2002) 026, hep-th/0206038.
C. Kokorelis, hep-th/0211091.
C. Kokorelis, hep-th/0309070.
J.F.G. Cascales, A.M. Uranga, JHEP 0305 (2003) 011, hep-th/0303024.
M. Cvetic, G. Shiu, A.M. Uranga, Nucl. Phys. B 615 (2001) 3, hep-th/0107166.
M. Cvetic, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801, hep-th/0107143.
M. Cvetic, I. Papadimitriou, Phys. Rev. D 67 (2003) 126006, hep-th/0303197.
R. Blumenhagen, L. Grlich, T. Ott, JHEP 0301 (2003) 021, hep-th/0211059.
G. Honecker, Nucl. Phys. B 666 (2003) 175, hep-th/0303015.
M. Cvetic, T. Li, T. Liu, hep-th/0403061.
M. Cvetic, I. Papadimitriou, G. Shiu, Nucl. Phys. B 659 (2003) 193, hep-th/0212177.
M. Bianchi, A. Sagnotti, Phys. Lett. B 247 (1990) 517.
E.G. Gimon, J. Polchinski, Phys. Rev. D 54 (1996) 1667, hep-th/9601038.
J.F.G. Cascales, et al., JHEP 0402 (2004) 031, hep-th/0312051.
C. Angelantonj, et al., Phys. Lett. B 387 (1996) 743, hep-th/9607229.
R. Blumenhagen, A. Wisskirchen, Phys. Lett. B 438 (1998) 52, hep-th/9806131.
S. Govindarajan, J. Majumder, JHEP 0402 (2004) 026, hep-th/0306257.
G. Aldazabal, et al., JHEP 0309 (2003) 067, hep-th/0307183.
R. Blumenhagen, JHEP 0311 (2003) 055, hep-th/0310244.
I. Brunner, et al., hep-th/0401137.
R. Blumenhagen, T. Weigand, hep-th/0401148.
G. Aldazabal, E.C. Andres, J.E. Juknevich, JHEP 0405 (2004) 054, hep-th/0403262.
G. Honecker, T. Ott, hep-th/0404055.
F. Marchesano, G. Shiu, hep-th/0409132.
F. Marchesano, G. Shiu, hep-th/0408059.
A.M. Uranga, Class. Quantum Grav. 20 (2003) S373, hep-th/0301032.
T. Ott, Fortschr. Phys. 52 (2004) 28, hep-th/0309107.
E. Kiritsis, Fortschr. Phys. 52 (2004) 200, hep-th/0310001.
D. Lst, hep-th/0401156.
M.R. Douglas, JHEP 0305 (2003) 046, hep-th/0303194.
B. Gato-Rivera, A.N. Schellekens, Commun. Math. Phys. 145 (1992) 85.
M. Kreuzer, A.N. Schellekens, Nucl. Phys. B 411 (1994) 97, hep-th/9306145.
J. Fuchs, et al., Ann. Phys. 204 (1990) 1.
J. Fuchs, et al., Phys. Lett. B 495 (2000) 427, hep-th/0007174.
W. Lerche, A.N. Schellekens, N.P. Warner, Phys. Rep. 177 (1989) 1.
J. Fuchs, C. Schweigert, J. Walcher, Nucl. Phys. B 588 (2000) 110, hep-th/0003298.
C.A. Lutken, G.G. Ross, Phys. Lett. B 214 (1988) 357.
A.N. Schellekens, S. Yankielowicz, Phys. Lett. B 242 (1990) 45.
T. Gannon, J. Math. Phys. 36 (1995) 675, hep-th/9402074.
R.E. Behrend, et al., Nucl. Phys. B 570 (2000) 525, hep-th/9908036.
C. Angelantonj, A. Sagnotti, Phys. Rep. 371 (2002) 1, hep-th/0204089.
G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 354 (1995) 279, hep-th/9503207.
A. Recknagel, V. Schomerus, Nucl. Phys. B 531 (1998) 185, hep-th/9712186.
J. Fuchs, A.N. Schellekens, C. Schweigert, Nucl. Phys. B 473 (1996) 323, hep-th/9601078.
L.R. Huiszoon, PhD thesis, 2002, available on request.
J. Fuchs, I. Runkel, C. Schweigert, hep-th/0403157.
M. Bianchi, J.F. Morales, JHEP 0003 (2000) 030, hep-th/0002149.
R. Blumenhagen, D. Lst, S. Stieberger, JHEP 0307 (2003) 036, hep-th/0305146.
L. Birke, J. Fuchs, C. Schweigert, Adv. Theor. Math. Phys. 3 (1999) 671, hep-th/9905038.
Y. Kazama, H. Suzuki, Nucl. Phys. B 321 (1989) 232.
L. Borisov, M.B. Halpern, C. Schweigert, Int. J. Mod. Phys. A 13 (1998) 125, hep-th/9701061.
P. Bantay, Nucl. Phys. B 633 (2002) 365, hep-th/9910079.
M. Kreuzer, H. Skarke, Adv. Theor. Math. Phys. 4 (2002) 1209, hep-th/0002240.
57
Asyzygies, modular forms,

and the superstring measure I
Eric DHoker a , D.H. Phong b
a Department of Physics and Astronomy, University of California, Los Angeles, CA 90095, USA
b Department of Mathematics, Columbia University, New York, NY 10027, USA

Abstract
The goal of this paper and of a subsequent continuation is to find some viable ansatze for the
three-loop superstring chiral measure. For this, two alternative formulas are derived for the two-loop
superstring chiral measure. Unlike the original formula, both alternates admit modular covariant
generalizations to higher genus. One of these two generalizations is analyzed in detail in the present
paper, with the analysis of the other left to the next paper of the series.
PACS: 11.25.-w; 11.25.Db; 11.25.Hf
1. Introduction
String perturbation theory is closely related to the mathematical theories of Riemann
surfaces and modular forms. In string theory, scattering amplitudes are given perturbatively
at loop order h by the moments of a measure on the moduli space of Riemann surfaces
of genus h. For example, the vacuum-to-vacuum amplitude, or cosmological constant, is
given by an expression of the form

A=
(1.1)
(det Im )5
c, d []() d [](),
Mh
Research supported in part by National Science Foundation grants PHY-01-40151 and DMS-02-45371.
E-mail address: phong@math.columbia.edu (D.H. Phong).
doi:10.1016/j.nuclphysb.2004.12.021
E. DHoker, D.H. Phong / Nuclear Physics B 710 (2005) 5882
59
where , denote independent spin structures, c, are suitable constant phases, and
d []() is a holomorphic form of maximal rank (3h 3, 0) on the moduli space of
Riemann surfaces of genus h. Here we have represented the Riemann surface by its period
matrix , after a choice of canonical homology basis. Since the integrand in the above
formula should be independent of the homology basis, it follows that d []() must
transform covariantly under Sp(2h, Z) modular transformations.
This correspondence between the superstring measure and modular forms is particularly
simple at one-loop, where d [] [](0, )4 /()12 , so that d []() is essentially
a -constant. No such simple relation is known in general. It has long been hoped that the
extensive mathematical literature on modular forms, combined with physical constraints
on the superstring measure such as factorization, can help pinpoint the measure. But this
hope has remained unfulfilled, perhaps because of the lack of a suitable ansatz.
In [14] (see also the review [5], and earlier work [6]), an explicit formula for the twoloop superstring measure was derived from first principles. It is given by
[]4 (0, ) 6 []()
d [] =
(1.2)
dI J .
16 6 10 ()
1I J 2

Here 10 () = even [](0, )2 is the well-known modular form introduced by Igusa
[7,8], and the key expression 6 []() is defined by

i |j
[i + j + k ]4 (0, ),
6 []()
(1.3)
1i<j 3
k=4,5,6
where we have expressed the even spin structure as the sum of three odd spin structures,
say = 1 + 2 + 3 , and 4 , 5 , 6 are the remaining odd spin structures. This prescription
clearly depends heavily on the fact that the surface has genus 2.
The main goal of the present paper is to provide two alternative formulas of 6 []()
which are no longer tied specifically to genus 2. One formula still retains the form (1.3) (so
that 6 [] is still expressed as a linear combination of the 4th powers of three -constants),
but the contributing spin structures i + j + 4 , i + j + 5 , i + j + 6 are now characterized as triples {m , n , p } which together with form a totally asyzygous quartet
(Theorem 1). The other formula is a more radical departure from (1.3), since 6 [] is now
expressed as a linear combination of the squares of six -constants. The corresponding
sextets of spin structures are characterized by the requirements that they do not contain ,
and can be divided into three pairs of spin structures, the union of any two is a totally
asyzygous quartet (Theorem 2).
The new formula in terms of 4th powers of -constants admits immediately a generalization to all genera h 2. The generalization to genus 3 of the formula in terms of squares
of -constants is already more delicate. We shall describe it only in the next paper of this
series, and devote the rest of this paper to the study of the 4th powers formula. We show
that the generalization of the 4th powers formula satisfies the same modular transformation laws as in genus 2 (Theorem 3) and the same relation with the basic modular forms
4 ()2 2h 8 (), upon summation over spin structures (Theorem 4). The expression
4 ()2 2h 8 () is well known to vanish in genus 1 by the Jacobi identity, and in genus
2 as a consequence of Igusas theorem on the ring of modular forms. Our numerical simulations indicate very strongly that it also vanishes in genus 3, and a natural conjecture is
60
that it actually vanishes for all genera. We also determine explicitly the degeneration limits
of the 4th powers formula in genus 3.
As we had stressed earlier, our main motivation for this study is to develop a suitable
ansatz for the superstring measure in higher genus. In the next paper of this series, we
shall show, based on the degeneration properties of the two formulas, that it is the formula in terms of squares of -constants of sextets of spin structures which leads to viable
candidates for the three-loop superstring measure. Nevertheless, both formulas exhibit remarkable properties, and it can be hoped that they will both be of interest in the theories of
Riemann surfaces and modular forms.
2. Alternative formulas for 6 []()

On a Riemann surface of genus h, there are 22h spin structures, i.e., square roots
of the canonical bundle of . Fixing a canonical homology basis, a spin structure can
be identified with a half-characteristic = ( | ), , {0, 12 }h . The 22h spin structures
can be divided into 2h1 (2h + 1) even and 2h1 (2h 1) odd ones, depending on the parity
p() = exp(4i ). The relative signature 1 |2 of a pair of spin structures 1 , 2 is
defined by

1 |2 = exp 4i(1 2 1 2 ) .
(2.1)
The relative signature of two spin structures 1 , 2 is not invariant under modular transformations (for their precise transformation laws, see Appendix A). However, for a triplet ,
, of spin structures, we can define the sign e(,

, ) by
e(,
, ) |

| |,
(2.2)
which does not depend on the ordering of the triplet, and is invariant under modular transformations. This can be seen from the relation between the parities of a triplet of spin
structures and their relative signature, which is a direct consequence of their definitions
p( +
+ ) = p()p(
)p() e(,
, ).
(2.3)
In genus h = 2, there are 10 even spin structures and 6 odd ones. It is convenient to
reserve the notation for even spin structures, and denote odd spin structures by . Each
even spin structure can be written as = 1 + 2 + 3 , where the i s are pairwise distinct
odd spin structures. If 4 , 5 , 6 is the complementary set of odd spin structures, then we
also have = 4 + 5 + 6 , and the mapping {1 , 2 , 3 } is exactly 2 to 1. For each
even spin structure in genus 2, say = 1 + 2 + 3 , the form 6 [] was defined in [1]
by the formula (1.3). In this section, we provide two more formulas for 6 [](). The
first one gives a different description of the even spin structures whose -constants enter
the right-hand side of (1.3). The second one replaces the product of 3 fourth powers of
-constants by the product of 6 second powers of different -constants. Both formulas do
not depend on the decomposition of even spin structures into odd spin structures in genus 2
and hence allow extensions to arbitrary genus.
61
Let (1 , . . . , N ) be an unordered N -tuplet of N distinct spin structures. We say that

the N -tuplet is a totally asyzygous N -tuplet if e(i1 , i2 , i3 ) = 1 for any subset of three
distinct spin structures i1 , i2 , i3 in the N -tuplet.
Theorem 1. The expression 6 []() in (1.3) can be rewritten as

1
6 []() =
|
|| [
]4 (0, )[]4 (0, )[]4 (0, ).
2
,
,, tot.asyz.
(2.4)
Here the summation is over all triples
, , which together with form a totally asyzygous
quartet. Each triplet (
, , ) is counted only once, irrespective of its ordering.
Proof. We have to show that, given an even spin structure = 1 + 2 + 3 , the quartets
{i + j + k }{k =i,j } ,
(2.5)
are asyzygous for any choice of ij among 12, 13, 23, and that conversely, any totally
asyzygous quartet {,
, , } containing must be of this form (or of the mirror form
{i + j + k }{k =i,j } if is represented as = 4 + 5 + 6 ).
We begin by showing that (2.5) is asyzygous. Recall that in genus 2, it follows immediately from the description of even and odd spin structures that i + j + k is even if
and only if the odd spin structures i , j , k are pairwise distinct. Thus the spin structures
in the quartet in (2.5) are all even. Furthermore, the sum of any distinct three of them is
3(i + j ) + k1 + k2 + k3 = i + j + k1 + k2 + k3 , and hence equal to the odd spin
structure k4 , where k4 is the unique index defferent from i, j , k1 , k2 , k3 . It follows now
from the relation (2.3) that the triplet {i + j + k }{k=k1 ,k2 ,k3 } is asyzygous, and hence the
quartets (2.5) are totally asyzygous.
We show now the converse. Since all the spin structures in {,
, , } are assumed to
be even, the totally asyzygous condition is equivalent, in view of the relation (2.3) to the
condition that the sum of any three distinct spin structures in the quartet be odd. We can
easily work this out for = 1 + 2 + 3 . Recall that each even spin structure admits two
representations, as the sum of a set of three distinct odd spin structures + + , or
as the complementary sum. Since the spin structures
, , are different from , we may
assume that, for each of them, at least one of the three spin structures , , is among
1 , 2 , 3 . By taking the complementary representation if necessary, we may assume that
at least two of them are in this set, and in fact exactly two, since they are distinct from .
Thus, after some suitable renaming, the spin structures
, , must fall into one of the
following three possibilities

= + + ,

= + + ,

= + + ,
1
(1)
= 1 + 3 + k ,
= 2 + 3 + l ,
(2)
= 1 + 2 + k ,
= 1 + 3 + l ,
(3)
= 1 + 2 + k ,
= 1 + 2 + l ,
(2.6)
with j , k , l {4, 5, 6}.

The case (3) gives a quartet of type (2.5), as claimed. To deal with the cases (1) and (2),
we again exploit the fact that the sum of the spin structures in an asyzygous triplet of
62
even spin structures must be odd. Applied to a triplet of the form { = 1 + 2 + 3 , 1 +

2 + j , 1 + 3 + k }, it implies that 3 + j + k must be odd. But this implies that
j = k (since they are both different from 3 ). Turning more specifically to the case (1),
we deduce that the spin structures j , k , l must all be the same in this case. Going to the
complementary representation for {,
, , }, we obtain a quartet of the form (2.5), with
ij given by one of the pairs 45, 46, or 56. Finally, for case (2), we note that the previous
observation implies that k = l . But then we have
+ + = 1 + 3 + j , which is even,
contradicting the fact that the triplet
, , is asyzygous.
It remains to re-express the sign factor i , j in (1.3) in the form of indicated in (2.4).
For this, consider an asyzygous quartet {,
, , } written under the form corresponding
to case (3) above. Then |
|| = |
+ + = |3 = 1 |3 2 |3 . Since
the triplet {1 , 2 , 3 } is asyzygous, this is equal to 1 |2 . 2
Theorem 2. The form 6 [](0, ) can also be expressed as

1
; {i }
[ij ](0, )2 ,
2
6
6 []() =
{i }
(2.7)
j =1
where the sum runs over all sextets {i1 , . . . , i6 } of even spin structures which are
-admissible in the following sense: (1)
/ {ij }; and (2) the sextet can be partitioned
into 3 disjoint pairs {i1 , i2 }, {i3 , i4 }, {i5 , i6 }, the union of any two pairs forming a
totally asyzygous quartet.
The modular group Sp(4, Z2 ) acts transitively on the space of 7-tuplets {; i1 , . . . , i6 }
satisfying the condition that {i1 , . . . , i6 } is -admissible. The expressions
(; i1 , . . . ,
i6 ) {1} are phases satisfying the following modular transformation law
6

M; {Mij }
(, M)4 =
; {ij }
(ik , M)2 ,
(2.8)
k=1
where
(, M) is the 8th root of unity entering transformation laws for -constants. The
phases
(; i1 , . . . , i6 ) are completely determined by one another by this transformation
law. The global relative sign in (2.7) is determined by the choice
(7 ; {3 , 4 , 5 , 6 , 9 ,
0 }) = 1, in the indexing of even spin structures defined in Appendix B. Their explicit
form will be given below, in (2.22) and (2.23).
Proof. The main tool in the proof is the Riemann relations for -functions. Riemann relations are usually applied to 4th powers of -constants, but they apply equally well to 2nd
powers, as we show now. Recall that the Riemann relations are given by [4, Section 2.4]
4

|[](1 )[](2 )[](3 )[](4 ) = 4
[](1 ),
(2.9)
i=1
where is expressed as linear combinations of the . The case of interest here is for
1 = 2 = and 3 = 4 = 0, where = 0 is a twist half-characteristic. Thus, the relations
63
reduce to 1 = 2 = and 3 = 4 = 0. Restricting attention to = odd, we are left with

0=
(2.10)
|[]()2 [](0)2 =
e4i |[ + ](0)2 [](0)2 .
This sum receives non-trivial contributions only from terms with both and + even.
Any of the 15 twists may be expressed (see, for example, [9]) as the sum of two odd spin
structures, = a + b for a = b, all such spin structures may be parametrized by
= a + j + k ,
+ = b + j + k ,
j, k
/ {a, b}.
(2.11)
Of the 6 possible odd spin structures, 4 yield trivially 0. The two (equivalent) non-trivial
cases are for = a and = b . Choosing the first, the Riemann identity may now be
recast in the following form,

(ab|j k)[a + j + k ](0)2 [b + j + k ](0)2 ,
0=
(2.12)
j =k
j,k =a,b
where

(ab|j k) = (ba|j k) = exp 4i a (j + k ) + b (j + k ) .
(2.13)
Here, the symmetry in a b is readily shown using the fact that j = k and that a, b
/
{j, k}. This form of Riemann identities is closely related to Riemann identities on Prym
varieties, which appeared in the study of the cosmological constant in certain orbifold
models with broken supersymmetry [9,11].
We can return now to the proof of Theorem 2 proper. The property (1) of -admissible
sextets described in Theorem 2 is clearly invariant under modular transformations. Both
properties (1) and (2) are invariant under the subgroup of modular transformations which
leave the spin structure unchanged. The transitivity of this subgroup can be read off from
the tables of sextets and modular transformations in Appendix B. The transformation law
for the phases
(; {ij }) implies that both sides of Eq. (2.7) transform in the same way
under modular transformations. Thus it suffices to prove the theorem for some fixed even
spin structure, say = 1 + 2 + 3 . This is 7 in the notation of Appendix B. In this case,
the sextets which satisfy the properties (1) and (2) are given by (see table in Appendix C)
{345960},
{192638},
{152048},
{125690},
{134589},
{234680},
(2.14)
where we have abbreviated the sextet {i1 , . . . , i6 } by {i1 . . . i6 }.

We rewrite now 6 [1 + 2 + 3 ] using the preceding version of Riemann identities
with squares of -constants. Using the shorthand (ij k) = [i + j + k ](0)2 , we have the
following Riemann relations needed to recast 6 [1 + 2 + 3 ],
(45|12)(124)(125) = (45|13)(134)(135) (45|23)(234)(235),
(56|12)(125)(126) = (56|13)(135)(136) (56|23)(235)(236),
(64|12)(124)(126) = (64|13)(134)(136) (64|23)(234)(236).
(2.15)
64
Multiplying the three left-hand sides together and the three right-hand sides together, and
using the identity
(45|ij )(56|ij )(64|ij ) = i |j ,
(2.16)
we readily see that

6 [1 + 2 + 3 ]
= (45|13)(56|13)(64|23)(134)(135)2 (136)(234)(236)
(45|13)(56|23)(64|13)(134)2 (135)(136)(235)(236)
(45|23)(56|13)(64|13)(134)(135)(136)2 (234)(235)
(45|23)(56|13)(64|23)(135)(136)(234)2 (235)(236)
(45|23)(56|23)(64|13)(134)(136)(234)(235)2 (236)
(45|13)(56|23)(64|23)(134)(135)(234)(235)(236)2 .
(2.17)
Combining lines 1 and 4, lines 2 and 6, and lines 3 and 5 using the same Riemann identities
(2.15), we get
6 [1 + 2 + 3 ] = cA (124)(125)(135)(136)(234)(236)
cB (125)(126)(134)(136)(234)(235)
cC (124)(126)(134)(135)(235)(236).
(2.18)
The signs are easily simplified, and we get

cA = (45|12)(56|13)(64|23)
= exp 4i{1 6 + 2 5 + 3 4 + 4 1 + 5 3 + 6 2 },
cB = (45|23)(56|12)(64|13)
= exp 4i{2 6 + 3 5 + 1 4 + 4 2 + 5 1 + 6 3 },
cC = (45|12)(56|13)(64|23)
= exp 4i{3 6 + 1 5 + 2 4 + 4 3 + 5 2 + 6 1 }.
(2.19)
To re-express the above result in terms of even spin structures, we need to choose a basis
for the even spin structures in terms of the odd ones. We choose the same basis as was
used in [4], but the actual values of the odd spin structures a are left arbitrary. Modular
transformations simply permute the odd spin structures. The basis may be expressed as
follows (up to complete periods),
1 = 1 + 4 + 6 = 2 + 3 + 5 ,
6 = 1 + 5 + 6 = 2 + 3 + 4 ,
2 = 1 + 2 + 6 = 3 + 4 + 5 ,
7 = 1 + 2 + 3 = 4 + 5 + 6 ,
3 = 1 + 2 + 5 = 3 + 4 + 6 ,
8 = 1 + 3 + 4 = 2 + 5 + 6 ,
4 = 1 + 4 + 5 = 2 + 3 + 6 ,
9 = 1 + 3 + 6 = 2 + 4 + 5 ,
5 = 1 + 2 + 4 = 3 + 5 + 6 ,
0 = 1 + 3 + 5 = 2 + 4 + 6 .
(2.20)
65
Table 1
Decomposition of odd into even within a 7 -admissible sextet
A
B
C
1
2
3
4
5
6
3 + 4 + 5
3 + 5 + 9
3 + 4 + 6
3 + 6 + 0
3 + 4 + 9
3 + 4 + 0
1 + 8 + 9
1 + 6 + 9
1 + 2 + 6
1 + 3 + 8
1 + 2 + 9
1 + 3 + 9
4 + 8 + 0
1 + 4 + 8
1 + 4 + 5
2 + 4 + 8
2 + 4 + 0
4 + 5 + 8
1 + 8 + 9
1 + 4 + 8
1 + 4 + 5
1 + 3 + 8
1 + 5 + 8
1 + 3 + 9
2 + 3 + 6
2 + 3 + 8
2 + 8 + 0
2 + 4 + 8
2 + 4 + 0
2 + 6 + 8
1 + 2 + 5
1 + 6 + 9
1 + 2 + 6
1 + 5 + 0
1 + 2 + 9
1 + 2 + 0
In total, there are 15 totally asyzygous quartets, 9 of which do not contain a given spin
structure. The 9 asyzygous quartets which do not contain, say 7 , may be grouped in terms
of two groups of three 7 -admissible sextets, given as follows,
A = [345960],
A = [183945],
B = [192638],
B = [283640],
C = [152048],
C = [125069].
(2.21)
The structure of the asyzygous quartets allows us to recast odd spin structures in terms of
asyzygous triplets of even ones, for each group, as follows in Table 1.
Re-expressing the signs cA , cB , cC , in terms of even spin structures, we obtain,
6 [7 ]
= (4 , 5 , 0 )(3 , 9 , 6 )[3 ]2 [4 ]2 [5 ]2 [6 ]2 [9 ]2 [0 ]2
(3 , 9 , 6 )(1 , 2 , 8 )[1 ]2 [2 ]2 [3 ]2 [6 ]2 [8 ]2 [9 ]2
(1 , 2 , 8 )(4 , 5 , 0 )[1 ]2 [2 ]2 [4 ]2 [5 ]2 [8 ]2 [0 ]2 ,
(2.22)
where the following combination enters,

(i , j , k ) exp 4i{i j + j k + k i }
(2.23)
and our final expression for 6 [7 ]. An alternative formula is obtained by replacing the
partitioning A, B, C by A , B , C . Averaging over the two formulas gives the desired
formula (2.7). Note that it is the full set of sextets in (2.14) which is an orbit under the
group of modular transformations fixing . Note also that the above expression in terms of
even spin structures is uniquely defined by the partitioning. This may be established using
the symmetries of the function . Computing the effect of a permutation, we find
(i , j , k )(j , i , k ) = i |j j |k k |i = e(i , j , k ).
(2.24)
Next, by inspection, we see that the triplets (i , j , k ) on which is evaluated in

each partition, namely (4 , 5 , 0 ), (3 , 9 , 6 ), and (1 , 2 , 8 ), are all syzygous, so that
(i , j , k ) is totally symmetric in its three arguments. 2
Remark 1. We note that, unlike in the expression for 6 [] in terms of 4th powers of
-constants, the phases
(; {i }) in the expression (2.7) cannot be expressed in terms of
simple modular invariant quantities such as the relative signature of two spin structures.
66
This will account for some important subtleties when we discuss the generalization of
expressions such as (2.7) to higher genus in [10].
Remark 2. Another
proof of Theorem 2 can be obtained, using the hyperelliptic representation s 2 = 6i=1 (x pi ) of genus 2 Riemann surfaces. If we view the odd spin
structures i as corresponding to the branch points pi , then the spin structure = 7
corresponds to the partition of the 6 branch points into two sets A = {p1 , p2 , p3 } and
B = {p4 , p5 , p6 }. We can then establish the following formula
6 []
(p1 p (4) )(p2 p (5) )(p3 p (6) )
(2.25)
=
,

1/2
2
[] 10
i =j A (pi pj ) i =j B (pi pj )
where runs over the 3 cyclic permutations of {4, 5, 6}. Expressing the right-hand side in
terms of -constants gives again (2.18), up to an overall sign, which cannot be determined
from the hyperelliptic representation.
Remark 3. We stress that the full orbit under the subgroup of modular transformations
leaving 7 unchanged requires all 6 sextets listed in (2.14). However, the fact that this orbit
can be broken into two groups, each already giving essentially 6 [], plays an important
role in the construction in [10] of modular forms using pairs of sextets.
3. A modular covariant form 6# []() in any genus
In the previous section, we have seen that the modular covariant form 6 []() admits
an expression in terms of totally asyzygous quartets of even spin structures. This notion
makes sense for all genera, and thus we can define an extension to all genera of 6 [](),
denoted by 6# [](), by the right-hand side of (2.4). The main goal of the present section
is to show that 6# []() satisfies modular transformation laws and identities similar to
those of 6 []() in genus 2. But before we do so, we wish to point out that the notion of
totally asyzygous odd spin structures could have been used as well.
Let {1 , 2 , 3 , 4 } be any quartet of spin structures, and define another quartet of spin
structures {1 , 2 , 3 , 4 } by

i =
(3.1)
j , 1 i 4.
j =i
The two
quartets determine each other uniquely, since the above equations imply that
j = i =j i . We claim now that {1 , 2 , 3 , 4 } is an asyzygous quartet of even spin
structures if and only if {1 , 2 , 3 , 4 } is an asyzygous quartet of odd spin structures.
Indeed, assume that {1 , 2 , 3 , 4 } is an asyzygous quartet of even spin structures, and
consider say p(1 ). Then p(1 ) = p(2 )p(3 )p(4 )e(2 , 3 , 4 ) = 1 which shows that
1 , and hence any i , is odd. Reversing the roles of i and i in the same formula shows
that p(1 ) = p(2 )p(3 )p(4 )e(2 , 3 , 4 ), which implies that e(2 , 3 , 4 ) = 1, since
the i s are even, and the i s are now known to be odd. Thus the quartet {1 , 2 , 3 , 4 }
is asyzygous. The same argument can clearly be reversed to show that {1 , 2 , 3 , 4 } is an
67
asyzygous quartet of even spin structures if {1 , 2 , 3 , 4 } is an asyzygous quartet of odd

spin structures, completing the proof of our claim.
In view of the above exact correspondence between asyzygous quartets of even and
odd spin structures, we obtain immediately the following equivalent characterization of
6# []() valid in any genus
6# [] =
1
6
[1 ,2 ,3 ,4 ] []| (1) | (2)
[1 ,2 ,3 ,4 ]
| (3) [ (1) ]4 [ (2) ]4 [ (3) ]4 .
(3.2)
Here the summation is over all totally asyzygous quartets [1 , 2 , 3 , 4 ] of odd spin structures. The indicator function is defined by

1, iff = (1) + (2) + (3) ,
[1 ,2 ,3 ,4 ] []
(3.3)
0, iff = (1) + (2) + (3) ,
where is any permutation of 1, 2, 3, 4.
We turn now to the modular properties of 6# []():
Theorem 3. Under modular transformations M, the expression 6# []() transforms as
)
=
(, M)4 det(C + D)6 6# [](),
6# [](
(3.4)
where
= M
and
(, M) is precisely the same 8th root of unity occurring in the modular
transformation for [](0, ).
Proof. We establish (3.4) for general modular transformations by establishing it for the
generators of Sp(2h, Z). A set of generators is given by the matrices MA , S, and MB
listed in Appendix A. Modular transformations map totally asyzygous quartets into totally
asyzygous quartets. Now the signature on pairs is invariant under MA and S, and the factors
(, MA ) and
(, S) are essentially trivial in this case, so we need only consider the case
of MB generators. We proceed as follows,
1
)
=

|

6# [](
|
|
[
]4 (0, )[
]
4 (0, )[
]
4 (0, ).
2

,,
[,
]
(3.5)
Next, we cast each spin structure as the transformation of its preimage, and use the constant transformation laws,
1
)
=

|

6# [](
|
|

4 (
, MB )
4 (, MB )
4 (, MB )
2
[,
,,]
[
]4 (0, )[]4 (0, )[]4 (0, ).
The signature factors worked out to be

h

|

(3i
i i i )Bi ,
|
|
= |
|| exp 2i
i=1
(3.6)
68

h

2

(
i ) + (i )2 + (i )2 Bi .
4 (
, MB )
4 (, MB )
4 (, MB ) = exp 4i
(3.7)
i=1
Assembling both factors, and using the fact that Bi are integers, we obtain

|

|
|
4 (
, MB )
4 (, MB )
4 (, MB )
=
4 (, MB )|
||

h

exp 2i
i (2i + 1) +
i (2
i + 1) + i (2i + 1)
i=1
+ i (2i

+ 1) Bi .
(3.8)
Now, in the last exponential factor, the combinations i (2i + 1) are integers whenever i
are half-integers, and thus the exponential factor is 1. 2
It follows from Theorem 3 that
6# [](
)
= det(C + D)8 []4 (0, )6# []()
4 (0, )
[]
(3.9)
#
4
and hence the expression 6 [][] is a modular form of weight 8. More precisely,
define the modular form 4k () by

4k () =
(3.10)
[](0, )8k .
For genus 2, where a complete classification of the polynomial ring of modular forms
exists [7], these modular forms are subject to relations such as 48 = 42 . In higher genus,
we have not found any analogous discussion in the literature. Remarkably, we have the
following.
Theorem 4. The form 6# []() satisfies the following identity in any genus h 2

6# [][]4 = 22h5 2h 8 42 .
(3.11)
Proof. In fact, we shall prove a more general result for the following more general quantity,

1
6A [] 5
|1 |2 |3 [1 ]4 [2 ]4 [3 ]4
2 3!
1 ,2 ,3

1 + a0 e(1 , 2 , 3 ) 1 + a1 e(, 2 , 3 )

1 + a2 e(, 1 , 3 ) 1 + a3 e(, 1 , 2 ) .
(3.12)
Here, A is an array of sign assignments A = [a0 ; a1 , a2 , a3 ], with a0 , a1 , a2 , a3 = 1. Since
e(1 , 2 , 3 )e(, 1 , 2 )e(, 1 , 3 )e(, 2 , 3 ) = +1, it follows that 6A is non-vanishing
only if a0 a1 a2 a3 = +1, which we henceforth assume. Two As are equivalent if they differ
69
by a permutation of their last 3 entries. The sign factor |1 |2 |3 would appear
to involve a specific choice. Actually, all other choices are equivalent to it. For example,
another modular covariant choice would be,
|1 2 |3 = |1 |2 |3 e(, 2 , 3 ) = a1 |1 |2 |3 .
(3.13)
But, since a1 is independent of i , this choice is equivalent to the one in (3.12). As a special
A
case, we have 6# [] = 6 [] for A = [; , , ].
Using the relation a0 a1 a2 a3 = +1, the form of 6A [] may be simplified as follows,
6A [] =

1
|1 |2 |3 [1 ]4 [2 ]4 [3 ]4 1 + a0 e(1 , 2 , 3 )
96
1 ,2 ,3

1 + a1 e(, 2 , 3 ) + a2 e(, 1 , 3 ) + a3 e(, 1 , 2 ) . (3.14)
We analyze this sum according to its ai -dependence. The term independent of ai is proportional to []12 by the Riemann identities. In the term proportional to a1 , we use
|1 |2 |3 e(, 2 , 3 ) = |1 2 |3 . The sums over 2 , 3 yield 4 by the Riemann identities, while the sum over 1 yields []4 . In the term proportional to a0 a1 , we
use
|1 |2 |3 e(1 , 2 , 3 )e(, 2 , 3 ) = 1 |1 |2 1 |3 .
The sums over 2 and 3 yield [1 ]8 and the remaining sum over 1 is a sum over [1 ]12 .
In summary, we have the following simplified form, valid for general genus h,

1
1
6A [] = 23h5 (1 + a0 )[]12 22h5 a []4 4 + a0
|1 [1 ]12
3
3
1

a0
1 e(1 , 2 , 3 ) |1 |2 |3 [1 ]4 [2 ]4 [3 ]4 ,

48
1 ,2 ,3
(3.15)
where we use
the notation a = a1 + a2 + a3 .
The sum 6A [][]4 may be calculated in terms of 4 and 8 using the Riemann
relations, in addition to one further result, which involves the last term in (3.15). Concentrating on the summation over , we have to carry out the following sum,

(3.16)
|1 |2 |3 [](0)4 =
|1 + 2 + 3 [](0)4 .
Using the fact that the triplet (1 , 2 , 3 ) is restricted to be asyzygous by the projection
prefactor in the last term of (3.15), the spin structure 1 +2 +3 must be odd, and therefore,
using the Riemann identities on the -sum, (3.16) cancels for all h 2. The final expression
is then

1
1
6A [][]4 = 23h5 (1 + a0 )(1 + a)8 22h5 a 42 2h 8 .
(3.17)
3
3
In the special case of A = A , we have a0 = 1 and a = 3, and we readily recover the

statement of Theorem 3. 2
70
Note that with the help of 6# [], the last sum in (3.15) may be eliminated, so that

1
1
6A [] = a0 6# [] 23h5 (1 + a0 )[]12 + 22h5 a0 a []4 4
3
3

1
22h5 a0 (a + 3)
|1 [1 ]12 .
(3.18)
3
1
4. A conjectured identity between 8 () and 4 ()

The identity in Theorem 3 suggests that the following relation between 8 () and
4 () should hold in any genus
2h 8 () 42 () = 0.
(4.1)
() 2h
for all k as Im(II ) +i. This number simply corresponds

We note that 4k
to the 2h spin structures for which I = 0. This explains why the factor of 2h in the relation
between 4 and 8 is needed. In genus 1, the relation (4.1) corresponds to the well-known
isospectrality of the root lattices of E8 E8 and SO(32), and follows immediately from the
Jacobi identity 34 = 24 + 44 . In genus 2, it is a consequence of Igusas classification of
genus 2 modular forms. To the best of our knowledge, the relation has not been established
for genus h 3, although there is an extensive literature on related topics (see, e.g., [12],
and references therein). We present here some strong evidence for it in genus 3. For this, it
is useful to define the following parameters for the period matrix
1
1

1
2 3
2 2
x = ei1 ,
u = ei1 ,
1
1
i
2
= 2 3 2
(4.2)
,
y=e
v = ei2 ,
2 1 ,
i
i3 .
3
1
1
,
z
=
e
w
=
e
3
2 2
2 1
4.1. Asymptotic evidence
The first type of evidence for the relation (4.1) is its vanishing asymptotics under completely separating degenerations. These correspond to i 0. More precisely, we shall
show that (4.1) holds for arbitrary x, y, z up to and including the orders 1 2 3 and 12 22 ,
22 32 , 12 32 .
Now the -function for the genus 3 period matrix can be written as

1
[](z, ) = exp
(4.3)
[i ](zi , i ),
(3 1 2 + 3 1 2 + 3 1 2 )
4i
i=1,2,3
where = (1 |2 |3 z = (z1 , z2 , z3 ) and i = /zi . Expanding the exponential into a

series, we see that even and odd i behave differently. We distinguish 2 types of even spin
structures : those for which all genus 1 components carry even spin structures i , and
those for which one component carries an even genus 1 spin structure but two components
carry the odd spin structure. For the latter, [] vanishes to order i j , which implies
that the lowest order contribution to 4 and 8 that is generated by these terms is i8 j8 .
As long as we work to orders less than 8, it will thus be necessary to consider only spin
structures for which all genus 1 components i are even.
t
)t ,
71
Taking then all i to be even and zi = 0, and using the heat equation, i2 [i ](zi , i ) =
4ii [i ](zi , i ), we deduce the following relations,
[](0, ) =

ki =0
2k
12k1 22k2 3 3
k2 +k3 [1 ]k21 +k3 [2 ]k31 +k2 [3 ]
(2k1 )!(2k2 )!(2k3 )! 1

ki =0
2k +1
12k1 +1 22k2 +1 3 3
(2k1 + 1)!(2k2 + 1)!(2k3 + 1)!
k12 +k3 +1 [1 ]k21 +k3 +1 [2 ]k31 +k2 +1 [3 ],
(4.4)
where we have set ki [i ] = ki [i ](0, i ). To linear order in 1 , the first term occurring
is obtained from the second term in the sum with k1 = k2 = k3 = 0. Its contribution to 4
and 8 is, respectively,

4 =
i8 (1 )j8 (2 )k8 (3 ){1 + 81 2 3 1 ln i 2 ln j 3 ln k },
i,j,k
8 =
i16 (1 )j16 (2 )k16 (3 ){1 + 161 2 3 1 ln i 2 ln j 3 ln k }.
(4.5)
i,j,k
Using now the genus 1 relation ( i i8 )2 = 2 i i16 , we readily see that the equation
42 = 88 is obeyed at genus 3 to this order.
The next order is of order 12 22 (and by symmetry, the terms of order 12 32 and 22 32
are dealt with in the same way). The contributions to both modular forms are as follows,

4 =
i8 (1 )j8 (2 )k8 (3 )
i,j,k
8 =

i,j,k

1 + 412 bj ck + 422 ai ck + 12 22 ai bj (2ck + 14(ck )2 ,
i16 (1 )j16 (2 )k16 (3 )

1 + 1 + 812 bj ck + 822 ai ck + 12 22 ai bj (4ck + 60(ck )2 ,
(4.6)
with the abbreviations, ai = 1 i (1 )/i (1 ), bi = 2 i (2 )/i (2 ), ci = 3 i (3 )/

i (3 ), ci = 23 i (3 )/i (3 ). Next, we use the following formulas valid in genus 1,

1
1
i (1 )8 ai = 1 4 (1 ),
i (1 )16 ai = 1 8 (1 ),
(4.7)
8
16
i
and similar formulas for bi and ci . Putting all together, we have

88 42 =
12 22
4 (1 )2 4 (2 )2 1 4 (1 )2 4 (2 )
256

242 (3 4 )2 +
k (3 )16 32ck + 480(ck )2
k

k (3 )4 4 (3 ) 16ck + 112(ck )2 .
(4.8)
72
Finally, using the relation

23 28 (3 ) 4 (3 )2 = 0,
(4.9)
the above contributions cancel, giving the asymptotics we claimed.

There is also evidence for the relation (4.1) in a different asymptotic region, namely
in an expansion of 2h 8 () 4 ()2 in powers of x, y, z, but exact in u, v, w. A 2 hrs
calculation using Maple 9 shows that the relation holds up to and including terms of order 4
in each variable, i.e., up to and including terms x 4 , y 4 , z4 , x 4 y 4 , x 4 z4 , y 4 z4 and x 4 y 4 z4 .
4.2. Evidence from special cases
For general x, y, z, w and u = v = 1 (or u = w = 1 or v = w = 1) the relation follows
from the corresponding properties at genus 2, as well as from the relation 42 = 28 at
genus 1.
4.3. Numerical evidence
The genus 3 -constants were computed numerically, up to and including 50 significant digits at random points in moduli space (this means picking points and verifying that
Im > 0). The Riemann relations are then verified numerically and were found to agree
to approximately 50 digits (recall that there are 36 terms, so a systematic error would cumulate in the sum and we should expect accuracy to 48 significant digits). Finally, the
relations (4.1) was verified numerically. It was found to hold within about 48 digits. This
has been carried out for seven points in moduli space; six with purely imaginary and one
with complex . The values of the points in moduli space were chosen so that Im > 0
(see Table 2). Notice that real x, y, z, u, v, w corresponds to purely imaginary . Numerically, this is the situation when the fewest cancellations occur and the highest precision
can be expected.
5. Degeneration limits
The behavior of a holomorphic modular covariant form near the divisor of Riemann
surfaces with nodes is usually of considerable interest. For example, before the Gliozzi
Table 2
Numerical evidence for 4 = 88 in genus 3
x
|8 42 /8|
0.3
0.3
0.3
0.3
0.02
0.0468
0.0468 + 0.1i
0.4
0.4
0.4
0.4
0.04
0.0666
0.0666 0.2i
0.5
0.5
0.5
0.5
0.05
0.0734
0.0734
1.1
1.1
1.1
1.1
0.11
0.962
0.962
1.2
1.2
1.2
1.2
0.12
1.11
1.11 0.07i
1.4
1.3
1.2
1.1
0.14
1.25
1.25
5.75 106
4.68 106
3.98 106
3.56 106
72.1
7.72
27.6 23.6i
< 1035
< 1035
< 1035
< 1036
< 1047
< 1046
< 1045
73
ScherkOlive projection, the superstring measure must exhibit a double pole, due to the
presence of the tachyon (see, e.g., [13] and [10] for a detailed discussion). In this section,
we shall analyze the behavior of 6# in the case of genus 3, when the underlying surface
separates into a genus 1 surface (1) and a genus 2 surface (2) .
Let (AI , BI ), 1 I 3, be a canonical homology basis for the genus 3 surface .
Assume that the basis splits up in an obvious way, with (A1 , A2 ; B1 , B2 ) associated with
the genus two surface while (A3 , B3 ) is associated with the genus 1 surface. The period
(3)
matrix I J , I, J = 1, 2, 3, may be parametrized as follows,

11 12 1
(3) = 21 22 2 ,
(5.1)
1
2 33
and in the separating limit described above, we have 1 , 2 0, while 33 and
{I J }1I,J 2 tend to the period matrices and of (1) and (2) .
It is natural to parametrize genus 3 spin structures following the decomposition of the
homology. Even spin structure fall into two classes {ia } and {m0 }, according to whether
the spin structures on the disconnected components is even/even or odd/odd,

i
ia
(5.2)
,
m0 m .
a
0
Here, a = 1, 2, 3 runs over the even spin structures at genus 1, i = 0, 1, . . . , 9 runs over
the even spin structures of genus 2 and m = 1, . . . , 6 runs over the odd spin structures of
genus 2. To save notation, we denote the unique odd spin structure of genus 1 by 0 , instead
of 0 as previously. The genus 3 -constants (non-zero only for even spin structures) have
a very simple behavior under the separating degeneration limit:
[ia ](0, ) 2 [i ](0, 2 ) 1 [a ](0, 3 ) 2 [i ]1 [a ],
[m0 ](0, ) 0.
(5.3)
Thus, to study the leading behavior of the constants at genus 3, we shall be interested
only in the spin structures ia and not m0 .
Theorem 5. Under a separating degeneration into a genus 2 and a genus 1 surface, the
limit of the genus 3 form 6# []( (3) ) for an even spin structure with even/even factorization on each disconnected component is given by

6# [] (3) = F ( )6 []() + 16(
(5.4)
)12 2A []() + S []() .
Here is the spin structure component of on the genus 2 surface, F ( ) is the genus 1
form defined by
F ( ) = 2 ( )12 + 3 ( )12 + 4 ( )12 + 34 ( )4 3 ( )8
+ 33 ( )4 2 ( )8 + 33 ( )4 4 ( )8 ,
and A [](), S []() are the genus 2 forms defined both by sums of the form

|
|| [
]4 (0, )[]4 (0, )[]4 (0, ),
(5.5)
(5.6)
74
but over different sets of triplets {

, , } of even spin structures. For A [](), the triplet
{
, , } is required to be asyzygous, while for S [](), it is required to be syzygous. For
both, {
, , } is also required to combine with into a quartet which contains exactly 2
asyzygous triplets.
Proof. Under Sp(6, Z) modular transformations, we can map any even spin structure into
= (0|0), in which case 6 []( (3) ) becomes

1
6# [] (3) =
(5.7)
[
]4 0, (3) []4 0, (3) []4 0, (3) .
2
[,
,,]
In the degenerating limit, only spin structures of the form ia will survive, so we may
immediately restrict
, , to be of that form,

i
j
k
,
j b
,
kc
.
ia
(5.8)
a
b
c
The asymptotics is now manifest,

26# [] (3)

2 [i ]4 2 [j ]4 2 [k ]4 1 [a ]4 1 [b ]4 1 [c ]4 .
(5.9)
[11 ,
ia ,j b ,kc ]
It remains to identify the totally asyzygous quartets that contribute and to parametrize them
in a convenient way. The conditions are
e(1 , i , j ) e(1 , a , b ) = 1,
e(1 , j , k ) e(1 , b , c ) = 1,
e(1 , k , i ) e(1 , c , a ) = 1,
e(i , j , k ) e(a , b , c ) = 1.
(5.10)
It is not hard to analyze the equation on the genus 1 spin structures. Denoting the choice
of the 3 genus 1 spin structures simply by their numbers abc, and listing the 4 values of e
above in a row vector, we have
111, 222, 333
+ + ++
123, 132
+ +
112, 121, 211
+ + ++
231, 321
+ +
113, 131, 311
+ + ++
312, 213
+ +
122, 212, 221
+ + ++
223, 332
+ +
133, 313, 331
+ + ++
232, 323
++
322, 233
+ +
(5.11)
Thus, up to permutations, there are only two cases. The case where all signs are + produces
genus 2 totally asyzygous quartets of the form [1 , i , j , k ], since we have
e(1 , i , j ) = 1,
e(1 , a , b ) = +1,
e(1 , j , k ) = 1,
e(1 , b , c ) = +1,
e(1 , k , i ) = 1,
e(1 , c , a ) = +1,
e(i , j , k ) = 1,
e(a , b , c ) = +1.
75
(5.12)
Thus {1 , i , j , k } form a totally asyzygous quartet, and we recognize the structure of

6 [1 ]( (2) ). Furthermore, the fact that they are totally asyzygous implies that these spin
structures must all be distinct. The corresponding contributions tend towards the first term
on the right-hand side of (5.4).
In the remaining contributions, with two + and two signs in (5.11), the genus 1 part is
actually always proportional to 12 . This is clear in the first three lines, where the product
24 34 44 = 1612 manifestly factors out. But it is also true in the last three lines, upon
adding up the contributions from the two cases listed. For example,
223 28 44 ,
332 24 48 .
(5.13)
Their sum is 28 44 + 24 48 = 24 34 44 = 1612 . It remains only to identify the genus 2 contribution. In the present case, the values e(1 , i , j ), e(1 , j , k ), e(1 , i , k ), e(i , j , k )
must contain exactly two signs + and two signs , i.e., the quartet {1 , i , j , k } contains exactly two syzygous triplets and two asyzygous triplets. These quartets come in two
groups, one group characterized by the fact that {i , j , k } is an asyzygous triplet, and
the other characterized by the fact that {i , j , k } is syzygous. Since the first three lines
of (5.11) correspond to {i , j , k } asyzygous and the next three to {i , j , k } syzygous,
Theorem 5 follows. 2
Appendix A. Modular transformations for any genus

Modular transformations in genus h form the infinite discrete group Sp(2h, Z), defined
by

M=
A
C

B
,
D
A
C
B
D

0
I
I
0

A
C
B
D
t

=
0
I

I
,
0
(A.1)
where A, B, C, D are integer valued h h matrices and the superscript t denotes transposition. To exhibit the action of the modular group on 1/2 characteristics (even or odd), it
is convenient to assemble the 1/2 characteristics into a single column of 2h entries and the
action of the modular group is then given by [14]

1
D C
CD T
=
+ diag
.
(A.2)
B A

AB T
2
Here and below, diag(M) of a h h matrix M is an 1 h column vector whose entries are
the diagonal entries on M. On the period matrix, the transformation acts by
= (A + B)(C + D)1 ,
while on the Jacobi -functions, the action is given by
(A.3)
76
t

[]
(C + D)1 ,
1
=
(, M) det(C + D) 2 ei
t (C+D)1 C
[](, ),
(A.4)
where = ( | ) and = ( | ). The phase factor

(, M) depends upon both and
the modular transformation M and obeys
(, M)8 = 1.
A.1. Convenient generators of the modular group
Using a Lemma 15 of [8], the group is generated by

A B
0 I
and
,
0 D
I 0
(A.5)
where I is the h h identity matrix. The first set is easily seen to be generated in turn by
the subgroups generated by matrices of the form MA and MB , so that the full Sp(2h, Z) is
generated by three subgroups,

A
0
I B
0 I
,
M
MA =
(A.6)
=
,
S
=
,
B
0 (At )1
0 I
I 0
1
where A SL(h, Z) and B t = B, the latter forming the translation group Z 2 h(h+1) . The
action on spin structures for each of the 3 subgroups is given by

MA ( | ) = (At )1 |A ,

1
MB ( | ) = |A B + diag B ,
2

S( | ) = ( | ).
(A.7)
A.2. Modular transformations of -constants
We shall be most interested in the modular transformations of -constants of even spin
structures , which are given by
=
(, M) det(C + D) 2 [](0, ),
[](0,
)
1
(A.8)
and we have
(, MA ) = 1,

(, MB ) = exp i t B + i t diag B ,
(, S) = exp{ih/4}.
A.3. Modular transformations of signatures
The signature on pairs is not invariant under all modular transformations; instead,
MA |MA
= |
,

MB |MB
= |
exp 2i(
)t diag B ,
(A.9)
S|S
= |
.
77
(A.10)
This transformation law means, however, that the signature for triplets is invariant under
all modular transformations. It suffices to check this for transformations MB ,
e(MB , MB
, MB ) = MB |MB
MB
|MB MB |MB = e(,
, ),
(A.11)
where the exponential factors arising from the transformation law of the three pairs cancel.
In particular, totally asyzygous N -tuples are mapped into totally asyzygous N -tuples under
modular transformations.
Appendix B. Spin structures and asyzygies in genus 2

The spin structures for genus 2 Riemann surfaces are described in detail in [4, Sections
2.12.3]. For the readers convenience, we reproduce here the notation, since it is used
extensively below. The odd spin structures may be labeled by

0 0
,
21 =
11

1 1
,
22 =
00

0 1
23 =
,
11

1 1
24 =
,
01

1 0
25 =
,
11

1 1
26 =
,
10
(B.12)
and the even spin structures by

0 0
,
00

0 0
25 =
,
10
21 =

0 0
,
01

0 1
26 =
,
10
22 =

0 1
,
00

1 0
27 =
,
00

1 0
29 =
,
10
23 =

0 1
,
01

1 0
28 =
,
01

1 1 (B.13)
20 =
.
11
24 =
Generators of genus 2 modular transformations M1 , M2 , M3 , , T , S are chosen as they

were defined in [4].
B.1. Tables of asyzygies in genus 2
The properties of asyzygies often require a case by case treatment. For genus 2, it is
convenient to simply make full tables of asyzygies, and check the desired properties by
inspection of these tables. In this appendix, we provide the tables which have been used in
this paper.
78
There are 120 distinct triples of which 60 are syzygous, e = +1,

(123)
(157)
(230)
(289)
(379)
(489)
(124)
(159)
(245)
(290)
(380)
(490)
(127)
(168)
(246)
(347)
(390)
(579)
(128)
(160)
(256)
(348)
(456)
(570)
(134)
(178)
(258)
(356)
(457)
(589)
(135)
(179)
(259)
(358)
(450)
(580)
(136)
(180)
(267)
(350)
(468)
(679)
(149)
(190)
(260)
(367)
(469)
(670)
(140)
(234)
(278)
(369)
(478)
(689)
(156)
(239)
(270)
(378)
(470)
(680)
(137)
(167)
(249)
(349)
(467)
(678)
(138)
(169)
(240)
(340)
(460)
(690)
(139)
(170)
(257)
(357)
(479)
(789)
(130)
(189)
(250)
(359)
(480)
(780)
(145)
(235)
(268)
(368)
(567)
(790)
(146)
(236)
(269)
(360)
(568)
(890)
and 60 are asyzygous, e = 1,

(125)
(147)
(237)
(279)
(370)
(569)
(126)
(148)
(238)
(280)
(389)
(560)
(129)
(158)
(247)
(345)
(458)
(578)
(120)
(150)
(248)
(346)
(459)
(590)
B.2. Asyzygous quartets in genus 2

The list of asyzygous quartets can be easily read off the table of asyzygous triplets given
above. It turns out that there are 15 asyzygous quartets in all. They are given in Table 3,
together with their transformations under modular transformations.
B.3. Admissible sextets in genus 2
We consider sextets of even spin structures in genus 2 satisfying the property listed in
Theorem 2: they can be written as the union of three pairs, the union of any two pairs
forming a totally asyzygous quartet. For convenience, we sometimes refer to sextets with
Table 3
Modular transformations of totally asyzygous quartets (TAQ)
TAQ
M1
M2
M3
1250
1269
1370
1389
1458
1467
2357
2368
2479
2480
3459
3460
5678
5690
7890
3460
3459
1370
1389
2368
2357
1467
1458
2479
2480
1269
1250
5678
5690
7890
2150
1269
2480
2479
2357
2368
1458
1467
1389
1370
3459
3460
5678
5690
7890
1269
1250
1389
1370
1467
1458
2368
2357
2480
2479
3460
3459
5678
5690
7890
1520
1584
1730
1467
1269
1389
2357
5678
3459
5690
2479
7890
2368
2480
3460
1370
1389
1250
1269
1467
1458
2357
2368
3459
3460
2479
2480
5678
7890
5690
1458
1467
1398
1370
1250
1269
3459
3460
2497
2480
2357
2368
5690
5678
7890
79
Table 4
Modular transformations and TAQ structure of admissible sextets
Sextet
M3
AB
B C
C A
123570
123689
124580
124679
125690
134589
134670
137890
145678
234579
234680
235678
247890
345690
567890
123689
123570
124679
124580
125690
134670
134589
137890
145678
234680
234579
235678
247890
345690
567890
+
+
+
+
+
+
123570
145678
125690
134589
124580
124679
137890
134670
123689
234579
567890
235678
345690
247890
234680
134589
134670
124580
124679
145678
123570
123689
137890
125690
234579
234680
345690
247890
235678
567890
123570
123689
134670
134589
137890
124679
124580
125690
145678
234579
234680
235678
345690
247890
567890
10
19
48
16
12
18
17
13
14
27
28
23
24
34
56
37
38
20
47
69
45
46
89
67
49
40
68
80
60
90
25
26
15
29
50
39
30
70
58
35
36
57
79
59
78
this property as admissible sextets. In the table below, we have listed all the 15 sextets
in genus 2 satisfying this property. The three totally asyzygous quartets that each sextet
contains have been denoted by A, B, and C. The three pairs can then be recaptured by
taking A B, B C, and A C.
Given a fixed spin structure , the admissible sextets split into two groups s c [] and
s[], characterized, respectively, as the sextets which include and do not include . For
each fixed , s[] always consists of 6 sextets, and s c [] of 9 sextets. In Table 4, the first 6
sextets are the sextets in s[7 ], and the remaining ones are the sextets in s c [7 ].
Appendix C. Genus 3 spin structures and modular transformations

Here, we provide a summary of the computer results obtained on genus 3 spin structures, as well as multiplets of spin structures. The genus 3 spin structures are labeled as
follows,
1 = [0, 0, 0, 0, 0, 0],
13 = [0, 1, 0, 0, 1, 0],
25 = [1, 0, 0, 1, 0, 1],
2 = [0, 0, 0, 0, 0, 1],
14 = [0, 1, 0, 1, 0, 0],
26 = [1, 0, 0, 1, 1, 0],
3 = [0, 0, 0, 0, 1, 0],
15 = [0, 1, 0, 1, 0, 1],
27 = [1, 0, 1, 0, 0, 0],
4 = [0, 0, 0, 1, 0, 0],
16 = [0, 1, 0, 1, 1, 0],
28 = [1, 0, 1, 0, 0, 1],
5 = [0, 0, 0, 1, 0, 1],
17 = [0, 1, 1, 0, 0, 0],
29 = [1, 0, 1, 0, 1, 0],
6 = [0, 0, 0, 1, 1, 0],
18 = [0, 1, 1, 0, 0, 1],
30 = [1, 0, 1, 1, 1, 1],
7 = [0, 0, 1, 0, 0, 0],
19 = [0, 1, 1, 0, 1, 0],
31 = [1, 1, 0, 0, 1, 1],
8 = [0, 0, 1, 0, 0, 1],
20 = [0, 1, 1, 1, 1, 1],
32 = [1, 1, 0, 1, 1, 1],
9 = [0, 0, 1, 0, 1, 0],
21 = [1, 0, 0, 0, 0, 0],
33 = [1, 1, 1, 0, 1, 1],
80
10 = [0, 0, 1, 1, 1, 1],
22 = [1, 0, 0, 0, 0, 1],
34 = [1, 1, 1, 1, 0, 0],
11 = [0, 1, 0, 0, 0, 0],
23 = [1, 0, 0, 0, 1, 0],
35 = [1, 1, 1, 1, 0, 1],
12 = [0, 1, 0, 0, 0, 1],
24 = [1, 0, 0, 1, 0, 0],
36 = [1, 1, 1, 1, 1, 0],
(C.14)
and we take the following basis of genus 3 modular generators

0
0
I bi
ai
Bi =
,
S
=
,
Ai =
I
0 I
0 (ait )1

I
,
0
(C.15)
where

1 0 0
b1 = 0 0 0 ,
0 0 0

0 1 0
b4 = 1 0 0 ,
0 0 0

0 1 0
a1 = 1 0 0 ,
0 0 1

1 1 0
a4 = 0 1 0 ,
0 0 1

0 0 0
0 0 0
b2 = 0 1 0 ,
b3 = 0 0 0 ,
0 0 0
0 0 1

0 0 1
0 0 0
b5 = 1 0 0 ,
b6 = 0 0 1 ,
0 0 0
0 1 0

1 0 0
0 0 1
a2 = 0 0 1 ,
a3 = 0 1 0 ,
0 1 0
1 0 0

1 0 1
1 0 0
a5 = 0 1 0 ,
a6 = 0 1 1 .
0 0 1
0 0 1
(C.16)
The action of all elementary modular generators (Ai , Bi , i = 1, . . . , 6 and S) are given in
Table 5. The action of the additional generator B1 SB1 SB1 is also given. The generators B4 , B5 , B6 , A1 , A2 , A3 , A4 , A5 , A6 , S, and form a subgroup leaving 1 invariant.
It is conjectured that these generators span the maximal subgroup of Sp(6, Z2 ) which
leaves 1 invariant.
C.1. Multiplets of spin structures
Asyzygous combinations play an important role. An asyzygous triplet (1 , 2 , 3 ) is
defined so that e(1 , 2 , 3 ) = 1. An asyzygous N -plet (with N > 3) is a set of spin
structures such that each triplet of distinct spin structures is asyzygous. The count is as
follows
asyzygous quartets
5040,
asyzygous quintets 2016,

asyzygous sextets
336,
asyzygous septets
0.
(C.17)
81
Table 5
Modular transformations of genus 3 even spin structures
B1
B2
B3
B4
B5
B6
A1
A2
A3
A4
A5
A6
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
11
12
13
14
15
16
17
18
19
20
1
2
3
4
5
6
7
8
9
10
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
4
5
6
1
2
3
7
8
9
10
14
15
16
11
12
13
17
18
19
20
24
25
26
21
22
23
27
28
29
30
32
31
33
34
35
36
2
1
3
5
4
6
8
7
9
10
12
11
13
15
14
16
18
17
19
20
22
21
23
25
24
26
28
27
29
30
31
32
33
35
34
36
1
2
3
4
5
6
17
18
19
20
11
12
13
14
15
16
7
8
9
10
24
25
26
21
22
23
34
35
36
33
32
31
30
27
28
29
1
2
13
4
5
16
7
8
19
20
11
12
3
14
15
6
17
18
9
10
22
21
31
25
24
32
28
27
33
36
23
26
29
35
34
30
1
2
6
4
5
3
8
7
10
9
11
12
16
14
15
13
18
17
20
19
21
22
26
24
25
23
28
27
30
29
32
31
36
35
34
33
1
2
3
11
12
13
21
22
23
31
4
5
6
14
15
16
24
25
26
32
7
8
9
17
18
19
27
28
29
33
10
20
30
34
35
36
1
4
7
2
5
8
3
6
9
10
11
14
17
12
15
18
13
16
19
20
21
24
27
22
25
28
23
26
29
30
34
35
36
31
32
33
1
11
21
4
14
24
7
17
27
34
2
12
22
5
15
25
8
18
28
35
3
13
23
6
16
26
9
19
29
36
31
32
33
10
20
30
1
2
3
14
15
16
7
8
9
20
11
12
13
4
5
6
17
18
19
10
27
28
29
34
35
36
21
22
23
32
33
30
31
24
25
26
1
12
3
4
15
6
7
18
9
20
11
2
13
14
5
16
17
8
19
10
23
31
21
26
32
24
29
33
27
35
22
25
28
36
30
34
1
2
9
5
4
10
7
8
3
6
11
12
19
15
14
20
17
18
13
16
21
22
29
25
24
30
27
28
23
26
33
36
31
35
34
32
1
3
2
7
9
8
4
6
5
10
21
23
22
27
29
28
24
26
25
30
11
13
12
17
19
18
14
16
15
20
31
33
32
34
36
35
1
2
3
4
5
6
7
8
9
10
21
22
23
24
25
26
27
28
29
30
11
12
13
14
15
16
17
18
19
20
31
32
33
34
35
36
References
[1] E. DHoker, D.H. Phong, Two-loop superstrings I, main formulas, Phys. Lett. B 529 (2002) 241255, hepth/0110247.
[2] E. DHoker, D.H. Phong, Two-loop superstrings II, the chiral measure on moduli space, Nucl. Phys. B 636
(2002) 360, hep-th/0110283.
[3] E. DHoker, D.H. Phong, Two-loop superstrings III, slice independence and absence of ambiguities, Nucl.
Phys. B 636 (2002) 6179, hep-th/0111016.
[4] E. DHoker, D.H. Phong, Two-loop superstrings IV, the cosmological constant and modular forms, Nucl.
Phys. B 639 (2002) 129181, hep-th/0111040.
[5] E. DHoker, D.H. Phong, Lectures on Two-Loop Superstrings, Hangzhou, Beijing, 2002, hep-th/0211111.
[6] E. DHoker, D.H. Phong, The geometry of string perturbation theory, Rev. Mod. Phys. 60 (1988) 9171065.
82
[7] J.I. Igusa, On the graded ring of theta constants, Amer. J. Math. 86 (1964) 219;
J.I. Igusa, On Siegel modular forms of genus two, Amer. J. Math. 84 (1962) 175;
J.I. Igusa, Modular forms and projective invariants, Amer. J. Math. 89 (1967) 817855.
[8] J.I. Igusa, Theta Functions, Springer-Verlag, Berlin, 1972.
[9] K. Aoki, E. DHoker, D.H. Phong, Two-loop superstrings on orbifold compactifications, Nucl. Phys. B 688
(2004) 369, hep-th/0312181.
[10] E. DHoker, D.H. Phong, Asyzygies, modular forms, and the superstring measure II, Nucl. Phys. B 710
(2005) 83116, the next article in this issue, hep-th/0411182.
[11] S. Kachru, J. Kumar, E. Silverstein, Vacuum energy cancellation in a nonsupersymmetric string, Phys. Rev.
D 59 (1999) 106004, hep-th/9807076;
S. Kachru, E. Silverstein, Self-dual nonsupersymmetric type II string compactifications, JHEP 9811 (1998)
001, hep-th/9808056;
S. Kachru, E. Silverstein, On vanishing two loop cosmological constants in nonsupersymmetric strings,
JHEP 9901 (1999) 004, hep-th/9810129;
G. Shiu, S.H.H. Tye, BoseFermi degeneracy and duality in non-supersymmetric strings, Nucl. Phys. B 542
(1999) 45, hep-th/9808095.
[12] E. Freitag, Hilbert Modular Forms, Springer-Verlag, Berlin, 1990;
E. Freitag, Siegelsche Modulfunktionen, Springer-Verlag, Berlin, 1983.
[13] A. Belavin, A. Polyakov, A. Zamolodchikov, Infinite conformal symmetry in two-dimensional quantum
field theory, Nucl. Phys. B 241 (1984) 333380;
D. Friedan, S. Shenker, The analytic geometry of two-dimensional conformal field theory, Nucl. Phys. B 281
(1987) 509545;
D. Friedan, S. Shenker, The integrable analytic geometry of quantum string, Phys. Lett. B 175 (1986) 287
296.
[14] J. Fay, Theta Functions on Riemann Surfaces, in: Lecture Notes in Mathematics, vol. 352, Springer-Verlag,
Berlin, 1973.
Asyzygies, modular forms,

and the superstring measure II
Eric DHoker a , D.H. Phong b
a Department of Physics and Astronomy, University of California, Los Angeles, CA 90095, USA
b Department of Mathematics, Columbia University, New York, NY 10027, USA

Abstract
Precise factorization constraints are formulated for the three-loop superstring chiral measure, in the
separating degeneration limit. Several natural ansatzes in terms of polynomials in theta constants for
the density of the measure are examined. None of these ansatzes turns out to satisfy the dual criteria
of modular covariance of weight 6, and of tending to the desired degeneration limit. However, an
ansatz is found which does satisfies these criteria for the square of the density of the measure, raising
the possibility that it is not the density of the measure, but its square which is a polynomial in theta
constants. A key notion is that of totally asyzygous sextets of spin structures. It is argued that the
ansatz produces a vanishing cosmological constant.
PACS: 11.25.-w; 11.25.Db; 11.25.Hf
1. Introduction
Recently, the superstring measure to two-loop order and for even spin structure was
computed from first principles [17]. The construction relies on a careful treatment of
supermoduli, chiral splitting and finite-dimensional gauge fixing determinants, and builds
on earlier work in this direction [8,9]. Although intermediate calculations are complex and
Research supported in part by National Science Foundation grants PHY-01-40151 and DMS-02-45371.
E-mail address: phong@math.columbia.edu (D.H. Phong).
doi:10.1016/j.nuclphysb.2004.12.020
84
intricate, the final form of the superstring measure turns out to be very simply expressed in
terms of a new modular object, denoted by 6 []() in [4].
At present, no analogous derivation is available to 3-loop order and beyond. Some of
the special simplicity of genus 2 does carry over to genus 3, in that no Schottky relations
need to be imposed on the period matrix. The structure of supermoduli, however, becomes
considerably more complex and, at present, the calculation appears formidable.
Therefore, the simplicity of the ultimate form of the two-loop superstring measure raises
the question as to whether the genus 3 superstring measure might have a comparatively
simple form in terms of natural modular objects. Constraints from holomorphicity, modular
invariance, and physical factorization will provide powerful restrictions on any candidate
measures. The precise form of the 2-loop measure gives a drastic constraint on the separating degeneration limits of the 3-loop measure.1
In this paper, we take a first step in the degeneration approach to the superstring measure by formulating a precise ansatz for the 3-loop measure and verifying that it satisfies
the correct factorization conditions when the worldsheet degenerates. Our ansatz for the
(chiral) superstring measure d[]( (3) ) can be described as follows. Set
[](0, (3) )4 6 []( (3) )

(3)
dI J .
d[] (3) =
8 4 9 ( (3) )
(1.1)
I J
Here is a fixed even spin structure, (3) = {I(3)

J } is the period matrix of the genus 3
(3) 2
worldsheet, 9 (I J ) is the modular form 18 ( (3) ) of weight 18 constructed in [13],

1
dI(3)
and the measure 9 (I(3)
J)
J has been shown to be holomorphic in [11]. The key
term 6 []( (3) ) is to be determined by the following constraints:
(i) 6 []( (3) ) is holomorphic in (3) on the Siegel upper half space;
(ii) 6 []( (3) ) is a modular covariant form of weight 6 in the sense that, under modular
(3)
(3)
we
transformations sending I J I J = (A (3) + B)(C (3) + D)1 , ,
have

(3) = (, M)4 det C (3) + D 6 6 [] (3) ,
6 []
(1.2)
where (, M) is the same phase factor as in the modular transformation for constants.
(iii) In the degeneration t 0, where the worldsheet separates into a genus 1 and a genus 2
surface of period matrices (1) and (2) , respectively, we must have

12

lim 6 [] (3) = (1) 6 [] (2) ,
(1.3)
t0
where 6 []( (2) ) is the main new factor in the genus 2 superstring measure found
in [1,4], and is the spin structure induced by on the genus 2 component.
1 The constraints of modular invariance were used along these lines to guess the bosonic string measure to 2-
and 3-loops in [10] and [11], respectively. A general theory based on constraints from modular invariance and
physical factorization was developed in [12].
85
The constraint (iii) on the degeneration limit of 6 []( (3) ) is a consequence of the
factorization properties of string amplitudes. To establish it, we require a precise formula

(3)
for the degeneration limit of the measure 9 ( (3) )1 dI J , formula which is also one
of the main results of this paper (see Theorem 1 below).
We should stress that the condition (iii) is very restrictive, since it applies to an arbitrary
separating degeneration. Thus we have to expect 6 []( (3) ) to be built of sums of many
terms, different groups of which would tend to 0 in different limits.
The original expression for 6 []( (2) ) derived in [14] depended very much on the
fact that the worldsheet had genus 2. Since then, two alternate expressions have been found
which can extend to higher genus [14]. A characterizing feature of these two expressions
is that one of them is a sum over fourth powers of -constants of triplets of spin structures,
while the other is a sum of second powers of -constants of sextets of spin structures. The
key to determining which N -tuplets {i } of spin structures should contribute to 6 []()
turns out to be the notion of total asyzygies. Recall that to any triplet of spin structures
{1 , 2 , 3 } is associated a modular invariant sign, namely, the product
e(1 , 2 , 3 ) = 1 |2 2 |3 3 |1
(1.4)
of relative signatures | = exp 4i( ). A triplet of spin structures is said to

be syzygous or asyzygous, depending on whether e is +1 or 1. The criteria for which
triplets or sextets should contribute to 6 []() turns out to be entirely expressible in
terms of asyzygies (see Section 5.1 below). Once the criteria for which triplets or sextets to
include has been identified, one needs to find phase assignments (; {i }) with which to
sum the contributions of various sextets {i }. The phase assignments have to be consistent
with modular invariance, which identifies them all up to a global phase.
These alternative descriptions of 6 []() suggest several possible generalizations to
genus 3, all involving summations over monomials in [i ]. They are listed in Section 5.2,
where we describe also in detail their viability as ansatzes for the genus 3 superstring chiral
measure 6 []( (3) ). The net outcome is the following:
A first ansatz is in terms of sums of products of three fourth powers only, such as
[i1 ]4 [i2 ]4 [i3 ]4 . Using in particular the degeneration formulas of [14], we show
that none of this form exists which satisfies the criteria (ii) and (iii). More generally, the
criterion (iii), requiring the appearance of ( (1) ), effectively prevents the rule for which
N -tuplets to be included to remain the same for all genera.

Next, we consider ansatzes involving sums of second powers, such as 6j =1 [ij ]2 .
In genus 2, the sextets which contribute to 6 []() can be characterized by the condition of -admissibility (see Section 5.1). This condition makes sense for all genera, but in
genus 3, the set of such sextets (called -admissible by extension) breaks up into many
orbits under the subgroup of modular transformations fixing a given spin structure . One
particularly important orbit is the set of sextets which do not contain , and which are totally asyzygous, in the sense that any of their sub-triplets is asyzygous. We refer to the other
orbits as partially asyzygous. The partially asyzygous orbits do not appear to have as simple a description as the orbit of totally asyzygous sextets, although they can be identified
by computer analysis.
86
The partially asyzygous orbits turn out not to be viable candidates for 6 []( (3) ):
computer analysis reveals that many of them do not admit consistent phase assignments
(; {i }). Even when they do, their degeneration limits do not satisfy the criterion (iii)
listed above. Thus we rule them out as ansatzes for 6 []( (3) ).
We found the criterion of totally asyzygous sextets to be much more compelling: its key
property is that the genus 2 sextets obtained by factorization from a totally asyzygous genus
3 sextet automatically satisfy the key condition of admissibility in genus 2 (see Lemma 1
in Section 6.1). Furthermore, although these genus 2 sextets may be admissible but not
-admissible, Lemma 2 in Section 6.2 shows that the contributions of such sextets sum
up to 0 if they are assigned phases consistent with modular invariance. Thus the ansatz in
terms of totally asyzygous sextets would satisfy the degenerating condition (iii) if phase
assignments exist which are consistent with (ii). However, perhaps surprisingly, such a
consistent phase assignment does not exist and (ii) cannot be satisfied. A simple example
is provided in Section 6.2.2.
Another
possible ansatz could be in terms of sums of products of twelve first powers
(3)
of , such as 12
i=1 [i ](0, ). The criterion for which dozens {i }1i12 to include
is difficult to guess from the genus 2 case. There is no consistent phase assignments if the
dozens are assumed to consist of a pair of totally asyzygous sextets, and more generally,
no consistent sign assignments appear possible.
Thus, we are led to believe that no candidate for 6 []( (3) ) exists which is a polynomial in . On the other hand, consistent modular covariant assignments (; {i }, {i })
do exist for suitable bilinear combinations of pairs of totally asyzygous sextets of [i ]2 .
This suggests that only 6 []( (3) )2 is a polynomial in -constants. We find that, for a
suitable integer normalization factor N , and a suitable choice of multiplicities Npq of the
orbits Qpq of pairs {i , i } of totally asyzygous sextets under the subgroup of Sp(6, Z)
fixing , the expression

2
1
6 [] (3) = 8
Npq
2 N pq
6

i=1
2 [i ]

; {i }, i
({i },{i })Qpq

6

2 i
(1.5)
i=1
does satisfy all the conditions implied by (i)(iii) for the square of 6 []( (3) ), for arbitrary separating degeneration limits. In particular, it is a highly non-trivial result that
in any separating degeneration limit of this form to a genus 2 and a genus 1 surface, the
limit becomes a perfect square. In general, these expressions will not admit holomorphic
square roots away from the separating degeneration limit. If there exists a specific choice
of multiplicities Npq (not all 0) which guarantees the existence of a holomorphic square
root, then (1.5) will single out a compelling candidate for the genus 3 superstring measure.
The existence of such a holomorphic square root is knownto occur at genus 3 in at least
one other instance, namely, the modular form 9 ( (3) ) = [](0, (3) )1/2 , which is
known to be the (unexpectedly) holomorphic square root of 18 ( (3) ).
87
The remainder of this paper is organized as follows. In Section 2, the general criterion
for physical factorization is spelled out for the superstring measure. In Section 3, the factorization properties of the bosonic factors in the genus 3 measure are derived. In Section 4,
the construction of the genus 3 superstring measure is formulated as a degeneration problem. In Section 5, the consistency with criteria (i), (ii), (iii) above of various candidates is
analyzed and (1.5) is constructed.
2. Factorization of the superstring measure

The main goal of this section is to derive the precise degeneration constraints which the
3-loop superstring measure must satisfy when a separating cycle in the worldsheet (3) is
pinched to a point, and (3) separates into a torus (1) and a genus 2 surface (2) .
2.1. Geometric picture of factorization
We begin with the geometric description of the moduli space of Riemann surfaces near
the divisor of surfaces with nodes, as provided by the following well-known construction [15].
Let (1) and (2) be two Riemann surfaces of genus h1 and h2 , let p1 (1) , p2
(2)
be two given points, and let |z1 | < 1, |z2 | < 1 be local coordinates on (1) and
(2) which are centered at p1 and p2 , respectively. Let S be the surface given by S =
{(X, Y, t); XY = t, |X| < 1, |Y | < 1, |t| < 1}, and construct the fibration C of surfaces
over the unit disk {t; |t| < 1} given by

C = (z1 , t); z1 (1) , |z1 | > |t| S (z2 , t); z1 (2) , |z2 | > |t| ,
(2.1)
with the following identifications

t
(z1 , t) z1 , , t
for z1 (1) , |t| < |z1 | < 1,
z1

t
(z2 , t)
, z2 , t
for z2 (2) , |t| < |z2 | < 1.
z2
(2.2)
For each t = 0, the fiber of S above t can be identified with the annulus At = {X; |t| <
|X| < 1}. Thus the fiber of C above t is a regular surface t of genus h = h1 + h2 , which
can be covered by the three overlapping charts (1) \{|z1 | > |t|}, At , and (2) \{|z2 | > |t|},
with the identifications
t
z1 X
(2.3)
for |t| < |z1 |, |z2 | < 1.
z2
2.2. Physical picture of factorization
In the physical picture, we view the surface t rather as the disjoint union
(1)
(2)
t = in
At out
,
(2.4)
88

(1)
(2)
where we have set in = (1) \{|z1 | < 1}, and out = (1) \{|z1 | < 1}. In a given con(1)
(2)
(1)
formal field theory, the surfaces with boundary in , out define two states in | and
(2)
|out . To make contact with the Hamiltonian picture, we can use the exponential map
X = t 1/2 e to identify the annulus At with a cylinder

1
1
1
1
< 0 < ln
.
= 0 + i1 ; 0 1 2, ln
(2.5)
2 |t|
2 |t|
Now the operators for time and space translations are the Hamiltonian H = L0 + L 0 and
the momentum operator P = L0 L 0 .2 If we view 0 as time, and 1 as space, then
the shift in time and the shift in space corresponding to the cylinder are given, respectively,
by the length of the cylinder and the phase shift in as the point X moves on a straight
line from X = |t| to X = 1. This gives ln |t| for the shift in time and arg(t) for the
1
1
1
1
shift in space, since = |t| 2 ei 2 arg(t) and = |t| 2 ei 2 arg(t) are the points on the cylinder
corresponding to X = |t|1/2 and X = 1. The cylinder corresponds then to the following
operator insertion

exp i arg(t)(L0 L 0 ) exp ln |t| (L0 + L 0 ) = t L0 tL0 ,

and hence the partition function Zt corresponding to the surface t is given by
(1)
(2)
Zt = in t L0 tL0 out .
(2.6)
(2.7)
To obtain the degenerating limit t 0, we insert a basis of states | diagonalizing t L0 tL0

(1)
(2)
in |t L0 tL0 | out .
Zt =
(2.8)
The descendant states | contribute lower order terms in the limit t 0. To identify
the leading contribution, we need thus to consider only primary states. In the case of
string propagation, before the GSO projection, the state with lowest m2 is the tachyon with
m2 = 2. By momentum conservation, its momentum must be k = 0 (it is not on-shell,
but intermediate states do not have to be on-shell). Since the vertex for tachyon emission
with momentum 0 is just the identity, the leading term for Zt is given by

Zt = Z (1) t 2 t2 Z (2) + O |t|3 ,
(2.9)
where Z (1) and Z (2) are the partition functions for the surfaces (1) and (2) .
To deal with spin structures, we start from surfaces (i) with canonical homology bases
(i)
(i)
(i)
(i)
(i)
(i)
(i)
(i)
AI , BI , #(AI BJ ) = I J , #(AI AJ ) = 0, #(BI BJ ) = 0 for 1 I, J hi .
Then the combined bases give a canonical basis for the genus h1 + h2 surface t . With
this choice of homology bases, a spin structure can be identified with an assignment of
either 0 or 1/2 to each homology cycle of t , and hence with a pair (1 , 2 ), with i a spin
2 Since all conformal anomalies ultimately cancel, we can ignore the contribution of the central charge when
we map the annulus into the cylinder.
structure on the surface (i)
= 2 .
1
89
(2.10)
In a conformal field theory where the fields are world sheet fermions requiring a spin
structure, the preceding degeneration formula becomes

Zt [] = Z (1) [1 ] t 2 t2 Z (2) [2 ] + O |t|3 .
(2.11)
2.3. Factorization of the genus 3 superstring measure
We formulate now the precise degeneration constraint for the superstring measure when
the worldsheet = t is of genus h = 3 and degenerates into two surfaces (1) and (2)
of genus h1 = 1 and h2 = 2.
We shall assume that, at loop order h, the vacuum-to-vacuum superstring amplitude is
of the form

5

(h) ,
det Im (h)
A=
(2.12)
c,
d[] (h) d[]
,
Mh
where c, are suitable phases, and the sum over the spin structures , corresponds to
the GSO projection, which projects out the tachyon and produces spacetime supersymmetry. The space Mh is the moduli space of Riemann surfaces of genus h. We always fix
a homology basis, and view each Riemann surface as characterized by its period matrix
(h)
(h) = {I J }1I,J h . The form d[]() is a (3h 3, 0) holomorphic form on Mh ,
transforming under modular transformations in such a way that the full expression above
is modular invariant. It is called the (chiral) superstring measure at genus h.
Near t = 0, the 3h 3 moduli parametrizing t can be chosen to be the 3h1 3 and
3h2 3 moduli for the surfaces (1) and (2) , together with the 3 parameters p1 , p2 and
t. The degeneration formulas derived above for conformal field theory suggest imposing
the following degeneration constraint for the chiral superstring measure
dt

d[]() = d[1 ] (1) 2 d[2 ] (2) dp1 dp2 + O t 1 .
(2.13)
t
As usual, these formulas hold for h1 , h2 2. When h1 = 1, the counting is slightly different, since p1 and its differential are no longer relevant due to translation invariance on
the torus. This is actually the case of main interest in the present paper, so we make the
above formula more explicit in this case: the moduli for (1) is then a single parameter
(1) , and the superstring measure for one-loop is 4 [1 ]( (1) )/25 4 12 ( (1) ) (see, e.g.,
[4], Eq. (8.2)). Thus the degeneration constraint for the chiral superstring measure at genus
h when the worldsheet separates into a torus (1) and a genus h 1 surface (2) is given
by
d[]() =

4 [1 ]( (1) )
dt
d (1) 2 d[2 ] (2) dp2 + O t 1 . (2.14)
5
4
12
(1)
t
2 ( )
90
2.4. Factorization of the genus 3 bosonic string measure

Although this paper is mainly concerned with the genus 3 superstring measure and its
degeneration limit, we take the opportunity to discuss also similar issues for the bosonic
string, partly as a check later on our method. The measure for the bosonic string in the
critical dimension is of the form

A=
(2.15)
(det Im )13 dB () dB (),
Mh
where dB () is holomorphic. Because the intermediate state of lowest mass is still the
tachyon, the measure dB () satisfies the same degeneration constraint as in (2.13).
When the worldsheet degenerates into a torus (1) and a surface of genus 2, the degeneration constraint can be written as
dB () =

d (1)
dt
2 dB (2) dp2 + O t 1 ,
12
24
(1)
(2) ( ) t
(2.16)
where (2)12 24 ( (1) ) d (1) is the genus 1 bosonic string measure, with the conventions of [8] and the normalization d 2 (1) /(8 2 Im (1) )2 for the SL(2, R) invariant
measure on the Siegel upper half space.
3. The measure
I J
(3)
dI(3)
J /9 ( ) in genus 3
An important feature of the chiral superstring measure d[]( (h) ) is that it is a holomorphic (3h 3, 0) form. To find it, we begin by constructing a natural holomorphic
(3h 3, 0) form dB ( (h) ) on Mh (later identified with the chiral bosonic measure, but
this is not essential for our considerations), so that the problem of finding d[]( (h) )
reduces to that of finding the density d[]/dB . In genera h = 2 and h = 3, we can exploit the fact that Mh and the Siegel upper half space of symmetric matrices with positive
imaginary part have the same dimension, and henceforth we consider only these cases.
3.1. The modular forms 18 ( (3) ) and 10 ( (2) )
Recall that on a surface of genus h, there are 22h spin structures, of which
+ 1) are even and 2h1 (2h 1) are odd. The parity of a spin structure corresponds to the parity in of the -function [](, (h) ), which is also the parity of
the number of independent holomorphic spinors of spin structure . The properties of
-functions which we need can be found in [4, Sections 2.12.3] and [7, Appendix B]. For
convenience, we restate here the transformations of spin structures and -constants
[](0, (h) ) [](0,

(h) ) under modular transformations

1
A B
(h)
(h)
(h)
= A + B C + D ,
(3.1)
M=
Sp(2h, Z).
C D
2h1 (2h
If we write = ( | ) and = ( | ), they are given by

1
D C
CD T
=
+ diag
B A

AB T

2
91
(3.2)
and by

= (, M) det C (h) + D 1/2 [] 0, (h) ,

[](0,
)
(3.3)
where (, M) is an eighth root of unity, which depends on both the spin structure and
the modular transformation M. There is no simple closed formula for (, M), but its
values for h = 2 on generators of Sp(4, Z) can be found in [4, Section 2.3].
The above transformation for -constants should be compared with the defining transformation law for modular forms () of a given weight w

w
(h) = det C (h) + D (h) ,
(3.4)
which do not involve roots of unity such as (, M). Nevertheless, the following natural
form can be defined using the even -constants

2k [] 0, (h) .
2h1 (2h +1)k (h) =
(3.5)
even
It has been shown by Igusa [13] that in genus h = 2 and h = 3, 2h1 (2h +1)k ( (h) ) are
modular forms of weight 2h1 (2h + 1)k when k = 1 and k = 1/2, respectively.
Let these forms be denoted by 10 ( (2) ) and 18 ( (3) ), respectively. It is well known
that the form 10 ( (2) ) has no zero inside the moduli space of Riemann surfaces of genus
2, while the form 18 ( (3) ) vanishes exactly of second order along the variety of hyperelliptic surfaces of genus 3 [11]. Indeed, 2h1 (2h +1) ( (h) ) vanishes if and only if a
-constant vanishes for some even spin structure . Since the parity of the number of independent holomorphic spinors is the same as the parity of , this means that there are at
least 2 independent holomorphic spinors of spin structure . By the RiemannRoch theorem, the number of zeroes of a holomorphic spinor is always (h 1). In genus h = 2,
a holomorphic spinor has then exactly one zero, and the ratio of two linearly independent
holomorphic spinors would be a meromorphic function with exactly one zero and one pole.
Such a function provides a one-to-one correspondence between the given Riemann surface
and the sphere, contradicting our initial assumption that h = 2. Similarly, when h = 3, a
holomorphic spinor has 2 zeroes, and the ratio of two linearly independent holomorphic
spinors is a meromorphic function with two zeroes and two poles. Such a function provides a two-to-one correspondencewith the sphere, and thus the Riemann surface must be
hyperelliptic. Conversely, if s 2 = 8i=1 (x ui ) is a hyperelliptic surface of genus 3, then
s 1/2 (dx)1/2 and xs 1/2 (dx)1/2 define two holomorphic spinors associated with an even
spin structure. Thus 18 ( (3) ) vanishes at such surfaces (in fact, to second order), and the
proof of the claim is complete.
Since the form 18 ( (3) ) vanishes of second order, we can follow [11] and obtain a
holomorphic character 9 ( (3) ) by taking its square root

2

9 (3) = 18 (3) .
(3.6)
92
In genus h = 2 and h = 3, the moduli space Mh and the Siegel upper half space have
the same dimension, which is 3 and 6, respectively. An integral over Mh can be identified
with an integral over a fundamental domain of the modular group Sp(2h, Z) in the Siegel
upper half space. On this space, we can introduce the following holomorphic (3h 3, 0)
forms3
1
10 ( (2) )
1
9 ( (3) )
(2)
for genus h = 2,
dI J ,
1I J 2
(3)
for genus h = 3.
dI J ,
(3.7)
1I J 3
Both measures are holomorphic on the Siegel upper half space. This is obvious when h = 2.

When h = 3, this is due to [11], who showed that the form I J dI(3)
J also vanishes
along the variety of hyperelliptic surfaces, so that the zeroes in the denominator 9 ( (3) )
are cancelled by the measure factor.
It follows from Igusas classification theorem for genus 2 modular forms that the
bosonic string measure is actually given in genus h = 2 by [10,11]

dB (2) =
c2
10 ( (2) )
(2)
dI J ,
(3.8)
1I J 2
where c2 is an overall constant. This constant was in fact evaluated in [4, Section 7.1], and
was found to be c2 = 12 . There is no such classification theorem in genus 3 or higher,
but cogent arguments have been proposed for the similar relation in genus 3 to hold [11]

dB (3) =
c3
9 ( (3) )
(3)
dI J
(3.9)
1I J 3
with c3 another overall constant. As part of our program for determining the genus 3 superstring measure, we shall present further evidence for this relation below.
3.2. Degeneration of 91 ( (3) )
I J
(3)
dI J
The superstring chiral measure will be identified by its density with respect to the basic

(3)
measure 9 ( (3) ) I J dI J . In order to reformulate the degeneration constraints (2.13)
for the superstring measure in terms of degeneration constraints for its density, we need

(3)
the precise degeneration limit of the measure 9 ( (3) ) I J dI J . This is given in the
following theorem:
3 The ordering of the forms d (h) in these measures is a matter of convention. We shall ignore the resulting
IJ
signs and sometimes denote the resulting volume form just by d h(h+1)/2 (h) .
Theorem 1. In the degeneration limit given by Section 2.1, t 0, we have

(2)

1
1
dt
d (1)
I J I J
(3)
d
=
2 dp2
IJ
6
(3)
(2)
(1)
24
(2) 10 ( )
t
9 ( )
( )
1I J 3

1
.
+O
t
93
(3.10)
Proof. We consider the parametrization of surfaces t degenerating into two surfaces

(1) and (2) described in Section 2.1. As indicated there, we choose canonical homol(i)
(i)
ogy bases (AIi , BIi ), so that the union of these cycles constitutes a canonical homology
basis for . Let (It 1 , It 2 ) be the basis of holomorphic Abelian differentials on t dual to
the (AI1 , AI2 ) cycles. Then these holomorphic differentials have the following asymptotic
behavior as t 0 [15]

(1)
I1 (z) + 4t I1 (p1 )p1 (z) + O(t 2 ), when z (1) ,
t
I1 (z) =
(2)
t
2
when z (2) ,
4 I1 (p1 )p2 (z) + O(t ),

(1)
t
2
when z (1) ,
t
4 I2 (p2 )p1 (z) + O(t ),
I2 (z) =
(3.11)
(2)
I2 (z) + 4t I2 (p2 )p2 (z) + O(t 2 ), when z (2) .
(i)
Here, pa refers to the meromorphic differential on surface (i) , i = 1, 2, with a double

pole at pi , while Ii refers to a basis of holomorphic differentials on surface (i) .
The components of the period matrix behave as follows [15]:
i
tI1 (p1 )J1 (p1 ),
2
i
(2)
tI2 (p2 )J2 (p2 ),
= I 2 J2 +
2
(1)
i
tI1 (p1 )J2 (p2 ),
2
i
tI2 (p2 )J1 (p1 ),
=
2
It1 J1 = I1 J1 +
It1 J2 =
It2 J2
It2 J1
(3.12)
where (i) refers to the period matrix on the surface (i) .

Henceforth, we consider the case where h1 = 1 and h2 = 2. It is convenient to set
(1) = , (2) = , and use the following notations:

11 12 1
1 = i
(3)
2 t1 (p2 )0 (p1 ),
= 12 22 2 ,
(3.13)
i
2 = 2 t2 (p2 )0 (p1 ).
1
2 3
Here I (p2 ) denote the genus 2 holomorphic differentials and 0 denotes the genus 1
holomorphic differential which is just the constant 1 in the usual parametrization of the
(1)
(1)
torus of modulus as C/Z + Z, since the homology basis (A1 , B1 ) has been fixed.
The -constants at genus 3 for even spin structures behave differently in the separating
limit depending on whether the spin structures on the genus 2 and genus 1 components are
both even or both odd. We have the following limits:

0, (3) = [](0, )[](0, ) + O(t),
94

t

0, (3) = 0 (p1 )1 (0, )h (p2 )2 + O t 2 .
0
4
(3.14)
Here (respectively, ) denote an even (respectively, odd) genus 2 spin structure, while
denotes an even genus 1 spin structure and 0 denotes the unique genus 1 odd spin
structure. Furthermore, we use the familiar notation
h (z)2 I (z) I [](0, ).
(3.15)
This square is defined for any surface, while its square root, h in single-valued only on a
surface with spin structure .
(a) The limit of 18
We are now in a position to study the limit of the modular form 18 and its square
root 9 . In genus 3, there are 36 even spin structures, of which 30 separate into two even
spin structures in genus 1 and 2, and 6 separate into two odd spin structures in genus 1
and 2. In the first group of 30, the spin structures obtained after degeneration run over all
10 genus 2 even spin structures and over all 3 genus 1 even spin structures. Similarly, in
the second group of 6, the spin structures obtained after degeneration run over all 6 genus 2
even spin structures. Thus we obtain

(3)
t
2

h (p2 ) 0 (p1 )1 (0, ) + O t 7 .
=
[](0, )[](0, )
18
4
,
(3.16)
In view of the well-known genus 1 identities,

1 (0, ) = 2( )3 ,
(3.17)
[](0, ) = 2( )3 ,
and the definition of 10 (), this can be rewritten as
6
(3)

6
3/2
3 10 t
18
= 10 ()
2( )
0 (p1 )6 2( )3
h (p2 )2
4

+ O t7

= 24 6 t 6 0 (p1 )6 10 ()3/2 ( )48
h (p2 )2 + O t 7 .
(3.18)
Taking the square root, we find

h (p2 ) + O t 4 .
9 (3) = 4 3 t 3 0 (p1 )3 10 ()3/4 ( )24
(3.19)
Notice that, while each h may not be single-valued on a surface with given spin structure
(or without specified spin structures), the product over all is single-valued on any surface.
(3)
(b) The limit of the volume factor d 6 I J

We turn next to the limit of the measure d 6 I(3)
J . In the above notation, we have
(3)
d 6 I J = d 3 d d1 d2 .
(3.20)
95
We now evaluate d1 d2 , using the definition of its ingredients,

2
t dt dp2 0 (p1 )2 1 (p2 )2 (p2 ) 2 (p2 )1 (p2 ) .
(3.21)
4
The combination in parentheses is a holomorphic 3-form in p2 . To evaluate it, we turn to
the hyperelliptic representation of Riemann surfaces of genus 2. Let the surface (2) be
given by
d1 d2 =
s2 =
6

(x ui ).
(3.22)
i=1
Then zJ 1 dz/s(z), J = 1, 2, is a basis of holomorphic differential forms. Let I J be the

(2)
change of bases matrix from this basis to the basis I2 (z) (which we abbreviate to I (z)
for the rest of the proof of Theorem 1),
2iI (z) =
I J
zJ 1 dz
.
s(z)
(3.23)
Hence, we have
1 (p2 )2 (p2 ) 2 (p2 )1 (p2 ) =
1
(dp2 )3
(det )
.
2
4
s(p2 )2
(3.24)
Thus the holomorphic

3-form manifestly has 6 simple zeros precisely at the branch points,

exactly as h (z). Thus, the p2 -dependence of these two forms is the same.
(c) Determining the constant of proportionality
We need to determine the constant of proportionality, which is moduli dependent, and
requires several precise coefficients of proportionality between the -function and the hyperelliptic representation of holomorphic spinors [4]. In the hyperelliptic representation,
each of the 6 odd spin structures i corresponds to a branch point ui , and the one-form
h2i (z) is proportional to the one-form (x ui ) dz/s(z). Set
h2i (z) = Ni (x ui )
dz
,
s(z)
where Ni is a moduli dependent constant. Then we have

1

(dp2 )3
=
h (p2 ).
2
1/2
s(p2 )
N
(3.25)
(3.26)
Combining all, we obtain

d1 d2 =
det
1
t dt dp2 0 (p1 )2 1/2
h (p2 ).
16
i Ni
Next, we have the following two identities:

(det )4 []8 = (ai aj )2 (bi bj )2 ,
i<j
(3.27)
96

24
12
(det ) []
2
10

Ni4
(ai aj )2 (bi bj )2 .
(3.28)
i<j
Here is an even spin structure. In the hyperelliptic representation, it corresponds to a

partition of the 6 branch points into two disjoint sets {a1 , a2 , a3 } and {b1 , b2 , b3 } of three
branch points each. The first identity is a classic Thomae formula [16, vol. II, Section 8]. To
establish the second identity, we make use of the following bilinear -constants, introduced
in [4, Eq. (2.38)]
Mi j = 1 [i ](0, )2 [j ](0, ) 2 [i ](0, )1 [j ](0, ).
(3.29)
Solving for I [i ](0, ) from (3.25) and using the formula (3.23) for I (z), we find
(det )Mi j = 4 2 Ni Nj (ui uj ).
(3.30)
Taking the products gives

(det )4 M212 M223 M231 M245 M256 M264 =
6
N4i
i=1
(ai aj )2 (bi bj )2 .
(3.31)
i<j
However, the M2i j have been determined completely explicitly in terms of -constants
in [4, Eq. (4.9)],

[i + j + k ]2 ,
Mi j = 4 []2
(3.32)
k=i,j
so that
M212 M223 M231 = M245 M256 M264 = 12 []4 10 ().
(3.33)
Substituting this into (3.31) gives the second identity in (3.28), and (3.28) is now established. Taking the ratio of the two identities in (3.28), we find
3 10 ()1/4 det =
6
N1/2
.
i
(3.34)
i=1
Comparing with (3.27), we obtain in this manner the following asymptotics for d1 d2
d1 d2 =

1
2 1
t
dt
dp
(p
)
h (p2 ) + O t 2 .
2
0
1
3
1/4
16
(3.35)
10
The theorem is now an immediate consequence of (3.19), (3.20), and (3.35).
3.3. Degeneration limit for the 3-loop bosonic string

We recall that the genus 3 bosonic string measure must satisfy the degeneration con1 3
straint (2.16). Since the genus 2 bosonic string measure is given by c2 10
d , Theorem 1 provides further evidence that the genus 3 bosonic string measure is given by
97
c3 91 ( (3) ) d 6 (3) . In fact, Theorem 1 also dictates what the coefficient of proportionality between the genera 2 and 3 must be
c3 =
c2
1
=
.
(2)6 26 18
(3.36)
As another check, we consider the separating degeneration limit of the tachyon amplitude, which is given by the following integral, where E(z, w) is the prime form,
2

dt

E(zi , zj )2ki kj .
(3.37)
t2
i<j
The behavior of the prime form when zi (2) and zj (1) is given by
E(zi , zj ) t 1/2 E(zi , p2 )E(p1 , zj ).
(3.38)
If the sum of the momenta on (2) is k,

k=
ki =
kj ,
i,zi (2)
(3.39)
j,zj (1)
then we have the following t-dependence,

2
dt p2
|t| 1 ,
t2
p2 2
(3.40)
which is the expected tachyon pole, with the correct value of the mass squared.
4. The genus 3 superstring measure as a degeneration problem

Using the formula for the genus 2 superstring measure found in [1] and the degeneration
formulas of Theorem 1, we can formulate now more concretely the constraints on the genus
3 superstring measure, in the degeneration limit where the worldsheet separates into a
torus (1) and a surface (2) of genus 2. Let be an even genus 3 spin structure. If, in this
degeneration, factorizes into two odd spin structures, the leading contribution of order
t 2 to d[]( (3) ) vanishes, and we need not consider this case. Henceforth, we assume
that factorizes into two even spin structures, and denote by 1 and by 2 the even
spin structures, respectively, on the torus and on the genus 2 surface (2) .
Let the genus 3 superstring measure be expressed under the form (1.1), for some density
6 []( (3) ) yet to be determined. Recall that in genus h = 2, the superstring measure
d[]( (2) ) was shown to be given by [1,4]

[]4 (0, (2) )6 []( (2) )
d[] (2) =
16 6 10 ( (2) )
dI(2)
J.
(4.1)
1I J 2
The main expression 6 []( (2) ) is given in [1, Eq. (7.1)]. We shall discuss it further
in the next section. The degeneration constraint (2.14), Theorem 1, and the degeneration
formulas (3.14) for -constants imply then that 6 []( (3) ) must satisfy the following
98
limit

12

lim 6 [] (3) = (1)
6 [] (2) .
t0
(4.2)
This is the condition (iii) formulated in the introduction.

We discuss next the issue of modular invariance for d[]( (3) ). The full integrand in
the amplitude (2.12) must be invariant under Sp(6, Z). Under the modular transformations
(3.1), we have

2
det Im (3) = det C (3) + D det Im (3) ,

4
(3)
(3)
(4.3)
d I J = det C (3) + D
dI J .
I J
I J
At first sight, in genus 3, we have difficulties due to the fact that the expression 9 ( (3) )
is defined only through its square, 18 ( (3) ). However, the ambiguity in taking square
roots here should not be relevant in string theory: for the superstring, 9 and its conjugate
appear in each chiral sector. This is also the case for the heterotic string, since we have
seen that 9 appears in the chiral measure for the bosonic string in the critical dimension,
and this is unaffected by compactification. Thus the sign ambiguity in 9 can be ignored.
In analogy with the genus 2 case, we shall impose then the modular transformation law (ii)
described in the introduction on the unknown term 6 []( (3) ).4 This condition implies
that the superstring chiral measure d[]( (3) ) transforms covariantly under modular
transformations without any phase factor

(3) = det C (3) + D 5 d[] (3) ,
d[]
(4.4)
so that a manifestly modular invariant GSO projection is given by

d[] (3) .
(4.5)
This completes our discussion of the three conditions (i)(iii) formulated in the introduction for the modular covariant form 6 []( (3) ).
5. Ansatzes for the superstring chiral measure

The goal of this section is to construct modular covariant forms in genus 3 satisfying
the constraints (ii) and (iii). The starting point is the expression 6 []( (2) ) in genus 2.
Our strategy is to find and analyze analogous expressions in genus 3. Although there are
several natural analogues, it will turn out that the degeneration condition (iii) is quite rigid,
and singles out a very small set of candidates.
4 A slightly less restrictive requirement is to allow in (ii) an additional phase (M) depending only on the
modular transformation M, but not on the spin structure . Such additional phases do not affect significantly our
subsequent construction of candidates for 6 []( (3) ).
99
5.1. The form 6 []( (2) ) in genus 2

We begin by recalling the form 6 []( (2) ) in genus 2. It was derived directly from
the gauge-fixed genus 2 superstring measure, and its original expression was heavily dependent on the fact that the worldsheet had genus 2 (see [1, Eq. (7.1)]). More recently,
two different expressions were found for 6 []( (2) ) which do admit generalizations to
higher genus [14]. To describe them, recall that a triplet {1 , 2 , 3 } is said to be asyzygous
if e(1 , 2 , 3 ) = 1 (respectively, syzygous when +1), using the usual definitions of the
signatures on pairs and triples of spin structures,

i |j exp 4i i j i j ,
e(1 , 2 , 3 ) 1 |2 2 |3 3 |1 .
(5.1)
More generally, we define as in [14] an N -tuple of spin structures of be totally asyzygous,

if any triplet of distinct spin structures in the N -tuple is asyzygous
{1 , . . . , N } totally asyzygous
{i , j , k } asyzygous, for all i, j, k pairwise distinct.
(5.2)
The notion of totally asyzygous N -tuple is modular invariant, since the cyclic product in
(5.1) of relative signatures for a triple of spin structures is.
Returning now to 6 []( (2) ), the first alternate expression involves 4th powers of
-constants (as does the original expression in [1, Eq. (7.1)]) and is given by

1
6 [] (2) =
2
3
4

|i [i ] 0, (2) .
(5.3)
{,1 ,2 ,3 } i=1
tot.asyz.
The notation indicates that, for given spin structure , the summation runs over all triples
{1 , 2 , 3 } such that {, 1 , 2 , 3 } forms a totally asyzygous quartet. The second alternate
expression for 6 []( (2) ) involves only squares of -constants, but it requires summation
over certain sextets of even spin structures. To identify which sextets, we define a sextet
{1 , . . . , 6 } of spin structures in genus 2 to be admissible if it can be decomposed into
three pairs
{1 , . . . , 6 } = {i1 , i2 } {i3 , i4 } {i5 , i6 },
(5.4)
with the union of any two pairs of the decomposition forming a totally asyzygous quartet.
For a given spin structure , we define the sextet {i } to be -admissible if it is admissible and it does not contain . With this definition, the second alternative expression for
6 []( (2) ) is given by
1

6 [] (2) =
2

{i } -adm.
6
2

; {i }
[i ] 0, (2) .
(5.5)
i=1
Here the signs (; {i }) are uniquely related by modular transformations

6

M; {Mi }
2 (i , M) = 4 (, M) ; {i } ,
i=1
M Sp(4, Z),
(5.6)
100
where 4 (, M) is the same factor occurring in the transformation law for 4 []. An explicit expression for the signs was given in [14].
5.2. Ansatzes in genus 3
The preceding formulas for 6 []() in genus 2 suggest several natural extensions to
genus 3. We discuss them below. The main issue is whether they can satisfy the desired
conditions (i), (ii), and (iii) listed in the introduction, which are required for any viable
ansatz for the genus 3 superstring chiral measure.
5.2.1. Ansatz in terms of asyzygous quartets of spin structures
The first alternate expression (5.3) for 6 []( (2) ) clearly makes sense for arbitrary genus, and in particular for genus 3. Thus we are dealing here with an ansatz for
6 []( (3) ) involving summations over spin structures {1 , 2 , 3 } which together with
the given spin structure , form a totally asyzygous quartet. The modular-covariant form
which it defines has been studied in [14], where it was denoted by 6# []( (3) ). However,
its degeneration limits, as determined in [14, Theorem 5], do not satisfy the degeneration
constraint (4.2) for the genus 3 superstring measure. Thus this ansatz in terms of asyzygous
quartets of spin structures must be dropped from contention.
5.2.2. Ansatzes in terms of admissible sextets of spin structures
We turn then to several ansatzes which can be viewed as generalizations to genus 3
of the expression (5.5) for 6 []() in terms of -admissible sextets. First, in complete
analogy with the genus case, we define a sextet {1 , . . . , 6 } to be admissible if it can be
decomposed as
{i1 , i2 } {j1 , j2 } {k1 , k2 }
(5.7)
with any two pairs constituting a totally asyzygous quartet. Given a spin structure , a
sextet is said to be -admissible if it is admissible, and it does not contain .
Despite the similarity in the definitions, there is in practice a fundamental difference
between admissible sextets of spin structures in genus 2 and in genus 3: if
s = {i1 , i2 } {j1 , j2 } {k1 , k2 }
(5.8)
is an admissible sextet in genus 2, then the triplets {i , j , k }, with , , = 1, 2,

are automatically syzygous. Furthermore, if the sextet is -admissible, then the following
triplet signatures are automatically determined:
e(, i1 , i2 ) = e(, j1 , j2 ) = e(, k1 , k2 ) = +1.
(5.9)
This may easily be inferred by inspection of Table 4 in [14].

This is no longer true for genus 3: in an admissible sextet {1 , . . . , 6 }, the triplets
{i , j , k } need not all be syzygous (or all asyzygous). Thus the admissible sextets
in genus 3 fall into 2 categories:
(1) All triplets {i , j , k } are asyzygous, so that the whole sextet {1 , . . . , 6 } is
totally asyzygous;
101
(2) At least one triplet {i , j , k } is syzygous. In this case, the relations (5.9) also do
not follow from -admissibility. In particular, one has a classification depending on
the following signs:
1 = e(, j1 , k1 ),
2 = e(, k1 , i1 ),
3 = e(, i1 , j1 ).
(5.10)
The four resulting cases, namely, [+ + +], [+ + ], [+ ] and [ ], are nonempty and the modular group acts within each case, though not necessarily transitively
so. For convenience, we refer to all these cases as cases of partially asyzygous sextets.
Ansatzes in terms of totally asyzygous sextets
We shall examine two ansatzes, in terms of totally asyzygous sextets with sign assignments.
(A)
6
2

; {i }
[i ] 0, (3) ,

6 [] (3)
{i } tot. asyz.
{
/ i}
(B)

6 [] (3)
i=1

; {i }, i
{i },{i } tot. asyz.

{
/ i },{i }
6
2
2

[i ] 0, (3) i 0, (3)
1/2
.
(5.11)
i=1
A key issue in these ansatzes is whether sign assignments exist which are consistent with
modular transformations. The first Ansatz (A) is simpler, and it will turn out that it does
satisfy the degeneration constraint (4.2) if a consistent assignment existed. However, this
turns out not to be the case, which is why the second Ansatz (B) is needed. This second
Ansatz (B) turns out to be the only viable candidate for 6 []( (3) ) among all the ones
examined in the present paper. Its full treatment requires the rest of the paper. We postpone
it then to the Section 6, and complete now the discussion of the remaining ansatzes, which
involve partially asyzygous sextets.
Ansatz in terms of partially asyzygous sextets
In these remaining cases, the natural ansatzes would be

6 [] (3)

[1 ,2 ,3 ]
6
2

; {i }
[i ] 0, (3) ,
(5.12)
i=1
where the summation would be over all -admissible sextets {1 , . . . , 6 } with some
fixed sign assignment [1 , 2 , 3 ].
The first task is to examine whether consistent phase assignments (; {i }) exist. One
does this orbit by orbit under the modular group which leaves invariant, and uses the
102
usual modular sign factors in the transformations of 2 . The results are as follows, where
we have numbered the genus 3 even spin structures as in Appendix C of [14],
1. For the cases [ ], [+ ] and [+ + ], all orbits produce inconsistent sign
assignments and are ruled out;
2. For the case [+ + +], one orbit with 1680 sextets (generated by sextet {2, 4, 5, 6, 33,
35}) and one orbit with 3360 sextets (generated by sextet {5, 7, 12, 13, 22, 30}) both
generate consistent sign assignments;
3. For the case [+ + +], there is one remaining orbit (generated by sextet {2, 4, 5, 9, 27,
32}) which produces an inconsistent sign assignment.
A simple example showing the non-existence of consistent phases for one of these orbits
of -admissible, partially asyzygous sextets is given in Section 6.2.3 below.
The actual sums in both cases of (2) above are non-vanishing. In the limit where the
surface degenerates to a genus 2 times genus 1 surface, both sums converge to the same
limit as the form 6# []( (3) ) of [14], which is inconsistent with the requirement (iii) in
the introduction. Thus, even though the sign assignments are consistent, the limits are not
and the cases are all ruled out.
Although this analysis rules out a construction of 6 []( (3) ) in terms of partially asyzygous -admissible sextets, it is in principle still possible that an ansatz for
6 []( (3) )2 can be obtained in terms of pairs of partially asyzygous -admissible sextets, just as we outlined in the preceding case (B) of totally asyzygous -admissible sextets.
However, there does not appear to be any clear way of recapturing 6 []( (2) )2 from the
degeneration limits of sums over pairs of partially asyzygous sextets.
5.2.3. Ansatz in terms of dozens of spin structures
Since the previous ansatzes have not produced viable candidates for 6 []( (3) ) itself
as a polynomial in 2 , we may ask whether polynomials in could work,
12

6 [] (3) =
; {i }1i12
[i ] 0, (3) ,
{i }
(5.13)
i=1
where the summation runs over a suitable set of dozens {1 , . . . , 12 } of spin structures.
Here the expression for 6 []() in genus 2 provides little guidance for choosing this set.
There are conceivably many possibilities. But, given the good degeneration limits of totally
asyzygous sextets, it is natural to consider products of pairs of totally asyzygous sextets,

6 [] (3) =
(, s1 , s2 )
{s1 ,s2 }Qpq
12

[i ] 0, (3) ,
(5.14)
i=1
where the sum is over pairs of totally asyzygous sextets,

s1 = {1 , 2 , 3 , 4 , 5 , 6 },
s2 = {7 , 8 , 9 , 10 , 11 , 12 },
(5.15)
103
and Qpq denote the different orbits of -admissible pairs of totally asyzygous sextets
under the modular subgroup leaving invariant. The orbits Qpq are described in detail in
Section 6.3.2 below.
Clearly, this construction can make sense only for orbits Qpq for which the phases
(, s1 , s2 )2 can be consistently defined. But this problem was already solved (by computer) in the treatment of the Ansatz (B) in terms of pairs of totally asyzygous sextets (see
Section 6.3.2 and subsequent discussions). It was found that consistent phases exist for the
orbits Q01 , Q02 , Q13 , Q20 Q21 , Q22 , and Q23 but not for the orbits Q11 , Q12 , and Q3 .
A computer calculation shows, however, that in none of these orbits the sign (, s1 , s2 )
can actually be consistently defined. Thus, this particular ansatz for 6 []( (3) ) is also
ruled out.
6. The ansatzes in terms of totally asyzygous sextets

To determine the degeneration behavior of the genus 3 candidates (A) and (B), we need
to determine the degeneration behavior of genus 3 totally asyzygous sextets of even spin
structures.
6.1. Degenerations of totally asyzygous sextets
The basic fact is the following:
Lemma 1. Let {i }1i6 be a sextet of genus 3 even spin structures. Assume that it is totally asyzygous and that each i degenerates into even spin structures in genera 2 and 1.
Let 1 , . . . , 6 be the 6 genus 2 even spin structures which arise in this manner. Then the
sextet {1 , . . . , 6 } is an admissible sextet of genus 2 even spin structures in the sense defined previously, that is, it can be divided into three pairs, the union of any two defines a
totally asyzygous quartet. Furthermore, we have
6

i=1
6
2
12
2

[i ] 0, (3) 24 (1)
[i ] 0, (2) .
(6.1)
i=1
Proof. We recall that there exist no totally asyzygous quintets at genus 2 (and thus no
genus 2 totally asyzygous sextuplets, etc.). This can be seen by direct inspection of the tables of asyzygies in genus 2 provided in [14]. Let 1 , . . . , 6 be the genus 1 spin structures
arising from the degeneration of 1 , . . . , 6 . By assumption, they are even. We examine
in turn all possible arrangements for 1 , . . . , 6 :
Assume that 1 , . . . , 6 take at most 2 distinct values amongst the 3 possible even
spin structures at genus 1. Then, it follows that e(i , j , k ) = +1 for any triplet of
s arising in the sextet. For the genus 3 sextet to be totally asyzygous, the genus 2
sextuplet 1 , . . . , 6 must be totally asyzygous, but this is impossible.
Assume that five of the six 1 , . . . , 6 (say 1 , . . . , 5 for definiteness) take at
most 2 distinct values amongst the 3 possible even spin structures at genus 1. Then
104
e(i , j , k ) = +1 for 1 i, j, k 5. For the genus 3 sextet to be totally asyzygous,

the genus 2 quintet 1 , . . . , 5 must be totally asyzygous, but this is impossible.
The only remaining possibility is that amongst the six 1 , . . . , 6 , each of the 3 distinct
genus 1 even spin structures (which we denote 2 , 3 , 4 by slight abuse of notation)
occurs precisely twice.
Thus, up to permutations of the s, we have

1
3
5
1 =
,
3 =
,
5 =
,
2
3
4

2
4
6
2 =
,
4 =
,
6 =
.
2
3
4
(6.2)
It is clear that the quartets {1 , 2 , 3 , 4 }, {1 , 2 , 5 , 6 }, {3 , 4 , 5 , 6 }, are totally asyzygous, and that they are the only totally asyzygous quartets within {1 , . . . , 6 }. This proves
the first part of Lemma 1. The second part follows immediately from the degeneration
formulas for -constants and from the identity (3.17). 2
6.2. Orbits of sextets
Since the genus 2 expression 6 []( (2) ) is built from genus 2 admissible sextets,
Lemma 1 shows that asyzygous sextets have the potential to produce a form 6 []( (3) )
tending to 6 []( (2) ) in the degeneration limit. Fix an even spin structure . In analogy
with the genus 2 case, we define a -admissible sextet of even spin structures to be a
totally asyzygous sextet {i } not containing . We can restrict then the sextets entering
the candidate for 6 []( (3) ) to the -admissible ones. This justifies the form given in
(5.11) for the Ansatzes (A) and (B).
We need to consider the degenerations of -admissible sextets {i }. We can assume
that
=
(6.3)
with both lower genus spin structures i and even, since otherwise will not contribute
to the leading asymptotics. Let {i } be the sextet of genus 2 spin structures obtained by
factoring {i }. We can assume that they are all even, since otherwise {i } will again not
contribute to the leading asymptotics. Now Lemma 1 guarantees that the sextet {i } is admissible in the genus 2 sense. However, the condition
/ {i } does not guarantee that
/ {i }, i.e., the -admissibility of the genus 3 sextet {i } does not guarantee the admissibility of the genus 2 sextet {i }. Thus we have to analyze the contributions of genus
2 admissible sextets which are not -admissible. We also have to determine the exact multiplicities with which -admissible and -not admissible sextets occur in the degeneration
of an Sp(6, Z) orbit of -admissible sextets in genus 3. The first issue is addressed by
Lemma 2 below. The second issue will be addressed by a computer listing of all possibilities. The results will be described in subsequent sections.
105
6.2.1. Orbits of admissible sextets in genus 2

The list of admissible sextets in genus 2 is provided in [14, Table 4]. By a simple inspection of that list and the actions of modular transformations in the same table, we find
that
There are no totally asyzygous quintets, and a fortiori, no totally asyzygous sextets in
genus 2;
In genus 2, there are 15 admissible sextets. The group Sp(4, Z) acts transitively on the
set of admissible sextets;
Given a genus 2 even spin structure , there are always exactly 6 sextets which are
admissible, and 9 which are not. We denote these sets of sextets by s[] and s c [],
respectively:

/ {i } ,
s[] = {i } admissible sextet;

s c [] = {i } admissible sextet; {i } ;
(6.4)
Let Sp[](4, Z) be the subgroup of Sp(4, Z) fixing . Then Sp[](4, Z) acts transitively
on both s[] and s c []. In particular, if the phases (; {i }) satisfy the transformation
(5.6), then all the phases in each orbit s[] or s c [] are uniquely determined by the
phase of any single element inside s[] and s c [].
Lemma 2. Let be a fixed genus 2 even spin structure. Assume that the phases (; {i })
satisfy the condition (5.6) for all M Sp[](4, Z). Then we have

6

; {i }
[i ]2 = 26 [] (2) ,
{i }s[]
(6.5)
i=1
6

; {i }
[i ]2 = 0.
{i }s c []
(6.6)
i=1
The sign in the first identity is a consequence of the fact that the phases (; {i }) in
each orbit s[] or s c [] are determined only up to a global sign.
Proof. The first identity in (6.5) is just a reformulation of (5.5), and was proved in [14].
To establish
the second identity, we go to the hyperelliptic representation.
Let s 2 = 6i=1 (x pi ) be a hyperelliptic representation for the surface (2) .5 As before, we identify the spin structure with a partition of the 6 branch points into two sets of
3 branch points each, say {a1 , a2 , a3 } {b1 , b2 , b3 }. The Thomae formula (for genus 2)
takes the following form:

xpi pj = pi pj .
[]2 = Cxa1 a2 xa2 a3 xa3 a1 xb1 b2 xb2 b3 xb3 b1 ,
(6.7)
5 The branch points p here should not be confused with the punctures p and p in the degeneration coni
1
2
struction of Section 2. The notation pi for the branch points is in accord with [14], which is used heavily in the
proof of Lemma 2.
106
Here, 4 = 1, and C is -independent. Actually, we need the explicit correspondence only

for the sextets themselves. Given the normalization of a single sextet, the correspondences
for all others may be derived using the action of modular transformations on both sides. We
fix the expression for one sextet, say (125690), to be C 6 , and determine the hyperelliptic
expressions for the others by modular transformations,
t1 +(125690) = +(p1 p6 )(p2 p4 )(p3 p5 )C 6 V (pi ),
t2 +(137890) = (p1 p3 )(p2 p5 )(p4 p6 )C 6 V (pi ),
t3 +(145678) = (p1 p4 )(p2 p3 )(p5 p6 )C 6 V (pi ),
t4 +(124580) = (p1 p4 )(p2 p6 )(p3 p5 )C 6 V (pi ),
t5 +(134670) = +(p1 p5 )(p2 p3 )(p4 p6 )C 6 V (pi ),
t6 +(123689) = (p1 p6 )(p2 p5 )(p3 p4 )C 6 V (pi ),
t7 (134589) = +(p1 p4 )(p2 p5 )(p3 p6 )C 6 V (pi ),
t8 (124679) = (p1 p6 )(p2 p3 )(p4 p5 )C 6 V (pi ),
t9 (123570) = +(p1 p2 )(p4 p6 )(p3 p5 )C 6 V (pi ),
t10 +(235678) = +(p1 p2 )(p3 p4 )(p5 p6 )C 6 V (pi ),
t11 +(247890) = (p1 p3 )(p2 p6 )(p4 p5 )C 6 V (pi ),
t12 (234579) = +(p1 p2 )(p3 p6 )(p4 p5 )C 6 V (pi ),
t13 +(234680) = (p1 p5 )(p2 p6 )(p3 p4 )C 6 V (pi ),
t14 +(345690) = +(p1 p5 )(p2 p4 )(p3 p6 )C 6 V (pi ),
t15 +(567890) = (p1 p3 )(p2 p4 )(p5 p6 )C 6 V (pi ).
The omnipresent factor V is the Vandermonde polynomial

V (pi )
xij2 =
(pi pj ).
1i<j 6
(6.8)
(6.9)
1i<j 6
Under a permutation of the branch points, V (pi ) is multiplied by the signature of this
permutation. The following modular transformations were used to establish these signs:
(t1 ) = +t2 ,
M3 (t4 ) = +t8 ,
S(t14 ) = +t11 ,
T (t1 ) = +t3 ,
S(t8 ) = +t7 ,
M1 (t8 ) = t12 ,
S(t3 ) = +t6 ,
T (t7 ) = +t9 ,
M3 (t12 ) = +t13 ,
T (t6 ) = +t5 ,
M1 (t3 ) = t10 ,
S(t13 ) = +t15 ,
(t5 ) = +t4 ,
T (t10 ) = +t14 .
(6.10)
Taking into account the behavior of V under permutations, modular invariance determines
the relative signs in the sums over s[] and s c []. Working this out for one of the spin
structures, say = 1 gives the following explicit formulas:

{i }s[1 ]
6

1 ; {i }
[i ]2 = 2t12 + 2t13 + 2t15 = 2t10 2t11 2t14 = 26 [1 ],
i=1
6
9

1 ; {i }
[i ]2 =
ti = 0.
{i }s c [1 ]
i=1
107
(6.11)
i=1
The modular covariance properties then yield these results for all spin structures and thus
completes the proof of the theorem. 2
6.2.2. Orbits of admissible sextets in genus 3 (totally asyzygous sextets)
We list here a number of results on the modular transformations of asyzygous multiplets
{i } in genus 3, all of which have been proven by computer calculations.
The sets of all asyzygous quartets, quintets and sextets transform transitively under the
full modular group acting on characteristics;
There are 5040 totally asyzygous quartets, 2016 totally asyzygous quintets, 336 totally
asyzygous sextets, and no totally asyzygous septets;
The set of all asyzygous sextets that do not contain a given spin structure transforms
transitively under the modular subgroup Sp[](6, Z) leaving invariant. In analogy
with the genus 2 case, we denote by S[] the set of asyzygous sextets not containing
a spin structure . For any , S[] consists of 280 elements;
Upon factorization, each sextet {i } of genus 3 spin structures produces a sextet {i }
of genus 2 spin structures. Consider the set of the 336 sextets {i } of genus 2 spin
structures which are obtained from factorization from the set of all 336 asyzygous
sextets in genus 3. Then the set of such {i } can be divided into 246 sextets which
contain at least some odd spin structure, together with 6 copies of all 15 genus 2
admissible sextets;
Similarly, let factorize into a genus 1 and a genus 2 spin structure as in (6.3),
and consider the set of all genus 2 sextets {i } arising from factorization of the 280
-admissible genus 3 sextets in S[]. Then the set of such {i } can be divided into
208 sextets which contain at least some odd spin structures, together with 6 copies of
s[] and 4 copies of s c [].
We can now consider the first ansatz in (5.11) for 6 []( (3) ), where the summation
is over the set S[] of -admissible sextets. For 6 []( (3) ) to transform as in (ii), we
impose the analogous condition to (5.6) in genus 3
6

2 (i , M) = 4 (, M) ; {i } ,
M; {Mi }
M Sp(6, Z).
(6.12)
i=1
Restricted to M Sp[](6, Z), this implies that all the phases (M; {Mi }) in the first
ansatz uniquely determine one another. Assuming the existence of such a consistent assignment of phases, the expression in the first ansatz is then uniquely determined up to a
global sign. The Sp[](6, Z) consistency of phases implies the Sp[](4, Z) consistency
of phases. Thus Lemmas 1 and 2 apply. Together with the numerology for the degeneration
of the orbit S[] found above, we obtain
lim
t0

{i }S[]
6

12

; {i }
[i ]2 0, (3) = 6 24 (1) 6 [] (2) . (6.13)
i=1
108
Here, we assume that all 6 copies of s[] obtained in factoring S[] lead to contributions
of the same sign. However, there is a more severe obstruction to the Ansatz (A):
There does not exist a phase assignment (; {i }) satisfying the condition (6.12)
and the sextets are totally asyzygous. This is in marked contrast with the genus 2 case,
where the phases (; {i }) satisfying (5.6) do exist. A counterexample in genus 3 is
obtained by considering the following 1 -admissible sextet,6
s1 = (2 , 8 , 14 , 16 , 25 , 30 ),
(6.14)
and the action of the composite modular transformation A1 B4 . From Table 6 of [14],
it is clear that A1 B4 leaves 1 , 3 , 4 , 6 invariant and maps 2 5 . Thus, the
-admissible sextet s1 , as a whole, is invariant under A1 B4 . The sign factor is also
easily computed, using
2 (i , A1 B4 ) = 2 (B4 i , A1 ) 2 (i , B4 ) = e4i(i )1 (i )2 ,
(6.15)
and we find
2 (i , A1 B4 ) = +1,
i = 2, 8, 14, 16, 25,
2 (30 , A1 B4 ) = 1.
(6.16)
But then the sextet contribution changes sign under a transformation that leaves the
sextet invariant, which means to no consistent sign can be defined.
6.2.3. Orbits of admissible sextets in genus 3 (partially asyzygous sextets)
A consistent phase assignment is also lacking in this case. A counterexample in genus
3 is obtained by considering the following 1 -admissible sextet,
s2 = (2 , 6 , 8 , 18 , 29 , 36 ).
(6.17)
The modular transformation A6 B6 A6 B6 leaves each of the spin structures in s2 , and thus
the entire sextet, invariant. The signs accompanying the transformation are easily computed, using
2 (i , A6 B6 A6 B6 ) = +1,
i = 2, 8, 18,
(i , A6 B6 A6 B6 ) = 1,
i = 6, 29, 36.
(6.18)
But then the sextet contribution changes sign under a transformation that leaves the sextet
invariant, which means to no consistent sign can be defined.
6.3. Orbits of pairs of sextets
In the preceding section, we have seen sums over -admissible sextets are not consistent
with the modular transformation (6.12). Thus we cannot construct 6 []( (3) ) directly
by the Ansatz (A). In this section, we shall show that certain sums over pairs of sextets
do admit consistent phase assignments, and that carefully chosen sums do lead to viable
candidates for 6 []( (3) )2 .
6 Throughout, we shall use the nomenclature for genus 3 spin structures and modular transformations given
in Appendix C of [14].
109
6.3.1. Orbits of pairs of admissible sextets in genus 2

Fix an external genus 2 even spin structure . Our first task is to identify the orbits of
pairs -admissible sextets under Sp[](4, Z). Clearly, for each integer p, the subset of pairs
{i }, {i } with p common spin structures is invariant under Sp[](4, Z). For -admissible
pairs of sextets, there is a finer partition which does give precisely all the orbits under
Sp[](4, Z):

Qp0,0 [] = {i }, i s[] s[]; # {i } i = p ,

Qp0,1 [] = {i }, i s[] s c []; # {i } i = p ,

Qp1,0 [] = {i }, i s c [] s[]; # {i } i = p ,

Qp1,1 [] = {i }, i s c [] s c []; # {i } i = p .
(6.19)
By inspecting the table of admissible sextets in genus 2, we find that only the values
a,b
p = 3, 4 and 6 produce non-empty sets Qa,b
p []. The sizes of the orbits Qp [] are given
0,0
0,0
0,0
0,1
1,0
by #Q3 [] = 12, #Q4 [] = 18, #Q6 [] = 6, #Q3 [] = #Q3 [] = 36, #Q40,1 [] =
#Q41,0 [] = 18, #Q31,1 [] = 36, #Q41,1 [] = 36, #Q61,1 [] = 9, which does add up to 152 =
225. In this counting, the pairs of sextets have been viewed as ordered pairs. For later
purposes, it is preferrable to count unordered pairs, in which case the sizes of the orbits
Qa,b
p [] become
Q30,0 = 6,
Q30,1 = 36,
Q31,1 = 18,
Q40,0 = 9,
Q40,1 = 18,
Q41,1 = 18,
Q60,0 = 6,
Q60,1 = 0,
Q61,1 = 9.
To each orbit
Qa,b
p [],
(6.20)
we can associate the following polynomial in -constants
6
6

2
pa,b ; {i }, i
[i ]2
i ,
({i },{i })Qa,b

p []
i=1
Fpa,b [] =
(6.21)
i=1
where the phases (; {i }, {i }) are required to satisfy

6
6

M; {Mi }, Mi
2 (i , M)
2 i , M = ; {i }, i .
i=1
(6.22)
i=1

Since Qa,b
p [] are orbits of Sp[](4, Z), the phases (; {i }, {i }) completely determine
a,b
each other within Qp []. We also find, by computer inspection, that a consistent assigna,b
ment of phases (; {i }, {i }) exist for each Qa,b
p []. Thus the expressions Fp [] exist,
and are uniquely determined by a single normalizing sign. We shall define this normalizing
sign below.
Remarkably, the expressions Fpa,b [] can be expressed very simply in terms of
6 []( (2) ) and two other polynomials in -constants, defined by

ti2 , s[1 ] = {10, 11, 12, 13, 14, 15},
F [1 ]
is[1 ]
110
F c [1 ]
ti2 ,
s c [1 ] = {1, 2, 3, 4, 5, 6, 7, 8, 9}.
(6.23)
is c [1 ]
Then we have
Lemma 3. Let the normalizing signs for Fpa,b [] be defined by Eq. (6.28). Then the expressions Fpa,b [] are given by
F30,0 [1 ] = 6 [1 ]2 F [1 ]/2,
F30,1 [1 ] = F c [1 ],
F31,1 [1 ] = F c [1 ]/2,
F40,0 [1 ] = 6 [1 ]2 ,
F40,1 [1 ] = F c [1 ],
F41,1 [1 ] = F c [1 ],
F60,0 [1 ] = F [1 ],
F61,1 [1 ] = +F c [1 ].
(6.24)
Proof. The following relations were established earlier:

6 [1 ] = t10 + t11 + t14 = t12 t13 t15 ,
(6.25)
0 = t1 + t 2 + t3 + t 4 + t5 + t 6 + t 7 + t8 + t 9 .
(6.26)
The relation (6.25) was established as a step in the proof of the alternative form (5.5)
of 6 []( (2) ) in [14]. The relation (6.26) is a reformulation of the second identity in
Lemma 2. Additional rearrangement formulas are as follows:
t1 = t2 + t3 + t5 + t7 = t14 + t15 ,
t2 = t1 + t3 + t4 + t8 = t11 + t15 ,
t3 = t1 + t2 + t6 + t9 = t10 + t15 ,
t4 = t2 + t5 + t6 + t8 = t11 + t13 ,
t5 = t1 + t4 + t6 + t7 = t13 + t14 ,
t6 = t3 + t4 + t5 + t9 = t10 + t13 ,
t7 = t1 + t5 + t8 + t9 = t12 + t14 ,
t8 = t2 + t4 + t7 + t9 = t12 + t11 ,
t9 = t3 + t6 + t7 + t8 = t12 + t10 ,
(6.27)
and they follow directly from the hyperelliptic representation; the equivalences are under
the relations (6.25), (6.26).
We define now the normalizing signs for Fpa,b [] promised earlier. Writing epa,b (1 ; ti ,
tj ) = epa,b [](i, j ) for simplicity, they are given by
30,0 [1 ](10, 11) = +1,
30,1 [1 ](1, 10) = +1,
31,1 [1 ](1, 2) = +1,
40,0 [1 ](10, 13) = +1,
40,1 [1 ](1, 14) = +1,
41,1 [1 ](1, 4) = +1,
60,0 [1 ](10, 10) = +1,
61,1 [1 ](1, 1) = +1.
The resulting polynomials are then as follows:

F30,0 [1 ](t) = +t10 t11 + t10 t14 + t11 t14 + t12 t13 + t12 t15 + t13 t15 ,
(6.28)
111
F40,0 [1 ](t) = +(t10 + t11 + t14 )(t12 + t13 + t15 ),

2
2
2
2
2
2
F60,0 [1 ](t) = +t10
+ t11
+ t12
+ t13
+ t14
+ t15
,
F30,1 [1 ](t) = +t1 (t10 + t11 + t12 + t13 ) + t2 (t10 + t12 + t13 + t14 )
+ t3 (t11 + t12 + t13 + t14 ) + t4 (t10 + t12 + t14 + t15 )
+ t5 (t10 + t11 + t12 + t15 ) + t6 (t11 + t12 + t14 + t15 )
+ t7 (t10 + t11 + t13 + t15 ) + t8 (t10 + t13 + t14 + t15 )
+ t9 (t11 + t13 + t14 + t15 ),
F40,1 [1 ](t) = +t1 (t14 + t15 ) + t2 (t11 + t15 ) + t3 (t10 + t15 )
+ t4 (t11 + t13 ) + t5 (t13 + t14 ) + t6 (t10 + t13 )
+ t7 (t12 + t14 ) + t8 (t11 + t12 ) + t9 (t10 + t12 ),
F31,1 [1 ](t) = +t1 t2
+ t 1 t 3 + t1 t 5 + t1 t 7 + t 2 t 3 + t2 t 4 + t2 t 8 + t3 t 6 + t3 t 9
+ t4 t 5 + t4 t 6 + t4 t 8 + t5 t 6 + t 5 t 7 + t6 t 9 + t7 t 8 + t7 t 9 + t 8 t 9 ,
F41,1 [1 ](t) = +t1 t4
+ t 1 t 6 + t1 t 8 + t1 t 9 + t 2 t 5 + t2 t 6 + t2 t 7 + t2 t 9 + t3 t 4
+ t3 t 5 + t3 t 7 + t3 t 8 + t4 t 7 + t 4 t 9 + t5 t 8 + t5 t 9 + t6 t 7 + t 6 t 8 ,
F61,1 [1 ](t) = +t12 + t22 + t32 + t42 + t52 + t62 + t72 + t82 + t92 .
(6.29)
The above expressions for F can be recast in the following, more systematic way,

ti tj ,
p = 3, 4, 6,
Fp0,0 [1 ](t) =
#(ij )=p,
ij ;i,j s[1 ]
Fp0,1 [1 ](t) =
ti tj , p = 3, 4,
#(ij )=p,
is[1 ],j s c [1 ]
Fp1,1 [1 ](t) =
ti tj ,
p = 3, 4, 6.
(6.30)
#(ij )=p,
ij,i,j s c [1 ]
Using the relations (6.25), (6.26), and (6.27), we can reduce these expressions to the linear
combinations of the 3 standard forms 6 [1 ]2 , F [1 ], and F c [1 ] given in Lemma 3. 2
6.3.2. Orbits of pairs of admissible sextets in genus 3
We consider next the same issue of orbits and consistency of phase assignments for
pairs of admissible sextets in genus 3. The following can be found by computer listings:
The set of all pairs of asyzygous sextets may be decomposed into 7 mutually exclusive
sets, according to whether the two sextets in the pair have 0, 1, 2, 3, 4, 5, or 6 spin
structures in common. Each of these sets of pairs transforms transitively under the
group of all modular transformations Sp(6, Z). The number of pairs in each category
is listed in the second column of the table below.
112
Table 1
Numbers of pairs of asyzygous sextets and modular orbits excluding 1
#
# pairs
# pairs 1
15120
10080
5400(1)
5400(2)
{2, 15, 17, 19, 22, 32}, {9, 10, 13, 16, 27, 28}
{4, 5, 10, 16, 26, 36}, {7, 11, 14, 20, 27, 33}
30240
21000
840
10080(1)
10080(2)
{3, 4, 12, 17, 25, 29}, {10, 20, 22, 25, 28, 35}
{5, 15, 22, 27, 29, 31}, {4, 11, 18, 19, 24, 31}
{2, 5, 17, 19, 28, 30}, {2, 8, 10, 15, 25, 32}
7560
5460
1260
1680
2520
{8, 12, 15, 20, 28, 33}, {4, 12, 13, 17, 24, 33}
{4, 5, 10, 16, 26, 36}, {3, 4, 7, 12, 22, 36}
{5, 14, 16, 20, 25, 35}, {6, 16, 24, 25, 30, 36}
3360
2800
2800
{1, 4, 10, 17, 27, 33}, {4, 10, 11, 17, 25, 26}
336
280
Orbits
Reference pair
280
any pair
Also, we shall need the number of pairs of sextets, such that neither sextet in the pair
contains a given spin structure 1 . The numbers of such pairs in each category is listed in
the third column of the table below. Under modular subgroup that preserves 1 , the sets
with 0, 1, and 2 spin structures in common are NOT transitive. Table 1 lists in the fourth
column the sizes of the orbits of the modular subgroup Sp[1 ](6, Z).
Consider the transformation law for sign assignments (; {i }, {i }) for pairs of
sextets in genus 3 given by the analogue of (6.22),
6
6

2 (i , M)
2 i , M = ; {i }, i ,
M; {Mi }, Mi
i=1
i=1
(6.31)
where M is any element of Sp(6, Z). With computer calculations, using all the generators S, MAi , and MBi , i = 1, . . . , 6, of the full Sp(6, Z), the following may be shown:
1. A unique (up to a global sign) and consistent sign assignment exists for all the
orbits in the sets with 0, 2 and 6 spin structures in common, as well as for the orbit
10080(2) in the set with 1 spin structure in common;
2. No consistent sign assignment exists for any of the other orbits.
6.4. Branching rules for Sp[](6, Z) orbits into Sp[](4, Z) orbits
In this section, we list the multiplicities of all the Sp[](4, Z) orbits which arise upon
factorization of the orbits of Sp[](6, Z). Recall that, in genus 3, the invariant set of pairs
of -admissible sextets with p common spin structures can be decomposed further into
irreducible orbits. Let these orbits be denoted by Qpq , with p indicating that the pairs of
-admissible sextets have p common spin structures, and q indicating which orbit is being
considered for given p. In Table 2, Npq denotes the multiplicity of the genus 2 orbit in the
113
Table 2
Branching rules for genus 3 orbits into genus 2 orbits
#(Q)
N01
N02
N21
Q30,0
12
Q30,1
36
Q31,1
Q40,0
Q40,1
Q41,1
Q60,0
Q60,1
Q61,1
18
18
18
6
0
9
Genus 2 orbit
Total number of pairs
N22
N23
N6
234
252
234
90
126
72
decomposition of the orbit Qpq . Also, #(Q) denotes the cardinality of the genus 2 even
spin structure orbit.
The computer analysis also shows that, in the above table, all copies of any given orbit
a,b
Qp [] always occur with the sign +. Thus there is no cancellation between the various
a,b
copies of any orbit Qa,b
p []. Of course, the global sign in front of each Fp [] is a matter
of convention, depending on the choice of global sign for the definition of Fpa,b [].
To each Sp[](6, Z) orbit Qpq , we can associate then a polynomial Ppq in genus 2
-constants, defined as the linear combination of the polynomials Fpa,b [], with coefficients given by the multiplicities with which the Sp[](4, Z) orbit Qa,b
p [] appears. Thus
P01 and P02 stand for the two polynomials corresponding to the two genus 3 orbits of pairs
with 0 common spin structures; P21 , P22 , P23 stand for the 3 orbits of pairs with 2 common
spin structures; and P6 stands for the single orbit of pairs with 6 common spin structures.
The overall sign of each polynomial is arbitrary. The relative signs are of course fixed by
the stabilizer group of the genus 3 spin structure . We have (we omit reference to
in F ),
P01 = 12F30,0 + 4F30,1 + 2F61,1 = 1262 + 6F 2F c ,
P02 = 4F30,1 + 4F31,1 + 6F60,0 = 6F + 6F c ,
P21 = +8F40,1 2F41,1 + 6F60,0 + 2F61,1 = 6F + 12F c ,
P22 = 4F40,0 2F41,1 + 3F60,0 = 462 + 3F + 2F c ,
P23 = 8F40,0 2F41,1 + 2F61,1 = 862 + 4F c ,
P6 = +6F60,0 + 4F61,1 = 6F + 4F c ,
(6.32)
114
where we have used (6.24) to express all of these in terms of the quantities 62 , F and F c .
The previous discussion results in the following lemma:
Lemma 4. Let the sign assignments (; {i }, {i }) satisfy the transformation (6.31) for
each orbit Qpq . Then we have

lim
t0
6
2
2

; {i }, i
[i ] 0, (3) i 0, (3)
({i },{i })Qpq
i=1
24

= 28 (1) Ppq (2) .
(6.33)
6.5. Candidates for 6 []( (3) )2

Each orbit Qpq contributes a consistent term to the candidate for the genus 3 superstring
measure, transforming covariantly under Sp(6, Z) transformations. Thus we can take an
arbitrary linear combination of these orbits and obtain a modular covariant expression

p,q
Npq
6
2
2

; {i }, i
[i ] 0, (3) i 0, (3) .
({i },{i })Qpq
[]( (3) )2
i=1
28 ( (1) )2
(6.34)
[]( (2) )2 .
must tend to
In view of Lemma 4,
Candidates for 6
6
the limit at t 0 of the linear combination (6.34) will be a multiple of ( (1) )2
6 []( (2) )2 if the multiplicities Npq satisfy
2N01 + 2N02 + 2N21 + N22 + 2N6 = 0,
N01 + 3N02 + 6N21 + N22 + 2N23 + 2N6 = 0,
in which case the limit is given by
2
2

28 (1) (12N01 + 4N22 + 8N23 )6 [] (2) .
(6.35)
(6.36)
It is convenient to summarize our findings in the following theorem:

Theorem 2. Let the genus 3 expression 6 []( (3) )2 be defined by (1.5), where Qpq are
the orbits of pairs of -admissible sextets from Table 2. Assume that the multiplicities Npq
satisfy the condition (6.35), and set N = 12N01 + 4N22 + 8N23 . Let factorize into
an even spin structure at genus 2. Then the expression 6 []( (3) )2 satisfies the three
conditions
(i ) 6 []( (3) )2 is holomorphic on the Siegel upper half space;
(3) )2 = det(C (3) + D)12 6 []( (3) )2 ;
(ii ) 6 [](

(iii ) limt0 6 []( (3) )2 = ( (1) )24 6 []( (2) )2 .
For example, an integer combination leading to a multiple of 28 ( (1) )2 6 []( (2) )2
by the square of an integer is N01 = 2, N02 = 4, N21 = 2, N23 = 1, in which case we
115
get
lim
t0
Npg
pg
({i },{i })Qpq
6
2
2

; {i }, i
[i ] 0, (3) i 0, (3)
i=1
24
2

= 16 28 (1) 6 [] (2) .
(6.37)
6.6. Vanishing of the genus 3 cosmological constant

We address a final issue of physical and mathematical significance, namely the behavior
of the genus 3 cosmological constant, defined by

4

6 [] (3) [] 0, (3) .
8
(6.38)
By its very construction, 6 [] transforms under the modular group Sp(6, Z) as

[](0, (3) )12 , and therefore the quantity 8 is a genus 3 modular form of weight 8.
An infinite family of modular forms of weight 4k may be generated as follows:

8k

[] 0, (3)
4k (3)
(6.39)
for k any positive integer. In [14], it was argued that 8 = 42 /8, based on asymptotic
identifications and numerical calculations. We shall assume that this is the only independent holomorphic modular form of weight 8, as we are not aware of any proof that this
statement is true. Given this assumption, as well as the asymptotic behavior established in
this paper for 6 []( (3) ), as the surface undergoes a separating degeneration, it is clear
that the modular form 8 must vanish in this limit. But 8 is non-zero in the same limit.
As a result, 8 = 0 throughout moduli space, and the cosmological constant vanishes to
three loop order.
Acknowledgements
We are happy to thank Edward Witten for stimulating discussions. Part of this work was
carried out while one of us (E.D.) was at the Aspen Center for Physics. All calculations
were carried out using M APLE 9.
References
[1] E. DHoker, D.H. Phong, Two-loop superstrings I, main formulas, Phys. Lett. B 529 (2002) 241255, hepth/0110247.
[2] E. DHoker, D.H. Phong, Two-loop superstrings II, the chiral measure on moduli space, Nucl. Phys. B 636
(2002) 360, hep-th/0110283.
[3] E. DHoker, D.H. Phong, Two-loop superstrings III, slice independence and absence of ambiguities, Nucl.
Phys. B 636 (2002) 6179, hep-th/0111016.
116
[4] E. DHoker, D.H. Phong, Two-loop superstrings IV, the cosmological constant and modular forms, Nucl.
Phys. B 639 (2002) 129181, hep-th/0111040.
[5] E. DHoker, D.H. Phong, Two-loop superstrings V, gauge slice independence of the N -point function, in
preparation.
[6] E. DHoker, D.H. Phong, Lectures on Two-Loop Superstrings, Hangzhou, Beijing, 2002, hep-th/0211111.
[7] K. Aoki, E. DHoker, D.H. Phong, Two-loop superstrings on orbifold compactifications, Nucl. Phys. B 688
(2004) 369, hep-th/0312181.
[8] E. DHoker, D.H. Phong, The geometry of string perturbation theory, Rev. Mod. Phys. 60 (1988) 9171065.
[9] E. DHoker, D.H. Phong, Superholomorphic anomalies and supermoduli space, Nucl. Phys. B 292 (1987)
317;
E. DHoker, D.H. Phong, Conformal scalar fields and chiral splitting on super Riemann surfaces, Commun.
Math. Phys. 125 (1989) 469;
E. DHoker, D.H. Phong, Superstrings, super Riemann surfaces, and supermoduli space, in: String Theory,
Rome, 1988, in: Symplectic Mathematics, vol. XXXIII, Academic Press, London, 1990, pp. 1767.
[10] G. Moore, Modular forms and 2-loop string physics, Phys. Lett. B 176 (1986) 369;
A. Belavin, V.G. Knizhnik, Algebraic geometry and the geometry of quantum strings, Phys. Lett. B 168
(1986) 201206.
[11] A. Belavin, V. Knizhnik, A. Morozov, A. Perelomov, Two and three loop amplitudes in the bosonic string
theory, Phys. Lett. B 177 (1986) 324328;
A. Morozov, Explicit formulae for one, two, three, and four-loop string amplitudes, Phys. Lett. B 184 (1987)
171176.
[12] A. Belavin, A. Polyakov, A. Zamolodchikov, Infinite conformal symmetry in two-dimensional quantum
field theory, Nucl. Phys. B 241 (1984) 333380;
D. Friedan, S. Shenker, The analytic geometry of two-dimensional conformal field theory, Nucl. Phys. B 281
(1987) 509545;
D. Friedan, S. Shenker, The integrable analytic geometry of quantum string, Phys. Lett. B 175 (1986) 287
296.
[13] J.I. Igusa, Modular forms and projective invariants, Amer. J. Math. 89 (1967) 817855.
[14] E. DHoker, D.H. Phong, Asyzygies, modular forms, and the superstring measure I, Nucl. Phys. B 710
(2005) 5882, the previous article in this issue, hep-th/0411159.
[15] J. Fay, Theta Functions on Riemann Surfaces, Springer Lecture Notes in Mathematics, vol. 352, Springer,
Berlin, 1973.
[16] D. Mumford, Tata Lectures on Theta I, Progress in Mathematics, vol. 28, Birkhuser, Boston, MA, 1983.
Nonperturbative states in type II superstring theory

from classical spinning membranes
Jan Brugues a , Joan Rojo a , Jorge G. Russo a,b
a Departament dEstructura i Constituents de la Matria, Universitat de Barcelona,
Diagonal 647, E-08028 Barcelona, Spain

b Instituci Catalana de Recerca i Estudis Avanats (ICREA), Spain
Received 2 September 2004; accepted 12 January 2005

Abstract
We find a new family of exact solutions in membrane theory, representing toroidal membranes
spinning in several planes. They have energy square proportional to the sum of the different angular momenta, generalizing Regge-type string solutions to membrane theory. By compactifying the
eleven-dimensional theory on a circle and on a torus, we identify a family of new nonperturbative
states of type IIA and type IIB superstring theory (which contains the perturbative spinning string
solutions of type II string theory as a particular case). The solution represents a spinning bound state
of D branes and fundamental strings. Then we find similar solutions for membranes on AdS7 S 4
and AdS4 S 7 . We also consider the analogous solutions in SU(N ) Matrix-theory, and compute the
energy. They can be interpreted as rotating open strings with D0 branes attached to their endpoints.
PACS: 11.25.-w; 11.25.Sq
1. Introduction
Understanding new aspects of M-theory may eventually lead to a powerful setup to uncover and clarify the nonperturbative physics of string theory (for a review of M-theory,
see [1]). An important piece of information about the theory is its mass spectrum. In eleven
E-mail address: jrusso@ecm.ub.es (J.G. Russo).
doi:10.1016/j.nuclphysb.2005.01.019
118
J. Brugues et al. / Nuclear Physics B 710 (2005) 117138
uncompact dimensions, determining the spectrum of quantum states is a hard problem,

there is no coupling constant and gravitational effects are always important. In compactifying the eleven-dimensional coordinate X 10 on a circle of radius R10 , one makes contact
3 / l 3 , where l is the elevenwith type IIA superstring theory with coupling gIIA = 2R10
P
P
dimensional Planck length [2,3]. For small radius, M-theory describes a weakly-coupled
string theory and it is possible to compute masses without the complication of gravitas
), s < 2 so that the gravitational
tional interactions, provided they are of order M = O(gIIA
2
forces (proportional to gIIA M) are negligible as gIIA 0. In particular, this is the reason
why there exists a simple weak coupling description for D0 branes, which have masses
1
).
O(gIIA
M-theory is known to contain membranes and five branes (see, e.g., [1,4]). In this paper
we will find new classical solutions in supermembrane theory [5,6] and compute their energy. In flat space, these solutions are similar to the spinning string solutions of [7], which
were found to be highly stable quantum mechanically in weakly-coupled string theory. For
a special choice of quantum numbers (such that the solution depends only on one spatial membrane coordinate) the solution reduces to the string solution of [7]. For general
quantum numbers, the energy has nontrivial dependence on the coupling constant gIIA of
1
). This means that for small radius R10 one can ignore gravitational
the form E O(gIIA
effects, so these solutions describe genuine quantum states with high-excitation quantum
numbers in M-theory, and new nonperturbative states in type IIA string theory. By further
compactifying the coordinate X 9 on a circle of radius R9 , one makes contact with type IIB
string theory. Ten-dimensional type IIB theory arises as M-theory on the torus (X 9 , X 10 )
in the limit that the area R9 R10 goes to zero at fixed gIIB = RR109 . Membranes states can
lead to various types of nonperturbative objects of type IIB theory, such as bound states
of fundamental strings and D strings [810]. The spinning solutions presented here lead to
a new class of nonperturbative states of type II string theory, representing rotating objects
which do not have a pure fundamental string interpretation. They have in general D brane
charges and fundamental string charges.
We then consider similar spinning solutions for membrane theory in AdS7 S 4 and in
AdS4 S 7 . These solutions are the membrane analogue of the general class of spinning
string solutions found in AdS5 S 5 in [11] (which includes some spinning string solutions appeared earlier in [12,13]). The energy admits a simple expansion at large angular
momenta.
It has been proposed [14] that M-theory in the light-cone frame is described by Matrixtheory, which can be viewed as a regularized theory of the supersymmetric membrane [15]
(for reviews see [16,17]). The fundamental degrees of freedom can be viewed as the D0
branes, which in the light-cone frame are expected to capture all the complicated dynamics
of M-theory. The Lagrangian is that of supersymmetric quantum mechanics described by
0 + 1 super-YangMills theory with sixteen supersymmetries. In the last part of the paper
we consider spinning solutions in Matrix-theory, found in [18], which are analogous to
the spinning membrane solutions. Here they are slightly generalized to incorporate rotation
in four different planes, and in addition we compute the energy, which was not done in [18].
There is an extensive literature on classical solutions in membrane/Matrix-theory. The
reader can look at [6,19] and more recently [18,2025]. Some of the present solutions are
similar to the ones obtained in [18]. In addition, we present generalizations and many new
119
solutions that include momentum and winding in the directions X9 , X10 , which allows to
make contact with type II string theories. These are in fact the most interesting solutions
that give rise to new nonperturbative states with rotation in type II string theory.
2. Spinning membranes solutions in flat space

2.1. General rotating ansatz
The starting point is the bosonic part of the action for the supermembrane [5] in flat
space

T2
S=
(2.1)
d 3 hh X X h , = 0, . . . , 10.
2
T2 is the membrane tension, T2 = (2lP3 )1 . The equations of motion for the metric on the
membrane worldvolume gives
h =
X X
,

, = 0, 1, 2.
(2.2)
In what follows we use the notation ( 0 , 1 , 2 ) (, , ). Treating the X and h as independent fields, we now fix the gauge in the usual way [6,26], by setting
h = h = 0,
h L2 = g,
(2.3)
where L is an arbitrary constant with units of length and g the determinant of the spacelike
part of the induced metric. The action (2.1) becomes

g ij
T2
h i X j X 1 .
S=
(2.4)
d 3 L X X +
2
L
Using the equation of motion for the induced metric, hij = i X j X the action (2.4)
reads

T2
1
3
{X , X }{X , X } ,
S=
(2.5)
d L X X +
2
2L
where the Poisson bracket is defined as usual as
{f, g} rs r f s g,
r, s = , ,
(2.6)
for any two differentiable functions f, g on the two-dimensional manifold.

In terms of the X fields, the constraints for the induced metric (2.3) that must be
imposed to the solutions are
X X
X X
=
= 0,
(2.7)

X X
X X 2
X X
X X
2
=
. (2.8)
L

120
Now we consider a general rotating ansatz of the following form

X0 = ,
Z1 (, , ) = X1 + iX2 = r1 (, )ei1 +i1 (,) ,
Z2 (, , ) = X3 + iX4 = r2 (, )ei2 +i2 (,) ,
Z3 (, , ) = X5 + iX6 = r3 (, )ei3 +i3 (,) ,
Z4 (, , ) = X7 + iX8 = r4 (, )ei4 +i4 (,) ,
X9 = X10 = 0.
(2.9)
This represents a membrane spinning in four orthogonal planes Z1 , . . . , Z4 . Using this

ansatz, the action for the membrane (2.5) reads
T2
S=
2

d L
3
4

a2 ra2
a=1

4

1
2
2 2
2
2
2
+
{ra , rb } + ra rb {a , b } + 2ra {a , rb }
.
2L
(2.10)
a,b=1
The equations of motion for the radial and the angular coordinates are given by

rc L2 c2
4

ra2 {c , a }2
+ {ra , c }

4

+
{rc , ra }, ra + ra2 {rc , a }, a
a=1
a=1
= 0,
(2.11)
4

2

ra {c , a }, a + {c , ra }, ra = 0.
(2.12)
a=1
Inserting the ansatz (2.9), the constraints (2.7), (2.8) take the form
4

a ra2 a =
a=1
a ra2 a = 0,
(2.13)
a=1
4

4

a=1
a2 ra2

4

1
1
{ra , rb }2 + ra2 rb2 {a , b }2 + ra2 {a , rb }2 .
2
2
(2.14)
a,b=1
2.2. Constant radius solution

A simplification of our original ansatz (that generalizes the rotating string solutions of
[7] to membrane theory) is considering solutions of constant radius and with the phases
a depending linearly on , . Assuming that the ra are constants, the equation of motion
(2.12) for the phases becomes

4

ra2 {b , a }, a = 0.
121
(2.15)
a=1
This is indeed solved by a (, ) = ka + la , i.e.,

Za (, , ) = ra eia +i(ka +la ) ,
a = 1, . . . , 4,
(2.16)
with ka , la integer numbers, as implied by the periodicity condition satisfied by a closed

membrane. The equation of motion (2.11) of the radial coordinates leads to
1 2
ra (lb ka la kb )2 = 0.
L
4
Lb2
(2.17)
a=1
This equation determines the frequencies b as a function of the radii ra and the different
winding numbers.
Using the ansatz (2.16), the constraints can be written as
4

a ra2 ka =
a=1
a ra2 la = 0,
a=1
4

4

a2 ra2
a=1

=
4

(2.18)
ra2 rb2 (ka lb kb la )2 .
(2.19)
a,b=1
Combining with Eq. (2.17), we get

=2
2
4

ra2 a2 .
(2.20)
a=1
The solution (2.16) appeared in [18] in the case of light-cone gauge membrane theory.
In the light-cone gauge the constraints (2.18) (which lead to constraints on the angular
momenta and winding numbers, see (2.27)) is absent. The reason is the following. This
constraint comes from (2.7). In the light cone gauge X + = p+ , this gives p+ s X =
X i s X i , which can be solved for X provided the integrability condition {X i , X i } = 0
is satisfied, which is automatically the case for the solution (2.16). In the present case,
instead, X0 = , X9 = X10 = 0, and this physical membrane state (2.16) exists only if
the angular momenta and winding numbers are constrained by (2.18) (i.e., (2.27)). The
light-cone gauge represents an extreme physical situation of infinite momentum, where the
theory typically gets simplified. In Section 2.4 we shall consider a generalization where the
constraint (2.7) can be solved by adding winding and linear momentum in the directions
X9 , X10 .
2.3. Energy and angular momenta
Now we evaluate the conserved quantum numbers of the rotating membrane solutions.
We start with the action (2.5),

4

2 2

1
T2
3
2
2
2
2
2
ra a + ra a + {ra , X0 }2
S=
d L X 0 r5 r6
2
L
a=1
122

6
1 1
1
2
2
2
2
2
{ra , rb } + (ra rb ) {a , b } + rb {ra , b }
+
,
L
2
2
(2.21)
a,b=1
with coordinates defined as

Za (, , ) = ra (, , )eia (,,) ,
X9 r5 ,
X10 r6 ,
a = 1, . . . , 4,
X0 .
The conserved quantum numbers are

S
,
E = d d 0 , 0
X0

S
Ja = d d a , a a .

(2.22)
(2.23)
(2.24)
In the present case, we obtain

E = 4 2 T2 L,
Ja = 4 2 T2 La ra2 .
(2.25)
The energy of our solution can be obtained from (2.20) (coming from the constraint (2.8)),
which determines ,

E 2 = 2 4 2 T2 L
(2.26)
Ja a ,
a
where a are determined by (2.17). Note that each term in the sum on the right-hand side
of (2.26) is positive definite, since Ja a = 4 2 T2 La2 ra2 . The constraints (2.18) become
4

Ja ka = 0,
a=1
4

Ja la = 0.
(2.27)
a=1
Let us now particular cases where some of the angular momenta vanish. Because of the
constraint (2.18), the minimum number of nonvanishing angular momenta is two. However,
in the case of rotation in two planes, i.e., Z3 , Z4 = const, it can be seen that Eqs. (2.17),
(2.18) lead to 1 = 2 = 0, so this solution is trivial. The first nontrivial case is the rotation
in three planes.1
In the case of rotation in three planes, we can write (2.17), (2.18) as
3

Ja ka = 0,
Ja la = 0,
(2.28)
3

Ja
(lb ka la kb )2 = 0,
a
(2.29)
a=1
b2
3

a=1
a=1
1 When the coordinates X , X are compact, there are solutions with rotation in one and two planes, de9
10
scribed in Section 2.4. In the case of eleven uncompact coordinates, there is a solution with rotation in two planes
in the light-cone gauge, where the constraint (2.18) is absent. In Section 5 we discuss the analogous solution in
the Matrix-theory.
123
where we have defined a (4 2 T2 L3 )1/3 a . In (2.29) we have expressed the radii in

terms of the angular momenta using (2.25). The reason is that the energy must be expressed
in terms of the conserved quantum numbers, angular momenta and winding numbers, characterizing the state (winding numbers are not conserved in interactions because they are
along contractible circles; nevertheless, they characterize the states of the free theory). The
parameters a are dimensionless (just as a ), but they depend only on Ja , la , ka through
(2.29). Eqs. (2.29) are coupled cubic equations, and the analytic solution is complicated. It
is easy to see that solutions with real frequencies (and real radii) exist. For example, with
the choice
(k1 , k2 , k3 ) = (1, 3, 4),
(l1 , l2 , l3 ) = (4, 3, 11),
(J1 , J2 , J3 ) = (4, 2, 2),
(2.30)
the constraints (2.28) are satisfied, and from (2.29) we obtain the real solution ( 1 , 2 ,
3 ) = (4.9, 21.5, 14.4) (for which also the radii are real, since Ja /a are positive, see
(2.25)). In conclusion, the energy in the case of rotation in three planes is
3
2 2/3
Ja a ,
E = 2 4 T2
2
(2.31)
a=1
where the a are determined by Eqs. (2.29) in terms of integer quantum numbers ka , la , Ja
obeying (2.28). As expected, the energy does not depend on the arbitrary length parameter
L introduced in the choice of gauge.
2.4. More general rotating solutions
We now consider the membrane wrapped around two compact directions, X 9 and X 10 ,
with linear momentum,
X9 = R9 (n9 + m9 ) + q9 ,
X10 = R10 (n10 + m10 ) + q10 .
(2.32)
For the coordinates Za , a = 1, . . . , 4, we use the same ansatz (2.16). The constraints (2.7),
(2.8) now take the form
4

a ra2 ka + R9 q9 n9 + R10 q10 n10 = 0,
(2.33)
a ra2 la + R9 q9 m9 + R10 q10 m10 = 0,
(2.34)
a=1
4

a=1

a

a2 ra2
q92
2
q10
= L2
d

2
ra2 a2 + R92 R10
(n9 m10 n10 m9 )2 , (2.35)
a=1
where in the last equation we have used the relation
124
c2
J2
a
2/3 2

(lc ka la kc )2 4 2 T2
R9 (m9 kc n9 lc )2
a
a
2 2/3 2
4 T2
R10 (m10 kc n10 lc )2 = 0,
(2.36)
which follows from the equation of motion for the radial coordinates. The parameters a
are defined as before, a (4 2 T2 L3 )1/3 a .
The linear momenta along X9 and X10 are

S
Pi = d d i , i
(2.37)
.
X i
We obtain
P9 = 4 2 T2 Lq9 ,
P10 = 4 2 T2 Lq10 .
(2.38)
Because X9 and X10 are compact coordinates, they are quantized,

P9
n 9
,
R9
P10
n 10
.
R10
(2.39)
The constraints take the form

4

Ja ka + n9 n 9 + n10 n 10 = 0,
a=1
4

Ja la + m9 n 9 + m10 n 10 = 0.
(2.40)
a=1
The energy is then obtained from (2.25) with determined by (2.35). We get
4
2/3
2

2
Ja a + 4 2 T2 R92 R10
(n9 m10 n10 m9 )2
E 2 = 2 4 2 T2
a=1
n 29
R92
n 210
2
R10
(2.41)
The a are numbers determined by (2.36) in terms of the integer quantum numbers Ja , lc ,
kc , m9 , n9 , m10 , n10 and the parameters R9 , R10 . Note that in the energy E there is no de2
pendence on the arbitrary constant L as expected. The second term proportional to R92 R10
is the usual contribution to the energy coming from the torus area, times membrane tension,
times membrane charge (equal to n9 m10 n10 m9 ).
2.4.1. Rotation in one plane
The simplest rotating solution is the case Z2 , Z3 , Z4 = const corresponding to rotation
in one plane Z1 , with the previous ansatz for X9 , X10 ,
X0 = ,
Z1 = rei +i(k +l) ,
X9 = R9 (n9 + m9 ) + q9 ,
X10 = R10 (n10 + m10 ) + q10 .
125
In this case, the constraints become,

kJ + n9 n 9 + n10 n 10 = 0,
lJ + m9 n 9 + m10 n 10 = 0,
2
(m10 k n10 l)2 .
L2 2 = R92 (m9 k n9 l)2 + R10
The energy is then

2

n 2
n 2
2
E 2 = 2 4 2 T2 L J + 4 2 T2 R92 R10
(n9 m10 n10 m9 )2 + 92 + 10
2
R9 R10

2

R2
1
= 4 2 T2 (n9 m10 n10 m9 )R9 R10 +
n 29 + n 210 29
.
R9
R10
(2.42)
Thus the energy is a complete square. This is related to the fact that the corresponding
quantum state is a BPS state. In fact, it is the same BPS state studied in [8,9]. In general, the
BPS state of [8,9] represents a nonmarginal bound state of fundamental string and D string
in type IIB string theory, with charges n 9 , n 10 and momentum n n9 m10 n10 m9 (see
also Section 3.2). It describes an excited string state (or excited membrane state), whose
specific quantum state can be any of the exponential number of states at this level. The
present solution represents one of these quantum states, namely the state with maximum
angular momentum.
2.4.2. Rotation in two planes
Let us now consider the solution describing rotation in two planes Z1 , Z2 , by setting
Z3 = Z4 = const. The constraints and the equations of motion become,
k1 J1 + k2 J2 + n9 n 9 + n10 n 10 = 0,
l1 J1 + l2 J2 + m9 n 9 + m10 n 10 = 0,
2/3 2

J2
2
12 =
R9 (m9 k1 n9 l1 )2 + R10
(k1 l2 k2 l1 )2 + 4 2 T2
(m10 k1 n10 l1 )2 ,
2
2/3 2

J1
2
22 =
R9 (m9 k2 n9 l2 )2 + R10
(k1 l2 k2 l1 )2 + 4 2 T2
(m10 k2 n10 l2 )2 .
1
The solution simplifies when the linear momenta along the directions coordinates X9 , X10
vanish, i.e., n 9 = n 10 = 0. In this case, we have
J1 k1 + J2 k2 = J1 l1 + J2 l2 = 0,
so that l1 k2 = l2 k1 , and we find an explicit simple expression for the frequencies

2/3 2

2
12 = 4 2 T2
(m10 k1 n10 l1 )2 ,
R9 (m9 k1 n9 l1 )2 + R10
2/3 2

2
22 = 4 2 T2
R9 (m9 k2 n9 l2 )2 + R10
(m10 k2 n10 l2 )2 .
Then, the energy of this solution is
2
2/3
2

2
Ja a + 4 2 T2 R92 R10
(n9 m10 n10 m9 )2
E 2 = 2 4 2 T2
a=1
(2.43)
(2.44)
(2.45)
126

2

2

2

k1
k1
2
2
|l1 J1 | + |l2 J2 |
= 2 4 T2 R9
m9 n9 + R10
m10 n10
l1
l1
2

2
(n9 m10 n10 m9 )2 .
+ 4 2 T2 R92 R10
(2.46)
We have added absolute
value bars to the two terms in the sum to account for the fact that
each term in the sum a Ja a is positive definite because Ja ra2 a . Note that, unlike
the previous case, the energy is not a complete square.
In the case X9 = n9 R9 and X10 = R10 , i.e., m9 = n10 = 0, m10 = 1, the energy
becomes

2

2 k1 |l J | + |l J | + 4 2 T 2 R 2 R 2 n2 .
E 2 = 2 4 2 T2 R92 n29 + R10
(2.47)
1 1
2 2
2
9 10 9
2
l1
We will return to this energy formula in the next section.
3. Nonperturbative states in type II string theory from rotating membranes

3.1. Type IIA string theory or M-theory on S 1
Type IIA string theory is obtained from M-theory by compactifying the eleventh dimension X 10 on a circle. The string tension and string coupling are related to the membrane
and M-theory parameters as follows:
=
1
4 2 R
10 T2
1
T1 = (2 )
2
11
= 16 5 lP9 ,
2
3
gIIA
= 4 2 R10
T2 ,
(3.1)
= 2R10 T2 ,
1 2 3/2
1

T2 = 2lP3
= 4
gIIA ,
(3.2)
2
R10
2
= gIIA
.
(3.3)
The perturbative solutions of type IIA string theory have the form dictated by the doubledimensional reduction ansatz [27],
X 10 = R10 ,
X = 0,
= 0, . . . , 9.
(3.4)
The membrane solutions of the previous sections have explicit dependence on the coordinate, and also momentum in the eleventh coordinate X 10 . One expects that in string
theory they arise as nonperturbative objects.
First, consider the type IIA limit in ten uncompact dimensions which is achieved by
setting R9 . In this case we must set n9 = m9 = 0 in the solution (2.16), (2.32). Now
the momentum P9 is continuous. In addition, we set n10 = 0, m10 = 1 to have X 10 =
R10 + q10 . In terms of the type IIA parameters, the energy (2.41) of the membrane
solution of Section 2.4 then takes the form
E
= P92
4

2/3
gIIA
a=1
Ja a (gIIA ) +
n 210
2
gIIA
(3.5)
127
where angular momenta and winding numbers satisfy the constraints

4

ka Ja = 0,
a=1
4

la Ja = n 10 .
(3.6)
a=1
The a are determined in terms of the coupling constant gIIA and the conserved quantum
numbers by the system of Eqs. (2.36), which in terms of string parameters reads
c2
J2
a
4/3
(lc ka la kc )2 gIIA kc2 = 0.
(3.7)
In order to solve explicitly the above equations, we consider again the case of rotation in
two planes. Then the equations for the frequencies are given by
J22
4/3
(l1 k2 k1 l2 )2 = gIIA k12 ,
2
J2
4/3
22 1 (l1 k2 k1 l2 )2 = gIIA k22 .
1
12
(3.8)
These equations admit a number of solutions, which simplify in the particular case of spins
of equal magnitude and opposite sign, J1 = J2 (which, by the constraint equations (3.6),
implies k1 = k2 ). In this case the frequencies are given by
8/3
2gIIA k12
1

,
4 k 2 /n
4
1 + 1 + 4gIIA
1 10

2
4
n 10
4gIIA k12
2 = 4/3 1 + 1 +
.
n 410
2gIIA
1 =
n 210
(3.9)
(3.10)
Inserting into (3.5), we obtain the exact expression for the energy of this configuration. At
weak coupling, gIIA 1, the energy has the following expansion
E 2 =
2J1 n 210
2
gIIA
2
+ 4J1 gIIA
k12
n 210
6
.
+ O gIIA
(3.11)
There are several points which are worth noticing. First, we again find a Regge-type formula. A priori, this is not obvious from (3.5), since the frequencies in general depend
on the angular momenta. The point is that the frequencies depend on the combination
2 (l k l k )2 which for J = J is equal to n
210 (see (3.6)). Second, also for this soJ1,2
1 2
2 1
1
2
1
, characteristic of D branes. As explained
lution the energy has the behavior E = const gIIA
in the introduction, this guarantees that one can ignore gravitational back reaction effects
in the weak coupling limit. The origin of the 1/gIIA behavior is a D0 brane charge, coming
from momentum of the membrane in the X10 direction. Because the membrane has zero
charge, now there is no D2 brane and no winding charge for the fundamental string. Thus
the state represents a rotating system of D0 brane and fundamental string in uncompactified
ten-dimensional type IIA string theory.
128
Now we consider the case of compact X9 coordinate, and the solution with energy given
by (2.47). In terms of type IIA string theory parameters, the analytical energy formula
(2.47)describing rotation in two planesbecomes

R 2 n2
k2
2 R92 n29
2
E =
+ 21 |l1 J1 | + |l2 J2 | + 9 29 ,
2
gIIA
l1
k1 J1 + k2 J2 = l1 J1 + l2 J2 = 0.
(3.12)
1
. This behavior is characteristic of
At weak coupling, the energy behaves as E = const gIIA
D branes. For this particular solution there is no D0 brane present, since we have set the D0
brane charge n 10 to zero. The origin of the 1/gIIA behavior is the presence of a D2 brane.
Indeed, for this solution, the membrane charge is equal to n9 m10 n10 m9 = n9 . Since the
membrane is also extended in the Z1 , Z2 and X10 directions, after dimensional reduction
one is left with a rotating bound state of a D2 brane and a fundamental string (with winding charge also equal to n9 ). Remarkably, the energy formula has a simple Regge-type
behavior, E 2 J .
Now consider the particular case n9 = 0. We get

2

E 2 = 2 4 2 T2 R10 |k1 J1 | + |k2 J2 | = |k1 J1 | + |k2 J2 | .
(3.13)
It is interesting that the energy of these states does not depend on l1 , l2 . In particular,
the energy is the same as in the case l1 = l2 = 0, which in the dimensionally reduced
theory corresponds to a string (see (2.16), (3.4)). Thus there is a family of membranes
(parametrized by l1 , with l2 = l1 J1 /J2 ) giving rise to states with the same energy in the
string theory.
The solutions with l1 = l2 = 0 and energy given by (3.13) are the solutions of [7]. These
solutions were found to be quantum mechanically very stable, with a lifetime proportional
2
to [7] gIIA
(mass)5 . The reason of this long life time is that the closed string cannot classically break due to the fact that during the evolution there is never contact between two
points of the string. It can only decay by emitting light modes. These states are probably the
most stable states in the string spectrum. We expect that the present membrane solutions
with n9 = 0 and energy given by (3.12) are also long-lived quantum mechanically.
An interesting question is what is the supergravity solution describing these spinning
membranes. Spinning M-brane supergravity solutions were given in [28]. However, none
of these solutions describe the present configurations. The reason is that the present membrane configurations rotate along the directions where the membrane is extended.
3.2. Type IIB string theory
Type IIB string theory arises by compactifying M-theory on the torus X9 , X10 . The
type IIB parameters are related to M-theory parameters as follows:
=
1
,
2
4 R10 T2
2
gIIB
=
2
R10
R92
R9B =

.
R9
(3.14)
129
In terms of type IIB parameters, the general expression of the energy (2.41) reads
2/3
4
R9B
1
Ja a + 2 (n9 m10 n10 m9 )2

2
gIIB
R9B
a=1

R2
n 210
2
+ 9B
+
n
.
9
2
2
gIIB
E2 = 2
(3.15)
Let us now consider particular cases. In the case of rotation on one plane, Eq. (2.42),
we have

2
n 210
1
R9B
2
2
+ n 9 + 2
,
E = (n9 m10 n10 m9 )
(3.16)
R9B
gIIB
describing the BPS nonmarginal bound state of D string (charge n 10 ), fundamental string
(charge n 9 ) and momentum n n9 m10 n10 m9 , as anticipated in Section 2.4.1.
In the case of the solution (2.47) describing rotation in two planes, we have

n29
n29
k12
2
|l
+
J
|
+
|l
J
|
+
.
E2 =
(3.17)
1
1
2
2
2
2
gIIB
l12
R9B
This is the T-dual of the D2 brane/fundamental string system described in Section 3.1.
Because the D2 brane is extended in X9 and Z1 , Z2 directions, T-duality in X9 should
produce a complicated rotating system involving D-strings, D3 branes and fundamental
strings, with momentum n9 . It is remarkable that the energy of such a system is given by a
simple Regge-type formula. As the type IIA T-dual counterpart, we expect that this system
is classically stable and long-lived quantum mechanically.
4. Spinning membranes in AdSp S q

In this section we construct analogous membrane solutions in a curved background
spacetime, AdSp S q , which are relevant for AdS/CFT correspondence applications [29].
First we obtain the general expression for our solutions, which are of the form (2.16), and
then we find an explicit expression for the energy of these solutions in the large angular
momentum limit. The solutions generalize the rotating circular strings of [11] to membrane
theory.
We consider membranes on AdS7 S 4 and AdS4 S 7 or, generically, AdSp S q . Let
Y , = 0, . . . , p be the embedding coordinates in the AdS space and X k , k = 1, . . . , q + 1

the embedding coordinates in the sphere. The action for the membrane reads

1
T2
h
S=
d 3 hh Y Y + X k X k +
2
2

k k

2
2
+ Y Y + RA + X X RS .
(4.1)
130
The Lagrange multipliers enforce the constraints

q+1

Xk2 = RS2 ,
2
Y Y = RA
,
(4.2)
k=1
so that they transform as h

where = (, +, . . . , +, ). We have defined ,
under world-volume reparametrizations. The radii of the sphere and of the AdS space are
given by

1/2
AdS4 S 7 ,
RS = 2RAdS = lP 25 2 N

RS = RAdS /2 = lP (N )1/3 AdS7 S 4 ,
(4.3)
where lP is the Planck length. From the action (4.1) one derives the following equations of
motion for the coordinates,

Xk = hh Xk ,
(4.4)

= hh Y .
Y
(4.5)
Now we consider the following ansatz,
Z1 X1 + iX2 = r1 ei1 +im1 +in1 ,
Z2 X3 + iX4 = r2 ei2 +im2 +in2 ,
..
.
Zd X2d1 + iX2d = rd eid +imd +ind ,
where d = 2 for
S4
and d = 4 for
S7.
(4.6)
For the AdS coordinates, we take
Z0 Y0 + iYp = RA ei0 .
(4.7)
We use the same gauge as in Section 2, given by h = 0, h = 0 and h L2 = g. For

our ansatz (4.6), these constraints read
d

ma a ra2 =
a=1
d

na a ra2 = 0,
(4.8)
a=1
2 2
RA
0
d

ra2 a2 =
1 2 2
ra rb (mb na ma nb )2 .
L2
(4.9)
b<a
a=1
The equations of motion that are derived from the action (4.1) give rise to the following
relations between the parameters:
= L02 ,
(4.10)
1
1
2
h m2a h n2a + h ma na
L
L
L
d
1 2
= Lwa2
rb (nb ma na mb )2 .
L
= La2
b=1
(4.11)
131
The energy and the angular momenta can be derived from (2.23), (2.24),
Ja = 4 2 T2 Lra2 a 4 2 T2 LRS2 Ja ,
(4.12)
E = 4 T2 LRA 0 4 T2 LRA E.

The constraint da=1 ra2 = RS2 can be written as
(4.13)
d

Ja
a=1
wa
= RS2 .
(4.14)
Solving (4.11), one finds the frequencies in terms of . Then is determined using (4.14).
Plugging the result into (4.9) (with ra2 = RS2 Ja /a ), one determines the energy (4.13) in
terms of Ja and winding numbers. The system of equations can be solved systematically
as power series in J1 ,
J
d

Ja .
(4.15)
a=1
Note that large J implies also large Lagrange multiplier ||. Indeed, since the radii ra
on the sphere are bounded, large angular momenta requires large frequencies, which, by
(4.11), implies large ||, and the scaling || J 2 . This situation is similar to that of
spinning strings, see [11]. So we first solve Eq. (4.11) for the frequencies (inserting ra2 =
RS2 Ja /a ) using perturbation theory and then plug the expression of the frequencies in the
constraint (4.14) to express in terms of the angular momenta and the winding numbers.
We find

d
RS2
1
2
,
wc = J +
(4.16)
J
(n
m
n
m
)
+
O
a c a
a c
2L2 J 2
J2
a=1
||1/2 = J

d
RS2
1
2
.
J
J
(n
m
n
m
)
+
O
a c c a
a c
2L2 J 3 a,c
J2
(4.17)
Hence we find the energy:

E
= 02
RS2
2
RA
J +
2
RS4
d

2 L2
J 2 RA
a,b=1

1
.
(ma nb na mb ) Ja Jb + O
J2
(4.18)
Taking the square root and expanding, we get

E

d

RS3
1
RS
2
.
J=
(m
n
n
m
)
J
J
+
O
a
b
a
b
a
b
RA
2J 3 RA L2
J2
(4.19)
a,b=1
Finally, expressing Eq. (4.19) in terms of the physical energy and spin, one obtains that the
dependence on the arbitrary constant L cancels, as expected, to give the result,

d
R 5 (4 2 T2 )2
J
1
2
,
E
(4.20)
= S
(m
n
n
m
)
J
J
+
O
a b
a b
a b
RS
2J 3
J2
a,b=1
supplemented with the constraints (4.8)
132

d

Ja ma =
a=1
d

Ja na = 0.
(4.21)
a=1
This spinning toroidal membrane solution is the analog of the circular spinning strings in
AdS5 S 5 found in [11]. For the spinning strings, in the large angular momentum limit,
one finds the formula
E J =
3
1 2
mi Ji + .
2J 2
(4.22)
i=1
As in the case of AdS5 S 5 (4.22), the energy (4.20) of configurations with angular momentum on the sphere exhibits the behavior E J , as opposed to the Regge behavior
E 2 J of flat space. In the framework of the AdS/CFT correspondence, the energy formula (4.20) gives the anomalous dimension of a CFT operator with SO(5) (for the 5 + 1
CFT) or SO(8) (for the 2 + 1 CFT) charges Ja .
In the particular case of rotation in two planes, there is a significant simplification,
because in this case the constraint (4.8) implies that n1 m2 m2 n1 = 0. As a result, the
energy is given by the leading order term without higher order corrections,
J
.
(4.23)
RS
This suggests that this configuration should be supersymmetric and that the corresponding
CFT operator should be BPS, since its bare dimension is exact in the classical membrane
approximation.
E=
5. Membranes and Matrix-theory

In this section we consider solutions of the matrix model equations which are the analogues of the classical membrane solutions of Section 2.2, and study their properties.
5.1. Membranes solutions and matrix model solutions
The starting point is the bosonic part of the supersymmetric quantum-mechanics Yang
Mills Lagrangian describing the dynamics of N D0 branes, given by

2

R10
1
i i
i
j 2
tr X X + 6 X , X
, i = 1, . . . , 8,
L=
(5.1)
2R10
2lP
where X i are matrices of the Lie algebra of SU(N ). The equations of motion are
8
2
k i i
R10
d2 k
X ,X ,X .
X
=
2
6
dt
lP
(5.2)
i=1
This equation of motion is supplemented with the self-consistency Gauss constraint,

8

[Xi , X i ] = 0.
i=1
(5.3)
133
In order to establish a dictionary between membrane solutions and matrix model solutions,
let us recall how the toroidal membrane arises as the N = limit of the SU(N ) matrix
model. The toroidal membrane coordinates can be expanded as
X i (, , ) =
i
Xnm
( )ein +im
n,m=
i
Xnm
( )Tnm ,
(5.4)
n,m=
where Tn1 n2 Tn are generators of the area-preserving diffeomorphism algebra of the

torus,
n m)T
n +m ,
{Tn , Tm } = (
(
n m)
n1 m2 m1 m2 .
(5.5)
The analogue of the Tn generators in the SU(N ) matrix model is a special basis of generators {Jn } of the Lie algebra of SU(N ) satisfying the algebra [30,31]

2
(
n m)
Jm+
[Jm , Jn ] = 2i sin
(5.6)
n , N = odd,
N

[Jm , Jn ] = 2i sin
(5.7)
(
n m)
Jm+
n , N = even.
N
In Appendix A we review the construction and properties of these algebras. A general
matrix on the Lie algebra can be expanded as
X i ( ) =
N
1

i
Xnm
( )Jnm ,
(5.8)
n,m=0
where prime means excluding the term (n, m) = (0, 0). In the N = limit, the algebra
(5.6), (5.7) approaches the area-preserving diffeomorphism algebra of the torus, and the
SU(N ) Matrix-theory is expected to reproduce exactly the same dynamics of membrane
theory in the light-cone gauge, since they are described by the same Hamiltonian.
Given any toroidal membrane classical solution, it can be expanded as in (5.4). Then
one can write down an ansatz for a matrix model solution of the form (5.8) by taking
i
the same coefficients Xnm
and extending the sum taking into account the periodicity,
J(m1 +N k1 ,m2 +N l1 ) = J(m1 ,m2 ) . For the rotating membrane solutions of Section 2, the ansatz
is straightforward since in complex coordinates Za there is a single term in the sum (5.4),
i.e., Za = ra eia eika +ila .
5.2. Rotating matrix model solution
Here we review the rotating solution of the matrix model given in [18]. The starting
point is the matrix model Lagrangian (5.1) in complexified coordinates as
L=
R10

1
tr Za Z a + 6 tr [Za , Zb ] Za , Zb + Za , Zb Za , Zb
2R10
8lP
a, b = 1, . . . , 4.
(5.9)
134
The equations of motion of the matrix model and the constraint take the form
2

R10
d2
Zb , [Zb , Za ] + Zb , Zb , Za ,
Z
=
a
2
6
dt
2lP

Z a , Za + Z a , Za = 0.
(5.10)
(5.11)
A solution with rotation in the planes Za can now be found by analogy with the membrane
case, i.e., replacing Tnm = ein +im by Jnm , with the matrix ansatz
Za = ra eia t Jn a ,
a = 1, . . . , d,
(5.12)
where 1 < d 4. Consider the case of even N . The equations of motion (5.10) give the
relation

d
2
R10
2
2
2
(
nb n a ) ,
rb sin
a = 4 6
(5.13)
N
lP
b=1
which is the analog of the relation (2.17). The Gauss constraint (5.11) is satisfied automatically by the ansatz (5.12), using that Jm = Jm .
5.3. Energy of this configuration
Let us now evaluate the light-cone energy P of the solutions that we have found above.
This is obtained by inserting the solution (5.12) into the Hamiltonian corresponding to the
matrix model (5.9). We obtain

d
d
N 2 2 N R10 2 2 2
n a n b .
a ra 6
ra rb sin
P = H =
(5.14)
2R10
N
lP
a=1
a,b=1
Using Eq. (5.13) for the frequencies, the energy can be expressed in a compact form as

d
N R10 2 2 2
P = 6
(5.15)
n a n b .
ra rb sin
N
lP
a,b=1
For large N , this has the same form as the membrane energy (2.26), which (using (2.17),
(2.25)) can be written as
2
Emem
= const
d

ra2 rb2 (lb ka la kb )2 .
a,b=1
Let us now consider explicit cases. In the case of SU(2), the simplest nontrivial example,
the generators of the Lie algebra (5.7) coincide with the usual Pauli matrices, as can be
seen in Appendix A. For this group we find that any nontrivial configuration (with nonzero
energy) has an energy given by

d
d
2R10
2R10
2 2
2
P = 6
(5.16)
= 6
ra rb sin
ra2 rb2 .
2
lP a,b=1, a=b
lP a,b=1, a=b
135
In a similar way one can obtain the energy for the case of SU(3), but now it can be seen
that there are configurations with different energy.
The matrix model solution can be viewed pictorially as follows. Recall that the diagonal
entries in the matrices Za represent strings which begin and end on the same D0 brane,
whereas the nondiagonal entries represent strings going from one D0 brane to a different
one. For the SU(3) case, there are three D0 branes. The Za can be proportional to any of the
eight generators listed in Appendix A. Solutions with a Za proportional to a generator with
three nondiagonal entries represent a configuration of three strings joining the three D0
branes, forming a triangle, which rotate in the plane Za . In another plane Zb , the solution
can have Zb proportional to a diagonal generator J(1,0) or J(2,0) , representing a rotating
system of three D0 branes with strings beginning and ending on the same D0 brane.
Acknowledgements
Work supported in part by the European Commission RTN programme under contract
HPNR-CT-2000-00131, by MCYT FPA 2001-3598, CIRIT GC 2001SGR-00065, and by
the MEC grant AP2002-2415.
Appendix A. Construction of SU(N) generators

In this appendix we review the construction and relevant properties of the SU(N ) generators in the representation that we are using [30]. First we do this in general and then
we give the explicit expressions for the SU(3) generators. These algebra becomes the areapreserving diffeomorphism algebra of the toroidal membrane in the N limit.
A basis for the SU(N ) algebras can be built from two unitary N N matrices,
1 0 0 ...
0
0 1 0 ... 0
0 0 ...
0 0 1 ... 0
0
. . . .
.. .. ..
0 0 2 . . .
0 ,
. . ... ,
g
h
. .
.. . .
..
.. ..
0 0 0 ... 1
.
.
.
1 0 0 ... 0
0 0 0 . . . N 1
g N = hN = 1,
(A.1)
where is an N th root of unity with period no smaller than N , that is, = ei4/N for N
odd and = ei2/N for N even. Now using hg = gh, it follows that the unitary N N
matrices
Jm =
m1 m2
2
g m1 hm2 ,
m
(m1 , m2 ),
(A.2)
span the algebra of SU(N ), that is, they close under multiplication
Jn Jm =
(
nm)

2
Jn +m ,
(A.3)
136
with as defined above, so that their commutation rules are given in terms of trigonometric
structure constants,

n m/2
2

nm/2

[Jm , Jn ] =
(A.4)
Jm+
m
n Jm+
n = 2i sin
n,
N
where the last equality follows for odd N (for even N one has instead of 2 ), and the
vector product is defined in the usual way, n m
= n1 m2 m1 n2 . Another interesting
property is that for the case of three generators whose winding number satisfy
n 1 + n 2 + n 3 = 0,
(A.5)
then the algebra of these generators gets simplified and we get a twisted SU(2) trigonometric algebra

2
(
n1 n 2 ) Jn 3 ,
[Jn 1 , Jn 2 ] = 2i sin
(A.6)
N
together with the corresponding permutations. This algebra is twisted SU(2) in the sense
that it is the algebra of SU(2) but with a single trigonometric structure constants, since due
to the condition (A.5) it follows that n 1 n 2 = n 1 n 3 = n 2 n 3 .
From their definition, one can check that all the generators are traceless. Consider, in
particular, the generators of the form
J(m1 ,0) ,
m1 < N.
(A.7)
The trace is given by

tr(Jm1 ,0 ) = tr g m1 =
N
1

k ,
(A.8)
k=0
which indeed vanishes, since is the N th root of unity. These generators are normalized
as follows,

tr Jn Jn = tr(Jn J n ) = tr 1 = N.
(A.9)
Now we compute the explicit expressions for the generators of the SU(2) and SU(3)
algebra in the basis of [30]. For SU(2) the generators reduce to the usual Pauli matrices,
J(1,0) = 3 ,
J(0,1) = 1 ,
J(1,1) = 2 .
(A.10)
For SU(3) the expressions are more involved, differing from the usual canonical basis.
These generators of the algebra of SU(3) are given by

1
1
i4/3
i2/3
J(1,0) =
,
J(2,0) =
,
e
e
ei2/3
ei4/3

1
1
J(0,2) = 1
,
J(0,1) =
1 ,
1
1

J(1,1) =
ei4/3
ei4/3
J(2,2) =
ei2/3
J(2,2) =
ei4/3
,
1

1 ,
ei4/3

1 ,

J(2,1) =
ei2/3
137
ei4/3
ei2/3

.
(A.11)
One can explicitly check with this equations the algebra of the generators, given by

2
[Jm , Jn ] = 2i sin
(A.12)
m
n Jm+
n
3
subject to the periodicity of the generators of the algebra, namely
J(m1 +3k1 ,m2 +3l1 ) = J(m1 ,m2 )
(A.13)
with ka , la integer numbers. One can explicitely check that all these generators are indeed
traceless, that is
tr(Jn ) = 0.
(A.14)
References
[1] P.K. Townsend, Four lectures on M-theory, hep-th/9612121.
[2] P.K. Townsend, The eleven-dimensional supermembrane revisited, Phys. Lett. B 350 (1995) 184, hep-th/
9501068.
[3] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
[4] M.J. Duff, Supermembranes, hep-th/9611203.
[5] E. Bergshoeff, E. Sezgin, P.K. Townsend, Supermembranes and eleven-dimensional supergravity, Phys. Lett.
B 189 (1987) 75.
[6] E. Bergshoeff, E. Sezgin, P.K. Townsend, Properties of the eleven-dimensional super membrane theory,
Ann. Phys. 185 (1988) 330.
[7] D. Chialva, R. Iengo, J.G. Russo, Decay of long-lived massive closed superstring states: exact results,
JHEP 0312 (2003) 014, hep-th/0310283;
D. Chialva, R. Iengo, Long lived large type II strings: decay within compactification, hep-th/0406271.
[8] J.H. Schwarz, An SL(2, Z) multiplet of type IIB superstrings, Phys. Lett. B 360 (1995) 13, hep-th/9508143;
J.H. Schwarz, Phys. Lett. B 364 (1995) 252, Erratum.
[9] J.G. Russo, A.A. Tseytlin, Waves, boosted branes and BPS states in M-theory, Nucl. Phys. B 490 (1997)
121, hep-th/9611047.
[10] J.G. Russo, BPS bound states, supermembranes, and T-duality in M-theory, hep-th/9703118.
[11] G. Arutyunov, J. Russo, A.A. Tseytlin, Spinning strings in AdS(5) S 5 : new integrable system relations,
Phys. Rev. D 69 (2004) 086009, hep-th/0311004.
[12] S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS(5) S 5 , Nucl. Phys. B 668 (2003) 77, hep-th/
0304255.
[13] G. Arutyunov, S. Frolov, J. Russo, A.A. Tseytlin, Spinning strings in AdS(5) S 5 and integrable systems,
Nucl. Phys. B 671 (2003) 3, hep-th/0307191.
[14] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[15] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988)
545.
[16] D. Bigatti, L. Susskind, Review of matrix theory, hep-th/9712072.
138
[17] W. Taylor, M(atrix) theory: matrix quantum mechanics as a fundamental theory, Rev. Mod. Phys. 73 (2001)
419, hep-th/0101126.
[18] M. Axenides, E.G. Floratos, L. Perivolaropoulos, Rotating toroidal branes in supermembrane and matrix
theory, Phys. Rev. D 66 (2002) 085006, hep-th/0206116.
[19] J. Hoppe, H. Nicolai, Phys. Lett. B 196 (1987) 451.
[20] J. Hoppe, Some classical solutions of membrane matrix model equations, hep-th/9702169;
J. Arnlind, J. Hoppe, More membrane matrix model solutionsand minimal surfaces in S 7 , hep-th/
0312062.
[21] M. Alishahiha, M. Ghasemkhani, Orbiting membranes in M-theory on AdS(7) S(4) background,
JHEP 0208 (2002) 046, hep-th/0206237.
[22] M. Alishahiha, A.E. Mosaffa, Circular semiclassical string solutions on confining AdS/CFT backgrounds,
JHEP 0210 (2002) 060, hep-th/0210122.
[23] S.A. Hartnoll, C. Nunez, Rotating membranes on G(2) manifolds, logarithmic anomalous dimensions and
N = 1 duality, JHEP 0302 (2003) 049, hep-th/0210218.
[24] P. Bozhilov, M2-brane solutions in AdS(7) S(4), JHEP 0310 (2003) 032, hep-th/0309215.
[25] J. Hoppe, S. Theisen, Spinning membranes on AdS(p) S q , hep-th/0405170.
[26] I. Bars, C.N. Pope, E. Sezgin, Massless spectrum and critical dimension of the supermembrane, Phys. Lett.
B 198 (1987) 455.
[27] 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.
[28] M. Cvetic, D. Youm, Rotating intersecting M-branes, Nucl. Phys. B 499 (1997) 253, hep-th/9612229.
[29] 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.
[30] D.B. Fairlie, P. Fletcher, C.K. Zachos, Trigonometric structure constants for new infinite algebras, Phys.
Lett. B 218 (1989) 203.
[31] E.G. Floratos, The HeisenbergWeyl group on the Z(N ) Z(N ) discretized torus membrane, Phys. Lett.
B 228 (1989) 335.
Comparative study of CP asymmetries

in supersymmetric models
E. Gabrielli a,c , K. Huitu a,b , S. Khalil d,e
a Helsinki Institute of Physics, P.O. Box 64, 00014 University of Helsinki, Finland
b Division of HEP, Department of Physics, P.O. Box 64, 00014 University of Helsinki, Finland
c CERN, PH-TH, Geneva 23, Switzerland
d IPPP, University of Durham, South Rd., Durham DH1 3LE, UK
e Department of Mathematics, German University in CairoGUC, New Cairo, El Tagamoa Al Khames, Egypt
Received 25 August 2004; received in revised form 3 December 2004; accepted 24 December 2004
Abstract
We systematically analyze the supersymmetric contributions to the mixing CP asymmetries and
branching ratios of B KS and B KS processes. We consider both gluino and chargino
exchanges in a model independent way by using the mass insertion approximation method. While
we adopt the QCD factorization approach for evaluating the corresponding hadronic matrix elements, a critical comparison with predictions in naive factorization one is also provided. We find
that pure chargino contributions cannot accommodate the current experimental results on CP asymmetries, mainly due to b s constraints. We show that charged Higgs contributions can relax
these constraints making chargino responsible for large asymmetries. On the other hand, pure gluino
exchanges can easily saturate both the constraints on B KS and B KS CP asymmetries.
Moreover, we also find that the simultaneous contributions from gluino and chargino exchanges could
easily account for the present experimental results on the mentioned asymmetries. Remarkably, large
experimentally allowed enhancements of B KS branching ratio can easily be achieved by the
contribution of two mass insertions in gluino exchanges. Finally, we analyze the correlations between the CP asymmetries of these processes and the direct CP asymmetry in b s decay. When
all experimental constraints are satisfied, supersymmetry favors large and positive values of b s
asymmetry.
E-mail address: emidio.gabrielli@helsinki.fi (E. Gabrielli).

doi:10.1016/j.nuclphysb.2004.12.036
140
E. Gabrielli et al. / Nuclear Physics B 710 (2005) 139188
PACS: 12.60.Jv; 13.25.Hw; 14.80.Ly
1. Introduction
The B-factories are producing interesting experimental results with continuously increasing integrated luminosities. Currently they offer one of the most promising routes to
test the KobayashiMaskawa ansatz for the CP violation. The understanding of the CP violating mechanism is one of the major open problems in particle physics. There are reasons
to believe that the Standard Model (SM) cannot provide a complete description for the CP
violating phenomena in nature. For instance, it is established that the SM mechanism of CP
violation cannot account for the observed size of the baryon asymmetry in the Universe,
and additional sources of CP violation beyond the SM are needed [1].
Recently, BaBar and Belle Collaborations [2] announced large deviations from the SM
expectations in the CP asymmetry of B 0 KS and branching ratio of B 0 K 0 . These
discrepancies have been interpreted as possible consequences of new physics (NP) beyond
the SM [312]. For B 0 and B 0 decays to a CP eigenstate fCP , the time dependent CP
asymmetries are usually described by rates afCP (t),
(B 0 (t) fCP ) (B 0 (t) fCP )
(B 0 (t) fCP ) + (B 0 (t) fCP )
= CfCP cos MBd t + SfCP sin MBd t,
afCP (t) =
(1)
where CfCP and SfCP represent the parameters of direct and indirect CP violations, respectively, and MBd is the B 0 eigenstate mass difference.
In the Standard Model, the angle in the unitary triangle of CabibboKobayashi
Maskawa (CKM) matrix [13], can be measured from B meson decays. The golden mode
B 0 J /KS is dominated by tree contribution and SJ /KS = sin 2 + O(3 ), where
is the Cabibbo mixing angle (see, e.g., [14]). These uncertainties are less than 1%.
The dominant part of the decay amplitudes for B 0 KS , KS is assumed to come
from the gluonic penguin [15,16], but some contribution from the tree level b uus
decay
is expected. The | is almost pure |s s and consequently this decay mode corresponds
also accurately, up to terms of orders O(2 ), to sin 2 in the SM [17]. The b uus
tree
level contribution to Bd K was estimated in [16]. It was found that the tree level
amplitude is less than 2% of the gluonic penguin amplitude. Thus also in this mode one
measures the angle with a good precision in the SM. Therefore, it is expected that NP
contributions to the CP asymmetries in B 0 KS , KS decays are more significant than
in B 0 J /KS and can compete with the SM one.
New sample of data have been recently analyzed by BaBar and Belle Collaborations
and the results with higher statistics have been now announced in [18,19]. The new experimental value of the indirect CP asymmetry parameter for B 0 J /KS is given by
[18,19]
SJ /KS = 0.726 0.037,
(2)
141
which does not differ much from the previous one [20,21,23], and agrees quite well with
the SM prediction 0.715+0.055
0.045 [22]. However, results of Belle on the corresponding sin 2
extracted for B 0 KS process has changed dramatically [18,19]
SKS = 0.50 0.25+0.07
0.04
(BaBar),
= 0.06 0.33 0.09 (Belle),
(3)
where the first errors are statistical and the second systematic, showing now a better agreement than before [23,24]. However, as we can see from Eq. (3), the relative central values
are still different. BaBar results [18] are more compatible with SM predictions, while
Belle measurements [19] still show a deviation from the cc measurements of about 2 .
Moreover, the average SKS = 0.34 0.20 is quite different from the previous one [25],
displaying now 1.7 deviation from Eq. (2).
Furthermore, the most recent measured CP asymmetry in the B 0 KS decay is found
by BaBar [18] and Belle [19] Collaborations as
S KS = 0.27 0.14 0.03 (BaBar),
= 0.65 0.18 0.04 (Belle),
(4)
with an average S KS = 0.41 0.11, which shows a 2.5 discrepancy from Eq. (2). For
the previous results see (BaBar) [26] and (Belle) [24].
It is interesting to note that the new results on s-penguin modes from both experiments
differ from the value extracted from the cc mode (J /), BaBar by 2.7 and Belle by
2.4 [18,19]. At the same time the experiments agree with each other, and even the central
values are quite close:
0.42 0.10 (BaBar),
0.43+0.12
0.11
(Belle).
On the other hand, the experimental measurements of the branching ratios of B 0

K 0 and B 0 K 0 at BaBar [26,27], Belle [28], and CLEO [29] lead to the following
averaged results [25]:

6
BR B 0 K 0 = 8.3+1.2
(5)
1.0 10 ,

0
+6.0
0
6
BR B K = 65.25.9 10 .
(6)
From theoretical side, the SM predictions for BR(B K)1 are in good agreement with
Eq. (5), while showing a large discrepancy, being experimentally two to five times larger,
for BR(B K) in Eq. (6) [30]. This discrepancy is not new and it has created a growing
interest in the subject. However, since it is observed only in B K process, mechanisms
based on the peculiar structure of meson, such as intrinsic charm [31] and gluonium [32]
content, have been investigated to solve the puzzle.
Supersymmetry (SUSY) is one of the most popular candidates for physics beyond
the SM [33]. In SUSY models there are new sources of CP violation besides the CKM
1 In order to simplify our notation, from now on everywhere, where the symbols B and K will appear, they
will generically indicate neutral B 0 (B 0 ) and K 0 (K 0 ) mesons, respectively.
142
phase [34]. The soft SUSY breaking (SSB) terms contain several parameters that may be
complex, as may also be the SUSY preserving parameter in the Higgs sector. Then,
new CP violating phases can naturally arise in the SSB sector of scalar-quarks (squarks)
and scalar-leptons (sleptons). These new phases have significant implications for the electric dipole moments (EDM) of the neutron, electron, and mercury atom [34,35] and can
be strongly constrained by the negative search of CP violation in EDM experiments.
Therefore, it remains a challenge for SUSY to provide an explanation for the above
mentioned discrepancies in B-decays, while keeping the above EDMs within their experimental ranges. This can be possible, for instance, in SUSY models where the CP violating
phases entering in b s transitions are independent from the corresponding ones affecting EDMs. An interesting example is provided by SUSY models with flavor dependent CP
violating phases [3537].
In SUSY framework, the main effect on B K is usually assumed to come from the
gluino loop contributions to s-penguin diagrams [69,38], if R parity is conserved.2 However, charginos could also be responsible for such discrepancy, as it has been discussed in
[10,11], although b s constraints strongly suppress their contribution [11]. Similarly
to B K, one would a priori expect large effects from SUSY corrections to B K
as well, which may contradict the experimental results reported in Eq. (4). Possible mechanisms to explain such behavior in the supersymmetric context have been proposed in
Refs. [8,9,39].
In this paper we perform a complete analysis of SUSY contributions to the CP asymmetries and branching ratios of B K and B K processes. Previous analyses on
the same issue have considered either gluino [6] or chargino exchanges [7,10,11]. We think
that in the framework of general SUSY models a complete analysis involving both effects
is in fact needed. Indeed, we find that a SUSY scenario in which both chargino and gluino
give sizeable contributions to CP violating processes, is an interesting viable possibility.
For instance, large effects of chargino contributions to CP asymmetries that one would
expect to be excluded by b s constraints [11], could be achieved taking into account
gluino exchanges. This is due to potentially destructive interferences between chargino and
gluino amplitudes in b s decay that eventually relax b s bounds.
In our analysis we consider both the effects of gluino and chargino exchanges by using
the mass insertion method (MIA) [40]. As known, this method is a useful tool for analyzing
SUSY contributions to flavor changing neutral current processes (FCNC) since it allows to
parametrize, in a model independent way, the main sources of flavor violations in general
SUSY models.
We take into account all the relevant operators involved in the effective Hamiltonian
for B = 1 transition, and provide analytical expression for the corresponding Wilson
coefficients. We analyze the most interesting scenarios in which one or two mass insertions
are dominant in both gluino and chargino sector.
2 Recently, SUSY effects to B K decays have been analyzed in the context of R-parity breaking models
[5]. In this case, as well as in models with extra dimensions [4], effects are induced at tree-level. We will restrict
our analysis to the R-parity conserving scenarios, where SUSY corrections to B K process always enter at
1-loop.
143
An important issue in these calculations is the method of evaluating the hadronic matrix
elements for exclusive hadronic final states, which may play a crucial role in CP asymmetries. Many studies have been done with naive factorization approach (NF) [41,42], for the
computation of two-body non-leptonic B decays. Recently, a new approach has been developed, called QCD factorization (QCDF) [4345], which offers the possibility to include
non-factorizable contributions and to calculate the strong phases. The drawback in this approach is that it includes undetermined parameters H,A and phases H,A , characterizing
the infrared divergences. We critically consider both approaches and analyze theoretical
uncertainties connected with SUSY predictions. We provide a comparative study of SUSY
contributions from chargino and gluino to B K and B K processes in NF and
QCDF approaches. We also analyze the branching ratios of these decays and investigate
their correlations with CP asymmetries. Finally, we discuss the correlations between CP
asymmetries of these processes and the direct CP asymmetry in b s decay [46].
This paper is organized as follows. In Section 2 we give the basic ingredients needed for
calculations, with many important formulas provided in Appendices AC. In Section 3 we
discuss the QCDF method in the evaluation of B-meson decay amplitudes of B K and
B K. Moreover, we compare SUSY predictions between NF and QCDF approaches.
Section 4 is devoted to the CP asymmetries of these decay modes. Both effects from gluino
and chargino exchanges are discussed. In Section 5 the branching ratios and their correlations to asymmetries are considered. Section 6 contains the analysis of SUSY contributions
to direct b s asymmetry, and correlations to the mentioned CP asymmetries are also
considered. Finally, Section 7 contains our main conclusions.
2. SUSY contributions to the effective Hamiltonian of B = 1 transitions
We start our analysis by considering the supersymmetric effect in the non-leptonic
B = 1 processes. Such an effect could be a probe for any testable SUSY implications
B=1 for these
in CP violating experiments. The most general effective Hamiltonian Heff
processes can be expressed via the operator product expansion (OPE) as [47]

10

GF
p
p
B=1
=
p C1 Q1 + C2 Q2 +
Ci Qi + C7 Q7 + C8g Q8g
Heff
2 p=u,c
i=3
i , Ci C i },
+ {Qi Q
(7)
, with V
where p = Vpb Vps
pb the unitary CKM matrix elements satisfying the unitarity
triangle relation t + u + c = 0, and Ci Ci (b ) are the Wilson coefficients at low
energy scale b O(mb ). The basis Qi Qi (b ) is given by the relevant local operators
renormalized at the same scale b , namely
p
Q2 = (pb)
V A (s p)V A ,

(qq)
V A ,
Q3 = (s b)V A
Q1 = (p b )V A (s p )V A ,

Q4 = (s b )V A
(q q )V A ,
Q5 = (s b)V A

q
(qq)
V +A ,
Q6 = (s b )V A

q
(q q )V +A ,
144
Q7 = (s b)V A
3
q
Q9 = (s b)V A
3
q
Q7 =
2
2
eq (qq)
V +A ,
Q8 = (s b )V A
3
q
eq (qq)
V A ,
e
mb s (1 + 5 )F b,
8 2
Q10 = (s b )V A
eq (q q )V +A ,
3
q
Q8g =
eq (q q )V A ,
gs
A
mb s (1 + 5 )GA
t b .
8 2
(8)
Here and
the SU(3)c
=
V A q
(1 5 )q, and q runs
Moreover, eq are quark electric charges in unity of e, (qq)
over u, d, s, c, and b quark labels. In the SM only the first part of right-hand side of Eq. (7)
p
(inside first curly brackets) containing operators Qi will contribute, where Q1,2 refer to
the currentcurrent operators, Q36 to the QCD penguin operators, and Q710 to the electroweak penguin operators, while Q7 and Q8g are the magnetic and the chromo-magnetic
i (b ) are obtained from Qi
dipole operators, respectively. In addition, operators Q i Q
by the chirality exchange (q1 q2 )V A (q1 q2 )V A . Notice that in the SM the coefficients
C i identically vanish due to the V A structure of charged weak currents, while in MSSM
they can receive contributions from both chargino and gluino exchanges [48,49].
Due to the asymptotic freedom of QCD, the calculation of hadronic weak decay amplitudes can be factorized by the product of long and short distance contributions. The first
ones, that will be analyzed in the next section, are related to the evaluation of hadronic
matrix elements of Qi and contain the main uncertainty of our predictions. On the other
hand, the latter are contained in the Wilson coefficients Ci and they can be evaluated in
perturbation theory with high precision [47]. For instance, all the relevant contributions of
particle spectra above the W mass (mW ) scale, including SUSY particle exchanges, will
enter in Ci (W ) at W O(mW ) scale.
The low energy coefficients Ci (b ) can be extrapolated from the high energy ones
Ci (W ) by solving the renormalization group equations for QCD and QED in the SM.
The solution is generally expressed as follows [47]

U ij (, W )Cj (W ),
Ci () =
(9)
A
stand for color indices, and t
color matrices,
2 i[ , ].
where U ij (, W ) is the evolution matrix which takes into account the re-summation of the
terms proportional to large logs (s (W ) log(W /b ))n (leading), (s2 (W ) log(W /b ))n
(next-to-leading), etc., in QCD.
In our analysis we include the next-to-leading order (NLO) corrections in QCD and
QED for the Wilson coefficients Ci=110 as given in Ref. [47], while for C7 () and
C8g () we include only the leading order (LO) ones.3 The expressions for the evolution
matrix U ij (, W ) at NLO in QCD and QED can be found in Ref. [47]. The reason for
retaining only the LO accuracy in C7 () C8g () is that the matrix elements of the dipole
operators enter the decay amplitudes only at the NLO [43].
3 However, in the evaluation of the BR of b s decay, the complete NLO corrections in C ( ) have
7 b
been taken into account [50].
145
Next we discuss the SUSY contributions to the effective Hamiltonian in Eq. (7). The
modifications caused by supersymmetry appear only in the boundary conditions of the
Wilson coefficients at W scale and they can be computed through the appropriate matching with one-loop Feynman diagrams where Higgs, neutralino, gluino, and chargino are
exchanged, see for instance Refs. [48,49,51]. Only chargino and gluino contributions can
provide a potential source of new CP violating phase in MSSM and could account for the
observed large deviations in B KS asymmetry. In principle, the neutralino exchange
diagrams involve the same mass insertions as the gluino ones, but they are strongly suppressed compared to the latter. For these reasons we neglect neutralino in our analysis. The
charged Higgs contributions cannot generate any new source of CP violation in addition to
the SM one, or any sizeable effect to operators beyond the SM basis Qi . However, when
charged Higgs contributions are taken into account together with chargino or gluino exchanges, their effect is relevant. In particular, as we will show in the next sections, due to
destructive interferences with chargino and gluino amplitudes, b s constraints can be
relaxed allowing sizeable contributions to the CP asymmetries.
The results for Wilson coefficients at W scale can be expressed as follows

g
Ci (W ) = CiW + CiH + 1
(10)
Ci + Ci ,
t
where CiW , CiH , Ci , and Ci correspond to the W , charged Higgs, chargino, and gluino ex
changes, respectively. In our analysis we will impose the boundary conditions for Ci , and
g
Ci at the scale W = mW , although they should apply to the energy scale at which SUSY
particles are integrated out, namely MS . However, these threshold corrections, originating
from the mismatch of energy scales, are numerically not significant since the running of s
from MS to mW is not very steep.
Finally, for NDR renormalization scheme, the electroweak contributions to the Wilson
coefficients are given by [48,51]
11 s
C2W = 1
,
6 4

1
s (W,H,)
1 (W,H,)
(W,H,)
(W,H,)
(W,H,)
C3
B
E
B
=
+
+
C
,
u
d
2
6 sin w
2
24
s (W,H,)
s (W,H,)
(W,H,)
(W,H,)
E
E
=
,
C5
=
,
C4
8
24

s (W,H,)
(W,H,)
(W,H,)
(W,H,)
E
4C
=
,
C7
=
+ D (W,H,) ,
C6
8
6
(W,H,)
C8
= 0,
(W,H,)
4C (W,H,) + D (W,H,)
=
C9
6

1
(W,H,)
(W,H,)
(W,H,)
Bd
,
+ 2
+ Bu
4C
sin w
C1W =
14s
,
16
(W,H,)
C10
= 0,
(W,H,)
C7
= M(W,H,) ,
(W,H,)
C8g
= Mg(W,H,) ,
(11)
where s and are evaluated at mW scale. The functions appearing above, include the contributions from photon-penguins (D), Z-penguins (C), gluon-penguins (E), boxes with
146
external down-quarks (Bd ) and up-quarks (Bu ), the magnetic (M ), and the chromomagnetic-penguins (Mg ). The corresponding SM results are [47] D W D(xt ), C W
C(xt ), E W E(xt ), and B W B(xt ) with xt = m2t /m2W , and analogously for charged
Higgs [51], D H DH (xH ), C H CH (xH ), E H EH (xH ), and B H BH (xH ) with
xH = m2t /m2H , where the loop functions B, C, D, E and BH , CH , DH , EH and are provided
in Appendix C. Regarding the SM and charged Higgs contributions to magnetic-penguins,
we have [48]

3
3
MW = xt F1 (xt ) + F2 (xt ) ,
MgW = xt F1 (xt ),
2
2

2
xH
2
MH =
F1 (xH ) + F2 (xH ) cot2 + F3 (xH ) + F4 (xH ) ,
2
3
3

x
H
F1 (xH ) cot2 + F3 (xH ) ,
MgH =
2
where the functions Fi (x) are reported in Appendix C. Finally, the gluino and chargino
exact contributions to the expressions appearing in Eq. (11) can be found in Refs. [48,51],
while here we will provide only the corresponding results in the so-called mass insertion
approximation.
Regarding the SUSY contributions to the opposite chirality operators Q i , we stress
that while in the SM the Wilson coefficients C i identically vanish, in SUSY models
chargino and gluino exchanges could sizeably affect these coefficients. However, in case
of charginos, these effects are quite small being proportional to the Yukawa couplings of
light quarks [11] and so we will not include them in our analysis. Moreover, we have also
neglected the small contributions to C1,2 coming from box diagrams, where both chargino
and gluino are exchanged [11,51].
As mentioned in the introduction, in order to perform a model independent analysis of
FCNC processes in general SUSY models, it is very convenient to adopt the mass insertion
approximation (MIA) method [40]. This method has been applied in many analyses of
FCNC processes in K, D, and B meson sector and leptonic sector, mediated by gluino and
neutralino exchanges, respectively [40]. More recently, it has been extended to the sector
of FCNC processes mediated by chargino exchanges [11,52,53]. In particular, in [11], the
chargino functions appearing in Eq. (11) have been calculated at the first order in MIA.
Let us briefly recall the main steps of this approximation. In MIA framework, one
chooses a basis (called super-CKM basis) where the couplings of fermions and sfermions to neutral gaugino fields are flavor diagonal. In this basis, the interacting Lagrangian
involving charginos is given by

diag

a
Lq q + = g
Vk1 Kba
Yd . K + ab dRa k+ u bL
dL k+ u bL Uk2
k
a,b

diag
Vk2 K . Yu ab dLa k+ u bR ,
(12)
where qR,L = 12 (1 5 )q, and contraction of color and Dirac indices is understood. Here
diag
Yu,d are the diagonal Yukawa matrices, and K stands for the CKM matrix. The indices a,
b and k label flavor and chargino mass eigenstates, respectively, and V , U are the chargino
mixing matrices defined by
147

2mW sin
,
1
2
(13)
where M2 is the weak gaugino mass, is the supersymmetric Higgs mixing term, and tan
is the ratio of the vacuum expectation value (VEV) of the up-type Higgs to the VEV of the
down-type Higgs.4 As one can see from Eq. (12), the Higgsino couplings are suppressed
by Yukawas of the light quarks, and therefore they are negligible, except for the stopbottom interaction which is directly enhanced by the top Yukawa (Yt ). In our analysis we
neglect the Higgsino contributions proportional to the Yukawa couplings of light quarks
with the exception of the bottom Yukawa Yb , since its effect could be enhanced by large
tan . However, it is easy to show that this vertex cannot affect dimension six operators
of the effective Hamiltonian for B = 1 transitions (operators Qi=110 in Eq. (7)) and
only interactions involving left down quarks will contribute. On the contrary, contributions
proportional to bottom Yukawa Yb enter in the Wilson coefficients of dipole operators
(C7 , C8g ) due to the chirality flip of b s and b sg transitions.
As mentioned in the case of MIA, the flavor mixing is displayed by the non-diagonal
q
entries of the sfermion mass matrices. Denoting by AB the off-diagonal terms in the
mass matrices for the up and down, respectively, where A, B indisfermion (q = u,
d)
cate chirality couplings to fermions A, B = (L, R), the AB squark propagator can be
expanded as
U M + V 1 = diag(m + , m + )
and M + =
M2
2mW cos
q

i(AB )ab
iab
+
+ O 2 ,
(14)
k2 m
2 (k 2 m
2 )2
where q = u, d selects up or down sector, respectively, a, b = (1, 2, 3) are flavor indices,
1 is the unit matrix, and m
is the average squark mass. As we will see in the following, it is
q
convenient to parametrize this expansion in terms of the dimensionless quantity (AB )ab
q
2
. At the first order in MIA, the penguin and box diagrams which contribute to
(AB )ab /m
the B = 1 effective Hamiltonian are given in Figs. 1 and 2, respectively. Evaluating the
diagrams in Figs. 1 and 2 by retaining only terms proportional to bottom- and top-quark
Yukawa couplings and performing the matching, the chargino contributions to the Wilson
coefficients in Eq. (11) can be determined from the following relations [11]

u
u

LL

Ka2 Kb3 LL ba RF +
Ka2 K33 RL 3a Yt RFRL
F =

q 1
2 1 AB ab
qAa qBb = i k 2 1 m
a,b

u
u

Yt RFLR + K32
K32
Ka3 LR
K33 RR
R RR + RF0 Yt2 ,
a3
33 F
(15)
where the symbol F {D, C, E, Bd , Bu , M , Mg }, and the detailed expressions for RFLL ,
RFLR , RFRL , RFRR , and RF0 can be found in Appendix A.5 Notice that the last term in
4 This tan should not be confused with the angle of the unitarity triangle.
5 The expression R 0 was missing in [11]. The contribution is O(2 ) and does not change the numerical
F
results.
148
Fig. 1. Penguin diagrams for B = 1 transitions with chargino (I+ ) exchanges at the first order in mass insertion. Here U , U = {u,
c,
t}, with indices h, k, m, n = {L, R} and I, J = {1, 2}. The cross symbol in the squark
propagator indicates the mass insertion. The corresponding diagrams at zero order in mass insertion are simply
obtained by removing the mass insertion in the propagators of up-type squarks (U ).
Fig. 2. Box diagrams for B = 1 transitions with chargino exchanges at the first order in mass insertion, where
U , U = {t, c,
u},
U = {c, u}, and D = {b, s, d}, where h, k, l, n, r, m, q = {L, R}.
Eq. (15), proportional to RF0 , is independent of mass insertions. This is due to the fact
that for chargino exchanges the super-GIM mechanism is only partially effective, when
only squarks but not quarks are taken to be degenerate.
Here we will just concentrate on the dominant contributions which turn out to be due to
the chromo-magnetic-penguin (Mg ) and Z-penguin (C) diagrams [11]. From the above expressions, it is clear that LR and RR contributions are suppressed by order 2 or 3 , where
= sin c 0.22, with c the Cabibbo angle. In our analysis we adopt the approximation
of retaining only terms proportional to order = sin c . In this case, Eq. (15) simplifies as
follows [11]
F = LL RFLL + Yt RL R RL
F ,
(16)
149
u ) + ( u ) and
u
u
where LL = (LL
32
RL = (RL )32 + (RL )31 .
LL 31
LL
RL
The functions RF and RF depend on the SUSY parameters through the chargino
masses (mi ), squark masses (m)
and the entries of the chargino mass matrix. For instance,
LL,RL
, respectively,
for Z and magnetic (chromo-magnetic) dipole penguins RCLL,RL and RM
(g)
we have

(0)
|Vi1 |2 PC (xi )
RCLL =
i=1,2

(1)

(2)
2
2 1
Ui1 Vi1 Uj 1 Vj 1 PC (xi , xj ) + |Vi1 | |Vj 1 |
PC (xi , xj ) ,
+
8
i,j =1,2
1
(0)
RCRL =
Vi2 Vi1 PC (xi )
2
i=1,2

(2)
(1)
Vj2 Vi1 Ui1 Uj1 PC (xi , xj ) + Vi1 Vj 1 PC (xi , xj ) ,
i,j =1,2
LL
RM
=
,g

i
RL
RM
,g
LL
|Vi1 |2 xW i PM
(xi ) Yb
,g

i
Vi1 Ui2 xW i
mi LR
P
(xi ),
mb M ,g
LL
Vi1 Vi2 xW i PM
(xi ),
,g
(17)
2 , xi = m
2 /m2i , and xij = m2i /m2j . The loop funcwhere xW i = m2W /m2i , xi = m2i /m
(1,2)
LL(LR)
tions PC (x, y), PM ,g (x) are provided in Appendix C. Finally, U and V are the
matrices that diagonalize the chargino mass matrix, defined in Eq. (13).
It is worth mentioning that the large effects of chargino contributions to C7 and C8g
LL and R LL , respectively, which are enhanced by m /m in
come from the terms in RM
i
b
Mg
Eq. (17). However, these terms are also multiplied by the Yukawa bottom Yb , which leads
to enhancing the coefficients of the LL mass insertion in C7 and C8g at large tan .
As we will see later on, this effect will play a crucial role in chargino contributions to
B ( )K decays at large tan .
We will also consider the case in which the mass of stop-right (mtR ) is lighter than other
squarks. In this case the functional form of the RF remains unchanged, while only the
expressions of RFRL should be modified in the way described in Appendix B.
Now let us turn to the gluino contributions in the b s transition. In the super-CKM
basis, the quarksquarkgluino interaction is given by:
Ld dg =

A
d PL g A dR d PR g A dL + h.c. ,
2gs T
(18)
where g A are the gluino Majorana fields, dR,L the squark fields, T A are the SU(3)c generators, and , are color indices. The dominant gluino contributions are due to the QCD
penguin diagrams, and the magnetic and chromo-magnetic dipole operators. At the first
order in MIA, the penguin and box diagrams are shown in Figs. 3 and 4, respectively. Performing the matching, the gluino contributions to the corresponding Wilson coefficients at
150
Fig. 3. Penguin diagrams for B = 1 transitions with gluino exchanges at the first order in mass insertion, where
h, k = {L, R}.
Fig. 4. Box diagrams for B = 1 transitions with gluino exchanges at the first order in mass insertion, where
q = {b, c, s, d, u} and h, k, m = {L, R}.
SUSY scale are given by [49]6

g
C3

d
1
5
1
1
LL 23 B1 (x) B2 (x) P1 (x) P2 (x) ,
=
9
9
18
2
2 2GF m2q
s2
6 Note that the Wilson coefficients of Eq. (19) are different from those reported in Ref. [49] by a minus sign and
a rescaling factor. This is due to the different convention for the Wilson coefficients in the effective Hamiltonian
of Eq. (7).
151

d
7
2
1
1
3
g
B
B
P
P
C4 = s 2 LL
(x)
+
(x)
+
(x)
+
(x)
,
1
2
1
2
23
3
3
6
2
2 2GF mq

d 10
s2
1
1
1
g
B1 (x) + B2 (x) P1 (x) P2 (x) ,
C5 =
LL 23
9
18
18
2
2 2GF m2q

2
d
2
7
1
3
g
C6 = s 2 LL
(x)
+
(x)
+
(x)
+
(x)
,
B
B
P
P
1
2
1
2
23
3
6
6
2
2 2GF mq

d
d mg
8s
g
C7 =
M
(x)
+
M
(x)
,
1
LL 23 3
LR 23
mb
9 2GF m2q

d
1
s
g
M
C8g =
(x)
+
3M
(x)
3
4
LL 23
3
2GF m2q

d mg 1
M
(x)
+
3M
(x)
,
+ LR
1
3
23 m
b 3
(19)
d ) . The functions
where C i,8g are obtained from Ci,8g by exchanging L R in (AB
23
appearing in these expressions can be found in Appendix C, with x = m2g /m2q .
g
Now we would like to comment about the chiral enhancement in C7 ,8g due to the
d
d )
(LR )23 mass insertion. As for chargino in Eq. (17), the term proportional to (LR
23 in
g
C7 ,8g in Eq. (19), has also the large enhancement factor mg /mb in front. Moreover, contrary to the chargino case, this term is not suppressed by the bottom Yukawa coupling. As
we will see later on, this enhancement factor will be responsible for the dominant gluino
effects in B ( )K decays.
In concluding this section, we emphasize that in the SM the b s transition process
is dominated by the top quark mediated penguin diagram, which does not include any
CP violating phase. Therefore, this process, as observed at the B-factories, opens up the
possibility to probe virtual effects from new sources of flavor structure and CP violation.
The SUSY contributions through gluino and chargino exchanges are independent. The ones
d ) ,
from gluino depend on the flavor structure of the down squark sector, namely (AB
23
u )
while the other ones from chargino depend on the up squark sector, particularly (AB
32
u ) . So, depending on the constraints imposed on the flavor structure of the down
and (AB
31
or up sector (for instance, from b s decay, KK and BB mixing), gluino or chargino
exchanges could give sizeable effects. As known, in many SUSY scenarios the lighter
chargino is expected to be one of the lightest supersymmetric particles. Thus, it could
contribute significantly in the one-loop processes. However, even though gluino in most
models is expected to be heavier than chargino, it is a strongly interacting particle, and
may give the dominant effect as well.
3. B ( )K in QCD factorization approach
The calculation of B ( )K decays involves the evaluation of the hadronic matrix
elements of related operators in the effective Hamiltonian, which is the most uncertain part
152
of this calculation. In the limit in which mb

QCD and neglecting QCD corrections in
s , the hadronic matrix elements of B meson decays in two mesons can be factorized, for
example, for B M1 M2 , in the form

M1 M2 |Qi B 0 = M1 |j1 B 0 M2 |j2 |0,
(20)
where M1,2 indicates two generic mesons, Qi are local four fermion operators of the effective Hamiltonian in Eq. (7), and j1,2 represent bilinear quark currents. Then, the final
results can be usually parametrized by the product of the decay constants and the transition form factors. This approach is known as naive factorization (NF) [41,42]. Then, the
hadronic matrix elements for B K are given by [42],

K 0 Q1 B 0 = 0,
4
K 0 Q4 B 0 = X,
3
0 0 3
K Q7 B = es X,
2
0
0

K Q10 B = es X,

K 0 Q2 B 0 = 0,
4
K 0 Q3 B 0 = X,
3
0 0
0 0 1
K Q5 B = X,
K Q6 B = X,
3
0 0 1
0 0
K Q8 B = es X,
K Q9 B = 3es X,
2

0
(21)
K Q7 B 0 = 0,
where for color number Nc = 3 one gets

X = 2F+BK m2 f m (pK ).
(22)
In Eq. (22), m is the meson mass, F+BK (m2 ) is the transition form factor evaluated
at transfered momentum of the order of m scale, pK stands for K momentum, and is
the polarization vector. In our analysis, for the parameters above, we will use the central
values m = 1.02 GeV, f = 0.233 GeV, and F+BK = 0.35 0.05, and for the scalar
product

2
1 2
mB
(pK ) =
mB m2K + m2
m2 13 GeV.
m
2mB
Since the evaluation of the matrix element of Q8g goes beyond the standard NF approach, we report here its result as given in [42]

0

s mb
1

K Q8g B 0 =
(23)
Q4 + Q6 Q3 + Q5 ,
4 q 2
3
where Qi K 0 |Qi |B 0 . In the derivation of Eq. (23), the following assumption was
made [42]
p
q = q2 b .
(24)
mb
The momentum q appearing in Eq. (24), is connected to the virtual gluon in the Q8g
operator, and it is given by q = pb ps , with pb(s) the momentum of b(s) quark, and
q 2 is an averaged value of q 2 . It has been shown that the physical range of q 2 in B
153

KS is m2b /4 < q 2 < m2b /2 [54]. Notice that the term 1/ q 2 in Eq. (23) has origin
from the propagator of the virtual gluon exchange between Q8g and external sources.
Here we stress that in the SM, the large uncertainty on q 2 does not necessarily convert
in a large uncertainty on the B K amplitude, since the contribution of Q8g is not
the dominant source in the SM. On the contrary, q 2 could cause a large uncertainty in
the numerical analysis of SUSY models. Indeed, as we will show in the next section, in
most relevant SUSY scenarios, C8g provides the dominant source to B K amplitude.
Moreover, the NF approach suffices a serious problem, namely, the decay amplitude in this
approximation is not scale independent. The hadronic matrix elements cannot compensate
for the scale dependence of the Wilson coefficients. This might be a hint for the necessity
of including higher order QCD corrections to the hadronic matrix elements. Furthermore,
due to the approximations used in NF, one cannot predict the direct CP asymmetries due to
the assumption of no strong re-scattering in the final state, thus leaving undetermined the
predictions of strong phases.
Recently, in the framework of QCD and heavy quark effective theory, a more consistent method for the determination of non-leptonic B meson decays has been developed
[43]. In this approach, called QCD factorization (QCDF), the hadronic matrix elements
can be computed from the first principles by means of perturbative QCD and QCD /mb
expansions. Final results can be simply expressed in terms of form factors and meson lightcone distribution amplitudes. Then, the usual NF is recovered only in the limit in which
QCD /mb and s corrections are set to zero. A nice feature of QCDF is that the strong
phases of non-leptonic two body decays can be predicted.
In QCDF the hadronic matrix element for B MK with M = , in the heavy quark
limit mb
QCD can be written as [43]

QCD
n
,
MK|Qi |BQCDF = MK|Qi |BNF 1 +
(25)
rn S + O
mb
n
where MK|Qi |BNF denotes the NF results. The second and third term in the bracket
represent the radiative corrections in S and QCD /mb . Notice that, even though at higher
order in s the simple factorization is broken, these corrections can be calculated systematically in terms of short-distance coefficients and meson light-cone distribution functions.
Now we briefly recall the main results of this method [43,44]. In QCDF the decay
amplitudes of B ( )K can be expressed as

A B ( )K = Af B ( )K + Aa B ( )K ,
(26)
where
10

GF
( )
Af B ( )K =
( )K Qi |BNF ,
Vpb Vps
ai
2 p=u,c 1=1
(27)
10

GF

( )
A B ( )K = fB fK f
Vpb Vps
bi
.
2
p=u,c i=1
(28)
and
a
154
The first term Af (B ( )K) includes vertex corrections, penguin corrections and hard
( )
spectator scattering contributions which are involved in the parameters ai
. The other
term Aa (B ( )K) includes the weak annihilation contributions which are absorbed
( )
. However, these contributions contain infrared divergences, and
in the parameters bi
the subtractions of these divergences are usually parametrized as [44]
1

dx
mB
iH,A
,
XH,A 1 + H,A e
ln
x
QCD
(29)
where H,A are free parameters expected to be of order of H,A O(1), and H,A
[0, 2]. As already discussed in Ref. [44], if one does not require fine tuning of the annihilation phase A , the A parameter gets an upper bound from measurements on branching
ratios, which is of order of A 2. Clearly, large values of A are still possible, but in this
case strong fine tuning in the phase A is required. However, assumptions of very large
values of H,A , which implicitly means large contributions from hard scattering and weak
annihilation diagrams, seem to be quite unrealistic.
Notice that the annihilation topology contribution due to the effective operator Q8g
has not yet been calculated, and it is expected to have as well a logarithmic divergence.
This effect should increase the theoretical uncertainty substantially specially in models
like supersymmetric ones, where the Q8g plays a crucial rule in the B K decay.
Following the scheme and the notation of Ref. [44] we write the decay amplitude of
B K as:
1 p
1 p
p
p
p
p 3 + 4 3,EW 4,EW + 3
2
2
p=u,c

1 p
1 p
p
3,EW + S3 S3,EW ,
2
2
A(B K) = iAK,
and A
where p = Vpb Vps
K, =
p
G
BK .
F m2 f F+
2 B
(30)
p
The quantities i i (K, ) and
i i (K, ) depend,7 in addition to the leading contribution of the b s transition, on

the one-loop vertex corrections, hard spectator interactions and penguin corrections [44].
As mentioned in Section 2, NP effects are parametrized in the Wilson coefficients, while
p
p
all the other functions involved in the definition of i and i depend on some theoretical
input parameters like the QCD scale QCD , the value of the running masses, and parameters of vector-meson distribution amplitudes. Therefore it would be very useful for future
analyses involving any NP scenarios, to provide a numerical parametrization of Eq. (30)
in terms of the Wilson coefficients at low energy. In this respect it is very convenient to
i according to the parametrization of the effective
define new Wilson coefficients Ci and C
7 Notice that in the notation of Ref. [44], the order of the arguments in the functions p (M , M ) and
1
2
i
p
i (M1 , M2 ) is fixed and it is determined by the order of the arguments (M1 , M2 ) in the pre-factor AM1 ,M2 ,
where M1,2 labels the final states. In Eq. (30) this order corresponds to M1 = K, M2 = .
155
Hamiltonian in Eq. (7)

GF
B=1
i Q i },
Heff
=
{Ci Qi + C
2 i
(31)
i are the same ones of Eq. (8).8 Fixing the experimental

where the operators basis Qi and Q
and SM parameters to their center values as given in Table 1 of Ref. [44], we can present
the explicit dependence of the decay amplitude on the Wilson coefficients relevant for this
process. In particular, for

GF
i ),
A(B K) = i m2B F+BK f
Hi ()(Ci + C
2
i=1...10,7 ,8g
(32)
we obtain
H1 () 0.0002 0.0002i,
H2 () 0.011 + 0.009i,
2
H3 () 1.23 + 0.089i 0.005XA 0.0006XA
0.013XH ,
H4 () 1.17 + 0.13i 0.014XH ,

2
H5 () 1.03 + 0.053i + 0.086XA 0.008XA
,
2
H6 () 0.29 0.022i + 0.028XA 0.024XA
+ 0.014XH ,
2
H7 () 0.52 0.026i 0.006XA + 0.004XA
,
2
H8 () 0.18 + 0.037i 0.019XA + 0.012XA
0.007XH ,
2
H9 () 0.62 0.037i + 0.003XA + 0.0003XA
+ 0.007XH ,
H10 () 0.62 0.037i + 0.007XH ,

H7 () 0.0004,
H8g () 0.047.
(33)
Notice that here both H7 and H8g do not have any dependence on XH,A , since, as we
mentioned, the hard scattering and weak annihilation contributions to Q7 and Q8g have
been ignored.
As can be seen from expressions in Eq. (33), terms proportional to XA,H represent
always small corrections for typical values of XH,A O(1). Moreover, if we set XA,H to
zero, the contribution to the amplitude is (incidentally) quite close to the corresponding one
in NF for q 2 = m2b /4. Remarkably, the strong phases appearing in the terms independent
on XH,A in Hi () are also negligible. Thus, in the limit XH,A 0, the NF result (where
strong phases are assumed to vanish) seems to be recovered for the choice q 2 = m2b /4.
Note that largest corrections in XA are contained in the H6 () term. In particular, for
2 in
values of A = 1 and A 0, we have XA 4 and so the term proportional to XA
H6 () becomes of the same order as the other term, which is independent on XA . Therefore, it is expected that the effect of the weak annihilation parameter would be important
when contributions to C6 becomes large. Clearly, this last consideration applies in a NP
8 Notice that in the notation of [44], the operators Q and Q correspond to our operators Q and Q ,
1
2
2
1
respectively.
156
context, where the new contribution to Wilson coefficients Ci could sizeably differ from
the corresponding SM ones.
Finally we would like to comment about the fact that contributions from C i and Ci to
the decay amplitude A(B K) are identically the same (with the same sign). This can
be simply understood by noticing that
0 0 0 0
i B
K Qi B = K Q
(34)
which is due to the invariance of strong interactions under parity transformations, and to
the fact that initial and final states have same parity. Indeed only terms proportional to
the chiral structure V V or A A in the four fermion operators of Qi basis contribute
to the matrix elements, and consequently property in Eq. (34) easily follows. Analogous
considerations apply to Q7 and Q8g as well.
Now we turn to the B K process. As known, the physicalstate of and are the
+ |d d)/
2 components:
mixture of SU(3) singlet s = |s s and octet q = (|uu

|
cos sin
|q
(35)
=
.
|
sin
cos
|s
q
Therefore, we have the following decay constants: f = fq sin and fs = fs sin and
from a fit to experimental data one finds [44]: fq = (1.07 0.02)f , fs = (1.34 0.06)f
and = 39.3 1.0.
In the NF approach, the hadronic matrix elements of the B K process are given by
[9,42]
1
0 0
K 0 Q1 B 0 = X2 ,
K Q2 B = X2 ,
3
0 0 1
0 0
4
2
4
K Q3 B = X1 + 2X2 + X3 ,
K Q4 B = X1 + X2 + X3 ,
3
3
3
3

0 0 R1
R
2
X1 2X2 1
X3 ,
K Q5 B =
3
3

0 0
2
1
R2 X 3 ,
K Q6 B = R1 X1 X2
3
3

0 0 1
R1 X 1
R2
X2 + 1
X3 ,
K Q7 B =
2
3
3

0 0 1
1
X2
K Q8 B = R1 X1
+
R2 X 3 ,
2
3
3

0 0 1
X1
4
K Q9 B =
+ X2 X3 ,
2
3
3

0 1
0
X2 4
K Q10 B = X1 +
(36)
X3 ,
2
3
3

with

X

X1 = m2B m2 F1B m2K fK ,
2
157

X
X2 = m2B m2K F1BK m2 f ,
2

2
X3 = mB m2K F1BK m2 2fK2 f2 Y ,
R1 =
2m2K
,
(mb ms )(ms + md )
R2 =
(2m2K m2 )
,
(mb ms )ms
where F1B (q 2 ) = 0.35 is the B transition form factor evaluated at q 2 scale, and
fK() = 0.16(0.13) GeV is the decay constant of K() meson. X = 0.57 and Y = 0.82,
which correspond to p = 20 , represent the rate of the uu + d d and s s component in
the . These matrix elements show that the B K process receives a small contribution
from color suppressed b uus
tree diagram, in addition to the b s qq
(q = u, d, s)
penguin diagrams. As in case of B K decay, for the matrix element of the chromomagnetic operator we have

0

0
s mb
1

Q4 + Q6 Q3 + Q5 .
K Q8g B =
(37)
4 q 2
3
In the QCDF approach, the decay amplitude of B K is given by [44,45]
A(B K)

1 p
i
p
p
p
p pu 2 + 23 + 3,EW + 2S3 S3,EW
= AK,q
2
2
p=u,c

1 p
1 p
1 p
1 p
p
p
p
p
+ iAK,s 3 + 4 3,EW 4,EW + 3 3,EW + S3 S3,EW
2
2
2
2

i
1
1
p
p
p
p
p
+ iAK,c pc 2 + 3 + Aq ,K 4 4,EW + 3 3,EW , (38)
2
2
2
where9
GF
q
AK,q = m2B F BK f
2
GF
AK,s = m2B F BK fs
2
GF 2 BK c
AK,c = mB F
f
2
GF

Aq ,K = m2B F B fK
2
6.8 106 GeV,

1.05 105 GeV,
2.3 107 GeV,
7.0 106 GeV,
(39)
where the third contribution is negative due to the definition of negative decay constant fc
in [45]. As in the case of B K, we will provide below the numerical parametrization
9 Same notation as in A(B K) has been adopted here for p and p , where p p (M , M ),
1
2
i
i
i
i
p
p
p
p
p
p
i i (M1 , M2 ). The expressions for i (K, ), i ( , K), i (K, ), and i ( , K) can be found in
Refs. [44,45].
158
in terms of the Wilson coefficients defined in Eq. (31). By fixing the hadronic parameters
with their center values as in Table 1 of Ref. [44], we obtain

GF
i ),
Hi ( )(Ci C
A(B K) = i m2B F+BK fs
(40)
2
i=1...10,7 ,8g
where
H1 ( ) 0.44 + 0.0005i,
H2 ( ) 0.076 0.064i + 0.006XH ,
2
+ 0.014XH ,
H3 ( ) 2.23 0.15i + 0.009XA + 0.0008XA
H4 ( ) 1.76 0.29i + 0.026XH ,

2
,
H5 ( ) 1.52 + 0.004XA + 0.008XA
2
H6 ( ) 0.54 0.29i + 0.006XA + 0.027XA
+ 0.026XH ,
2
H7 ( ) 0.078 + 0.001XA 0.004XA
,
2
H8 ( ) 0.58 + 0.02i + 0.004XA 0.014XA
0.004XH ,
2
H9 ( ) 0.44 + 0.054i + 0.005XA 0.0004XA
0.007XH ,
H10 ( ) 0.80 + 0.02i 0.004XH ,

H7 ( ) 0.0007,
H8g ( ) 0.089.
(41)
i appearing in Eq. (40) is due to the fact that, contrary

The sign difference between Ci and C
to the B K transition, initial and final states have here opposite parity. Thus, due to
the invariance of strong interactions under parity transformations, only V A or A V
structures of four-fermion operators will contribute to the hadronic matrix elements, and
so
K|Qi |B = K|Q i |B.
(42)
The comparison between the coefficients Hi () and Hi ( ) in Eq. (41) tells us that,
apart from the sign difference of the Wilson coefficients C i in the amplitudes of B K
and B K, in QCDF these amplitudes are expected to be different, unlike the case of
NF as discussed in Ref. [9]. However, notice that the values of Hi () and Hi () are quite
sensitive to the scale where they have been evaluated. Our results in Eqs. (33), (41)
pole
correspond to the choice = mb = 4.5 GeV.
4. CP asymmetry of B ( )KS : gluino/chargino
Here we analyze the supersymmetric contributions to the time dependent CP asymmetries in B KS and B KS decays in the framework of mass insertion approximation, in gluino and chargino dominated scenarios.
New physics could in principle affect the B meson decay by means of a new source of
CP violating phase in the corresponding amplitude. In general this phase is different from
the corresponding SM one. If so, then deviations on CP asymmetries from SM expectations can be sizeable, depending on the relative magnitude of SM and NP amplitudes. For
159
instance, in the SM the B KS decay amplitude is generated at one loop and therefore it
is very sensitive to NP contributions. In this respect, SUSY models with non-minimal flavor structure and new CP violating phases in the squark mass matrices, can easily generate
large deviations in the B KS asymmetry.
As mentioned in the introduction, the time dependent CP asymmetry for B KS can
be described by
(B 0 (t) KS ) (B(t) KS )
(B 0 (t) KS ) + (B(t) KS )
= CKS cos MBd t + SKS sin MBd t,
aKS (t) =
(43)
where CKS and SKS represent the direct and the mixing CP asymmetry, respectively, and
they are given by
CKS
2
|(K
S )| 1
,
=
2
|(K
S )| + 1
SKS =
2 Im[ pq (K
S )]
2
|(K
S )| + 1
(44)
The parameter (K
S ) is defined by
(K
S) =
A(K
S)
,
A(KS )
(45)
0 and B 0 mesons, respectively.

where A(K
S ) and A(KS ) are the decay amplitudes of B
Here, the mixing parameters p and q are defined by |B1 = p|B + q|B 0 , |B2 = p|B
q|B 0 where |B1(2) are mass eigenstates of B meson. The ratio of the mixing parameters
is given by
V Vtd
q
= e2id tb ,
p
Vtb Vtd
(46)
where d represent any SUSY contribution to the BB 0 mixing angle. Finally, the above
amplitudes can be written in terms of the matrix element of the B = 1 transition as

B=1 0
B=1 0
B .
B ,
A(K
(47)
A(KS ) = KS | Heff
S ) = KS |Heff
In order to simplify our analysis, it is useful to parametrize the SUSY effects by introducing the ratio of SM and SUSY amplitudes as follows
SUSY
A
R ei ei ,
(48)
ASM KS
and analogously for the KS decay mode
SUSY
A
R ei ei ,
ASM KS
(49)
SUSY
where Ri stands for the corresponding absolute values of | AASM |, the angles , are the
SM SUSY parametrize the strong
corresponding SUSY CP violating phase, and , = ,

,
(CP conserving) phases. In this case, the mixing CP asymmetry SKS in Eq. (43) takes the
160
following form
SKS =
sin 2 + 2R cos sin( + 2) + R2 sin(2 + 2)
(50)
1 + 2R cos cos + R2
and analogously for B KS

S KS =
sin 2 + 2R cos sin( + 2) + R2 sin(2 + 2)

1 + 2R cos cos + R2
(51)
Assuming that the SUSY contribution to the amplitude is smaller than the SM one, i.e.,
R( ) 1, one can simplify the above expressions as:

2
S( )KS = sin 2 + 2 cos 2 sin ( ) cos ( ) R( ) + O R,(
(52)
) .
It is now clear that in order to reduce SKS smaller than sin 2, the relative sign of sin
and cos has to be negative. If one assumes that sin cos 1, then R 0.1 is
required in order to get SKS within 1 of the experimental range.
We will work in a minimal scheme of CP and flavor violation. This means that, in the
framework of MIA, we will assume a dominant effect due to one single mass insertion.
In this case the CP violating SUSY phase will coincide with the argument (module ) of
corresponding mass insertion. Moreover, in the following we will generalize this scheme
by including two (dominant) mass insertions simultaneously, but assuming that their CP
violating phases are the same. We will perform this analysis in both gluino and chargino
scenarios.10
In the following subsections we present and discuss our results separately for B KS
and B KS processes.
4.1. CP asymmetry in B KS
As shown in Eq. (3), the most recent world average measurement of the CP asymmetry
SKS , indicates a 1.7 deviation from the cc measurement. In particular,
SKS = 0.34 0.20
(53)
allows also for negative values of CP asymmetry at 2 level. In the light of these results,
we will analyze here, in a model independent way, the SUSY scenarios, which are more
favored to explain such deviations. Let us start first with gluino contributions to the CP
asymmetry in B KS .
These effects have been analyzed in Refs. [6,8,9] in the framework of NF or QCDF and
by adopting the MIA method. However, from these works a direct comparison between the
10 There are corrections to the effective Hamiltonian B = 1, mediated by both chargino and gluino box dia-
grams in which both up and down mass insertion contribute. See Refs. [11,51] for further details. Then, chargino
or gluino contributions cannot be completely disentangled by taking active only one mass insertion per time.
However, these corrections affect only the Wilson coefficients C1 and C2 in Eq. (7) and their effect is quite small
in comparison to the other SUSY contributions. In our analysis we can safely neglect them.
161
d ) ] (top) and arg[( d ) ] (bottom) with gluino contribution of one mass

Fig. 5. SKS as a function of arg[(LL
23
LR 23
d
d
insertion (LL )23 (top) and (LR )23 (down). Left and right plots correspond to NF and QCDF, respectively. The
region inside the two horizontal lines corresponds to the allowed experimental region at 2 level.
results in NF and QCDF predictions is not always possible, mainly because of different approaches used in analyzing the SUSY contributions. For this reason we reanalyze here the
gluino contributions in both NF and QCDF approaches in a unique SUSY framework, by
comparing the predictions for the CP asymmetries versus the relevant SUSY CP violating
phases. In QCDF we have included the last updated results for hard spectator scattering
and weak annihilation diagrams [44], as explained in Section 3.
We present our numerical results for the gluino contributions to CP asymmetry SKS
in Fig. 5. The left and right plots correspond to the evaluation of amplitudes by means of
NF and QCDF methods, respectively. In all the plots, regions inside the horizontal lines
indicate the allowed 2 experimental range. In the top and bottom plots only one mass ind ) |<1
sertion per time is taken active, in particular this means that we scanned over |(LL
23
d ) | < 1. Then, S
and |(LR
23
KS is plotted versus , which in the case of one dominant
d ) ].
mass insertion should be identified here as = arg[(AB
ij
We have scanned over the relevant SUSY parameter space, in this case the average
squark mass m
and gluino mass mg , assuming SM central values [55]. Moreover, we re-
162
quire that the SUSY spectra satisfy the present experimental lower mass bounds [55]. In
> 300 GeV. In addition, we impose that the branching ratio
particular, mg > 200 GeV, m
(BR)11 of b s and the BB mixing are satisfied at 95% C.L. [59], namely 2 104
d ) | and |( d ) | are obBR(b s ) < 4.5 104 . Then, the allowed ranges for |(LL
23
LR 23
tained by taken into account the above constraints on b s and BB mixing.
In the plots corresponding to QCDF, we have also scanned over the full range of the
parameters A,H and A,H in XA and XH , respectively, as defined in Eq. (29). We remind
here that XA and XH take into account the (unknown) infrared contributions in the hard
scattering and annihilation diagrams. Regarding the allowed range of , as can be seen
from results in Eq. (33), the dominant effect is due to the annihilation contributions proportional to XA . As discussed in Section 3, the total width will grow as A4 log4 (mb /) for
large A . Therefore, as we will show in Section 5, by requiring that the SUSY contribution
to the branching ratio and asymmetries of B KS is inside the 2 experimental range,
an upper bound on A of order A 2 is obtained.12
In the corresponding plots evaluated in NF, in order to maximize the SUSY contributions to CP asymmetry, we have fixed the average of gluon momenta to its minimum
value q 2 = m2b /4. We remind here that q 2 enters as free parameter in matrix element of
the chromo-magnetic operator Q8g , see Eq. (23). As we will show later on, the dominant
SUSY effect to B KS amplitude is given by the SUSY contribution to the Wilson coefficient C8g . Therefore, the smaller the q 2 value is, the larger the SUSY contribution to
the CP asymmetry can be.
In the framework of NF, the strong phase, which comes from the hadronic matrix elements, cannot be predicted. For this reason we set both SM and SUSY strong phases to
zero. Therefore, the phase in Eq. (44) can take only the values 0 or , corresponding to
the relative sign of SM and SUSY amplitudes.
By comparing the scatter plots in NF and QCDF in Fig. 5 we see that the predictions are
d ) have chances to drive
quite similar. Only the gluino contributions proportional to (LR
23
d ) just
SKS toward the region of large and negative values, while the pure effect of (LL
23
approach the negative values region.
This result can be easily understood by noticing that the dominant SUSY source to the
B KS decay amplitude is provided by the chromo-magnetic operator Q8g . In particular, as already mentioned in Section 2, gluino contributions to C8g , which are proportional
d ) , can be very large with respect to the SM ones, being enhanced by terms of
to (LR
23
order mg /mb . In addition, large gluino effects in C8g may still escape b s constraints
[58]. This is a remarkable property and it is due to the fact that in the gluino contributions
to dipole operators Q7 ,8g , the ratio of |C8g /C7 | is enhanced by color factors with respect
to typical contributions of W or chargino exchanges [48].
11 The branching ratio (BR) of b s is evaluated at the NLO in QCD, as provided in Refs. [50,56]. However,
we have not included the 2-loop threshold corrections of SUSY contributions at W scale. For these corrections
see Ref. [57] for more details.
12 We would like to stress here that in the literature the allowed range of has been sometime overestimated.
A
For instance, in the analysis of [8] the experimental upper bound on BR(B KS ) was not imposed, leaving
the possibility of larger values of A 8.
163
u ) ] (top) and arg[( u ) ] (bottom) with chargino

Fig. 6. As in Fig. 5, but for SKS as a function of arg[(LL
32
RL 32
u ) (top) and ( u ) (down).
contribution of one mass insertion (LL
32
RL 32
Now we discuss the chargino effects to SKS , which are summarized in Fig. 6. These
contributions have been first analyzed, in the framework of MIA, in Ref. [11], but only
using NF approach for evaluating the hadronic matrix elements. Here, we extend our previous analysis in [11] by including the corresponding predictions in the QCDF approach.
In Fig. 6, SKS is plotted versus the argument of the relevant chargino mass insertions,
u ) and ( u ) . Same conventions as in Fig. 5 have been adopted for left and
namely (LL
32
RL 32
right plots.
As in the gluino dominated scenario, we have scanned over the relevant SUSY parameter space, in particular, the average squark mass m,
weak gaugino mass M2 , the term, and
the light right stop mass m
tR . Also tan = 40 has been assumed and we take into account
the present experimental bounds on SUSY spectra, in particular m
> 300 GeV, the lightest
chargino mass M > 90 GeV, and m
tR 150 GeV. As in the gluino case, we scan over
u ) and ( u ) , by considering the
the real and imaginary part of the mass insertions (LL
32
RL 32
constraints on BR(b s ) and BB mixing at 95% C.L. The b s constraints impose

u ) , specially at large tan [11]. Finally, as in the other plots, we
stringent bounds on (LL
32
scanned over the QCDF free parameters A,H < 2 and 0 < A,H < 2 .
164
From these results we can see that also for the chargino dominated scenario the predicu ) ones
tions in NF and in QCDF are quite close, apart from a slight difference in the (RL
32
that we will discuss below. The main conclusion in this scenario is that negative values of
SKS cannot be achieved neither in NF nor in QCDF.
The reason why extensive regions of negative values of SKS are excluded here, is
only due to the b s constraints [11]. Indeed, as shown in our previous work [11],
u )
the inclusion of (LL
32 mass insertion can generate large and negative values of SKS ,
by means of chargino contributions to chromo-magnetic operator Q8g which are enhanced
by terms of order m /mb . However, contrary to the gluino scenario, the ratio |C8g /C7 |
is not enhanced by color factors and large contributions to C8g leave unavoidably to the
breaking of b s constraints.
u )
On the other hand, the contribution of (RL
32 is independent of tan and so large
effects in R that could drive SKS toward the region of negative values cannot be achieved.
u ) and ( u ) contributions
However, as can be seen from Fig. 6, while both in NF (RL
32
LL 32
u
are within the 2 experimental range, in NF the (RL )32 contribution fits better than in
u )
QCDF. This can be explained by the fact that (RL
32 mainly gives contribution to the
electroweak operators, whose matrix elements are more sensitive, with respect to the other
operators, to the approach adopted for their evaluation.
As shown in our previous work [11], by scanning over the two relevant mass insertions
u ) and ( u ) , the b s constraints on ( u ) are a bit more relaxed, but a large
(RL
32
LL 32
LL 32
amount of fine tuning between SUSY parameters is necessary if small values of SKS are
required.
In order to understand the behavior of these results, it is useful to look at the numerical
(BKS )
in terms of the relevant
parametrization of the ratio of amplitudes R AASUSY
SM (BKS )
mass insertions. Below, we present numerical results for R in both NF and QCDF. In
QCDF we set to zero the effect of annihilation and hard scattering diagrams, corresponding
to the choice of A,H = 0 and A,H = . In this case, we expect QCDF predictions to be
quite close to the NF ones. In particular, for a gluino mass and average squark mass of
order m
= mg = 500 GeV, we obtain

d
d
R |NF
g 0.08 LL 23 120 LR 23 + {L R},
d
d

+ {L R},
R |QCDF
(54)
0.14 ei0.1 LL
127 ei0.08 LR
g
23
23
while in the case of chargino, by using gaugino mass M2 = 200 GeV, = 300 GeV, m
tR =
150 GeV, and tan = 40, we have
u
u
u
u
R |NF
1.83 LL 32 0.32 RL 32 + 0.41 LL 31 0.07 RL 31 ,
u
u
1.89 ei0.07 LL
0.11 ei0.17 RL
R |QCDF
32
32
u

i0.17 u
+ 0.43 ei0.07 LL
0.02
e
(55)
RL 31 ,
31
where the first symbol R |NF
g means that the corresponding quantity R has been calculated in NF approach including only gluino contributions. Analogously for the other cases
of QCDF and for chargino ( ) exchanges.
From results in Eqs. (54), (55), it is clear that the largest SUSY effect is provided by
the gluino and chargino contributions to the chromo-magnetic operator which are propor-
165
d ) and ( u ) , respectively. However, the b s constraints play a crucial

tional to (LR
23
LL 32
role in this case. For the above SUSY configurations, the b s decay set the following
d ) | < 0.019 and
(conservative) constraints on gluino and chargino contributions: |(LR
23
u
|(LL )32 | < 0.20. Implementing these bounds in Eqs. (54), (55), we see that gluino contribution can easily achieve larger values of R = |R | (see Eqs. (48), (50)) than chargino
one, and this is the main reason why extensive regions with negative value of SKS are
favored and disfavored in Figs. 5 and 6, respectively.
4.2. CP asymmetry in B KS decay

Recent measurements of CP asymmetry in B KS show another discrepancy with
SM predictions. In particular from Eq. (4), the world average is
S KS = 0.41 0.11,
(56)
and which is about 2.5 deviation from SM expectations. From these results we see that,
similarly to what happens in B KS decay, large deviations from SM are possible.
Since SUSY contributes to both CP asymmetries with the same CP violating source, it is
possible that the SUSY effects driving SKS towards negative values, could also sizeably
decrease S KS . The main reason for that is because the leading SUSY contributions to
the amplitudes of B KS and B KS enter through the Wilson coefficient of C8g
and the operator Q8g has a comparable matrix elements in both processes. However, since
NP corrections enter through the quantity R , the role of the SM contribution will be
crucial. Indeed, while the B KS amplitude is purely generated at one loop in the
SM, the B KS one receives tree-level contribution from the SM by means of nonvanishing matrix element of Q2 . Therefore, the increase of SUSY contributions to C8g is
now compensated in R , by the large SM amplitude contribution.
We show our results for gluino in Fig. 7, where we have just extended the same analysis
of B KS . Same conventions as in figures for B KS have been adopted here. As we
can see from these results, there is a depletion of the gluino contribution in S KS , precisely
for the reasons explained above. Regions of negative values of S KS are more disfavored
with respect to SKS , but a minimum of S KS 0 can be easily achieved. Comparing NF
and QCDF, we can also see that SUSY predictions in NF and QCDF are very close.
Finally, in Fig. 8 we present our results for chargino contributions. Here we see that
charginos can produce at most a deviation from SM predictions of about 20%, and the
u ) . These results again show the relevant role
most conspicuous effect is achieved by (LL
32
played by the chromo-magnetic operator.
As in the case of B KS decay, we present below the parametrization of R
ASUSY (B KS )
ASM (B KS ) , by using the same SUSY inputs adopted in Eqs. (55), (54). For gluino
contributions we have

d
d
R |NF
g 0.08 LL 23 79 LR 23 {L R},
d

d
{L R},
0.07 ei0.24 LL
64 LR
R |QCDF
g
23
23
(57)
166
Fig. 7. As in Fig. 5, but for S KS .
while for chargino exchanges we obtain

u
u
u
u
R |NF
1.2 LL 32 0.01 RL 32 + 0.27 LL 31 0.002 RL 31 ,
u
u
0.95 LL
0.025 ei0.19 RL
R |QCDF
32
32

u
i0.19 u
+ 0.21 LL
0.006
e
RL 31 .
31
(58)
Notice that in the second curly brackets in Eq. (57) there is a minus sign in front. This
takes into account for the minus sign in the matrix elements of Q i operators as shown in
Eq. (42).
d )
In both cases of gluino and chargino contributions, we see that the coefficient of (LR
23
u
and (LL )32 mass insertions is smaller in comparison than the corresponding ones in B
KS , see Eqs. (54), (55). As explained before, this depletion comes from the fact that the
SM amplitude of A(B KS ) receives contribution from tree-level W exchanges and it
is larger than A(B KS ).
This general behavior is going in the right direction to explain the experimental data.
In order to show better this effect, we plotted in Fig. 9 the correlations between SKS versus S KS for both chargino (left plot) and gluino (right plot) in QCDF. For illustrative
167
Fig. 8. As in Fig. 6, but for S KS .
u ) (left)
Fig. 9. Correlation of asymmetries SKS versus S KS with the contribution of one mass insertion (RL
32
d
and (LR )23 (right), for chargino (left) and gluino (right) exchanges. Region inside the ellipse corresponds to the
allowed experimental ranges at 2 level.
168
d ) ] = arg[( d ) ] (left) and arg[( d ) ] = arg[( u ) ] (right), with

Fig. 10. As in Fig. 9, but for arg[(LR
23
RL 23
LR 23
LL 32
d ) & ( d ) (left) for gluino exchanges, and ( d ) & ( u )
the contribution of two mass insertions (LR
23
RL 23
LR 23
LL 32
(right) for both gluino and chargino exchanges.
Fig. 11. As in Fig. 9, but taking into account a charged Higgs exchange with mass mH = 200 GeV and tan = 40.
purposes, in all figures analyzing correlations, we colored the area of the ellipse corresponding to the allowed experimental range at 2 level.13 In Fig. 9 we scanned over the
u ) (left plot)
real and imaginary part of the single relevant mass insertions, namely (RL
32
d
and (LR )23 (right plot) for chargino and gluino, respectively.
In conclusion, as can be seen from the results in Fig. 9, pure chargino exchanges have
no chance to fit data at 2 level, while gluino can fit them quite well. At this point we
want to stress that sizeable chargino contributions to the CP asymmetries, in particular
u )
from (LL
32 mass insertion, are ruled out by b s constraints. Therefore, it would
be interesting to see if the effects of a light charged Higgs exchange could relax these
constraints allowing chargino contributions to fit inside the experimental ranges. For this
purpose, we present in Figs. 1113 the impact of a light charged Higgs in both chargino
13 All ellipses here have axes of length 4 . As a first approximation, no correlation between the expectation
values of the two observables have been assumed.
169
d ) . In the right plot the effect

Fig. 12. As in Fig. 9, but for gluino contributions with single mass insertion (LL
23
of a charged Higgs exchange, with mass mH = 200 GeV and tan = 40, has been taken into account.
u ) .
Fig. 13. As in Fig. 12, but for chargino contribution with single mass insertion (LL
32
and gluino exchanges. In particular, plots in Fig. 9 should be directly compared with the
corresponding ones in Fig. 11, where a charged Higgs with mass mH = 200 GeV and
tan = 40 has been taken into account. From these results we can see that the effects of
u ) mass insertion are negligible, as we expect
charged Higgs exchange in the case of (RL
32
u ) in b s and b sg amplitudes are not
from the fact that terms proportional to (RL
32
d ) or ( d ) (see
enhanced by tan . On the other hand, in gluino exchanges with (LR
23
LL 23
Figs. 11, 12), the most conspicuous effect of charged Higgs contribution is in populating
the area outside the allowed experimental region. This is clearly due to a destructive interference with b s amplitude, which relaxes the b s constraints. The most relevant
u )
effect of a charged Higgs exchange is in the scenario of chargino exchanges with (LL
32
mass insertion. In this case, as can be seen from Fig. 13, a strong destructive interference
with b s amplitude can relax the b s constraints in the right direction, allowing
chargino predictions to fit inside the experimental region. Moreover, we have checked that,
170
for tan = 40, charged Higgs heavier than approximately 600 GeV cannot affect the CP
asymmetries significantly.
In a recent paper [9], it has been shown that there exists a particular scenario in which
gluino contribution can sizeably decrease SKS providing on the same time a very modest
d )
d
effect in S KS . This can be achieved if one assume that both (LR
23 and (RL )23 mass
insertions are of the same order, including their CP violating phase. In this case gluino
i operators in Eq. (7), where Q i are the
contributes with the same weight to both Qi and Q
operators with opposite chirality. Due to the different parity of the final states, as explained
in Section 3, the contribution of the corresponding Wilson coefficients to the amplitude of
B KS and B KS enter in combination as Ci + C i and Ci C i , respectively.
However, the analysis of Ref. [9] was based on the NF approach, and one may wonder,
if their results still hold in QCDF. For this reason we repeat here the same analysis, but in
QCDF and scanning over the strong CP phases A,H and A,H < 2. Results of this scenario
are shown in Fig. 10 (left plot), where we scanned over two mass insertions simultaneously,
d )
d
namely (LR
23 and (RL )23 , by assuming that their CP violating phases are the same.
From these results we can see that a large number of SUSY configuration fitting inside the
2 experimental region, can be obtained.
We have also considered another scenario in which both chargino and gluino exchanges
d )
are assumed to contribute simultaneously with relevant mass insertions, namely (LR
23
u
and (LL )32 . We plot the corresponding results for the correlations between SKS versus
S KS in the right plot of Fig. 10. As in the case of Fig. 9, we assume a common SUSY CP
violating phase between the two mass insertions. From these results we can see that also
in this case a large number of configurations can fit inside the 2 experimental regions.
u )
This result shows a remarkable fact. The stringent bounds on (LL
32 from the experimental limits on BR(B Xs ) are relaxed when one considers both gluino and chargino
contributions, which come with different sign. This does not happen, if two chargino mass
insertions are considered. Then some configurations with large tan are allowed and therefore chargino can contribute significantly to the CP asymmetries SKS and S KS .
Finally, we would like to comment about the stability of SUSY predictions for SKS
against the low renormalization scale b , where Wilson coefficients are evaluated. In NF
the scale dependence in the Wilson coefficients is not compensated by the corresponding
one in the matrix elements, so a large uncertainty is expected. However, we have noticed
that also in QCDF this uncertainty still persists. In particular the coefficients in the parametrizations in QCDF in Eqs. (54), (55) can vary up to 3040% when the b scale is changed
from b = mb /2 to b = mb . All the numerical results in this paper correspond to the
pole
choice b = mb = 4.5 GeV. In both NF and QCDF, this residual scale dependence in
R is mainly due to the scale dependence in the SM amplitude and in the C8g (b ) contribution to the SUSY amplitude. However, we noticed that the main sensitivity to the scale
in R comes from the SM contribution. On the other hand, the main source of sensitivity
in R comes from C8g (b ) in the SUSY amplitude, since the SM one, receiving tree-level
contributions, is less sensitive to the renormalization scale.
In the next section we are going to consider the correlations of SKS and S KS asymmetries versus the corresponding branching ratios of B KS and B KS .
171
5. SUSY contributions to BR(B ( )K)

In this section we discuss the impact of large SUSY contribution, which is required
to explain the deviations from SM in CP asymmetries of S( )KS , on the branching ratios
(BR) of B K and B K. As shown in Eqs. (5), (6), the experimental measurements
of these BRs at BaBar, Belle, and CLEO lead to the following averaged results [25]:

6
BR(B K) = 8.3+1.2
(59)
1.0 10 ,

6
BR(B K) = 65.2+6.0
(60)
5.9 10 ,
which mean that BR(B K) is in good agreement with SM predictions, while the experimental value of BR(B K) is two to five times larger than the SM one. Since these
two processes are highly correlated and both are based on the b s transition, it seems a
challenge for the SUSY contribution to enhance BR(B K) by a factor of two or more,
while leaving other BRs and asymmetries inside their experimental ranges.
The branching ratio of B MK, with M = or , can be written in terms of the
corresponding amplitude MM as
BR(B MK) =
1 |P |
1
|MM |2
,
2
8 MB
tot
(61)
B=1 |B
0 and
where MM = M K 0 |Heff
|P | =
[(MB2 (mK + mM )2 )(MB2 (mK mM )2 )]1/2

.
2MB
The inclusion of SUSY corrections modifies the BR as

2
BRtotal (B MK) = BRSM (B MK) 1 + 2 cos(M M )RM + RM
SUSY
(62)
(63)
SUSY
(BMK)
(BMK)
where RM = | AASM (BMK)
|, M = arg[ AASM (BMK)
] and M is the corresponding
strong phase. The input parameters that we have used in the previous section with A,H = 1
and A,H = , i.e., XA,H = 0 lead to BRSM (B K) = 2.76 106 and BRSM (B
K) = 11.1 106 .
It is remarkable that in order to produce the experimental result of SKS by SUSY
SUSY
(BKS )
] has to be around /2 and the strong
contribution, the phase = arg[ AASM (BK
S)
phase should be of order or 0, as shown in Section 4. These particular values for
the phases suppress the leading term in Eq. (63). Therefore, for typical scenarios in which
R 1, the SUSY contributions do not enhance BR(B K) much. At most a total
branching ratio of order 3 BRSM (B K) 8.3 106 can be achieved, as we will
explain below, which is still compatible with the experimental measurements. Clearly, if
R
1 then the total branching ratio would exceed the experimental bound in Eq. (59).
However, as shown in Section 3, the inclusion of b s constraints moderates possible
large SUSY contributions to R .
Next we briefly discuss the dependence of the branching ratio on the QCDF free parameters A,H , related to annihilation and hard scattering diagrams. The only relevant parameter here is the A one, while regarding H , the BR shows a moderate dependence [44].
172
Fig. 14. BR(B K) as function of the annihilation parameter A and for H = 1, A,H = 0, for one mass
d ) (left) and chargino ( u ) (right). Lighter dashed line corresponds to
insertion contribution of gluino (RL
23
LL 32
the SM case. The region inside the two horizontal lines corresponds to the allowed experimental range at 2
level.
By using dimensional analysis, one can see that for large values of A , BRSM (B K)
scales like A4 . This result suggests that one can set very strong upper bounds on A by
requiring that BRSM (B K) does not exceed experimental range. Clearly, when new
corrections beyond the SM ones are included, upper bounds on A could be relaxed due to
possible negative interferences between SM and new physics amplitudes.
In order to understand the impact of the annihilation diagrams on BR(B K), we
show below the explicit dependence of the SM amplitude on the parameters A and A :
2
iASM (B KS ) 1.8 108 + 7.9 1010 XA 6.1 1010 XA
+ 2.2 1010 XH ,
(64)
where numbers in the right-hand side of Eq. (64) are in unity of GeV, and XA,H are defined
in Eq. (29). We have used SM central values as in Table 1 of [44], and omitted contributions
from imaginary parts which are quite small.
As can be seen from Eqs. (64) and (29), for A
1 the SM amplitude could be doubled and hence the SM branching ratio would be about four times the result with A < 1,
therefore exceeding the present experimental range. As shown in Ref. [44], and in Fig. 14
(see lighter dashed line), the constraint on A is already obtained for moderate values of
A 2.
It would be interesting to see how much the upper bounds on A could be modified
by the inclusion of SUSY corrections, while satisfying all the experimental constraints.
For this purpose, we plot in Fig. 14 the total branching ratio of B K as a function
of A for gluino and chargino contribution. We present the gluino results for the case of
LR dominant scenario, and the chargino ones with the LL dominant case. The values of
these mass insertions are varied in the allowed range as done for the other plots. Also
we scanned over the different values of the strong phases A,H , squark masses, gluino
and chargino masses. In these plots, the dashed lines correspond to the SM predictions
for BR(B K). As can be seen from this figure, by applying the 2 constraints on
BRSM (B K) one can set very stringent upper bounds on A , namely 2. On the
173
Fig. 15. As in Fig. 9, but for correlations of SKS versus BR(B K) (left) and S KS versus BR(B K)
d ) has been considered.
(right). Gluino contributions with one single mass insertion (LR
23
other hand, this bound can be relaxed to 4 and 5, in the case of LL and LR mass
insertion contributions in up- and down-squark, respectively. However, it is important to
note that most of the configurations in both gluino and chargino cases that allow for > 2
lead to CP asymmetry SKS outside the 2 range of the experimental results. Therefore,
as mentioned above, we consider < 2 as a conservative bound.
Regarding the BR(B K), we find that it is less sensitive to A than BR(B K).
This can be easily observed from the dependence of the amplitude of B K on XA .
Analogously to Eq. (64), we get
2
iASM (B K) 3.6 108 + 0.19 1010 XA + 4.2 1010 XA
+ 7.7 1010 XH ,
(65)
where, as in Eq. (64), numbers in the r.h.s. of Eq. (65) are expressed in units of GeV. Thus,
comparing Eqs. (64) and (65), we see that the effect of A in ASM (B K) is suppressed
by two orders of magnitude with respect to the same one in ASM (B K). Hence the
strongest bound one can obtain on A will come only from BR(B K). Therefore, in
all plots of the present work, including the QCDF ones in Section 4, we scanned over A
by requiring A 2.
As mentioned before, the experimental measurements of BR(B K) represent another large discrepancy with the SM prediction. There have been various efforts to explain
the large observed branching ratio in the B K process, based on the peculiarity of
meson. For instance, intrinsic charm [31] or gluonium contents of [32], have been
investigated as possible new sources for such an enhancement. Clearly, NP contributions
could also be responsible of such discrepancy in BR(B Ks ), or at least for a part of it.
In this respect, we are going to analyze next the maximum effect that one can obtain from
SUSY contributions to BR(B K), by taking into account the experimental constraints
on the BR(B K), SKS , and S KS .
In Fig. 15 we plot the CP asymmetries SKS versus BR(B K) and S KS versus
BR(B K), where the area of colored ellipse corresponds as usual to the allowed experimental region for the correlations at 2 level. We consider here the dominant gluino
174
d )
d
Fig. 16. As in Fig. 15, but for gluino contributions with two mass insertions (LR
23 and (RL )23 , with the
d
d
assumption of arg[(LR )23 ] = arg[(RL )23 ].
d ) and scan over the other parameters as before. One can see from
contribution due to (LR
23
this figure that, gluino contribution can fit quite well inside the 2 ellipse in both left and
right plots. It is worth noticing that, gluino effects could also rise BR(B K) to very
large values ( 1.52 105 ), which are outside the allowed region.
In the correlation between S KS and BR(B K), with just one mass insertion,
SUSY contributions can explain the measured large BR(B K) and at the same time
fit the 2 range for S KS . As can be seen from this figure, for S KS within even 1 region, the BR(B K) can be large as 7 105 . However, in this case some fine tuning
between SUSY parameters is necessary.
In Fig. 16, we present the same correlations as in Fig. 15, but with two simultaneous
d ) and ( d ) in gluino sector. In this case, we can
contributions of mass insertions (LR
23
RL 23
easily see that gluino scenario can saturate simultaneously both SKS and BR(B K)
within their experimental ranges, covering also larger areas with respect to the ones in
Fig. 15.
Finally, the combination from gluino and chargino exchanges on BR(B K) and
BR(B K) versus the respective CP asymmetries, are shown in Fig. 17. We scan on
d )
the allowed range of the most relevant mass insertions for these two contributions: (LR
23
u
(gluino) and (LL )32 (chargino). We also vary the other parameters as in the previous cases.
The message we can learn from these results is that both gluino (LR) and chargino (LL)
contributions can easily accommodate the experimental results for the CP asymmetries
and branching ratios of BR(B K) and BR(B K). As already stressed in Section 4, this phenomenon is related to the fact that b s constraints are more relaxed for
chargino contributions at large tan , due to potentially destructive interferences of gluino
and chargino amplitudes in b s decay. The main difference emerging from results
in Figs. 16 and 17, is that for the latter scenario large and negative values of asymmetry
(S, 1) can also be easily achieved, even though they are now outside the allowed
experimental range.
It is crucial to see, if regions of points which fit inside the two 2 ellipses in Figs. 16
and 17, actually correspond to the same SUSY configurations. For this purpose, in Fig. 18
175
d ) and ( u ) , with
Fig. 17. As in Fig. 15, but for gluino and chargino contributions with mass insertions (LR
23
LL 32
d
u
the assumption of arg[(LR )23 ] = arg[(LL )32 ].
Fig. 18. Correlations of BR(B K) versus BR(B K) for gluino contributions with two mass insertions
d )
d
d
(LR
23 and (RL )23 (left) and for gluino and chargino contributions with two mass insertions (LR )23 and
u ) (right). The 2 constraints on S
(LL
and
S
are
applied.

KS
32
KS
we plot correlations of BR(B K) versus BR(B K) for the same scenarios considered in Figs. 16 and 17, where all points satisfy the constraints on SKS and S KS at
2 level. As we can see from results in Fig. 18, only the scenario in which both LR and
RL mass insertions in gluino exchanges contribute, can naturally enhance BR(B K)
inside the allowed experimental range, while respecting all the other constraints on CP
asymmetries. On the contrary, in the scenario in which both LR (gluino) and LL (chargino)
are contributing, this enhancement is ruled out by simultaneous constraints on CP asymmetries.
As already mentioned, BR(B K) suffers from large theoretical uncertainties due to
the peculiar structure of meson. Therefore, it is not a conservative approach to constrain
NP models from the measurements of BR(B K), until the role of potential new mechanisms responsible of the enhancement of BR(B K) in the SM will be not completely
clarified.
176
6. Direct CP asymmetry in b s versus S( )KS

In this section we analyze the correlation for SUSY predictions between the direct CP
asymmetry ACP (b s ) in b s decay and the other ones in B ( )KS . The CP
asymmetry in b s is measured in the inclusive radiative decay of B Xs by the
quantity
ACP (b s ) =
(B Xs ) (B Xs )
.
(B Xs ) + (B Xs )
(66)
The SM prediction for ACP (b s ) is very small, less than 1% in magnitude, but known
with high precision [46]. Indeed, inclusive decay rates of B mesons are free from large
theoretical uncertainties since they can be reliably calculated in QCD using the OPE. Thus,
the observation of sizeable effects in ACP (b s ) would be a clean signal of new physics.
In particular, large asymmetries are expected in models with enhanced chromo-magnetic
dipole operators, like for instance supersymmetric models [46].
The most recent result reported by BaBar Collaboration for ACP (b s ) is [60]
bs
ACP
= 0.025 0.050 (stat) 0.015 (syst)
(67)
which corresponds to the following allowed range at 90% confidence level:

bs
6% < ACP
< 11%.
(68)
Clearly, the present experimental sensitivity is not accurate enough to strongly constrain
the SM predictions at percent level.
Recently it has been shown that the SUSY contribution to the CP asymmetry in the b
s decay, even with the inclusion of experimental constraints on electric dipole moments
and branching ratio of B Xs , could be much larger than the SM expectation [36].
Therefore, in the light of present experimental results, it would be challenging to analyze
the SUSY predictions for the correlation between ACP (b s ) and S,( )KS , since in
SUSY models these asymmetries are strongly correlated.
The relevant operators of the effective Hamiltonian in Eq. (7) that play a role in
ACP (b s ), are given by Q2 , Q7 , Q8g . We remind here that Q2 , defined in Eq. (8), is
the usual currentcurrent operator induced at tree-level. Then, the expression for ACP (b
s ) at the NLO accuracy, is given by [46]
ACP (b s )

8z

s (mb ) 40

Im C2 C7
v(z) + b(z, ) Im (1 + s )C2 C7
=
2
9
|C7 | 81

4
8z

Im C8g C7
+ b(z, ) Im (1 + s )C2 C8g
,
9
27
where z = (mc /mb )2 and v(z) is [46]

28 4
2
2
2
2
+ ln z
z+
ln z z2
v(z) = 5 + ln z + ln z
3
3
9
3
3
+O z .
(69)
(70)
177
The function b(z, ) is defined as b(z, ) = g(z, 1) g(z, 1 ), where the parameter is
related to the experimental cut on the photon energy, E > (1 )mb /2, and finally g(z, y)
is given by

2

y
y
2
+
1
g(z, y) = (y 4z) y 4yz + 6z ln
4z
4z

3y(y 2z)
4z
(71)
.
1
4
y
In the SM, the Wilson coefficients C2 , C7 , and C8g are real, in particular C2SM 1.08,
C7 = 0.318, and C8g = 0.151, so that the only source of CP violation comes from the
s parameter inside Eq. (69) which is defined in terms of the CKM matrix elements as
s =
V

Vus
ub
2 (i ) O 102 .

Vts Vtb
(72)
In general SUSY models, C7 and C8g may be complex, and the corresponding phase
bs
can provide the dominant CP violating source to ACP . In MIA framework, the SUSY
d )
contributions to the Wilson coefficients C7 and C8g are given in terms of (AB
23 (for
u
gluino) and (AB )32 (for chargino). Therefore, the SUSY phase in these mass insertions
is the only source of CP violation in this process. As pointed out in previous sections, in
8g , which have
general SUSY models, effects induced by the dipole operators Q 7 and Q
opposite chirality to Q7 and Q8g , are non-negligible. In the MSSM these contributions are
suppressed by terms of order O(ms /mb ) due to the universality of the A-terms. However,
in general models one should take them into account, since they might be of the same order
than C7 and C8g . Denoting by C 7 and C 8g the Wilson coefficients multiplying the new
7 and Q
8g , the expression for the asymmetry in Eq. (69) will be modified by
operators Q
making the replacement
Ci Cj Ci Cj + C i C j .
(73)
Notice that, since14 C 2 = 0, the only modification in the numerator of Eq. (69) is due
to the new term C 7 C 8g . However, if only one single mass insertion is taken into account, both C 7 and C 8g will be proportional to the same mass insertion, and therefore
) = 0. Therefore, the effects of these new operators will enter only through
Im(C 8g C 7
C 7 in the denominator, by means of the shift |C7 |2 |C7 |2 + |C 7 |2 .
It is worth mentioning that in the scenarios dominated by one single mass insertion, the
leading SUSY contribution to the CP asymmetry ACP (b s ) is due to the C7 . This can
be understood as follows. Using our previous inputs for charm and bottom masses and also
assuming, for instance, an energy resolution 0.3 so that b(z, ) 0.17, Eq. (69) can be
14 In principle, C might receive some contributions from RGE due to the mixing of the other Q
i operators
2
2 . However, the radiative effects induced by C i on C 2 are quite small and so we will neglect them in our
with Q
analysis.
178
Fig. 19. Correlations of SKS versus ACP (b s ) (left) and S KS versus ACP (b s ) (right), for gluino
d ) .
contributions with one single mass insertion (LR
23
written as
ACP (b s )

s (mb )
0.045 Im C7
0.44 Im C7 C8g
+ 0.006 Im C8g
.

2
|C7 |
(74)
at scale are related to the

We remind here that the low energy coefficients C7 and C8g
b
high energy ones by
SUSY
SUSY
SUSY
(mb ) 0.66C7
(W ) + 0.09C8g
(W ),
C7
(75)
SUSY
SUSY
C8g
(mb ) 0.69C8g
(W ).
(76)
As shown in Section 2, the kind of mass insertions that will appear in C7 (W ) are the
same that determine C8g (W ). In the case of SUSY contribution from one single mass
( ) will acquire the same phase and so the second term
insertion, both C7 (b ) and C8g
b
in Eq. (69) identically vanishes.
In conclusion, the relevant mass insertions contributing to ACP (b s ) give also the
dominant effects in SKS and S KS . Therefore, it is expected that these three CP asymmetries should be highly correlated, since all of them depend on the same source of SUSY CP
violation.
In Fig. 19 we plot the correlations between SKS and ACP (b s ) and also S KS and
d ) . As we can
ACP (b s ) for gluino exchanges with one mass insertion, namely (LR
23
see from results in Fig. 19, a specific trend emerges from this scenario. The simultaneous combination of SKS and S KS constraints, favors the SUSY predictions to be in the
region of positive values of ACP (b s ) asymmetry.15 As shown in Fig. 20, analogous
results are also obtained when two mass insertions in gluino sector are taken simultane15 This result is also in agreement with the corresponding predictions for the correlation of S
KS versus
ACP (b s ) in the article of Kane et al. in Ref. [6].
179
d ) and ( d ) .
Fig. 20. As in Fig. 19, but for gluino contributions with two mass insertions (LR
23
RL 23
d )
u
Fig. 21. As in Fig. 19, but for gluino and chargino contributions with mass insertions (LR
23 and (LL )32 ,
respectively.
bs
d ) and ( d ) . However, in this case also negative values of A

ously, namely (LR
23
LR 23
CP
are allowed. Although these points are favored by SKS , they are disfavored by S KS .
d )
u
Finally, in Fig. 21 we show our results for two mass insertions (LR
23 and (LL )32
with both gluino and chargino exchanges. In this case we see that SKS constraints do not
set any restriction on ACP (b s ), and also large and positive values of ACP (b s )
asymmetry can be achieved. However, by imposing the constraints on S KS , see plot on
the right side of Fig. 21, the region of negative ACP (b s ) is disfavored in this scenario
as well.
7. Conclusions
The B-meson decays to K, K, and to XS provide a clean window to the physics
beyond the SM. In this paper we have systematically analyzed the supersymmetric contributions to the CP asymmetries and the branching ratios of these processes in a model inde-
180
pendent way. The relevant SUSY contributions in the b s transitions, namely chargino
and gluino exchanges in box and penguin diagrams, have been considered by using the
mass insertion method.
We have provided analytical expressions for all the relevant Wilson coefficients. Both
naive factorization and QCD factorization approximation to determine the hadronic matrix
elements have been employed. We showed that the SUSY predictions for the mixing CP
asymmetry of SKS and S KS are not very sensitive to the kind of approximation adopted
for evaluating hadronic matrix elements.
We found that due to the stringent constraints from the experimental measurements of
BR(b s ), the scenario with pure chargino exchanges cannot give large and negative
values for CP asymmetry SKS . Indeed, by combining present experimental constraints on
SKS and S KS asymmetries at 2 level, pure chargino contributions can be almost ruled
out. On the other hand, it is quite possible for gluino exchanges to account for SKS and
S KS at the same time. Interestingly, we have shown that the charged Higgs, if not too
heavy, may change the chargino and gluino contributions to enhance the CP asymmetries
considerably.
The branching ratios of these decays have also been considered. It has been noticed that
the parameter is strictly bounded by the B K branching ratio, and this influences
strongly the possible SUSY contributions to the asymmetries. Investigating the correlations
between CP asymmetries and branching ratios, pointed out that the most favored SUSY
scenarios, which easily satisfy all the experimental results, are the ones with inclusion of
two mass insertions. In particular, a pure gluino dominated scenario, in which both LR
and RL mass insertions are contributing and the one in which both gluino and chargino
contribute with LR and LL mass insertions in down- and up-squark sectors, respectively.
In the latter scenario we show that chargino exchanges provide sizeable effects. This is due
to the fact that b s constraints could be relaxed by potentially destructive interferences
between gluino and chargino amplitudes. Finally, it is remarkable to notice that in the
scenario in which both LR and RL down squark mass insertions dominate, the observed
large enhancement of BR(B K) could be naturally explained, while respecting all the
other experimental constraints on CP asymmetries.
We also discussed the correlations between the CP asymmetries of these processes and
the direct CP asymmetry in b s decay. In this case, we found that the general trend of
SUSY models, satisfying all the experimental constraints, strongly favors large and positive
contributions to b s asymmetry.
More precise measurements would certainly allow us to draw more definite conclusions
on the SUSY models that can accommodate these data.
Acknowledgements
The authors thank Debrupa Chakraverty for her efforts in the initial stages of this work.
We would like to thank G. DAgostini, E. Kou, and K. sterberg for very useful discussions. E.G. acknowledges the CERN PH-TH division and IPPP, CPT of Durham University
for their kind hospitality during the preparation of this work. The authors appreciate the
181
financial support from the Academy of Finland (project Nos. 104368 and 54023) and from
the Associate Scheme of ICTP.
Appendix A. Chargino contributions in MIA

Here we provide the expressions for chargino contributions, at leading order in MIA, to
the RFAB quantities in Eq. (15), where F refers to D, E, C, B(u,d) , and M ,g [11].

LL
RL
=
|Vi1 |2 xW i PD (xi ),
RD
=
Vi2 Vi1 xW i PD (xi ),
RD
i=1,2
RR
RD
i=1,2
RL
= RD
,
|Vi2 | xW i PD (xi ),
LR
RD
|Vi1 |2 xW i PE (xi ),
RL
RE
=
|Vi2 | xW i PE (xi ),
LR
RE
i=1,2
LL
RE
=
i=1,2
RR
RE
Vi2 Vi1 xW i PE (xi ),
i=1,2
RL
= RE
,
i=1,2
RCLL
(0)
|Vi1 |2 PC (xi )
i=1,2

1
Ui1 Vi1 Uj1 Vj1 PC(2) (xi , xj ) + |Vi1 |2 |Vj 1 |2
PC(1) (xi , xj ) ,
8
i,j =1,2
1
(0)
Vi2 Vi1 PC (xi )
2
i=1,2

(2)
(1)
RCRL =
i,j =1,2
RCLR
RCRR

= RCRL ,

(2)
(1)
=
i,j =1,2
RBLL
=2
u
Vi1 Vj1 Ui1 Uj1 xWj xij PBu (xj , xij ),
i,j =1,2
RBRL
u
= 2
RBLR
u

= RBRL
,
u
i,j =1,2
RBLL
=
d
RBRL
d
RBRR
=2
u
i,j =1,2
|Vi1 |2 |Vj 1 |2 xWj PBd (xj , xij ),
i,j =1,2
i,j =1,2
Vi2 Vi1 |Vj 1 |2 xWj PBd (xj , xij ),
182

RBLR
= RBRL
,
d
d
LL
RM
=
,g

i
LR
RM
=
,g
RBRR
=
d
Vi2 Vi1 Vj1 Vj 2 xWj PBd (xj , xij ),
i,j =1,2
LL
|Vi1 |2 xW i PM
(xi ) Yb
,g

i
LL
Vi1 Vi2 xW i PM
(xi ) + Yb
,g
RL
RM
=
,g
RR
RM
=
,g
Vi1 Ui2 xW i
mi LR
P
(xi ),
mb M ,g
Vi2 Ui2 xW i
mi LR
P
(xi ),
mb M ,g
LL
Vi1 Vi2 xW i PM
(xi ),
,g
i
LL
|Vi2 |2 xW i PM
(xi ).
,g
(A.1)
Regarding the RF0 terms at the zero order in mass insertion, we have
0
RD
= 2xW
|Vi2 |2 D (xi ),
0
RE
= 2xW
i=1,2
RC0 = 2
|Vi2 |2 E (xi ),
i=1,2
Ui1 Uj1 Vi2 Vj2 C(2) (xi , xj ),
i,j =1,2
RB0 d =
RB0 u =
0
RM
1
xWj Vi2 Vi1 Vj1 Vj 2 B(d) (xj , xj , xij ),
2
i,j =1,2
(u)
xWj Vi2 Vj2 Ui1
Uj 1 B (xj , xj , xij ),
i,j =1,2
= xW
|Vi2 |
i=1,2
0
RM
= xW
g

2
F1 (xi ) + F2 (xi ) ,
3
|Vi2 |2 F2 (xi ),
(A.2)
i=1,2
2 , xi
where Yb is the Yukawa coupling of bottom quark, xW i m2W /m2i , xW m2W /m
2
2
2
2
2
2
mi /m
, xi m
/mi , and xij mi /mj . The chargino mixing matrices U and V , are
,
defined in Eq. (13). The loop functions of penguin PD,E,C , D , E , C , box B (u,d,g)
(u,d,g)
LL , P LR are
, and magnetic and chromo-magnetic penguin diagrams F1,2 , PM
PB
M ,g
,g
reported in Appendix C.16

16 The minus sign appearing in front of the right-hand side of equation above for R 0 , takes into account
Bd
the correction of a sign mistake in the first reference of [51] for the chargino contributions to down-type box
diagrams. This mistake was also pointed out in the second reference of [51]. This sign correction has also been
re-absorbed in the PBd function given in Appendix C.
183
Appendix B. Chargino contributions in MIA for the light tR case

In this appendix we provide the relevant formulas for the case in which the mass of the
stop-right (m2t ) is lighter than the average squark mass m
2 . We recall here that this will
R
modify only the expressions of RFRL , RFRR , and RF0 , since the light stop right does not affect
0
0
RR
RR
RFLL . In the case of RD,E,C
and RD,E,C
the functional forms of RD,E,C
and RD,E,C
remain
unchanged, while the arguments of the functions involved are changed as xW xW t , xi
xit , and xi xit , where xW t m2W /m2t , xit m2i /m2t , xit m2t /m2i . Regarding the
R
(u,d)
and RB0 u,d , the functions PB

box-type contributions17 RBRR
u,d
should be replaced as follows
(u,d)
PB
(xj , xij ) PB
(u,d)
(u,d)
and B
appearing there
(xj t , xj , xij ),
B(u,d) (xj , xj , xij ) B(u,d) (xj t , xj , xij ),
(B.1)
where
1
(u,d)
PB (x, y, z) = x B(u,d) (x, y, z).
(B.2)
2 x
In the case of RFLR and RFRL the analytical expression of loop functions of penguin PD,E,C ,
(u,d)
LL and P LR , rebox PB , and magnetic and chromo-magnetic penguin diagrams PM
M ,g
,g
spectively, should be replaced with the following ones
PD,E (xi ) PD,E (xi , xit ),
(1,2)
PC
PC0 (xi ) PC0 (xi , xit ),
(1,2)
(xi , xj ) PC
(xi , xit , xj , xj t ),
(u,d)
(u,d)
PB (xj , xij ) PB (xj , xj t , xij ),
(B.3)
where

2
xit D (xit ) xi D (xi ) ,
xt 1

2
xit E (xit ) xi E (xi ) ,
PE (xi , xit ) =
xt 1

2 (1,2)
(1,2)
C (xj t , xit ) C(1,2) (xj , xi ) ,
PC (xi , xit , xj , xj t ) =
xt 1

4
(0)
C(1) (xit , xi ) C(1) (xi , xi ) ,
PC (xi , xit ) =
xt 1
(u)

1
(u)
B (xj t , xj , xij ) B(u) (xj , xj , xij ) ,
PB (xj , xj t , xij ) =
2(xt 1)
(d)

1
(d)
B (xj t , xj , xij ) B(d) (xj , xj , xij ) ,
PB (xj , xj t , xij ) =
2(x 1)
t

1
2
2
LL
xit F1 (xit ) + F2 (xit ) xi F1 (xi ) + F2 (xi ) ,
PM (xi , xit ) =
xt 1
3
3
PD (xi , xit ) =
17 Here we have corrected the formula in [11]. The effects are O(2 ) and do not affect the numerical results.
184

1
2
2
xit F3 (xit ) + F4 (xit ) xi F3 (xi ) + F4 (xi ) ,
xt 1
3
3

1
LL
xit F2 (xit ) xi F2 (xi ) ,
PM
(xi , xit ) =
g
xt 1

1
LR
PM
(B.4)
xit F4 (xit ) xi F4 (xi ) .
(xi , xit ) =
g
xt 1
LR
PM
(xi , xit ) =
The expressions for the loop functions involved in Eqs. (B.2) and (B.4) can be found in
Appendix C.
Appendix C. Basic SM and SUSY loop functions

Here we report the loop functions entering in the SM and SUSY diagrams at 1-loop.
SM functions
x 2 (25 19x) 3x 4 + 30x 3 54x 2 + 32x 8
+
log x,
36(x 1)3
18(x 1)4
x(x 6) x(3x + 2)
C(x) =
+
log x,
8(x 1)
8(x 1)2
x(x 2 + 11x 18) 9x 2 + 16x 4
+
log x,
E(x) =
12(x 1)3
6(x 1)4
x
x
+
log x.
B(x) =
4(x 1) 4(x 1)2
D(x) =
(C.1)
Charged Higgs functions

x(47x 2 79x + 38) x(3x 2 + 6x 4)
+
log x,
108(x 1)3
18(x 1)4
1
CH (x) = B(x),
2
x(7x 2 29x + 16) x(3x 2)
EH (x) =
+
log x.
36(x 1)3
6(x 1)4
DH (x) =
(C.2)
Chargino functions in MIA

PD (x) =
PE (x) =
2x(22 + 60x 45x 2 + 4x 3 + 3x 4 3(3 9x 2 + 4x 3 ) log x)

27(1 x)5
2
3
x(1 + 6x 18x + 10x + 3x 4 12x 3 log x)
,
9(1 x)5
x(3 4x + x 2 + 2 log x)
(0)
PC (x) =
,
8(1 x)3

2
x (x 1 log x) y 2 (y 1 log y)
1
(1)
PC (x, y) =
,
8(x y)
(x 1)2
(y 1)2

(2)
PC (x, y) =
PBu (x, y) =
185

xy
x(x 1 log x) y(y 1 log y)
,
4(x y)
(x 1)2
(y 1)2
y x(1 3x + y)
x(x 3 + y 3xy + y 2 ) log x
4(x 1)2 (x y)2

2(x 1)3 (x y)3
xy log y
,
+
2(x y)3 (y 1)
x(3y x(1 + x + y)) x(x 3 + (x 3)x 2 y + y 2 ) log x
PBd (x, y) =
4(x 1)2 (x y)2

2(x 1)3 (x y)3
xy 2 log y
,
+
2(x y)3 (y 1)
x(2 9x + 18x 2 7x 3 + 3x(x 2 3) log x)
LL
PM
(x)
=
,
9(1 x)5
x(13 20x + 7x 2 + (6 + 4x 4x 2 ) log x)
LR
PM
(x)
=
,
6(1 x)4
x(1 + 9x + 9x 2 17x 3 + 6x 2 (3 + x) log x)
LL
PM
(x)
=
,
g
12(1 x)5
x(1 4x + 5x 2 2x(2 + x) log x)
LR
PM
(x) =
.
g
2(1 x)4
(C.3)
Chargino functions in MIA with a light tR

43x 2 + 101x 52 2x 3 9x + 6
+
log x,
108(x 1)3
18(x 1)4
2

x
y2
1
(1)
log x
log y ,
C (x, y) =
16(y x) x 1
y1

xy
x
y
(2)
C (x, y) =
log x
log y ,
8(y x) x 1
y 1
D (x) =
11x 2 + 7x 2
x3
+
log x,
36(x 1)3
6(x 1)4

B(u) (x, y, z) = F (x, y, z) + F (y, z, x) + F (z, x, y) ,
E (x) =
B(d) (x, y, z) = xF (x, y, z) + yF (y, z, x) + zF (z, x, y),

x log x
,
F (x, y, z) =
(x 1)(x y)(x z)
x 3 6x 2 + 3x + 2 + 6x log x
F1 (x) =
,
12(x 1)4
2x 3 + 3x 2 6x + 1 6x 2 log x
F2 (x) =
,
12(x 1)4
x 2 4x + 3 + 2 log x
x 2 1 + 2x log x
F3 (x) =
,
F
(x)
=
.
4
2(x 1)3
2(x 1)3
(C.4)
186
Gluino functions in MIA

P1 (x) =
1 6x + 18x 2 10x 3 3x 4 + 12x 3 ln(x)
,
18(x 1)5
7 18x + 9x 2 + 2x 3 + 3 ln(x) 9x 2 ln(x)
P2 (x) =
,
9(x 1)5
1 + 4x 5x 2 + 4x ln(x) + 2x 2 ln(x)
M1 (x) =
,
2(1 x)4
5 + 4x + x 2 2 ln(x) 4x ln(x)
M2 (x) =
,
2(1 x)4
1 + 9x + 9x 2 17x 3 + 18x 2 ln(x) + 6x 3 ln(x)
M3 (x) =
,
12(x 1)5
1 9x + 9x 2 + x 3 6x ln(x) 6x 2 ln(x)
M4 (x) =
,
6(x 1)5
1
B2 (x) = xM2 (x).
B1 (x) = M1 (x),
4
(C.5)
References
[1] A.G. Cohen, D.B. Kaplan, A.E. Nelson, Annu. Rev. Nucl. Part. Sci. 43 (1993) 27;
M.B. Gavela, P. Hernandez, J. Orloff, O. Pne, C. Quimbay, Nucl. Phys. B 430 (1994) 345;
M.B. Gavela, P. Hernandez, J. Orloff, O. Pne, C. Quimbay, Nucl. Phys. B 430 (1994) 382;
A.D. Dolgov, hep-ph/9707419;
V.A. Rubakov, M.E. Shaposhnikov, Phys. Usp. 39 (1996) 461, Usp. Fiz. Nauk 166 (1996) 493 (in Russian).
[2] B. Aubert, et al., BaBar Collaboration, Nucl. Instrum. Methods A 479 (2002) 1;
S. Mori, et al., Belle Collaboration, Nucl. Instrum. Methods A 479 (2002) 117.
[3] R. Fleischer, T. Mannel, Phys. Lett. B 511 (2001) 240;
C.W. Chiang, J.L. Rosner, Phys. Rev. D 68 (2003) 014007.
[4] S. Khalil, R. Mohapatra, hep-ph/0402225, Nucl. Phys. B, in press.
[5] A. Datta, Phys. Rev. D 66 (2002) 071702;
B. Dutta, C.S. Kim, S. Oh, Phys. Rev. Lett. 90 (2003) 011801;
A. Kundu, T. Mitra, Phys. Rev. D 67 (2003) 116005;
B. Dutta, C.S. Kim, S. Oh, G. Zhu, hep-ph/0312388;
B. Dutta, C.S. Kim, S. Oh, G. Zhu, hep-ph/0312389.
[6] E. Lunghi, D. Wyler, Phys. Lett. B 521 (2001) 320;
M.B. Causse, hep-ph/0207070;
G. Hiller, Phys. Rev. D 66 (2002) 071502;
M. Ciuchini, L. Silvestrini, Phys. Rev. Lett. 89 (2002) 231802;
S. Khalil, E. Kou, Phys. Rev. D 67 (2003) 055009;
K. Agashe, C.D. Carone, Phys. Rev. D 68 (2003) 035017;
G.L. Kane, P. Ko, H.B. Wang, C. Kolda, J.H. Park, L.T. Wang, Phys. Rev. Lett. 90 (2003) 141803;
C. Dariescu, M.A. Dariescu, N.G. Deshpande, D.K. Ghosh, Phys. Rev. D 69 (2004) 112003;
M. Ciuchini, E. Franco, G. Martinelli, A. Masiero, M. Pierini, L. Silvestrini, hep-ph/0407073.
[7] Z. Xiao, W. Zou, hep-ph/0407205.
[8] M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 67 (2003) 075016;
M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 68 (2003) 079901, Erratum.
[9] S. Khalil, E. Kou, Phys. Rev. Lett. 91 (2003) 241602.
187
[10] S. Baek, Phys. Rev. D 67 (2003) 096004;

Y. Wang, Phys. Rev. D 69 (2004) 054001.
[11] D. Chakraverty, E. Gabrielli, K. Huitu, S. Khalil, Phys. Rev. D 68 (2003) 095004.
[12] J.F. Cheng, C.S. Huang, X.H. Wu, Phys. Lett. B 585 (2004) 287.
[13] J. Charles, et al., CKMfitter Group, hep-ph/0406184.
[14] R. Fleischer, hep-ph/0405091.
[15] D. London, R. Peccei, Phys. Lett. B 223 (1989) 257.
[16] D. London, A. Soni, Phys. Lett. B 407 (1997) 61.
[17] Y. Grossman, G. Isidori, M. Worah, Phys. Rev. D 58 (1998) 057504.
[18] M.A. Giorgi, BaBar Collaboration, Plenary talk at XXXII Int. Conference on High Energy Physics, Beijing,
China, 1622 August 2004, http://ichep04.ihep.ac.cn/.
[19] Y. Sakai, Belle Collaboration, Plenary talk at XXXII Int. Conference on High Energy Physics, Beijing,
China, 1622 August 2004, http://ichep04.ihep.ac.cn/.
[20] T.E. Browder, hep-ex/0312024, Proceedings of 21st Int. Symposium on Lepton and Photon Interactions at
High Energies, Batavia, IL, 1116 August 2003, in press.
[21] B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 89 (2002) 201802;
K. Abe, et al., Belle Collaboration, Phys. Rev. D 66 (2002) 071102(R).
[22] A.J. Buras, hep-ph/0210291;
A.J. Buras, F. Parodi, A. Stocchi, JHEP 0301 (2003) 029.
[23] B. Aubert, et al., BaBar Collaboration, hep-ex/0403026.
[24] K. Abe, et al., Belle Collaboration, Phys. Rev. Lett. 91 (2003) 261602.
[25] Heavy Flavor Averaging Group, http://www.slac.stanford.edu/xorg/hfag/.
[26] B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 91 (2003) 161801.
[27] B. Aubert, et al., BaBar Collaboration, Phys. Rev. D 69 (2004) 0111102.
[28] K. Abe, et al., Belle Collaboration, Phys. Rev. Lett. 91 (2003) 201801;
H. Aihara, Belle Collaboration, Talk at FPCP, 2003.
[29] R.A. Briere, et al., CLEO Collaboration, Phys. Rev. Lett. 86 (2001) 3718;
S.J. Richichi, et al., CLEO Collaboration, Phys. Rev. Lett. 85 (2000) 520.
[30] E. Kou, A.I. Sanda, Phys. Lett. B 525 (2002) 240.
[31] I.E. Halperin, A. Zhitnitsky, Phys. Rev. Lett. 80 (1998) 438.
[32] D. Atwood, A. Soni, Phys. Lett. B 405 (1997) 150.
[33] P. Fayet, S. Ferrara, Phys. Rep. 32 (1977) 249;
H.P. Nilles, Phys. Rep. 110 (1984) 1.
[34] Y. Grossman, Y. Nir, R. Rattazzi, Adv. Ser. Dir. High Energy Phys. 15 (1998) 755, hep-ph/9701231.
[35] S. Abel, S. Khalil, O. Lebedev, Nucl. Phys. B 606 (2001) 151.
[36] D. Bailin, S. Khalil, Phys. Rev. Lett. 86 (2001) 4227.
[37] S. Abel, D. Bailin, S. Khalil, O. Lebedev, Phys. Lett. B 504 (2001) 241;
S. Khalil, hep-ph/0202204.
[38] S. Bertolini, F. Borzumati, A. Masiero, Nucl. Phys. B 294 (1987) 321;
M. Ciuchini, et al., Phys. Rev. Lett. 79 (1997) 978;
R. Barbieri, A. Strumia, Nucl. Phys. B 508 (1997) 3;
S.A. Abel, W.N. Cottingham, I.B. Whittingham, Phys. Rev. B 58 (1998) 073006;
A. Kagan, in: Proceedings of the 7th Int. Symposium on Heavy Flavor Physics, Santa Barbara, CA, 711
July 1997, hep-ph/9806266.
[39] B. Dutta, C.S. Kim, S. Oh, G. Zhu, hep-ph/0312388;
B. Dutta, C.S. Kim, S. Oh, G. Zhu, hep-ph/0312389.
[40] L.J. Hall, V.A. Kostelecky, S. Raby, Nucl. Phys. B 267 (1986) 415;
J.S. Hagelin, S. Kelley, T. Tanaka, Nucl. Phys. B 415 (1994) 293;
E. Gabrielli, A. Masiero, L. Silvestrini, Phys. Lett. B 374 (1996) 80;
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
[41] H. Simma, D. Wyler, Phys. Lett. B 272 (1991) 395;
G. Kramer, W.F. Palmer, H. Simma, Nucl. Phys. B 428 (1994) 77;
G. Kramer, W.F. Palmer, H. Simma, Z. Phys. C 66 (1995) 429;
G. Kramer, W.F. Palmer, Phys. Rev. D 52 (1995) 6411;
188
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
G. Kramer, W.F. Palmer, Y.L. Wu, Commun. Theor. Phys. 27 (1997) 457;
D. Du, L. Guo, Z. Phys. C 75 (1997) 9;
N.G. Deshpande, B. Dutta, S. Oh, Phys. Rev. D 57 (1998) 5723.
A. Ali, G. Kramer, C.D. L, Phys. Rev. D 58 (1998) 094009.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914;
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313.
M. Beneke, M. Neubert, Nucl. Phys. B 675 (2003) 333.
M. Beneke, M. Neubert, Nucl. Phys. B 651 (2003) 225.
A.L. Kagan, M. Neubert, Phys. Rev. D 58 (1998) 094012.
G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125.
S. Bertolini, F. Borzumati, A. Masiero, G. Ridolfi, Nucl. Phys. B 353 (1991) 591.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
K. Adel, Y.P. Yao, Phys. Rev. D 49 (1994) 4945;
A. Ali, C. Greub, Phys. Lett. B 361 (1995) 146;
M. Misiak, M. Mnz, Phys. Lett. B 344 (1995) 308;
C. Greub, T. Hurth, D. Wyler, Phys. Rev. D 54 (1996) 3350;
C. Greub, T. Hurth, Phys. Rev. D 56 (1997) 2934;
N. Pott, Phys. Rev. D 54 (1996) 938;
K. Chetyrkin, M. Misiak, M. Mnz, Phys. Lett. B 400 (1997) 206;
K. Chetyrkin, M. Misiak, M. Mnz, Phys. Lett. B 425 (1998) 414, Erratum;
A.J. Buras, A. Kwiatkowski, N. Pott, Nucl. Phys. B 517 (1998) 353;
P. Gambino, M. Misiak, Nucl. Phys. B 611 (2001) 338;
A.J. Buras, A. Czarnecki, M. Misiak, J. Urban, Nucl. Phys. B 631 (2002) 219.
E. Gabrielli, G.F. Giudice, Nucl. Phys. B 433 (1995) 3;
E. Gabrielli, G.F. Giudice, Nucl. Phys. B 507 (1997) 549, Erratum;
A.J. Buras, P. Gambino, M. Gorbahn, S. Jager, L. Silvestrini, Nucl. Phys. B 592 (2001) 55.
E. Gabrielli, S. Khalil, Phys. Rev. D 67 (2003) 015008.
S. Khalil, O. Lebedev, Phys. Lett. B 515 (2001) 387.
J.M. Gerard, W.S. Hou, Phys. Rev. D 43 (1991) 2909.
S. Eidelman, et al., Phys. Lett. B 592 (2004) 1.
A. Kagan, M. Neubert, Eur. Phys. J. C 7 (1999) 5.
M. Ciuchini, G. Degrassi, P. Gambino, G.F. Giudice, Nucl. Phys. B 527 (1998) 21;
M. Ciuchini, G. Degrassi, P. Gambino, G.F. Giudice, Nucl. Phys. B 534 (1998) 3;
P. Ciafaloni, A. Romanino, A. Strumia, Nucl. Phys. B 524 (1998) 361;
F. Borzumati, C. Greub, Phys. Rev. D 58 (1998) 074004;
F. Borzumati, C. Greub, Phys. Rev. D 59 (1999) 057501;
G. Degrassi, P. Gambino, G.F. Giudice, JHEP 0012 (2000) 009;
M. Carena, D. Garcia, U. Nierste, C.E.M. Wagner, Phys. Lett. B 499 (2001) 141;
F. Borzumati, C. Greub, Y. Yamada, Phys. Rev. D 69 (2004) 055005.
A. Kagan, Phys. Rev. D 51 (1995) 6196;
M. Ciuchini, E. Gabrielli, G.F. Giudice, Phys. Lett. B 388 (1996) 353;
M. Ciuchini, E. Gabrielli, G.F. Giudice, Phys. Lett. B 393 (1997) 489, Erratum.
S. Chen, et al., CLEO Collaboration, Phys. Rev. Lett. 87 (2001) 251807;
R. Barate, et al., ALEPH Collaboration, Phys. Lett. B 429 (1998) 169;
K. Abe, et al., Belle Collaboration, Phys. Lett. B 511 (2001) 151;
B. Aubert, et al., BaBar Collaboration, hep-ex/0207076;
M. Steinhauser, hep-ph/0406240.
B. Aubert, et al., Phys. Rev. Lett. 93 (2004) 021804.
Towards quantum dielectric branes:

curvature corrections in Abelian beta function
and non-Abelian BornInfeld action
Pedro Bordalo a , Lorenzo Cornalba b , Ricardo Schiappa c,d
a LPTHE, Universit Paris VI, 4 Place Jussieu, F-75252 Paris cedex 05, France
b Instituut voor Theoretische Fysica, Universiteit van Amsterdam,
Valckenierstraat 65, 1018 XE Amsterdam, The Netherlands

c CAMGSD, Departamento de Matemtica, Instituto Superior Tcnico,
Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal

d Faculdade de Engenharia, Universidade Catlica Portuguesa, Estrada de Talade,
2635-631 Rio de Mouro, Lisboa, Portugal

Received 20 September 2004; received in revised form 2 December 2004; accepted 14 December 2004
Abstract
We initiate a programme to compute curvature corrections to the non-Abelian BornInfeld action. This is based on the calculation of derivative corrections to the Abelian BornInfeld action,
describing a maximal brane, to all orders in F = B + 2 F . An exact calculation in F allows
us to apply the SeibergWitten map, reducing the maximal Abelian brane point of view to a minimal non-Abelian brane point of view (replacing 1/F with [X , X ] at large F ), resulting in matrix
model equations of motion in the considered background. We first study derivative corrections to the
Abelian BornInfeld action and compute the two loop beta function for an open bosonic string in a
WZW (parallelizable) background. This beta function is the first step in the process of computing
open string equations of motion, which can be later obtained by either computing the Weyl anomaly
coefficients or the partition function in the given background. The beta function for the gauge field is
exact in F and computed to orders O(H, H 2 , H 3 ) (where H = dB and the curvature is R H 2 ) and
O(F , 2 F , 3 F ). In order to carry out this calculation we develop a new regularization method
for two loop graphs. We then relate perturbative results for Abelian and non-Abelian BornInfeld
E-mail addresses: bordalo@lpthe.jussieu.fr (P. Bordalo), lcornalb@science.uva.nl (L. Cornalba),

schiappa@math.ist.utl.pt (R. Schiappa).
doi:10.1016/j.nuclphysb.2004.12.019
190
P. Bordalo et al. / Nuclear Physics B 710 (2005) 189254
actions, by showing how Abelian derivative corrections yield non-Abelian higher order commutators
and vice versa, at large F . We begin the construction of a matrix model describing corrections to
Myers dielectric effect. This construction is carried out by first setting up a perturbative classification of the relevant non-Abelian tensor structures, which can be considerably narrowed down by the
physical constraint of translation invariance in the action and the possibility for generic field redefinitions. The final matrix action is not uniquely determined and depends upon two free parameters.
These parameters could be computed via further calculations in the Abelian theory.
PACS: 11.25
1. Introduction and summary

The study of D-branes in curved backgrounds has been the subject of a great deal of
activity in recent years. D-branes seem to be quite interesting objects once we move away
from the realm of flat space, as they generically display the ability to change their shape
and their dimension, among many other phenomena. These properties arise as one embeds
the branes in nontrivial closed string backgrounds, with perhaps the most renowned aspect
being the dielectric effect [1]. While there has been a great deal of research dedicated to
this aspect of D-brane physics, it has all been performed for branes within weakly curved
backgrounds (see, however, [2,3]). It is therefore clear that a most pressing question concerns what happens as the closed string curvature is increased. An answer to this question
necessarily requires a detailed analysis of curvature corrections to the open string equations
of motion.
The dielectric effect is also undoubtedly associated to the fact that we are dealing with
many D-branes, rather than a single one. The study of quantum corrections to the dielectric
effect (corrections in ) then seems as a great challenge: one would need to generalize the
non-Abelian BornInfeld (BI) action to curved backgrounds. Still, if one wishes to proceed
along these lines, there are a couple of routes one can follow. The original approach in [1]
constrains the non-Abelian BI action via an analysis of open string T-duality. In order
to extend this work to higher order in the background curvature one would first require
a set of corrected open string T-duality rules, which are not yet available. Thus, an
attempt along these lines would first entail a study of open string T-duality to higher loop
order in the worldsheet. Another approach to this problem was the one followed in [47]
which started with an one loop calculation in matrix theory [8]. Here one can of course
imagine performing higher loop calculations and, following the same approach, expect to
obtain a quantum corrected non-Abelian BI action. This is however a long road to follow
(other approaches within the setting of matrix theory were also attempted, e.g., [915]).
Finally, one can of course just do the standard string theory calculation, by computing the
non-Abelian string effective action. This is also well known to be a hard calculation (see,
e.g., [1620]).
The approach we shall use is based on recent work concerning noncommutative geometry in string theory [2126]. In this context one has the so-called SeibergWitten (SW) map
relating commutative and noncommutative descriptions of the BI action [25], and the main
idea is simple: a commutative description corresponds to a maximal brane analysis (thus,
191
an Abelian BI) while a noncommutative description corresponds to a minimal brane analysis (thus, the non-Abelian BI). Knowledge of the Abelian equations of motion will yield
the non-Abelian ones, as long as the SW map is known. That such a procedure can be used
to compute curvature corrections to actions based on matrix models was first suggested
in [24], and was later studied in detail in [2729] within the context of the non-Abelian BI
action in flat space. In here, we wish to start a programme which extends that analysis to
curved backgrounds.
The starting point is the calculation of curvature corrections to the Abelian BI equations of motion, to all orders in the gauge field strength. We will work in the bosonic open
string sigma model, where the beta function of the gauge field living on the D-brane is
computed perturbatively in derivatives of background fields and is further expected
to have corrections at every loop order.1 Observe that one key aspect of this calculation
is that it needs to be exact in F , the gauge field strength on the D-brane. Indeed, if one
has any hope of performing the SW limit, open string parameters G and will need to
be properly identified before such a limit can be taken [25]. This contrasts with previous calculations performed in the literature, which were mostly perturbative in F . Beta
functions for bosonic open string sigma models were computed up to two loops (in flat
space) in [3033]. Later, and in the superstring context, curvature corrections to the BI
and WessZumino (WZ) actions were computed in [3437], but no F or B field on the
brane were considered. These calculations allowed a better understanding of the physics
of Abelian branes in curved spaces, but are of no help on what concerns the non-Abelian
case. Inclusion of the F (or B) field came later. In flat space, analysis of the BI and WZ
to all orders in F was carried out in [38,39], and in curved space a perturbative analysis
in F , to first order, was done in [40,41]. Computations in curved backgrounds, which are
exact in F , were performed in the context of the bosonic open string in [42] (but where
the result turned out not to be target space covariant), and in the context of type II string
theory in [43], where corrections to the string effective action were obtained. A different
line of research concerning the computation of corrections to the non-Abelian BI action,
in the supersymmetric case, has been carried out in [4447]. Let us make one more remark
concerning the fact that our calculation deals with bosonic string theory. At the BI level,
this will differ from the superstring calculation via the appearance of terms in F n with n
odd, where there is a proposal to sum up these odd contributions [48].
In this paper we shall compute the beta function for the gauge field on a maximal
D-brane, at two loop order, using a short distance cutoff as a regulator. The curved closed
string background is always kept on-shell. Here one could in principle choose any bosonic
closed string background (without a dilaton, as while we are computing corrections, we
remain at small string coupling gs ). The sigma model analysis of closed bosonic strings
in [4951] however points out that generic backgrounds will be extremely complicated
at two loops, due to the immense possibilities in both curvature and NeveuSchwarz
NeveuSchwarz (NSNS) H -field couplings (with H = dB). It would thus be of some
interest to restrict to some class of backgrounds which would be more tractable on a first
approach. A particular class of backgrounds was pointed out precisely in [49], the class of
1 This is not the case for a supersymmetry preserving brane in a supersymmetric background.
192
parallelizable backgrounds (this is a large class of backgrounds which includes all Wess
ZuminoWitten (WZW) group manifold models). In this regard, it is also interesting to
note that the non-Abelian example in [1] of D0-branes expanding into a fuzzy sphere,
due to the presence of RamondRamond (RR) flux, is precisely matched by the example
in [52,53] of D0-branes expanding into a fuzzy sphere, this time around due to the presence
of NSNS flux. Hence, one can consider the SU(2) WZW model of [52,53] as the prototype
model of the physics of Myers dielectric effect.
Regarding this last point, it is worth noting that the question of how will curvature
corrections contribute to the dielectric effect on non-Abelian branes has received some
attention, with special emphasis towards possible corrections to the standard fuzzy sphere
result. Indeed, the initial result in [52,53] for the SU(2) WZW model at level k showed
that at infinite level the D0-branes expand into a fuzzy sphere. It moreover pointed out
that, at finite level, the fuzzy spheres are expected to be deformed in some manner closely
i
related to quantum groups at a root of unity, q = e k+2 , at least for integer level k. Another
point of interest arose in later work [54,55] where it was shown that D-brane charge is only
well defined modulo k + 2. This can be understood physically by the statement that S2
branes inside S3 SU(2), made out of D0-branes, can only grow up to a certain radius. As
finite level corrections to the infinite level result amount to corrections to the dielectric
effect, this series of works seem to imply that the non-Abelian branes get corrected to some
sort of quantum group deformation. In [56], this WZW model was further investigated
from the maximal brane point of view, where it was found that the semiclassical analysis
yields precise agreement with the conformal field theory (CFT) analysis,2 and further clues
pointing towards a quantum group relation were uncovered. Still, it was not until [58
62] that an attempt was done in order to try to put the quantum group relation on firmer
grounds. The authors of these papers started from the conjecture that the matrix model
describing the full quantum corrected dielectric branes of the SU(2) level k WZW model
i
was a quantum group deformed matrix model at the root of unity q = e k+2 , and tried to
match to both the CFT and the semiclassical results previously mentioned. A derivation of
such a quantum matrix model directly from string theory remains an open problem.
Having computed the two loop beta function for an open bosonic string in a WZW
(parallelizable target manifold) closed string background, to all orders in F = B + 2 F
(which is the correct gauge invariant combination on the worldsheet boundary), one would
like to relate it to the open string equations of motion (or the open string effective action).
In fact, it is not the vanishing of the beta functions, = 0, which should be equivalent to the
string equations of motion, but rather the vanishing of the Weyl anomaly coefficients, = 0
(i.e., the ones which ensure the absence of the Weyl anomalythe standard beta functions
only ensure the absence of the scale anomaly) [6365]. This can be accomplished in two
distinct ways. On one hand, one can compute the missing terms in the relation between the
beta function and the Weyl anomaly coefficient, = + , and thus find the equations
of motion. On the other hand, one may wish to compute the string effective action directly.
This can be done by using the techniques of boundary string field theory [66,67] which tell
us how to relate the string effective action to the partition function, precisely via the beta
2 These results were extended to all compact groups in [57].
193
function

S[A] = 1 + [A]
Z[A].
A
Because any of the aforementioned calculations is at the same level of difficulty as the
two loop beta function calculation, we shall not perform them in this paper and will leave
them for future work. Nevertheless, we shall carefully describe how to extract perturbative
information concerning the non-Abelian BI action in curved backgrounds, given higher
derivative corrections to the Abelian BI action. This is the first step in a programme which
may eventually lead to new insight into the dielectric effect.
As we have explained, our method uses the SW map to connect commutative and noncommutative descriptions of the BI action. Derivative corrections in the Abelian action
can be understood, at large field strength F on the Abelian D-brane worldvolume, as corrections to the standard multiplication of functions in such a way that this multiplication
actually gets deformed to a noncommutative star product. This is achieved via a particular change of variables (from commutative to noncommutative fields) which precisely
corresponds to the SW map [25]. The noncommutative description is then simpler to understand as a matrix model, as one represents functions with operators, star products with
operator multiplication and integrals get replaced by traces [28,68,69]. This calculation,
while naturally perturbative from the Abelian BI point of view, also induces a straightforward perturbation theory for the calculation of curvature corrections to the non-Abelian
BI action. As it turns out, and we shall show this later on, a simple understanding of the
non-Abelian perturbation theory leads to a classification of all possible curvature corrections which, constrained by translation invariance and field redefinitions, are extremely
few in number. In this way, one can infer many things about the non-Abelian BI action in
curved backgrounds without having to carry out the (Abelian) calculation to its very end.
Naturally, absolute conclusions and results will only be available once all coefficients are
fixed. Here, we shall content ourselves with two free parameters. Future work should then
concentrate on taking the programme we set out to its full extent.
This paper is organized as follows. In Section 2 we begin with a review of the perturbation theory expansion which is required in order to perform the two loop calculation we
purpose to do. We explain what propagator should be used in order for the calculation to be
exact in F , and list all the one and two loop diagrams that arise from the possible vertices.
Many details concerning the perturbative expansion of the worldsheet action, as well as
technical details concerning the propagator are left to Appendices A and B. In Section 3
we shall perform the main calculation in this paper as we obtain the two loop beta function
for the maximal D-brane gauge field, A, presented in Section 3.9 (many technical details of
the calculation are presented in Appendix C). In order to compute the two loop beta function we first have to devise a new regularization scheme, which is actually valid to any loop
order, that will allow us to resum the infinite series of powers in F we are obtaining (due to
exactness of the result with respect to this field) in terms of open string tensors, G and .
After having obtained the maximal brane result, we set up its translation into a minimal
brane language. This is done in Section 4. This is a descriptive section where we explain
how derivative corrections to the Abelian BI can be used to obtain curvature couplings in
the non-Abelian BI action, setting up the programme of computing corrections to the di-
194
electric effect in string theory. Mostly motivated by the Abelian results we then describe a
perturbative scheme within which one classifies tensor structures which may contribute (at
first non-trivial order) to curvature couplings in the non-Abelian BI action (we also include
Appendices E and F dealing with some known couplings of the non-Abelian BI action).
We list possible non-Abelian tensor structures appearing at two loops in perturbation theory, and show how they correspond, at large F , to Abelian tensor structures appearing in
the Abelian BI action. These structures can be further constrained by the requirement of
translation invariance of the action, and by generic field redefinitions. It turns out that in the
end one is left with surprisingly few allowed tensor structures. Concentrating on the SU(2)
model at level k, a prototype for Myers dielectric effect, there are in fact only two curvature corrections to consider. This allows us to write a matrix model for quantum dielectric
branes (still dependent on the two free parameters), and which we present in Section 4.4.
We end with some directions for future work.
2. Setting up the calculation

We begin this section by presenting the open string sigma model, which we shall expand
using the background field method in order to yield local covariant interaction vertices. The
background field method is described at great length in Appendix A, to which we refer the
reader for details and notation. It is from this expanded sigma model action that we shall
perform the perturbation theory analysis which will allow us to compute the required two
loop beta function. We shall also discuss in this section the one and two loop diagrams
relevant for the computation. The open string sigma model action contains a bulk contribution

i
1
g
(X)
dX
dX
+
B (X) dX dX ,
S=
4
4
as well as a boundary contribution

SB = i
A (X) dX ,
where is the worldsheet Hodge dual. B is the gauge potential for the NSNS 3-form H
field, H = dB. We shall restrict our study to backgrounds which are parallelizable, in the
sense that
1
R + H H = 0,
4
H = 0.
These manifolds represent exact closed string backgrounds, to all orders in [49] (examples of such manifolds are the group manifolds and some of their orbifolds, see [70]). In
the following, we list the various ingredients required for the perturbation theory.
195
2.1. The propagator

We shall follow the standard background field method by choosing a classical solution
to the string equations of motion x ( a ), which satisfies3
Da a x = 0,

g dx + iF dx = 0,
where
F = B + 2 (dA)
and where we have defined the following covariant derivatives on sections of the target
space tangent bundle
D = d + dx ,
i
D = D dx H .
2
We will take the fluctuating quantum field to be a section, , of the target space tangent
bundle, defining the geodesic which starts from the classical solution and must be followed
for an unit of affine parameter in order to reach the displaced configuration of the string.
In terms of the quantum field , the quadratic part of the action then becomes (see
Appendix A)

1
i
F dx + F D .
g
D
D
+

4
4
In order to put the bulk part of the quadratic action in the standard form (from where
one can easily identify the worldsheet propagator) we introduce, on the worldsheet, the
spacetime vielbein EA , such that (here gAB is constant)
g (x) = EA (x)EB (x)gAB ,
and we shall reparametrize the quantum fluctuations with the tangent space vector A given
by
A = EA .
It turns out that, whenever the background is parallelizable, we can always choose EA such
that [49]
EA D = d A .
(2.1)
...
Thus, more generally, given any section S11...n m we have that

M1 ...Mm
1
n
...m
dSN
= EM11 EMmm EN
EN
DS11...
.
n
n
1 ...Nn
1
(2.2)
3 Let a be coordinates on the string worldsheet, , with disk topology and endowed with a flat fixed metric.
196
In particular, on the boundary of the worldsheet

1 ABC

A
D
D = EA d H
B FCD dx ,
2
(2.3)
where, clearly,
dx A EA dx .
We are now in a position to write the quadratic part of the action in a final form

i
gAB
A
B
d d +
FAB A d B
4
4
i
+
4

1
DE
A FBC FAD HB FCE A B dx C .
2
(2.4)
One can now discuss the propagator for the quantum field A . Since the tensors FAB
and HABC depend on the position on the worldsheet, we shall consider only the bulk part
of the action as defining the free theory, taking the 2-vertices on the boundary as interaction
vertices. In Appendix B we derive the propagator for a slightly more general kinetic term,
corresponding to the first line of (2.4) with FAB constant (and when is the unit disc in the
complex plane). We shall later see in the discussion that this case is actually the relevant
one for the computation of the beta function. Let us quote the result for the propagator,
which is written in terms of the open string tensors4 G and ,
1
g+F
AB

=
1
G
AB
+ AB ,
where GAB and AB are, respectively, symmetric and antisymmetric. Introducing the
functions

1 zw
A(z, w) = ln
,
B(z, w) = 2 ln |1 zw|,
1 z w

zw
1
,
D(z, w) = (zz + w w),
C(z, w) = ln

1 zw
2
defined for z, w C, one has the propagator on the disk as

1 A
(z) B (w) = AB A(z, w) + GAB B(z, w) + g AB C(z, w) + D(z, w) .
If considering only the bulk part of (2.4) as the free action, one will simply have the above
formula but with GAB = g AB and AB = 0.
4 We shall interchangeably use both GAB and ( 1 )AB for the inverse open string metric.
G
197
2.2. The vertices

Next we need to enumerate the relevant vertices, for the perturbative calculation we
wish to perform. In the last subsection we have already discussed the 2-vertices. One can
now use the results in Appendix A for the action written in terms of , together with
Eqs. (2.1) and (2.3), to quickly arrive at the results below. Recall that the vertices below
have an explicit minus sign with respect to the expansion of the action, since the weight
factor for the path integral is eS .
2.2.1. Bulk vertices
We begin by listing the relevant vertices arising from the bulk terms in the perturbative
expansion of the sigma model action. The resulting vertices relevant for the beta function
calculation are given by

1 1
TI J K I d J d K ,
2 2

1 1
TI J KL I J d K d L ,
2 4
where the tensors TI J K and TI J KL are given by

i
TI J K = HI J K ,
3

1
1
TI J KL =
HI KM H M J L + HJ KM H M I L = (RI KJ L + RJ KI L ).
12
3
Note that TI J K is totally antisymmetric, whereas TI J KL is explicitly symmetric in indices
I , J and K, L.
2.2.2. Boundary vertices
The boundary vertices are obtained in a similar way, starting from the expansion of the
action in Appendix A, and fall into two distinct classes. First of all we have the vertices

1
MI J I d J ,
2

1 1
MI J K I J d K ,
2 2!

1 1
MI J KL I J K d L ,
2 3!
where the tensors MA1 ...An are explicitly symmetric in the first n 1 indices A1 , . . . , An1 ,
and are given by
MI J =
i
FI J ,
2
198
i
MI J K = (I FJ K + J FI K ),
3

i
1
I J FKL FI M H M J N H N KL + SymI J K .
MI J KL =
8
12
The second class of boundary vertices is given explicitly by

1 1
NI J I J ,
2 2!

1 1
NI J K I J K ,
2 3!

1 1
NI J KL I J K L .
2 4!
The tensors NA1 ...An are explicitly symmetric in all the indices A1 , . . . , An , and are worldsheet 1-forms, since they explicitly contain the 1-form dx A . They are given by

i
1
1
NI J =
I FJ K + J FI K FI M HJ MN FKN FJ M HI MN FKN dx K ,
2!
2
2
by
NI J K

i
1
I J FKL FI M H M J N H N KL I FJ M H M KN F N L
=
3!
4

+ SymI J K dx L ,
and finally by

i
3
3
I J K FLM I FJ A H A KB H B LM I J FKA H A LB F B M
NI J KL =
4!
4
2

1
A
B
C
D
+ FI A H J B H KC H LD F M + SymI J KL dx M .
8
The above vertices, together with the free propagator described previously, are all we shall
need in order to compute the two loop beta function.
2.3. The diagrams
To conclude this section we turn our attention to the diagrams contributing to the beta
function at both one and two loops, in parallelizable backgrounds. The diagrams which
contribute in this case may have either one explicit boundary vertex of N -type, containing
a dx M insertion, or none.5 Let us comment on this further. Since we are interested in counterterms to the gauge potential, AM , we are looking for divergences in the graphs of the
5 Of course all diagrams to consider are 1PI graphs.
199
form M dx M , which clearly contain a single dx M term. Diagrams with two insertions
of dx M would contribute to the closed string sector beta functions, and these have already
been taken into account as the closed string physics is on-shell (new diagrams arising from
the open string sector with two insertions of dx M will only contribute at string loop level
as is well known [32]). The reason why diagrams with no insertion of dx M may contribute
is the following: we are, in general, looking for divergences in correlators of operators at
different points on the boundary of the string worldsheet. We can then have ultralocal divergences, proportional to derivatives of the delta function, which will contribute to the beta
function upon integration by parts, thus producing the missing insertion of dx M [33]. We
shall then denote the diagrams with one insertion of dx M as diagrams of -type, and those
with no dx M insertion as diagrams of -type (the reason for the notation - and -type,
related to the type of local divergence, will become fully clear in the next section).
We begin by listing the one loop diagrams. There is one -type diagram and one -type
diagram (where the thick horizontal line denotes the disk boundary):
(2.5)
Note that we are not showing the insertions, along the propagators, of the 2-vertex of Mtype,

1
MI J I d J ,
2
since, as will become clear in the next section, we are going to absorb its effects to all
orders in MI J in the propagator.
2.3.1. Two loop -type graphs
Below are the two loop diagrams with one explicit insertion of dx M . It is clear from the
structure of the vertices previously presented that there are only insertions of dx M at the
disk boundary, in this parallel background case, due to the N -type vertex (in the general
case, there will also be diagrams with bulk insertions of dx M ). Of the following two loop
graphs, diagrams 3, 4, 6, 7 and 8 contribute to the beta function:
(2.6)
200
2.3.2. Two loop -type graphs

These are the two loop diagrams with no insertion of dx M , and in which we shall look
for ultralocal divergences. Observe that graphs 4 and 5, although they look as purely bulk
graphs, implicitly have an arbitrary number of 2-vertex insertions of M-type on the propagators, and therefore could have ultralocal contributions on the boundary. Nonetheless we
shall show that the diagrams contributing are 1, 2 and 3:
(2.7)
Let us make one comment on counterterm graphs, which we have not considered. When
performing a two loop computation one should, in principle, include all the one loop counterterms in the original action and further consider their fluctuations. This yields further
vertices and diagrams at two loops. However, and for what concerns the beta function calculation we are interested in, we may consider these terms to be zero. This reason for doing
so is that these counterterm diagrams only have divergences in ln2 , and not in ln , with
the ultraviolet cutoff. Thus, they will not contribute to the beta function [33]. As such, the
graphs above are all the graphs one needs to take into consideration.
3. The two loop beta function

In order to compute the beta function associated with the gauge field, AM , we must consider divergences arising in vacuum diagrams at one and two loops built from the vertices
described in the previous section. We shall then be interested in divergences of the form
()
1
ln
M (A, H ) dx M ,
2

()
M
where
is the contribution to the beta function at loops, and where is the ultraviolet cutoff scale. The above form is dictated by dimensional analysis. As is clear, we are
looking for a divergence given by the integral of a local function of the string fields, M ,
multiplying the 1-form dx M . As seen in the last section, a generic vacuum graph will then
fall into two large categories. It either contains a single insertion of a N -type vertex, which
explicitly contains a dx M , or it will not contain an N -type vertex, and then the dx M factor
will appear from differentiation in a way which we shall shortly explain. We call the two
types of diagrams - and -graphs, respectively.
3.1. Graphs of -type
Let us first focus on graphs of -typei.e., with a single insertion of a vertex of
N -typesince these graphs are the simplest ones conceptually. We will first describe the
201
general computational scheme and then explicitly work out two particular two loop graphs,
leaving the full computation to Appendix C. Also, we shall work from now on with units
such that
= 1.
In general, we are considering expressions of the following schematic form

d d1
dn
N x( ) O1 x(1 ) On x(n ) C(, 1 , . . . , n ),

2 2
2
where , 1 , . . . , n , are the angular positions of the n + 1 vertices on the unit circle and
where the N -type vertex is located at . We are suppressing all dependence on the radial
coordinate on the disk, as well as all explicit reference to tangent space indices on the
various tensors, N and Oi (which are of M or of T -type), which are just spectator indices
as far as the arguments below are concerned. Finally, and most importantly, we are calling
C to the correlator of the A (, ) fields corresponding to the graph in question (again
with tangent space indices understood). By standard quantum field theory arguments, the
divergence in this correlation function C will be local in the angular coordinates, , i , and
will be givenby dimensional argumentsby
c ln (2)n
( i ),
where c is a constant tensor (recall we are suppressing indices). It is for this reason that we
are dubbing these graphs as -type graphs. Therefore, although the vacuum graph is highly
nonlocal in the background spacetime fields, the divergent part is local and is given by

d
c ln
N O1 On x( ) ,
2
thus yielding a contribution to the beta function of
M dx M = cNO1 On .
We therefore need to find the tensor c. The simplest way to do this is to consider the integral

dn
d d1
C(, 1 , . . . , n ) = c ln + regular.
2 2
2
If we work instead in momentum space, in the angular direction ein , where n is the discrete angular momentum, we will be considering the insertion of the composite operators
in the various vertices at zero momentum, and we will be looking at the divergent part.
Therefore, in practical terms, one must take all the vertices and replace tensors multiplying the fields, which depend on the spacetime position x(, ), by constant tensors, and
only then proceed to compute the graph (where vertices are now most easily displayed in
momentum space).
202
3.2. The vertices in momentum space

To show how this works, let us work out the form of the vertices in momentum space.
Let us first consider the M-type vertices, with momentum W flowing into the vertex (as
discussed above, for the -type graphs one needs W = 0, but for the moment we will keep
the discussion slightly more general, since we shall later need this extension for the -type
graphs). We have

1
d iW A1
AN1 AN ( )
MA1 A2 ...AN
e
(N 1)!
2
i
(nN MA1 A2 ...AN + cyc1...N )nA11 nANN i ni W ,
=
N!
where cyc1...N stands for the sum over cyclic permutations, and the sum over the outgoing
momenta ni is implied. As we argued above, the tensor MA1 ...AN should be thought of as
constant. We therefore have the vertices
i(nN MA1 A2 ...AN + cyc1...N )i ni W .
The 2-vertex is simple, since MAB = MBA , and one has
i(n2 n1 )MA1 A2 n1 +n2 W .
The N -type vertices have no derivatives and are simply given by
NA1 A2 ...AN i ni W .
We also have two bulk T -type vertices. First of all, the 4-vertex (here, we are parametrizing
the unit disk with radial coordinate [0, 1] and angular coordinate [0, 2])

1
1
d iW A1 A2
A3 A4 + 2 A3 A4
e
TA1 ...A4 d

4
2

1
d A1 A2 A3 A4
= TA1 ...A4
|n3 n4 | n3 n4 i ni W
4
n1 n2 n3 n4
(n3 n4 |n3 n4 |)
1
i ni W .
= TA1 ...A4 nA11 nA22 nA33 nA44
4
i |ni |
Note that, in the above, we are able to perform the -integration due to the fact that the
-propagator factorizes as follows (more details in the next subsection)
A
( ) |n| |n| .
nA ( )n
Then, denoting (ij kl) TAi Aj Ak Al (nk nl |nk nl |), the vertex is given by

1
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413)
i |ni |

+ (1423) + (2314) i ni W .
Finally we have the 3-vertex

|n2 |n3
d iW A1
e
n W .
TA1 A2 A3 d
x A2 A3 = iTA1 A2 A3 nA11 nA22 nA33
2
i |ni | i i
203
Using the total antisymmetry of TABC , the 3-vertex finally reads

iTA1 A2 A3
|n1 |(n2 n3 ) + |n2 |(n3 n1 ) + |n3 |(n1 n2 ) i ni W .
i |ni |
3.3. The dressed propagator
Recall that, due to the kinetic term

gAB
d 2 ab a A b B ,
4
the propagator on the disk is given by
A

(z) B (w) = g AB B(z, w) + C(z, w) + D(z, w) .
The function D(z, w) only contributes to the zero momentum part of the correlator, since
it only depends on the radial distances zz and w w.
It will therefore not contribute in the
computation of the leading divergences in the high momentum regime, where the sums
representing the various graphs diverge. The function C(z, w), on the other hand, vanishes
if either z or w are on the boundary. It therefore only contributes in graphs where one
internal line connects two bulk vertices of T -type (or a T -vertex to itself). We shall argue
that the C-part of the propagator does not contribute to beta function. The reason comes
from noticing that the divergence associated to the M function is associated to singular
behavior when all operators collide at the same point on the boundary of the worldsheet,
exactly where the contribution of C vanishes. Therefore, for the purpose of computing the
beta function, we consider instead the propagator

A
(z) B (w) = g AB B(z, w),

|zw||n|
B
,
= g AB
nA n
|n|
where, in the second line, we have given the momentum space representation, using the
fact that

1
B xei , yei =
(xy)|n| ein() .
|n|

n
=0
As we had previously noted, the propagator factorizes as |zw||n| and the radial part has then
already been absorbed in the definition of the vertices. We are only left with the angular
part, which is simply

1
B
= g AB .
nA n
|n|
Following the above discussion we are then instructed, in the computation of -type graphs,
to use the vertices in the previous subsection at zero momentum (with W = 0) and glue
them with the above propagator.
Recall that we have two distinct 2-vertices

i(n2 n1 )MA1 A2 n1 +n2 ,
NA1 A2 n1 +n2 ,
204
and that we want to add their contribution to the propagator in order to have an exact
result in F . On the other hand, since in our graphs we shall have at most one insertion
of the 2-vertex N , we shall rather find it convenient to include only the M-type 2-vertex
in the propagator. As usual in quantum field theory, the inverse propagator sums with the
2-vertex, to give the full propagator
1
1

= |n|gAB + nFAB
.
|n|gAB 2inMAB
Introducing open string tensors as usual

AB AB
1
1
=
+ AB ,
g+F
G
we obtain the final dressed propagator

1
1
B
nA n
+ AB nAB .
= GAB
|n|
n
3.4. The regularization scheme

The last ingredient one needs in order to compute the graphs is a regularization scheme.
Introducing a regulator > 0, we consider a generic vacuum graph contributing to a
-graph. In order to regulate the sums defining the graph at high momenta, let us introduce a factor of
e|n|
for every internal line connecting two vertices, where we shall consider as a single line any
line with insertions of 2-vertices. Note that this is consistent due to the fact that all vertices
are inserted at zero momentum and therefore 2-vertices do not alter the momentum flowing
through a single line (which is therefore a well-defined quantity, n).
We shall then be interested, as already advocated, in graphs which diverge primitively as
ln , when 0. In fact, graphs which diverge as ln2 and higher (already present at two
loops) will be made finite by the presence of the lower order counterterms, as described for
instance in [33].
3.5. One loop result
As warm-up, let us rederive the standard one loop result (2.5). When we later discuss
-graphs, we shall see that they also contain a possible contribution at one loopthis,
nonetheless, vanishes upon explicit computation. Therefore we compute the full one loop
beta function by computing the -graph.
The single graph contributing is shown in (2.5), and is given by

1
nAB e|n| .
NAB
2
n
=0
The factor of 12 comes from the symmetry factor of the graph. The sum is explicitly symmetric in n n, and therefore the only contribution comes from the GAB term in nAB ,
205
thus yielding
NAB GAB
1
n>0

en = NAB GAB ln 1 e NAB GAB ln().
Therefore, the one loop contribution to the beta function is given by the standard result [32]
AB
1
(1)
M
(3.1)
dx M = NAB
.
G
3.6. Examples of two loop -graphs
We will next consider two examples of two loop diagrams, however leaving the full
computation to Appendix C. Let us first focus on an -type graph, which we shall choose
to be [4] (see (2.6)). It is clearly given, using the Feynman rules we have just discussed
and the symmetry factor of 4, by
1 AR SR CD |n||m|
n n m e
4
n,m
=0

i m(MABCD MDABC ) + n(MCDAB MBCDA ) NRS .
(3.2)
In order to proceed, we first quote some results on double sums (reviewed in Appendix C),
which read
ea(n+m)
ea(n+m)
0.
ln ,
2
nm
n
n,m1
n,m1
Notice that, in the first sum, we are isolating the ln term part. Moreover, the second sum
diverges as ln2 and vanishes for the purpose of computing the beta function, since we are
looking for primitive ln divergences. It is therefore easy to see that the last two terms in
the second line of (3.2) vanish. Now using that MABCD is symmetric in ABC and that the
sum over m is odd for the remaining terms, so that one can replace mCD with m1 AB , we
arrive at the result
2iMABCD (GNG + N)AB CD ln .
Let us next focus on the computation of a -type graph, for instance [6] (see (2.6)). It
has a symmetry factor of 4 and is given by

1

|n||m||p|
n+m+p nAA mBB (N )CC
TABC TA B C
p e
4
n,m,p
[|n|(m p) + |m|(p n) + |p|(n m)]2
.
(|n| + |m| + |p|)2
We must split the sum into three regions I, II, III, given by n, m > 0, n, p > 0 and m, p > 0,
alongside with the corresponding regions given by (n, m, p) (n, m, p). We are
then led to consider the function
F (n, m, p) =
1 [|n|(m p) + |m|(p n) + |p|(n m)]2 |n||m||p|

e
nmp 2
(|n| + |m| + |p|)2
206
and the corresponding sums

SI =
F (n, m, n p) =
n,m>0
SII =

n,m>0
F (n, n p, n) =
n,p>0
SIII =
(n m)2 2(n+m)
e
,
nm(n + m)2

n,p>0
F (m p, m, p) =
m,p>0
(n p)2 2(n+p)
e
,
p 2 n(n + p)
m,p>0
(m p)2 2(m+p)
e
.
p 2 m(m + p)
The above sums can be easily computed using results from Appendix C, and are given by
SI 4 ln ,
SII ln ,
SIII ln .
Now let us consider the propagators nAA mBB (N )CC

p . Since the full sum must be
invariant under (n, m, p) (n, m, p), we will obtain the following tensor structures
1

AA BB (GNG + N )CC ,
2
nmp
1

GAA GBB (GNG + N )CC ,
2
|n||m|p
1

GAA BB (GN + N G)CC ,
|n|mp|p|
1

AA GBB (GN + N G)CC ,
n|m|p|p|
which, in turn, give the following total sums
2(SI + SII + SIII ) 4 ln ,
2(SI SII SIII ) 12 ln ,
2(SI + SII SIII ) 8 ln ,
2(SI SII + SIII ) 8 ln ,
where the factor of 2 comes from regions with (n, m, p) (n, m, p). Combining all
of the above facts, we finally have the result for the graph [6], which is
TABC TA B C ( + 3GG)AA ,BB (GN G + N )CC ln

+ 2TABC TA B C (G + G)AA ,BB (GN + N G)CC ln .

3.7. Graphs of -type
We now move out attention to the more complex case of the -type graphsi.e.,
those graphs which do not contain an N -type vertex and which therefore do not contain
an explicit dx M in the vertices. Again, we are considering expressions of the following
207
schematic form

dn
d d1
O x( ) O1 x(1 ) On x(n ) C(, 1 , . . . , n ),

2 2
2
where, as before, we are suppressing explicit reference to all indices and radial dependence.
To compute these graphs we again need to understand the nature of the short-distance
singularities in the correlator C. Introducing the functions

( j ),
DI (, i ) = (2)n ( I )
j
=I
with I = 1, . . . , n, it is simple to see, by power counting, that the short distance singularity
of C proportional to ln must be of the form

cI DI ,
ln
I
where the cI s are constant tensors. Therefore we have a contribution to the beta function
given by

M dx M =
cJ OO1 dOJ On .
J
(note that the arbitrary choice of O and OJ is immaterial, since different choices change
the beta function by an exact 1-form). Let us comment on the expression dOJ . Assume, for
M . Then, using (2.2)
simplicity of exposition, that the tensor OJ is given by some section SN
and (2.3), one concludes that

1
1
M
M

dSN = E EN DS = E EN DS + H F dx S H F dx S .
2
2
Since moreover D = dx , we arrive at the final explicitly covariant result

1
1
M
M
P
= L S N
+ H M P A F A L SN
H P N A F A L SPM dx L ,
dSN
2
2
which can be easily extended to general tensors OJ . We see explicitly that, due to the
presence of ( J ) in the short-distance singularity of the correlator C, one obtains an
explicit factor of dx M in the expression for the graph. To find the tensors cI , we consider
the integral

dn
1
d d1
C(, 1 , . . . , n ) eiW (J ) eiW (J )

2iW
2 2
2
= cJ ln + regular.
The first term inserts momentum W in the vertex at and momentum +W at the vertex
at J . Again working in momentum space we see that one must consider the vertices of
section (3.2), with the following differences:
(1) all vertices Oi , i
= J , are at zero momentum W = 0;
208
(2) the vertex corresponding to OJ has momentum W and has the external tensor M, T ,
replaced by dM, dT ;
(3) the vertex corresponding to O has momentum W .
We then subtract the result with W W , and divide by 2iW . In practice, we shall write
down the initial expressions for arbitrary W and will use the fact that the singular part is W independent, in order to simplify the results. We shall, more specifically, use the expression
above formally for W 0, by substituting it with

.
W W =0
In doing so, we shall use the following facts

1
AB = nAB ,
W W =0 n+W
n

|p|
e|pW | = e|p| .
W W =0
p
(3.3)
The regularization scheme adopted for the -graphs easily carries over to the -graphs,
with one minor extension. Consider an internal line, possibly with insertions of 2-vertices.
If all of the 2-vertices are at zero momentum, the momentum flowing along the line is just
a given value n and we regularize it with e|n| . If, on the other hand, we have K 1 insertions of 2-vertices with nonvanishing momentum, then the momentum flowing through the
line will change, and will acquire K distinct values n1 , . . . , nK . Then, we shall regularize
the line with the factor
e K (|n1 |++|nK |) .
Finally, and as far as the propagator is concerned, it is clear that whenever the 2-vertex
MAB is inserted at zero momentum it can be resummed as before, to give the previously
discussed dressed propagator.
3.8. Examples of two loop -graphs
First we want to show that the -contribution to the one loop beta function vanishes. It
is given by
1

AA
(2n W )2 nW
nB B MAB dMA B e 2 (|n|+|nW |) ,
2 n
|W =0 . Using (3.3), one only obtains a contribution from the

on which we must act with W
acting on the regulator, and it is

-dependent part, coming from W
n|n|nAA nB B MAB dMA B e|n| .
209
Keeping only the even part of the sum under n n, we obtain

e|n| .
2MAB (G dM + dM G)AB
n>0
The above does not contribute to the beta function for two reasons. Firstly, MAB is
odd
AB BA, and (G dM + dM G)AB is even. Secondly, the expression
under|n|
= e /(1 e ) is regular as 0, and therefore does not constitute a
n>0 e
diverging term in the correlator.
Let us now compute, as an example of a two loop -graph, the contribution to the beta
function coming from [1] (see (2.7)). The graph, whose symmetry factor is 4, has the
following expression (readily obtained from the Feynman rules previously described)

1
i MABCD (m) + MBCDA (n W ) + MCDAB (n) + MDABC (+m)
4 n,m
AR
nSB mCD e 2 (|nW |+|n|+2|m|) .
i(2n W ) dMRS nW
Computing the derivative W

|W =0 one obtains the two terms (in the following expressions

we will omit, for simplicity, the sum n,m and the regulator e(|n|+|m|) )

1
CD
m(MABCD MDABC ) + n(MBCDA + MCDAB ) ( dM )AB
n m
4
and, from the regulator,

|n|
CD
m(MABCD MDABC ) + n(MCDAB MBCDA ) ( dM )AB
n m .
4

1
In the first expression, the second term vanishes, since n,m nm
0. The first term gives
(MABCD MDABC )(G dM G + dM )AB CD ln

which itself vanishes, this time around due to symmetry of MABCD and antisymmetry of
(G dM G + dM )AB . The second expression, coming from the regulator, gives the
following two contributions
(MDABC MABCD )(G dM G + dM )AB CD ln ,
(MBCDA MCDAB )(G dM + dM G)AB GCD ln .
The first term vanishes as before and the remaining term yields the unique contribution to
the beta function, given by (renaming indices)
(MABCD MBCDA )(G dM + dM G)AD GBC ln .
3.9. Final result for the two loop beta function
Following the techniques just described one can proceed to compute the contribution
to the beta function from all of the - and -graphs. The full computation is carried out
in Appendix C and leads to the results we shall now present. The following graphs give a
210
vanishing contribution
[1] = [2] = [5] = 0,
[4] = [5] = 0.
The nonvanishing contributions to the beta function from the -type graphs are

1 1 AR,BS,CT
1 1
[3] = MABC (3MRST + MRT S )
N + N
G G
GG

AR,BS,CT

1 1
N + N
+ MABC (MRST MRT S )
G G

1
1
1 AR,BS,CT
1
N + N
+
+ MABC (MRST 3MRT S )
,
G
G
G
G

AB
1 1
N + N
[4] = 2i MABCD
CD ,
G G

AA ,BB ,CC

1 1
1 1
N + N 3
+
[6] = TABC TA B C
G G
GG

1
1
1
1 AA ,BB ,CC
+ 2TABC TA B C
,
N + N
+
G
G
G
G

1 1
[7] = 4TABC MA B C 2 N + N

G G

AA ,BB ,CC
1
1
1
1
,
N + N
+
G
G
G
G

AB CD
1 1
1
N + N
[8] = TABCD
.
G G
G
The nonvanishing contributions to the beta function from the -type graphs are

1 AD 1 BC
1

dM + dM
[1] = (MABCD MBCDA )
,
G
G
G
1
[2] = MA1 A2 A3 dMB1 B2 B3
6
A1 B1 ,A2 B2 ,A3 B3
1 1
1 1
1 1
2 + 3
+3 4
GG
G G
GG
1
+ MA1 A2 A3 dMB2 B3 B1
3
A1 B1 ,A2 B2 ,A3 B3
1 1
1 1
1 1
+ 2
,
5 +2
GG
G G
GG

1
1
1
[3] = 6iMABC TA B C

dM + dM
G
G
G

AA ,BB ,CC

1
1
dM + dM
.
G
G
We then have that the two loop contribution to the beta function is given by

(2)
M dx M =
[i] +
[j ].
i
211
(3.4)
Recall that the above terms must be summed to the one loop contribution (3.1),
AB
1
(1)
M
M dx = NAB
.
G
4. The non-Abelian BornInfeld action

We are now in possession of the Abelian beta function for the open string gauge field.
At this stage it is important to stress that = 0 are not the equations of motion for the
massless open string mode [63]: indeed it is not the vanishing of the beta functions, = 0,
which should be equivalent to the string equations of motion, but rather the vanishing of
the so-called Weyl anomaly coefficients, = 0 (i.e., the ones which ensure the absence
of the Weyl anomalythe standard beta functions only ensure the absence of the scale
anomaly). For the closed string the Weyl anomaly coefficients have been studied in [63]
and it is known that they are given by the beta functions plus some additive terms. These
additive terms, up to two loops, are given in terms of derivatives of the dilaton and covariant derivatives of H (at three and higher loops they have not been computed) so that if
the dilaton is constant and H is covariantly constant, as is the case for the parallel backgrounds, the Weyl anomaly coefficients will actually coincide with the beta functions. For
the open string there are also some results [65], although not as complete. Here, the Weyl
anomaly coefficients are both diffeomorphism and gauge invariant [33] (the beta functions
are not) so that sigma model conformal invariance can always be rephrased as the vanishing of the beta functions up to terms which can be generated by diffeomorphisms or
gauge transformations. The general expression for the open string equations of motion is
thus [33]
I dx I I dx I + FI J M J dx I + dL = 0,
(4.1)
where M I and L are functions of FI J and its derivatives. Given the two loop beta function
we have computed, finding the two loop open string equations of motion is thus equivalent
to determining these unknown functions, M I and L, also at two loops.
An alternative way to finding the open string equations of motion is to compute the
string effective action directly, although this is not necessarily simpler than computing
the aforementioned Weyl anomaly coefficients. This calculation entails making use of the
boundary string field theory formalism [66,67], where besides the beta function one is also
required to compute the partition function of the sigma model theory at two loops (in the
superstring case one is only required to compute the partition function). Knowledge of both
these functionals then yields the open string effective action via the well-known expression

Z[FI J ].
S[FI J ] = 1 + M [FI J ]
AM
212
It is important to point out that either Weyl anomaly coefficients or partition function, one
must always compute to two loops and nonperturbatively in F . This is to say, any of the
aforementioned calculations will be nontrivial, and we shall leave them as the next step
in our general programme to compute curvature couplings in the non-Abelian BI action,
a step to which we hope to return sometime in the future. At this stage, we want to proceed
with the explanation of our strategy to compute these non-Abelian couplings given the
Abelian equations of motion. We shall thus assume we are in possession of the Abelian
equations of motion and outline the general procedure from there.
These massless open string mode Abelian equations of motion correspond to a commutative, or maximal brane, description of the BI action. Our goal now is to translate this
Abelian description of the BI action to a non-Abelian description, and this is done via the
SW map. The method we shall deploy has been studied at length in [2729] within the
context of the non-Abelian BI action in flat space, and has actually led to a detailed set of
constraints that could eventually make way to a solution of the non-Abelian BI. Here, we
shall begin by reviewing the strategy of [2729] as it generalizes to our situation of curved
backgrounds. We will spell out how one should proceed in order to move from the Abelian
to the non-Abelian description, for the case of parallel backgrounds, and will also comment on the possibility to constraint the result to possible field redefinitions and ordering
ambiguities.
4.1. On the SeibergWitten map
The main idea behind the SW map [2426] is that derivative corrections in the Abelian
action can be understood, at large field strength (or B-field) B on the Abelian D-brane
worldvolume, as corrections to the standard multiplication of functions so that multiplication actually gets deformed into a noncommutative star product. This is achieved via a
particular change of variables (from commutative to noncommutative fields) which precisely corresponds to the SW map [25]. The noncommutative description is then simpler
to understand as a matrix model, as one represents functions with operators, star products
with operator multiplication and worldvolume integrals get replaced by traces [28,68]. The
shift in notation is as follows

xA X A,
V (B) dx Tr,
where V (B) = det B(1 + ) is a volume form which makes the integral act as a trace
with respect to the star product. BI J is a symplectic form in the usual case of the Moyal
star product, and dots correspond to loop diagrams which contribute when the Poisson
structure is not constant [68,69].
Because we are going to higher order in derivatives than in [25], let us pause for a
moment and explain the distinction between the several star products one can consider
(see [25,28,68,69] for applications of the different possibilities). The standard BI situation
of constant B was dealt in [23,25] and leads to the well-known Moyal star product
i
f g(x) = e 2
AB x y
A B

f (x)g(y)x=y .
213
When B is no longer constant but the target is still flat space, there are derivative corrections
to the standard Abelian BI action and in order for the SW map to work properly the star
product should be written using Kontsevichs formula [28,69] (defined for nonconstant
Poisson structure)
1
i
f g = f g + AB A f B g AC BD A B f C D g
2
8

1 AL
BC
L (A B f C g B f A C g) + O 3 .
12
Kontsevichs formula can also be of use even when is no longer Poisson. This corresponds to the case where the target space is no longer flat [68]. We should point out that
in this paper we shall not use the point of view in [68]. Even though we believe that there
should be a definition of star product that will compute correlation functions at this order,
just like what happened at order O(H ) in [68], in here we wish to keep the background explicit in all formulas (using the approach of [68] the background would be fully translated
into the star product itself). Even so, there are derivative corrections to B arising from our
beta function calculation. This means that one should be careful when translating derivative corrections to star products as the correct star product to use could be Kontsevichs
one.
Let us begin by justifying our method in general terms [2729], explaining how the
SW limit takes us from the standard BI equations of motion to the noncommutative matrix
model equations of motion. Consider an Abelian action S, written in terms of a gauge field
AM (x), and a SW map given by (here, is the pull-back map)
B = F,
where BMN is constant, and F = B + 2 dA. Also, let the variation of S under AM (x)
be given by (defining the Weyl anomaly coefficients from an effective action point of
view)

S = Tp d n x M (x)AM (x).
If one writes the SW map as usual [25,26]
X M = x M + MN A N (x),
X M = x M ,
then one can also write the variation of the action in terms of a variation of the coordinates
X M , as

S = Tp d n x det B M X M = Tp d n x det F M X M .
At large B field, matrix model equations of motion can be easily translated into equations
depending on Poisson brackets rather than commutators (we shall see this in greater detail
in the following). So, matrix model equations of motion let us know about M . What is
then left to understand in the above expression is X M . We have [26]

X M (x) = AB A X M B X N AN X (x) ,
214
so that X M (x) = ( B1 )MN AN (x) and one obtains

1 MN
det F M
= N (x).
B
We have thus shown that in general it is always possible to move between Abelian and nonAbelian descriptions of D-brane physics. All one has to do is compare Abelian and nonAbelian equations of motion as expressed above. In practice, we shall see in the following
how to implement this reasoning.
In order to establish a proper perturbative framework, the first thing one needs to do is
to have a consistent set of conventions on dimensions of the various fields. We will take
the target fields gI J and BI J to be dimensionless. In this case FI J has length dimension
L2 . Next, turn to the beta function. In the standard perturbative calculation is taken
as a loop counting parameter, so that different orders in perturbation theory correspond to
different overall powers of . In order for things to properly match later on, we shall take
the following convention for the overall factors of in front of the -loop perturbative
Abelian beta function (the same holds true for the Weyl anomaly coefficients)
( )1 (1) + ( )0 (2) + + ( )2 () + .
Now, each () can still be expanded for large field strength F . This is due to the fact that
we are computing the beta function with a propagator which is exact in F . One will thus
have
() =
n()
,
( )n
n
where each n does not depend on . This deals with conventions for dimensions.
Let us begin with an illustration in the case of flat space [2729]. Start with the nonAbelian description and the standard quartic action
()
S=

1
Tr X M , X N X M , X N ,
4( )2
which has the matrix equations of motion

( )2 X M , X M , X N = 0.
Here and in the following, repeated indices are always summed (i.e., contracted with the flat
metric gAB ). If one now uses the star product to leading order, one has that [X M , X N ] =
i MN and [X M , [X M , X N ]] = i[X M , MN ] = MK K MN + . Observe that in here
MN ( F1 )MN and indeed has length dimension L2 as expected. In terms of the Poisson
structure the previous matrix equations of motion can thus be written as
( )2 MK K MN + = 0.
(4.2)
Let us now turn to the dual description and try to obtain the exact same result starting from
the beta function for the Abelian brane (at one loop there is no distinction between beta
function and anomaly coefficients, for the open string). First, one should take notice that
215
the open string tensors, at large F , behave as

MN

1
1 1 MN
1
1
g
=
+ =
MK KN + ,
lim

2
F G
(2 ) F F
(2 )2
MN
1
1
1 MN
MN
=
+ =
+ .
lim
(4.3)
F
2 F
2
Now write the contraction of the one loop beta function with the open string tensor
as (recall indices are raised with the open string Poisson tensor as one translates between
Abelian and non-Abelian equations of motion)
KL
1
(1)
K FLN ,
MN N = MN
G
and compute the large F limit as
(1)
lim MN N
F

1
1 MN
J K KL
J (2 FLN ) +
=

2
(2 )2
1
MN J K KL J FLN + .
=
(2 )2
Recalling the basic matrix fact that ( F1 ) = F1 F F1 it is now simple to obtain that the
beta function equation of motion, at large F , is precisely given by
( )2 KJ J KM + = 0,
which is the exact same expression we have previously obtained when starting from the
non-Abelian matrix description. This simple example thus illustrates the basics of our
method.
In conclusion, at large field strength F there is a map that allows us to obtain a matrix
action from a loop corrected, higher derivative, beta function result. We should point out
that given a generic loop corrected beta function it may be extremely hard to actually carry
out this programme. An advantage in our favor is, of course, the fact that we are dealing
with (WZW) parallel target space backgrounds, where the geometry simplifies immensely
as compared to the generic background situation. In order to make it easier for a determination of the non-Abelian equations of motion from the Abelian ones, we shall now
run through the expectations, based on general grounds, of which possible structures can
appear for the quantum corrected matrix model. Because we are starting from loop corrections to the Abelian action, it is only natural to expect that these non-Abelian structures
will also be organized in a consistent perturbative expansion, to which we now turn.
4.2. Perturbative classification of tensor structures
Let us consider parallel backgrounds. Generic monomials that can appear in the action
should be dimensionless and should thus have the generic form (the indices refer to powers
of H and X )
nm n m
H X .
216
Table 1
X2
X3
X4
X5
X6
H0
= 1 YES
= 2 OK
= 1 NOT
= 2 OK
H1
= 1 YES
= 2 OK
= 1 NOT
= 2 OK
H2
= 1 NOT
= 2 NOT
= 1 NOT
= 2 OK
H3
= 1 NOT
= 2 OK
H4
= 1 NOT
= 2 OK
Because we are doing a perturbative calculation, we need to understand how these monomials come into play as one carries forth the perturbation theory expansion. In order to
understand this issue with greater clarity let us focus for the moment on a WZW model at
level k. In this case, the
radius (characteristic size) of the group manifold target space is of
order H 1 R s k. On the other hand the radius (characteristic size) of the conjugacy
s
class (the D-brane inside the target space) is of order X n R

k n k , at fixed integer n.
The above powers of H and X thus behave as

nm n m
1 n+m
H X
.

k
What we learn is that there are two distinct expansions. The expansion, which is organized according to powers of n m; and the level expansion, which is organized according
to powers of n + m. Indeed this is to be expected [53]: one knows that in WZW models
the relevant low energy limit is a combined limit in which 0 and k . It is then
only natural that the perturbative expansion arranges itself in this double expansion. We are
now in a position to make a guess on the structure of the non-Abelian monomials which
will emerge from the beta function loop calculation. The perturbative structure we expect
to find is spelled out in Table 1.
We have listed above all possible monomials which can appear in the action, including
all candidates at both one and two loops. Because these are monomials in the action they
must have all their tensorial indices contracted so that not every possible entry is allowed
(we are not considering the H 0 X 2 monomial as our method only computes the potential
terms in the action, not the kinetic term to which this monomial contributes). Denoted
with a cross, , are monomials which cannot appear in the action (it is simple to see that
they would always have free indices). Denoted with dots, . . . , are monomials that will
only be relevant at higher loop level (as we will explain in the following). Each monomial
has a distinct power of n m associated to the expansion and a distinct power of n + m
associated to the level expansion. For the allowed entries, one needs to understand at which
loop level in perturbation theory will they appear and how could they contribute. Having
in mind the case of WZW models, where the level expansion organizes the perturbative
expansion, we can now make the following conjecture. New contributing monomials at
217
each order, , in perturbation theory correspond to diagonal lines of fixed n + m, where

the loop order is = n+m
2 1. So, the one loop beta function can contribute with new
monomials starting in the n + m = 4 diagonal and then all entries to the right of this
diagonal. Likewise, the two loop beta function can contribute with new monomials starting
in the n + m = 6 diagonal and then all entries to the right of this diagonal. Observe that the
keyword here is new monomials. What we mean by this is that, for instance, the two loop
beta function can also contribute to monomials to the left of the n + m = 6 diagonal but
it will not generate any new monomials with respect to the one loop resultit will only
generate derivative corrections to monomials which already exist from lower order results
in the perturbative calculation. As an illustration of this point observe that the way the two
loop results can affect the left side of the n + m = 6 diagonal is via derivative corrections
to the Kontsevich star product, in the monomials that were first generated at one loop. As
a consequence, if a monomial which should be first generated at one loop is in fact not
generated (i.e., it has coefficient zero from the string theory calculation), then there can be
no contribution to the corresponding entry arising from any higher order in perturbation
theory. In summary, the conjecture is that the () function, at loop level , contributes to
the action with terms of order k N with N 1 + , or n + m 2 + 2. Therefore, our
computation at two loops can reliably fix terms in the non-Abelian action with n + m = 4
and n + m = 6.
With these ideas in mind, let us then understand the remaining details in the table above.
Loop monomials are separated by diagonal lines and we have denoted with = 1 YES
terms which we know to exist from the one loop result (i.e., from the Myers action). These
have been previously computed from both Abelian [31,32] and non-Abelian points of
view [1,52]. In particular, we have denoted with = 1 NOT all the monomials which
we know not to be generated by the one loop beta function (see Appendix E). One of these
lies in the n + m = 4 diagonal and thus, according to the previous paragraph, there can
also be no contribution to this monomial from the two loop beta function. This fact we
have denoted with the label = 2 NOT. Finally, we have denoted with = 2 OK all
monomials which can have contributions from the two loop beta function. The contribution
of the two loop beta function to the monomials in the n + m = 4 diagonal will be only at
the level of corrections to the star product expansion. The contribution of the two loop beta
function to monomials in the n + m = 6 diagonal will be that of generating new monomials
with respect to the one loop result. The two loop beta function may also generate terms to
the right of the n + m = 6 diagonal. These terms would play a distinctive role in a three
loop calculation, a subject on which we shall have nothing to say. Let us comment on the
various terms which can be generated at one and two loops:
The H 0 X 6 term starts at two loops and corresponds to Tr F 3 in the non-Abelian BI
action for open bosonic strings. It arises from a matrix term of the type (sum over
repeated indices)
3

1
Tr X M , X N X N , X L X L , X M ,
and the result one should thus expect to see from the two loop beta functioni.e., in
the equations of motionis of the type (more on this later)
218
(2) +
1
3

N L L N K KM + .
Observe that the normalization of Tr F 3 in the non-Abelian BI action is known, so

that the full result for the two loop equations of motion should reproduce the above
expectations, including all factors. A detailed discussion of this term is included in
Appendix F.
The monomials H 1 X 5 , H 2 X 4 , H 3 X 3 and H 4 X 2 also start at two loops. The only
contribution from the star product to these monomials should be the leading one, i.e.,
all one should worry about is f g = f g + 2i MN M f N g, higher order terms not
being necessary and only making an appearance at three loops and higher. In particular
one need not worry about higher order derivative corrections to the Kontsevich star
products for these monomials.
The term H 2 X 2 is not present at one loop (see Appendix F), even though dimensional
analysis says it could have appeared. One thus concludes that this monomial should
also not be present at two loops, since the two loop calculation should contribute to
new monomials H n X m only for n + m 6.
The terms H 1 X 3 and H 0 X 4 start at one loop. They are the well-known terms in the
Myers action. One thus expects that the two loop result should only correct the star
product in these terms from Moyal to Kontsevich. We expect no new monomials arising from the two loop calculation. For example, the equations of motion for the term
H 1 X 3 could have corrections of the form

HABC X B X C = HABC BC + M BN N CM + ,
where the first term arises from (1) and the second from (2) with some constant
to be computed (and which will yield information on the star product). The dots are
possible higher loop corrections.
This is the generic set up for obtaining the non-Abelian equations of motion. The procedure itself is straightforward to implement: given the loop beta function result, one
should start by expanding all possible closed string background fields, gI J , BI J and IKJ ,
in terms of the local coordinates, x A (including the closed string fields which are present
inside the open string tensors). This must be done in order to properly identify the matrix
model, as it will depend on the noncommutative version of the coordinates. The expansion
of these tensors is most simply done in Riemann normal coordinates (RNC) (see Appendix A). Once the expansion is done, one should then write the beta function in powers
of I J (which is to say, powers of F1 ). These steps should not entail any difficulties. The
final step is clearly the trickier as one needs to rewrite the derivative expansions in the
beta function in terms of star products. Here, one will need to reorganize monomials as
X A1 X An . A point that may make this calculation slightly easier is the fact that,
as we have just seen when listing the possible monomials appearing in the action, most
of the time one is actually able to get along only with the leading terms in the star product.
219
4.3. Field redefinitions and ordering ambiguities

One thing one should always keep present in string theory is that the coefficients of the
low energy effective action, for the massless string modes, may be ambiguous. Here, we
shall have a pragmatic approach on this issue. Of all the structures which can appear in
the effective action, based on the general arguments we have just seen, we wish to find out
which ones can or cannot be removed via field redefinitions. The final set of structures with
unambiguous coefficients will be our result for the corrected non-Abelian action.
As we have seen, there are five monomial structures arising at order n + m = 6, which
are expected to make their appearance in a two loop calculation in the Abelian theory.
These monomials correspond to specific tensor structures and, because they are matrix expressions, there are matrix ordering issues to be dealt with when classifying the structures
that correspond to each of the two loop monomials. In other words, we would now like to
have a list of all the possible tensor structures (including all possible orderings) that can
appear from the two loop calculation. For each of the five monomials in the action we have
classified the possible tensor structures (Table 2).
Repeated indices are summed and an overall trace is implied. Note that for the H 2 X 4
column other possible structures one could write down are not independent as the background field H satisfies a Jacobi identity. Also, we have omitted the H 4 X 2 column, as we
shall briefly see that it does not contribute to the action. One still has to classify the independent tensor structures in the monomial H 0 X 6 . Here we will make use of a property we
shall explain later: the action should be translation invariant. As we shall see shortly, this
amounts to the requirement that only commutator structures can appear in the classification
of this term (a fact which should not come as a great surprise since this term is the known
Tr F 3 coupling in the BI actionsee Appendix E). The independent structures thus are
M N N S S M
X ,X
X ,X X ,X ,
A B C A B C
X , X ,X
X , X ,X
,
A B C B A C
X , X ,X
(4.4)
X , X ,X
.
Let us stress once more that there are no issues concerning orderings of matrices left to
resolve: we have listed all possible matrix orderings in the previous classification of tensor
structures.
Having solved the issue of ordering ambiguities in the two loop monomials, we now
turn to the ambiguities arising from field redefinitions. We shall begin by analyzing the
Table 2
H 3X 3
HAMN HBML HCN L
H 2X 4
X AX B X C
HABM HCN L HMN L X A X B X C
H 1X 5
X AX B X C X D
HABC X A X B X C X D X D
HAMN HBMN X A X C X B X C
HABC X A X B X D X C X D
HABM HMCD
H 2 HABC X A X B X C
X AX B X C X C
HAMN HBMN
H 2X AX B X AX B
H 2X AX AX B X B
220
constraints due to translation invariance of the matrix action. Generically speaking, an

action is an integral (i.e., trace) over spacetime, and integrals have the property of translation invariance. This is a feature we would thus like to see made explicit in our matrix
actionnot all of the above tensor structures will be invariant under translations; indeed
one expects that commutator structures must be present in order for such a property to hold.
Let us begin with a slightly more general point of view, and let us analyze what happens to
monomials H n X m under a change of coordinates X M X M + M (X ), where M is the
vector generating the diffeomorphism (see [11,15] for further discussions on these issues).
The reason to start with this more general analysis is the following: the translational invariant tensor structures at one loop could, under the above field redefinition, generate new
terms in the action whose role would be to affect the listed tensor structures at two loops in
such a way as to make the whole action translational invariant, i.e., in such a way that one
would no longer need the requirement that the two loop structures need to be translational
invariant by themselves. We shall see that this, in fact, does not hold truethe two loop
tensors must also have the property of translational invariance and thus be expressed via
commutators.
Recall the matrix model we are computing is set up in RNC, so that this fact must
still remain true after the above diffeomorphism. The geometric requirement associated
to the coordinate change is thus that we move from RNC to RNC. If one recalls that the
metric transforms as gMN gMN M N N M , and if one uses the formulas in
Appendix A, it is simple to check that the requirement of having RNC for the transformed
metric translates to a constraint on M which itself translates to the fact that the field
redefinition acts on the coordinate matrices as
1
X M X M + M (0) N (0)HN A L HLB M X A X B + ,
6
and on the monomials present in the matrix action the effect of the above constant and
quadratic terms in the field redefinition is H n X m H n X m1 and H n X m H n+2 X m+1 ,
respectively. There are, of course, higher order terms in the above transformation which
generate many other possibilities, but for our two loop expansion this is all we care about.
This action allows for two sets of movements in Table 1: moving one box to the left
at fixed horizontal line as H n X m H n X m1 , or moving one box to the right and two
boxes down as H n X m H n+2 X m+1 . As a consequence, it is simple to observe after a
glance at Table 1 and immediate application of the two rules above, that none of the one
loop monomials (the n + m = 4 diagonal) can affect (under these movements) any of the
two loop monomials (the n + m = 6 diagonal). Moreover, for the two loop monomials,
the H n X m H n X m1 movement cannot produce any effect and this immediately implies that the full two loop diagonal must consist of monomials which are translational
invariant. Thus, translation invariance requires that the X M coordinates are organized into
commutator structures. We shall now return to that classification and check which (if any)
of the structures are translation invariant (recall that the H 0 X 6 term was already dealt with
in a translation invariant waythus the commutator structures) (Table 3).
The first thing one notices is that the number of independent structures has been narrowed down. For the monomial H 4 X 2 one can check that either the possible tensor structures are schematically of the form SymTensor| X X and thus vanish, or the trace ends
221
Table 3
H 3X 3
H 2X 4
HAMN HBML HCN L
[X A , X B ]X C
HABM HCN L HMN L
[X A , X B ]X C
H 2 HABC [X A , X B ]X C
H 1X 5
HABM HMCD
[X A , X B ][X C , X D ]
HABC [X A , X B ][X C , X D ]X D
HAMN HBMN
[X A , X C ][X B , X C ]
H 2 [X A , X B ][X A , X B ]
up killing the commutator term, so that none of the tensor structures for this monomial
leads to translation invariant terms. A similar reasoning applies to other structures. Altogether, we have managed to reduce the apparently very complex problem of determining
the non-Abelian BI action in a curved parallel target space, at two loops, to the [still complicated but simpler] problem of determining two constants at one loop and ten constants at
two loops. The constants to be determined are precisely the coefficients of the mentioned
tensor structures as they appear in the action and as determined, e.g., via the beta function
calculation we have set up. Of course the value of the coefficients may still be zero and
further simplify the action.
There is one further point to mention. We have constrained the action to field redefinitions associated to translational invariance. This is a geometric constraint based on
general expectations for the form of an effective action. String theory, however, also allows for completely generic field redefinitions and this may in fact imply that some of
the two loop coefficients could still be ambiguous. Let us now try to further understand
this point. The most general field redefinition one can perform on the coordinate matrices
X M X M + X M in target WZW parallel backgrounds consists of terms in the following
monomials
X H X 2 H 2X
X 3 H X 4 H 2X 3
X 5 H X 6 H 2X 5
..
..
..
.
.
.
In this generic field redefinition, only movements to the right and movements down
are allowed in Table 1. This implies that field redefinitions of two loop structures will only
affect higher order structures (for which we care not). On the other hand, field redefinitions of one loop structures may affect the two loop tensors. This can happen by action
of the X 3 , the H X 2 and the H 2 X monomials in the list above (the remaining infinite set
of possibilities will only contribute at higher order). Indeed it is simple to observe that the
redefinition X X 3 will produce a shift H n X m H n X m+2 , which corresponds to moving two boxes to the right at fixed horizontal line, the redefinition X H X 2 will produce
a shift H n X m H n+1 X m+1 , which corresponds to moving one box to the right and one
box down, and the redefinition X H 2 X will produce a shift H n X m H n+2 X m , which
corresponds to moving two boxes down at fixed vertical line. In practical terms, some of
the two loop tensor structures may thus be washed away. The redefinitions in question are
properly written as

X M X N , X N , X M ,

X M HMN S X N , X S ,
222
Table 4
H0
H1
H2
H3
H4
X2
X3
X4
X5
X6
X M H 2 X M ,
X M HMN L HN LS X S ,
their form naturally constrained by translational invariance requirements (i.e., while the
field redefinitions by themselves need not be translation invariant, they must still be such
that the action will remain translation invariant after we have performed the redefinition).
This freedom to perform four completely generic field redefinitions thus removes four
coefficients from the possible ten coefficients at two loops. At this stage we must pay some
attention to the known results concerning the non-Abelian BI action in flat backgrounds.
Here we know that the X 3 redefinition is used to remove one of the three H 0 X 6 structures,
and we further know that once this has been done, the coefficient of the [X , X ]3 term is
fixed and nonzero while the coefficient of the remaining [X , [X , X ]]2 term is fixed and
zero. We will choose to do this very same field redefinition in curved space, to enjoy the
fact that the H 0 X 6 coefficients will then be all known. We shall further make use of the
H X 2 redefinition to remove the H 1 X 5 one possible tensor structure. Then, we shall make
use of the two H 2 X field redefinitions to remove one of the three tensor structures at order
H 3 X 3 and to remove one of the three tensors structures at order H 2 X 4 . We are thus left
with Table 4.
Here it is important to realize that both one loop coefficients are known as are both
H 0 X 6 coefficients (of which one of them is zero). So, at this stage, we are only left with
four unknown coefficients. Of course an unambiguous determination of these four coefficients can only be achieved via a string theory calculation as the one we have set up in
the present paper. We shall next see that, when considering Myers dielectric effect, the situation gets even better. In such a context there are only three coordinate indices, making
the two H 3 X 3 tensor structures to be equal, and likewise for the two H 2 X 4 structures.
Thus, when studying quantum corrections to the dielectric effect we are only left with two
unknown coefficients.
4.4. Matrix model for quantum dielectric branes
We shall now concentrate in the seminal example of [1] (see also [52,53]). For Myers dielectric effect we only need to select three coordinate matrices, with indices
{I, J, K, . . .} {i, j, k, . . .} = {1, 2, 3}. This is of course in strict correspondence with the
example of the SU(2) WZW model at level k [52,53]. In this case Hij k = H ij k , where ij k
is the totally antisymmetric Levi-Civita symbol, and where H 1 . Let us also stress
k
once again that due to the RNC expansion of target tensors, all background tensors which
appear in the matrix model are c-numbers.
223
In this three-dimensional situation both the H 3 X 3 and the H 2 X 4 monomials collapse

to only two independent tensor structures. It simply follows that
Haij Hbik Hcj k = H 3 aij bik cj k = H 3 aij (bc ij bj ic ) = H 3 acb = H 3 abc ,
Habi Hcj k Hij k = H 3 abi cj k ij k = H 3 abi 2ci = 2H 3 abc ,
H 2 Habc = H 3 ij k ij k abc = 3!H 3 abc ,
Habi Hicd = H 2 abi icd = H 2 (ac bd ad bc ),
Haij Hbij = H 2 aij bij = 2H 2 ab ,
H 2 = H 2 ij k ij k = 3!H 2 .
This means that we are left with only two unknown coefficients for the matrix model which
describes corrections to the dielectric brane solutions of open string theory. These are
the coefficients associated to the following two curvature couplings

H 3 ij k Tr X i , X j X k and H 2 Tr X i , X j X i , X j .
Let us finally, carefully understand how everything comes together in the construction
of the matrix model for quantum dielectric branes. First, recall that for the one loop Myers
matrix model the energy consists of two tensor couplings with well-known coefficients (we
shall explicitly display powers of ) [1]
(1)
Veff
[X ] =

1
i
H ij k Tr X i , X j X k .
Tr X i , X j X i , X j +
2
6
4()
The equations of motion which follow are

i i a i
+ H aij X i , X j = 0,
X , X ,X
2
where an ansatz for a fuzzy solution of the type [X i , X j ] = Aij k X k leads to
i
2A2 ai X i + 2AH ai X i = 0
2
i
A = 0 A = H.
2
Picking the nontrivial solution, it is simple to compute its energy as Veff [Xfuzzy ] =
2
H24 Tr X i X i < 0, so that the stable, minimal energy, solution corresponds to the famous
Myers dielectric effect [1].
Proceeding at two loops, one has three tensor couplings. The H 0 X 6 coupling has
well-known coefficient, fixed via the non-Abelian BI action for the bosonic string (see
Appendix F). The other two tensor couplings have undetermined coefficients. One can
write the two loop contribution to the quantum matrix model as
1
(2)
Veff [X ] =
6 2 ()3

Tr X i , X j X j , X k X k , X i

1 2 i j i j i2 3
H Tr X , X X , X +
H ij k Tr X i , X j X k
4
6
224
with unknown coefficients 1 and 2 . These coefficients can only be obtained once the beta
function programme is carried out to its very end. The energy thus becomes

1
1 + 1 H 2 Tr X i , X j X i , X j
2
4()

i
H + 2 H 3 ij k Tr X i , X j X k
+
6

1
Tr X i , X j X j , X k X k , X i .
2
3
6 ()
Veff [X ] =
(4.5)
One can moreover compute equations of motion as usual and, in this case, it is not too hard
to get

i

1 + 1 H 2 X i , X i , X a + H + 2 H 3 aij X i , X j
2
1 i i j j a
X , X ,X , X ,X
= 0.
2 2
(4.6)
This is our final result for the quantum matrix model describing the first nontrivial curvature
corrections to Myers dielectric effect. It is definitely clear that, in the end, the answer is
much simpler than we first expected when set out to do this calculation. On the other hand,
the programme we have spelled in this work still needs completion, as one should soon
address the question of determining the unknown coefficients 1 and 2 . A parallel path
of research should also try to obtain solutions to the above quantum matrix equations,
possibly from a quantum group deformed fuzzy sphere point view (perhaps along the lines
of [6062]). The one thing which remains clear is that there is still much to learn on what
concerns the physics of dielectric branes.
4.5. Mapping non-Abelian to Abelian branes
In order to set the pace for the full calculation we have described, regarding two loop
tensor structures in the non-Abelian BI action, we shall conclude this section by exemplifying how to compute some of these structures in the flat space limit. The final calculation
of all structures, thoroughly explained earlier, will be left for future work. We shall begin
with the non-Abelian result, at the level of the equations of motion, and obtain an Abelian
result which is necessarily nonperturbative in F . These Abelian tensor structuresderived
from the non-Abelian actionmust be present in the two loop Abelian equations of motion. Comparison of these terms with the Abelian equations of motion then yields the
required coefficients in the non-Abelian matrix action, as we have previously described at
length.
The procedure is similar to the one at the beginning of this section, where we have addressed the quartic YangMills term in the non-Abelian action. Indeed, it was quite simple
to write the matrix equations of motion associated to this term at large F , as (recall (4.2))
1 M M N
1
X , X ,X
= 2 MK K MN + = 0.

2
( )
( )
225
In order to obtain the Abelian results (rather than the non-Abelian, as we exemplified earlier) one now needs to rewrite -derivatives as F -derivatives, obtaining
1 N T RM MS
R FST + = 0,
( )2
and finally, making use of the limits (4.3), one can conjecture the Abelian term in the
equations of motion as being (this is not guaranteed to be the exact result to all orders in F ,
as our procedure only fixes the large F behavior uniquely)
RS
1
4 2 N T
R FST + = 0.
G
Of course at one loop all is simple. At two loops the structures get more intricate, but the
process of computation is the same. In flat space, the non-Abelian action is known (see
Appendix F)

1
Tr X I , X J X I , X J
4( )2

1
Tr X I , X J X J , X K X K , X I ,
2

3
6 ( )
Veff [X ] =
with the matrix equations of motion

M M K
X , X ,X
1 M M N N K
X , X ,X , X ,X
= 0.
2 2
Following the exact same procedure as before (albeit via a much cumbersome calculation),
one can then conjecture the contribution of the previous two loop structure to the Abelian
equations of motion as being

1
2 2 ( )3

X M, X M,X N , X N,X S

1 M M 1 N N AB
= 8
M (A FMN
B FN S )
G
G
M M N N
1
1
AA B B
G
G

M FAB
A FMN
B FN S + A FN S A FBN
M FMB
2
S S
+ A FN S A FMB
M FBN
+ A FMN
A FB S M FN B

+ .
+ A FMN
A FN B M FB S
This is an amusing expression as it can also yield a conjecture on how to obtain the Weyl
anomaly coefficients at two loops, in flat space. Indeed, if one writes the expression for the
two loop beta function, (3.4), in flat space and subtracts it from the result above (which is
in fact a conjecture on the final expression for the two loop Weyl anomaly coefficients),
then, and according to (4.1), whatever is left is the completion of the beta function to yield
226
the Weyl anomaly coefficients. As we have previously explained at greater length, computing this expression from first principles in sigma model theory (or in boundary string field
theory), is the most pressing problem concerning the full calculation of the non-Abelian
BI action.
5. Future directions
Of course the most pressing question arising from our work concerns the full Abelian
equations of motion. The beta function we have computed is the required first step, but
one will eventually have to address either the Weyl anomaly coefficients or the partition
function, as we have explained in the text. Only after this is done our programme can be
brought to conclusion, following the guidelines we have spelled out in the paper. Much
remains to be learned from the non-Abelian BI action in curved space, and in particular in
what concerns corrections to the dielectric effect. Parallel to this, we believe that other
lines of research can follow along the lines presented in this work, and end with a few
suggestions for future research:
One obvious generalization concerns the supersymmetric case. This is not too hard as
the RNS sigma model is well known to order O(H 2 ) [49] and one could easily perform
the same type of calculation we perform in here, for the quantum dielectric branes of
type II string theory. In this regard, the result in [43] may be particularly relevant, as
the R 2 open string effective action is presented to all orders in the boundary field F . It
would be interesting to investigate what sort of corrections one obtains in superstring
theory.
Another reason why the study of superstrings is of interest is, of course, the study of
the quantum dielectric effect due to the presence of RR background fields (this was the
original example in [1]). At first sight this may seem like a hard problem, as RR fields
are notoriously difficult to handle. However, the recent work in [71] may be of help as
it lays the ground to a noncommutative analysis (as the one we perform here) in the
realm of RR string theory backgrounds.
Once we are extending these results to other situations, one which is of obvious interest
is for the dielectric effect on the fundamental string itself. This was studied in the
context of matrix string theory [72] in [73,74]. It would be interesting to compute
quantum corrections to those results and investigate what sort of deformations they
yield on F -strings.
A harder problem would be to tackle the full BCFT. Here we only point out that the
extension of the results in [68] to higher orders in H may provide for a full algebraic
description of the open string massless fields embedded in curved closed string backgrounds via deformation theory (i.e., in terms of traces and star products).
The method described in this paper is, in a certain sense, a T-duality shortcut, as one
is moving from a maximal brane to a minimal brane in one unique step. It would be
interesting to use the results in this paper to derive corrections to the open string
T-duality rules. This would be helpful, for instance, in generalizing the beta function
we have obtained for the maximal D-brane to arbitrary Dp-branes.
227
If one would obtain such corrected T-duality rules, one could then envisage in extending this action to nonconformal backgrounds. T-duality transformations of closed
string fields in nonconformal backgrounds turn out to be compatible with renormalization group (RG) flows of the two-dimensional sigma model field theory [7577].
Extending this work to open strings is harder [78,79], but it could prove of interest to
generalize these works to higher orders: a possible solution, at all orders, for the Tduality rules would, using the compatibility of duality and the RG flow, possibly allow
one to uniquely determine the solution, at all orders, for the beta function equations of
motion.
Acknowledgements
P.B. wishes to thank S. Ribault and K. Graham for useful discussions. R.S. would
like to thank Troels Harmark and Niclas Wyllard for useful discussions and comments
at different stages of this project. P.B. is supported by the grant SFRH/BD/799/2000 (Portugal). L.C. is supported by a Marie Curie fellowship, under the European Commissions
Improving Human Potential Programme (HPMF-CT-2002-02016). R.S. is supported in
part by funds provided by the Fundao para a Cincia e a Tecnologia, under the grant
SFRH/BPD/7190/2001 (Portugal). L.C. and R.S. are supported in part by funds provided
by the Fundao para a Cincia e a Tecnologia, under the grant POCTI/FNU/38004/2001
(Portugal).
Appendix A. Perturbative expansions

In order to set up a two-dimensional sigma model perturbation expansion of the quantum fields in the open string action, some sort of geometrical expansion is required on the
target manifold fields (the sigma model couplings), both for the closed and the open string
sectors. The standard expansion employs the background field method [3032,49]. There
is, however, a different method based on the radial gauge expansion of the target fields (the
Riemann normal coordinate (RNC) expansion for the metric) [80,81], which in some cases
is better suited to the problem at hand (see, e.g., [68]). We wish to explain both methods in
here, as applied to our problem. In the paper, we shall use the background field method in
the main computation, but the RNC expansion also proves to be of interest for subsequent
calculations.
Let us begin with a bosonic string action in a background metric g (X) and NSNS
2-form field B (X),

1
i
S=
(A.1)
g
(X)
dX
dX
+
B (X) dX dX ,
4
4
where is the Hodge dual on the worldsheet . The factor of i signals a choice of euclidean signature on the worldsheet. This action is to be supplemented with a boundary term
228
as we are describing the open bosonic string action

SB = i A (X) dX .
(A.2)
Throughout the paper we shall consider backgrounds with constant dilaton field, =
const, and will disregard this field. This we can do as we will be working in maximally
symmetric target spaces. The background equations of motion for the closed bosonic string,
to first order in , are
1
R + H H = 0,
H = 0.
4
The connection which is mostly used is the one emerging from the equations of motion of
the worldsheet string action, which is = 2i H (we shall consider the plus sign in the
following). The curvature of the connection follows as
i
i
1
1
R = R + H H + H H H H .
2
2
4
4
For parallelizable manifolds (parallelizable by inclusion of torsion, which we pick to be
the field H ), the connection is flat (but torsionfull). In this case one has R = 0,
which implies
1
H = 0,
H[ H ] = 0.
R + H H = 0,
4
Here is the covariant derivative with respect to the Levi-Civita connection. This solves
the closed string background equations of motion to all orders, in both the bosonic and
supersymmetric cases [49], so that the closed string fields will be on-shell throughout, as
we analyze open strings in this class of backgrounds. Parallelizable manifolds are a large
class of manifolds which include in particular all Lie groups (of course with the same
choice of torsion). This means that our results will hold for generic WZW models.
While the most common expansion of the metric field in covariant tensors is probably
the RNC expansion, the background field method is the one we shall employ and to which
we first turn. The fourth order background field expansion for the worldsheet bulk fields
was done in [49] and later generalized to worldsheet boundary fields in [30,33], where it
was precisely applied in beta function calculations. We shall briefly review this method and
show how to apply it to the open string sigma model (A.1) and (A.2) to quartic order in the
fluctuations (which is the required expansion in order to perform the two loop calculation
of the open string beta function in closed string background fields).
We shall take for worldsheet the disk = D, parameterized by polar coordinates and
with a flat euclidean metric. Greek indices are assigned to spacetime, Latin indices are
for the worldsheet and 01 = 1. The aim of the background field expansion is to maintain
target general coordinate covariance explicit both at the level of the sigma model local
couplings and at the level of the fluctuating field (which will become a tangent vector
in target space). The two-dimensional quantum field X is written as a perturbation of
a fixed classical background field x ( ) by a quantum fluctuating field ( ), such that
X ( ) = x ( ) + ( ). The idea is then to change coordinates to the tangent vector
229
along the geodesic linking x to : consider the geodesic from the background field x to
the background field plus fluctuation x + , which will be denoted by (s) (with s
[0, 1]) such that x = (s = 0) and x + = (s = 1). The tangent vector to this geodesic,
equation
(s) =
d
(s),
ds
at s = 0 is a target manifold vector field. This is the quantum field in powers of which we
shall expand the action. Because it is a precise geometrical entity in target space, one is
guaranteed that the perturbative expansion will be in terms of target tensor fields (evaluated
at the classical background, x). It turns out that, at least up to order O( 4 ), this expansion
can be written largely in terms of the generalized Riemann tensor R . The procedure
to carry out such an expansion is to consider the action (A.1) as depending on the geodesic
field

1
S (s) =
g (s) d (s) d (s)

4

i
B (s) d (s) d (s),
+
(A.3)

4
such that the diverse terms in the perturbative expansion of the action (given the split of
the field into background and fluctuation), as S = S (0) + S (1) + S (2) + , will be given by
their power series expansion coefficients

1 dn
(n)
S =
S (s) .
n! ds n
s=0
The exact same reasoning naturally applies to the boundary piece of the open string action

SB (s) = i A (s) d (s).
(A.4)
For the closed string worldsheet, this expansion is carefully carried out in the appendix
of [49]. For the open string one needs to take care with partial integrations in order not
to miss any boundary contributions. This has been carried out in [30] to quadratic order
and later in [33] to quartic order but in flat target backgrounds. Here we wish to obtain
the full open string action to fourth order in curved backgrounds. Let us first concentrate
on (A.3), in particular in the boundary terms that will be generated from this worldsheet
bulk contribution.
At zeroth order one obtains just the classical constant action (here and in the following,
all target fields are evaluated at x)

1
i
S (0) =
g
dx
dx
+
B dx dx .
4
4
230
At first order the action is linear in , and includes a boundary contribution from a partial
integration

1
d d g Da a x
S (1) [ ] =

2

1
+
d g x + iB x ,
2
where we have introduced worldsheet coordinates {, }, derivatives Da = Da +
i
b

2 H ab x and Da = a + a x , and where (in order to cancel tadpoles) the bulk piece yields the background equations of motion, Da a x = 0, while the
boundary piece yields the boundary conditions on the background field, (g x +
iB x )| = 0. Likewise at quadratic order one obtains the expected bulk contribution [49] with a new boundary term

1
(2)
d d g Da Da
SWZW [ ] =

4

i
+
d B x + B D .

4
In here, is the target covariant derivative associated to the Levi-Civita connection. We

have also particularized for the case of parallelizable backgrounds where R (the curvature of the connection) vanishes. In this particular case, to third order one obtains the
standard bulk term [49] with a new boundary contribution, which we have computed as

i
(3)
SWZW [ ] =
d d H ab Da Db
12

i
1
+
d B B H H x
12
4

i
+
d B D .
6
Proceeding to fourth order (which is required in order to compute the full two loop beta
function), we have computed the following contribution (where again the bulk piece had
been previously computed in [49])

1
(4)
d d H H Da Da
SWZW [ ] =
48

i
3
+
d B B H H x
48
4

i
1
B
+
d
H
H
D .

16
12
231
This concludes the analysis of the expansion of (A.3) to the order we need in order to perform our beta function calculation. One still needs to address the open string action (A.4).
This term is expanded in the exact same way as the previous one so that we shall limit ourselves to presenting the results. We have performed the expansion in a completely generic
(parallelizable or not) background and found

(0)
SB = i d A x ,
(1)
SB [ ] = i
d F x ,
(2)
SB [ ] =
i
2

d F x + F D ,
i
SB(3) [ ] =
6

d F + F R x
i
+
3
d F D ,
and
(4)
SB [ ] =
i
24

d F + 3 F R + F R x
i
8

1
d F + F R D .
3
When specialized to parallelizable backgrounds and added to the previous bulk plus boundary expansion, it is simple to check that it just amounts to the replacement B
B + 2 F in the WZW action, which is precisely what one expects from generic
D-brane arguments. It is this resulting perturbative expansion that we use throughout the
text for the beta function computation.
This is the essence of the background field expansion. However, once one has managed
to carry out the string perturbative calculations, it is still necessary to expand the target
tensors in order to obtain non-Abelian matrix actions. This is now a purely target space
geometric problem, where the most common expansion of the metric field in covariant
tensors is probably the RNC expansion, defined via the radial gauge prescription for the
metric field
X g (X) = X g (0),
(X) = 0. We shall now turn to analyze this expansion method in
or equivalently, X X
the backgrounds we are interested in. This radial gauge prescription can be applied to the
NSNS B-field as (see, e.g., [68])
X B (X) = X B (0).
232
While for the metric field the solution of the radial gauge equation is a bit more complicated
(see, e.g., [80]), for the NSNS field one can solve in H = dB as [68]
1
B (X) = B (0) + X
s 2 H (sX) ds.
0
For the metric field there is an algorithmic process that computes the RNC expansion to
any desired order, e.g., [80,81]. The expansion to quartic order in the coordinates is (see
[80,81] for higher order terms)
1
1
g (X) = g + R X X + R X X X
3
6
1
2
+ R X X X X + R R X X X X + .
20
45
One can likewise obtain an expansion for the Levi-Civita connection coefficients, which is
as follows

1
(X) = R + R X
3

1
1
1
1
R + R R R R
6
2
2
2
X X + .
For the special case of symmetric manifolds where R = 0 the RNC expansion
for the metric field simplifies to a closed form expression (see, e.g., [81])
+
g (X) = g +
g +
1 22k+2
f1 f 1 2 f 2 3 f k1
2
(2k + 2)!
1
2
k=1
+

k=1
22k+2
(f )k ,
(2k + 2)!
where we have defined

f1 = R1 X X ,
f i i+1 = R i i+1 X X ,
f k1 = R k1 X X .
If one further deals with (WZW) parallelizable manifold backgrounds, where both
R = 0 and R + 14 H H = 0, the previous formula can be written as
+

(1)k
M1 M 1 2 M 2 3 M 2k1
g (X) = g + 2
(2k + 2)!
k=1
g + 2
+

k=1
i 2k
(M )2k ,
(2k + 2)!
233
where we have defined the antisymmetric matrix M H X , to appear again in the

following. In this particular case of WZW backgrounds, the Levi-Civita connection coefficients can also be simplified to

1
H H + H H X + .
(X) =
12
Observe that in this expression it is not very useful to use the matrix M . This is clear
also from the above expression for the metric: when computing the connection coefficients
there will always be a free floating H piece, due to the derivatives of the metric. One
can develop similar methods to handle the NSNS field. The main concern here is to obtain
a closed form expression valid for (WZW) parallelizable backgrounds. If one computes
generic symmetrized derivatives of H = dB to second order, and in RNC at the origin
H (0) = H ,
( ) H (0) = ( ) H + 3( )[ H ]
= ( ) H + R ()[ H ] ,
...
one can then use the previous definition of B in terms of H to compute the expansion to
cubic order in the coordinates
1
1
1
B (X) = B + H X + H X X + H X X X
3
4
10
1
+ R [ H] X X X + .
15
For the special case of WZW backgrounds, where besides the previously stated conditions
on the curvature one has H = 0, and where one can write derivative expansions
(in RNC) for (1 2 ) , (1 2 3 ) , . . . , and higher symmetrized derivatives [80,81], it

follows that one can then compute symmetrized arbitrary derivatives of the H field (an odd
number of symmetrized derivatives yields zero)
( ) H = R ()[ H ] ,
3
2
( ) H = R (| R | )[ H ] + R (| R | ) H
5
3
2
2
+ R (| R | ) H + R (| R | ) H ,
3
3
...
Finally, one can use the definition of B in terms of H to compute the expansion to fifth
order in the coordinates, obtaining
2
2
2
M M M M M M M M M .
3!
5!
7!
For the (WZW) parallelizable backgrounds one can then guess the following closed form
expression for the expansion of the NSNS field
B (X) = B
B (X) = B 2
+

k=1
1
(M )2k1 .
(2k + 1)!
234
This concludes the analysis of the closed string sector (A.1). Using these expansions
and further expanding the quantum field X in a zero mode and a quantum fluctuation,
X = x + , one can devise a graphical procedure to perturbatively compute correlation
functions in the open string BCFT (this procedure was done to order O(H ) in [68], which
naturally generalizes to all other orders). Observe that the matrix M will also include a
term in the fluctuating field, , thus generating interacting vertices of the two-dimensional
bulk quantum theory. If one wishes to include the boundary gauge field in this expansion
one further needs to deal with (A.2).
In gauge theory radial gauge is also known as FockSchwinger gauge, where one aims
at explicit covariance in perturbative calculations. Just as before, this gauge can be defined
via the condition
X A (X) = X A (0),
and just like for the B field one can also solve in F = dA as
1
A (X) = A (0) + X
sF (sX) ds.
0
One now proceeds as before. In RNC we know how to compute arbitrary symmetrized
derivatives of the metric connections, (1 2 ) , (1 2 3 ) , . . . , [80,81], so that one

can proceed to compute symmetrized derivatives of the covariant F field, obtaining
F (0) = F ,
2
( ) F (0) = ( ) F R ()[ F] ,
3
( ) F (0) = ( ) F 2R ([ ) F] ,
...
In this way the previous integral formula for the gauge field can be perturbatively expanded
in terms of target tensors, thus obtaining a covariant perturbative expansion as
1
1
1
A (X) = A (0) + X F (0) + X X F (0) + X X X F (0)
2
3
8

1
+ X X X X F (0) + O X 5
30
1 1
X X X R (0)F (0)
6 4

1
+ X X X X R (0) F (0) + O X 5 + O R 2 .
5
This is the essence of the radial gauge procedure. This method was employed in [68] for
computing general correlation functions of open string fields in curved backgrounds. It is
employed in this paper for expanding background fields in the matrix model limit (and we
can also envisage that it could be of interest for computing correlation functions in more
general backgrounds than the ones in [68]).
235
Appendix B. The propagator on the unit disk

The beta function calculation we perform in this paper is done with worldsheet coordinates on the unit disk. Because most previous works on these subjects deal mainly with the
upper half plane computation, we address in this appendix our conventions for the free field
CFT propagator on the disk. Let us start by considering the two point function, defined by
AB (z, w) =

1 A
(z) B (w)
on the unit disk in the complex plane C. This is a symmetric function AB (z, w) =
BA (w, z) of its arguments. If one considers AB as a function of z (with w fixed in the
interior of the disk), one has that
AB = 2(z w)g AB 2g AB ,
since we are considering a sigma model with bulk kinetic term
where = 4 ,
1
A d B . Also, note that the second term after the equality comes from
g
d
4 AB
the excluded zero mode, which acts as a background charge distribution of uniform density. The propagator also satisfies the boundary condition (recall that dz = i dz and
d z = i d z )

gAC d CB + iFAC d CB = 0,
(B.1)

i
A d B , with F
where we are assuming the boundary term 4
FAB d
AB constant.
The disk is defined by zz = 1, so that z d z + z dz = 0, and the previous boundary condition (B.1) becomes
CB = 0.
(g F)AC z CB + (g + F)AC z
Let us now consider the modified propagator P AB , defined by
1
AB .
AB (z, w) = P AB (z, w) + (zz + w w)g
2
It satisfies the differential equation
P AB = 2(z w)g AB ,
(B.2)
supplemented with the modified boundary condition

B
CB = A
.
(g F)AC zP CB + (g + F)AC z P
(B.3)
The leading singularity of this modified propagator is easily seen to be given by

P AB (z, w) g AB ln |z w|, so that the basic claim we now present is that

1 zw
AB
AB
AB
AB z w
,
2G ln |1 zw|
P
= ln
g ln
1 z w
1 zw
where AB and GAB are the standard open string tensors [23,25]. One observes from
this expression that the leading singularity is correct, and therefore Eq. (B.2) is satisfied.
236
Moreover, the derivatives of P AB can be computed as

(g F)AC z P
CB
CB
(g + F)AC z P

z
zw
zw
1
B
CB
(g F)AC g
+
,
=
1 zw A 2
z w 1 zw

z w
z
z w B 1
A (g + F)AC g CB
+
.
=
1 z w
2
z w
1 z w
Using the fact that zz = 1 one can then sum the two expressions above and obtain the
correct equation for the modified boundary condition (B.3). Let us consider the functions
that play a role in the previous result for P AB (z, w),

1 zw
,
A(z, w) = ln
1 z w
B(z, w) = 2 ln |1 zw|,

zw

,
C(z, w) = ln
1 zw
1
D(z, w) = (zz + w w).
2
Note that C vanishes whenever z or w are on the boundary of the disk. The Fourier transforms of A and B, on the unit disk , are given by

1
(xy)|n| ein() ,
A xei , yei =
n
n
=0

1
B xei , yei =
(xy)|n| ein() .
|n|
n
=0
The basic result we have obtained for the propagator on the disk is that
AB (z, w) = AB A(z, w) + GAB B(z, w) + g AB C(z, w) + g AB D(z, w).
We thus conclude by quoting
result. With the appropriate
dof the above
the Fourier transform
conventions of F () = n Fn ein and Fn = 2
F ()ein , one finally gets that

1 A
B
|n|
AB 1
AB 1
G
+
+ (n
= 0),
(x)
(y)
=
(xy)
n
n
|n|
n

1
1 A
0 (x)0B (y) = x 2 + y 2 g AB + ,
2
where the terms in come from the function C and vanish when either x or y are 1 (i.e.,
when either point is at the disk boundary).
Appendix C. Computation of two loop graphs

All the computations in this appendix are carried out in units where = 1. We recall the
vertices used in the subsequent computation. First we have the boundary M- and N -type
237
vertices
i(nN MA1 A2 ...AN + cyc1...N )i ni W ,
NA1 A2 ...AN i ni W .
Since MAB = MBA , the two M-vertex reads
i(n2 n1 )MA1 A2 n1 +n2 W .
We then have two bulk T -vertices

iTA1 A2 A3
|n1 |(n2 n3 ) + |n2 |(n3 n1 ) + |n3 |(n1 n2 ) ,
i |ni |

1
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413) + (1423) + (2314) ,
i |ni |
where we are using the compact notation (ij kl) TAi Aj Ak Al (nk nl |nk nl |).
C.1. Computation of -type graphs
C.1.1. Graph [1]
There is only one term

1
1
GAB GCD
NABCD
e(|n|+|m|) ,
4
|n||m|
m,n
which diverges as ln2 . Therefore its contribution to the beta function is zero.
C.1.2. Graph [2]
The graph reads (in what follows we omit the explicit reference to the ultraviolet cutoff , which we consider as understood)

1
MABC NRST
(ip)nAR mBS pCT m+n+p .
2
m,n,p
All the relevant sums are of the two forms
1

1
,
,
nm
m(m + p)
m,n1
m,p1
which both diverge in ln2 . So the contribution of this graph to the beta function is zero.
C.1.3. Graph [3]
To compute this graph, we introduce the notation
AB

1
AB
n =
= nAB nAC NCD nDB + .
|n|g + nF + N
238
The result is then compactly written as

1
MABC MRST
p 2 nAR mBS pCT + 2mp nAR mBT pCS m+n+p .
4
m,n,p
Expanding the propagators we have, to order N ,

1
BS CT
AR BT CS
p 2 (N )AR
MABC MRST
n m p + mp(N )n m p m+n+p
2
m,n,p

1
MABC MRST
p 2 nAR mBS (N )CT
p m+n+p
4
m,n,p

1
CS
AR BT
CS
MABC MRST
mp nAR (N )BT
m p + n m (N )p
2
m,n,p
m+n+p .
The second
1 and third lines do not have a leading ln divergence, since the sums are of the
form
nm m+n+p , as in the graph [2]. The nonvanishing contribution comes from the
first line. Consider first the following sums
S
p
m+n+p ,
n2 m
1
m+n+p ,
n2
in the summation regions I, II, III, defined by n, m > 0, by n, p > 0 and by m, p > 0.
Denote the three sums by SI , SII , SIII , and RI , RII , RIII . We have that
SI ln ,
SII ln ,
SIII ln ,
RI ln ,
RII ln ,
RIII ln .
The result then has the basic tensor (the 2 comes from the sums which are doubled under
(n, m, p) (n, m, p))
1
2 MABC MRST
2
multiplied, for the first term (S sums), by the following tensor structures
AR,BS,CT
(GNG)GG
(N )
(GN G)
(N )GG
(GN)G
(N G)G
(GN )G
(NG)G
+
+
II III
+
+ +
+ +
+

+
+

3 ln
ln
ln
3 ln
ln
ln
ln
ln
239
and, for the second term (R sums), by the following tensor structures
AR,BT ,CS
(GNG)GG
(N )
(GN G)
(N )GG
(GN)G
(N G)G
(GN )G
(N G)G
I
+
+
+
+
+
+
+
+
II III
+ +
+ +
+ +
+ +
+
+
+
+
ln
ln
ln
ln
3 ln
3 ln
3 ln
3 ln
We arrive at the following rather lengthy result

AR,BS,CT
ln
[3] = MABC (3MRST + MRT S ) (GN G + N )GG

AR,BS,CT
+ MABC (MRST MRT S ) (GN G + N )
ln

AR,BS,CT
+ MABC (MRST 3MRT S ) (GN + N G)(G + G)
ln .
C.1.4. Graph [4]
This graph has been computed in the text, in Section 3.6.
C.1.5. Graph [5]
The graph reads

1
d
d
A
B
C
TABC NRST x dx
.
R S T
6
2
2
The radial part gives
1
0
dx x |n|+|m|+|p|1 =
1
,
|n| + |m| + |p|
whereas the angular part contributes as

6i|m|pnAR mBS pCT n+m+p .
The total contribution is then

p|m|
nAR mBS pCT n+m+p .
iTABC NRST
|n|
+
|m|
+
|p|
m,n,p
The graph vanishes by symmetry. In fact, the most general result for the above sum is given
by
TABC NRST [aGG + b + cGG + dG G]AR,BS,CT .
Moreover, under the exchange m p in the sum, we see that c = d. Recall that N is symmetric in the indices (R, S, T ), whereas the indices (A, B, C) of T are antisymmetrized.
Therefore the above result vanishes irrespective of a, b, c = d, due to symmetry under
S T and antisymmetry under B C. Thus, this graph does not contribute to the beta
function.
240
C.1.6. Graph [6]

C.1.7. Graph [7]
The graph reads
1
TABC MA B C
2
d
2
A x B C dx
A B C
d
,
2
and the radial integration is readily done

1
dx x |n|+|m|+|p|1 =
.
|n| + |m| + |p|
After doing the angular integrations, one gets

1

n+m+p p 2 |m| nAA mBB pCC
|n|
+
|m|
+
|p|
m,n,p

+ mp|m| nAB mBC pCA + np|m| nAC mBA pCB .
TABC MA B C
Using the total antisymmetry of TABC , and the symmetry of MA B C in A B , we obtain

TABC MA B C
F (n, m, p) nAA mBB pCC n+m+p ,
m,n,p
where
F (n, m, p) =

1 p 2 |m| + np|p| + mp|n|
(m n) .
2
|n| + |m| + |p|
We must then consider the two following sums

n+m+p
m,n,p
F (n, m, p)
,
n2 mp

m,n,p
n+m+p
F (n, m, p)
,
nmp 2
in the three regions I, II, III (recall, with n, m > 0, n, p > 0 and m, p > 0), where we obtain
SI =
1
1 nm
ln ,
2
2
2
n m
m,n1
1
np
1
SII =
ln ,
2
2
2
n (n + p)
n,p1
SIII =
mp
1
ln ,
2
(m + p)2 m
m,p1
and
RI =
mn
1
= 0,
2
nm(n + m)
m,n1
RII =
241
np
1
= 0,
2
np(n + p)
n,p1
pm
1
= 0.
RIII =
2
(m + p)mp
m,p1
Finally we can compute the only non-vanishing contribution to the graph itself, given by

BB CC
2TABC MA B C
F (n, m, p)(N )AA
n m p n+m+p .
m,n,p
This is given by
4TABC MA B C
multiplied by the following tensor structures
AA ,BB ,CC
(GNG)GG
(N )
(GNG)
(N )GG
(GN)G
(N G)G
(GN )G
(NG)G
+
+
II III
+
+ +
+ +
+

+
+

0
2 ln
2 ln
0
ln
ln
ln
ln
We finally obtain

[7] = 4TABC MA B C 2(N + GN G)
AA ,BB ,CC
ln .
(GN + NG)(G + G)
C.1.8. Graph [8]
This graph is computed from the expectation value of

d
1
d
1
TABCD NRS A B x C x D + 2 C D x dx
.
R S
8
2
2
x
In what follows we concentrate on the expectation values of the s, omitting the term
1
8 TABCD NRS . First of all, the radial part always contributes as
1
x dx x 2|n|+2|m|2 =
0
1
1
.
2 |m| + |n|
The angular part then gives three contributions

BS CD
2 2m2 nAR n
m ,
2 AB CR DS
2 2n m n n ,
242

CS
8 |m||n| mn mBD nAR n
.
The combined result is then

2 AR BS CD
1
1
DS
m n n m + n2 mAB nCR n
TABCD NRS
4
|m|
+
|n|
n,m

CS
.
+ 2 |m||n| mn mBD nAR n
The second and the third terms diverge in ln2 , so their contribution to the beta function is
zero. The first term, on the other hand, contributes as
[8] = TABCD (GNG + N)AB GCD ln .
C.2. Computation of -type graphs
C.2.1. Graph [1]
C.2.2. Graph [2]
We look at the graph, with symmetry factor of 6, given by

i2
A3 B3
(p W )MA1 A2 A3 + nMA2 A3 A1 + mMA3 A1 A2 nA1 B1 mA2 B2 pW

W W =0 6

(p W ) dMB1 B2 B3 n dMB2 B3 B1 m dMB3 B1 B2 n+m+p .
We then obtain the following terms (a factor of n+m+p is understood)
1 1 2
p MA1 A2 A3 dMB1 B2 B3
6p

+ (nMA2 A3 A1 + mMA3 A1 A2 )(n dMB2 B3 B1 + m dMB3 B1 B2 ) nA1 B1 mA2 B2 pA3 B3 ,
and, from the regulator
1 |p|
(pMA1 A2 A3 + nMA2 A3 A1 + mMA3 A1 A2 )nA1 B1 mA2 B2 pA3 B3
12 p
(p dMB1 B2 B3 + n dMB2 B3 B1 + m dMB3 B1 B2 ).
Using the symmetry of MABC and dMABC in the first two indices and combining terms,
we get
1 A1 B1 A2 B2 A3 B3
m p
12 n

2p 2 2p 2 2p 2
|p| 2
|m| 2
|n| 2
+
+
+ p +
p + p
MA1 A2 A3 dMB1 B2 B3
p
m
n
p
m
n

2pn
|p|pn
|m|pn
|n|pn
+
+
+
.
+ 2MA1 A2 A3 dMB2 B3 B1
m
p
m
n
243
Doing the sums, we obtain

1
[2] = MA1 A2 A3 dMB1 B2 B3
6
[2 + 3GG + 3GG 4GG]A1 B1 ,A2 B2 ,A3 B3
1
+ MA1 A2 A3 dMB2 B3 B1
3
[ + 2GG 5GG + 2GG]A1 B1 ,A2 B2 ,A3 B3 .
The second contribution is given by

i3

(p W )MA1 A2 A3 + nMA2 A3 A1 + mMA3 A1 A2

W W =0 4
A C
nA1 B1 mA2 B2 pW3 pDB3

(2p W ) dMCD (pMB1 B2 B3 nMB2 B3 B1 mMB3 B1 B2 )n+m+p .
It leads to the following terms
i
3 B3 p(pM
MA1 A2 A3 nA1 B1 mA2 B2 ( dM )A
B1 B2 B3 + nMB2 B3 B1 + mMB3 B1 B2 )
p
2
(C.1)
and

i
1 + |p| (pMA1 A2 A3 + nMA2 A3 A1 + mMA3 A1 A2 )
4
3 B3 (pM
nA1 B1 mA2 B2 ( dM )A
B1 B2 B3 + nMB2 B3 B1 + mMB3 B1 B2 ).
p
The first expression, (C.1), gives a vanishing contribution. The second also gives a vanishing contribution, due to the following symmetry argument. Notice that the possible
contractions are between the tensor

(aGG + b)( dM + G dM G)
A B ,A B ,A B
+ (cG + d G)(G dM + dM G) 1 1 2 2 3 3
(with constants a, b, c, d) and an expression which is given by a product of two tensors
MABC , and which is symmetric under the interchange Ai Bi . On the other hand, the
tensor above is antisymmetric in the same interchange, and we therefore get a vanishing
result.
C.2.3. Graph [3]
The graph has a symmetry factor of 2, and reads

i
(p W )MABC + nMBCA + mMCAB TA B C dMRS
2

1
|n|(m p) + |m|(p n) + |p|(n m)
|n| + |m| + |p|

CR
(W 2p)nAA mBB pW
pSC .
244
W |W =0
Take the derivative
to get
i
(pMABC + nMBCA + mMCAB )TA B C
2

1
|n|(m p) + |m|(p n) + |p|(n m)
|n| + |m| + |p|

nAA mBB ( dM )CC

p ,
and also, from the regularization,
i
(pMABC + nMBCA + mMCAB )TA B C
2

|p|
|n|(m p) + |m|(p n) + |p|(n m)
|n| + |m| + |p|

nAA mBB ( dM )CC

p .
1
1
1
pR,
nS and
mS, where
We must consider the sums
nmp 2
nmp 2
nmp 2

(1 + |p|)
|n|(m p) + |m|(p n) + |p|(n m) ,
|n| + |m| + |p|

(1 + |p|)
|n|(m p) + |m|(p n) + |p|(n m) .
S
|n| + |m| + |p|
We obtain the following three tables. Firstly

nm
nm
pRI
0,
mn(n + m)
nm
R
n,m1
pRII
np
1
pn
+
ln ,
np(p + n)
n(n + p) 2
n,p1
pRIII
m,p1
mp
1
mp
ln ,
mp(m + p)
(m + p)m
2

where we use the fact that n,m1 n e(anbm) b1 ln , and that, within our regularization scheme, in -type graphs a = b = 2, whereas in -type graphs a = b = 1. Secondly

3
nm
nm
ln ,
+
nSI
2
m(n + m) 2
m(n + m)
n,m1
nSII
n,p1
nSIII
1
np
np
ln ,
+
(n + p)p 2
p 2 (n + p)
mp
mp
ln ,
+
mp
p2 m
m,p1
and finally
mSI

n,m1
3
nm
nm
ln ,
+
2
n(n + m)
2
n(n + m)
mSII
245
pn
pn
ln ,
+
2
np
p n
n,p1
mSIII
pm
m,p1
p 2 (m + p)
1
pm
ln .
(m + p)p
2
Using the table for the R sums

AA ,BB ,CC
I
+
+
+
+
II III

+ +
+ +

+
+
+
+
I
GG(G dM G) +
( dM ) +
(G dM G) +
GG( dM ) +
G(G dM )
G( dM G)
G(G dM )
G( dM G)
II III

+ +
+ +

+
+
+
+
GG(G dM G)
( dM )
(G dM G)
GG( dM )
G(G dM )
G( dM G)
G(G dM )
G( dM G)
0
0
0
0
ln
ln
ln
ln
and that for the S sums

AA ,BB ,CC
AA ,BB ,CC
GG(G dM G)
( dM )
(G dM G)
GG( dM )
G(G dM )
G( dM G)
G(G dM )
G( dM G)
0
3 ln
3 ln
0
2 ln
ln
ln
2 ln
I
+
+
+
+
II III

+ +
+ +

+
+
+
+
0
3 ln
3 ln
0
ln
2 ln
2 ln
ln
the various terms combine as follows

3i(MBCA MCAB )TA B C ()AA,BB (G dM G + dM )CC ln ,

and
i(MABC )TA B C (G G)AA ,BB (G dM + dM G)CC ln

+ i(MBCA )TA B C (2G G)AA ,BB (G dM + dM G)CC ln

+ i(MCAB )TA B C (G + 2G)AA ,BB (G dM + dM G)CC ln .

Using the fact that MABC + MBCA + MCAB = 0, that TABC is totally antisymmetric, and
that MABC is symmetric in AB, we can combine the various terms and obtain, after reshuffling indices,

[3] = 6iMABC TA B C (G dM + dM G)G
AA ,BB ,CC
(G dM G + dM )
ln .
246
C.2.4. Graph [4]

This graph has symmetry factor 4, and is given by
i

CR
TABC TA B C dMRS (W 2p)nAA mBB pW
pSC
4

1
|n|(m p + W ) + |m|(p n W ) + |p W |(n m)
|n| + |m| + |p W |

1
|n|(m p) + |m|(p n) + |p|(n m) .
|n| + |m| + |p|
Taking the derivative W

|W =0 one gets (excluding terms from the regularization)
i
4 TABC TA B C multiplying by
2|p|

nAA mBB ( dM )CC
p
3
(|n| + |m| + |p|)

2
|n|(m p) + |m|(p n) + |p|(n m) ,
and by
2

nAA mBB ( dM )CC
p
2
(|n| + |m| + |p|)

|n|(m + p) |m|(p + n) |p|(n m)

|n|(m p) + |m|(p n) + |p|(n m) .
Notice that the factors multiplying the propagators are even under the exchange
(n, m, p) (n, m, p), and therefore, one concludes that the general form of the
tensor multiplying TABC TA B C will be

(aGG + b)( dM + G dM G)
AA ,BB ,CC
+ (cG + dG)(G dM + dM G)
,
for some constants a, b, c, d. The expression TABC TA B C is symmetric when interchanging primed and unprimed indices, whereas the above tensor is antisymmetric. Therefore
the contribution above vanishes.
Consider finally the terms coming from the regulator. They read
|p|
i

TABC TA B C nAA mBB ( dM )CC
p
4
(|n| + |m| + |p|)2

2
|n|(m p) + |m|(p n) + |p|(n m) .
This vanishes for the same reasons as above.
C.2.5. Graph [5]
The graph, which has symmetry factor 4, is given by

1 1
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413) + (1423) + (2314)
4 i |ni |
i(n1 n2 ) dMB1 B2 nA11 B1 nA22 B2 nA33 A4 ,
247
where
n1 = n,
n2 = n W,
n3 = m,
n4 = m.
Let us now compute the derivative

coming from the regularization
W |W =0
and get, excludingfor the momentterms

i 1
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413)
4 i |ni |

+ (1423) + (2314) dMB1 B2 nA11 B1 nA22 B2 nA33 A4 ,
and

i |n2 |
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413)
2 (i |ni |)2

+ (1423) + (2314) dMB1 B2 nA11 B1 nA22 B2 nA33 A4 ,
and finally

i 1
(3412) + (1324) + (1423) dMB1 B2 nA11 B1 nA22 B2 nA33 A4 .
2 i |ni |
We then have the following six terms
1
i
1 A2 A3 A4
( dM )A
n
m
8 |n| + |m|
times (Ai i)

|n|
2
T1234 2m
+1 ,
|n| + |m|

|m|
,
T3412 2n2
|n| + |m|

|m|
,
T1324 nm + |nm|
|n| + |m|

|n|
+1 ,
T2413 nm |nm|
|n| + |m|

|m|
,
T1423 nm + |nm|
|n| + |m|

|n|
+1 .
T2314 nm |nm|
|n| + |m|
The first and second line vanish due to symmetry. In fact dMAB is antisymmetric, and
therefore the symmetric part of dM is dM G + G dM , but it comes with a
(n|n|)1 , which is odd under n n, whereas the rest of the sum is even. The other
four term can be rewritten as

4n2 |m|
,
SS1234
|n| + |m|
248
SA1234 (4nm),

AS1234 4|nm| ,

4nm|n|
AA1234
,
|n| + |m|
where SS is symmetric in both pairs 12 and 34, SA is symmetric in pair 12 and antisymmetric in pair 34, AS is antisymmetric in pair 12 and symmetric in pair 34, etc. The
only
terms (due to symmetry) are the second and third, but they both vanish as
surviving
1
n,m n(n+m) 0.
Finally let us compute the terms coming from the regularization

i |n1 |
TA1 ...A4 n3 n4 |n3 n4 | + (3412) + (1324) + (2413)
4 i |ni |

1 A2 A3 A4 ,
+ (1423) + (2314) ( dM )A
n
m
where

|n|
i
2
1 A2 A3 A4 2m2 T
( dM )A
1234 2n T3412
n
m
8 |n| + |m|

+ nm |nm| (T1324 + T2413 ) nm + |nm| (T1423 + T2314 ) .
The first two terms vanish for the same reasons as before. The other four terms resum to

|n|
i
1 A2 A3 A4 4|nm|SS
( dM )A
1324 + 4nmAA1234 ,
n
m
8 |n| + |m|
which also vanishes by symmetry.
Appendix D. Formul for useful sums

In the computation of the beta function, the only results we shall need are the ln contributions in the following sums
1
ln ,
n2
n,m1
n,m1
1
ln ,
(n + m)2
1
1
ln ,
n
a
n,m1
together with the vanishing contributions of the following ones

1
0,
nm
n,m1
n,m1
1
0,
n+m
(D.1)

n,m1
1
0.
n(n + m)
249
(D.2)
Note that all the sums are regulated using the factor
ea(n+m) .
Denoting with
x = ea ,
the three sums in (D.1) read
x
2 1
+ ln ,
1x
6a
ln(1 x) DiLn(1 x) ln ,
1
x
ln ,
ln(1 x)
1x
a
DiLn(1 x)
where we have used that

DiLn(1 x) =
xn
n1
n2
2
+ a ln + O().
6
The first and second sum in (D.2) read, on the other hand,
ln2 (1 x) ln2 0,
x
+ ln(1 x) 0.
1x
To evaluate the third sum in (D.2), we observe that
x
1

x ln(1 x)
1
=
=
,
x
n(n + m)
n
1x
n,m1
n,m1
so that the sum is given, after integration, by

1 2
ln (1 x) + const ln2 0.
2
Appendix E. BornInfeld action at one loop
In the main text we describe a general procedure to extract curvature couplings for the
non-Abelian BI action, given the two loop Abelian BI equations of motion computed nonperturbatively in F . This procedure is implemented at the level of the equations of motion,
and when the curved background is described in terms of a RNC expansion. Here, and as
a complement to what is done in the main text, let us briefly analyze what are the general
expectations from the [one loop] Abelian BI action. For the WZW parallelizable target
manifolds we are working with, one can define MAB (x) gAB (x) + BAB (x) in a RNC
250
expansion as
1
1
MAB (x) = gAB + HABC x C HACK H K DB x C x D + ,
3
12
where we extracted the constant part of the NSNS 2-form field, and where one raises and
lowers indices with the constant metric gAB (which can be chosen to be AB ). In this case
the Abelian BI action is written

S = Tp d n x det MI J (x) + BI J + 2 FI J (x) .
Next, one applies the SW map to the previous expression to learn which non-Abelian
couplings are present in this one loop result. First, change coordinates to i such that

BI J + 2 FI J (x) dx I dx J = Bij d i d j ,
so that the dynamical variables become the functions x I ( i ). In these new coordinates the
BI action becomes (we have denoted B1 )

S = Tp d n det Bij + i x I j x J MI J (x)

= Tp d n det B det ji + ik k x I j x J MI J (x) .
We are now in a position to obtain a matrix model from the previous expression. Use the
Poisson bracket, {f, g} = ij i f j g, and the fact that (recall that the determinant can be
expanded in powers of the trace)

n
n
Trij ik k x I j x J MI J (x) = TrI J x I , x K MKJ (x) ,
to conclude that the Abelian action can be finally written as

S = Tp d n det B det JI + x I , x K MKJ (x) .
Our point here concerns the power expansion of this action: it is simple to see that it
contains the two known terms which are fixed at one loop, it contains none of the terms
which are fixed at two loops, and it contains infinite other terms which can only be fixed at
higher loop order (see the main text). Indeed, one obtains upon expansion

1

1
S = Tp d n det B 1 + x I , x J x I , x J HI J K x I , x J x K + .
4
6
A would-be non-Abelian BI action can be obtained from the previous expression
I
J
I
J
upon
familiar deformation quantization map of {x , x } i[X , X ] and also
n the
d det B Tr (where the trace is now over the matrices X ). It includes the terms
in the one loop diagonal, but no terms in the two loop diagonal (see the main text)

I J K
1 I J I J i
S = Tp Tr 1 X , X X , X + HI J K X , X X + .
4
6
251
Appendix F. Two loop matrix model in a flat background

In this appendix we analyze the two loop matrix model we have found, when the background is flat (which is basically the flat space non-Abelian BI action for a bosonic string).
A question we want to address is whether or not the standard solution of commuting Dbranes can get replaced with some other, possibly fuzzy, solution. Let us begin precisely
with the flat space non-Abelian BI action for a bosonic string (see, e.g., Polchinskis vol. 1,
p. 188, for details and notation)

2i J K I
2
1
.
S AaI = 2 d 26 x Tr FI J F I J
Tr FI FJ FK
4
3
go
We have disregarded the open string tachyon above. As we compactify from maximal
Abelian brane to minimal non-Abelian brane, one has [82]

i
XI
XI XJ
,
FI J i
,
=
[XI , XJ ].
AI
2
2 2
(2 )2
After factoring out the volume of spacetime, and disregarding the kinetic term, the matrix
action follows

2

1
i
2
S[X ] = 2
Tr [XI , XJ ] X I , X J

2
4
go
(2 )
3

i
2i
J
K
I
.
XK , X
Tr XI , X XJ , X
3
(2 )2
One should now recall that the effective potential relates to the action as Veff [X ] = S[X ],
so that by factoring out a (strictly positive) common term in the expression above one
obtains the final expression we use in this paper

1
Tr X I , X J X I , X J

2
4( )

1
Tr X I , X J X J , X K X K , X I ,
2

3
6 ( )
Veff [X ] =
where repeated indices are summed (contracted with the flat metric I J ). It is straightforward to obtain the equations of motion which follow from this potential, and they are
M M K
X , X ,X
1 M M N N K
X , X ,X , X ,X
= 0.
2 2
One immediately observes that commuting branes, where [X M , X N ] = 0, satisfy the equations of motion with energy Veff [Xcommuting ] = 0. On the other hand, one can also try a
noncommuting fuzzy solution, where [X M , X N ] = AMN K X K (let us consider only three
coordinate matrices). In this case, the equations of motion tell us that
1
2A4 KM X M = 0 A = 0 A2 = 2 2 ,
2 2
where the A = 0 case corresponds to the previously analyzed commuting solution and
one should pick nonzero A for the fuzzy solution. The fuzzy branes satisfy the equations
2A2 KM X M
252

2
of motion with energy Veff [Xfuzzy ] = 3 Tr X M X M > 0. It thus follows that the stable,
minimal energy solution, corresponds to the standard solution of commuting D-branes, as
should have been expected from scratch in a flat space background.
References
[1] R.C. Myers, Dielectric branes, JHEP 9912 (1999) 022, hep-th/9910053.
[2] B. Janssen, Y. Lozano, A microscopical description of giant gravitons, Nucl. Phys. B 658 (2003) 281, hepth/0207199.
[3] B. Janssen, Y. Lozano, D. Rodriguez-Gomez, A microscopical description of giant gravitons, II: the AdS5
S 5 background, Nucl. Phys. B 669 (2003) 363, hep-th/0303183.
[4] D. Kabat, W. Taylor, Linearized supergravity from matrix theory, Phys. Lett. B 426 (1998) 297, hepth/9712185.
[5] W. Taylor, M. Van Raamsdonk, Supergravity currents and linearized interactions for matrix theory configurations with fermionic backgrounds, JHEP 9904 (1999) 013, hep-th/9812239.
[6] W. Taylor, M. Van Raamsdonk, Multiple D0-branes in weakly curved backgrounds, Nucl. Phys. B 558
(1999) 63, hep-th/9904095.
[7] W. Taylor, M. Van Raamsdonk, Multiple Dp-branes in weak background fields, Nucl. Phys. B 573 (2000)
703, hep-th/9910052.
[8] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[9] M.R. Douglas, D-branes and matrix theory in curved space, Nucl. Phys. B (Proc. Suppl.) 68 (1998) 381,
hep-th/9707228.
[10] M.R. Douglas, A. Kato, H. Ooguri, D-brane actions on Khler manifolds, Adv. Theor. Math. Phys. 1 (1998)
237, hep-th/9708012.
[11] J. De Boer, K. Schalm, General covariance of the non-Abelian DBI action, JHEP 0302 (2003) 041, hepth/0108161.
[12] M. Van Raamsdonk, Blending local symmetries with matrix nonlocality in D-brane effective actions,
JHEP 0309 (2003) 026, hep-th/0305145.
[13] S.F. Hassan, N = 1 worldsheet boundary couplings and covariance of non-Abelian worldvolume theory,
hep-th/0308201.
[14] J. De Boer, K. Schalm, J. Wijnhout, General covariance of the non-Abelian DBI action: checks and balances,
hep-th/0310150.
[15] D. Brecher, K. Furuuchi, H. Ling, M. Van Raamsdonk, Generally covariant actions for multiple D-branes,
JHEP 0406 (2004) 020, hep-th/0403289.
[16] A. Hashimoto, I.R. Klebanov, Scattering of strings from D-branes, Nucl. Phys. B (Proc. Suppl.) 55 (1997)
118, hep-th/9611214.
[17] M.R. Garousi, R.C. Myers, Worldvolume interactions on D-branes, Nucl. Phys. B 542 (1999) 73, hepth/9809100.
[18] M.R. Garousi, Noncommutative worldvolume interactions on D-branes and DiracBornInfeld action, Nucl.
Phys. B 579 (2000) 209, hep-th/9909214.
[19] M.R. Garousi, R.C. Myers, Worldvolume potentials on D-branes, JHEP 0011 (2000) 032, hep-th/0010122.
[20] M.R. Garousi, Transformation of the DiracBornInfeld action under the SeibergWitten map, Nucl. Phys.
B 602 (2001) 527, hep-th/0011147.
[21] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory: compactification on
tori, JHEP 9802 (1998) 003, hep-th/9711162.
[22] Y.-K.E. Cheung, M. Krogh, Noncommutative geometry from D0-branes in a background B-field, Nucl.
Phys. B 528 (1998) 185, hep-th/9803031.
[23] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/9903205.
[24] L. Cornalba, R. Schiappa, Matrix theory star products from the BornInfeld action, Adv. Theor. Math.
Phys. 4 (2000) 249, hep-th/9907211.
[25] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
253
[26] L. Cornalba, D-brane physics and noncommutative YangMills theory, Adv. Theor. Math. Phys. 4 (2000)
271, hep-th/9909081.
[27] L. Cornalba, Corrections to the Abelian BornInfeld action arising from noncommutative geometry,
JHEP 0009 (2000) 017, hep-th/9912293.
[28] L. Cornalba, On the general structure of the non-Abelian BornInfeld action, Adv. Theor. Math. Phys. 4
(2002) 1259, hep-th/0006018.
[29] L. Cornalba, Some computations with SeibergWitten invariant actions, Mod. Phys. Lett. A 16 (2001) 227,
hep-th/0101015.
[30] H. Dorn, J. Otto, Open bosonic strings in general background fields, Z. Phys. C 32 (1986) 599.
[31] A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields, Nucl. Phys.
B 280 (1987) 599.
[32] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions, Nucl. Phys. B 288
(1987) 525.
[33] O.D. Andreev, A.A. Tseytlin, Two loop beta function in the open string sigma model and equivalence with
string effective equations of motion, Mod. Phys. Lett. A 3 (1988) 1349.
[34] C.P. Bachas, P. Bain, M.B. Green, Curvature terms in D-brane actions and their M-theory origin, JHEP 9905
(1999) 011, hep-th/9903210.
[35] C.A. Scrucca, M. Serone, A note on the torsion dependence of D-brane RR couplings, Phys. Lett. B 504
(2001) 47, hep-th/0010022.
[36] A. Fotopoulos, On ( )2 corrections to the D-brane action for nongeodesic worldvolume embeddings,
JHEP 0109 (2001) 005, hep-th/0104146.
[37] S. Corley, D. Lowe, S. Ramgoolam, EinsteinHilbert action on the brane for the bulk graviton, JHEP 0107
(2001) 030, hep-th/0106067.
[38] N. Wyllard, Derivative corrections to D-brane actions with constant background fields, Nucl. Phys. B 598
(2001) 247, hep-th/0008125.
[39] N. Wyllard, Derivative corrections to the D-brane BornInfeld action: non-geodesic embeddings and the
SeibergWitten map, JHEP 0108 (2001) 027, hep-th/0107185.
[40] A. Fotopoulos, A.A. Tseytlin, On gravitational couplings in D-brane action, JHEP 0212 (2002) 001, hepth/0211101.
[41] A. Barabanschikov, Boundary sigma model and corrections to D-brane actions, Phys. Rev. D 67 (2003)
106001, hep-th/0301012.
[42] F. Ardalan, H. Arfaei, M.R. Garousi, A. Ghodsi, Gravity on noncommutative D-branes, Int. J. Mod. Phys.
A 18 (2003) 1051, hep-th/0204117.
[43] M. Wijnholt, On curvature squared corrections for D-brane actions, hep-th/0301029.
[44] P. Koerber, A. Sevrin, Testing the 3 term in the non-Abelian open superstring effective action, JHEP 0109
(2001) 009, hep-th/0109030.
[45] M. de Roo, M.G.C. Eenink, P. Koerber, A. Sevrin, Testing the fermionic terms in the non-Abelian D-brane
effective action through order 3 , JHEP 0208 (2002) 011, hep-th/0207015.
[46] P. Koerber, A. Sevrin, The non-Abelian D-brane effective action through order 4 , JHEP 0210 (2002) 046,
hep-th/0208044.
[47] A. Sevrin, A. Wijns, Higher order terms in the non-Abelian D-brane effective action and magnetic background fields, JHEP 0308 (2003) 059, hep-th/0306260.
[48] P.C. Argyres, C.R. Nappi, Spin 1 effective actions from open strings, Nucl. Phys. B 330 (1990) 151.
[49] E. Braaten, T.L. Curtright, C.K. Zachos, Torsion and geometrostasis in nonlinear sigma models, Nucl. Phys.
B 260 (1985) 630.
[50] S. Mukhi, Finiteness of nonlinear sigma models with parallelizing torsion, Phys. Lett. B 162 (1985) 345.
[51] S. Mukhi, The geometric background field method, renormalization and the WessZumino term in nonlinear
sigma models, Nucl. Phys. B 264 (1986) 640.
[52] A.Y. Alekseev, A. Recknagel, V. Schomerus, Noncommutative worldvolume geometries: branes on SU(2)
and fuzzy spheres, JHEP 9909 (1999) 023, hep-th/9908040.
[53] A.Y. Alekseev, A. Recknagel, V. Schomerus, Brane dynamics in background fluxes and noncommutative
geometry, JHEP 0005 (2000) 010, hep-th/0003187.
[54] A. Alekseev, V. Schomerus, RR charges of D2-branes in the WZW model, hep-th/0007096.
254
[55] S. Fredenhagen, V. Schomerus, Branes on group manifolds, gluon condensates, and twisted K-theory,
JHEP 0104 (2001) 007, hep-th/0012164.
[56] C. Bachas, M. Douglas, C. Schweigert, Flux stabilization of D-branes, JHEP 0005 (2000) 048, hepth/0003037.
[57] P. Bordalo, S. Ribault, C. Schweigert, Flux stabilization in compact groups, JHEP 0110 (2001) 036, hepth/0108201.
[58] H. Steinacker, Aspects of the q-deformed fuzzy sphere, Mod. Phys. Lett. A 16 (2001) 361, hep-th/0102074.
[59] H. Steinacker, Quantum field theory on the q-deformed fuzzy sphere, hep-th/0105126.
[60] J. Pawelczyk, H. Steinacker, Matrix description of D-branes on 3-spheres, JHEP 0112 (2001) 018, hepth/0107265.
[61] J. Pawelczyk, H. Steinacker, A quantum algebraic description of D-branes on group manifolds, Nucl. Phys.
B 638 (2002) 433, hep-th/0203110.
[62] J. Pawelczyk, H. Steinacker, Algebraic brane dynamics on SU(2): excitation spectra, JHEP 0312 (2003)
010, hep-th/0305226.
[63] A.A. Tseytlin, Sigma model Weyl invariance conditions and string equations of motion, Nucl. Phys. B 294
(1987) 383.
[64] A.A. Tseytlin, Sigma model approach to string theory, Int. J. Mod. Phys. A 4 (1989) 1257.
[65] K. Behrndt, H. Dorn, Weyl anomaly and CurciPaffuti theorem for sigma models on manifolds with boundary, Int. J. Mod. Phys. A 7 (1992) 1375.
[66] E. Witten, Some computations in background independent off-shell string theory, Phys. Rev. D 47 (1993)
3405, hep-th/9210065.
[67] S.L. Shatashvili, Comment on the background independent open string theory, Phys. Lett. B 311 (1993) 83,
hep-th/9303143.
[68] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes in curved
backgrounds, Commun. Math. Phys. 225 (2002) 33, hep-th/0101219.
[69] M. Kontsevich, Deformation quantization of Poisson manifolds I, Lett. Math. Phys. 66 (2003) 157, qalg/9709040.
[70] P. Bordalo, Discrete torsion and WZW orbifolds, Phys. Lett. B 582 (2004) 86, hep-th/0310029.
[71] L. Cornalba, M.S. Costa, R. Schiappa, D-brane dynamics in constant RamondRamond potentials, S-duality
and noncommutative geometry, hep-th/0209164.
[72] R. Dijkgraaf, E. Verlinde, H. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 43, hep-th/9703030.
[73] R. Schiappa, Matrix strings in weakly curved background fields, Nucl. Phys. B 608 (2001) 3, hepth/0005145.
[74] D. Brecher, B. Janssen, Y. Lozano, Dielectric fundamental strings in matrix string theory, Nucl. Phys. B 634
(2002) 23, hep-th/0112180.
[75] P.E. Haagensen, Duality transformations away from conformal points, Phys. Lett. B 382 (1996) 356, hepth/9604136.
[76] P.E. Haagensen, K. Olsen, T-duality and two loop renormalization flows, Nucl. Phys. B 504 (1997) 326,
hep-th/9704157.
[77] P.E. Haagensen, K. Olsen, R. Schiappa, Two loop beta functions without Feynman diagrams, Phys. Rev.
Lett. 79 (1997) 3573, hep-th/9705105.
[78] H. Dorn, J. Otto, Remarks on T-duality for open strings, Nucl. Phys. B (Proc. Suppl.) 56 (1997) 30, hepth/9702018.
[79] K. Olsen, R. Schiappa, Heterotic T-duality and the renormalization group, Int. J. Mod. Phys. A 14 (1999)
2257, hep-th/9805074.
[80] U. Mller, C. Schubert, A.E.M. van de Ven, A closed formula for the Riemann normal coordinate expansion,
Gen. Relativ. Gravit. 31 (1999) 1759, gr-qc/9712092.
[81] A. Hatzinikitas, A note on Riemann normal coordinates, hep-th/0001078.
[82] W. Taylor, D-brane field theory on compact spaces, Phys. Lett. B 394 (1997) 283, hep-th/9611042.
Quantum dynamics of a bulk-boundary system

Shoichi Ichinose a , Akihiro Murayama b
a Laboratory of Physics, School of Food and Nutritional Sciences, University of Shizuoka,
Yada 52-1, Shizuoka 422-8526, Japan

b Department of Physics, Faculty of Education, Shizuoka University, Shizuoka 422-8529, Japan
Received 26 July 2004; received in revised form 29 November 2004; accepted 16 December 2004
Abstract
The quantum dynamics of a bulk-boundary theory is closely examined by the use of the background field method. As an example we take the MirabelliPeskin model, which is composed of
5D super-YangMills (bulk) and 4D WessZumino (boundary). Singular interaction terms play an
important role of canceling the divergences coming from the KK-mode sum. Some new regularization of the momentum integral is proposed. An interesting background configuration of scalar fields
is found. It is a localized solution of the field equation. In this process of the vacuum search, we
present a new treatment of the vacuum with respect to the extra coordinate. The supersymmetric
effective potential is obtained at the 1-loop full (w.r.t. the coupling) level. This is the bulk-boundary
generalization of the ColemanWeinbergs case. Renormalization group analysis is done and the correct 4D result is reproduced. The Casimir energy is calculated and is compared with the case of the
KaluzaKlein model.
PACS: 11.10.Kk; 11.27.+d; 12.60.Jv; 12.10.-g; 11.25.Mj; 04.50.+h
Keywords: Bulk-boundary theory; MirabelliPeskin model; Effective potential; ColemanWeinberg potential;
Brane model; Casimir energy; 5D supersymmetry
E-mail addresses: ichinose@u-shizuoka-ken.ac.jp (S. Ichinose), edamura@ipc.shizuoka.ac.jp

(A. Murayama).
doi:10.1016/j.nuclphysb.2004.12.023
256
S. Ichinose, A. Murayama / Nuclear Physics B 710 (2005) 255308
1. Introduction
Through recent several years of development, it looks that the higher-dimensional approach has obtained the citizenship as an important building tool in constructing the unified
theories. It appears with some names such as RandallSundrum model, brane world,
extra dimension model, orbifold model, etc. Before the appearance of the new approach, supersymmetry (SUSY) was the main promising tool to go beyond the standard
model. Among many ideas in the higher-dimensional approach, the system of bulk and
boundary theories becomes a fascinating model of the unification. A boundary is regarded
as our world. It is inspired by the M, string and D-brane theories [1]. One pioneering paper,
which concretely describes the model, is that by Mirabelli and Peskin [6]. They take the
5D supersymmetric YangMills theory as a bulk theory and make it couple with a boundary matter. The boundary couplings (with the bulk world) are uniquely fixed by the SUSY
requirement. They demonstrated some consistency in the bulk quantum theory by calculating self-energy of the scalar matter field. Here we examine the effective potential and the
vacuum energy of this system. We investigate further closely the role of the bulk fields and
the singular interactions.
A field theoretical analysis of the bulk and boundary system was recently done in the
work by Goldberger and Wise [2]. They try to tackle the problem by the generalized use
of the renormalization group. Randall and Schwarts [3,4] also attacked the same problem
by introducing a special regularization of the ultraviolet divergences guided by the idea of
the holography. We take a different approach to the bulk and boundary system. (We will
see, however, some similar results.) It is, at present, hard to show any consistency (such
as renormalizability, unitarity, etc.) in the higher-dimensional quantum field theory. It is,
at least perturbatively, unrenormalizable. We would rather regard the bulk world as an external heat-reservoir which gives some freedom to the boundary world and define or
regularize the 4D dynamics. The external world of the bulk classically and quantumly
affect our world of 4D, and vice versa. In this circumstance we focus on the renormalization properties of the 4D world. We examine a way to treat the linear (power) divergences
coming from the bulk quantum effect. Another important aspect is the present treatment of
the extra axis. We will find some freedom in the definition of the vacuum. Z 2 -symmetry
plays an important role there. We can naturally introduce the singular behaviour for some
scalars. One important merit of the bulk-boundary approach is that the anomaly phenomenon (of the 4D world) is naturally accepted as a current flow which goes out through the
wall or comes into the 4D world [5].
Contrary to the motivation of the original work of Ref. [6], we do not seek the SUSY
breaking mechanism, rather we keep the supersymmetry and make use of the SUSY invariant properties in order to make the analysis as simple as possible. The SUSY symmetry is
so restrictive that we only need to calculate some small portion of all possible diagrams.
As the analysis of the effective potential of the 5D model, we recall that of the Kaluza
Klein (KK) model [7]. The dynamics quantumly produces the effective potential which
describes the Casimir effect. The situation, however, is contrastively different from the
present case in some points.
257
(1) The present approach realizes the 4D reduction by the localized (along the extra
axis) configuration (kink, soliton, delta-function), whereas it is done, in the KK model, by
the shrinkage of the radius of the extra S1 space.
(2) Z2 -symmetry is not used in the KK model whereas the present one exploits it in
order to make a singular structure at x 5 = 0, l (fixed points) where the 4D worlds are. The
discrete symmetry imposes a nontrivial boundary condition on the vacuum.
(3) In the KK model scalar fields are set constant in order to find the vacuum configuration, whereas, in the present case, we do a new treatment of the vacuum by allowing the
extra-coordinate dependence on some scalars.
(4) The present model is supersymmetric, whereas the KK model is not.
(5) In the KK model, the scalar field comes from the (5, 5)-component of the 5D metric
(and partially from the dilaton). The 5D quantum effect produces the effective potential
which can be interpreted as the Casimir force induced by the vacuum polarization between
the l separated objects. On the other hand, in the present case, the scalar components come
from various places: the 5th component of the bulk vector, the bulk scalar and the boundary
scalar fields. Hence the vacuum structure becomes much richer.
(6) The present model has, as the characteristic length scales, the thickness parameter
(brane tension) besides the period of the extra space. As the vacuum energy calculation,
we should see the dependence on both lengths.
The present model shares common properties with those of the RS-model in the points
such as localization, Z2 -symmetry, bulk-boundary relation, etc. (The comparative aspect of
the KK model and the RS-model is explained in Ref. [8].) We could regard the present result about the Casimir energy as some RS counter-part of the result obtained by Appelquist
and Chodos for the case of the KK-model.
The concrete object we will obtain is the effective potential. The formalism itself is very
orthodox. The new point is its application to the 5D bulk-boundary system. The system is
much extended from the ordinary field theory. The effective potential is well-established
in the field theory. Especially in the middle of 70s much literature appeared. One of the
famous outcome is the ColemanWeinberg potential [9]. In the SUSY theories, Miller
proposed a useful method, called AFTM [10], based on the tadpole diagram method by
Weinberg [11]. It was applied to unified models [12]. We will take another formalism, the
background field method, by DeWitt [13] and t Hooft [14].1 The new formulation of the
bulk-boundary system is another aim of this paper.
We summarize the new points as follows:
(1) background-field formulation of the bulk-boundary theory;
(2) (0) singularity problem is solved;
(3) a new proposal for resolving the UV-divergences in the 5D quantum S 1 /Z2 orbifold
theory;
1 The use of the background field method in the brane world analysis is stressed by Randall and Schwarz [3,4].
They develop a perturbative treatment in the AdS5 5D bulk theory. They try to solve the similar problems to the
present ones. Especially perturbative treatment, log versus power divergences, regularization, renormalization
group running of the coupling. They do not use a SUSY theory. They focus on the bulk gauge field theory.
258
(4) new treatment of the vacuum in the presence of the extra-space;

(5) Casimir energy calculation.
Some of the present results are briefly reported in Ref. [15].
The paper is organized as follows. In Section 2, we introduce the present formalism of
the background field method, in the analysis of the effective potential. The simple model
of WessZumino is taken as an example. Here we explain the supersymmetric effective
potential. MirabelliPeskin model is explained in Section 3. It is a typical bulk-boundary
model based on 5D SUSY. In Section 4, we quantize the model using the background field
method. A new treatment of the background field, in relation to the extra coordinate, is
presented. This leads to an interesting background solution (vacuum) which describes the
field-localization. Feynman rules are obtained for the perturbative analysis in Section 5.
The singular vertices, which involve the delta function, appear. Some Feynman diagrams
are explicitly calculated. We take into account both bulk and boundary quantum effects.
We will find that the singular interaction terms play the role of the counter-terms to
cancel the divergences coming from the KK-mode sum. In Section 6, the mass matrix
appearing in the 1-loop Lagrangian is obtained. This is the preparation for the 1-loop full
calculation of the next section. Assumption of the form of the background field about its
extra-coordinate dependence is crucial for the present analysis. In Section 7, the effective
potential is obtained. Two typical cases, (A) and (B), are considered. In case (A) we look
at the potential from the vanishing vacuum of the brane matter-field. The final form of
the potential is similar to the 4D super-QED. In the intermediate stage, we find a new
type Casimir energy which is characteristic for the brane world. In case (B), we obtain the
potential for the no brane configuration. In the intermediate stage, we find the ordinary
type Casimir energy. The effective potential has rich structure. We conclude in Section 8.
We relegate some important detailed explanation to three appendices. Appendix A treats
the super-QED which is a good reference point in the analysis of the bulk-boundary theory
in the text. Appendix B provides the calculation of the eigenvalues of the mass matrix of
Section 6. The results are used in Section 7. Appendix C explains the concrete form of
the present background fields. They satisfy the field equation with the required boundary
condition.
2. Effective potential of WessZumino model

In order to explain the background field approach to obtain the effective potential, we
take the simplest 4D SUSY theory, that is, the WessZumino model
m m A + F F
L[, A, F ; , m] = im m + A

1
+ m AF + (AAF A) + h.c. ,
2
2
(1)
where (mn ) = diag(1, 1, 1, 1). The notation is basically the same as the textbook by
WessBagger [24]. is a Majorana fermion, A is a complex scalar field and F is an
(complex scalar) auxiliary field. The general background field method [13,14,16] tells us
that the (DeWittWilsonian) effective action S eff [, a, f ] is given by

exp iS eff [, a, f ]

L[] I
4
= D DA DF exp i d x L[ + , a + A, f + F ]
,
I b
259
(2)
where ( I ) (, A, F ) are the quantum fields and their background fields ( I )|b
(, a, f ). We define the effective potential V eff as the non-derivative part of S eff . A simple
and practical way to pick up the part is to consider the case:
= 0,
a = const,
f = const,
(3)
where we put = 0 from the requirement of the Lorentz invariance of the vacuum and
const means a constant:

exp iV eff [a, f ]

= exp i V0eff D DA DF exp i d 4 x L2 + order of (quant. field)3 ,

2
2
eff
V0 = f f + ma + a f + ma + a f L0 ,
2
2

L I
= f + ma + a 2 F + f A + h.c. L1 , m + a,
I b
2
1
1
+ Q MQ,
L2 = im m ( + )
2
2
2 f 0
f 2 0
m
M =
(4)
, 2 = m ,
0
1 0
0 0 1
F, F ),
where the scalar quantum fields are denoted by the column matrix Q: QT = (A, A,
A, F , F ). The matrix M is the same as the matrix appearing in Eq. (15) of
Q = (A,
Ref. [10]. There is the special case, called on-shell, of the background values a, f :
f = (m + a)f = 0,
f + ma + a 2 = 0,
(5)
2
which satisfies the field equation and makes L1 vanish. When the above background values
(a, f ) satisfy the on-shell condition above, V0eff reduces to

V0eff on-shell = ff 0,
(6)
which shows the positive semi-definiteness. This shows the characteristic aspect of the
supersymmetric configuration. The (classical) vacuum is given by: f = f = 0, a(m +
2
2 a) = 0 (a = 0 or a = m). In the following, except when explicitly stated, we do not
require the on-shell condition (5). We regard a and f not as specific constants (specific
vacuum) but as the general source (external) fields appearing in the effective potential. It
is an off-shell generalization but is the most natural one based on the background field
method.
260
Let us now evaluate the 1-loop quantum effect. First we can integrate out the auxiliary quantum-fields F and F using a squaring equation: F F + AF + A F =

Then the quadratic-part Lagrangian L2 reduces to L :
(F + A)(F + A)
AA.
2

L2 = im m + A2A

0
1
1
A
(A A)M
A
2
2

f
M=
.
(7)
f
The eigenvalues of M are given as

+ ff ,
ff .
m2 =
m2+ =
(8)
eff , from the bosonic part (scalar

The contribution to the 1-loop effective potential V1-loop
loop), is evaluated as

1
A
DA DA exp i d 4 x A2A
( A A ) M
A
2

m2i
1
2
2 1/2
= exp Tr
ln 1
= det 2 m+ 2 m
2
2
i=+,

eff
.
= exp i d 4 x V1-loop
(9)
The V1eff
-loop above lacks the fermionic 1-loop contribution:

1
D D exp i d x im m
2

+1
= det(2 )

0
(10)
This part does not depend on f and f. It says the 1-loop effective potential calculated
only by the scalar part is correct up to the f -independent terms. As far as the f -dependent
part is concerned, the scalar part result (9) is sufficient. If we trace the source of this
phenomenon, it is simply that the auxiliary fields f and f have the higher physical dimension, M 2 . They cannot have the Yukawa coupling with fermions. (F has the mass
eff
dimension 5.) This fact means that dV1eff

-loop /df (or dV1-loop /d f ) is definitely determined
only by the scalar part. Miller [10,17] utilized this fact, that is, F-tadpole or D-tadpole
[11] in general SUSY theories are rather simply obtained. In the present case, (1-loop)
F-tadpole corresponds to dV1eff
-loop /df . He noticed, if the SUSY is preserved in the quantization, the f -independent part can be fixed by the following boundary condition.2 We
2 This reminds us of the similar situation of 2D WZNW model and 2D induced gravity. Polyakov and
Wiegman [18] obtained the former effective action not by integrating the quantum field fluctuation but by
solving the chiral anomaly equation in 2D QED. Polyakov [19] obtained the latter effective action by solv-
261
follow Millers idea. Looking at the tree-level (on-shell) result (6), and taking into account
the quantum stableness of the SUSY theory, we are allowed to take the supersymmetric
boundary condition:
The SUSY effective potential vanishes at f = 0.
(11)
Normalizing at f = 0, the 1-loop effective potential is finally obtained as

V eff V eff
1-loop
1-loop f =0

2 ff
d 4k
ln 1 2
(2)4
(k + )
2

1
1
d 4k
2 ff
+ O 3 .
4
2
2
2
2
(2) (k + m )
1
=
2
(12)
In this simple example, we can explicitly see the subtracting term V1eff
-loop |f =0 is just given
by the fermion-loop contribution (10). The SUSY condition recovers the ignored (1-loop)
contribution in V eff . The middle expression of (12) is the same as Eq. (26) of Ref. [10]
The quadratic divergences appearing in the intermediate stages, as in (9) and (10), cancel
and the logarithmic divergence only remain in the final expression (12). It is absorbed by
the wave-function renormalization of the auxiliary fields as follows.3 In order to do the
renormalization, we first introduce a counterterm L in the following form:
1-loop
VR
eff
V1eff
-loop V1-loop |f =0 L,
1-loop
VR V0eff + VR
V0eff = ff + ,
L = Z ff,
(13)
where Z 1 + Z is the wave function renormalization factor of f and f. The 0th (classical) part, V0eff , is added. Now we fix Z by demanding the following renormalization
condition:4

d 4k
1
dVR
1 2
1
=
1
+ Z,

4 (k 2 + m2 )2
2
(2)
d(ff ) f =f=0,a=a=0
|k|
hence Z 1 + Z = 1
ln ,
2
m
16
(14)
ing the Weyl anomaly equation. They treated the gauge-field tadpole ( /A ) and the Weyl-mode tadpole
( / = g /g ), respectively.
3 No divergences for the coupling operator, AAF , and the mass term, mAF , are consistent with the nonrenormalization theorem (see a textbook [20]). The F-term part does not receive radiative correction. See, for
example, the Wests textbook [20].
4 We follow Ref. [9] in the choice of the renormalization condition. In (14), by setting the coefficient in front
of the term ff , appearing in the effective (renormalized) potential, we define the present renormalization. No
new mass parameter (such as the renormalization point) is introduced.
262
where is the momentum cutoff |k 2 | 2 . The anomalous dimension of the auxiliary

field is given by
Fb = ZF,

2
ln Z =
anomalous dimension of F : F =
(15)
+ O 4 .
2
ln
16
We see the quantum effect in the SUSY theory apparently appears in the scaling behaviour of the auxiliary field. It implies the structure (shape) of the effective potential is very
sensitive to the quantization. The final form, after the renormalization, is given by

1-loop
VR |on-shell = V0eff + VR
,
on-shell

eff
V0 on-shell = f f,
1-loop
VR
on-shell

1
()
2
2
2

=
ln
f
f
)
64 2
( ff )( + ff )

+ ff

ff ln
+ 2
ff

( ff )( + ff )
2
,
+ f f ln
m4

2
where ff = ma + a 2 , = |m + a|2 .
(16)
2
For the pure imaginary case of a = ib (b is a real number), the above potentials, V0eff |on-shell
2
1-loop
and V
|on-shell , are depicted in Fig. 1. In this case we have ff = m2 b2 + b4 , =
R
m2 + 2 b2 . The precise shape of the quantum correction depends on the renormalization

condition. However some characteristic features are considered meaningful. The shape of
1-loop
the 1-loop correction is not the ColemanWeinberg type. The total shape of VR
|on-shell is
eff
similar to the tree potential V0 |on-shell . This shows that the SUSY invariant vacuum, b = 0,
is stable against the quantum effect.5 The positive definiteness is preserved after the 1-loop
correction. The form of the potential does not essentially change. This result typically
shows a general feature of SUSY theories. It was confirmed before in the counter-term
calculation [21].
Super-QED is similarly treated in Appendix A. In this case the matter sector is vectorlike by introducing a pair of chiral multiplets. The SUSY boundary condition is taken at
D = 0. The anomalous dimension of the D-field and the -function of the coupling are
obtained. The 1-loop effective potential is explicitly obtained and its SUSY invariant properties are confirmed. The result will become an important reference in the analysis of the
bulk-boundary theory in the following sections.
5 Here we should be careful for the meaning of V |
R on-shell . It is still an off-shell quantity in the sense that
the true vacuum is realized at b = 0 only. Only at this vacuum, SUSY is preserved. We call VR |on-shell the
SUSY-invariant effective action because it is, at its vacuum, SUSY-invariant.
263
Fig. 1. The effective potential of the WessZumino model (16). The case a = ib (b: real) and the mass parameter
1-loop
|on-shell V1 , right) are
m = 1. The tree part (V0eff |on-shell V0 , left) and the 1-loop correction part (VR
depicted for the coupling parameter = 0.1 (up), = 1 (down). The horizontal axis is b. The potentials are even
functions of b.
3. MirabelliPeskin model
As a toy model of a bulk-boundary model, Mirabelli and Peskin proposed the following
system. Let us consider the 5-dimensional spacetime. The space of the fifth component is
taken to be S1 , with the periodicity 2l:
x 5 x 5 + 2l.
(17)
We also require the system to be (anti)symmetric with respect to the Z2 -symmetry:

Z2 transformation: x 5 x 5 .
(18)
This makes the two points, x 5 = 0 and x 5 = l, fixed points under Z2 -transformation. The
extra space is S 1 /Z2 orbifold. Let us consider 5D bulk theory Lbulk which is coupled with
4D matter theory Lbnd on a wall at x 5 = 0 and with Lbnd on the other wall at x 5 = l.
The boundary Lagrangians are, in the bulk action, described by the delta-functions along
the extra axis x 5 :

S = d 5 x Lblk + x 5 Lbnd + x 5 l L bnd .
(19)
We make use of the SUSY symmetry in order to make the problem simple. Both bulk
and boundary quantum effects are taken into account.
264
(i) 5D super-YangMills theory

We take, as the bulk dynamics, the 5D super-YM theory which is made of a vector
field AM (M = 0, 1, 2, 3, 5), a scalar field , a doublet of symplectic Majorana fields
i (i = 1, 2), and a triplet of auxiliary scalar fields X a (a = 1, 2, 3). The metric is
(MN ) = diag(1, 1, 1, 1, 1). We basically follow the notation of [22]:

2
1
LSYM = tr FMN 2 (M )2 i i M M i + X a + g i , i ,
2
FMN = M AN N AM + ig[AM , AN ],
M = M + ig[AM , ],

i
i
i
M = M + ig AM , ,
(20)
where all bulk fields are the adjoint representation of the gauge group G. The SU(2)R
index i is raised and lowered by the antisymmetric tensors ij and ij :6
AM = AM T ,
= T ,
i = i T ,

1
T , T = if T ,
tr T T = ,
2
X a = Xa T ,
(21)
where T is the generator of the group and f is the structure constant. (As for the
group indices , , . . . , there is no distinction between the upper one and the lower one.)
The bulk Lagrangian LSYM of (20) is invariant under the following SUSY transformation:
AM = i i M i ,
= i i i ,
i

i = MN FMN + M M i + i X a a j j ,
i
i

X a = i a j M M j + i , i a j j ,
(22)
where MN = 14 [ M , N ], and the SUSY global parameter i is the symplectic Majorana

spinor. This system has the symmetry of 8 real supercharges.7
As the 5D gauge-fixing term, we take the Feynman gauge
2
2

1
Lgauge = tr M AM = M AM
.
2
The corresponding ghost Lagrangian is given by

Lghost = 2 tr M c M (A)c = 2 tr M c M c + ig AM , c ,
(23)
(24)
where c and c are the complex ghost fields. We take the following bulk action:
Lblk = LSYM + Lgauge + Lghost .
(25)
6 The present notation: [A , ] = if A
M
M T = i(AM ) T = iAM , tr{[AM , ]5 } =
(i/2)f
AM 5 = i tr{(AM )5 }. Hence M = M gAM .
7 Two Dirac spinors 1 and 2 has a reality condition. The total number of the independent real SUSY-
freedom is 8, which is the same as that of one Dirac spinor.
265
Table 1
Z2 -parity assignment
P = +1, L
P = 1, R
AM
Am
A5
Xa
X3
X 1,2
(ii) Z2 -structure
In order to consistently project out N = 1 SUSY multiplet which has 4 real supercharges
(4Qs), we make use of the Z2 symmetry (18) which divides the 8Qs system into two
4Qs systems. We assign Z2 -parity, even (P = +1) or odd (P = 1) under the Z2 transformation, to all fields in accordance with the 5D SUSY (22). A consistent choice
is given as in Table 1. Note that Am , (m = 0, 1, 2, 3), is the 4D components of the bulk
vector AM . A symplectic Majorana field is expressed by two Weyl spinors. We can write
i and i as follows:

(L )
(R )
2
=
1 =
,
,
( R )
( L )

(L )
(R )
2
=
1 =
(26)
,
.
(R )
(L )
LSYM of (20) is invariant under the Z2 -transformation (18). On the wall (x 5 = 0), all oddparity states vanish. The parity odd fields, A5 and in Table 1, will play an important role
in the effective potential. The SUSY transformation (22) reduces to the following one of
even-parity states generated by L :
Am = i L m L + iL m L ,

L = mn Fmn L + i X 3 5 L ,

X 3 5 = L m m L L m m L .
(27)
(Note that the odd parity field appears in the x 5 -derivative form.) This is a N = 1 (4D)
vector multiplet (Am , L , X 3 5 ) transformation in the WZ-gauge although all fields
depend on the extra coordinate x 5 . Especially D X 3 5 plays the role of D-field. The
multiplet defined in the 5D world can be expressed in the form of the 4D vector superfied:
1
m + i 2 L i 2 L + 2 2 D, D = X 3 5 .
V = m A
(28)
2
The odd-parity fields ( + iA5 , i 2R , X 1 + iX 2 ) transform as a chiral (adjoint) multiplet:
( + iA5 ) = 2L (i 2R ),

R = i m F5m m m L + i X 1 + iX 2 L ,

X 1 + iX 2 = 2L m m R i5 L 2i[, L L ].
(29)
266
This multiplet can be expressed in the 4D superspace as

= ( + iA5 ) + 2 (i 2R ) + 2 X 1 + iX 2 .
(30)
We will not use this multiplet in this paper.

(iii) Matter Lagrangian on the wall
Let us introduce matter fields on the walls. We consider two cases: (a) chiral matter,
(b) vector-like matter.
(iiia) Chiral matter
We introduce, on the x 5 = 0 brane, a 4 dim chiral multiplet (, , F ) where is a complex
scalar field, is a Weyl spinor and F is an auxiliary field of complex scalar. This is the
simplest case as a matter candidate and was taken in the original theory [6]. The chiral
superfield is introduced:
= + 2 + 2 F .
= + 2 + 2 F,
(31)
Using the N = 1 SUSY property of the bulk fields (Am , L , D = X 3 5 ), we can find
the bulk-boundary coupling:

(a)
2gV 2 2
Lbnd = e

m
= m i m m + F F + 2ig( L L ) + g D,
(32)
where m m + igAm , D = X 3 5 . All matter fields are taken to be the fundamental
representation of the internal group G. We may add the following superpotential term to
the above Lagrangian:

1
1

m + + h.c.,
LSupPot =
(33)
2
3!
2
where the primed Greek suffixes ( , , . . .) show those of the fundamental representation.
(iiib) Vector-like matter
We introduce, as the 4D matter fields on the x 5 = 0 brane, one pair of 4 dim chiral multiplets, S = (S , S , FS ) and R = (R , R , FR ).

2gV S + R e2gV R 2 2 + m S R | 2 + S R | 2
L(b)
bnd = S e

+ m m
S + S m R R + FS FS + FR FR
= m

1
+
S i m m
S + R i m m
R + h.c.
2
+ 2g(i S L S i R L R + h.c.)

+ g S DS R DR + m{S FR + R FS S R + h.c.},
(34)
= igA . This is the bulk-boundary generalization of super-QED or QCD.
where m
m
m
We can identify the matter fermions (S , R ) as one Dirac fermion (electron, quark).
On the other brane x 5 = l, we introduce another WZ-multiplet(s), ( , , F ) for

case (a), and (S , S , FS ) and (R , R , FR ) for case (b). The bulk-boundary couplings
267
are fixed in the same way. The quadratic (kinetic) terms of the vector Am , the gaugino
spinor L and D = X 3 5 are in the bulk world. We note here the interaction between the bulk fields and the boundary ones is definitely fixed from SUSY. In the ordinary
standpoint of the field theory, the boundary theory (32) or (34) is perturbatively unrenormalizable because the coupling g has the physical dimension of M 1/2 (unrenormalizable
coupling).
4. Quantization using background field method

From the results of Section 2 (and Appendix A), we may put, for the purpose of obtaining the 1-loop effective potential, the following conditions on Lblk :
Am = 0 (m = 0, 1, 2, 3),
i = i = 0.
(35)
A5
Here the extra (fifth) component of the bulk vector

does not taken to be zero because it
is regarded as a 4D scalar on the wall. Then Lblk = LSYM + Lgauge + Lghost reduces to

3
Lred
blk , X , A5

= tr M M + X 3 X 3 M A5 M A5 + 2g(5 A5 )

A5 ] + irrel. terms,
g 2 (A5 )(A5 ) + Lred
ghost [c, c,

5

red
m
A5 ] = 2 tr m c c + 5 c c + ig A5 , c ,
Lghost [c, c,
(36)
where we have dropped terms of X1 X1 , X2 X2 as irrelevant terms because they decouple
from other fields. As for the boundary part, we may impose the conditions:
(a) chiral matter:
Am = 0,
(b) vector-like matter: A = 0,

m
= 0,
L = 0;
S = R = 0,
L = 0.
(37)
Lbnd reduces to
(a) chiral matter:

3
Lred
bnd(a) , , D = X 5

= m m + g X3 5 + gf A5 T + F F

m
( F + F )
+
2

+
( F + F + F ) + h.c. ;
3!
(b) vector-like matter:

3
Lred
bnd(b) S , S , R , R , D = X 5
= m S m S m R m R

+ g X3 5 T S
S R R
+ FS FS + FR FR + m(S FR + R FS + h.c.),
(38)
268
where , are the suffixes of the fundamental representation. In the same way, the boundred
= ( , F
ary Lagrangians at x 5 = l, Lbnd(a) and Lbnd(b) , reduce to Lbnd(a),(b)

3

F in (38)). Now we expand all scalar fields (, X , A5 ; , F, , F ), except ghost fields,
into the quantum fields (which are denoted again by the same symbols) and the background
fields (, 3 , a5 ; , f, , f ).
+ ,
X3 3 + X3 ,
A5 a5 + A5 ,
+
,
F
f
+
F
and
primed
ones.
chiral matter
S S + S , R R + R , FS fS + FS , vector-like matter
FR fR + FR and primed ones.
(39)
We treat the ghost fields as quantum ones.

In Section 2, for the purpose of obtaining the effective potential we consider the case that
the background fields are constant (in order to pick up the nonderivative part of the effective
action). In the present case of the 5D spacetime, we have the extra coordinate x 5 . Because
4D (x m -space) scalar property is independent of the extra space, we take into account the
x 5 -dependency of the vacuum configuration. The distribution along the extra coordinate is
important to make the localized configuration. We require that the background fields may
be constant only in 4D world, not necessarily in 5D world. We may allow the background
field to depend on the extra coordinate x 5 . This gives us an interesting possibility to the
extra space model. (See also the beginning paragraph of Section 6 where the necessity of
the present treatment is explained using an explicitly-x 5 -dependent solution (62).)
When the background fields (, 3 , a5 ; , f, , f ) satisfy the field equations derived
red
5 red
5
from Lred
blk + (x )Lbnd + (x l)Lbnd , using (36) and (38), the situation is called onshell. The equations (on-shell condition) are given as
:
5 2 + gf 5 a5 gf 5 (a5 ) g 2 f f a5 a5

+ g5 x 5 T + g5 x 5 l T

+ g 2 x 5 T + x 5 l T f a5
= 5 Z g(Z a5 ) = 0,
A5 :
5 2 a5 + gf 5 g 2 f f a5

+ g 2 x 5 T + (x 5 l) T f
= 5 2 a5 g( Z) = 0,
X3 :

3 + g x 5 T + x 5 l T = 0,

:

1
d T + m f + f = 0
2
( , f f in the left equation),
F
269

F :
1
f + m + = 0 ( , f f in the left equation), (40)
2
where d = ( 3 5 + ga5 ) is the background (4-dimensional) D-field and

Z g((x 5 ) T + (x 5 l) T ) 5 + gf a5 . In deriving the above
equations, we assume, based on the statement of the previous paragraph on the background
field, = (x 5 ), 3 = 3 (x 5 ), a5 = a5 (x 5 ), = const, = const. The total symmetricity
of m and with respect to the suffixes is also assumed. In the above derivation,
we use the fact that total divergences vanish from the periodicity condition. As stated below (6), we do not assume the above on-shell condition except when we state its use. We
regard the background fields as general external fields or as off-shell fields.
The quadratic part w.r.t. the quantum fields (, X 3 , A5 ; , F, , F ) gives us the 1-loop
quantum effect. That part of Lred
blk is given as

L2blk , A5 , X 3

= tr M M + X 3 X 3 M A5 M A5

+ 2g tr (5 A5 ) + (5 a5 ) + (5 A5 )

2g 2 tr (a5 )(A5 ) g 2 tr(a5 + A5 )2 + L2ghost [c, c],

L2ghost [c, c]
= 2 tr m c m c + 5 c 5 c + ig a 5 , c .
(41)
The quadratic part of Lred

bnd is given by

L2bnd(a) , , F, F ; , A5 , X 3
= m m + gd T ig 2 [A5 , ] T

+ g X3 5 ig[a5 , ] ig[A5 , ] T + T + F F

( f + 2 F ) + h.c. ,
+ m F +
2

d 3 5 ig[a5 , ] ,
L2bnd(b) = m S m S m R m R

+ g d S T S R T R ig[A5 , ] S T S R
T R

+ X3 5 ig[a5 , ] ig[A5 , ]

T R R T R
S T S + S T S R
+ FS FS + FR FR + m(S FR + R FS + h.c.),
(42)
2
2
where T (T ) . Lbnd(a)
and Lbnd(b)
are the same as L2bnd(a) and L2bnd(b)

except the replacement: , F F , , f f . Now we can integrate out the
270
2 . Using a squaring equation:8

auxiliary field X3 in L2blk + (x 5 )L2bnd + (x 5 l)Lbnd

1 3 3
X X + g x 5 T + T + x 5 l T + T X3
2
2

1
= X3 + g x 5 T + T + g x 5 l T + T
2

2

1
g 2 x 5 (0) T + T
2
2

1
g 2 x 5 l (0) T + T ,
(43)
2
we obtain the final 1-loop Lagrangian, necessary for the present purpose, as
Sa2 [, A5 ; , F ]

= d 5 X L2blk X3 =0

+ x 5 m m + gd T g5 T + T

+ m F +
( f + 2 F ) + h.c.
2
2

g
2
+ F F (0) T + T
2

5

+ x l { , , F F } ,

+ tr M M M A5 M A5
L2blk X3 =0 = L2ghost [c, c]

+ 2g (5 A5 ) + (5 a5 ) + (5 A5 )

2g 2 (a5 )(A5 ) g 2 (a5 + A5 )2 ,
(44)
for the chiral matter model.9 Similarly we obtain, for the vector-like matter, as
Sb2 [, A5 ; S , R ]

5
5
2
m S m S + m R m R
= d X Lblk X3 =0 + x

+ gd S T S R T R

g5 S T S + S T S R
T R R T R
+ FS FS + FR FR + m(S FR + R FS + h.c.)
8 Note the relation: (x 5 )(x 5 l) = 0.
9 Here we may omit the terms, ig 2 [A , ] T , ig 2 ([a , ] + [A , ] )( T + T ) in
5
5
5
2
Lbnd(a) of (42). From the Z2 -odd property of a5 , A5 , and , the above terms, with the term (x 5 ) or (x 5 l)
multiplied, have no contribution to the 1-loop effect. The same thing is used in (45).

2
g2

(0) S T S + S T S R
T R R T R
2

+ x 5 l { , , F F } .
271
(45)
For simplicity we consider the case of no superpotential: m = = 0, hereafter.10
5. Bulk and boundary quantum effects

Before the full 1-loop calculation of the next section, it is useful to look at some
important diagrams appearing in the perturbation w.r.t. the coupling g. We can express
propagators and vertices as in Fig. 2 (for the bulk part) and Fig. 3 (for the boundary and
mixed parts). All double lines express the background fields. The corresponding terms in
the Lagrangian, from which the Feynman rules can be easily read, are given by, for the
bulk part (Fig. 2):
Fig. 2. Feynman graphs for the bulk fields. (5P1), . . . , (5V5) of (46).
10 We examine the case with the superpotential in Ref. [23].
272
Fig. 3. Feynman graphs for the boundary at x 5 = 0 and bulk-boundary-mixed fields. (4P), . . . , (45V2) of (47).
For all vertex graphs, the overall factor, (x 5 ), is omitted. See (47) for detail.
1
(5P1): M M ,
2
1
(5P2): M A5 M A5 ,
2
(5V1): gf {5 A5 + 5 A5 },
(5V2): g 2 f f a5 A5 g 2 f f a5 A5 ,
(5V3): gf 5 a5 ,
g2
f f a5 a5 ,
2
g2
(5V5): f f A5 A5 .
2
(5V4):
(46)
They can be read from terms in L2blk |X3 =0 of (44). Those for the boundary at x 5 = 0 and
mixed parts (Fig. 3) are given by -function parts of (44):

(4P):
x 5 m m ,

(4V1):
x 5 gd T ,

2
1
(4V2):
g 2 x 5 (0) T ,
2
(4V3):
(4V4):
(45V1):
(45V2):

2
1
g 2 x 5 (0) T ,
2

g 2 x 5 (0) T T ,

x 5 g5 T ,

x 5 g5 T .
273
(47)
Those for the boundary at x 5 = l and mixed parts are the same as above except the replacement: (x 5 ) (x 5 l), , .
(i) Boundary (4D) quantum effect
All divergent diagrams (for the chiral matter model) up to the order of g 3 are listed up in
Fig. 4. All are 1-loop diagrams within the x 5 = 0 brane.
The diagram (a) is interesting because its presence says FayetIliopoulos D-term appears
in the boundary due to the radiative correction. It is quadratically divergent. The term is
proportional to Tr T , hence it exists only when the gauge group G involves U (1). If the
appearance really happens it could give a dynamical SUSY breaking (see a textbook [24]).
(Note that the tadpole diagram of massless field in 4D vanishes in the dimensional regu-
Fig. 4. Divergent Feynman graphs for the boundary part up to the order of g 3 .
274
Fig. 5. Divergent Feynman graphs for the bulk part up to the order of g 3 .
larization [14].11 Hence the presence of the FI D-term is rather subtle.) This D-term does
(b)
not appear in the vector-like matter case, Lbnd , because the S and R contributions cancel
each other. (The situation is the same as the super-QED.)
The diagram (b) was considered in Ref. [6]. It contributes, with (f) of Fig. 5 explained later,
to the self-energy of the scalar matter. This diagram (b) is independent of d, hence does
not contribute to the effective potential under the SUSY boundary condition.
The diagram (c) gives the renormalization of D-field. The tree part is in the bulk as
tr(X 3 X 3 5 5 ). (The corresponding part appears in super-QED. See d 2 -term of
Eq. (A.15).)
The diagram (d) gives, with (g) of Fig. 5, the renormalization of the gauge coupling g,
and contributes to the -function (g). (This part is very contrasting with the corresponding part of super-QED (da a-term).
We will discuss it in the final part of this section as

(g)/Fig. 5 + (d)/Fig. 4 part).
11 On the other hand, the dimensional regularization is generally considered non-appropriate for the SUSY
theories because the totally antisymmetric tensor is essentially involved with SUSY symmetry [25]. This
looks to obscure the presence of the valid calculation of the tadpole diagram.
The contribution to the effective potential of each diagram (of Fig. 4) is given by

d 4k 1
,
(a) and (b): i gd T g 2 (0) T T
(2)4 k 2

1
i2 2 1
d 4k
g d d
(c):
,
4
2!
2
(2) (k 2 )2

1
i2
d 4k
2 g 3 (0) d T T T
(d):
.
4
2!
(2) (k 2 )2
275
(48)
(ii) Bulk (5D) quantum effect

Among bulk quantum propagations, we do not, at present, consider those which propagate
between one brane and the other brane. Those diagrams play an important role as the
gauge mediation mechanism in Ref. [6]. The present interest, however, is not the SUSY
breaking mechanism. This simplification is admitted because we can control the ignored
contribution by adjusting the length l between the two branes.
All divergent diagrams are listed up in Fig. 5 up to the order of g 3 . Only the diagram (g)
contributes, others do not in the SUSY boundary condition.12 The diagram (f) was analysed
in Ref. [6] for the calculation of the matter-field self energy. The diagram (g) gives the bulk
contribution to the -function of the coupling g.
The contribution to the effective potential of diagrams (f) and (g) are given by

i2
(k 5 )2
1
2
,
(f): g 2 T T
2
5
2
2!
k + (k ) k 2
k5

g 3 T T T d (1)
(g):

k5
where

k5
d 4k 1
.
(2)4 2l 5
(k 5 )2
1
,
5
2
(k ) (k 2 )2
k 2
(49)
(50)
k lZ
The k 5 -summation comes from the KK-expansion for the bulk field , which will be
explained in (65). The result (f) of (49) is consistent with the corresponding term in (24)
and (25) of Ref. [6].
Using a formula

nZ
= coth(x),
x
x 2 + n2
(51)
12 D = X 3 is defined in the bulk. It plays the role of D-field on the boundary theory L
5
bnd . Its background field d = 3 5 (a) can be taken independent of 5 (a). The purely bulk diagrams (e) and (h) (of
Fig. 5), which contains 5 , are treated as d-independent ones. They do not contribute the effective potential in
the SUSY boundary condition.
276
the k 5 summation part in (f) and (g) can be evaluated as (k 0 = ik 4 , k 2 = (k 1 )2 + (k 2 )2 +

(k 3 )2 + (k 4 )2 k 2 )

(k 5 )2
(k 5 )2
=
5
2
(k )
k 2 + (k 5 )2
k 5 l Z
k 5 l Z

2

l
1
=
1 + k 2
l
k 2 ( )2 + n2
nZ
nZ

= 2l(0) + l k 2 coth l k 2 ,
(52)
where the Wick-rotation of k 0 -axis is done and the relation

1 = 2l(0),
(53)
k 2
nZ
is used. The above result tells us the contribution from the singular parts in the boundary,
that is, (0) parts in (b) and (d) of Eq. (48) (see also Fig. 4), cancel those in the bulk,
that is, (f) and (g) of (49) (see also Fig. 5). This phenomenon was pointed out for the selfenergy diagrams in [6]. In Appendix B, it is shown that the cancellation phenomenon more
generally (at the full order of the coupling within 1-loop) occurs in the effective potential.13
The final results are obtained as
(f)/Fig. 5 + (b)/Fig. 4:

1
id 4 k 1 2
g 2 T T
coth
l k 2
,
k
(2)4 2
k 2
(g)/Fig. 5 + (d)/Fig. 4:

1
id 4 k 1 2
g 3 T T T d
coth
l k 2
.
k
(2)4 2
(k 2 )2
(54)
2
We
note the
above results give correct 4D expressions in the limit of l k 1:
l k 2 coth(l k 2 ) 1. (See the da a-part
of super-QED, (A.15).) Hence the present bulkboundary system can be regarded as some deformation of the corresponding 4D theory.14
If we write the main deformation factor as follows

2
e2nl k ,
coth l k 2 = 1 2
(55)
n=0
the role of the extra space becomes clear. Because the spectrum above (En = n k 2 )
shows that of the harmonic oscillator in the temperature T = 1/(2l), it can be translated as
the whole (4D Euclidean) system is exposed to the heat-bath and in the equilibrium state
13 In Ref. [2], (x 5 ) is called classical singularity and the treatment of its divergence (0) is discussed using
a generalized renormalization group.
14 The similar propagator form appears also in 5D bulk (AdS ) approach at some limit [4].
5
277
with temperature
T = 1/(2l).15 The size of the extra space gives the inverse temperature.
The l k 1 limit in the previous paragraph corresponds to the high temperature limit.
Here the role of the singular term becomes clear. It is a counterterm to cancel the
divergences coming from the KK-mode summation. The (f) + (b) part is independent of d,
hence it does not contribute to the SUSY effective potential.16 (We expect (f) + (b) part is
canceled by the vector- and spinor-loop contribution.) This fact implies the above smoothing phenomenon takes place independently of the SUSY requirement. We should note that,
after the above cancellation, divergences, due to 4D-momentum integral, still remain. They
correspond to the ordinary divergences due to the (SUSY) local interaction.
Let us obtain renormalization group quantities from the previous result, T d -term.
The deformed propagator still makes the ultraviolet behaviour worse than the usual 4D
propagator:

1
d 4 k 2
1
l
(56)
=
{l ln 2 ln sinh l},
k coth l k 2
4
(2)
(k 2 )2 16 2
. (We do not
where and is the ultraviolet and infrared cutoffs respectively, |k|
care about the infrared divergence because it can be cured by taking the massive matter
multiplet.) It is linearly divergent. This result is reasonable from the power counting. Now
we consider the scaling behaviour of the gauge coupling in the renormalization procedure.
This problem is generally hard because the 5D gauge theory is regarded (perturbatively)
nonrenormalizable. In the present case, however, we expect this model reduces to a 4D
SUSY gauge theory in the limit l 0. We precisely define the limit as
g2
l
l 0,
(fixed) 1,
g 0,
g 2 l,
(57)
where the dimensionless coupling is introduced instead of g 2 . From the relation g 2 =

l, we are naturally led to introduce another cutoff instead of as
l ln .
(58)
(scaling)
This relation connects two transformations, scaling and translation: e
versus + / l (translation). Then the renormalization group -function of the dimensionless coupling is obtained as

1
gb = g + g = g 1 +
ln ,
8 16 2

2
1
1
2
b = (g + g) = 1 +
,
ln + O
l
4 16 2

d
2
d
1
0=
+
O
,
ln b = 1 + O()
ln +
4 16 2
d(ln )
d(ln )
15 The same thing is commented in Ref. [4] from the AdS approach.
5
16 This is consistent with N = 1 SUSY non-renormalization theorem.
278
d(ln )
ln =
1
,
4 16 2
(59)
where gb and b are bare quantities and G = SU(2) is taken (T T T = T /4, see
Eq. (B.3)). The above result coincides with that of the ordinary 4D chiral-gauge SUSY
theory (see textbooks [20,26]). We confirm here that the correct 4D renormalization works
although 5D quantum loop expression (56) is linearly divergent.
The previous paragraph confirms that the renormalization procedure works well at the
4D limit. In this paragraph, we argue a renormalization procedure for the general case of l
(not limited to the l 0 case) and propose a practical approach to define a finite quantity
from a divergent one (such as (54), (56)) coming from the 5D quantum effect. In the present
standpoint the extra axis is regarded as a regularization axis.17 We have already pointed
out that the macro parameter 1/ l plays a role of the temperature T = 1/(2l) to smooth the
UV behaviour. We recall a historically-famous fact in the beginning of the quantum mechanics. In the Planck distribution of the energy spectrum in a cavity, the light behaves like
a wave in the high temperature region compared with its own energy (RayleighJeanss
region, kB T h, where kB and h are the Boltzman constant and the Planck constant,
respectively) whereas it behaves like a particle in the low temperature region (Wiens region, kB T h). In the present case, we are treating 5D quantum dynamics which can
be regarded
as a system composed of 4D Euclidean system of light (oscillator) with en
ergy k 2 . They are thermally distributed in the heat-bath with the temperature T = 1/(2l).
Now we take a natural restriction on the present treatment of the 5D quantum field theory:
we cannot treat it in the Wiens region because, in this region, the light behaves like a
particle and some new mass (energy) unit (probably the Planck mass) should exist in the
theory. At present, however, we do not have such mass unit and it implies the present field
theory treatment breaks down in the Wiens region. (In other words, quantization in the
phase space of and l is lacking. ) We must switch to an unknown treatment in order to
obtain a meaningful quantity from the divergent ones (54), (56). Let us propose a condition
on the 4D momentum cutoff . We should choose in such a way that the structure of
the extra space (the circle in the present case) cannot be recognized in the 4D world:
T =
1
.
2l
(60)
If we adopt this idea,18 the integral (56) becomes well-defined (l 1). Here we propose
a sort of renormalization for the present 5D quantum system. Note here that the UV cut is essentially given by the inverse of the IR cutoff parameter
off , of the 4D momentum k,
l of the extra space. In this way, we have the following relation:

|k|
1
= T.
2l
(61)
17 The standpoint is the same as that appeared in the domain wall fermion of the lattice field theory [2729].
18 A comparative treatment was proposed by Randall and Schwarz [3,4]. They examined the UV-divergence
problem in 5D YangMills theory on the AdS5 geometry. In the analysis, 4D-momentum/extra-space-coordinate

propagator is taken. They take such a regularization that the 4D-momentum cutoff depends on the extra coordinate. They claim the linear divergence reduces to log-divergence.
279
This is a sort of mass hierarchy relation which appears in extra-dimensional models. The
relation (61) reminds us of the similar one that appears in the regularization of fermion
determinant (the domain wall fermion or the overlap formalism) in the lattice field theory.
(See (32) of Ref. [30], Eq. (29) of Ref. [31], and Eq. (26) of Ref. [32].) The chiral symmetry
in the fermion system corresponds to Z2 -symmetry in the present case.
In this section, we have confirmed that the renormalization works well as far as the 4D
world is concerned. Aside from the (5D) renormalization problem, we next examine the
vacuum structure.
6. Vacuum in the brane world and mass matrix

First we examine the vacuum in the present 5D approach. The relevant scalars are a5
and , that is, the background fields of A5 (the extra component of the bulk vector) and
(the bulk scalar), respectively. They should be, in principle, given by solving the (renormalized) equation of motion. They describe the vacuum. We usually take the following
procedure in order to obtain a vacuum.
Ordinary procedure for the vacuum search [33]
(1) First we obtain the effective potential assuming the scalar property of the vacuum
(as described in (35), (37)) and the constancy of the scalar vacuum expectation values.
(2) Take the minimum of the effective potential.
At the present case, however, we should take into account the x 5 -dependency and the
Z2 -property of the vacuum expectation value. We take the following forms of a5 (x 5 ) and
(x 5 ), which describe the localized (around x 5 = 0, l) configuration and a natural generalization19 of the ordinary treatment stated above:

a5 x 5 = a x 5 ,
x 5 = x 5 ,
5
5 +1 for 2nl < x < (1 + 2n)l,
x = 0
(62)
n Z,
for x 5 = nl,
5
1 for (2n 1)l < x < 2nl,
where (x 5 ) is the periodic sign function with the periodicity 2l. a and are positive
constants. See Figs. 6 and 7. It is shown, in Appendix C, that the above forms of a5 (x 5 )
and (x 5 ) satisfy the field equation of the present model. The periodic sign function can
be regarded as the thin-wall limit of a kink solution and shows the localization of the
19 The condition of constant is generalized to piece-wise constant. This is required from the necessity of a
non-trivial vacuum and the consistency with the Z2 odd property. We stress here the present generalization, that
is, the allowance of x 5 -dependence on the vacuum scalars (a5 and ), makes it possible to naturally introduce the
piece-wise constant ((x 5 )) in the theory. It is consistent with SUSY because the configuration (62) is obtained
as a solution of the present SUSY theory. See Appendix C. This situation should be compared with that appeared
in the work by Bergshoeff, Kallosh and Van Proeyen [34]. They replace some constants (masses, couplings, . . .)
with supersymmetric singlet fields which behave as piece-wise constants. They have to newly add (D 1)-form
field in order to keep SUSY.
280
Fig. 6. Behaviour of the background field a5 (x 5 ) (the extra component of the bulk vector).
Fig. 7. Behaviour of the background field (x 5 ) (the bulk scalar).
bulk scalar and the extra component of the bulk vector. This generalization is also natural
from the viewpoint that the present theory starts with the singular interaction (-function
term of (19).). We may use the piecewise-continuous or piecewise-smooth functions as the
theoretical materials, which is required from Z2 -property [35].
We now begin to prepare for the full (with respect to the coupling order) calculation of
the 1-loop effective potential. The 1-loop action, (44), can be expressed as

Sa2 = Safree + S ghost + d 5 X

1
A5
2

M
M 0
M
M M
M 0

M M
M MA5
0
0
MA5 MA5 A5
A5

+ and involving terms ,

free
Sa = d 5 X tr M M M A5 M A5 x 5 m m ,

S ghost = d 5 X M c M c + igf 5 c a5 c ,
where each component is read from (44) as

(M ) = g x 5 d T g 2 (0) x 5 T T ,
(63)
281

(M ) = g x 5 d T g 2 (0) x 5 T T ,

(M ) = g 2 (0) x 5 T T ,

(M ) = g 2 (0) x 5 T T ,

(M ) = (M ) = g5 x 5 T ,

(M ) = (M ) = g5 x 5 T ,
(M ) = g 5 f a5 + gf a5 5 g 2 f f a5 a5 ,
(MA5 A5 ) = g 2 f f ,
(MA5 ) = gf 5 gf 5 g 2 f f a5 g 2 f f a5 ,
(MA5 ) = gf 5 + gf 5 + g 2 f f a5
g 2 f f a5 .
(64)
In the present analysis, as mentioned in Section 5, we ignore the quantum propagation

between the two branes. We consider only the case that the quantum-loops propagate between the x 5 = 0 brane and the bulk or purely within the x 5 = 0 brane. Hence we may
ignore the (x 5 l)-terms in the above expression.
From the periodicity (x 5 x 5 + 2l) and the Z2 -odd property, the bulk fields
(X), A5 (X) can be KK-expanded as

1
n 5
5
x ,
n (x) sin
x, x =
l
l n=1

1
n 5
5
A5 x, x =
(65)
x ,
An (x) sin
l
l
n=1
where the normalization is taken in the way:
2
n=1 An (x) .
We evaluate the action term by term.
l
2 dx 5
2
n=1 n (x) ,
l
l A5
2 dx 5
(i) Free part of the bulk and boundary system

The free part Safree can be obtained as
!
2
2

k
k
free
4
Sa = d x
tr m k m k
k2 m Ak m Ak
A2k
l
l
k=1
"
m m .
(66)
From the Z2 -odd property, zero KK-mode does not appear. All quantum modes are massive
with the order of l 1 .
(ii) M , M , M , M (boundary part)
The boundary part can be read from (64).
282
(iii) M , M , M , M (bulk-boundary mixed part)

1
2

d 5 X (M ) =
1
2

d 5 X (M )

n
g
=
d 4 x T
n (x),
l
2 l
n=1

1
1
d 5 X (M ) =
d 5 X (M )
2
2

n
g
n (x).
=
d 4x T
l
2 l
(67)
n=1
The remaining ones are bulkbulk contribution.

(iv) M

int is evaluated as
The one part of 12 d 5 X (M ) Siv
int
Siv1
g 2

d 5 X f f a5 a5
g2
= f f a a
2

d 4x
n (x)n (x),
(68)
n=1
where we use a5 a5 = a 5 a 5 (x 5 )2 = a 5 a 5 . The other part can be expressed as

int
Siv2
g
2

d 5 X f a5 (5 + 5 )

= gf a
l
4
d x

dx 5 x 5 5 .
(69)
Using the Fourier expansion of the periodic sign function,

(x) =
(2n + 1)
4 1
sin
x ,
2n + 1
l
n=0
we obtain
int
Siv2
=
l
l
2g
f a
l

d 4x
m mQmn n ,
m=1 n=1

5
m 5
n 5
2l
5
dx x cos
x sin
x = Qmn ,
l
l
(70)
m n = odd,
0,
m n = even
1
1
{1 (1)mn } mn
,
= 2
0,
Qmn =
283
1
mn ,
m = n,
m = n.
(71)
We note the antisymmetricity: Qmn = Qnm .

(v) MA5 A5

1
int
Sv
d 5 X A5 (X)(MA5 A5 ) A5 (X)
2

g2
= f f
d 4 x An (x)Ak (x)
2l
n=1 k=1
l

k 5
n 5
x sin
x .
dx 5 x 5 x 5 sin
l
l
Using the localized form of (x 5 ) given in (62), we obtain

g2
d 4 x An (x)An (x).
Svint = f f
2
(72)
(73)
n=1
(vi) MA5 , MA5

This group consists of four terms

1
1
int
Svi
d 5 X (MA5 ) A5 =
d 5 X A5 (MA5 )
2
2
int
int
int
int
= Svi1
+ Svi2
+ Svi3
+ Svi4
,

1
int
=
Svi1
d 5 X gf 5 A5 ,
2

1
int
Svi2 =
d 5 X gf 5 A5 ,
2

1
int
=
Svi3
d 5 X g 2 f f a5 A5 ,
2

1
int
=
Svi4
d 5 X g 2 f f a5 A5 .
2
Here we note the relation

5 = 2 x 5 x 5 l ,
(74)
(75)
which expresses the localization of the bulk scalar. (x) is the periodic (periodicity 2l)
delta function. See Fig. 8. Using the above equation, we can evaluate the first term as
follows:

int
Svi1 = gf d 4 x [ A5 |x 5 =0 A5 |x 5 =l ] = 0.
(76)
284
Fig. 8. Behaviour of 5 (x 5 ).
The third and fourth terms are evaluated as

g2
int
Svi3
= f f a
d 4 x n (x)An (x),
2
n=1

2

g
int
= f f a
d 4 x n (x)An (x).
Svi4
2
(77)
n=1
Using the relation (71), we obtain

int
Svi2

g
= f
m mQmn An .
l
(78)
m=1 n=1
The background fields we take, (62), satisfy the required boundary condition. They also
satisfy the on-shell condition (40) for an appropriate choice of a,
,
, and 3 . Explanation is given in Appendix C.
We summarize the results of (i)(vi) as follows:

1
Sa2 = S ghost + d 4 x m Am
2

M
M 0
M
M M
M 0 n

M M
M MA
n
0
0
MA MAA mn
An
m

+ and involving terms ,
(79)
where the integer suffixes m and n runs from 1 to , and each component is described as

M = 2 + gd T g 2 (0) T T ,

M = g 2 (0) T T ,

M = g 2 (0) T T ,

M = 2 + gd T g 2 (0) T T ,
285
n
g
M = T
= M ,
n
l
n
l
g n
= Mn ,
M n = T
l
l

2
n
Mm n = 2 +
mn g 2 f f a a mn
l
4g
+ f a mQmn ,
l
2g
Mm An = g 2 f f a mn g 2 f f a mn f mQmn
l
= MAn m ,

2
n
mn g 2 f f mn ,
MAm An = 2 +
(80)
l
where the kinetic (free) part, Safree , is included ( 2 m m is the 4D Laplacian) in the
mass matrix. The repeated indices imply the Einsteins summation convention.20
7. Effective potential of bulk-boundary system

As shown in Section 2 and Appendix A for simple models, the effective potential is
obtained from the eigenvalues of the relevant mass-matrix obtained by the background expansion. Let us obtain the effective potential from the mass matrix M of (79) and (80).
It is composed of three field values , , d , two wall heights, a , , the gauge coupling, g and the boundary parameter l. The full explicit calculation, even at 1-loop level, is
technically hard. We obtain some interesting sections of the full result: case (A), = 0,
= 0; case (B), a = 0, = 0. Detailed explanation is given in Appendix B.
Case (A): = 0, = 0
We look the potential with the suppression of the scalar matter dependence. (Or we
may say we look the potential from the = = 0 point.) In this case the mass matrix
M has the following properties: (1) In M, the boundary part and the bulk one decouple
each other; (2) All (0)-terms disappear. The boundary-loop quantum effect gives rise to
the following potential before the renormalization procedure:

d 4k
g2 d 2
eff
,
ln 1
V1-loop =
(81)
4 (k 2 )2
(2)4
where we take G = SU(2) as the internal gauge group. The behaviour is similar to the
super-QED explained in Appendix A. The above expression, when treated perturbatively,
is logarithmically divergent. Noting the relation: d = ( 3 5 + ga5 ) , we realize
the renormalization procedure connects the boundary and the bulk phenomena through the
20 For the convenience, we list the physical dimensions of various quantities. [] = [ ] = M, [a]
= []
=
m
M 3/2 , [d] = M 5/2 , [g] = M 1/2 , [l] = M 1 , [(0)] = M, [] = [ ] = [n ] = [An ] = M.
286
field renormalization of X 3 and although we do not touch on the renormalization of the

bulk fields.21
The bulk-loop quantum effect does not give the d -dependence to the vacuum energy.
Hence it does not contribute to the effective potential after the use of the SUSY boundary condition. It gives, however, an important result: the scalar-loop contribution to the
vacuum energy which depends on the wall heights a and .
22 (We expect some part of
the contribution appear when the bosonfermion balance breaks down due to some SUSY
breaking mechanism.) We can regard it as a new type Casimir energy, because a and can
be regarded as different-type boundary parameters from l [36]. For the large circle limit
l , the final result of the new Casimir energy, per one KK-mode, is

1 eff
g2
(82)
VCasimir = 3 1 2 + 2 a 2 + 3 a + O g 4 ,
l
l
where 1 , 2 and 3 are some constants. The new points, compared with the ordinary
Casimir energy [7], are (1) the potential depends on the circle radius as l 3 ; (2) the potential
depends on the gauge coupling g; (3) the potential depends on the wall heights, and a.
We expect the above quantity (82) does not depend on the gauge we have chosen [37]. This
contribution from the scalar loop, however, is expected to be canceled by those from the
fermion and vector loops in the present SUSY-invariant setting.
Case (B): a = 0, = 0
In this case the brane structure disappears. The situation is similar to the case of Appelquist and Chodos (AC). From the bulk modes of and A5 , we have AC-type eigenvalues
2
n
2
, n = 1, 2, 3, . . .
n = k
(83)
l
(see (B.26) and (B.36)). This gives the famous form of the Casimir energy
1 eff
const
V
(84)
= 5 .
l Casimir
l
This is the scalar-loop contribution and is expected to be canceled by other nonscalar fields
effect.
The eigenvalues for the boundary part is obtained as a complicated expression involving
the following terms:

2
S ,
d 2 = d d ,
d V d ,
T ,
V 2 T . (85)
We have the full expression in the computer file. In the manipulation of eigenvalues search
(determinant calculation), we face the following combination of terms:
(0) +
(m/ l)2
1
.
l
k 2 (m/ l)2
(86)
m=1
21 The importance of the communication between the bulk and boundary renormalizations was stressed by
Goldberger and Wise [2].
22 No Casimir energy in the SUSY invariant theory is reasonable from the general result about the vanishing
energy of the SUSY vacuum.
287
The first term comes from the singular terms in M, the second from the KK-mode sum.
Using the relation mZ 1 = 2l(0), the above sum leads to a regular quantity
+ k2
1
2
2l
+ k + (m/ l)2
mZ

1
2
2
2 + k coth{l + k },
= 1
2
2
2 k cot{l k },
(0)|sm =
> k 2 ,
k 2 > .
(87)
We have confirmed this smoothing phenomenon occurs at the 1-loop full level.
The effective potential induced on the boundary comes from the eigenvalues depending
on the field d . When we look at d 2 -part, the following ones are obtained as the dominant
part
g 2
= k 2
(88)
d .
2
This is the same as case (A).
When we look at the d V d T part, the eigenvalues are dominated by the solutions of the following equation:

+ k2
g3
2 2
coth l + k 2 = 0.
+k d V
(89)
2
2
In the perturbative approach, this equation gives, in the O(g 3 ) order, two eigenvalues 1 , 2
which satisfy

2 2
g3
k 2 coth l k 2
1 d V
1 2 = k
(90)
4
(k 2 )2
(see (B.48)). This is the same as (54). The eigenvalues obtained as the full solutions of
(89) gives the effective potential at the 1-loop full level. The corresponding diagrams are
shown in Fig. 9. The figure is a bulk-boundary generalization of the ColemanWeinbergs
case [9].23
We succeed in obtaining the full 1-loop eigenvalues induced by the bulk-boundary quantum effect.
8. Conclusion
We have analyzed the quantum structure of a bulk-boundary system by taking the example of the MirabelliPeskin model. The analysis is newly formulated by the background
field method. Feynman rules for the perturbative calculation are presented. We focus on the
(1-loop) effective potential and the vacuum energy. It is confirmed that the singular terms
well behave with the KaluzaKlein modes summation. The whole effect can be regarded
k 2 1, in (90), we see the result essentially reduces to the ordinary type appearing in 4D theory (such as a term, (A.15), in 4D super-QED).
23 If we take the 4D-limit, l
288
Fig. 9. The diagrams contributing to the effective potential at the full (w.r.t. g) 1-loop level. The dotted lines
represent quantum propagation of the bulk scalar , the directed single lines represent that of the boundary (4D)
scalars , . The double lines represent boundary background fields , and d. The wall represents the
x 5 = 0 brane. The corresponding eigenvalues are given by (89). This figure is the bulk-boundary generalization
of Fig. 2 of Ref. [9].
as some deformation of the 4D quantization. Its 4D reduction by l 0 is confirmed in the
renormalization group calculation. The characteristic relation among the 4D-momentum k,

UV-cutoff , and the IR-cutoff (S 1 radius) l appears. It comes from the requirement to escape from the linear divergence. The relation is the same one as in the lattice domain wall
fermion. In addition to the bulk scalar , the extra component of the bulk vector A5 plays
an important role in determining the vacuum. Especially their localized configurations are
exploited. In the treatment, the vacuum is generalized in the sense that scalars may depend on the extra coordinate x 5 . In the intermediate stage, we have obtained a new type
Casimir energy in addition to the ordinary type by Appelquist and Chodos. The obtained
result of the effective potential includes the bulk-boundary generalization of the Coleman
Weinbergs case.
We hope the present analysis advances further development of the brane world physics.
Acknowledgements
The authors thank N. Sakai for valuable comments when this work, still at the primitive stage, was presented at the Chubu Summer School 2002 (Tsumagoi, Gunma, Japan,
2002.8.309.2). Parts of the present results were presented at the seminar of DAMTP,
Univ. of Cambridge (2003.1.24). The encouraging comment by G.W. Gibbons is much
appreciated. Some results were also presented at the annual meetings of the Physical Society of Japan (Rikkyo University, Tokyo, Japan, 2002.9.1316; Thohoku Gakuin University, Sendai, Japan, 2003.3.2831; Miyazaki World Convention Center Summit, Miyazaki,
Japan, 2003.9.912) and SUSY04 (Tsukuba, Japan, June 1723). The authors thank the
audience for stimulated comments.
289
Appendix A. Effective potential of super-QED

Now we consider the super-QED. The action is most concisely described by one vector
superfield V and two chiral ones S (charge g) and R (charge g):

1
4
2
2 gV
gV
4
2
W W + mSR + h.c. ,
R +
S = d x d d Se S + Re
d xd
4
1
1
W = D 2 D V , V = m vm + i i + D,
4
2
S = AS (y) + 2 S (y) + FS (y),
R = AR (y) + 2 R (y) + FR (y),

y m = x m + i m .
(A.1)
We focus on the scalar sector of the effective potential, based on the following points:
(1) Lorentz invariance of the vacuum, (2) the 1-loop contribution from the nonscalar fields
(spinors, vectors) can be recovered by taking supersymmetric boundary condition. We
put the condition
vm = 0,
= = 0,
S = R = 0.
(A.2)
Then the Lagrangian of the super-QED reduces to the simple form

L[AS , FS ; AR , FR ; D: m, g]

= A S m m AS + FS FS + m(AR FS + A R FS ) + R S

1 2 1
D + gD(AS AS AR AR ) .
+
2
2
(A.3)
Now we expand all scalar fields around the background constants (aS , fS , . . .)
AS aS + AS ,
A S a S + A S ,
FS f S + FS ,
FS fS + FS ,
D d + D,
(A.4)
and similarly for AR , A R , FR , FR .

f, f, d] is defined as
The effective potential V eff [a, a,

f, f, d]
exp i d 4 x V eff [a, a,

= (DA DF DA DF )S,R DD

L I
4
exp i d x L[a + A, a + A, f + F, f + F , d + D]
I b

= exp i d 4 x V0eff
(DA DF DA DF )S,R DD

exp i d 4 x L2 + (quantum field)3 and higher-order ,
(A.5)
290
where ( I ) (AS,R , FS,R , A S,R , FS,R , D) are treated as the quantum fields and ( I )|b
(aS,R , fS,R , a S,R , fS,R , d) are as the background fields. V0eff is the tree (zeroth order) part
and is given below. L2 is the quadratic part and will be given in (A.10). The zeroth order is
L0 = V0eff

1
g
= fS fS + m(aR fS + a R fS ) + R S + d 2 + d(a S aS a R aR ).
2
2
The first order is given by

L I
L1 =
I b

= FS (fS + ma R ) + (fS + maR )FS + R S

g
g
+ AR mfS d a R + AS mfR + d a S + h.c.
2
2

g
+ D d + (a S aS a R aR ) .
2
(A.6)
(A.7)
There is a special choice of the background constants, the on-shell condition:

fR + ma S = 0,
fS + ma R = 0,
g
g
mfR + d a S = 0,
mfS d a R = 0,
2
2
g
d + (a S aS a R aR ) = 0,
(A.8)
2
and their complex conjugate. This is the solution of the field equation: L1 = 0. When the
background constants satisfy the above equations, the tree effective potential V0eff takes

1
V0eff on-shell = fS fS + fR fR + d 2 0.
(A.9)
2
From the results (A.8) and (A.9), the (classical) vacuum is given by the solution: fS =
fR = 0, d = 0, aR = aS = 0 where m = 0 is assumed. In the following analysis, however,
we consider the general case of the background constants. (We do not require the on-shell
condition: L1 = 0.)
The second order part L2 is given by taking the quadratic terms with respect to the
quantum fields (AS,R , FS,R , A S,R , FS,R , D). L2 can be expressed in the following form,
where F, F -involved terms are separated:24
AS
A S
1
L2 = ( A S AS A R AR D ) A
AR
2

AR
D
24 f and f disappear at this stage because the F and F -auxiliary fields appear in (A.3) only as quadratic terms.

+ FS FS + m(AR FS + A R FS ) + R S ,
2 + dg/2
0
0
0
gaS /2
0
2 + dg/2
0
0
g a S /2
A=
0
0
2 dg/2
0
gaR /2 .
0
0
0
2 dg/2 g a R /2
gaS /2
g a R /2
gaR /2
1
g a S /2
291
(A.10)
The above matrix A is the same as that in Ref. [17].25 Integrating out all auxiliary fields
D, FS , FR , FS , FR using squaring equations:

1
1
g(aS A S aR A R )D + c.c. + D 2
2
2

2
g
1
g2
D + (aS AS aR AR + c.c.) (aS A S aR A R + c.c.)2 ,

=
2
2
8
FS FS + m(AR FS + A R FS ) + (S R)
= (FS + mAR )(FS + mA R ) m2 AR A R + (S R),
L2 reduces to
GaS a S + d m2
GaS aS
M=
GaS a R
GaS aR
AS
A S
AR ) M
AR ,
A R
1
L2 = A S 2AS + A R 2AR + ( A S
2
2 = m m ,
(A.11)
AS
A R
Ga S a S
Ga S aS + d m2
Ga S a R
Ga S aR
GaR a S
GaR aS
GaR a R d m2
GaR aR
Ga R a S
Ga R aS
.
Ga R a R
2
Ga R aR d m
(A.12)
where
d dg/2. The four eigenvalues of M are obtained as
g
1 = d m2 = d m2 ,
2
g
g2
2
2 = d m 2GaS a S = d m2 aS a S ,
2
2
g
2
2
3 = d m = d m ,
2
g
g2
4 = d m2 2GaR a R = d m2 aR a R .
2
2
Then the 1-loop contribution is given as

(DA DA)S,R exp i d 4 x L2

G g 2 /4,
25 In the paper, however, the contribution from F and F is not taken into account.
(A.13)
292

1/2
= det(2 + 1 )(2 + 2 )(2 + 3 )(2 + 4 )
= exp i
1
Tr
ln(2 + i ).
2

d 4 x V1eff
-loop ,
V1eff
-loop =
(A.14)
i=1
Normalizing V1eff
-loop at d = 0, from the requirement of the supersymmetric boundary condition (Section 2), we finally obtain

eff
V1eff
-loop V1-loop d=0

2

g
d2
1
d 4k
ln 1
=
2
2 (k 2 + m2 )2
(2)4

2
d2
g
+ ln 1
2 (k 2 + m2 + g 2 aS a S /2)(k 2 + m2 + g 2 aR a R /2)
2
d(aS a S aR a R )
gg
+
2 2 (k 2 + m2 + g 2 aS a S /2)(k 2 + m2 + g 2 aR a R /2)
4

1
g
g3
d 4k
d 2 + d(aS a S aR a R )
(A.15)
+ O g4 .
4
2
2
2
4
8
(2) (k + m )
The last approximate form is logarithmically divergent. We introduce a counterterm L
as in the following form:
eff
VR = V0eff + V1eff
-loop V1-loop |d=0 L,
1
1
V0eff = d 2 gd(a S aS a R aR ) + ,
2
2
1
1
2
L = Zd + gd(a S aS a R aR ),
2
2
(A.16)
where V0eff is the tree part of the potential (A.6). Z = 1 + Z and gb = g + g are the
wave-function renormalization constant of D and the bare coupling constant, respectively.
We fix Z and g by imposing the following renormalization condition:

dVR
1
= ,
2
d(d 2 ) d=0,a=a=0

dVR
1

(A.17)
= g.

d[d(a a a a )]
2
S S
R R
d=0,a=a=0
Hence Z and g are fixed as

2

g2
1
d 4k
1
g
2
Z =
=
ln
,
2
2 16 2 m2
(2)4 (k 2 + m2 )2
|k|
g =
g3
4
|k|
d 4k
1
=
(2)4 (k 2 + m2 )2

g 3 1
2
ln
,
4 16 2 m2
(A.18)
293
where is the momentum cutoff, |k 2 | 2 . Then the -function of the coupling and the
anomalous dimension of the D field are given as

g2 1
2
,
gb = g + g = g 1
ln
4 16 2 m2
g 2 1

d
d ln g
ln gb =
1 + O g2
0
+ O g2 ,
2
d(ln )
d(ln )
2 16
(g)
g2 1
1 dg
=
,
g d ln
2 16 2

g2
ln Z =
(A.19)
+ O g4 .
2
ln
16
Finally the on-shell potential is obtained as

1-loop
eff
VR
V1eff
-loop V1-loop d=0 L,

1-loop
VR |on-shell = V0eff + VR
,
on-shell

1
V0eff on-shell = fS fS + fR fR + d 2 (Eq. (A.9)),
2
1-loop
VR
on-shell

1
g2 2
g
g
2 2
4
=
d 1
d
g d + m + d ln 1 +
4
64 2
2m2
2m2

2
g
g2
1 2m
1 + 2m
g2
2d
4 fS fS
2
4
gm d ln
f
ln
f
m 1+
S
S
g
2
g
g
2m4
1 + 2m2 d
1 + 2m
4 fS fS + 2m2 d
Z = 1 + Z,
2
g2
1 + 2m
g2
4 fR fR
m 1+
fR fR ln
2
4
g
g
2m
1 + 2m
4 fR fR 2m2 d

g2
g2
g
g2
g
+ d 2 ln 1 +
f
+
d
1
+
f
d
f
f
S
S
R
R
4
2m4
2m2
2m4
2m2

g2
g
g2
f
d
f
fS fS +
+ gm2 d 1 +
ln
1
+
S
S
2m4
2m4
2m2

g2
g
g2
2
gm d 1 +
d .
fR fR ln 1 +
fR fR
2m4
2m4
2m2
4
(A.20)
For the case: m = 1, fS fS = fR fR ff ; the above potentials, V0eff |on-shell and

1-loop
|on-shell , are depicted in Fig. 10. The shape of the 1-loop correction is not the
VR
ColemanWeinberg type. Positive definiteness is preserved after the renormalization. The
potential minimum does not change. The minimum (fS = fR = 0, d = 0, aS = aR = 0)
corresponds to the SUSY invariant vacuum. This shows a characteristic feature of the
SUSY theory, that is, it is stable against the quantum effect. The form of the potential VR
does not essentially change from the tree one (A.9). The stableness of the vacuum was
294
Fig. 10. The effective potential of SQED (A.20) for the case: m = 1, fS fS = fR fR ff . The tree part
1-loop
|
V1 , below) are depicted for g = 0.3
(V0eff |on-shell V0 , above) and the 1-loop correction part (VR
on-shell
(left), g = 1 (right). The axis-ranges are 0 d 1.9, 0 |f | ff 3.
already pointed out, in the counter-term calculation, by Barbieri et al. [21]. We confirm it
by the explicit form of the renormalized potential.
Appendix B. Eigenvalues of mass matrix M and effective potential of the
MirabelliPeskin model
The effective potential of the present bulk-boundary model can be obtained from the
eigenvalues of the mass matrix M of (79) and (80). It is made of three field values
, , d , two wall heights, a , , the gauge coupling, g and the boundary parameter l.
The general case is hard to analyze explicitly. Here we consider two interesting sections:
(A) = 0, = 0 (bulk-boundary decoupled case); (B) = 0, a = 0 (bulk-boundary coupled case).
B.1. Effective potential from matter field vanishing pointcase (A) = 0, = 0
In this configuration, the interaction term gD T of (32) does not contribute to the
bulk-boundary loop. The bulk and boundary are decoupled in the quantum fluctuation. It
295
turns out, however, that the renormalization procedure to deal with the divergences due to
the boundary-loop makes connection between the bulk and boundary. (See the explanation
below (B.7).) M has the following form:

M
0
0 0
0 0 n
0
M
.

(B.1)
0 0
M MA
0 0 m
MA MAA mn
The components are given, from (80), as

M = 2 + gd T ,

M = 2 + gd T ,

2
n
mn g 2 f f a a mn
Mm n = 2 +
l
+
4g
f a mQmn ,
l
Mm An = g 2 f f a mn g 2 f f a mn
MAm An
2g
f mQmn
l
= MAn m ,

2
n
2
mn g 2 f f mn ,
= +
l
(B.2)
where the integer indices m, n run from 1 to . Qmn is defined in (71). The singular terms,
(0)-terms, disappear. The bulk and the boundary are decoupled, hence the eigenvalues can
be obtained separately.
For simplicity we take SU(2) as the gauge group G (f = ) and the doublet
representation for the matter fields ( = 1, 2)

T , T = i T ,
1
T = ,
2
1

Tr T T = ,
2
(B.3)
where ( = 1, 2, 3) is the Pauli sigma matrices, and 123 = 1.

(Ai) boundary part
The eigenvalues of

(M ) =

g
2 (d1 id2 )
,
k 2 g2 d3
(B.4)
d 2 d1 2 + d2 2 + d3 2 .
(B.5)
k 2 + g2 d3
g
2 (d1
+ id2 )
are
= k 2
g 2
d ,
2
296
The same ones are obtained for (M

). The effective potential can be obtained as

2 g 2
1
Tr ln(+ )2 ( )2 = Tr ln k 2 d 2
2
4

4
2
2
2 2
d k
g d
ln
k
.
=
+
ln
1
4 (k 2 )2
(2)4

V1eff
-loop =
(B.6)
Compare this result with the super-QED case (the first line of (A.15)). Taking the SUSY
condition (see the explanation given above (12)), we reach the final answer:

4
g2
d 4k
g2 d 2
d 4k d 2
V1eff
(B.7)
=
=
ln
1
+
O
g .
-loop
4 (k 2 )2
4
(2)4
(2)4 (k 2 )2
The last approximated form corresponds to (c) of (48), and is logarithmically divergent.
Because d is given by d = ( 3 5 + ga5 ) , the UV divergence of (B.7) is renormalized by the bulk wave function of X 3 and . Here the 4D worlds connection to the
bulk world appears. The quantum fluctuation within the boundary influence the bulk world
through the renormalization. The boundary dynamics does not close within the brane. We
do not touch on the renormalization of the bulk fields. After an appropriate renormalization, we expect the effective potential (B.7) leads to a similar potential to that given in
Appendix A for the case m = 0, ff = 0. (Note that the boundary theory treated in this
section is chiral, whereas the super-QED treated in Appendix A is vector-like.) We may
conclude that, in the vacuum specified by = 0, = 0, the renormalization works well as
far as the boundary world is concerned. The 4D theory is well defined.
(Aii) Bulk part
Let us evaluate the eigenvalues from the bulk part. Because it does not depend on d , this
part does not contribute to the effective potential in the SUSY boundary condition. However it is important to see what terms are quantumly induced by the scalar fields. (Those
terms are expected to be canceled by the fermion and vector fields contribution.) The result
depends on the heights of the 4D scalars, a and ,
in addition to the periodicity 2l. Generally that part of the vacuum energy which depends on the boundary parameters is called
Casimir energy. We regard a and ,
besides l, as those parameters. They correspond
to the brane tension and the brane thickness in the brane world. One of most important
points of the brane model is how to treat the KK-modes. We can see such a point in this
calculation.
The eigenvalue equation can be written as

n
Mm n Mm An
m
(B.8)
=
.
MAm n MAm An
Am
A n
From the symmetry, we can take the following general form as an eigenvector.
n = f1 (n)a + f2 (n) + f3 (n)f a ,
A n = g1 (n)a + g2 (n) + g3 (n)f a ,
(B.9)
297
where fi (n) and gi (n) are scalar quantities (with respect to the internal group transformation) which may depend on a and .
26,27 Through the above relation, the eigenvalue
equation for and A is replaced by that for fi (n) and gi (n). The eigenvalues are obtained
from the zeros of the determinant of the following matrix:
f1 (n) f2 (n) f3 (n) g1 (n) g2 (n) g3 (n)
a,
m
,
m
a ,
m
a,
m
,
m
a ,
m
c
(B.10)
The above 4 matrices are given as follows.

The first row equation of (B.8), Mm n n + Mm An A n = m , gives two matrices a and c as
( k 2 (m/ l)2 )

g 2 a
4gmQ a /
l
mn
a=
0
0
2g 2 a
mn
2
2
c=
2g a mn
2gmQmn / l 4
mn
mn
( k 2 (m/ l)2 g 2 a 2 )mn

4gmQmn a 2 / l
,
2
2
4gmQmn / l
( k (m/ l)2 g 2 a 2 )mn
g 2 2 mn
2gmQmn 2 / l
(B.11)
g 2 a
mn 2gmQmn a /
l .
2
0
g a
mn
The second row equation of (B.8), MAm n n + MAm An A n = A m , gives two matrices d and b as
2
g a
mn 2g 2 2 mn 2gnQnm 2 / l
d = g 2 a 2 mn
2g 2 a
mn 2gnQnm a /
l ,
4
2
2gnQnm / l
0
g a
mn
( k 2 (m/ l)2 g 2 2 )

0
0
mn
b=
g 2 a
mn
( k 2 (m/ l)2 )mn
( k 2 (m/ l)2 g 2 2 )mn
.
(B.12)
For convenience, let us introduce two quantities xm , ym ;

xm k 2 (m/ l)2 g 2 2 ,
ym k 2 (m/ l)2 ,
xm ym = g 2 2 .
(B.13)
26 The physical dimensions of the coefficients functions are as follows; [f ] = [f ] = M 1/2 , [f ] = M 2 ,

1
2
3
[g1 ] = [g2 ] = M 1/2 , [g3 ] = M 2 .
27 The change of the vector space of the eigenfunctions makes the number of eigenvalues change. We can,
however, choose proper values from the consistency with the perturbative results of Section 5.
298
Then the following relations are obtained:
0
0
xm mn
b = g 2 a
mn ym mn
0 ,
0
0
xm mn
1
0
0
xm mn
2
b1 = gx a
mn y1m mn
0 ,
m ym
0
det b =
2
xm ym .
1
xm mn
(B.14)
m=1
c,
Using a useful formula about general matrices a,
b,
and d (det b = 0, det a = 0);

a c
det = det a cb 1 d det b = det a det b d a 1 c ,
(B.15)
d b
we can decompose the determinant of the matrix (B.10). The components of the matrix
A a cb1 d are explicitly written as
f1 (n)
m
A11
A = a,
,
m
A21
a ,
m A31
f2 (n) f3 (n)
A12 A13
A22 A23
A32 A33

g 2 2
2 a 2
4
2 1
2 +
+
mn
(A11 )mn = ym mn g (a )
xm
ym
ym
Qmj Qnj
g 2 2
4 2 mn
,
xj
l
j

2
2
2
mn ,
(A12 )mn = g a 1 + 2g
xm

n
a
m
,
Qmn 4gm + 2g 3 2
+
(A13 )mn =
l
xm xn

Qmj Qnj

2 g 2 2 2
g 2 a
4a
2
(A21 )mn = g
a +
a (a )
mn + 4 2 mn
,
xm
ym
xj
l
j

2g 4 2 2
2 2
2
(A22 )mn = ym g a
2a (a )
mn ,
xm

n
Qmn
m 2 2
2
3
2
(A23 )mn =
a (a )
4g a m + 2g
,
l
xm xn

n
g 3 a
m
(A31 )mn = 2 4 Qmn
,
xm xn
l

Qmn
3 2m
4gm
+
4g
,
xn
l4

Qmj Qnj
)
2
g 2 2
2 2
4 (a
mn 4 2 mn
(A33 )mn = ym g a g
,
xm
xj
l
299
(A32 )mn =
(B.16)
where the repeated integer suffixes do not mean the summation.28 The summation should
be taken only where the symbol j appears.
Let us evaluate the eigenvalues from the zeros of det A. General case is technically
difficult. We consider the following special cases.
We consider the following limit:
g 2
g2
= fixed 1,
l
l a = fixed,
l = fixed,
l .
(B.17)
This is the situation where the circle is large compared with the inverse of the domain wall
height. (a and have the dimension of M.) Then the elements of A reduce to

g 2 2
2 a 2
1
2
2 +
+
mn ,
(A11 )mn = ymn g 4 (a )
x
y
y

2
mn ,
(A12 )mn = g 2 a 1 + 2g 2
x
(A13 )mn = 0,

a 2 g 2 2 2
a +
a (a )
2 mn ,
(A21 )mn = g 4
x
y

4

2g 2 2
2a (a )
2 mn ,
(A22 )mn = y g 2 a 2
x
(A23 )mn = 0,
a =
(A31 )mn = 0,
(A32 )mn = 0,

(a )
2
mn ,
(A33 )mn = y g 2 a 2 g 4
x
(B.18)
where x k 2 g 2 2 , y k 2 . We notice, in this limit, Qmn -terms disappear. In

the propagator terms xm , ym , KK-mass terms m2 2 / l 2 disappear. All KK-modes equally
contribute to the vacuum energy. The condition 0 = det A = det A33 det A22 det(A11
A12 A22 1 A21 ) gives us the following eigenvalues:
(i) det A33 = 0 gives two eigenvalues 1 , 2 whose product is given by

2.
1 2 = k 2 + g 2 2 k 2 + g 2 a 2 g 4 (a )
(B.19)
28 Because of this, the product Q ( n + m ) appearing in (A )

mn xm
13 mn of (B.16), does not vanish in spite of the
xn
antisymmetricity of Qmn .
300
(ii) det A22 = 0 gives 3 , 4 whose product is

2
2 .
3 4 = k 2 + k 2 g 2 2 + a 2 + g 4 3 2 a 2 + (a )
(B.20)
(iii) det(A11 A12 A22 A21 ) = 0 gives 5 , 6 , 7 , 8 , 9 whose product is given by

2

2 a 2 2
5 6 7 8 9 = k 2 + g 2 2 k 2 + g 4 (a )
2

k 2 + g 2 k 2 a 2 + 2 + 3g 4 (a )
(B.21)
2 a 2 2 .
In particular, for the special case a = 0, the nontrivial factor is only k 2 + g 2 2 . Hence each
KK-mode equally contribute to the vacuum energy as

2
d 4k
eff
2
.
ln
1
+
g
V1KK-mode
(B.22)
(2)4
k2
This quantity is quadratically divergent. After an appropriate normalization, which we do
not know precisely, the final form should become, based on the dimensional analysis, the
following:

2
a 2
a
1 eff
+ O g 4 ,
V1-loop = g 2 c1 3 + c2 3 + c3 3
(B.23)
l
l
l
l
where c1 , c2 and c3 are some finite constants. This is a new type Casimir energy. Comparing
the ordinary one (B.27) explained soon, it is new in the following points: (1) it depends on
the brane parameters and a besides the extra-space size l; (2) it depends on the gauge
coupling; (3) it is proportional to 1/ l 3 .
B.2. Effective potential with no brane structurecase (B) = 0, a = 0
Let us evaluate the case (B), = 0, a = 0. In this case 5D vacuum does not have the
brane structure. The situation is similar to the case of Appelquist and Chodoss work. The
matrix M has the form

M
M 0
M
M M
M 0 n
,

(B.24)
M M
M
0
0
0
0
MAA mn
m
where each component is described as

M = 2 + gd T g 2 (0) T T ,

M = g 2 (0) T T ,

M = g 2 (0) T T ,

M = 2 + gd T g 2 (0) T T ,
n
g
= M ,
M = T
n
n
l

l
n
g
M n = T
= Mn ,
l
l

2
n
mn ,
Mm n = 2 +
l

2
n
MAm An = 2 +
mn ,
l
301
(B.25)
where the integer indices m and n run from 1 to . Qmn -terms disappear. Let us find the
eigenvalues of the above matrix.
MAA part is decoupled with others, hence the eigenvalues of the part is obtained as

n = k 2
n
l
2
,
n = 1, 2, 3, . . . .
(B.26)
These correspond to the massive KK-modes of the fifth component of the bulk vector.
The eigenvalues (B.26) and another ones (B.36) explained soon, have the same form as
that appearing in the work by Appelquist and Chodos [7]. They contribute to the Casimir
energy:
1 eff
const
= 5 .
V
l Casimir
l
(B.27)
The eigenvalue equation for the other parts can be written as

M M
M1 M M
( M M )m
= .
m

M

M n

( M )mn
n
(B.28)
We can take the following form as the eigenvector, from the transformation property.

= h1 + h2 d + h3 V + h4 d V T ,

= k1 + k2 d + k3 V + k4 d V T ,
n = f1 (n)d + f2 (n)V + f3 (n) d V ,
(B.29)
where V T and hi , ki and fi (n) are functions which are made of , and
, ),
det(M1 I ) = 0, can be rewritten by that
d .29 The eigenvalue equation for (,
for (hi , ki , fj (n)). The eigenvalues are obtained from the zeros of the determinant of the
29 Their physical dimensions are as follows: ([h ], [h ], [h ], [h ]) = ([k ], [k ], [k ], [k ]) =
1
2
3
4
1
2
3
4
(M 0 , M 5/2 , M 2 , M 9/2 ), ([f1 ], [f2 ], [f3 ]) = (M 3/2 , M 1 , M 7/2 ).
302
following matrix M2 :
h1 h2 h3 h4 k1 k2 k3 k4 f1 (n) f2 (n) f3 (n)

d (T )
V (T )
d V (T )

d ( T )
V ( T )
d V ( T )
d , m
V , m
d V , m
c1
c2
d1
d2
b1
(B.30)
The components in each box are displayed in the following. For the purpose, we introduce here the following quantities which turn out to constitute the final result of the
effective potential:
4-dim scalar mass term: S = ,
D mass term: d 2 = d d ,
3-body term: d V = d V ,
4-dim 4-body term: V 2 = V V ,
where V = T .
(B.31)
The 9 matrices in (B.30) are given by as follows:

The first row equation of (B.28), M + M + M n = ,

gives three matrices a, c, c1 as
g
2
k
g
a=
2
g (0)
0
0
c= 2
g (0)
0
g
4d V
2
4d
2
k 2 g4 (0)S
0
2
k 2 g4 (0)S
2i g 2 (0)
ig
2
0
0
2
g4 (0)S
i 2
2 g (0)
0
0
0
0
.
gn
0
0
gn
2
g4 (0)S
0
1 gn
c1 =
0
l l
0

i
2
2
2 (gd V + g (0)V )
,
2i (gd 2 + g 2 (0)d V )
2
k 2 g4 (0)S
2i g 2 (0)V 2
,
i 2
g
(0)d
V
2
g2
4 (0)S
(B.32)
The second row equation of (B.28), M + M

303
+ M n n = ,
gives three matrices d, b, c2 as
0
0
d= 2
g (0)
0
k 2
g
b=
2
g (0)
0
2
g4 (0)S
2i g 2 (0)
0
1 gn
c2 =
0
l l
0
0
i 2
2
2 g (0)V
,
2i g 2 (0)d V
2
g4 (0)S
g 2
4d
2
k 2 g4 (0)S
0
0
2
g4 (0)S
g
4d V
,
i (gd 2 + g 2 (0)d V )
2i (gd V + g 2 (0)V 2 )
2
k 2 g4 (0)S
i 2
2 g (0)
ig
2
0
0
0
0
.
gn
0
0
gn
2
k 2 g4 (0)S
(B.33)
We note the relations a = b , c = d , c1 = c2 .

The third row equation of (B.28), Mm + M
m
+ Mm n n = m ,
gives three matrices d1 , d2 , b1 as
0
g4 mS
1
d1 =
gm
0
l l
0
i g2 m
0
g4 mS
1
d2 =
gm
0
l l
0
i g2 m
m 2 2
2
( k ( l ) ) mn
0
b1 =
0
0
g
4 mS
0
0
g4 mS
0
i g2 mV 2
i g2 md V ,
g4 mS
i g2 mV 2
g
i 2 md V ,
g4 mS
0
2 2
( k 2 ( m
l ) ) mn
0
0
.
0
2 )2
)
( k 2 ( m
mn
l
(B.34)
We note d1 = d2 .
Using the formula (B.15), the determinant of M2 (B.30) decomposes as follows:

c1
c2 ,
A
M2 =
d1 d2 b1

c1
b 1 ( d1
det(M1 I ) det M2 = det A
c2 1
a
d

c
,
b

d2 ) det b1 .
(B.35)
304
The last expression is a product of two determinants. The eigenvalues from the right determinant det b1 = 0 gives

m = k 2
m
l
2
,
m = 1, 2, 3, . . . ,
(B.36)
which correspond to the massive KK-modes of the bulk scalar . As for the left determinant, the matrix in the inside can be evaluated using the explicit expressions of (B.32),
(B.33) and (B.34). Here we find a smoothing procedure of the singular term takes place as
follows. We write the matrix A in (B.35) as A((0)) to show the (0) dependence explicitly. Then we find the following renormalization-like relation with respect to the singular
quantity (0):

A (0)

c1
b 1 ( d1
c2 1

d2 ) = A (0)|sm ,
(m/ l)2
1
l
k 2 (m/ l)2
m=1

+ k2
1
= (0) +
1+
.
2l
+ k 2 + (m/ l)2
(0)|sm = (0) +
mZ
Using the relation
mZ 1 = 2l(0),
(B.37)
mZ
(0)|sm becomes a finite (regular) quantity
+ k2
1
2l
+ k 2 + (m/ l)2
mZ

1
2
2
2 + k coth{l + k },
= 1
2
2
2 k cot{l k },
(0)|sm =
> k 2 ,
k 2 > .
(B.38)
The above relation manifestly shows that the tower of the massive KK-modes smooths the
(0) singularity appearing in the boundary part of the mass matrix M, (B.24). (In the
perturbative analysis of Section 5, the present smoothing phenomenon corresponds to the
cancellation of singularity appearing in Eqs. (49)(54).) In the limit | + k 2 | , (0)|sm
reduces to (0).
Next we evaluate det A (A A((0)|sm )) in order to find remaining 8 eigenvalues. We
repeat the formula: det A = det(a c (b )1 d ) det b , where the primed quantities are
defined by those ones which are obtained by replacing (0), in matrices A = (a, c/d, b),
by (0)|sm . Now we may deal with 44 matrices a c (b )1 d and b . We can explicitly
calculate det A (using an algebraic soft) and indeed obtain the expression. det A is the
function composed of the background quantities: S, d 2 , d V and V 2 defined in (B.31). It
is better to see some sections rather than the full result in order to see the structure of the
effective potential.
305
(i) d = 0 (d 2 = 0, d V = 0)
This case gives the normalization value of the effective potential in the SUSY boundary
condition:

3

g2

2 5
2

+ k + (0) S ,
det A = + k
2
sm
15 = k 2
(5-fold),

g2
(B.39)
S + k 2 coth l + k 2 = 0.
4
Let us look at the above full result from the perturbative approach and relate it to the
result of Section 5. First we do the propagator (1/k 2 ) expansion because the perturbative
approach is based on the expansion around the free theory: g = 0,

g2 2
1
l k2
2
2
2
coth l k
+ k + S k coth l k + 2
4
k 2
2(coth l k 2 )2

1
= 0.
+O
(B.40)
(k 2 )2
6 , 7 , 8 (3-fold):
+ k2 +
Secondly we restrict the coupling and the considered configuration as follows:

g2
= fixed 1,
l

l k 2 1.
(B.41)
The second equation is required for the validity of 1/k 2 expansion and it says the 4d
momentum integral should have the UV cutoff 1/ l. Taking into account the perturbative
order up to the 1st order w.r.t. g 2 / l and the 0th order w.r.t. 1/k 2 , we obtain

g2
k 2 coth l k 2
2
= k 1 + S
(B.42)
.
4
k2
This eigenvalue is consistent with the first part of (54). We must pick up one eigenvalue
from 1 5 , and three ones 6 , 7 and 8 (3-fold) in order to be consistent with the perturbative result.
(ii) V 2 = 0, others = 0 (d 2 = 0, d V = 0, S = 0)
We examine the part that is composed of purely the 4-body interaction term operator V 2 :

8
det A = + k 2 .
(B.43)
The term V 2 does not appear. This is desirable from the renormalization point of view.
The absence of the 4-body interaction term in the SUSY normalization part implies the
renormalization of this term works well without SUSY.
(iii) d 2 = 0, others = 0 (d V = 0, V 2 = 0, S = 0) (equivalently = = 0)
This is a special case of (A), the decoupled case.

g2 2 4

2 2
,
det A = + k d
4

g
d 2.
= k 2
(B.44)
2
306
Both + and are 4-fold eigenvalue. We pick up two eigenvalues of + (2-fold) and
another two ones (2-fold). This result is consistent with case (A).
(iv) d V = 0, others = 0 (S = 0, d 2 = 0, V 2 = 0)
We examine the part that is composed of purely the 3-body interaction operator d V :

g3

2 6
2 2
+ k (0)|sm d V ,
det A = + k
2
16 = k 2 (6-fold),

+ k2
g3
2 2
7 , 8 : + k d V
coth l + k 2 = 0.
2
2
(B.45)
The perturbative values are obtained as in (i). 1/k 2 -expansion gives

g3
1
l k2
2 2
2
2
2
coth l k
+ k d V k coth l k + 2
4
k 2
2(coth l k 2 )2

1
= 0.
+O
(B.46)
(k 2 )2
Taking the terms up to the 0th order w.r.t. 1/k 2 and up to the 1st order w.r.t. g 2 / l, we
obtain

2 g 3

+ k 2 d V k 2 coth l k 2 = 0.
(B.47)
4
This is a quadratic equation w.r.t. . The two roots 7 , 8 satisfy

2 2
k 2 coth l k 2
g3
7 8 = k
(B.48)
1 d V
.
4
(k 2 )2
This is consistent with (54). As for the four eigenvalues, we pick up 1 , 2 (2-fold) and
7 , 8 .
(v) S = 0, others = 0 (d 2 = 0, d V = 0, V 2 = 0)
We examine the part that is composed of purely the mass term operator S = . The form
of det A is the same as the case (i). Hence the effective potential is the same as (i). The
4D scalar mass term appears in the intermediate procedure, but it disappears in the SUSY
boundary condition. This shows the renormalization about the scalar mass term works with
the help of SUSY.
Appendix C. Background fields and on-shell condition

We show the background fields taken in Section 6 satisfy the field equation of (40), the
on-shell condition, for a special case given below. The assumed forms are

a5 x 5 = a x 5 ,
x 5 = x 5 ,
= const,
= const,
d = 3 5 + g(a5 ) = const,
(C.1)
307
where (x) is the periodic sign function defined by (62). First we stress that the total
derivative terms, appearing in the derivation of the field equation (40), can be safely put
to 0 because of the periodicity property. Using the relation (75) and the condition m =
= 0, the equations in (40) can be expressed as

2 5 x 5 x 5 l g 2 x 5 (a )
a

+ g5 x 5 T + g5 x 5 l T

+ g 2 x 5 T + x 5 l T a x 5 = 0,

2a
5 x 5 x 5 l g 2 x 5 (a )

g 2 x 5 T + x 5 l T x 5 = 0,

3 + g x 5 T + x 5 l T = 0,

g 3 2 x 5 x 5 l + g(a )
T = 0,

g 3 2 x 5 x 5 l + g(a )
(C.2)
T = 0.
We note the following things.
= f a = 0.
(1) When a , the following relations hold: (a )
(2) 5 ((x 5 ) (x 5 l)) const = 0 with the Neumann boundary condition:
5 (A5 )|x 5 =0 = 5 (A5 )|x 5 =l = 0.
(3) (x 5 )2 = 1, (x 5 )3 = (x 5 ), 5 ((x 5 )) = 2((x 5 ) (x 5 l)), 12 5 {(x 5 )2 } =
((x 5 ) (x 5 l))(x 5 ) = 0.
Then we can conclude that (C.1) is a solution of the field equation (40) for the following
choice:
g
g
const a = = T = T ,
2
2

3 = g x 5 x 5 l T .
(C.3)
In this choice d = 0 is concluded. The more general solution is given in [38].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209.

W.D. Goldberger, M.B. Wise, Phys. Rev. D 65 (2002) 025011, hep-th/0104170.
L. Randall, M.D. Schwartz, Phys. Rev. Lett. 88 (2002) 081801, hep-th/0108115.
L. Randall, M.D. Schwartz, JHEP 0111 (2001) 003, hep-th/0108114.
C.G. Callan, J.A. Harvey, Nucl. Phys. B 250 (1985) 427.
E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
T. Appelquist, A. Chodos, Phys. Rev. D 28 (1983) 772;
T. Appelquist, A. Chodos, Phys. Rev. Lett. 50 (1983) 141.
[8] S. Ichinose, Phys. Rev. D 66 (2002) 104015, hep-th/0206187.
[9] S. Coleman, E. Weinberg, Phys. Rev. D 7 (1973) 1888.
[10] R.D.C. Miller, Phys. Lett. B 124 (1983) 59.
308
[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]
S. Weinberg, Phys. Rev. D 7 (1973) 2887.

A. Murayama, Int. J. Mod. Phys. A 13 (1998) 4257.
B.S. DeWitt, Phys. Rev. 162 (1967) 1195, 1239.
G. t Hooft, Nucl. Phys. B 62 (1973) 444.
S. Ichinose, A. Murayama, Phys. Lett. B 587 (2004) 121, hep-th/0302029.
L.D. Faddeev, A.A. Slavnov, Gauge Fields: An Introduction to Quantum Theory, second ed., Addison
Wesley, Menlo Park, CA, 1991.
R.D.C. Miller, Nucl. Phys. B 229 (1983) 189.
A.M. Polyakov, P.B. Wiegman, Phys. Lett. B 131 (1983) 121.
A.M. Polyakov, Mod. Phys. Lett. A 2 (1987) 893.
P. West, Introduction to Supersymmetry and Supergravity, second ed., World Scientific, Singapore, 1990.
R. Barbieri, S. Ferrara, L. Maiani, F. Palumbo, C.A. Savoy, Phys. Lett. B 115 (1982) 212.
A. Hebecker, Nucl. Phys. B 632 (2002) 101, hep-th/0112230.
J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton Univ. Press, Princeton, NJ, 1992.
I. Jack, D.R.T. Jones, Regularization of supersymmetric theories, hep-ph/9707278, LTH 400.
P.G.O. Freund, Introduction to Supersymmetry, Cambridge Univ. Press, Cambridge, 1986.
D.B. Kaplan, Phys. Lett. B 288 (1992) 342.
K. Jansen, Phys. Lett. B 288 (1992) 348.
Y. Shamir, Nucl. Phys. B 406 (1993) 90.
S. Ichinose, Phys. Rev. D 61 (2000) 055011, hep-th/9811094.
S. Ichinose, Nucl. Phys. B 574 (2000) 719, hep-th/9908156.
S. Ichinose, Prog. Theor. Phys. 107 (2002) 1069, hep-th/9911079.
S. Ichinose, Nucl. Phys. B 231 (1984) 335.
E. Bergshoeff, R. Kallosh, A. Van Proeyen, JHEP 0010 (2000) 033, hep-th/0007044.
S. Ichinose, Phys. Rev. D 65 (2002) 084038, hep-th/0008245.
S. Ichinose, A. Murayama, Casimir effect in the brane world, in: K. Sato, S. Nagataki (Eds.), Proc. 6th RESCUE Int. Symp. on Frontier in Astroparticle Physics and Cosmology, Univ. of Tokyo, Japan, 47 November
2003, Universal Academy Press, Tokyo, 2004, p. 395, hep-th/0401015.
S. Ichinose, Phys. Lett. B 152 (1985) 56.
D-branes and extended characters in SL(2, R)/U (1)

Angelos Fotopoulos a,1 , Vasilis Niarchos b , Nikolaos Prezas c
a Theoretical Physics Group, Imperial College, Prince Consort Road, London SW7 2BZ, UK
b Department of Physics, University of Chicago, 5640 S. Ellis Av. Chicago, IL 60637, USA
c Institut fr Physik, Humboldt-Universitt zu Berlin, Newtonstrae 14, D-12489 Berlin, Germany
Received 12 July 2004; received in revised form 18 October 2004; accepted 22 December 2004
Abstract
We present a detailed study of D-branes in the axially gauged SL(2, R)k /U (1) coset conformal
field theory for integer level k. Our analysis is based on the modular bootstrap approach and utilizes
the extended SL(2, R)/U (1) characters and the embedding of the parafermionic coset algebra in the
N = 2 superconformal algebra. We propose three basic classes of boundary states corresponding to
D0-, D1- and D2-branes. We verify that these boundary states satisfy the Cardy consistency conditions and discuss their physical properties. The D0- and D1-branes agree with those found in earlier
work by Ribault and Schomerus using different methods (descent from the Euclidean AdS3 model).
The D2-branes are new. They are not, in general, space-filling but extend from the asymptotic circle
at infinity up to a minimum distance min 0 from the tip of the cigar.
PACS: 11.25.Hf; 11.25.-w
1. Introduction
Understanding the D-brane spectrum of string theory is a problem of paramount importance for well-known reasons. For string backgrounds that admit an exact conformal field
theory description, this spectrum can be uncovered by utilizing techniques from boundary
E-mail addresses: angelos.fotopoulos@cpht.polytechnique.fr (A. Fotopoulos),
vniarcho@theory.uchicago.edu (V. Niarchos), prezas@physik.hu-berlin.de (N. Prezas).
1 Also at the Centre de Physique Theorique, Ecole Polytechnique, Palaiseau 91128, France.
doi:10.1016/j.nuclphysb.2004.12.030
310
A. Fotopoulos et al. / Nuclear Physics B 710 (2005) 309370
conformal field theory. In this formalism, the physical properties of a D-brane are encoded
in a boundary state. The boundary states are generalized coherent states on the Hilbert
space of first quantized closed strings and they are specified by a set of gluing conditions
between the left- and right-moving chiral algebra generators.
In general, the classification of the full set of boundary states is a difficult open problem.
Boundary states need to obey various consistency conditions, which are mostly understood
in the case of rational conformal field theories. An important condition of this sort comes
from worldsheet duality and is known as the Cardy consistency condition [1]. This condition is the following statement. The general amplitude between a pair of boundary states
is a tree-level cylinder amplitude in closed string theory describing the emission, propagation and absorption of a closed string, and it can be written in terms of the characters that
appear in the torus partition sum. After a worldsheet duality transformation the cylinder
amplitude can be restated as a one-loop open string amplitude and one should be able to
express it in terms of open string characters with positive integer multiplicities. There is
a general way to solve this set of constraints in rational conformal field theories [1]. The
resulting boundary states are in one-to-one correspondence with the conformal blocks of
the theory and can be written as appropriate linear combinations of a set of coherent states
(Ishibashi states) satisfying a prescribed set of gluing conditions [2]. The coefficients of
this expansion are dictated by the S-modular transformation matrix elements, and the fact
that the Cardy consistency condition is satisfied is a consequence of the Verlinde formula
[3].
The above construction, usually referred to as the Cardy ansatz, has been employed
with great success in a variety of solvable rational conformal field theories, like Wess
ZuminoWitten (WZW) models based on compact Lie groups and their cosets (c.f. [4]
for a review and further references). For example, D-branes in the parafermion theory
SU(2)/U (1) were studied in detail in [5] with special emphasis on the interplay between
their geometrical interpretation and their algebraic conformal field theory treatment. It was
found that there are two types of parafermionic branes, which were called A-type and Btype in analogy to the branes of N = 2 theories [6]. The A-type includes D0-branes sitting
at special points on the boundary of the disc-shaped target space and D1-branes stretching
between them. The B-type contains (mostly unstable) D0- and D2-branes located at the
center of the disc.
The Cardy conditions are supplemented by a further set of sewing constraints [7,8],
which again are better understood only in rational theories. In more general (possibly
non-rational) conformal field theories it is mostly unclear how to implement the above constraints in a useful way. In certain cases, however, considerable progress has been made in
the last few years. The Cardy consistency conditions and disk 1-point function factorization constraints have been analyzed successfully in the bosonic Liouville theory [911] and
the N = 1 Liouville theory [12,13]. Similar work on the AdS3 WZW model has appeared
in [1418].
In this paper we would like to study D-branes in a non-compact, non-rational version of
the SU(2)/U (1) parafermion theory. This is the SL(2, R)/U (1) coset conformal field theory described by a gauged WZW model with axial U (1) gauging [1925]. The Lorentzian
version of this theory describes strings propagating on a two-dimensional black hole [26].
Its Euclidean counterpart, with a cigar-like target space geometry, has a great wealth of ap-
311
plications in string theory; it is a building block for exact string backgrounds with N = 4
superconformal symmetry [27,28], it appears in the holographic description of little string
theories [29,30], and it plays an important role in the worldsheet description of Calabi
Yau compactifications near singularities [31]. At the same time, it provides a useful testing
ground for ideas that might generalize to other non-rational conformal field theories.
The range of applications of SL(2, R)/U (1) renders a good understanding of its D-brane
spectrum obligatory. Further motivation comes from the recent reinterpretation of matrix
models for non-critical strings as effective field theories living on Liouville branes [32
35]. It would be very interesting to see if a similar story can be said for the SL(2, R)/U (1)
theory and the corresponding matrix model [36].
The exact algebraic analysis of D-branes in SL(2, R)/U (1) is hampered, however, by
the usual difficulties that one encounters in non-rational conformal field theories. Namely,
the existence of an infinite (and usually continuous) set of primaries in conjunction with
the absence of an analogue of the Verlinde formula, implies that one should go beyond the
Cardy construction. Nevertheless, the recent progress in bosonic and N = 1 Liouville theory [913] suggested that the naive extension of the Cardy ansatz to non-rational theories
may still yield consistent boundary states. This idea was further explored in [37,38], where
it was used to determine branes in the N = 2 Liouville theory. Our goal in this paper is
to apply a similar modular bootstrap method to the SL(2, R)/U (1) theory. We formulate
boundary states using the Cardy ansatz and subsequently verify that all possible overlaps
between these boundary states satisfy the Cardy consistency conditions.
Part of this work has been also motivated by a recent study on extended characters for
cosets SL(2, R)k /U (1) with integer level k [39]. These characters are direct analogues
of the extended characters of the N = 2 superconformal algebra, which were the basic
ingredient in the modular bootstrap analysis of [37]. The extended characters are useful
because they exhibit simple modular transformation properties and reorganize the representations of the theory in a compact way. In some sense, this feature rationalizes the
boundary state analysis and exhibits various similarities with the corresponding analysis of
the SU(2)/U (1) case.
D-branes on the cigar have been studied previously by several authors. A semiclassical
analysis based on the DiracBornInfeld (DBI) action appeared in [40]. More recently, [41]
proposed a set of boundary states for SL(2, R)/U (1), departing from a boundary conformal
field theory analysis of the H3+ (Euclidean AdS3 ) model [17]. The latter is an Euclidean
continuation of the SL(2, R) WZW model and it reduces to SL(2, R)/U (1) upon gauging
a U (1) subgroup. Extending standard techniques for the construction of boundary states
in rational coset models [42], the authors of [41] obtained SL(2, R)/U (1) boundary states
from the boundary states of [17] by descent. Three families of branes were obtained in this
way: D0-, D1-, and D2-branes. The analysis of the respective boundary states vindicated
the semiclassical expectations and several of the corresponding cylinder amplitudes were
shown to satisfy the Cardy consistency condition.
Our approach can be considered as complementary to that of [41] and is relevant for
cosets SL(2, R)k /U (1) at fractional level k. We find four classes of branes. Two of them
are the same as the D0- and D1-branes of [41]. For the D0-branes, in particular, we find
only one boundary state instead of the infinite multitude that appears in [41]. The authors
of that paper noticed several puzzling features for most of these branes and conjectured that
312
only one of them is consistent. This single D0-brane is precisely the one we obtain with
our methods. Moreover, we find D1-branes and D2-branes that differ from those of [41].
This extra class of D1-branes, which arises from a direct application of the Cardy ansatz,
exhibits some puzzling semiclassical features and it is not entirely clear if it satisfies further
factorization constraints. Our D2-branes cover the cigar only partially and are marginally
stable. They correspond to a set of extra solutions of the DBI action [40], which were not
considered in [41]. The D2-branes of [41] cover the whole cigar and can be contrasted
with a different set of D2-branes in our construction, which turn out to be inconsistent.
Comments on this comparison appear in Section 5.
The layout of this paper is as follows. In Section 2, we discuss several aspects of
the SL(2, R)/U (1) theory that will be relevant for our subsequent analysis. We present
the standard and extended coset characters and briefly review the embedding of the
SL(2, R)/U (1) parafermion algebra into the N = 2 superconformal algebra. This will
be useful in the definition of the coset Ishibashi states later in the paper. In addition, we
discuss in detail the torus partition function of the coset in terms of the standard and extended characters. In Section 3 we review some useful generalities on the Cardy ansatz
and the modular bootstrap method, lay out our notation, and then apply everything to
the SL(2, R)/U (1) case. First, we define the appropriate coset Ishibashi states and then
consider various boundary states that follow from the application of the Cardy ansatz. In
Section 4 we consider the Cardy consistency conditions. We analyze the overlaps between
our boundary states and derive the corresponding open string densities and multiplicities.
We find that the majority of the boundary states of Section 3 satisfy the Cardy consistency
conditions. Only one class is found to be inconsistent and we discuss possible alternatives.
In Section 5 we summarize the results of Sections 3 and 4, analyze the properties of each
of our boundary states and comment on their geometrical interpretation. More specifically,
we discuss the corresponding wavefunctions, the open string spectrum, and the relation of
our results with semiclassical expectations. We also compare our findings to those of [41].
We conclude in Section 6 with a brief discussion of our results and various ideas for further applications and open problems. There are also three appendices. In Appendix A we
demonstrate in detail how the coset partition sum can be written in terms of the standard
and extended SL(2, R)/U (1) characters. In Appendix B we derive the S-modular transformation properties for the extended identity characters of the coset. This is a small add-on to
the analysis of [39]. Finally, in Appendix C we provide several explicit expressions for the
extended identity characters, which are useful in the analysis of the open string spectrum
of the D0-brane.
2. SL(2, R)/U (1) characters and the torus partition sum
2.1. Representations and standard characters
The SL(2, R)k /U (1) coset is a two-dimensional conformal field theory with central
charge
ccs =
3k
1.
k2
(2.1)
313
It can be obtained from the SL(2, R) WZW model at level k by gauging a U (1) subgroup
of SL(2, R). Different choices of the U (1) subgroup yield different target space geometries. A non-compact U (1) subgroup results in a manifold with Lorentzian signature, the
2D black hole [26]. A compact U (1) leads instead to a manifold with Euclidean signature.
Moreover, for a given U (1) subgroup there are two possible anomaly free U (1) gauge symmetries: the axial g hgh, and the vector g hgh1 , where g(z, z ) are group elements
of SL(2, R) and h an element of the U (1) subgroup that is being gauged.
For a compact U (1) the axial gauging produces a two-dimensional manifold with the
cigar geometry
e = e0 cosh .
ds 2 = d 2 + tanh2 d 2 ,
(2.2)
is a positive real number and is periodic, i.e., + 2 . The tip of the cigar lies
at = 0, where the dilaton reaches its maximum value = 0 . In the asymptotic region
we obtain the geometry of a cylinder in the presence of a linear dilaton with
2
background charge Q2 = k2
.
The vector gauging yields a two-dimensional manifold with trumpet geometry
e = e0 sinh .
ds 2 = d 2 + coth2 d 2 ,
(2.3)
2
k . Unlike the cigar, the trumpet geom-
In this case, is a periodic coordinate with period

etry is singular and the dilaton diverges at = 0. It is believed, however, that the two
backgrounds are T-dual to each other when one considers worldsheet instanton corrections
to the trumpet geometry [30,45]. In this paper, we focus on the cigar geometry.
As a non-compact analog of the SU(2)/U (1) parafermion model, the coset SL(2, R)/
U (1) defines a non-compact parafermion algebra, which is known to possess the following
unitary representations2 and characters [41,4650].
I : (j = 0, k , r Z)
(i) Identity representation Dj,r
2

1
r2
1 2
Ir ( ) = ( )2 q 4(k2) q |r|+ k 1 +
(1)s q 2 (s +(2|r|+1)s2|r|) 1 + q |r|
s=1
(2.4)
with charge
= r.
: (0 < j < k2 , r Z)
(ii) Discrete representations Dj,r
J03
dj,r ( ) = ( )2 q
(j 12 )2
k2
(j +r)2
k

1
(1)s q 2 s(s+2r+1)
(2.5)
s=0
with charge J03 = j + r.

(iii) Continuous representations Cs,+r : (s R, 0 < 1, r Z)
s2
c1 +is,+r = ( )2 q k2 q
(+r)2
k
with charge J03 = + r.

2 Corresponding unitary representations can also be found in SL(2, R).
(2.6)
314

ccs
The above characters are defined, as usual, by the traces Tr q L0 24 over the corresponding
1
n
coset modules and q = e2i . The Dedekind eta function reads ( ) = q 24
n=1 (1 q )
3
and J0 is the U (1) generator of the parent SL(2, R) theory, which is being gauged.
The torus partition sum of the axially-gauged (cigar) coset can be written explicitly
in terms of the above continuous and discrete characters. We will return to this issue
later in this section to discuss it further and various important details will be provided
in Appendix A. For now, we make a small detour to remind the reader of the well-known
connection between the SL(2, R)/U (1) parafermion algebra and the N = 2 superconformal algebra. This will help establish our notation and later define and analyze the extended
coset characters, which will be the cornerstone of our boundary conformal field theory
analysis.
2.2. Embedding the SL(2, R)/U (1) parafermion theory in N = 2
One way to obtain the Virasoro characters appearing in Eqs. (2.4)(2.6) is by embedding the SL(2, R)/U (1) parafermion algebra into an N = 2 superconformal algebra. This
embedding will be central in the following considerations, so we take a moment to remind
the reader of a few relevant details. For further discussion see for example [48,55,56].
On the level of currents there is a well-known connection between SL(2, R) and the
N = 2 superconformal algebra. Indeed, we may introduce a free time-like chiral boson Y
and write the SL(2, R) currents in the form

i 2 Y
k
3
k .
J = i
(2.7)
Y,
J = k e
2
In this definition Y (z) is a canonically normalized scalar with the OPE
Y (z)Y (w) log(z w).
(2.8)
Then, we can decompose an affine SL(2, R) primary field j,m into an SL(2, R)/U (1)
parafermion j,m times a U (1) primary by writing

j,m = e
2
k mY
j,m .
(2.9)
By replacing the time-like boson Y with a new space-like boson , again canonically normalized as (z)(w) log(z w), we are able to construct the currents of an N = 2
superconformal algebra
J = i c,
G =
2c
e
i
c
1
T = TSL(2,R)/U (1) ()2 .
2
(2.10)
For SL(2, R) primaries in the D j+ discrete representation the J03 charge is m = j + r

with r 0. Then, corresponding N = 2 primary fields can be constructed as
Zj,j +r = e
+r)
i 2(j
k2
1
c
j,j +r .
(2.11)
315
In each of the above relations

k
c =
(2.12)
k2
denotes the SL(2, R) central charge at level k divided by 3. The dimension and U (1)R
charge of the Zj,j +r operators are
j (j 1) + (j + r)2
2(j + r)
,
Qj +r =
.
(2.13)
k2
k2
In order to account for the full discrete spectrum of the N = 2 algebra, we also have to
consider affine primaries belonging in the D j representation of SL(2, R) or, equivalently,
affine descendants of D j+ . In that case we have m = j + r + 1 with r < 0 and the conformal
weight and U (1)R charge of the corresponding N = 2 representation reads [39]
hj,j +r =
hj,j +r =
1
j (j 1) + (j + r)2
r ,
k2
2
Qj +r =
For the principal continuous series of SL(2, R) with j =

corresponding N = 2 representations have
1
2
2(j + r)
1.
k2
(2.14)
+ is, s 0 and 2m Z, the
j (j 1) + m2
2m
,
Qm =
.
(2.15)
k2
k2
The above discussion makes clear that there is an intimate connection between N = 2
and SL(2, R)/U (1) representations. Indeed, the unitary highest weight representations of
the N = 2 algebra and their characters fall also into three classes [57,58] (it is enough to
consider only the NS sector of the N = 2 theory).3
hj,m =
(i ) Identity (vacuum) representation: (h = Q = 0)
chI (, z) = q (c1)/8
1q
(1 + yq 1/2 )(1 + y 1 q 1/2 )
3 (, z)
.
3 ( )
(2.16)
This representation and its spectral flowed analogs to be discussed in the next subsection have two null vectors. The corresponding SL(2, R)/U (1) representations in
(2.4) have j = 0 and j = k2 .
(ii ) Discrete representations: (hj,j =
|Qj |
2 ,
0 < j < k2 )
chd (hj,j , Qj ; , z) = q hj,j (c1)/8

y Qj
1
3 (, z)
.
1/2
1+y q
3 ( )
(2.17)
These representations are built upon the primary vertex operators Zj,j and they
have one null vector. The states Zj,j and Zj,j are respectively chiral and antichiral. The corresponding SL(2, R)/U (1) representations are given by the discrete
. The extra index r coincides with the N = 2 spectral flow parameter (see
series Dj,r
below).
3 In [37] these representations are called graviton, massless matter and massive respectively.
316
(iii ) Continuous representations: (hj,m >
|Qm |
2 )
chc (hj,m , Qm ; , z) = q hj,m (c1)/8

y Qm
3 (, z)
.
3 ( )
(2.18)
For j = 12 + is, s 0, m = + r with 0 < 1 and r Z these representations

correspond to the continuous SL(2, R)/U (1) series Cs,+r .
In the above expressions we use the standard theta function notation
3 (, z) =

1
1
n2
1 q n 1 + yq n 2 1 + y 1 q n 2 =
ynq 2 ,
(2.19)
n=
n=1
where y = e2iz .
These N = 2 characters decompose naturally into a sum of products between appropriate U (1) and SL(2, R)/U (1) characters. One can show the following decomposition
formulae. For the identity representation we have

k2 2
y n q 2k n In ( ),
chI (, z) = 1 ( )
(2.20)
nZ
for the discrete representations

chd (hj,j , Qj ; , z) = 1 ( )
2j
y k2 +n q
2j
k2
2
2k ( k2 +n)
dj,n ( ),
(2.21)
nZ
and for the continuous representations

chc (h 1 +is,+r , Q+r ; , z)
2
2(+r)
k2 2(+r)
2
y k2 +n q 2k ( k2 +n) c1 +is,n(+r) ( ).
= 1 ( )
nZ
(2.22)
These decompositions are helpful for defining extended coset characters and for deriving
their modular transformation properties using known facts about the N = 2 extended characters. The extended coset characters will be the basic building block of our boundary
conformal field theory analysis in Section 3 and we now turn to their definition.
2.3. Extended characters
The standard N = 2 characters presented above have an obvious drawback. Under modular transformations they generate a continuous spectrum of U (1)R charges. This is an
undesirable feature, especially if one is interested in the formulation of superstring theory on a background, part of which is described on the worldsheet by the N = 2 theory
at hand. The GSO projection requires integral U (1)R charges and the modular transformations would spoil this requirement. Therefore, it is desirable to find a different set of
extended characters that possess integral U (1)R charges and which form a closed set
under modular transformations [5963]. Such characters can be defined in the following
way [37,39].
317
Consider a general N = 2 theory with rational central charge

2K
(K, N N).
(2.23)
N
In such cases, extended characters can be defined by summing the standard characters
over integer spectral flows. It is well known that the spectral flow generators U define an
automorphism of the N = 2 superconformal algebra with the form
c = 1 +
c
U1 Lm U = Lm + Jm + 2 m,0 ,
2
U1 Jm U = Jm + c
m,0 ,
(2.25)
U1 G
r U = Gr
(2.26)
(2.24)
and they act on a general N = 2 character to produce the new characters

c 2
ch (, ; , z) q 2 y c
ch (; , z + ).
(2.27)
In the notation of the previous subsection and in the special case of integer flow parameter
= n Z we simply get
ch (hj,m , Qm , n; , z) = ch (hj,m+n , Qm+n ; , z).
(2.28)
Extended N = 2 characters could be defined from the standard ones by summing over
all integers n in (2.28), but, as it turns out, this is not completely satisfying. Such extended
characters would include theta functions at fractional levels and they would still behave
badly under modular transformations [37]. This problem can be avoided by taking mod
N partial sums over spectral flows.
k
and it is fractional only for fractional
In our case, the central charge is given by c = k2
level k. This is not a terrible compromise because most (if not all) physically interesting
cases correspond to fractional levels. For example, critical bosonic string theory on the
coset SL(2, R)/U (1) demands k = 94 . Also, in many applications that involve the supersymmetric version of the SL(2, R)/U (1) coset in superstring theory (e.g., for applications
relevant to the physics of string propagation near CalabiYau singularities or string propagation in the near horizon limit of various NS5-brane configurations) the level k is an
integer. In what follows, we focus our discussion only on integer levels k. More general
fractional levels can be handled with appropriate modifications.
Consequently, let us set N = k 2 and K = 2 in (2.23). Then, following [37], for each
of the previously presented representations4 we introduce the N = 2 extended characters

chI (n; ),
I (r; ) =
(2.29)
nr+(k2)Z
d (j, r; ) =
chd (hj,j +n , Qj +n ; ),
(2.30)
nr+(k2)Z
4 We consider only the positive discrete representations here. The negative discrete representations can be
treated in an analogous manner.
318
c (s, m; ) =
chc (h 1 +is,n , Qn ; ),
nm+(k2)Z
(2.31)
where we restrict to z = 0 for simplicity. Note that in these definitions r Zk2 and 2m
Z2(k2) . Our notation for the continuous extended characters c (s, m; ) is related to that
of [37] by the equation

2
c (s, m; ) =
(2.32)
s, 2m; .
k2
Through the decomposition formulae (2.20), (2.21) and (2.22) the extended N = 2
characters naturally give rise to extended characters in the SL(2, R)/U (1) coset. These
extended characters have been considered recently in [39] and it is straightforward to derive their form. For example, let us consider in detail the identity characters. From (2.29)
we obtain

chI (r + (k 2)n; )
I (r; ) =
nZ
= 1 ( )
k2 2r+2(k2)n
+s)2
2k (
k2
k2 2r
2
2k ( k2 +s)
Is+r+(k2)n ( )
n,sZ
= 1 ( )
Is+r+kn ( ).
(2.33)
n,sZ
With the extended coset character definition

Ir ( )
Ir+kn ( ),
(2.34)
nZ
and the index r now restricted in Zk , we obtain the decomposition

k2 2r
2
I (r; ) = 1 ( )
q 2k ( k2 +n) In+r
nZ
2r+n(k2), k(k2) ( )In+r ( ).
(2.35)
We use the standard definition of the classical theta function

m 2
m,k ( ) =
q k(n+ 2k ) .
(2.36)
( )
nZk
nZ
We can also bring (2.35) into the form

I (r; ) = 1 ( )
2r+n(k2), k(k2) ( )In+r ( ).
nZ
(2.37)
A similar analysis can be performed for the discrete and continuous representations. For
the relevant details we refer the reader to [39]. The extended discrete coset characters are
defined by the equation

dj,r+kn ( )
dj,r ( )
(2.38)
nZ
319
with r Zk and they satisfy the decomposition formulae

d (j, r)( ) = 1 ( )
2(j +r)+n(k2), k(k2) ( )dj,n+r ( )
2
nZ
= 1 ( )
k2 2(j +r)
2
2k ( k2 +n)
dj,n+r ( )
nZ
( )
2(j +r)+n(k2), k(k2) ( )dj,n+r ( ).
(2.39)
nZk
Notice that as a consequence of the identity dj,r = dk
2 j,r
for the ordinary coset charac-
ters, one can obtain a similar relation for the extended ones: dj,r = dk
2 j,r
. This will be
useful for our computations in Section 4. We should add that the respective representations
are conjugate but not equivalent. This fact is easy to see by checking, for example, that
they have different S-transformation matrix elements (c.f. Section 2.5) [64].
Finally, the continuous characters are defined by the equation

c1 +is,m+kn ( ),
c1 +is,m ( ) =
(2.40)
2
nZ
where m = + r with = 0, 12 and r Zk . They satisfy the decomposition formulae

2m+n(k2), k(k2) ( )c1 +is,n+m ( )
c (s, m)( ) = 1 ( )
2
nZ
= 1 ( )
nZ
= 1 ( )
nZk
k2 2m
2
2k ( k2 +n)
c1 +is,n+m ( )
2
2m+n(k2), k(k2) ( )c1 +is,n+m ( ).

2
(2.41)
The above decompositions involve theta functions at half-integer level k(k2)

2 . These
functions have good modular transformation properties only for integer level, i.e., k even.
For odd k we can use appropriate theta function identities to rewrite everything in terms
of integer level theta functions. We will not discuss this case here and in what follows we
simply restrict ourselves to even integers k. Again, we want to stress that this is done only
for purposes of simplicity and that the present technology can be used to analyze any case
of fractional level k.
In the next section we intend to use the above defined extended coset characters as the
basic building blocks for the construction of boundary states in the coset SL(2, R)/U (1)
CFT. Our motivation for using the extended characters is based on the following observations.
(1) The extended coset characters have nice modular transformation properties. For
the continuous and discrete representations these properties have been derived recently in
[39]. For the identity representation we derive them in Appendix B. The extended coset
characters are closed under S-modular transformations and this situation bears many common characteristics with the case of the rational conformal field theories. In particular,
320
the infinity of standard coset characters for a given Casimir, is organized into a finite set
parameterized by a half-integer periodic charge taking values in Zk . In that respect, the
extended characters are expected to be better suited for the application of the Cardy construction in the case of irrational conformal field theories.
(2) We are focusing on the case of integer (or fractional) levels k. In many respects,
our analysis is similar to that of a compact boson at integer (or fractional) squared radius,
say R 2 = 2k, where the standard U (1) symmetry is enhanced to the affine U (1)k , or the
case of the bosonic parafermion theory given by the coset SU(2)k /U (1)k . In both cases
the extended characters play a vital role in organizing the (Virasoro) representations of the
theory into finite sets according to the enhanced symmetries. The SL(2, R)k /U (1) coset
with integer k shares many similarities with the above cases. Moreover, in the next subsection we show that when k is integer, the torus partition sum of the cigar CFT can also be
written in terms of the extended characters.
2.4. Character decomposition of the torus partition function
It is instructive to express the torus partition function of the axially-gauged (cigar) coset
in terms of the above standard and extended characters. This will clarify certain aspects of
the closed string spectrum in this theory and will provide a clear basis for the use of the
above extended characters in the case of integer level k. We present two results: the expression of the torus partition sum in terms of the standard characters appearing in Eqs. (2.5)
and (2.6) and the torus partition sum written in terms of the extended characters appearing
in Eqs. (2.38) and (2.40). The first expression was already implicit in [52]. Here we clarify
and elucidate certain aspects of that analysis. The supersymmetric version of the second
expression has appeared recently in Refs. [53,54] and it shares many common characteristics with our bosonic analysis.
In Appendix A, we determine that the torus partition sum of the cigar CFT can be recast
in terms of the standard continuous and discrete characters as
Z(, ) = 16

n,wZ 0
ds (s; m, m)
c1 +is,m ( )c1 +is,m ( )
2
1
k1
2 <j < 2
w,r,r Z
(2j + r + r + kw)dj,r ( )dj,r ( ),
(2.42)
where m and m
are quantum numbers related to the integers n and w by Eq. (2.44). The
spectral density of the continuous representations is given by

(is + 12 m)(is + 12 + m)
1 d
1
log +
log
(s; m, m)
=2
.
2
2i 4ds
(+is + 12 m)(+is + 12 + m)
(2.43)
The result (2.42) can be derived from another expression of the torus partition sum, which
follows from a direct path integral computation in the corresponding gauged WZW theory.
We should mention that Eq. (2.42) represents only part of the full answer. It includes the
full contribution of the discrete representations, but only part of the continuous. Roughly
321
speaking, the continuous contribution includes two pieces. One that has an infrared divergence and needs to be regularized and another that is finite. Eq. (2.42) includes only the
contribution of the first diverging term. The remaining finite part cannot be written in terms
of the standard characters and has not been included in (2.42). The interpretation of these
extra states is unclear [54], as they do not seem to fit in any of the known SL(2, R)/U (1)
representations. For more details we refer the reader to Appendix A.
In (2.42) only continuous and discrete representations appear. We can think of the continuous quantum number s as the momentum in the radial direction of the cigar, and
the quantum numbers m and m
as the J03 and J03 charges under the unbroken global
U (1) SL(2, R) symmetry that remains after gauging the axial U (1). These can be reexpressed in terms of the angular momentum n and winding number w in the compact
U (1) direction of the cigar. The precise relation is the following
n + kw
n kw
,
m
=
.
(2.44)
2
2
The reason for the different identification of charges between the left and right movers can
be traced back to the form of the BRST constraint of the axially gauged WZW theory [25].
This will be important later, when we discuss the different types of boundary conditions
on . The continuous part of the spectrum corresponds to wave-like modes propagating
along the cigar, whereas the discrete part consists of states localized near the tip.
The conformal weights of the primary fields of the cigar CFT can be read off directly
from the torus partition sum (2.42):
Continuous:
m=
hm =
s 2 + 14
m2
+
,
k2
k
Discrete:

j
hm =

j
h m =
1)
j (j
k2 +
1)
j (j
k2 +
1)
j (j
k2
1)
j (j
k2
+
+
j
h m =
s 2 + 14
m
2
+
;
k2
k
m2
k ,
m2
k + |m j |,
m j,
m
2
m
j,
k ,
m
2
j |,
k + |m
(2.45)
m < j,
m
< j.
(2.46)
The origin of the extra terms in the scaling weights of the primary fields with m, m
<j
can be traced to the leading order behavior of the function
Sr =

1
()s q 2 s(s+2r+1)
(2.47)
s=0
which appears in the definition of the discrete characters dj,r (see Eq. (2.5)). This behavior
depends crucially on the sign of r: for r > 0 we get Sr = 1 q 1+r + , but for r < 0 we
have instead Sr = q r q 2r+1 + .
The presence of discrete representation primary fields with m, m
< j in (2.46) might
appear to be in contradiction with the constraints on the allowed values of n and w appear-
322
ing in [25,51].5 The authors of [25] considered coset primary fields that can be written in
the form

V (z, z ) = TSL(2,R) (z, z ), L (z, z ), R (z, z ) eiqL Y (z)+iqR Y (z)
(2.48)
and by analyzing the normalizability properties of the SL(2, R) wavefunctions TSL(2,R) (z, z )
they concluded that the physical discrete coset primaries should satisfy the constraint
|n| < |kw|. In particular, this implies that no discrete states with zero winding should
be allowed. The authors of [51] derived this bound in the exact theory by looking at the
LSZ poles of non-normalizable vertex operators of the form (2.48). They argued that such
poles correspond to normalizable states of the theory and showed that they have to satisfy
the constraint |n| < |kw|.
We want to emphasize that this bound does not, in fact, contradict the result appearing in
(2.46). The states appearing in (2.48) correspond to affine primaries of the parent SL(2, R)
WZW theory. For such states the bound |n| < |kw| is, of course, correct. In the torus
partition sum the contributing coset states descend from SL(2, R) states that belong either
to D + D + or to D D [52]. Here D + and D denote affine SL(2, R) modules based
on lowest-weight and highest-weight discrete representations of SL(2, R), respectively. For
the special coset primaries (2.48) that descend from affine SL(2, R) primaries the charges
m and m
are given by (2.44) and they should have the same sign, i.e., they can be either
m = j + r, m
= j + r with r, r > 0 or m = j r, m
= j r with r, r > 0. In both cases,
this results to the constraint |n| < |wk|.
The torus partition sum (2.42) implies, however, that the coset conformal field theory
has more primary fields. Indeed, there are coset primary fields descending from affine
SL(2, R) descendants. This is precisely what we see in the discrete part of (2.42). States
with m < j appear in the physical spectrum, but they cannot be descending from affine
SL(2, R) primaries in D + , because the latter have necessarily m j . For example, the
coset primary state with m = j 1 would come from the affine descendant state J1
|j, j ,
which can easily be shown to be a Virasoro primary (for both SL(2, R) and the coset). Here
|j, j denotes the lowest-weight state of D + . The vertex operator for this state contains
derivatives of the target space fields and does not fall into the class of primaries considered
by [25,51] (see Eq. (2.48)). Finally, notice that we can also interpret the extra m < j states
.
as descending from = 1 spectral flowed affine primaries of D
k
2 j
For integer level k it is possible to rewrite (part of) the above partition sum (2.42) in
terms of the extended coset characters. This is straightforward for the discrete part. Starting
from (2.42) we can write
Zd (, ) = 8

1
k1
2 <j < 2

2j + r + r + k(n + n + w)
w,n,nZ
r,r Zk
dj,r+kn ( )dj,r +k n ( )
5 We would like to thank E. Kiritsis, D. Kutasov, and J. Troost for helpful discussions on these issues.
=8
1
k1
2 <j < 2
=8
1
k1
2 <j < 2
w,n,nZ
r,r Zk

323
(2j + r + r + kw)dj,r+kn ( )dj,r +k n ( )
(2j + r + r + kw)dj,r ( )dj,r ( ).
(2.49)
wZ r,r Zk
The continuous contribution involves a series of subtle steps, which are being explained in
detail in Appendix A. Again, the continuous contribution that can be written in terms of the
extended characters comes from that part of the full partition sum that is infrared divergent
and needs to be regularized. The final result takes the form

Z(, ) = 16
ds

gZ2k
(s; g)c1 +is, g ( )c1 +is, g ( )

2

1
k1
2 <j < 2
wZ r,r Zk
+8
(2j + r + r + kw)dj,r ( )dj,r ( )
with spectral density
(is +
1 d
1
log +
log
(s; g) = 2
2
2i 4ds
(+is +
1
2
1
2
g2 )(is +
g2 )(+is +
1
2
1
2
g2 )
g2 )
(2.50)
(2.51)
2.5. Modular transformations of the extended coset characters

The S-modular transformation properties of the continuous and discrete extended coset
characters have been derived recently in [39]. The identity representation does not appear
in the torus partition sum (2.50) and lies outside the physical spectrum of the coset. It
can appear, however, in the open string spectrum and for that reason we need to know
its S-modular transformation matrix as well. The necessary modular identity has been
derived in detail in Appendix B. For quick reference, we summarize here the S-modular
transformation properties of all the extended characters.
For the identity representation we have
Ir

sinh( 2s
1
k2 ) sinh(2s ) c
2i rm
k
=
ds e
1 +is , m ( )
m
| cosh (s + i 2 )|2
k(k 2) mZ
2
2
2k
2
+
k(k 2)
k2
sin
2j =2 r Zk
(2j 1) 2i r(2r +2j ) d

k
e
j ,r ( ). (2.52)
k2
For the discrete representations we get

(j +r)m
1
1
d
=
e2i k
j,r
k(k 2) mZ
2k
ds
m
cosh (s k4j
k2 + i 2 )
cosh (s + i m2 )
c1 +is , m ( )
2
324
k2

4i
1
4i
1
i

e k (j +r)(j +r ) e k2 (j 2 )(j 2 ) dj ,r ( )
k(k 2)

2j =2 r Zk
4i
1
i

e k (j +r)( 2 +r ) d1 ,r ( )
+
2 k(k 2)
2
r Zk

4i
1

e k (j +r)( 2 +r ) dk1 ,r ( ) ,
(2.53)
and for the continuous

1
c1 +is,+r

2
4ss
2 i
m(r+)

k
=
c1 +is , m ( ).
e
ds cos
k2
2
2
k(k 2) mZ
2k
(2.54)
Notice that the identity character does not appear on the r.h.s. of any of the above modular transforms. Hence, for a generic extended character labeled by we have an S-modular
identity of the form

1

c
m
= ds
S s , c1 m ( )

2
2 +is , 2

0
m Z2k
k1

2j =1
r Zk
S d (j , r | )dj ,r ( ).
(2.55)
3. Modular bootstrap for SL(2, R)/U (1)

3.1. Generalities on the modular bootstrap
D-branes in string theory can be formulated with the use of the boundary state formalism. The boundary states are specified by a set of gluing conditions that relate the left- and
right-moving generators of the chiral algebras (Virasoro or extended) of the bulk conformal
field theory. These conditions are intimately related to the boundary conditions imposed at
the ends of the open strings.
The construction of a boundary state begins with the formulation of Ishibashi states
[2]. In rational conformal field theories the spectrum of bulk primary fields is finite and
discrete, and for each primary field we construct one Ishibashi state as a coherent state
satisfying appropriate gluing conditions. The general gluing conditions respect conformal
invariance and they read
(Ln L n )|R = 0.
(3.1)
|R is the Ishibashi state built upon the primary field R. In general, one can impose more
symmetric boundary conditions that preserve in addition some affine symmetry.
325
In generic conformal field theories the spectrum typically contains an infinite set of
fields, some of which are labeled by a continuous parameter. In that case, one would like
to construct analogous Ishibashi states that will be labeled either by a continuous index
s or by a discrete index i. An orthonormal set of such Ishibashi states would satisfy the
following cylinder amplitudes:
s|eT H |s = (s s )s (iT ),
(c)
i|eT H |i = i,i i (iT ),

(c)
s|eT H |i = 0,
(c)
(3.2)
c
H (c) = L0 + L 0 12
is the closed string Hamiltonian, with T being the length of the
cylinder and s , i denote respectively the continuous and discrete characters of the chiral
algebra.
A boundary state can be constructed as a linear combination of the above Ishibashi
states. Let us denote the general boundary state as |X; . X characterizes the boundary
conditions and specifies what linear combinations of Ishibashi states appear in the boundary state. The problem of determining |X; reduces to the problem of determining the
consistent choices of . In rational conformal field theories, this problem has a well-known
solution. The cylinder amplitude between any two boundary states |X; 1 and |X; 2 can
be written in terms of a finite discrete set of characters i appearing in the torus partition
function consistently with the gluing conditions. Worldsheet duality transforms the cylinder amplitude into an annulus amplitude, where it can be viewed as a 1-loop open string
partition sum. As such, it should take the form

(c)
N (i; 1 |2 )i (it)
X; 1 |eT H |X; 2 =
(3.3)
i
with t = 1/T the annulus modulus and N (i; 1 |2 ) positive integer multiplicities. This
form is quite restrictive and it gives rise to a very useful consistency condition. This condition can be satisfied by the Cardy ansatz, which employs the S-transformation matrix
elements to determine the coefficients of the Ishibashi state expansion of the boundary
states (see below for more detailed expressions).
In non-rational conformal field theories the annulus amplitude receives extra contributions from an infinite set of characters

(c)
N (i; 1 |2 )i (it).
X; 1 |eT H |X; 2 = ds (s; 1 |2 )s (it) +
(3.4)
i
The generalization of the above condition would demand positive definite spectral densities
(s; 1 |2 ) and positive integer multiplicities N (i; 1 |2 ). Satisfying these conditions is a
highly non-trivial problem and it is not known if this modular bootstrap programme can be
implemented successfully in general. Recent progress [913] in bosonic and N = 1 Liouville theory, however, has led to a set of boundary states that pass the Cardy consistency
conditions (3.4) as well as a variety of other consistency conditions (various factorization
constraints on the disk). This provides further motivation for studying the solutions of the
Cardy conditions in other non-rational conformal field theories. In this spirit, the authors of
[37,38] formulated recently various classes of boundary states in N = 2 Liouville theory.
326
In this paper, we consider the Cardy consistency conditions in the SL(2, R)k /U (1) coset
conformal field theory for integer levels k. As we mentioned in Section 2, this situation
shares many common characteristics with the rational SU(2)k /U (1) case and the use of
the extended coset representations leads to a more controlled application of the modular
bootstrap method. For example, the extended Ishibashi states appearing in this section have
a finite range of charges and the whole construction stands in close resemblance with that
of [5] for the SU(2)k /U (1) parafermionic model. Further confidence for the validity of
our results is obtained by the recent analysis of [41], where coset boundary states were
determined from consistent H3+ boundary states [17] by descent.
3.2. Extended coset Ishibashi states from N = 2 decomposition
There are two special types of boundary conditions in a theory with N = 2 superconformal symmetry [6]:

A-type: (Jn Jn )|B = 0,

(3.5)
Gr i G
r |B = 0,

r |B = 0.
B-type: (Jn + Jn )|B = 0,
(3.6)
Gr i G
Both are compatible with the diagonal N = 1 superconformal symmetry
(Ln L n )|B = 0,
G+
r )|B = 0,
(Gr i G
(3.7)
G .
+
In these equations |B denotes a generic N = 2 boundary state.
where G =
The A-type/B-type boundary conditions impose Dirichlet/Neumann boundary conditions
on the free boson associated to the U (1) current J .
Extended N = 2 Ishibashi states for each of the above boundary conditions have been
presented recently in [37]. For example, let us consider in detail the A-type boundary
conditions and let us set in (2.23) N = k 2 and K = 1. As we mentioned in the previous
section, these values will be relevant for the analysis of the SL(2, R)k /U (1) coset. We
need to consider only Ishibashi states for the continuous and discrete representations. These
representations form a maximal subset closed under modular transformations and they will
be the only ones relevant for the subsequent analysis of the coset theory.
The A-type Ishibashi states will be denoted as

A; s, m , s 0, m Z2(k2) ,
(3.8)

2

A; j, r , 2 2j k 2, r Zk2
(3.9)

and they satisfy the orthogonality conditions

m
m
m
c
2(k2)
A; s, eT H A; s ,
= (s s )m,m c s, ; iT ,
2
2
2

T H c
A; j , r = j,j k2
A; j, r e

r,r d (j, r; iT ),

m
c
= 0.
A; j, r eT H A; s,
2
(3.10)
327
M denotes the usual Kronecker delta modulo M. A-type N = 2 boundary states can be
n,n

written as appropriate linear combinations of the above Ishibashi states. The specific form
of these combinations has been determined in [37] by imposing the following modular
bootstrap equations:
A; 0|eT H |A; = (it),

c
A; 0|e
T H c
(3.11)
|A; 0 = I (r = 0, it).
(3.12)
is a label that parameterizes the N = 2 boundary states which solve these modular bootstrap equations. A special set of solutions employs what is known as the Cardy ansatz. This
ansatz leads to boundary states of the form

|A; =
ds
0

mZ2(k2)

k2

m

+
(s, m)A; s,
C (j, r)|A; j, r
2
(3.13)
rZk2 2j =2
with wavefunctions
S c (s, m2 | )
(s, m) =
,
S c (s, m2 |0)
(3.14)
S d (j, r| )
C (j, r) =
,
S d (j, r|0)
(3.15)
directly expressible in terms of the S-transformation matrix elements of the representation

labeled by . In total, this ansatz yields three classes of consistent branes corresponding to
the continuous, discrete, and identity representations. For more details we refer the reader
to [37].
3.2.1. Ishibashi state decompositions: N = 2 A-type coset B-type
Now let us examine how the above N = 2 Ishibashi states can be decomposed into
appropriate coset Ishibashi states. The A-type N = 2 Ishibashi states have equal left- and
right-moving U (1)R charges. This implies that the left- and right-moving coset charges are
also equal, i.e., m = m
(see, e.g., Eqs. (2.13)(2.15)).6 Given the relation of those charges
with the left- and right-moving momenta along the compact direction of the cigar (2.44),
we conclude that such boundary conditions impose Neumann boundary conditions on .
By convention we call the corresponding coset Ishibashi states B-type. On the other hand,
coset Ishibashi states with Dirichlet boundary conditions on will be called A-type. From
now on we will concentrate mostly on boundary conditions or Ishibashi states of the coset
theory and the A/B-type terminology will refer to the above conventions.
A more precise relation between extended N = 2 and extended coset Ishibashi states
follows from the character decomposition formulae (2.39) and (2.41). In the case of N = 2
6 We should point out that the whole discussion is algebraic. The N = 2 theory has two realizations depending
= 2m identification to the SL(2)/U (1) charges. The plus sign corresponds to
on the right moving charge Q
k2
the N = 2 Liouville theory and the minus sign to its supercoset cigar mirror. Nevertheless the discussion in this
paper is totally independent of the choice of the N = 2 theory. We therefore choose the plus sign.
328
A-type boundary conditions, we may write7

A; s, m , m
A; m + n(k 2), m + n(k 2)
=

U (1)
2 2 N =2
nZk

m
m

B; s, n + , n +
2
2
(3.16)
for the continuous representations, and

A; 2(j + r) + n(k 2), 2(j + r) + n(k 2)
|A; j, r, rN =2 =
U (1)
nZk
|B; j, n + r, n + r
(3.17)
for the discrete representations. The boundary states appearing on the r.h.s. of these decom
positions are respectively U
(1) k(k2) Ishibashi states and extended coset Ishibashi states.
2
The overlaps of the latter will be discussed shortly. Notice that we label the Ishibashi
coset charges in the same way that we did for the corresponding characters in Section 2.
The actual (extended) coset charges appearing on the r.h.s. of the above equations are
(n + m2 , n + m2 ) for the continuous representations and (j + r n, j + r n) for the

discrete representations. The definition of the U
(1) k(k2) Ishibashi states can be found,
2
among other places, in a related context in [5].

The above decompositions result in a certain set of B-type coset Ishibashi states of
which only a subset is physical. The physical boundary states are allowed to couple only to
closed string modes that appear in the torus partition sum (2.50). This requirement yields
the following constraints on the charges:
m+m
= 0 mod 2k
(3.18)
for the continuous representations, and

2j + r + r = 0 mod k
(3.19)
m m
2 , 2
for the discrete. Therefore, we are led to conclude that an Ishibashi state |B; s,
is acceptable only when m = m
= 0 mod k with m Z2k . Similarly, an Ishibashi state
|B; j, r, r is acceptable only when 2(j + r) = 0 mod k and r = r . To summarize, we find
that the only allowed B-type Ishibashi states are:

B; s, k , k , s 0,
Continuous: |B; s, 0, 0,
(3.20)

2 2

B; j, k j, k j , 1 j k 1.
Discrete: |B; j, j, j ,
(3.21)

2
2
2
Notice that in the discrete B-type Ishibashi states, j takes only integer values. This is a
consequence of the relation 2(j + r) = 0 mod k and the fact that k has been assumed to be
even.
7 In these relations, we present explicitly the left- and right-moving charges of the Ishibashi states to make our
discussion as transparent as possible.
329
The only non-zero overlaps of these boundary states are the following:
B; s1 , 0, 0|eT H |B; s2 , 0, 0 = (s1 s2 ) 1 +is1 ,0 (iT ),
2

k k
k k
c
B; s1 , , eT H B; s2 , ,
= (s1 s2 )c1 +is , k (iT ),
1 2
2 2
2 2
2
c
(3.22)
(3.23)
B; j1 , j1 , j1 |eT H |B; j2 , j2 , j2 = j1 ,j2 dj1 ,j1 (iT ),

(3.24)

k
k
k
k
c
B; j1 , j1 , j1 eT H B; j2 , j2 , j2 = j1 ,j2 dj , k j (iT ).
1 2
1
2
2
2
2
(3.25)
c
3.2.2. Ishibashi state decompositions: N = 2 B-type coset A-type

The construction of discrete N = 2 B-type Ishibashi states and their decomposition into
coset Ishibashi states is slightly more tricky. The N = 2 B-type gluing requires Qj +r =
j +r .8 Together with the condition of conformal invariance a gluing of this sort can be
Q
satisfied only for j = k/4 and r = r 1, r 0. For these values, however, the discrete
N = 2 Ishibashi states decompose into coset Ishibashi states based on representations that
do not appear in the physical spectrum of the cigar. Indeed, the relevant coset charges are
m = j +r = k4 +r and m
= j + r +1 = k4 r, and they yield m+ m
= k2 . On the other hand,
the leftright coupling dictated by the torus partition sum (2.42) requires m + m
= wk
with w integer. This implies the full absence of the corresponding discrete Ishibashi states
on the cigar.
An analogous problem does not exist for the continuous N = 2 B-type Ishibashi states.
The N = 2 B-type gluing translates into A-type (Dirichlet) boundary conditions on the
direction of the cigar, i.e., m = m,
and the corresponding boundary states couple to closed
string modes with zero winding w. The relevant Ishibashi state decomposition reads

B; s, m , m
B; m + n(k 2), m n(k 2)
=

U (1)
2
2 N =2
nZk

m
m
A; s, n + , n
(3.26)
2
2
and the only acceptable A-type coset Ishibashi states are the following

m m
, s 0, m Z2k .
Continuous: A; s, ,
2
2
(3.27)
These boundary states satisfy the requirement (3.18) coming from the torus partition sum
automatically and their overlaps are given by

m
m m T H c
m
2k
c
A; s, , e
A; s , 2 , 2 = (s s )m,m 1/2+is, m2 (iT ). (3.28)
2
2
8 We use the same quantum number j for left- and right-moving modes. This choice is dictated by the physical
spectrum of the coset theory, where j = j.
330
3.3. Modular bootstrap and the Cardy ansatz for coset boundary states
3.3.1. A-type boundary states
The A-type Ishibashi states presented above are in one-to-one correspondence with the
continuous physical primaries of the coset theory and it might be useful to think of them
as a direct analogue of the N = 2 A-type Ishibashi states that appeared in (3.8). In order
to determine Cardy consistent linear combinations of these boundary states it would be
tempting to impose the direct analogue of the modular bootstrap equations (3.11)
A; 0|eT H |A; = (it),
c
A; 0|e
T H c
|A; 0 = I0 (it)
(3.29)
(T = 1/t)
(3.30)
but the basic element of this approach is now missing. Because of the absence of discrete
A-type Ishibashi states in the coset, we cannot construct a boundary state |A; 0 based on
the identity representation and we cannot implement the above equations directly.
Nevertheless, it is interesting to ask whether we can still formulate Cardy consistent
branes with the use of the Cardy ansatz appearing in Eq. (3.14). Such boundary states
would be the analogues of the A-type, class 2 boundary states of the N = 2 Liouville
analysis of [37] and they would take the form

m
m

m
A; s,
A;
s
=
,
ds
(s
,
m
)
,
s,m

2
2
2
0
(3.31)
m Z2k
with wavefunctions

s,m (s , m ) = e
i(s ,m )
S c (s , m2 |s, m2 )

.

S c (s , m2 |r = 0)
(3.32)
The precise form of the S-transformation matrix elements can be found in Section 2. Note

that we have explicitly allowed for an extra phase ei(s ,m ) , which may also depend on
the variables s and m . This extra phase will be fixed partially in a moment in a way
that accommodates the Cardy consistency conditions of the next section. More generally,
a phase ambiguity of this sort always exists in the modular bootstrap approach [37] and
extra information is needed to fix it.
Besides the Cardy inspired wavefunctions (3.32) we would like to consider also a slight
variant motivated by the analysis of the D1 boundary states of [41] (in Section 5 we provide
a detailed comparison of our boundary states with those of [41]). These boundary states
are not of the Cardy form, i.e., they cannot be written in terms of the S-matrix elements.
They have been derived in [41] from corresponding boundary states in the H3+ conformal
field theory by descent and they possess the expected semiclassical properties. It is useful
to consider them also in the context of the present discussion.
In the next section, we check the Cardy consistency of these two classes of boundary
states by studying all possible overlaps among them. In the process, we find that consisi

tency between the two classes fixes the phase factor in (3.32) to ei(m ) = e 2 (m mod 2) . To
summarize, we propose the following classes of A-type boundary states on SL(2, R)/U (1).
331
Class 2 |A; s, m2 , s 0, m Z2k :

| cosh (s + i m )|2
mm
2
4ss

i(m ) i k
2

s,m (s , m ) =
e
e
cos
.

)
k2
(k(k 2))1/4
sinh( 2s
)
sinh(2s
k2
(3.33)
Class 2 |A; s, m2 , s 0, m Z2k :
4iss

(s , m ) =
s,m
e k2 + (1)m e
2
i mm
k
e
2
(k(k 2))1/4

| cosh (s + i m )|2
2

.

)
sinh( 2s
)
sinh(2s
k2
4iss
k2
(3.34)
Notice that for even m the above wavefunctions are the same. For odd m , however, the

4ss

factor cos( 4ss
k2 ) in class 2 is replaced by sin( k2 ) in class 2 .
3.3.2. B-type boundary states
In a previous subsection we derived various B-type Ishibashi states. Some of them are
based on the continuous representations (3.20) and others on the discrete representations
(3.21). None of them, however, is in one-to-one correspondence with the respective physical primaries of the coset conformal field theory and for that reason it is not straightforward
to impose directly the modular bootstrap equations (3.29). Instead, we can directly apply
the Cardy ansatz (3.14), (3.15) and then check that all the Cardy consistency conditions
are satisfied.
There is a well-known analogue of this situation in the rational SU(2)/U (1) case of [5].
In the analysis of that paper the B-type SU(2)/U (1) boundary states were also constructed
with the use of the Cardy ansatz and the Cardy consistency conditions were checked. For
example, the self-overlap of the identity (j = 0) B-type boundary state was found to be
(see Eq. (3.21) in [5])
B; 0|eT H |B; 0 =
c
2k1
0,n (it),
(3.35)
n =0
where 0,n (q) are the appropriate SU(2)/U (1) characters. In the present SL(2, R)/U (1)
case we find a very similar result. The identity boundary state will be denoted as |B and
its self-overlap takes the form
B|eT H |B =
c

rZk
Ir (it).
(3.36)
332
More generally, we consider boundary states of the form

|B; =

ds (s, 0)|B; s, 0 +

k
ds (s, k)B; s,
2
0
k/21
D (j, j )|B; j, j +
j =1
k/21

j =1

k
k

D j, j B; j, j (3.37)
2
2
and wavefunctions of the Cardy type

(s , m ) =
S c (s , m2 | )

S c (s , m2 |r = 0)
D (j , r ) =
S d (j , r | )
S d (j , r |r = 0)
(3.38)
with m = 0, k for the continuous representations and r = j, k2 j for the discrete representations. is a label that parametrizes the possible B-type boundary states of the above
form.
In this way, we find the
following three candidate classes of B-type boundary states (an
extra normalization factor k is inserted for later consistency).
Class 1 |B:

sinh( 2s
k
k2 ) sinh(2s )

c

(s , 0) = k S (s , 0|r = 0) =
(3.39)
,
(k(k 2))1/4
| cosh(s )|2

sinh( 2s ) sinh(2s )

k
k
k2

c

, (3.40)
(s , k) = k S s , r = 0 =
2
(k(k 2))1/4
| cosh((s + i k ))|2
2

2k
(2j 1)
d
,
D(j, j ) = k S (j, j |r = 0) =
(3.41)
sin
k2
(k(k 2))1/4

k
k
2k
(2j 1)
d

D j, j = k S j, j r = 0 =
.
sin
1/4
2
2
k2
(k(k 2))
(3.42)
Class 2 |B; s, , s 0, = 0,
1
2:
| cosh(s )|2
4 k
4ss
s, (s , 0) =
cos
,

1/4

k2
(k(k 2))
sinh( 2s
k2 ) sinh(2s )

| cosh((s + i k ))|2
k
4
4ss

2i
2

e
cos
,
s, (s , k) =

k2
(k(k 2))1/4

(3.43)
(3.44)
k2
Ds, (j, j ) = 0,

k
Ds, j, j = 0.
2
(3.45)
(3.46)
Class 3 |B; j1 , 1 2j1 k 1:

1
cosh (s k4j
k
| cosh s |2
k2 )

j1 (s , 0) =
,

cosh s
(k(k 2))1/4 sinh( 2s
k2 ) sinh(2s )

| cosh((s + i k2 ))|2
k

j1 (s , k) =

(k(k 2))1/4 sinh( 2s
k2 ) sinh(2s )
333
(3.47)
k
1
e2i(j1 4 ) cosh (s k4j
k2 + i 2 )
,
cosh((s + i k2 ))
4i
1
1
1
i

Dj1 (j, j ) =
e k2 (j1 2 )(j 2 ) ,
2(k 2) sin (2j 1)
k2

4i
1
1
k
i
1

e2ij1 e k2 (j1 2 )(j 2 ) .
Dj1 j, j =
2
2(k 2) sin (2j 1)
k2
(3.48)
(3.49)
(3.50)
We would like to emphasize that we obtain only one class 1 brane with the above ansatz.
A priori, one might expect to obtain a larger set of class 1 boundary states |B; r (with
r Zk ) by using the more general matrix elements S c (s, 0|r), S c (s, k2 |r), S d (j, j |r),
S d (j, k2 j |r). It is straightforward to check, however, that such boundary states will be
identical to |B |B; r = 0. We find a similar result for the class 3 branes.
The class 2 branes fall into two categories depending on whether the parameter m in
S c (s , 0|s, m) and S c (s , k/2|s, m) is even or odd. It is straightforward to verify that the
functions S c (s , 0|s, m) are independent of m, whereas S c (s , k/2|s, m) depend on m only
through a phase e2ia with a = 0, 12 according to the parity of the parameter m.
Finally, notice that our B-type class 3 branes are parametrized by a single discrete label.
The reader familiar with [37] may recall that the authors of that paper proposed class 3
boundary states parametrized by two discrete parameters. The underlying reason for the
appearance of a second discrete label is the fact that the S-transformation property of
discrete representations includes the extra characters with j = 12 and j = k1
2 , which lie
outside the range of the allowed Ishibashi states. An appropriate linear combination of two
discrete boundary states allows for the cancellation of the extra characters. This cancellation turns out to be automatic for B-type boundary conditions in the SL(2, R)/U (1) coset
and there is no a priori reason to include a second discrete label. The possibility of a second discrete label will be discussed further in Section 4, where we determine which of the
above boundary states pass the Cardy consistency checks.
4. Cardy consistency conditions

In this section we compute several amplitudes between the above boundary states. We
are going to focus only on AA and BB overlaps. Mixed overlaps between A-type and
B-type branes involve a different set of characters, whose properties will not be discussed
here.
334
The modular transformation of these amplitudes from the closed string channel (parameter T ) to the open (parameter t = 1/T ) yields the explicit forms of the spectral densities
and degeneracies of the open strings stretching between the various branes. In general, we
find annulus amplitudes of the form
1 |eT H |2

= ds
(s, m; 1 |2 )c1 +is, m (it)
(c)
mZ2k
k1
N (j, r; 1 |2 )dj,r (it) +
2j =1 rZk
M(r; 1 |2 )Ir (it)
(4.1)
rZk
with positive definite real spectral densities (s, m|1 , 2 ) and positive integer multiplicities N (j, r|1 , 2 ) and M(r|1 , 2 ). The identity representation and the discrete representations with j = 12 , k1
2 do not appear in the closed string spectrum of the coset, but they
can appear in the open string channel of the above amplitudes and they have been included
in (4.1).
4.1. AA overlaps
class 2 class 2
For the overlap of two class 2 boundary states we find

m1 T H c
m2
A; s1 ,
e
A;
s
,
2

2
2

= ds 1 (s ; s1 |s2 )c1 m1 m2 (it) + 2 (s ; s1 |s2 )c1
2 +is ,

2 +is ,
m1 m2 +k
2

(it) (4.2)
with spectral densities

8
1 (s ; s1 |s2 ) =
k2

=
cosh(2s)
2s
sinh( k2
) sinh(2s)
2
k2
4ss2
4ss
4ss1
cos
cos
ds cos
k2
k2
k2

dp

1 ,2 =1
cos(2pp )
sinh(Qp) sinh( 2p
Q )

1
+ i1 p1 + i2 p2 ,
cosh 2p
Q
(4.3)
8
2 (s ; s1 |s2 ) =
k2

335
4ss2
4ss
4ss1
cos
cos
ds cos
k2
k2
k2
1
2s
sinh( k2
) sinh(2s)

2
cos(2pp )
=2
dp
cos 2p(p1 + p2 ) ,
2p
k2
sinh(Qp) sinh( Q ) =1
0
(4.4)
where Q2 =
identity
2
k2 .
We made use of the change of variables p = sQ and the trigonometric
cosh A cosh B cosh =
1
4
cosh(A + 1 B + 2 )
(4.5)
1 ,2 =1
to bring the spectral densities into a form that is already familiar from the analysis of the
boundary states in the N = 2 Liouville theory [37].
Let us briefly sketch the derivation of the above expressions. From the wavefunctions
given in (3.33) we can easily obtain the overlap

m1 T H c
m2
e
A; s1 ,
A; s2 , 2
2

4ss2
4ss1
4
cos
cos
= ds
k2
k2
k(k 2)
0
mZ2k
eim(m1 m2 )/k
| cosh (s + i m2 )|2
2s
sinh( k2
) sinh(2s)
c1 +is, m (iT ).
2
(4.6)
Then, with the use of the modular transformation property (2.54) and the trigonometric
identity

2

cosh s + i m = 1 cosh(2s) + 1 eim + eim
(4.7)

2
2
2
we can derive (4.2).
As we said, the spectral densities appearing here are similar to those of [37]. Strictly
speaking, these densities are divergent because of the diverging integrands near the point
p = 0. There are two ways to deal with this infrared divergence. The first, which will be
adopted here and was also used in [37], is to regularize the densities by explicitly subtracting the divergent piece. Another approach considers relative spectral densities, i.e., one is
instructed to subtract the amplitude of a reference boundary state with fixed parameters
s1 and s2 (e.g., s1 = s2 = 0). This method is based on the universal nature of the divergence and has been used previously in Refs. [17,41]. Further comments on this approach
will appear in Section 5, where we also show that the relative densities following from
1 (s ; s1 |s2 ) and 2 (s ; s1 |s2 ) above are equivalent to those appearing in [41].
336
The finite piece of the above densities can be written in terms of the so-called q-gamma
function Sb (x), which is defined as [9,37]

log Sb (x) =

dt sinh(t (b2 + 1 2bx)) b2 + 1 2bx
.
2t
sinh(b2 t) sinh(t)
b2 t
(4.8)
In our case the parameter b is given by b = Q =

2
1 (s ; s1 |s2 ) reads
1 (s ; s1 |s2 )|fin

i
=Q
2
=
i
2
1 .
k2
Hence, the finite part of

b
i

0 p log Sb
+ (0 p + 1 p1 + 2 p2 )
2
2
0 ,1 ,2 =1

b
+ ib(0 s + 1 s1 + 2 s2 ) ,
0 s log Sb
2
(4.9)
(4.10)
0 ,1 ,2 =1
while that of 2 (s ; s1 |s2 ) is

2 (s ; s1 |s2 )|fin
2i
=Q
2i

1
b
i

+ + (p + 1 p1 + 2 p2 )
p log Sb
2b 2
2
1 ,2 =1

1
b
s log Sb
+ + ib(s + 1 s1 + 2 s2 ) .
2b 2
(4.11)
(4.12)
1 ,2 =1
We can also write an explicit integral representation of these finite densities with the use
of the defining relation (4.8)
1 (s ; s1 |s2 )|fin
1
=
2i
s

1 ,2 =1
s

+ 1 s1 + 2 s2
,
t
2 (s ; s1 |s2 )|fin
1
=
2i

dt cosh(t) sin(2tb2 (s + 1 s1 + 2 s2 ))
2s i
t
2 sinh(b2 t) sinh(t)

1 ,2 =1
(4.13)

dt sin(2tb2 (s + 1 s1 + 2 s2 ))
4s i
t
2 sinh(b2 t) sinh(t)
0

+ 1 s1 + 2 s2
.
t
(4.14)
337
class 2 class 2
This overlap can be computed along the same lines as the previous one. We obtain

m1 T H c
m2
e
A;
s
A; s1 ,
,
2

2
2

= ds 3 (s ; s1 |s2 )c1 m1 m2 (it) + 4 (s ; s1 |s2 )c1 m1 m2 +k (it)
2 +is
2 +is
(4.15)

3 (s ; s1 |s2 )

4ss
8
4ss1
cos
=
ds cos
k2
k2
k2
0
4ss2
2
2
cosh2 (s) cos( 4ss
k2 ) + sinh (s) sin( k2 )
2s
sinh( k2
) sinh(2s)
(4.16)
(4.17)
4 (s ; s1 |s2 )

4ss
8
4ss1
cos
=
ds cos
k2
k2
k2
0
4ss2
2
2
cosh2 (s) cos( 4ss
k2 ) sinh (s) sin( k2 )
2s
sinh( k2
) sinh(2s)
class 2 class 2
Similarly we find

m1 T H c
m2
e
A;
s
A; s1 ,
,
2

2
2

= ds 5 (s ; s1 |s2 )c1 m1 m2 (it) + 6 (s ; s1 |s2 )c1
2 +is ,

2 +is ,
m1 m2 +k
2

(it)
(4.18)

5 (s ; s1 |s2 )

4ss2
1
cosh2 (s) cos( 4ss
8
4ss
k2 ) cos( k2 )
=
ds cos
2s
k2
k2
sinh( k2
) sinh(2s)
0
4ss2
1
sinh2 (s) sin( 4ss
k2 ) sin( k2 )
2s
sinh( k2
) sinh(2s)

,
(4.19)
338
6 (s ; s1 |s2 )
8
=
k2
4ss2
1
cosh2 (s) cos( 4ss
4ss
k2 ) cos( k2 )
ds cos
2s
k2
sinh( k2
) sinh(2s)
4ss2
1
sinh2 (s) sin( 4ss
k2 ) sin( k2 )
2s
sinh( k2
) sinh(2s)

.
(4.20)
These spectral densities can be written more compactly in terms of the previously defined
densities 1 and 2 as

1
1 (s ; s1 s2 |0) + 2 (s ; s1 + s2 |0) ,
2

1
6 (s ; s1 |s2 ) = 1 (s ; s1 + s2 |0) + 2 (s ; s1 s2 |0) .
2
5 (s ; s1 |s2 ) =
(4.21)
(4.22)
The Cardy consistency conditions demand that the spectral densities are positive functions of their arguments. For the densities appearing in this subsection this is indeed true
since the leading divergence in the defining integrals goes as s12 with a positive coefficient.
4.2. BB overlaps
class 1 class 1
There is a single B-type, class 1 boundary state. The self-overlap of this boundary state
in the open string channel can be computed easily and yields the following answer:
B|eT H |B =
c
Ir (it).
(4.23)
rZk
As we mentioned before, the form of this amplitude is very similar to that of the B-type
parafermion branes in [5] (see, e.g., (3.35)).
class 1 class 2
This overlap reads
B|eT H |B; s, =
c

mZ2k
(1 + ei(m+2) )c1 +is, m (it).

2
(4.24)
class 1 class 3
The overlaps between class 1 and class 3 boundary states are
B|eT H |B; j =
c

rZk
dj,r (it).
(4.25)
339
class 2 class 2
For the overlap of two class 2, B-branes we obtain
B; s1 , 1 |eT H |B; s2 , 2

=
ds 1 (s ; s1 |s2 ) + 2 (s ; s1 |s2 )
c
mZ2k 0

1 + ei(m+21 +22 ) c1 +is , m (it)
2
(4.26)
where 1 (s ; s1 |s2 ) and 2 (s ; s1 |s2 ) are the previously defined densities in Eqs. (4.3) and
(4.4), respectively.
class 3 class 3
We compute these overlaps in some detail. Starting from the defining relations (3.47)
(3.50), we obtain
B; j1 |eT H |B; j2

1
k 4j1
k
ds
cosh s
=
2s
k2
k(k 2)
sinh( k2
) sinh(2s)
0

k 4j2 c
1 +is,0 (iT ) + e2i(j2 j1 ) c1 +is, k (iT )
cosh s
k2
2
2
2
c
4i
1
2 1

e k2 (j 2 )(j2 j1 ) d
k
j,j (iT ) + e2i(j1 j2 ) dj, k j (iT ) .
+
2j
1
2
2 k(k 2) j =1
sin k2
(4.27)
In these relations j is an integer by construction.
The discrete character terms can be recast into a simpler form with the use of the character identity dj,r = dk
. The resulting expression in the closed string channel is
2 j,r
2
2 1

cos k2
(2j 1)(j2 j1 ) d
k
j,j (iT ).
1
k(k 2) j =1
sin 2j
k2
(4.28)
For simplicity let us consider in detail the consistency of the general self-overlap. For
j1 = j2 the modular transformation of dj,j (iT ) in (4.28) yields several contributions.
One of them involves the discrete characters dj ,r (it), which appear with multiplicities
(using the character identity once again)
k
2 1
sin k2
(2j 1)(2j 1)
2
.
k2
sin 2j 1
j =1
(4.29)
k2
Note that there are values of j in this expression for which the multiplicities are negative.
Hence, we are led to conclude that the B-type, class 3 boundary states (3.47)(3.50) are
inconsistent.
340
The existence of consistent A-type class 3 branes in N = 2 Liouville theory [37] suggests a possible alternative with boundary states labelled by a pair of discrete labels. Thus,
in analogy with the analysis of [37] we consider the class 3 branes.
Class 3 |B; j1 , j2 , with j1 , j2 = 1, 32 , . . . , k1
2 and

k
| cosh s|2
j1 ,j2 (s, 0) =
2s
(k(k 2))1/4 sinh( k2
) sinh(2s)
j1 ,j2
k
s,
2
k4j2
1
cosh (s k4j
k2 ) + cosh (s k2 )
cosh s

k
(k(k 2))1/4
| cosh s|2
2s
sinh( k2
) sinh(2s)
2ij2 cosh (s k4j2 )

1
e2ij1 cosh (s k4j
k2 ) + e
k2
i k
(4.30)
cosh s
(4.31)
1

Dj1 ,j2 (j, j ) =
1/4
(2j
1)
2(k(k 2))
sin
k2
4i
1
1
4i
1
1
e k2 (j1 2 )(j 2 ) + e k2 (j2 2 )(j 2 ) ,
i k
1
k

Dj1 ,j2 j, j =
1/4
2
(2j
1)
2(k(k 2))
sin k2

4i
1
1
4i
1
1
e2ij1 e k2 (j1 2 )(j 2 ) + e2ij2 e k2 (j2 2 )(j 2 ) .
(4.32)
(4.33)
These are simple linear superpositions of the class 3 branes (3.47)(3.50), i.e.,
|B; j1 , j2 = |B; j1 + |B; j2 .
(4.34)
The overlaps of these branes follow easily from the overlaps computed above. In particular,
c
for self-overlaps of the form B; j1 , j2 |eT H |B; j1 , j2 we simply have to sum over
c
T
H

|B; j with (j, j ) = (j1 , j1 ), (j1 , j2 ), (j2 , j1 ) and
four overlaps of the type B; j |e
(j2 , j2 ). The contribution of discrete characters (in the closed string channel) now becomes
k
2
2 1

(1 + cos k2
(2j 1)(j1 j2 )) d
2k
j,j (iT ).
1
k(k 2) j =1
sin 2j
k2
(4.35)
After a modular transformation to the open string channel and the use of the character
, we obtain the following multiplicity for the discrete character
identity dj,r = dk
2 j,r
dj ,r (it):
N (j , r ; j1 , j2 |j1 , j2 )
k
2 1
sin k2
(2j 1)(2j 1) sin2 ( k2
(2j 1)(j1 j2 ) 2 )
4
.
=
k2
sin 2j 1
j =1
k2
(4.36)
341
The last summation can be performed with the use of the Verlinde formula for the
SU(2)k0 affine algebra with k0 = k 4 0. Recall that the modular transformation matrix
for the SU(2)k0 affine characters s ( ) is

k0
1
2
s
s
r
=
sin
(2r + 1)(2s + 1) (4.37)
Sr s ( ), with Sr =
k0 + 2
k0 + 2
2r=0
and the Verlinde formula is the statement

k0

Ssr Str Sru
= Nstu
S0r
(4.38)
2r=0
with the fusion coefficients

1, s + t + u Z and |s t| u min{s + t, k0 s t},
u
Nst =
0, otherwise.
(4.39)
In each of these relations9 r, s, t, u = 0, 12 , . . . , k20 . In our case, however, the summation

variable r can take only integer values. The appropriate modification of the Verlinde formula reads (c.f. [5])
k0
2

Sr Sr Su
r=0
s t r
S0r
k0
1 u
u
Nst + Nst2
.
2
(4.40)
Now, in (4.40) set k0 = k 4 as well as 2r = 2j 2, 2s = 2j 2, and 2t = 2u =

(j1 j2 ) + k20 . With this substitution Eq. (4.36) takes the simple form
k2

2j =2
r Zk
N (j , r ; j1 , j2 |j1 , j2 )dj ,r (it)
k0
k0

u(j1 ,j2 )
d
2 u(j1 ,j2 )
Nsu(j1 ,j2 ) + Nsu(j
s+1,(s+1) (it),
,j
)
1 2
(4.41)
2s=0
where u(j1 , j2 ) =
j1 j2
2
k0
4.
Thus, we see that the only potentially consistent class 3

u(j ,j )
k0
u(j
1 ,j2 )
2
boundary states are those labeled by (j1 , j2 ) such that Nsu(j11 ,j22 ) = Nsu(j
1 ,j2 )
= 0 for
k0
2.
all values of s. This is the case only when u < 0 or u >

Equivalently, this requires
|j1 j2 | > k4
.
Taking
into
account
the
range
of
the
parameters
j1 and j2 , we conclude
2
that there are two possibilities
1
j1 = ,
2
j2 =
k1
,
2
or
j1 =
k1
,
2
1
j2 = ,
2
(4.42)
9 The present indices r and s should not be confused with the charge of the discrete characters or the Casimir
of the continuous ones in previous Sections.
342
which are in fact equivalent since |B; j1 , j2 |B; j2 , j1 . Note that for these values of
j1 and j2 the class 3 boundary states (3.47) simply reduce to the class 2 boundary states
(3.43) with s = 0 and = k4 mod 1. Hence, we are again led to conclude that the class 1
and class 2 are the only B-type boundary states of Section 3.3.2 that satisfy the Cardy
consistency conditions. This should not be interpreted as the statement that there are no
B-type boundary states based on the discrete representations. It may still be possible to
construct appropriate linear combinations of the above class 3 boundary states that satisfy
the Cardy consistency conditions. A similar case by case analysis was performed in the
N = 2 Liouville case in [37].
5. Physical interpretation
In this section we discuss the physical properties of the SL(2, R)/U (1) boundary states
that satisfy the Cardy consistency condition and their relation to the boundary states proposed in [41].
As we have already stressed in various places of the previous discussion, the modular
bootstrap approach can determine uniquely only the modulus of the one-point functions
but not the overall phase. Although we will not do this explicitly, for the couplings of the
boundary states to continuous closed string modes we can partially restrict this ambiguity
by employing the reflection property of the one-point functions as in [37,41]
(s, m, w) = R(s, m, w) (s, m, w),
(5.1)
where
R(s, m, w) = (b )2is
(1 + 2is)(2isb2 ) ( 12 +
(1 2is)(2isb2 ) ( 12 +
mkw
2
mkw
2
is)( 12 +
+ is)( 12
m+kw
2
m+kw
2
is)
+ is)
(5.2)
with
b2 =
1
,
k2
b =
(1 b2 )
.
(1 + b2 )
(5.3)
For discrete representations the analytic continuation of the reflection amplitude R(s, m, w)
exhibits a series of poles. These singularities can be attributed to the specific normalization
of the discrete primary fields. An alternative normalization would lead to the same amplitudes, since the divergent terms cancel in physical quantities [41]. Note that the approach
we followed for the construction of boundary states in this paper picks out a naturally
divergent-free normalization for the one-point functions of the discrete states.
For quick reference we begin with a short summary of the results of the previous sections.
5.1. Summary of Cardy consistent boundary states
The analysis of Sections 3 and 4 provided the following Cardy consistent boundary
states.
343
A-type, class 2 |A; s, m2 , s 0, m Z2k :

| cosh (s + i m )|2
mm
2
4ss

i(m ) i k
2

.
e
e
cos
s,m (s , m ) =

)
k2
(k(k 2))1/4
sinh( 2s
)
sinh(2s
k2
(5.4)
m

A-type, class 2 |A; s, 2 , s 0, m Z2k
4iss

s,m
(s , m ) =
e k2 + (1)m e
2
i mm
k
e
2
(k(k 2))1/4

| cosh (s + i m )|2
2

.

)
sinh( 2s
)
sinh(2s
k2
4iss
k2
(5.5)
B-type, class 1 |B:

sinh( 2s
k
k2 ) sinh(2s )
,
(k(k 2))1/4
| cosh(s )|2

sinh( 2s
k
k
k2 ) sinh(2s )

,
(s , k) = k S c s , r = 0 =
2
(k(k 2))1/4
| cosh(s )|2

2k
(2j 1)
D(j, j ) = k S d (j, j |r = 0) =
,
sin
1/4
k2
(k(k 2))

k
k
2k
(2j 1)
d
D j, j = k S j, j r = 0 =
.
sin
1/4
2
2
k2
(k(k 2))

(s , 0) = k S c (s , 0|r = 0) =
B-type, class 2 |B; s, , s 0, = 0, 12 :
| cosh(s )|2
4 k
4ss

,
cos
s, (s , 0) =

1/4

k2
(k(k 2))
sinh( 2s
k2 ) sinh(2s )
| cosh(s )|2
4 k
4ss

2i
e
cos
,
s, (s , k) =

k2
(k(k 2))1/4
(5.6)
(5.7)
(5.8)
(5.9)
(5.10)
(5.11)
k2
Ds, (j, j ) = 0,

k
Ds, j, j = 0.
2
(5.12)
(5.13)
5.2. D-branes on SL(2, R)/U (1) from AdS3

The geometry of D-branes in the SL(2, R) WZW model, which describes strings propagating in AdS3 , has been studied semiclassically in [65]. In that paper it was found that
solutions of the DBI action correspond to regular and twined conjugacy classes of SL(2, R).
More specifically, one finds D1-branes with AdS2 worldvolumes and two types of D2branes with either H2 or dS2 worldvolumes. The first type of D2-branes has a D-instanton
344
Table 1
SL(2, R)
Cigar
Embedding equations
Moduli
H2
D2
C = sin 2
k N, N N
dS2
D2
AdS2
?
D1
D0
2 F = k C tanh
cosh2 C 2
2 F = k C tanh
cosh2 C 2
sinh
sin( 0 ) = sinhmin
=0
C = cosh min 1, A
min 0, 0 [0, 2 )
none
density on its worldvolume and the second is unphysical due to a supercritical worldvolume electric field.
D-branes in coset models G/H have worldvolumes localized on the projection of the
product of a conjugacy class of G with a conjugacy class of H onto the coset [42]. In our
case of interest, namely SL(2, R)/U (1), we expect that the cigar D-branes will be projections on a constant plane of the SL(2, R) D-branes. Here denotes the time coordinate of
AdS3 that is being gauged. The semiclassical aspects of cigar D-branes have been analyzed
in [40,41].10 Here we review some of their features for comparison with our D-branes.
For D1-branes on the cigar the worldvolume can be written as a function ( ) and
we need to minimize the energy of the system E = L
L. Here the dot denotes a
derivative with respect to and L is the DBI Lagrangian. For the D2-branes one has to
solve the equations of motion for the gauge field F . The analysis is straightforward and
is summarized in Table 1. There we give the embedding equations for the various branes
expected semiclassically as well as the SL(2, R) branes from which they descend. k is the
level, N is the quantized D0-brane charge on the D2-branes, min is the minimum distance
from the tip, and 0 , 0 + are the points where the two branches of the D1-branes reach
the asymptotic circle. Notice that by Dp-brane here we mean a p-dimensional object in
the cigar geometry. In Fig. 1 we depict the geometry of these branes on the cigar as well as
the SL(2, R) branes from which they descent.
We see that the cigar D1-branes, which descend from AdS2 branes in AdS3 , have two
continuous classical moduli, 0 and min . The DBI analysis for the D2-branes that cover the
whole cigar implies that they carry a D0-brane charge, which leads to the quantization of
the modulus C. Since they have trivial topology, they do not admit a non-trivial Wilson line.
Finally, for the D2-branes that descend from dS2 branes we expect a continuous modulus
and the possibility to have a Wilson line due to the non-trivial topology. Notice that in
this semiclassical analysis we have not been able to identify the SL(2, R) brane that yields
the D0-branes on the cigar. On the other hand, the analysis of [41] has a natural, although
geometrically obscure, candidate: the spherical branes of [17] with imaginary radius. These
branes preserve the rotational symmetry of the cigar.
The boundary state analysis of the previous sections yields a similar set of branes. The
class 1 branes have an open string spectrum that includes only the identity module and they
have Dirichlet boundary conditions along . The class 2 branes have Neumann boundary
10 See also [66] for a recent analysis of the supersymmetric coset in a similar spirit.
345
Fig. 1. D-branes on the cigar and corresponding AdS3 branes.
conditions in the radial direction. Now, we proceed to analyze in detail each of these boundary states.
5.3. A-type, class 2 and class 2
These boundary states describe D1-branes extending along the radial direction of the
cigar. They have Dirichlet boundary conditions in the coordinate and the simple pole of
their wavefunctions at s = 0 suggests Neumann boundary conditions on . This is in line
with the interpretation of the N = 2 A-type, class 2 boundary states of [37] as analogues
of the FZZT Liouville branes that extend along the Liouville direction [9,10].
As expected semiclassically, the type-A, class 2 and class 2 branes possess two
branches. In our exact conformal field theory analysis, this is manifest from the computation of the open string spectrum (4.2), (4.18) with s1 = s2 and m1 = m2 , i.e., for a single
D1-brane. This spectrum includes open string states with either integer or half-integer
winding w. Indeed, these are precisely the spectra encoded in the extended continuous
characters c1
and c1
k in the annulus amplitudes (4.2), (4.18).
2 +is,0
2 +is, 2
The class 2 boundary states are by construction the ones appearing in [41]. The class 2
are closely related. In order to make the comparison with the result of [41] more transparent, we can rewrite the denominator in our class 2 (2 ) boundary states in terms of
functions. With the use of (4.7) and the following identities
(1 + x + iy)(1 x + iy)(1 + x iy)(1 x iy)
2 2 (x 2 + y 2 )
,
cosh(2y) cos(2x)
x
sinh(x) =
,
(1 + ix)(1 ix)
=
346
we obtain (up to an unimportant phase factor) the class 2 wavefunctions in the form
s,m (s , m ) =
2is

( k2
)(1 + 2is)
2
i(m ) i mm
k
e
e
3
1/4
1
m
(k(k 2) )
( 2 + 2 + is)( 12 m2 + is)

4ss
.
cos
k2
(5.14)
The corresponding wavefunctions of [41], which motivated our class 2 boundary states,
read

2is
(1 + 2is)( k2
)
j
(1 b2 ) is
ein0
n (r, ) =
1
n
1
0
(k(k 2)3 )1/4 (1 + b2 )
( 2 + is + 2 )( 2 + is n2 )
where b2 =
1
k2
(5.15)
and j = 12 + is. Using the identifications
m
,
k
s
r = 2
,
k2
0 =
e2irs + (1)n e+2irs

,
2
(5.16)
(5.17)
and setting n = m + 2kl, l Z, m Z2k , we obtain full agreement between our class 2
boundary states and (5.15) and partial agreement between the class 2 expression (5.14) and
(5.15). Incidentally, notice that the cylinder amplitude (4.17) in [41] can be recast in terms
of the extended characters for integer levels k, therefore pointing towards the extended
Ishibashi states we have been using throughout this paper.
The basic differences between our class 2 D1-branes and those of [41] are the following.
First, recall that the latter branes reach the asymptotic circle of the cigar at on two
opposite points 0 and 0 + and the minimum distance from the tip is parametrized by r =
min 0. In [41] both of these parameters are continuous. Matching them to the parameters
of our boundary state (see Eq. (5.16)) shows that our angle 0 is discrete, because m Z2k .
Hence, our D1-branes can reach the circle at infinity only at 2k special points. This is a
simple manifestation of the fact that our branes preserve a larger chiral symmetry than
the Virasoro, and it is precisely what one expects from a construction based on extended
characters.
The most important difference, however, concerns the coupling of each brane to the
closed string modes. The wavefunctions agree for even momentum n, but they disagree
for odd momentum n. This difference affects crucially the semiclassical properties of the
class 2 states. In order to make this point clear, we can compare the spectral densities
appearing on each class.
First of all, the densities obtained in Section 4 for the class 2 class 2 overlaps should
be in perfect agreement with those appearing in [41], if we choose the same regularization
scheme. This should be true by construction. Indeed, considering for simplicity the selfoverlaps s1 = s2 and m1 = m2 and subtracting the spectral densities of a reference brane
347
with s = 0, m = 0, yields the relative spectral densities

1
2 (s ; 2s1 |0) 2 (s ; 0|0)
2
for states with integer winding, and
5 (s ; s1 |s1 ) 5 (s ; 0|0) =
(5.18)

1
1 (s ; 2s1 |0) 1 (s ; 0|0)
(5.19)
2
for states with half-integer winding. It is now a simple exercise to check that under the identification (5.17) and the explicit expressions (4.13) and (4.14) the densities above match (up
to a factor of 4 expected from the different normalization of the wavefunctions) with the
corresponding relative spectral densities of [41] (relations (4.8), (4.9), (4.10), and (4.11) in
that reference). Of course, one needs to perform a summation over l in the result of [41] in
order to rewrite it in terms of extended characters.
As has been discussed in [41], the above densities possess the expected semiclassical behavior, i.e., 5 (s ; s1 |s1 ) 5 (s ; 0|0) is finite in the limit k , whereas
6 (s ; s1 |s1 ) 6 (s ; 0|0) is not well defined, because it receives contributions from purely
stringy states that decouple in the semiclassical limit. The situation is reversed, however,
when we consider the class 2 boundary states. In particular, the regularized density for zero
winding (field theory) modes 1 (s ; s1 |s1 ) 1 (s ; 0|0) is ill-defined in the large k limit,
implying that the semiclassical interpretation of these boundary states is problematic. This
pathology could be an indication that the class 2 branes are unphysical.
In order to elucidate further the physical properties of the class 2 branes, we can compute the semiclassical limit of the one-point functions in the coordinate system introduced
in [17] for H3+ . First, we write the metric of H3+ as

ds 2 = d 2 + cosh2 e2 d 2 + d 2
(5.20)
6 (s ; s1 |s1 ) 6 (s ; 0|0) =
= d 2 + e2 d d ,
(5.21)
with an angular coordinate and , [0, ) radial coordinates. In this coordinate system the AdS2 branes are given by surfaces of constant = r. They descend, as argued in
[17,41], to D1-branes on the cigar. The one-point function of the resulting branes, ignoring
factors unimportant to the present calculation, is conveniently written from (5.15) as

j

j
n (r, ) 1 + b2 (2j + 1) dn n0 cosh r(2j + 1) n1 sinh r(2j + 1)
(5.22)
0
with the definitions

(2j + 1)
,
(1 + j + n2 )(1 + j n2 )

1 , n: even,
n =
,
n: odd.
j
dn =
(5.23)
Now we can Fourier transform between the j, n, w and the j, u, u coordinates using the
following formulae [17]:

j
Fn,w = d 2 u ein arg(u) |u|2j +kw f (u),
(5.24)
C
348
f (u) = ik(2)2
dw ein arg(u) |u|2j 2kw Fn,w .

j
(5.25)
nZ R
One can also show the identity

j
2
d 2 u ein arg(u) |u|2j 2kw |u + u|
2j sign (u + u),
dn n (w) = (2)
(5.26)
which will be used in a moment.

The integral of the Laplace operator eigenfunctions j (u|, , ) [17] over the semiclassical one-point functions gives the geometrical shape of the corresponding D-brane.
For the AdS2 branes and by using (5.26), this computation yields

2j
dj d 2 u j (u|, , ) |u + u|

cosh r(2j + 1) sign(u + u)
sinh r(2j + 1) ( r),
(5.27)
therefore describing a D1-brane as in Table 1.
One can determine the shape of the class 2 boundary states in the semiclassical limit
in a similar way. First we need to write down the corresponding wavefunctions in a form
similar to (5.22):

j

j
n (r, ) 1 + b2 (2j + 1) dn n0 cosh r(2j + 1) in1 cosh r(2j + 1) . (5.28)
0
Integrating them as above with the use of the Eqs. (5.24) and (5.26) gives

2j
dj d 2 u j (u|, , ) |u + u|

cosh r(2j + 1) i sign(u + u)
cosh r(2j + 1)

dP sin 2P ( + r) + sin 2P ( r) + ( r) + ( + r).
(5.29)
The integral in the second line comes from the term involving sign(u + u)
and it can be
traced back to the contribution of odd momenta to n1 . This integral is not well defined and
does not lead to a sensible geometrical interpretation of the class 2 D1-branes.
A possible way out is to consider a linear combination of the A-type class 2 wavefuncj
j
j
tion (3.33) with its complex conjugate, in particular (n )even
(r,0 ) = (n )(r,0 ) + (n )(r,0 ) .
In this case, the only non-zero couplings to closed strings are the ones with even momentum n. As a consequence, we only get the function terms in (5.29). This boundary state
describes two D1-branes with parameters (r, 0 ) and (r, 0 ). This, however, is exactly the
same boundary state that one would obtain from a linear superposition of two D1-branes
of [41]. Therefore, it seems that employing the Cardy ansatz does not lead to D1-branes
with nice semiclassical features and we are probably left only with the class 2 branes of
[41].
Finally, we can check if the class 2 states satisfy the H3 factorization constraints derived
in [17]. It was shown there that the allowed one-point functions for the AdS2 branes (and
correspondingly for their descendants on the cigar), have the following form

i

j+ 1
j u
= |u + u|
2j b 2 1 + b2 (2j + 1) E (j |r)
2 r
349
(5.30)
The function E (j |r) is restricted

with b a normalization constant and = sign(u + u).
by factorization constrains to have the form

E (j |r) = e0 (j |r) + e1 (j |r) er(2j +1)
(5.31)
with e (1j |r) = (1) e (j |r). The general A-type, class 2 boundary state with Fourier
transformed wavefunction (5.28) is not of the above form and hence does not solve the H3
factorization constrains. It would be an important but challenging task to determine the
consistency of these boundary states by analyzing the factorization constraints directly in
the cigar conformal field theory.
5.4. B-type, class 1
We find a single B-type, class 1 brane. This boundary state has Neumann boundary
conditions on the angular direction of the cigar (it is B-type) and is localized in the radial
direction (it is class 1). A circular D1-brane wrapped around the cigar will slip towards the
tip to lower its energy. Therefore, our B-type, class 1 brane will eventually be localized
at the tip of the cigar, where it should rather be interpreted as a D0-brane. Perhaps it is
useful to think of this brane as the analog of the SU(2)/U (1) B-type brane with j = 0
that sits at the center of the SU(2)/U (1) disk [5]. However, unlike those branes, which
are unstable and can move away from the center of the disk, the D0-brane of the cigar is
stable and cannot move from the tip. Analogous branes were formulated for the cigar CFT
in [41]. These branes were parametrized by a single discrete parameter m = 1, 2, . . . , but
as explained in that paper, only the m = 1 brane appeared to be physical. A comparison
of the properties of our B-type, class 1 brane with those of the m = 1 brane of [41] below
indicates that these are the same branes. Notice that we have restricted our analysis only to
unitary representations. As a result, our approach does not yield the m > 1 boundary states
of [41], which carry non-unitary open string spectra.
The B-type, class 1 boundary state couples to both continuous and discrete closed string
modes. The appropriate wavefunctions are listed in Eqs. (5.6)(5.9) and they are in agreement with the expressions appearing in [41]. For the continuous part the wavefunctions
read
(j, w)D0
m=1
1
kw

1/4 (1 b2 ) j + 2 (j + kw
2 )(j 2 )
,
= () k(k 2)
(1 + b2 )
(b2 (2j + 1))(2j )
(5.32)
w
1
where b2 = k2
and j = 12 + is in the notation of [41]. Indeed, up to a phase factor
that does not affect the analysis of the cylinder amplitudes, we find that these wavefunctions depend on w only through its parity (i.e., on whether w is even or odd). With a few
straightforward algebraic manipulations one can show that they can be brought into the
350
form

1
ak

1/4 (1 b2 ) is ( 12 + ak
2 is)( 2 2 is)
(s, a) = k(k 2)
(1 + b2 )
(b2 2is)(1 2is)
(5.33)
with a = 0, 1. Similarly, for the discrete representations one obtains the wavefunctions

1/4

k
ak
D j,
sin b2 (2j + 1) .
j = 2
2
k2
(5.34)
The above wavefunctions yield a cylinder amplitude B|eT H |B, which agrees with
the result following from Eqs. (5.6)(5.9). There is a small difference in the contribution
of the discrete representations, which we now flesh out.
The discrete part of the above cylinder amplitude involves a summation over the quantum numbers j and w. In [41] (see Eq. (3.18)) this summation runs over the restricted set
d := [ 1 , k1 ] ( 1 |kw| N). Our result involves a similar summation over j and w,
J0,w
2
2
2
d should be replaced by the unrestricted interval ( 1 , k1 ). Notice, however, that this
but J0,w
2
2
difference affects only the w = 0 contribution. Indeed, we are dealing with even levels k, so
d
can set k = 2, where is some positive integer. Then, the condition that j belongs in J0,w
requires the existence of a positive integer n, such that j = n + 1/2|kw| = n + |w|
and such that it satisfies the inequality 1/2 + (|w| 1) n |w| 1/2. n will be necessarily positive for any k > 2. It should be clear that for each w Z we can always find
an appropriate integer n in the range defined above and for such ns we get the same values
of j for every possible non-vanishing value of w. The value w = 0 is the only exception.
It is therefore excluded in the computation of [41], but it is included in our computation.
Unfortunately, we cannot offer an explanation for this discrepancy.
Despite this minor difference, the cylinder amplitudes agree perfectly in the open string
channel. The result appearing in (4.23) can be recast in terms of the standard dj,r discrete
characters, using the identity (C.11). One finds
c

rZk
Ir ( ) =

ccs
1
1
d0,l ( ) d1,l1 ( ) = q 24 1 + 2q 1+ k + q 2 + O q 2+ k .
lZ
(5.35)
In this expression d0,n denotes the extrapolation of the discrete character (2.5) to j = 0 and
not the character (2.4) of the identity representation. This is precisely the result found in
Eq. (3.6) of [41].
As we see, the open string spectrum of the B-type, class 1 brane is discrete and there is
no open string winding. This is a simple manifestation of the fact that the B-type, class 1
brane is rotationally invariant. For finite level k the one-loop result in (5.35) suggests that
the open string spectrum does not contain any marginal modes. This is consistent with the
semiclassical analysis, which fixes the position of the D0-brane at the tip of the cigar. In
the flat space limit, however, two states with scaling dimension h = 1 appear signaling the
free motion of a D0-brane in a flat 2-dimensional space.
351

These branes are new and have not been considered before in the boundary conformal
field theory context. They are labeled by a continuous parameter s [0, ) and they have
Neumann boundary conditions both in the angular and radial directions. Hence, their shape
is that of 2-dimensional D-branes. Since they do not couple to any discrete modes, which
are localized near the tip of the cigar, it is natural to expect that in general these branes
do not reach the tip and hence they are candidates for the D2-branes descending from dS2
branes of AdS3 in Table 1. Such branes develop a circular opening at a distance min away
from the tip and terminate there. The parameter min is directly related to the continuous
parameter s. For example, we can easily verify that both parameters scale to infinity when
we send k . For finite k notice that when we send s 0, the circular opening of the
D2-brane shrinks to zero radius and the brane covers the whole cigar.
According to the semiclassical analysis of Section 5.2 these D2-branes can have a
non-vanishing Z2 Wilson line because of the non-trivial cycle on their worldvolume. The
anticipation of a Z2 -valued, instead of R-valued, Wilson line will be better clarified in a
moment with the use of a T-duality argument. The possibility of the Wilson line is realized
explicitly by our B-type, class 2 branes. The one-point functions in (3.43) depend on the
extra Z2 -valued parameter = 0, 12 . The Cardy computation in (4.26) shows clearly that
the overlap of two branes with the same values 1 = 2 results in open string states with
even index m. On the other hand, the same computation for branes with different values of
1 and 2 involves open string states with odd m. The explicit form of the extended character c1
m in (C.4) reveals that even m indices correspond to integer momentum modes in
2 +is, 2
the angular direction, whereas odd m indices correspond to half-integer momentum modes.
Therefore, this open string spectrum suggests that parameterizes a Z2 Wilson line. It is
known (see, e.g., [67]) that if we have a circular D1 string, the presence of a Z2 Wilson
line results, under a 2 translation around the S 1 , into a minus sign in the wavefunction of
an open string ending on the D1-string. In our case, that would imply half-integer angular
momenta for open strings stretching between an = 0 and an = 12 brane. As a consequence, in the field theory limit one would expect, for example, a tachyon field of the form
[67]

1
Tn ()ei(n+ 2 ) .
T (, ) =
(5.36)
nZ
Indeed, this is what we observe in the open string spectra of our branes. Eq. (C.4) with
s = 0 and m = 2n + 1 (n Z) shows that the open string spectrum includes l = 0 modes
(n+ 1 )2
with scaling dimensions h = k2 . In the semiclassical limit of large k these are precisely
the scaling
dimensions of tachyon modes with half-integer angular momentum on a circle
of radius k.
The Z2 Wilson line can also be derived by the following T-duality argument. As we
briefly mentioned in Section 2, the trumpet geometry can be obtained from the cigar by
T-duality. This T-dual background can also be viewed as a Zk orbifold of the cigar. This
is most apparent in the asymptotic cylinder region, where the Zk orbifold reduces the
asymptotic radius Rcig to Rcig /k. To avoid the subtleties due to the singularity in the
352
trumpet geometry we can restrict our discussion only on the asymptotic cylinder region. On
the trumpet we can formulate D1-branes, which are the analogs of the A-type, class 2 (2 )
branes on the cigar (see [40]). Again, these branes will be parameterized by a continuous
parameter s and geometrically will correspond to D1-branes starting from the asymptotic
infinity and returning back to it. They will be perturbative and stay far from the trumpet
singularity for large parameters s. The boundary states of these D1-branes can be deduced
easily from the analysis of the A-type, class 2 (2 ) branes on the cigar. They are simply
Zk orbifold projections of those branes. In other words, they can be obtained by summing
over k images of the A-type, class 2 (2 ) boundary states.11 To be definite, let us consider
only the A-type, class 2 cigar branes, which have better semiclassical behavior.12 There
are two possible sums over k images that we can write and one can easily show that
1
1
|A; s, l = |B; s, = 0,
|A; s, 1trumpet
2
k lZ
k

1
1
1
1
A; s, l +
|A; s, 2trumpet
= B; s, = .
2
2
2
k lZ
(5.37)
(5.38)
This result is precisely what we expect from T-duality. The A-type trumpet D1-branes
were expected to yield B-type D2-branes on the cigar and this is exactly what we find. The
Wilson line is naturally Z2 -valued, because there are only 2k D1-branes on the cigar and
the Zk orbifold leaves behind a remaining Z2 degree of freedom. Hence, the fact that we
do not have an R-valued Wilson line, as would be expected on geometric grounds, can be
traced back to the use of extended Ishibashi states that preserve more symmetry than just
the Virasoro.
As a final comment on the B-type class 2 branes notice that the selfoverlaps in (4.26)
yield an open string spectrum that has a marginal deformation for s = k 2 and m = 0.
This indicates that our D2-branes are marginally stable to a change of the s modulus. Such
a marginal mode is also expected from the semiclassical analysis. The DBI action has a flat
direction of solutions parameterized by min . Moreover, one can compute the radial stress
energy tensor component T :
T =

cosh (F )2
L
F L = TD2
cosh tanh2 + (F )2
F
tanh2 + (F )2
(5.39)
and show, using the equations of motion from Table 1, that it is zero (and therefore continuous) at the minimum distance from the tip min . This provides further confirmation for
the existence of the above marginal deformation.
11 A similar argument was employed in [5] to uncover B-type branes in the parafermion model.
12 It turns out that for the purposes of the present argument we could also use the class 2 branes and obtain the
same result.
353

The analysis of Section 4 revealed that the class 3 (and most of the class 3 ) boundary
states have negative open string multiplicities and hence do not satisfy the Cardy consistency conditions. Nevertheless, there are some striking similarities between the class 3
boundary states of this paper and the cigar covering D2-branes of [41] and it is instructive
to compare the two results. For those branes we use the form of the closed string couplings
that appear in [43]13

kw
(j + kw
1
b2
i
i
2 )(j 2 )
= (b )j + 2
wj
,
2 2
4
(2j 1)(1 (2j 1)b2 )
i (2j 1) sin (j kw )
ei (2j 1) sin (j + kw
2 )+e
2
sin (2j 1) sin b2 (2j 1)
2
kw b
1
= 2(1) 2
(b )j + 2
4
kw
(j + kw
2 )(j 2 ) sin j cos (2j 1)
.
(2j 1)(1 (2j 1)b2 ) sin (2j 1) sin b2 (2j 1)

(5.40)
For 2j1 = (k 2) + k2 , and 2j2 = (k 2) + k2 , the class 3 boundary state wavefunctions (3.47)(3.50) become
1
4
k
cosh 2 s
,
j1 ,j2 (s, 0) = 2
(5.41)
k2
sinh 2sb2 sinh 2s
1
4
k
cosh 2 s
k
,
ei 2
j1 ,j2 (s, k) = 2
(5.42)
k2
sinh 2sb2 sinh 2s
1
4
i
k
1
i

e 2 (2j 1) cos (2j 1), (5.43)
Dj1 ,j2 (j, j ) = 2
2
k
2
2
sin b (2j 1)
4
i
k
i
1
k

e 2 (2j 1)
Dj1 ,j2 j, j = 2
2
2 k2
sin b2 (2j 1)
k
ei 2 cos (2j 1).
(5.44)
m
Notice that since 1 2ji k 1 we can easily obtain the quantization condition = k2
with the bound [0, 2 ).

It is not difficult to compare the above couplings with those of (5.40). One finds that the
continuous couplings are the same except for several phase factors that do not affect the
cylinder amplitudes. The discrete couplings, on the other hand, appear to be different. This
should be expected. From a similar comparison involving the B-type class 1 branes [41],
we know that for discrete representations properly normalized one-point functions should
13 Again, we specialize in the case of even k, but similar manipulations should also go through for odd level k
with minor modifications.
354
be considered. More precisely, in computing the cylinder amplitudes using (5.40) one has
to consider discrete contributions of the form:

(j, w) (j, w) d
j, kw (iT ),
Res
(5.45)
j
j
2
0,w 0,w
w,j
j
where one uses the bulk two-point function of the operator 0,w to normalize the cylinder
contributions for each closed string mode. It should be obvious from the last equation in
(5.40) that the dependence on is the same as that in (5.41). Moreover, the functions in
the numerator of (5.40) exhibit discrete poles for j kw
2 = n, n N. For even levels k
this yields contributions from all the integer values of j [ 12 , k1
2 ], precisely as expected
from the cylinder amplitude computation of Section 4.
Despite the apparent similarity of the above one-point functions, the expression (4.36)
for the self-overlap differs from the annulus amplitude computed in [41,43]. In particular,
these papers find that the contribution of discrete open string states vanishes in the case
of self-overlaps. This result is in contrast to (4.36) except for = 2 , in which case our
class 3 B-type boundary states reduce to class 2 boundary states. This vanishing, combined
with the negative open string multiplicities of the remaining cross-overlaps, led the authors
of [43] to conclude that only a single D2-brane should be allowed.
We should point out, however, that the analysis of [41,43] is, strictly speaking, valid
only for non-rational levels k. For rational levels there are additional subleties involving
contributions due to extra poles, which were not taken into account in the discussion of
[41,43]. Hence, it is plausible that the results of those papers might not apply directly to
the case of integer levels k that we study in this work. If this is the case, then it would be
interesting to see if the extra pole contributions can explain the discrepancy between our
results and those appearing in [41,43].
6. Summary and conclusions

In this paper we formulated D-branes on the SL(2, R)k /U (1) coset conformal field
theory at integer level k. The basic building blocks of our analysis were the extended coset
characters of [39]. With the use of the corresponding extended Ishibashi states we proposed
Cardy-like boundary states and verified the Cardy consistency conditions. Although the
discussion we presented here applies directly only to integer levels k, the extension to
rational levels should be straightforward.
Our analysis yielded the following results. We found two basic types of boundary states:
A- and B-type. We presented two A-type boundary states, which we called class 2 and
class 2 . Both of them are based on the continuous representations and geometrically
correspond to D1-branes starting and terminating at the asymptotic infinity. The class 2
boundary states resulted from a direct application of the Cardy ansatz and exhibit some
puzzling semiclassical features. It is not completely clear whether these branes are fully
consistent. The class 2 boundary states are a slight variant of the Cardy ansatz and they
have appeared previously in [41], where they were derived from AdS2 boundary states in
the Euclidean AdS3 . These branes have the expected semiclassical behavior.
355
We also presented two classes of B-type boundary states. These were called class 1 and
class 2. We found a single B-type class 1 boundary state, which is based on the identity
representation and corresponds to a rotationally invariant D1-brane that wraps the circular
direction of the cigar and slips toward the tip in order to minimize its energy. In this way
it becomes effectively a pointlike object, i.e., a D0-brane. The class 2 branes are based
only on continuous representations and they are D2-branes (with a Z2 -valued Wilson line)
covering part of the cigar (see Fig. 1). An interesting question concerns the formulation
of D2-branes covering the whole cigar. These are expected to be based on the discrete
representations. We formulated such branes in Section 3 as class 3 boundary states, but
they were found in Section 4 to violate the Cardy consistency conditions and they had to
be excluded. Nevertheless, it would still be interesting to see if one can postulate Cardy
consistent class 3 boundary states as appropriate linear combinations of the class 3 boundary states of Section 3, or with wavefunctions that do not follow from the application of
the Cardy ansatz. The simple linear superposition that yielded the class 3 boundary states
produced a set of D2-branes that share many similarities with the cigar covering D2-branes
of Refs. [41,43]. Only one of them was found, however, to be Cardy consistent. A similar
observation was made in [43].
In conclusion we would like to mention a few open problems. The analysis we presented
was based exclusively on the study of the Cardy consistency conditions. Hence, it should
be supplemented by a detailed study of the disk 1-point function factorization constraints.
This would provide additional evidence for the results presented in this paper and would
clarify important aspects of the open string theory on the D-branes we considered. The
analysis of the 4-point function conformal blocks is expected to be the difficult part of
this task. Another related task would be the study of the boundary two- and three-point
functions of the open string theories presented here.
Moreover, it would be intriguing to study the implications of our results for twodimensional holography and the connection with the matrix model of [36] in the
spirit of [3234]. On a related context, one could also explore the possibility of deriving the connection to the matrix model through an auxiliary Riemann surface similar to the analysis that has been performed for the minimal (super)string theory
in [68].
Finally, it would be worthwhile to obtain boundary states for the supersymmetric orbifold
SL(2, R)k /U (1) SU(2)k /U (1)
Zk
(6.1)
by extending the techniques used here. This background appears in the exact CFT description of NS5-branes (put symmetrically on a topologically trivial circle) and provides a holographic dual of little string theory in the double scaling limit [30]. It is
therefore of obvious importance to understand the corresponding D-branes. Furthermore, analyzing the dynamics of these branes along the lines of [69] and in connection with HananyWitten type configurations [70,71] is another interesting open problem.
356
Note added
While this paper was being prepared for submission, two interesting preprints appeared
in the hep-th archive discussing related issues. In [43] the D-brane spectrum of the supersymmetric coset and its N = 2 Liouville mirror is analyzed by extending the methods of
[41]. In [44] conformal boundary conditions for the N = 2 Liouville theory are studied
using conformal and modular bootstrap methods.
Acknowledgements
We would like to thank D. Israel, E. Kiritsis, D. Kutasov, J. Troost, and S. Ribault
for valuable discussions. The work of A.F. is supported in part by the PPARC SPG grant
00613. The work of V.N. is supported in part by the DOE grant DE-FG02-90ER-40560.
The work of N.P. is supported by the Deutsche Forschungsgemeinschaft under the project
number DFG Lu 419/7-2 and in part by the European Communitys Human Potential Programme under contract HPRN-CT-2000-00131 Quantum Spacetime.
Appendix A. The coset partition sum in terms of standard and extended characters
The spectrum of string theory on the axially-gauged (cigar) coset involves only the
continuous and discrete representations. For a semiclassical discussion of this spectrum
we refer the reader to [25]. A more recent discussion, from the point of view of the torus
partition sum, can be found in [52]. In this appendix we review the basic features of that
computation and show explicitly how the torus partition sum can be recast in terms of the
standard and extended characters. The standard character form did not appear explicitly in
[52], but it was already implicit in the considerations of that paper. The supersymmetric
version of the partition sum expression in terms of extended characters appeared recently
in [53,54].
Our starting point is the torus partition sum (3.9) of [52]

1/2
Z(, ) = 2 k(k 2)
1
ds1 ds2
m,wZ
(q q)
2/24
| sin((s1 s2 ))|2
2 2
2
k
|(s1 +w) (s2 +m)| + ((s1 s2 ))
2

.
(1 e2ir 2i(s1 s2 ) )(1 e2ir +2i(s1 s2 ) )2
r=1
(A.1)
This result can be derived from a direct path integral computation in the corresponding
gauged WZW theory.
357
With the use of the Poisson resummation formula in m it is possible to recast (A.1) into
the form
1
Z(, ) = 8 (k 2)2 |( )|
2
0
ds1 ds2
2ins2
1
4k (nk(w+s1 ))
e22 s1
|1 (s1 s2 | )|2
1
q 4k (n+k(w+s1 )) ,
(A.2)

1 q m 1 e2iz q m 1 e2iz q m .
(A.3)
n,wZ
where
1 (z| ) = 2ei/4 sin z
m=1
We can expand this function in the denominator of the partition sum, with the help of the
formula (y = e2i(s1 s2 ) )
iy 1/2 r
1
= 3
y Sr ( ),
1 (s1 s2 | ) ( )
(A.4)
rZ
where
Sr ( ) =

1
()s q 2 s(s+2r+1) .
(A.5)
s=0
This expansion is valid only for |q| < |z| < 1, which is indeed the case for 0 < s1 < 1.
Plugging the expansion (A.4) into the partition sum (A.2) and integrating out s2 yields

1
Z(, ) = 8 (k 2)2
|( )|4
q
1
r kw)2
4k (r
ds1
e22 (
k2 2
r +kw))
2 s1 +s1 (1+r+
r,r ,wZ
1
r +kw)2
4k (r
Sr ( )Sr ( )
(A.6)
with the constraint n = r r . Following [52] we can introduce an auxiliary variable s to

linearize the s1 integral. This can be achieved with the use of the identity

k2 2
(k 2)2 e22 ( 2 s1 +s1 (1+r+r +kw))

2
s 2s1 (is+2 (1+r+r +kw))
=
ds e (k2)2
.
(A.7)
Then, we can easily perform the s1 integral to obtain the partition sum in the form
8
Z(, ) =
|( )|4
ds
q 4k (rr kw) q 4k (rr +kw) Sr ( )Sr ( )

1
r,r ,wZ
s2

e (k2)2
1 e2(is+2 (1+r+r +kw)) .
2(is + 2 (1 + r + r + kw))
(A.8)
358
In the above expression, the contribution of the discrete representations comes from the
second term. The exponent of the exponential inside the parenthesis can be completed to a
square by shifting s and by defining the new variable t = s + i(k 2)2 we can write the
second term as

1
1
4
2
2
dt
q 4k (rr kw) q 4k (rr +kw) Sr ( )Sr ( )
4
i|( )|
r,r ,wZ
(k2)
t 2 22 ( k +r+r +kw)
2
2
1
,
t i2 (k 1 + r + r + kw)
(A.9)
where C denotes the contour C = {t | t = s + i2 (k 2), s R}. We may close the contour
at to form an infinite parallelogram between the real line and C. Then, the integrand
has poles at the locations t = i2 (k 1 + r + r + kw) inside the parallelogram, i.e., it
has poles for those integers r, r and w for which 0 < k 1 + r + r + kw < k 2. The
contribution of these poles gives the contribution Zd of the discrete representations to the
torus partition sum. After setting 2j kw r r it is straightforward to show that
Zd (, ) = 8
1
k1
2 <j < 2
w,r,r Z
(2j + r + r + kw)dj,r ( )dj,r ( ).
(A.10)
Since we assume that k is an (even) integer we see that 2j Z. Notice that in principle
we can replace dj,r by dk
since they share the same conformal weights. However,
2 j,r
their respective J03 charges are different and choosing the second one would not lead to a
sensible interpretation of these charges as left- and right-moving momenta of a free compact boson, which is required by the BRST constraint of the corresponding gauged WZW
model. In general one can check that the discrete part cannot be written in terms of characters with j = j. This will be important when we discuss the allowed coset Ishibashi
states.
The contribution of the continuous representations can be obtained in a similar way. The
integral over the first term in (A.8) plus the integral along the real line that remains from
(A.9) give14
16
Zc (, ) =
|( )|4
s2
1
s2
1
2
2
ds
rr ,n q k2 + 4k (nkw) q k2 + 4k (n+kw)
2
r,r Z n,wZ
Sr ( )Sr ( )

1 k
(q q)
2 ( 2 +r+r +kw)
1
2is + r + r + kw + 1 2is + r + r + k(w + 1) 1
(A.11)
Compared to (A.8) and (A.9) we have rescaled the integration variable as s 22 s. This
expression can also be written as

14 The integral in (A.9) can be written as
C dt = ds 2 i Res with s R.
16
Zc (, ) =
|( )|4
ds
rr ,n Sr ( )Sr ( )
2
r,r Z n,wZ
s2
s2
q k2 + 4k (nkw) q k2 + 4k (n+kw)
2is + r + r + kw + 1
s2
359
s2
q k2 + 4k (nk(w+1)) +r + q k2 + 4k (n+k(w+1))
2is + r + r + k(w + 1) 1
1
2 +
r
(A.12)
Shifting w w 1 in the second term and using the relation q r Sr = Sr yields

16
Zc (, ) =
|( )|4
s2
1
s2
1
2
2
ds
rr ,n q k2 + 4k (nkw) q k2 + 4k (n+kw)
2
r,r Z n,wZ

Sr Sr
Sr Sr
.
2is + r + r + kw + 1 2is + r + r + kw 1
(A.13)
This result can be simplified further with the use of the identity Sr = 1 Sr1 and by
shifting r r + 1, r r + 1 in the second term. This allows us to bring the above expression into the form
16
Zc (, ) =
|( )|4
s2
ds
rr ,n
2
r,r Z n,wZ
s2
q k2 + 4k (nkw) q k2 + 4k (n+kw)
1
Sr + Sr 1
.
2is + r + r + kw + 1
(A.14)
In terms of the continuous characters appearing in (2.6) the previous expression can be
written as
Zc (, ) =
16
2
ds
n,wZ
Q(s; n, w|, )c1 +is, nkw ( )c1 +is, n+kw ( ),

2
(A.15)
where Q(s; n, w|, ) is a function given by the formula

Q(s; n, w|, ) =
rr ,n
r,r Z

rZ
Sr + Sr 1
2is + r + r + kw + 1

Sr 12
Sr 12
+
.
2is + 2r + kw + 1 n
2is + 2r + kw + 1 + n
r Z
(A.16)
360
Let us focus on the first term of this function. We have

rZ
Sr 12
2is + 2r + kw + 1 n
=
=
=
1

Sr 12
Sr 12
+
2is + 2r + kw + 1 n r= 2is + 2r + kw + 1 n
r=0

r=0

Sr1 12
Sr 12
+
2is + 2r + kw + 1 n 2is 2r + kw 1 n
Sr
r=0
1
2

1
1
,
+
2r + 2is + kw + 1 n 2r 2is kw + 1 + n
(A.17)
where in the last line we made use of the identity Sr1 = 1 Sr . For r > 0 the leading
term of Sr is O(1) and the above sum is divergent. It can be regularized with the introduction of a damping exponential er

d
er
= log
log (A) + O().
r +A
dA
(A.18)
r=0
Hence, for the summations appearing in (A.17) this regularization gives

r
rZ

(is + 12 + nkw
Sr 12
1 d
1
2 )
2 log +
log
.
=
2is + 2r + kw + 1 n 4
i ds
(is + 12 nkw
2 )
(A.19)
Similarly, for the second term of (A.16) we obtain

r Z
r
(is + 12 n+kw
Sr 12
1
1 d
2 )
log
.
2 log +
=
2is + 2r + kw + 1 + n 4
i ds
(is + 12 + n+kw
2 )
(A.20)
Therefore, the leading contribution to the function Q(s; n, w|, ) is , independent and
reads
Q(s; n, w|, )
= log +
(is +
1 d
log
4i ds
(is +
(is +
1 d
log
= log +
4i ds
(is +
1
2
1
2
1
2
1
2
nkw
1
n+kw
2 )(is + 2 2 )
1
n+kw
nkw
2 )(is + 2 + 2 )
+ m)(is + 12 + m)
(A.21)
1
m)(is + 2 m)
are defined as
where the J03 and J03 charges m, m
m=
n kw
,
2
m
=
n + kw
.
2
(A.22)
We conclude that since c1
2 +is,m
= c1
2 +is,m
and c1
2 +is,m
= c1
2 is,m
361
, the leading contri-
bution to (A.15) can be written in terms of the density of states (s; m, m):

Zc (, ) = 8
n,wZ
where
ds
2(s; m, m)
c1 +is,m ( )c1 +is,m ( ),
2
(A.23)
(is +
1 d
1
log +
log
(s; m, m)
=2
2
2i 4ds
(+is +
1
2
1
2
m)(is +
m)(+is +
1
2
1
2
+ m)
+ m)
(A.24)
This result is in perfect agreement with [52].
To summarize, we have shown that the torus partition sum of the coset SL(2, R)/U (1)
can be written in terms of the standard characters c1
and dj,r in the form
2 +is,m
Z(, ) = 16

n,wZ 0
ds (s; m, m)
c1 +is,m ( )c1 +is,m ( )
2
1
k1
2 <j < 2
w,r,r Z
(2j + r + r + kw)dj,r ( )dj,r ( ).
(A.25)
Notice that by keeping only the , -independent part of the function (A.16), we discard a
set of states from (A.15) whose interpretation is unclear [54], as they do not seem to fit in
any of the known SL(2, R)/U (1) representations.
For integer level k it is possible to rewrite (part of) the above partition sum in terms of
the extended coset characters. This is actually quite straightforward for the discrete part.
Starting from (A.10) we can write
k1
Zd (, ) = 8
2
r,r Zk
j = 12 w,n,nZ
k1
=8
2
r,r Zk
j = 12 w,n,nZ
(2j + r + r + k(n + n + w))dj,r+kn ( )dj,r +k n ( )
(2j + r + r + kw)dj,r+kn ( )dj,r +k n ( )
k1
=8
2

(2j + r + r + kw)dj,r ( )dj,r ( ).
(A.26)
j = 12 wZ r,r Zk
The continuous contribution, on the other hand, involves a series of subtle steps. Initially, in Eqs. (A.15) and (A.16) we can set n kw = 2l1 k + g, and n + kw = 2l2 k + g
with l1 , l2 Z and g Z2k . Then, by shifting r r + l1 k the first term in (A.16) can be
written as

Sr 12
Sr+l1 k 12
(A.27)
=
.
2is + 2r + kw + 1 n
2is + 2r + 1 g
rZ
rZ
362
In a similar way, the second term gives

r Z

Sr 12
Sr +l2 k 12
=
.
2is + 2r + kw + 1 + n
2is + 2r + 1 + g
(A.28)
r Z
Recall that in the treatment of the standard characters we kept only the constant, ,
independent, terms in Sr 12 . A similar approach will not work for the extended characters
because of the following reason. From the expansion of the functions Sr in (A.17) it is clear
that the constant term of Sr is dependent on the sign of r. More specifically, for r > 0 the
constant term arises only from the s = 0 term in the expansion and it gives Sr 12 12 . For
r < 0 the constant term arises from two terms with opposite signs, s = 0 and s = 2r 1.
So, it gives Sr 12 1 1 12 = 12 . This makes the r.h.s. of Eqs. (A.27) and (A.28)
l1 - and l2 -dependent and spoils the decomposition of the partition sum in terms of the
. This complication does not arise for the part of
continuous extended characters c1
2 +is,m
the contributions (A.27) and (A.28) that comes only from the s = 0 term in the expansion
of the functions Sr+kl1 and Sr+kl2 . This part is , - and l1 , l2 -independent. As before, it
gives divergent terms and the sums over r and r should be regularized. With the regulator
appearing in Eq. (A.18) we can obtain

Sr+l1 k 12
(is + 12 + g2 )
1 d
1
r

2 log +
e
log
(A.29)
=
2is + 2r + 1 g s=0 4
i ds
(is + 12 g2 )
rZ
and

r Z
e r

(is + 12 g2 )
Sr +l2 k 12
1 d
1

2
log

+
log
.
=
2is + 2r + 1 + g s=0 4
i ds
(is + 12 + g2 )
(A.30)
We have inserted the notation |s=0 to keep track of the fact that we keep only the s = 0
constant term in the expansions of the Sr+kl1 and Sr+kl2 functions. For these contributions
the corresponding terms in the torus partition sum (A.15) give

Zcs=0 (, ) = 8
ds
0

gZ2k
2(s; g)c1 +is, g ( )c1 +is, g ( )

2
with spectral density
(is +
1 d
1
log +
log
(s; g) = 2
2
2i 4ds
(+is +
1
2
1
2
(A.31)
g2 )(is +
g2 )(+is +
1
2
1
2
g2 )
g2 )
(A.32)
The remaining contributions in the torus partition sum cannot be written in terms of the
extended characters. Similar partition function decompositions in terms of extended coset
characters for the supersymmetric SL(2, R)/U (1) were obtained in [54].
Appendix B. Derivation of the Ir ( ) S-modular transformation
The authors of [39] discussed the coset extended characters for the continuous and
discrete representations. The identity representation does not appear in the closed string
363
Hilbert space, but it does appear in the open string sector. In this appendix we discuss
the corresponding extended characters, and determine how they transform under the Stransformation 1 . Our computation starts with the known modular transformation
properties of the corresponding N = 2 extended characters I (r; ) and utilizes the character decompositions (2.37), (2.39) and (2.41) that appear in Section 2.
The N = 2 identity characters transform in the following manner [37]:

1 z
p ,j
i cz
2 /
=e
I r; ,
dp Sr c (p , j ; , z)

j Z2(k2) 0
k2

Srr ,s d (r , s ; , z)
(B.1)
r Zk2 s =2
In this expression, we have set

p ,j
Sr
sinh(Qp ) sinh(2p /Q)

1

e2irj /(k2)
=
,

2(k 2)
| cosh ( p + i j )|2
Srr ,s =
(s 1) 2i r(s +2r )
2
k2 ,
sin
e
k2
k2
(B.2)
(B.3)

2
with Q = k2
. The substitution of the character expansions of Section 2 in (B.1) gives
(for simplicity we set z = 0)

n(k2)+2r
1
1
e2im k(k2) m , k(k2) ( )In+r
k(k 2)

nZ m Zk(k2)
p ,j
dp Sr
nZ j Z2(k2) 0
k2

nZ r Zk2 s =2
n(k2)+j , k(k2) ( )c1/2+is(p ),nj /2 ( )
Srr ,s n(k2)+s +2r , k(k2) ( )ds /2,n+r ( ).
(B.4)
Next we find the linearly independent terms. On the l.h.s. of (B.4) this can be achieved
easily by setting n = k n + p, with n Z and p Zk . On the r.h.s. we have two terms.
For the first term, involving the continuous characters, again set n = k n + p, with n Z
and p Zk and split j as j = (k 2)q + t, where q Z2 and t Zk2 . Then shift
p + q p. For the moment we leave the second term as it is. In this way, Eq. (B.4) takes
the form

(k2)p+2r
1
1
e2im k(k2) m , k(k2) ( )Ik n+rp
k(k 2)

pZk m Zk(k2)
nZ

nZ
pZ
k q Z2

tZk2 0
p ,(k2)q +t
dp Sr
p(k2)+t,
k(k2) ( )
364
c1 +is(p ),k n+
( )
p
k q t
2
k2
p Zk r Zk2 s =2

Srr ,s p (k2)+s +2r , k(k2) ( )ds
( )
++r
,k np
(B.5)
We notice that the sums over n convert the unextended characters to extended ones and
they give

(k2)p+2r
1
1
e2im k(k2) m , k(k2) ( )Irp
k(k 2)

pZk m Zk(k2)

pZ
k q Z2
p ,(k2)q +t
dp Sr
tZk2 0
k2

p Zk r Zk2 s =2
c
p(k2)+t,
( )
k(k2) ( ) 1
+is(p ),p
k q t
2
Srr ,s p (k2)+s +2r , k(k2) ( )ds

2
,p +r
( ).
(B.6)
Now we can fix m and equate those terms that multiply the same -functions. From
the second line in (B.6) this entails setting m = p(k
2) + t. In the third line, we keep all
the terms with p (k 2) + s + 2r = m = p(k
2) + t mod k(k 2). Hence, we get

1
1
p
e2im k Irp
k(k 2)
pZk
=e
2r
2im k(k2)
p ,(k2)q +t
dp Sr
q Z2 0
k2

p Zk r Zk2 s =2 nZ
c1 +is(p ),p
( )
k q t
2

Srr ,s p (k2)+s +2r ,m +nk(k2) ds
( )
,p +r
. (B.7)
In this expression, we use the standard Kronecker delta and the extra summation over an
integer n takes into account the k(k 2)-periodicity of the functions. We have also
moved to the right the part of the exponential that depends on r.
Now, set m = pk
+ t 2p = pk
+ and sum over Zk . In the left side this keeps k
copies of the term with p = 0. Plugging in the explicit form of the modular S-matrices for
the N = 2 characters (B.2), (B.3):
p ,(k2)q +t
Sr
Srr ,s =

rt sinh(Qp ) sinh(2p /Q)
1
e2i k2
=
,

2(k 2)
| cosh ( p + i (k2)q +t )|2
(s 1) 2ir s +2r
2
k2
sin
e
k2
k2
(B.8)
(B.9)
we see that in the part corresponding to the continuous characters at the right, the p dependence drops out. So far we have obtained the equation
365

1
k
Ir
k(k 2)

r sinh(Qp ) sinh(2p /Q)
1
=
e2i k
dp

2(k 2)
| cosh ( pQ + i kq 2+ )|2
q Z2 Zk
0
c1
2 +is(p
), +kq
2

( )
k2

e2im
2r
k(k2)
Srr ,s p (k2)+s +2r ,pk+
+nk(k2)
Zk p Zk r Zk2 s =2 nZ
ds
2
,p +r
( ).
(B.10)
With the definition l = + kq Z2k we can recast it into the form
Ir

rl sinh(Qp ) sinh(2p /Q)
1
=
dp e2i k
c1 +is(p ), l ( )

p
l 2
2
2
2k lZ
|
cosh
(
+
i
)|
2
Q
2k 0

2
+
k(k 2) Z
k
sin
p Zk r Zk2 s =2 nZ
) k(k2) 2ir
e2i(pk+
e
2r
k2

(s 1)
k2
s +2r
k2
d
p (k2)+s +2r ,pk+
+nk(k2) s
2
,p +r
( ).
(B.11)
We can further manipulate the second term at the right as follows. For simplicity, we
may fix p = 0. From the Kronecker delta we deduce that the charge p + r in the character can be replaced by ( p k s )/2 (remember that the extended characters have
periodicity k). It is crucial that k is even and that the term nk(k 2) is an even multiple
of k. The exponents can be treated in a similar manner. Neglecting, for the moment, some
multiplicative constants, the second term in (B.11) can be written as

k2

Zk p Zk r Zk2 s =2 nZ
ds
2
,( p ks )/2
( ).
sin
(s 1) 2i r
k
e
p (k2)+s +2r +nk(k2),
k2
(B.12)
Since k is even, we set k = 2l and we may write p (k 2) + 2r + nk(k 2) =

2(n2l(l 1) + p (l 1) + r ). Furthermore, since p Zk we can write p = 2n + q
with n Zl , q Z2 and eventually we get p (k 2) + 2r + nk(k 2) = 2m + 2q(l 1)
with m Z. The summations over n, n and r can be replaced by a single summation over
the integer m and then we get
k2

Zk s =2 qZ2 mZ
sin
(s 1) 2i r
d
k
( ). (B.13)
e
2m+2q(l1)+s , s

k2
2 ,( qks )/2
366
Because of the Kronecker delta, which implies = 2m + qk 2q + s , this can also be

written as
k2

sin
Zk s =2 qZ2 mZ
(s 1) 2i r(2m2q+s )
k
e
2m+2q(l1)+s , ds
( ).
k2
2 ,mq
(B.14)
By shifting m m + q and then decomposing it as m = nk + t, with n Z, t Zk we find
k2

Zk s =2 qZ2 tZk nZ
sin
(s 1) 2i r(2t+s )
k
2nk+2t+qk+s , ds ( ).
e
k2
2 ,t
(B.15)
Finally, notice that 2nk + qk recombines into a new integer m Z and the summations
over q, and n can be substituted with a single summation over m:
k2

s =2
tZk mZ
sin
(s 1) 2i r(2t+s )
k
e
2t+s +m,0 ds ( ).
k2
2 ,t
(B.16)
For each pair (t, s ) there is always a unique integer m such that 2t + s + m = 0. Hence,
we can just drop the delta function and the summation over m and the final result reads
k2
sin
s =2 tZk
(s 1) 2i r(2t+s ) d
k
s ( ).
e
k2
2 ,t
(B.17)
Recombining everything in (B.11) gives the modular identity
Ir

rl sinh(Qp ) sinh(2p /Q)
1
=
dp e2i k
c1 +is(p ), l ( )

p
l 2
2
2
2k lZ
|
cosh
(
+
i
)|
2
Q
2k 0
k2

2
(s 1) 2i r(2t+s ) d
k
+
e
sin
s ( ).
k2
k(k 2) tZ
2 ,t
s =2
(B.18)
Appendix C. Explicit expressions for the extended characters Ir

In this appendix we derive an explicit expression for the extended coset character Ir .
This is needed for the identification of various open string spectra. Our derivation starts
from the defining equation (2.35)
k2 2r
2
I (r; ) = 1 ( )
(C.1)
q 2k ( k2 +n) In+r ( ).
nZ
To proceed we could either write down I explicitly and then expand the denominators
using the geometric series expansion, or we could use an identity that relates I with the
continuous and discrete characters and then deduce the expansion of the identity character
367
d
Ir from the known expansions of the coset characters c1
m and j,r . We follow the
2 +is, 2
15
latter approach.
The N = 2 character identity we would like to use can be found in Eq. (2.24) of [37]. It
has the form

i
I (r; ) = c , 2r; d (r, k 2; ) d (r 1, 2; ).
(C.2)
2
The values in this expression make no difference in the final result. Plugging on the r.h.s.
the expansions (2.39) and (2.41) we find the coset character identity
Ir ( ) = c1 1 ,r ( ) dk2 ,r+1 ( ) d1,r1 ( ).

2
(C.3)
The next step is to use the explicit expansions of the continuous and discrete characters
that appear in [39]. They have the following form:

1
m 2
c1 +is, m = 2 ( )q 4(k2)
(C.4)
q k(l+ 2k ) ,
2
lZ
dj,r ( ) = 2 ( )q
(j 21 )2
k2
(j +r+kl)2
k
Sr+kl ( ),
(C.5)
lZ
where
Sr ( ) =

1
(1)s q 2 s(s+2r+1) .
(C.6)
s=0
Plugging them into (C.3) we find

1

1
2
Ir ( ) = 2 ( )q 4(k2)
q k (r+kl) 1 q r+kl+1 Sr+kl+1 ( ) Sr+kl1 ( ) .
lZ
(C.7)
By using the identities
q r Sr ( ) = Sr ( ),
(C.8)
Sr ( ) = 1 Sr1 ( )
(C.9)
and
we can write
Ir ( ) = 2 ( )q 4(k2)
1

1
2
q k (r+kl) Srkl ( ) Srkl1 ( ) .
(C.10)
lZ
Finally, after summing over r we obtain

1 2

1
Ir ( ) = 2 ( )q 4(k2)
q k n Sn ( ) Sn1 ( ) .
rZk
(C.11)
nZ
15 A quick derivation of the identity extended characters I also follows directly from the expression of the
r
standard identity character, see Eq. (2.4) in Section 2. The alternative derivation of this appendix gives a further
consistency check of our formulae.
368
References
[1] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[2] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4 (1989) 251.
[3] E. Verlinde, Fusion rules and modular transformations in 2D conformal field theory, Nucl. Phys. B 300
(1988) 360.
[4] V. Schomerus, Lectures on branes in curved backgrounds, Class. Quantum Grav. 19 (2002) 5781, hepth/0209241.
[5] J.M. Maldacena, G.W. Moore, N. Seiberg, Geometrical interpretation of D-branes in gauged WZW models,
JHEP 0107 (2001) 046, hep-th/0105038.
[6] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996) 407,
hep-th/9606112.
[7] D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries, Nucl. Phys.
B 372 (1992) 654.
[8] J.L. Cardy, D.C. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett. B 259 (1991)
274.
[9] V. Fateev, A.B. Zamolodchikov, A.B. Zamolodchikov, Boundary Liouville field theory, I: boundary state
and boundary two-point function, hep-th/0001012.
[10] J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138.
[11] A.B. Zamolodchikov, A.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[12] T. Fukuda, K. Hosomichi, Super-Liouville theory with boundary, Nucl. Phys. B 635 (2002) 215, hepth/0202032.
[13] C. Ahn, C. Rim, M. Stanishkov, Exact one-point function of N = 1 super-Liouville theory with boundary,
Nucl. Phys. B 636 (2002) 497, hep-th/0202043.
[14] A. Giveon, D. Kutasov, A. Schwimmer, Comments on D-branes in AdS(3), Nucl. Phys. B 615 (2001) 133,
hep-th/0106005.
[15] A. Parnachev, D.A. Sahakyan, Some remarks on D-branes in AdS(3), JHEP 0110 (2001) 022, hepth/0109150.
[16] P. Lee, H. Ooguri, J.W. Park, Boundary states for AdS(2) branes in AdS(3), Nucl. Phys. B 632 (2002) 283,
hep-th/0112188.
[17] B. Ponsot, V. Schomerus, J. Teschner, Branes in the Euclidean AdS(3), JHEP 0202 (2002) 016, hepth/0112198.
[18] S. Ribault, Two AdS(2) branes in the Euclidean AdS(3), JHEP 0305 (2003) 003, hep-th/0210248.
[19] K. Bardakci, E. Rabinovici, B. Saering, String models with C < 1 components, Nucl. Phys. B 299 (1988)
151.
[20] D. Altschuler, K. Bardakci, E. Rabinovici, A construction of the C < 1 modular invariant partition functions,
Commun. Math. Phys. 118 (1988) 241.
[21] M. Rocek, K. Schoutens, A. Sevrin, Off-shell WZW models in extended superspace, Phys. Lett. B 265
(1991) 303.
[22] G. Mandal, A.M. Sengupta, S.R. Wadia, Classical solutions of two-dimensional string theory, Mod. Phys.
Lett. A 6 (1991) 1685.
[23] S. Elitzur, A. Forge, E. Rabinovici, Some global aspects of string compactifications, Nucl. Phys. B 359
(1991) 581.
[24] A. Giveon, Target space duality and stringy black holes, Mod. Phys. Lett. A 6 (1991) 2843.
[25] R. Dijkgraaf, H. Verlinde, E. Verlinde, String propagation in a black hole geometry, Nucl. Phys. B 371
(1992) 269.
[26] E. Witten, On string theory and black holes, Phys. Rev. D 44 (1991) 314.
[27] C. Kounnas, Four-dimensional gravitational backgrounds based on N = 4, c = 4 superconformal systems,
Phys. Lett. B 321 (1994) 26, hep-th/9304102.
[28] I. Antoniadis, S. Ferrara, C. Kounnas, Exact supersymmetric string solutions in curved gravitational backgrounds, Nucl. Phys. B 421 (1994) 343, hep-th/9402073.
[29] A. Giveon, D. Kutasov, O. Pelc, Holography for non-critical superstrings, JHEP 9910 (1999) 035, hepth/9907178.
[30] A. Giveon, D. Kutasov, Little string theory in a double scaling limit, JHEP 9910 (1999) 034, hep-th/9909110.
369
[31] H. Ooguri, C. Vafa, Two-dimensional black hole and singularities of CY Manifolds, Nucl. Phys. B 463
(1996) 55, hep-th/9511164.
[32] J. McGreevy, H. Verlinde, Strings from tachyons: the c = 1 matrix reloaded, JHEP 0312 (2003) 054, hepth/0304224.
[33] I.R. Klebanov, J. Maldacena, N. Seiberg, D-brane decay in two-dimensional string theory, JHEP 0307 (2003)
045, hep-th/0305159.
[34] J. McGreevy, J. Teschner, H. Verlinde, Classical and quantum D-branes in 2D string theory, JHEP 0401
(2004) 039, hep-th/0305194.
[35] E.J. Martinec, The annular report on non-critical string theory, hep-th/0305148.
[36] V. Kazakov, I.K. Kostov, D. Kutasov, A matrix model for the two-dimensional black hole, Nucl. Phys. B 622
(2002) 141, hep-th/0101011.
[37] T. Eguchi, Y. Sugawara, Modular bootstrap for boundary N = 2 Liouville theory, JHEP 0401 (2004) 025,
hep-th/0311141.
[38] C. Ahn, M. Stanishkov, M. Yamamoto, One-point functions of N = 2 super-Liouville theory with boundary,
Nucl. Phys. B 683 (2004) 177, hep-th/0311169.
[39] D. Israel, A. Pakman, J. Troost, Extended SL(2, R)/U (1) characters, or modular properties of a simple
non-rational conformal field theory, hep-th/0402085.
[40] A. Fotopoulos, Semiclassical description of D-branes in SL(2)/U (1)gauged WZW model, Class. Quantum
Grav. 20 (2003) S465, hep-th/0304015.
[41] S. Ribault, V. Schomerus, Branes in the 2D black hole, JHEP 0402 (2004) 019, hep-th/0310024.
[42] S. Fredenhagen, V. Schomerus, D-branes in coset models, JHEP 0202 (2002) 005, hep-th/0111189.
[43] D. Israel, A. Pakman, J. Troost, D-branes in N = 2 Liouville and its mirror, hep-th/0405259.
[44] C. Ahn, M. Stanishkov, M. Yamamoto, ZZ-branes of N = 2 super-Liouville theory, hep-th/0405274.
[45] E.B. Kiritsis, Duality in gauged WZW models, Mod. Phys. Lett. A 6 (1991) 2871.
[46] P.A. Griffin, O.F. Hernandez, FeiginFuchs derivation of SU (1, 1) parafermion characters, Nucl. Phys.
B 356 (1991) 287.
[47] K. Sfetsos, Degeneracy of string states in 2D black hole and a new derivation of SU(1, 1) parafermion
characters, Phys. Lett. B 271 (1991) 301.
[48] I. Bakas, E. Kiritsis, Beyond the large N limit: nonlinear W as symmetry of the SL(2, R)/U (1) coset
model, Int. J. Mod. Phys. A 7 (1992) 55, hep-th/9109029.
[49] K. Itoh, H. Kunitomo, N. Ohta, M. Sakaguchi, BRST analysis of physical states in two-dimensional black
hole, Phys. Rev. D 48 (1993) 3793, hep-th/9305179.
[50] A. Pakman, BRST quantization of string theory in AdS(3), JHEP 0306 (2003) 053, hep-th/0304230.
[51] O. Aharony, A. Giveon, D. Kutasov, LSZ in LST, hep-th/0404016.
[52] A. Hanany, N. Prezas, J. Troost, The partition function of the two-dimensional black hole conformal field
theory, JHEP 0204 (2002) 014, hep-th/0202129.
[53] T. Eguchi, Y. Sugawara, SL(2, R)/U (1) supercoset and elliptic genera of non-compact CalabiYau manifolds, hep-th/0403193.
[54] D. Israel, C. Kounnas, A. Pakman, J. Troost, The partition function of the supersymmetric two-dimensional
black hole and little string theory, hep-th/0403237.
[55] J.D. Lykken, Finitely reducible realizations of the N = 2 superconformal algebra, Nucl. Phys. B 313 (1989)
473.
[56] L.J. Dixon, M.E. Peskin, J. Lykken, N = 2 superconformal symmetry and SO(2, 1) current algebra, Nucl.
Phys. B 325 (1989) 329.
[57] V.K. Dobrev, Characters of the unitarizable highest weight modules over the N = 2 superconformal algebras,
Phys. Lett. B 186 (1987) 43.
[58] E. Kiritsis, Character formulae and the structure of the representations of the N = 1, N = 2 superconformal
algebras, Int. J. Mod. Phys. A 3 (1988) 1871.
[59] T. Eguchi, A. Taormina, Character formulas for the N = 4 superconformal algebra, Phys. Lett. B 200 (1988)
315.
[60] T. Eguchi, A. Taormina, On the unitary representations of N = 2 and N = 4 superconformal algebras, Phys.
Lett. B 210 (1988) 125.
[61] T. Eguchi, H. Ooguri, A. Taormina, S.K. Yang, Superconformal algebras and string compactification on
manifolds with SU(N ) holonomy, Nucl. Phys. B 315 (1989) 193.
370
[62] S. Odake, Extension of N = 2 superconformal algebra and CalabiYau compactification, Mod. Phys. Lett.
A 4 (1989) 557.
[63] S. Odake, Character formulas of an extended superconformal algebra relevant to string compactification, Int.
J. Mod. Phys. A 5 (1990) 897.
[64] D. Gepner, Field identification in coset conformal field theories, Phys. Lett. B 222 (1989) 207.
[65] C. Bachas, M. Petropoulos, Anti-de Sitter D-branes, JHEP 0102 (2001) 025, hep-th/0012234.
[66] N. Nakayama, K. Sugiyama, N = 2 supersymmetric sigma models and D-branes, hep-th/0404227.
[67] A. Sen, SO(32) spinors of type I and other solitons on braneantibrane pair, JHEP 9809 (1998) 023, hepth/9808141.
[68] N. Seiberg, D. Shih, Branes, rings and matrix models in minimal (super)string theory, JHEP 0402 (2004)
021, hep-th/0312170.
[69] S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici, G. Sarkissian, D-branes in the background of NS fivebranes, JHEP 0008 (2000) 046, hep-th/0005052.
[70] A. Hanany, E. Witten, Type IIB superstrings, BPS monopoles, and three-dimensional gauge dynamics, Nucl.
Phys. B 492 (1997) 152, hep-th/9611230.
[71] S. Ribault, D3-branes in NS5-branes backgrounds, JHEP 0302 (2003) 044, hep-th/0301092.
On the photon energy moments and their bias

corrections in B Xs +
D. Benson a , I.I. Bigi a,b , N. Uraltsev c,1
a Department of Physics, University of Notre Dame du Lac, Notre Dame, IN 46556, USA
b INFN, LNF, Frascati, Italy
c INFN, Sezione di Milano, Milan, Italy
Received 12 November 2004; accepted 24 December 2004

Abstract
Photon energy moments in B Xs + and the impact of experimental cuts are analyzed, including the biases exponential in the effective hardness missed in the conventional OPE. We incorporate
the perturbative corrections fully implementing the Wilsonian momentum separation ab initio. This
renders perturbative effects numerically suppressed while leaving heavy quark parameters and the
corresponding light-cone distribution function well defined and preserving their physical properties.
The moments of the distribution function are given by the heavy quark expectation values of which
many have been extracted from the B Xc decays. The quantitative estimates for the biases in
the heavy quark parameters determined from the photon moments show they cannot be neglected
for Ecut 1.85 GeV, and grow out of theory control for Ecut above 2.1 GeV. Implications for the
moments in the B Xc decays at high cuts are briefly addressed.
PACS: 13.20.He; 12.38.Lg; 12.38.Cy; 11.15.Tk
E-mail address: nikolai.uraltsev@mib.infn.it (N. Uraltsev).

1 On leave of absence from Department of Physics, University of Notre Dame, Notre Dame, IN 46556, USA
and from St. Petersburg Nuclear Physics Institute, Gatchina, St. Petersburg 188300, Russia.
doi:10.1016/j.nuclphysb.2004.12.035
372
D. Benson et al. / Nuclear Physics B 710 (2005) 371401
1. Introduction and basics

The decays of beauty hadronsrecognized for a long time as an area allowing high
sensitivity probes of fundamental dynamicshave entered now also the realm of precision
physics on both the experimental and theoretical side. Inclusive semileptonic decays of B
mesons allow the accurate extraction of the CKM parameters Vcb and Vub through their
integrated rates and of other basic heavy quark parameters through the moments of their
spectra. These heavy quark parameters include the heavy quark masses mb and mc which
are external to QCD, and B expectation values of local heavy quark operators like the kinetic 2 or chromomagnetic 2G . The latter are controlled by the strong forces and thus can
in principle be calculated within QCD once theoretical control has been established over
nonperturbative dynamics. The motivation for knowing the heavy quark parameters accurately goes beyond wanting to extract Vcb and Vub : their values provide challenging tests
for the available treatment of strong dynamics. There has been significant recent progress
in this direction both on the theory and experiment [13] side, with BaBar establishing
new benchmarks by combining the high statistics and quality of their data with a robust
theoretical treatment.
Detailed studies of the shape of the photon spectrum in B Xs + transitions produce
complementary information: (i) They allow a systematically independent determination of
2 and maybe even higher order terms. (ii) They lead to a more direct and thus potentially
more accurate extraction of mb than in semileptonic distributions which depend on a combination of mb and mc . (iii) They provide immediate access to the distribution function of
the b quark inside the B meson, the knowledge of which reduces the model dependence in
extracting Vub from inclusive B Xu transitions.
Claims have recently appeared in the literature challenging the classical OPE results
relating the moments of the inclusive distributions to the universal local heavy quark expectation values in the decaying hadrons. They assert that the expectation values extracted
from the b c decays cannot be related to the similar moments of the light-cone distribution function controlling the heavy-to-light decays like b s + , due to intervention
of the perturbative effects. We find no justification for such revisions, and contend that the
perturbative corrections to the moments in both types of the decays can be treated in the
standard OPE approach, see Ref. [4] for a recent analysis. The heavy quark parameters
extracted from B Xc definitely can be used to constrain the light-cone distribution
function.
In practice, when measuring the photon energy spectrum in B Xs + experimenters
have to impose lower cuts on the photon energy to suppress backgrounds. Those cuts are
presently in the range 1.8 to 2.1 GeV with little realistic hope for lowering them in the
near future. Yet, as emphasized recently [5], such cuts degrade the treatment of inclusive B
decays in the short-distance expansion. High cuts give rise to a new mass scale parameter
Q in the expansion, which we refer to as hardness, along with the original parameter mb .
Say, for the inclusive B Xs + decays we have
Q MB 2Ecut .
(1)
373
Some nonperturbative contributions are then described in inverse powers of Q rather than
mb . For high Ecut Q becomes too small for a 1/Q expansion to be reliable.
The usually employed practical version2 of the operator product expansion (OPE) often
manifests this deterioration in the power corrections explicitlyin particular for semileptonic decaysthrough the Wilson coefficients of the higher-dimensional operators: they
are functions of Ecut /mb , which effectively transform 1/mkb into 1/Qk terms. A similar
reduction of the effective hard scale is present also in the purely perturbative contributions
and can be traced through a decrease in the effective normalization scale for s , once the
BLM corrections are included.
The OPE-based theory usually leaves out, however, exponential terms like eM/had
with had denoting the typical momentum scale of nonperturbative QCD, and M stands
for the generic high mass scale used in the expansion. Those are insignificant as long as M
is driven by the b quark mass scale and exceeds a few GeV. Yet, as the presence of high
cuts causes Q to decrease towards 1 GeV, the terms which are exponential in Q rather than
in mb could quickly become significant.
The simplicity of two-body kinematics for b s + prevents the cuts from affecting
the Wilson coefficients of the higher-dimensional operators. They do not directly reveal,
therefore deterioration of the expansion parameter with high cuts. The cut dependence will
surface and manifest this once the perturbative corrections to these power terms are incorporated, yet this still is seen as a future level for theory. Let us note that these exponential
effects are not connected with the potential growth of the perturbative corrections to the
Wilson coefficients when the cuts are raised. We have found indications that in the actual
B decays such perturbative corrections are insignificant, once a physical renormalization
scheme is used.
The pilot study of Ref. [5] has shown that using the conventional OPE expressions
which lack such exponential terms introduces substantial biases into the values of mb and
2 extracted from truncated photon energy momentsi.e., moments evaluated with a cut
on the photon energyfor Ecut 2 GeV. Furthermore, applying the corresponding bias
corrections, even estimated in a simplified manner, resulted in a surprisingly good agreement in the OPE description between b s + and b c + , and thus removed
an otherwise apparent inconsistency. Therefore, in the present paper we want to present a
more careful theoretical evaluation of the photon moments with cuts, even if it is somewhat
model-dependent, and to put them into a form suitable for a global analysis of various B
decay data.
It had been appreciated quite some time ago that the usual application of the OPE to
radiative B decays must be incomplete, since it predicts the nonperturbative component of
the photon energy moments to be independent of Ecut . Ref. [7] tried to assess the potential
effect of the biases, in our terminology, by applying general inequalities based only on
the positivity of the distribution function. Being very general, the inequalities however
are saturated by quite unphysical functions having a support consisting of isolated points.
Therefore they are too weak to be of interest in practice. Furthermore, these inequalities do
2 A detailed discussion of the principal simplifications employed in what we conventionally refer to as the
practical OPE, can be found, e.g., in the review by Shifman [6], and references therein.
374
not generally hold in the presence of the perturbative corrections. Hence this approach has
limited utility.
In Ref. [7] it was also stated that the inequalities are violated once the cut is placed too
high, and this was interpreted as physical evidence for the OPE breaking down for high
cuts. We do not think such an interpretation can be correct. The discrepancy noted there
has a purely algebraic nature. It was actually caused by the Darwin term in Eq. (5.11) of
Ref. [7] entering with the wrong sign. Correcting it leads to the inequality being trivially
satisfied at arbitrary Ecut and thus eliminates the contradiction, as it has to be for a purely
arithmetic bound.3
Alternative numerical estimates based on the AC2 M2 ansatz [8] led to a gross underestimate: the biases emerged as negligible at cuts below 2 GeV. The perturbative corrections
were not incorporated at this point, and too low nonperturbative parameters were used as a
result of following the pole scheme of HQET.
Similar complications plagued attempts to provide definite numerical estimates in
Ref. [9]. The analysis there did not use consistently defined Wilsonian heavy quark parameters, and operated in terms of somewhat indefinite HQET analogies. The perturbative corrections in that scheme are not stable and significantly change subtle effects like
biases in the photon moments. On the other hand, the values of the nonperturbative parameters determining the assumed distribution function were varied in an overly wide
range including unphysically small values. Therefore it was difficult to arrive at conclusive estimates, and Ref. [9] came up with no certain result: the estimates ranged from
the biases being totally negligible even at Ecut = 2 GeV to very significant down to
Ecut = 1.8 GeV.
As described in Ref. [5], relying on the Wilsonian implementation of the OPE allows
to derive more robust statements. The perturbative corrections to physical effects are moderate, and the nonperturbative parameters have definite values, often known with small
numerical uncertainties. For instance, BaBar has recently presented the results of a comprehensive fit to the moments they have measured in inclusive semileptonic B decays,
obtaining [2]
mb (1 GeV) = 4.61 0.05|exp 0.04|HQE 0.02|th GeV,
2 (1 GeV) = 0.447 0.035|exp 0.038|HQE 0.010|th GeV2 ,
(2)
even without incorporating independent theoretical constraints on the heavy quark parameters (see, e.g., [10]). The results are in good agreement with earlier preliminary DELPHI
values [1,11].
3 The bound follows from the positivity of the integrals of F (z)(1 z)(1 tz)2 for arbitrary t , which is a
quadratic polynomial in t . Its determinant is therefore negative or zero, which yields the inequality. Since the
difference 1 Pn (Ecut ), with Pn the weight used in Ref. [7] belongs to the above class of functions (t = 1),
Eq. (5.11) is trivially satisfied.
375
The analysis of B Xs + is complicated by the fact that it is induced not only by

the genuinely local weak operator
e
mb s F (1 + 5 )b,
O7 =
(3)
32 2
but depends also on additional terms in the effective Lagrangian

4GF
ci Oi
Lbsg = Vtb Vts
2
i=1
8
(4)
(we use the standard definition of Oi which is given, e.g., in Ref. [9]). In particular, decays
mediated by the operator O2 = 14 s k (1 5 )bl cl (1 5 )ck (with the superscripts k,
l denoting color indices to be summed over) contribute through a (real or virtual) cc-pair
converting into a photon. The effect of operators other than O7 , most notably of O2 , is
essential for a precise evaluation of the decay rate [12]. However, it is expected that the
normalized photon energy moments are not changed noticeably. The corresponding refinements of Refs. [9,13] modify them by negligible amounts and therefore are irrelevant in
practice. While illustrating their effect for completeness, we need not dwell on the theoretical subtleties associated with their evaluation.
The effects not captured directly by the standard OPE also affect the b c inclusive
semileptonic decay rates, once the applied cut on the charged lepton energy becomes too
high thus decreasing the hardness Q. These effects are conceptually similar to the biases
in the photon energy moments, however the size of the two cannot be related to each other
unambiguously.
The remainder of the paper will be organized as follows: Section 2 presents the standard
OPE for the photon energy spectra and their moments, based on the Wilsonian implementation with an explicit scale separation. We address in detail the nontrivial impact of
experimental cuts on the OPE in Section 3 and how they induce biases in the values of
such truncated moments; the interplay with the perturbative corrections is studied. A stepby-step illustration of the computational procedure is provided. We discuss the effect of
the nonvalence four-quark operators on the inclusive B Xs + characteristics in Section 4. In Section 5 we comment on the corresponding bias effects in semileptonic decays
before summarizing in Section 6. While the numbers and plots in the paper use the exact formulae, only simplified expressions are presented in the main text. Many of the full
expressions are too cumbersome and are relegated to Appendix A. In Appendix A.1 we
describe in some technical detail the calculation of the photon spectrum in a formulation
where the Wilsonian prescription for the OPE has been implemented from the start. Additional non-O7 perturbative corrections are given in Appendix A.2, and in Appendix A.3
we give the numerical evaluation of the bias effects.
While the present paper was finalized the preprints [14] appeared which analyzed the
B Xs + decay heavily relying on a SCET-motivated approach. We largely disagree
with the findings of the author, which basically claim large perturbative uncertainty in the
decays and challenge accurate calculability of the integrated characteristics derived from
the B Xs + spectrum. We have added Section 2.1 to briefly address this controversy,
however the detailed discussion of the SCET-based approach goes beyond the scope of the
present paper.
376
2. Standard OPE for the photon energy moments

The purely perturbative spectrum for b s + has been calculated in Ref. [13] (in
the pole scheme) to first order in s and including the BLM corrections 0 s2 (0 =
11 23 nf = 9). The results for the moments read

E pert
E >Ecut

s
0 s 2
mb
,
1 a1 (x; ) + a 1 (x)
b1 (x; ) + b1 (x)
2
2
2
pert

E E 2
E >Ecut
2

s
0 s 2
mb
a2 (x; ) + a 2 (x)
+ b2 (x; ) + b2 (x)
=
,
2
2
(5)
where
x=
2Ecut
.
mb
(6)
Existing calculations of the perturbative corrections do not allow to specify which b quark
mass enters x in Eq. (6), since the results in different normalization schemes differ only
in order s2 (without 0 ). However, on general grounds we know that the pole mass cannot appear in short-distance calculations; additional arguments can be found in Ref. [21].
Therefore, we assume mb in Eq. (6) to be the running b quark mass normalized at the scale
1 GeV.
Terms proportional to a 1 (x), b1 (x), a 2 (x), b2 (x) correspond to contributions from operators other than the dominant O7 . While those are noticeable for the total decay width,
they are expected to be insignificant for the photon energy spectrum. For instance, they
are found to shift the first and second moment by about 0.004 GeV and 0.0008 GeV2 ,
respectively, which translates into mb 8 MeV and 2 0.01 GeV2 , i.e., far below
the accuracy of the perturbative piece alone. Thus there is no need for a detailed discussion
of these effects.
The values of the coefficients a1 (x; ) b2 (x; ) in Eqs. (5) depend on the normalization point . They can be fixed by the fact that the total moments (including nonperturbative contributions) are independent of and by the condition that, to a given order
in s , the expressions at = 0 reproduce the result in the pole scheme which assumes no
infrared cutoff in the perturbative contributions. This yields
3
4
13
CF 3 ,
a1 (x; ) = a1 (x; 0) CF
3
mb 18
mb

M
M
8
319 3
4
13
+
+
b1 (x; ) = b1 (x; 0) CF ln
CF ln
,
3
2 3 mb 18
2 130 m3b
2
3
1
4
a2 (x; ) = a2 (x; 0) CF 2 + CF 3 ,
3
mb 9
mb
377

M
M
13 2
1 3
1
4
ln
+
C
+
2
b2 (x; ) = b2 (x; 0) CF ln
+
.
F
3
2
6 m2b 9
2
20 m3b
(7)
(M is the normalization scale for s in the MS scheme; we take it to be mb .) The analytic

expressions for a1,2 (x; 0) and b1,2 (x; 0) are given in Appendix A.1, Eqs. (A.15).
The expressions in Eqs. (7) are approximate in their -dependenceeven to a given
order in s in two respects. They account for terms only through order 3 /m3b . Since
/mb is a truly small parameter at 1 GeV, these are tiny corrections. Secondly and
more importantly, Eq. (7) is valid only at sufficiently low Ecut . Otherwise, the simple prescription described above does not reproduce the actual -dependence of the moments: it
introduces systematic shifts that are the perturbative analogue of the biases. In the one-loop
approximation (including all its BLM corrections) this happens for 2Ecut > mb 2, i.e.,
where the hardness Q descends below 2. At high Ecut these perturbative biases become
significant.
Therefore we do not rely on the simplified expressions (7), but rather compute the moments explicitly integrating the perturbative spectrum calculated with the Wilsonian cutoff
at the separation scale . We have derived it both to order s and to all orders in BLM, as
described in Appendix A.1. The analytic expressions for the first-order moments ai (x; )
are reasonably compact. For the BLM corrections the analytic expressions have also been
obtained, yet they are too lengthy and we do not give them explicitly in this paper. We have
checked that at low cuts the -dependence indeed follows Eqs. (7).
The Wilsonian spectrum and, in particular its variation upon changing has some remarkable features. In contrast to the usual spectrum, it contributes even above mb ()/2.
The center of gravity of the spectrum upon varying does not change to the first approximation; modifications arise in order 2 /mb , as mandated by the OPE applied to
perturbative effects.
The nonperturbative pieces in the energy moments are given by
3 7 3
3
2 2G 5D
c2 LS
LS
,
4mb
c7 54m2c
12m2b

3 3
2
2D
2
1
LS
2 power
.
E E
=
+O
12
12mb
m2b
E power =
(8)
(9)
We have calculated them for the local weak decay operator O7 ; our results agree with the
expressions in the revised version of Ref. [7]. In principle, there are perturbative corrections
to the coefficients of these power terms; they have not been calculated for realistic decays.
The last term in Eq. (8) stands for certain effects of the nonperturbative conversion of
a cc
pair into a photon where the underlying weak decay is b ccs.
It scales only like
1/m2c . Such terms cannot arise in totally inclusive rates; they rather represent final state
interaction among the channels mediated by O2 and O7 .
One should note that there exists no (local) OPE for such rescattering widths. Thus, for
instance, there is no reason for mc entering there to be a short-distance charm mass. One
may think that the meson mass MD is actually appropriate there; this choice is adopted in
our numerical estimates. Fortunately, again this effect is negligible, yielding only E
0.003 GeV, i.e., potentially shifting mb by as little as 6 MeV.
378
Without accounting for the perturbative corrections to the Wilson coefficients of higherdimensional (nonperturbative) operators the total moments are given by the sum of the
perturbative and nonperturbative contributions. In the absence of the former, the energy
moments obtained in the 1/mb expansion do not depend on Ecut , which clearly cannot
hold for the actual spectrum. Therefore, the thus computed moments are biased:
1
b,
E E pert + E power + m
2
2
pert 2
power
1
2.
E E 2 E2 E 2
+ E E 2

12
(10)
2 are defined as the corresponding corrections to the moments

b ,
Note that the biases m
calculated in the standard OPE in terms of the actual heavy quark parameters. We have
introduced them with the coefficients motivated by the relation between the moments and
the heavy quark parameters in the large-mb limit. For this reason the biases are close
yet strictly speaking not identicalto the shifts to be applied to the values of mb and
2 naively extracted using the conventional prescription. In Section 3 we formulate the
suggestions for evaluating these biases.
2.1. Comments on the literature
When the present study was completed papers [14] appeared which addressed inclusive b s + decays from a different perspective motivated by the technique of the
so-called soft-collinear effective theory (SCET).4 Their basic statement appears rather
paradoxicalthe perturbative uncertainties were claimed to be quite significant, in conflict with our findings. Since we have performed explicit calculations, we conclude the
large uncertainties in the treatment of Refs. [14] are mainly an artifact of their approach,
intrinsically associated with the adopted SCET framework. A critical discussion of the latter goes beyond the scope of our paper. Instead, we shall try to elucidate the problem for
the case at hand with minimal technical details. With SCET being a fast evolving field aiming at developing a uniform approach to the light-front processes in B decays (b u,
b s + ), certain controversial issues remain there. In particular, we may not agree with
some statements in SCET. It is therefore difficult to provide an unbiased thorough review
of the literature on the subject. As a partial solution, we refer the reader to the recent papers [1416] for the novel applications and for references to the earlier papers and to the
alternative approaches.
The so-called soft-collinear factorization in QCD is highly emphasized in Refs. [14] as
a theoretical advantage; according to it the spectrum is represented in the form involving
the hard function H , jet function J and the soft shape function S:
d
H J S,
dE
(11)
4 There are actually a number of competing versions of SCET, sometimes marked with the numbers: SCET-I,
SCET-II, etc. A partial list of references can be found, for instance, in Refs. [14].
379
where symbolically denotes a convolution. S is just the familiar nonperturbative heavy

quark distribution function; in our Wilsonian scheme it is F (k+ ). In fact, the usual factorization of the soft modes employed in the Wilsonian OPE (i.e., between the effects from
small and large distances) is sufficient for this purpose. Moreover, H J is nothing but the
pert
(Wilsonian) perturbative spectrum d
dE . Contrary to the impression given in Ref. [14],
an expression for the total spectrum in this form was presented already in 1993 [17] (see
also Ref. [18]), not to mention a number of later papers.
Soft-collinear factorization, which goes back to the Low theorem [19] and its generalization in the GribovLow theorem [20], in this context refers merely to how to calculate
the perturbative spectrum itself. Regardless of anything else, it is well known how to calculate the perturbative corrections in B decays, and this has been done.
The principal point of the analysis of Refs. [14] is dividing the dynamics driving inclusive b s + decays into three domains of essentially different mass scale, namely
mb , mb hadr and hadr . The perturbative effects residing in the first two regions
were treated separately applying the renormalization group methods which implies a strong
hierarchy among the scales.
A critical discussion of the validity of a similar approach to b s + has already
been given in Ref. [21]. Rather than recapitulating the details here, we mention only a few
obvious points. With the genuine Wilsonian cutoff at 1 GeV the gluon energy in the perturbative diagrams cannot be below 1 GeV. Kinematically, on the other hand, no gluon with
E > m2b 2.3 GeV can be emitted, and even those with an energy approaching 2 GeV are
strongly suppressed. This leaves a limited phase space even for single gluon emission and
suppresses multiple emissions even more. The approximation where both the individual
gluon momenta and their aggregate energy are much smaller than the jet energy mb /2,
which is assumed in the Sudakov-type treatment, can never be realized in B decays even
remotely.
The Sudakov treatment aims at log-enhanced physical effects, it allows their resummation. For that one employs a reorganization of the perturbative series in powers of 1/ ln Q
rather than s , with Q denoting a supposedly large ratio of relevant scales. With s ln Q
then being of order unity, the expansion in its powers has to be resummed. Yet in B mesons
the large logs hardly reach unity. One cannot rely on an expansion valid at large ln Q,
when instead ln Q < 1. The large-log series might not even converge in that domain; there
is no reason to expect such a procedure to yield a meaningful result. This general problem
becomes quite manifest in b s + decays [21]: for instance, the NLO terms dominate over the leading order turning the physically expected Sudakov suppression into an
enhancement.
Furthermore, the renormalization group treatment per se is inefficient in fixing the precise scale at which to evaluate the running strong coupling, especially in the case of a
limited momentum interval. The concept of the renormalization invariance itself has nothing to say about whether, for instance, s () appearing in Refs. [14] is actually s (), or
s (2) or s (/2). On the other hand, s runs fast in the considered low-energy domain.
This is a major factor contributing to the large uncertainties of Refs. [14]. This ambiguity
would only be resolved in the RG framework by calculating the matching coefficients to
higher orders. At the same time, the improvement achieved directly in QCD by adding
380
straightforward BLM corrections, without subdividing the perturbative domain, does this
far more reliably [21].
Let us recall that a detailed illustration of how a formally correct large-log renormalization-group treatment yields a numerically misleading result for an energy scale gap
typical in B physics, was given already 10 years ago [22], in a similar context of the
zero-recoil b c transitions. That time a similar NLO renormalization-group improvement accomplished by Neubert was claimed to yield a precision perturbative factor A =
0.986 0.006 [23], while a thoughtful application of the direct one-loop QCD calculation yielded A 0.96a value later confirmed by the explicit higher-order calculations.
We think the observed large uncertainties in the calculation of the photon spectrum in
Refs. [14] root in a similar misconception.
An additional advantage of our approach over SCET is that the (quite significant) 1/mb
corrections, including those in the perturbative coefficients, are automatically included;
the KLN-type relations are respected at every order, which does not necessarily hold in the
truncated RG approach. We emphasize that, using a commensurate language, the required
anomalous dimensions, the BLM corrections to the SCET matching coefficients are all
automatically incorporated in the full QCD perturbative corrections we use; they actually
can be read off from the analytic expressions of Appendix A.1.
The most accurate evaluation for B decays is obtained in the full-QCD fixed-order
perturbative calculations. Implementing the Wilsonian approach significantly reduces the
corrections and makes them well behaved; the effect of the running of the strong coupling
in different kinematic domains is readily incorporated by calculating the BLM corrections
in the same framework. This strategy had already been advocated in Ref. [21] and is performed in this paper.
3. Bias corrections
3.1. Basic ansatz
As described in Ref. [5] and stated in the Introduction, there is no unambiguous way to
determine precisely the biases in the photon moments, unless the nonperturbative dynamics
for B mesons have been brought under full theoretical control. On the other hand with
B mesons constituting the ground state we can expect the b quark distribution function
F (k+ ) describing its intrinsic motion, to be essentially a positive and smooth function with
a pronounced peak near E = mb /2. There is no unique choice for such a function. Yet
once the average and the width of the distribution are fixed by the values of the heavy quark
parameters, the function is sufficiently constrained to allow reasonably accurate estimates
of the cut-induced biases for naturally shaped distributions.
For proper perspective let us recall a few basic facts. The tail of F (k+ ) corresponding
to the very low E and its behavior near maximal k+ corresponding to E approaching
MB /2 depend on the high-dimension expectation values and are determined by the subtle
details of the bound-state dynamics. That implies that nothing reliable can be said about
biases at Ecut mb /2, nor are they under full theoretical control at very low Ecut . Yet
for low Ecut their size is sufficiently small to render them insignificant. Our prescription
381
is therefore meant to be applied for Ecut in the range from about 1.7 to 2.1 GeV. Those
numbers follow from the expected values mb 4.6 GeV and 2 0.45 GeV2 [1,2].
The relevant parameters describing the shape of the b quark distribution function are its
mean and variance which (in terms of E ) are mb /2 and 2 /12, respectively, in the heavy
quark limit. For finite quark mass they have power corrections, which can be read off from
b and 2 with
Eqs. (8), (9). For use in F (k+ ) we can then replace mb and 2 by m
m
b mb +
3 7 3
2 2G 5D
LS
,
2mb
6m2b
2 2 r3
3 3
2D
LS
.
mb
(12)
In the strict 1/mb expansion one has r3 = 1. However, the 1/mb corrections are not negligible as is seen from the expression for the second moment. On the other hand, it must
be positive at any mb . This is the reason to expect that the power expansion for it is
sign-alternating, and that the 1/m2b correction partially offsets the last term in Eq. (12).
Therefore, we use in practice r3 = 0.75 0.25 as an educated guess of these effects.
Following Refs. [5,21] we estimate the biases in Eqs. (10) using two anstze for the
heavy quark distribution function:
F1 (k+ ) = N1 ( k+ ) eck+ ( k+ ),
2
F2 (k+ ) = N2 ( k+ ) ed(k+ ) ( k+ ),
(13)
with the parameters fixed by Eqs. (12). We will see that the two classes of functions, despite
their different functional dependence, yield very similar results once Eqs. (12) are imposed
fixing their mean and variance. Therefore we take their average as the central value, and
include the difference into the estimate of the theoretical uncertainty.
In general, the adopted distribution functions depend on the values of mb and 2 , and
the biases additionally on Ecut . Within the 1/mb expansion the latter depend primarily on
=m
the combination Q
b 2Ecut (directly related to the hardness Q MB 2Ecut ) rather
than on Ecut and mb separately. By dimensional arguments they are actually functions of
two variables since the absolute scale mb does not enter. Yet even with these simplifications
the resulting expressions are cumbersome, and it would not be enlightening to present them
here. Fortunately it is not necessary; for we have verified that the biases follow simple
scaling
behavior with reasonable accuracy sufficient for

our purposes. Once calibrated by
2 :
2 , the biases depend primarily on the ratio q Q/
b
m

= f (q, r) fs (q),
2
= g(q, r) gs (q),
2
(14)
with the residual dependence on the remaining ratio r = 2 / 2 rather weak (

b ). The functions fs and gs are shown in Fig. 1. (For illustration in Section 5
MB m
we also give the corresponding estimate for the third moment of the distribution.) The difference between the two curves which represent the results for the two classes of functions
defined in Eq. (13) gives a sense of the scale of the theoretical uncertainty, albeit cum
grano salis: having them coincide at some point does of course not mean there is then no
382
2 ) 2
b /( 2 ) 2 and
2 / 2 as functions of the cut scale variable q = Q/(
Fig. 1. The dimensionless ratios m
as described in the text. The middle blue curve gives the central value, the average of the biases obtained using
the two distribution functions. (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this article.)
uncertainty. The functions fs (q) and gs (q) can adequately be described by the following
simple functions for realistic values of q and r:

fs (q) 0.415 exp 1.3q 0.28q 2.5 ,

gs (q) 0.80 exp 0.60q 0.27q 2.5 .
(15)
3.2. Biases and perturbative corrections
The above bias evaluation was based purely on the nonperturbative primordial distribution. In reality cuts are applied to the observable spectrum resulting from the convolution of
the perturbative and the nonperturbative pieces. Perturbative corrections modify the bias.
Moreover, the effect of the perturbative corrections depends on their implementation, in
383
particular on the separation scale . Indeed, the moments of the combined spectrum are
-independent at a given cut, as it has to be for an observable. The biases computed with
the nonperturbative distribution alone, on the other hand, would depend on the choice of
since the distribution function and, in particular mb depend on the normalization scale.
It has been observed [5] that with the prototype Wilsonian perturbative spectrum of
Ref. [21] perturbative contributions had little impact on the biases. Having calculated
the perturbative spectrum explicitly with the Wilsonian cutoff, in this paper we adopt
a more accurate direct approach. Namely, we determine the complete spectrum starting
from this Wilsonian perturbative kernel. Simultaneously, we confront the results with the
corresponding OPE predictions which include complete expressions for the perturbative
moments derived from this spectrum, see Appendix A.1.
Again we find that for 1 GeV the biases are only weakly modified by the perturbative corrections. This is illustrated by Fig. 2, where we have assumed mb (1 GeV) =
4.6 GeV and 2 (1 GeV) = 0.45 GeV2 . Even so in our analysis we retain the perturbative
b and
2 . The red curve is the complete bias using the full
Fig. 2. The perturbative effects on the biases m
spectrum. The green curve is the bias calculated without perturbative corrections, and the blue line shows the
difference between the two curves. (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
384
shifts for the sake of completeness. At the same time, due to their smallish size we just
2 by functions of only the single variable q,
b and in
approximate these shifts in m
obtained at the central values expected for the parameters:
pert fs (q) 0.0103q 3 0.0729q 2 + 0.165q 0.111,
pert gs (q) 0.0138q 3 + 0.0.0437q 2 + 0.0316q 0.042.
(16)
If necessary, these can be calculated anew at the final fitted values; the procedure converges
well since the shifts are small.
3.3. Procedure
Let us summarize now the proposed way to obtain the numerical theory predictions for
the photon energy moments. The inputs are the (true) values of the heavy quark parameters;
they are assumed to be normalized at = 1 GeV. The procedure consists of three steps.
First, the biased, i.e., uncorrected moments are evaluated by adding the nonperturbative
and perturbative contributions. The former are given by Eqs. (8), (9). The perturbative ones
are described in detail in Appendix A.1; they refine the previously applied Eqs. (5).5 The
first-order integrals ak are given in Eqs. (A.5) and (A.6) with B1 defined in Eq. (A.2). The
similar BLM pieces of the moments, bk , are obtained by integrating Eq. (A.11) (and, for
b1 , additionally subtracting the value of B2 given by Eq. (A.12)). Alternatively, one can
use the approximate numerical interpolations Eqs. (A.16) and (A.17).
b and
2
Next with m
b and 2 defined in Eq. (12) we can determine the biases m
using Eq. (14)

b = 2 fs (q),
2 = 2 gs (q)
m
(17)

2 , where Q = m
with the dimensionless scaling variable q = Q/
b 2Ecut .
Finally, we may apply additional small adjustments accounting for the perturbative impact on the biases. While obtained in a lengthy way using complicated expressions, they
can be approximated quite adequately by simple polynomials as in Eqs. (16). To account
for these effects one would replace, at the previous step
fs (q) fs (q) + pert fs (q),
gs (q) gs (q) + pert gs (q).
(18)
The thus corrected photon energy moments can then be confronted with the data.
Here we illustrate the procedure using the central values of the heavy quark parameters
[24] very close to those reported by BaBar [2]
mb (1 GeV) = 4.61 GeV,
3
D
(1
2 (1 GeV) = 0.41 GeV2 ,
GeV) = 0.17 GeV .

3
(19)
(20)
Likewise we use the central number for the chromomagnetic value inferred from the
B B mass difference, and we fix the one for the LS term to its theoretical expectation:
2G (1 GeV) = 0.35 GeV2 ,
3
LS
= 0.15 GeV3 .
5 The numerical difference appears insignificant, though.
(21)
385
For Ecut = 2.0 GeV we have x = 0.8677 and for the perturbative contributions we find
from Eqs. (A.5), (A.6), (A.10)(A.12) and (A.18)

1
+ a 1 (0.8677) = 0.0511,
a1 0.8677;
4.61

1
b1 0.8677;
+ b1 (0.8677) = 0.0909,
4.61

1
+ a 2 (0.8677) = 0.00447,
a2 0.8677;
4.61

1
+ b2 (0.8677) = 0.00538;
b2 0.8677;
(22)
4.61
using s = 0.22 and 0 = 9 yields

E pert E >2.0 GeV = 2.328 GeV,
pert
2

= 0.00390 GeV2 .
E E 2
E >2.0 GeV
(23)
The nonperturbative power corrections given by Eqs. (8) and (9) are
power
2
E E 2
= 0.0253 GeV2 .
E power = 0.00715 GeV,
(24)
Adding the perturbative corrections and these power corrections results in

2
biased
E |biased
E E 2.0 GeV = 0.0292 GeV2 .
2.0 GeV = 2.321 GeV,
(25)
The last step is to correct for the biases. For our set of parameters, Eqs. (12) yield
m
b = 4.602 GeV or = 0.677 GeV and, using r3 = 0.75, we obtain 2 = 0.330GeV2 .
2
At Ecut = 2.0 GeV the value of the effective normalized hardness is then q = Q/
2
2
1.05 and we have r = / 0.72. We take the scaling form Eqs. (17) for the biases
parametrized in Eqs. (15), and apply the shifts Eqs. (16) due to the perturbative effects
according to Eqs. (18)

b = 2 fs (q) + pert fs (q) = 0.041 GeV,
m

2 = 2 gs (q) + pert gs (q) = 0.111 GeV2 .
(26)
We arrive at the final bias-corrected values of the moments combining Eqs. (25) and
(26) according to Eqs. (10):
b
m
= 2.342 GeV,
2
2

2

2
biased
E E 2.0 GeV = E E 2.0 GeV = 0.020 GeV2 .

(27)
12
If we follow the similar procedure for other cuts, we find, for the biased average photon
energy and the variance, the following values for Ecut = 1.8, 1.9, 2.0 and 2.1 GeV:

2
biased
E |biased
E E 1.8 GeV 0.0357 GeV2 ,
1.8 GeV 2.305 GeV,
E |2.0 GeV = E |biased
2.0 GeV +
386
E |biased
1.9 GeV 2.313 GeV,
E |biased
2.0 GeV 2.321 GeV,
E |biased
2.1 GeV 2.329 GeV,

2
biased
E E 1.9 GeV 0.0321 GeV2 ,

2
biased
E E 2.0 GeV 0.0293 GeV2 ,

2
biased
E E 2.0 GeV 0.0271 GeV2 .
(28)
The bias corrections read

b |1.8 GeV 0.015 GeV,
m
b |1.9 GeV 0.024 GeV,
m
b |2.0 GeV 0.041 GeV,
m
b |2.1 GeV 0.070 GeV,
m

2
0.0574 GeV2 ,
1.8 GeV

2
0.0790 GeV2 ,
1.9 GeV

2
0.111 GeV2 ,
2.0 GeV

2
0.152 GeV2 .
2.1 GeV
(29)
Then we arrive at the following final, i.e., bias-corrected predictions for the average and
variance,

2

E |1.8 GeV 2.312 GeV,
E E 1.8 GeV 0.0309 GeV2 ,

2

E |1.9 GeV 2.325 GeV,
E E 1.9 GeV 0.0255 GeV2 ,

2

E |2.0 GeV 2.342 GeV,
E E 2.0 GeV 0.020 GeV2 ,

2

E |2.1 GeV 2.364 GeV,
(30)
E E 2.1 GeV 0.0145 GeV2 .
These numbers, in principle, can be compared with some published measurements
(a more detailed analysis of the Belle data has also been performed6 ):
E |1.8 GeV 2.292 0.026 0.034 GeV

2

E E 1.8 GeV 0.0305 0.0074 0.0063 (GeV)2
Belle [25],
E |2.0 GeV 2.346 0.032 0.011 GeV

2

E E 2.0 GeV 0.0226 0.0066 0.0020 (GeV)2
CLEO [26],
Belle [25],
CLEO [26].
(31)
In view of the uncertainties in the experimental numbers and their high correlations we
refrain from detailed comparison; yet there appears a clear trend that the bias corrections
bring the predictions into close agreement with the data, in particular for the variance.
We want to emphasize the following features in the estimated values for biases:
The biases increase steeply with the energy cuts.
While for Ecut = 1.8 GeV they are rather insignificant, for Ecut 2.0 GeV they become comparable to or even larger than the accuracy with which the corresponding
parameters have been be extracted from B Xc .
The uncertainties inherent in the evaluation of biases remain rather small for mb , and
even for 2 they are of moderate size.
6 O. Buchmueller, private communication.
387
4. Effect of the four-quark operators at order s /m3b

i s), (s i b) do not contribute
As noted previously [17,21,27], four-quark operators (b
to b s + at tree level, unlike for b u. Yet hard gluon emission induces Wilson
coefficients for them already at order s . Here we give their contributions. Similar shortdistance corrections affect, in general, Wilson coefficients of all other operators, including
those determining the moments of the light-cone distribution function. The four-quark effects, however are usually numerically enhanced, at least where they are allowed at the tree
level.
To order s the effect is expressed in terms of two expectation values

1
1
SV = ik lj ij lk
B|b i bk s l s j |B,
3
2MB

1
1
SA = ik lj ij lk
(32)
B|b i 5 bk s l 5 s j |B,
3
2MB
where the i, j , k, l denote color indices, and yields the following contributions to the first
three moments:

2
s SV
1 SA
,
ss E E =
4 mb 3 mb

SV
s
SA
ss E =
2 +3 2 ,
2
mb
mb

1
SV
11 SA
(33)
;
s s BX = 2s
+
BX
3 m3b
m3b
the appropriate value of s here is about 0.3. SV ,A are related to the nonvalence four-quark
parameters V ,A , V ,A of Ref. [28] through
8
2
SV = V V ,
9
3
8
2
SA = A A .
9
3
(34)
It should be noted that the contributions from the four-quark operators are 1/m3b suppressed only in the total width. In the higher moments an enhancement arises over naive
expectations one might infer considering the fact that the four-quark operators contribute
to the spectrum only in the end-point region where E m2b hadr , and that their overall
effect on the width is 1/m3b . The diagrams with a gluon exchanged between the light quark
legs generate additional powers of = mb 2E in the denominator in the forward scattering amplitude; this directly affects the higher moments, but not the total rate. Accordingly,
we find the corrections to the rate negligible, yet not necessarily so for E and, in partic i s)(s i b) can
ular, for the second moment determining 2 . Since contributions from (b
arise only via nonvalence configurations in the B meson (for the KM favored modes), they
must be somewhat suppressed; yet the degree of suppression is uncertain.
The impact on the second moment and, therefore the related shift in the apparent value
of 2 could a priori be significantif the generic nonvalence color-straight operators (in
the terminology of Ref. [28]) are estimated through the vacuum chiral condensate s s0
388
according to
1
1
B|b i bi s k s k (0)|B
B|b i bi (0)|Bs s0 (200 MeV)3 .
(35)
2MB
2MB
A dedicated discussion of the nonfactorizable four-fermion expectation values can be
found in Ref. [28]. It turns out, though that the value of SV is expected to be particularly
suppressed compared to the estimate (35). SV , to leading order in 1/Nc , reduces to the
color-straight vector operator, nonvalence expectation values for which (giving the s-quark
number density at the origin) are probably small, since the density must vanish identically
when integrated over the whole volume. The 1/Nc color-octet piece of SV contributing to
3 , is probably larger, yet still suppressed.
the Darwin expectation value D
A larger value is expected for the spin-dependent axial matrix element SA , for which
Eq. (35) can be conservatively taken as an upper bound. Yet the coefficient for SA is
appreciable only in E and in the total width where the effects are 1/m2b - and 1/m3b suppressed, respectively. Therefore we think that the related corrections in E are below
10 MeV in mb and safely at the sub-percent level for the total width.
5. Cut-related uncertainties in b c
Lower cuts imposed on the charged lepton energy in B Xc decays likewise degrade the hardness of the transition. Yet the final state in b c exhibits kinematics more
complex than in b s + , which makes this effect less transparent. Furthermore the dynamics is considerably more complicated as well, since the relevant distribution describing
the dependence on the charged lepton energy cut, generally is neither the light-cone distribution nor the temporal distribution describing the effects of the Fermi motion in the
small velocity (SV) kinematics. There is no reason to expect an analogous distribution to
be universal for different moments; the power corrections are probably significant in the
realistic settings, and probably vary depending on the moments as well.
Nevertheless there are some common features between B Xs + and B Xc , in
particular due to the conceptual similarity of the OPE treatment of both cases. Furthermore
the full impact of the degraded hardness might not reveal itself in the few lowest terms in
the 1/mb and s expansion that one is, in practice, limited to in the theoretical description.
As argued in Ref. [5], in this situation it is reasonable to adopt a cautious approach and
use the bias estimates for B Xs + merely to assess the potential magnitude of additional cut-related uncertainties in the theoretical predictions for the b c transitions.
Namely, the conventional OPE predictions for a particular moment, with or without the cut,
depend on the heavy quark parameters. Assigning a cut-induced uncertainty to the values
of those parameters (even though they have definite values in actual QCD), would result in
an additional variation in the theoretical predictions, which can be used as an estimate of
the scale of possible cut-related exponential pieces.
To implement this prescription in practice, we need a commensurate definition for the
hardness in both transitions and how that scale is degraded by energy cuts. We advocate
the following ansatz which is more accurate at low hardness:

2
2
vs.
Q MB Ecut Ecut + MK

Qsl MB Ecut
2
2 ;
Ecut + MD
389
(36)
equating the two yields the prescription
sl
Ecut Ecut
+
2 M2
MD
K

sl
sl
2 )
2(Ecut + (Ecut )2 + MD
(37)
for identifying equivalent energies in radiative and semileptonic decays. The resulting desl we use Eq. (37) to determine the
pendence is practically linear, Fig. 3. For a given Ecut
corresponding Ecut . This allows us to estimate the size of biases associated with mb , 2 ,
3 , . . . at this hardness. Using the explicit dependence of the semileptonic moments in
D
question on these heavy quark parameters, one can estimate the degree to which the theoretical accuracy deteriorates due to the insufficient energy release. It is interesting to note
that Refs. [29] concluded that when the lepton cut is raised to 1.7 GeV theory looses control over MX2 . According to Fig. 3 the lepton cut at 1.7 GeV corresponds to the cut on E
at about 2.1 GeVand this is just the limit where the OPE for B Xs + totally breaks
down, according to the present estimates.
A possible physical motivation behind this procedure has been mentioned in Ref. [5]:
the lower cuts both on E or on E kinematically reduce or even eliminate the contribution
of high mass hadronic final states in the decay, thus introducing an effective normalization scale Q. By virtue of the heavy quark sum rules, the integrals of the inclusive
decay probabilities yield the values of the nonperturbative heavy quark parameters, with
the upper cutoff on the mass of the produced states playing the role of the normalization
point for the operators [27]. For large enough values of the hardness, explicit perturbative
corrections account for this scale dependence, yet the possible nonperturbative dependence
at limited Q is missed. This ansatz can therefore be interpreted as accounting for a certain
nonperturbative evolution of the heavy quark parameters at relatively low normalization
point , provided it is to some extent universal, similar to the universality of the perturbative -dependence of a given nonperturbative expectation value at in the perturbative
domain.
Fig. 3. Cuts in B Xs + and in B Xc yielding the same overall hardness Q.
390
3 : the dimensionless ratio

3 /( 2 )3/2 for the two distribution functions and
Fig. 4. The expected bias for D
D
their average (blue). (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
For the sake of completeness we illustrate in Fig. 4 the expected size of the similar bias
3 in the Darwin expectation value, which following Eqs. (10) is defined (up to a factor
D
24) as the cut-related deficit in the third moment of the photon spectrum. As anticipated,
the predictions for the bias itself are less certain, yet we see that the corresponding effect
can be significant down to rather low cuts.
6. Conclusions
In the present paper we addressed in detail the theoretical evaluation of the photon
energy moments in B Xs + decays with lower cuts on the photon energy, in terms
of the underlying heavy quark parameters. These moments are particularly important in
connection to the precision determination of the latter. Two general aspects have to be
addressed in calculating the momentsperturbative contributions which act as corrections,
and the nonperturbative effects dominating the measured low moments. We have improved
the former by computing the perturbative spectrum with the explicit Wilsonian infrared
cutoff on the low-energy gluon modes. This has been accomplished analytically including
the BLM corrections.
We believe that this approach ensures the most accurate and trustworthy results for the
heavy flavor decays with the heavy quark mass in a few GeV range, as it is for the actual
beauty hadrons. Extensions of the method including resummation of the Sudakov radiation
effects would be required for much heavier quarks, with masses exceeding 20 GeV.
Adding to what has been stated in Section 2.1, an essential element of the uncertainty
in the numerical outcome of Ref. [14] is related to the unphysical, somewhat loose7 shape
function S() arrived at in SCET in place of the truly Wilsonian F (k+ ). Ref. [14] sug7 The model-independent exact results for its asymptotic behavior emphasized in Ref. [14] in fact stem from
the chosen scheme which, in some respects, is unsatisfactory itself. Thus their physical significance is doubtful.
391
Fig. 5. The significance of the higher-order BLM corrections in the Wilsonian perturbative spectrum ( = 1 GeV).
Blue curve is the first-order spectrum evaluated with s = 0.35. Yellow curve is the usual s plus 0 s2 approximation, and the (barely visible) green line is obtained by adding the 02 s3 terms. Red curve shows the result of
the full BLM resummation which can be performed precisely in the Wilsonian approach. (For interpretation of
the references to colour in this figure legend, the reader is referred to the web version of this article.)
gested that the properties of S() required for constraining its possible form, may be
related to the true Wilson heavy quark expectation values measured, say, in B Xc
decays. Yet in practice such relations suffer from poorly controlled perturbative corrections and seem too crude. It is unclear if their accuracy can realistically be improved. The
advantage of employing the heavy quark distribution function F (k+ ) is that its moments
are directly related to the extracted expectation values; thus there is no significant uncertainty here.
Accordingly, we view our calculations of the spectrum and its moments as in all respects
superior theoretically to the SCET-based evaluations of Ref. [14] and as more accurate
numerically. This conclusion is illustrated, for instance, by Fig. 5 showing the perturbative
(Wilsonian) photon spectrum in b s + . It suggests that the fraction of events with
E < 1.8 GeV is not only small, but also reliably predicted, at least in what concerns the
perturbative effects.
Doubts have been aired recently that the heavy-to-light transitions may not be treated
with the conventional OPE, even abstracting from the complications associated with the
high experimental cuts. In particular, the principal OPE results established more than a
decade ago, were challenged. These results relate the moments of, say the photon energy
spectrum to the local heavy quark expectation values which describe the nonperturbative bound-state dynamics. The OPE ensures universalityi.e., process-independenceof
these expectation values as it should be for the characteristics of the initial bound state.
Consequently, the same expectation values can be measured, say through the moments of
the B Xc distributions.
We find such agnostic claims unfounded. The recent paper [4] showed that, in spite of
the well-known peculiarities of the light-like jet processes associated with the collinear
bremsstrahlung, the perturbative corrections to the moments of the photon spectrum are
392
calculable in the usual short-distance expansion, without encountering new types of the
nonperturbative effects. Our new perturbative calculations implement this idea in practice.
We state that the photon moments can be precisely calculated in terms of the analogous expectation values determined from B Xc moments, as long as the lower experimental
cut on E is placed appropriately.
On the nonperturbative side the so-called bias effects due to significant lower cuts are in
the focus of our study. They are missed in the routinely used simplified OPE expressions,
yet turn out to be significant in practice. We showed that for actual B decays these biases
are affected very little by the perturbative corrections in the adequate Wilsonian approach
with the separation scale chosen around the usual 1 GeV. This significantly simplifies incorporating the biases in practice.
The main phenomenological message can be expressed rather concisely:
Imposing a lower cut on the photon energy in B Xs + reduces the hardness Q of
the transition.
This impacts the evaluation of the energy moments in ways that are not fully described
by the usual OPE expressions, thereby inducing systematic shifts, or biases.
With the presently available accuracy of the measurements those biases can no longer
be ignored.
While those biases cannot be evaluated in a completely model-independent way yet,
different anstze for the heavy quark distribution functions, when constrained by the
OPE, yield a consistent picture:
For Ecut 1.7 GeV they are more or less irrelevant.
For Ecut > 2.1 GeV they are in all likelihood large, yet beyond accurate theoretical
control.
In the range 1.8 Ecut 2 GeV they are significant, yet can be estimated with
reasonable accuracy. Thus they can be corrected for.
Lower cuts on the lepton energy in B Xc conceptually have a similar impact
on truncated lepton energy and hadronic mass moments. Yet the resulting biases in the
values of the moments cannot be evaluated at present. We have suggested a simple prescription to obtain semi-quantitative estimates of the potential uncertainties associated
.
with the high values of Ecut
We have presented the generally rather cumbersome theoretical evaluation of the truncated photon moments in a reasonably compact form suitable even for global fits of various
data on inclusive B decays. In our opinion, such data analyses must be done by experimentalists who can take care of numerous experimental subtleties. As the numerical application
of our analysis, we only note that we find good agreement between the recently published
Belle data on the photon spectrum with the cut as low as 1.8 GeV, and the predictions based
3 extracted by BaBar from B X [2]. The data and
on the values of mb , 2 and D
c
theory predictions coincide within 1 of only the statistical error, cf. Eqs. (30) and (31),
even without invoking other error bars or the uncertainties inherent in the theory calculations.
393
Acknowledgements
We are grateful to Th. Mannel for discussions and to our experimental colleagues
U. Langenegger, V. Luth and in particular to O. Buchmueller for invaluable exchanges.
Additional comments on the difference in our treatment with the SCET emerged after informative discussions with M. Neubert at the Belle Workshop and at KEK and his talks
there, which are gratefully acknowledged. This work was supported in part by the NSF
under grant number PHY-0087419.
Appendix A
A.1. Perturbative spectrum with the Wilsonian scale separation
In Ref. [30] it has been described in detail how the required Wilsonian separation of
soft and hard gauge field modes can be accomplished. The analysis of the semileptonic
data based on this approach [31,32] proved to be efficient and successful [2]. To order s it
of the exchanged gluon: only those
corresponds to a separation based on the energy = |k|
with > are included into the Feynman integrals for the Wilson coefficients. While in
higher orders the situation becomes considerably more involved, it can be handled for all

BLM improvements similar to the first order case with only the relation between and |k|
being modified.
Adopting this separation scheme one computes the usual one-gluon decay probability,
). A massless gluon in the decay necessarily has |k|
< mb ,
yet with the extra factor (|k|
2
mb
hence taking > 2 would eliminate all bremsstrahlung. In what follows we will always
assume 2 < mb ; in practice, with around 1 GeV we have mb 0.22. The integration
over the thus restricted phase space is straightforward and yields
d pert ( ; )
d

s
s
(1 ) + A(1 ) + CF B1
(1 ),
= CF f 1 ;
mb
mb
pert
(A.1)
where the second term stands for the tree-level two-body decay modified by virtual corrections. Its strength A is determined by the condition that the normalized spectrum in
the l.h.s. integrated over the whole domain from 0 1, is unity. The last term with
(1 ) can be viewed as the reflection of the fact that the kinetic mass mb () in the
presence of perturbative corrections differs from the energy of the quark at rest (a rest
energy mass). The latter determines the perturbative end-point, in particular the photon
energy in the two-body decay:
B1 () =
4 2 1
+
ln +
3
2
2

1 + 2 + 1 + 2 .
(A.2)
394
The one-loop expression for the continuous part of the spectrum is

3 2
2
2 3 6 + ( 3) ln(1 ),
4(1 )
2(1 )
f1 ( ; ) =
(12)(4 2 22 +4 2 6 +2 2) + ( 2 3) ln(2),
4(1 )
2(1 )
1 2,
> 1 2.
(A.3)
These allow to readily calculate the required moments of the perturbative spectrum:
0 (x; )
1
1 d pert ( ; )
=
d
pert
d
x

2 +3x+30)
2
1 CF s x(2x 12
+ (x 2x+10)
ln(1 x) + 12 ln2 (1 x) ,
for x 1 2,
12
= 1 CF s 12 (6x 2 122 x + 6x 6x 83 + 22 + 19 + 6),

2
2
2 +8+7
+ 4 8+44 ln(2)+3 ln(1 x) + x 2x4

ln(2) 12 ln2 (2) ,
for x > 1 2,
(A.4)
CF s a1 (x; )
1
d pert ( ; )
(1 ) d
pert
d
x
1
CF s 72
(484 643 + 722 72 9x 4 + 22x 3 + 60x 2 96x + 23)
+ (1x)(x +x8) ln(1 x) B1 () , for x 1 2,

6
=
(A.5)
s (1x)(12)
2 + 62 x 5x + 3x 8 3)
C
(4x
12
(1x)(x 2 +x8)
ln(2) B1 () , for x 1 2,
+
6
CF s a2 (x; )
1
d pert ( ; )
(1 )2 d
pert
d
x
1
CF s 1440
(7685 7204 + 6403 4802 + 144x 5 495x 4 340x 3

2
2
+ 1650x 2 1020x + 61) + (1x) (3x24+2x17) ln(1 x) ,
for x 1 2,
=
(A.6)
(1x)2 (12)
2
2
2
s
(6x + 8 x 8x + 4x 2 7 4)
CF
24
2
2
+ (1x) (3x24+2x17) ln(2) , for x 1 2.
395
The usually considered BLM corrections can easily be calculated to arbitrary order
with merely technical modifications. The corresponding technique has been reviewed in
Refs. [30,33]. First the one-gluon decay rate must be calculated with a non-vanishing fictitious gluon mass mg reflecting the gluon conversion into the q q or gg pairs. This modifies
) is replaced by ( ) = (k2 + m2g 2 ). The integrals
the cutoff condition: (|k|
are still easily taken. All the functions additionally depend now on the new dimensionless parameter
m2g
m2b
(note that in Ref. [30] this ratio was denoted by 2 ). Generalizing
Eq. (A.1) we then set
d pert (, mg ; )
pert
d
s
s
= CF f1 (, ; ) (1 ) + A(1 ) + CF B1 (; ) (1 )
(recall that <
(A.7)
1
2
is assumed; otherwise the kinematic constraints are modified), with

3

(2 ) 2
4 2
+ arcsin
+
B1 (; ) =
1 2
3
2
2
4

1
(2 ) (4 )
4

2

( 2) 2
arctan
arctan
(4 )
(4 )(2 + 1)

1
(4 ) 2 ln + 2 ln
8

2 4 + 2 2
ln + 1 + 2
4

2 2
+1 .
(A.8)
2
To calculate the BLM corrections, we must integrate over the gluon mass. The upper
end of the -independent lower region of the spectrum now rises depending on : the
perturbative spectrum is independent of if > 2(1 ) (1 )2 :
f1 ( ; ; ) =
( + 1)
4(1 )3 ( 2 2 + + 1)

2 3 2 3 2 2 + 2 2 + 4 4 14 3 + 17 2 11
2 2 + 2 3
+ 4 + 2 5 9 4 + 9 3 + 7 2 15 + 6 +
2(1 )

at 2(1 ) (1 )2
ln 1 +
1
(2 1)
=
4(1 )2 (2 )

2 2 2 + 2 2 4 3 + 12 2 + 4 9 +
396
+ 8 3 + 82 2 20 2 + 4 2 122 + 10

8 + 42 + 2 + 4
+
2 2 + 2 3
ln(2 )
2(1 )
at 2(1 ) (1 )2 .
(A.9)
The above -dependent functions replace the usual one-loop ones of Eqs. (A.1)(A.3).
Using Eqs. (A.8), (A.9) it is not difficult to obtain the BLM corrections both to the spectrum
and to its moments. For instance, for the continuous part of the spectrum, at 1 2
we integrate the upper of the two expressions in Eq. (A.9) over directly from = 0 to
= 1 . At > 1 2 the integral from = 0 to = (1 )( 1 + 2) uses the lower
expression, while the integrand in the integral from = (1 )( 1 + 2) to = 1
is the -independent upper expression. The subtraction piece employing = 0, of course
is integrated over from 0 to , cf. Eq. (A.12).
Representing the spectrum as
d pert ( ; )
pert
d

s
s
+ 0 f2 ;
+ (1 ) + A(1 )
f1 ;
= CF
mb
mb

s
s
+ 0 B2
+ (1 ),
B1
+ CF
mb
mb
1
(A.10)
f1 and B1 are given in Eqs. (A.3) and (A.2), respectively, while for the second-order BLM
parts we obtain
f2 ( ; )
38 3 93 2 + 6 36 6 4 31 3 + 24 2 30 + 18
=
ln(1 )
96(1 )
48(1 )

3( 2 3)
2
ln2 (1 ) + Li2 ( ) , 1 2,
16(1 )
3
=
6 3 (1282 16) 2 (1283 2442 + 52 20)

96(1 )
3
2
52 + 12 15 20
+
96(1 )
2
(8 4) 2 + (83 162 + 10 2) 43 3 + 2
+
16(1 )

2 3 3 2 6
ln(1 ) + ln(2 + 1) +
ln(2 + 1)
16(1 )
10 2 2(122 + 3 + 3) + 242 24 + 6
ln(2)
+
48(1 )
4 3 + 2(42 6 + 1) 2 2(42 9 + 1) 6
+
16(1 )
397

ln 2 + 2 2 + 1

( 2 3)
ln(1 ) ln(2 + 1)
+
8(1 )
1
ln(1 ) ln(2) ln(2) ln(2 + 1) ln2 (1 ) Li2 ( )
2

(1 )(2 + 1)
(2 + 1)
+ Li2
Li2
, > 1 2.
2
1
The BLM correction to the coefficient of the
integrating over
1
B2 () =
4

0
(1 )
(A.11)
can be obtained by numerically

d B1 (; 0)
B1 (; ) .
1 + e5/3
(A.12)
The complete -dependent analytic expressions for the BLM corrections to the moments contain polylog functions at worst, but are somewhat cumbersome, and we do not
quote them here. At = 0 the order-s spectrum and its second-order BLM correction
reproduce the standard expressions [13]
f1 ( ; 0) =
f2 ( ; 0) =
( 2 3)
2 3 3 2 6
+
ln(1 ),
4(1 )
2(1 )
(A.13)
38 3 93 2 + 6 36 6 4 31 3 + 24 2 30 + 18
ln(1 )
96(1 )
48(1 )

3( 2 3)
2
(A.14)
ln2 (1 ) + Li2 ( ) .
16(1 )
3
Their moments, ak (x; 0) and bk (x; 0), respectively are given by

x
f1 ( ; 0) d
0
=
x
x(2x 2 + 3x + 30) (x 2 2x + 10)

1
+
ln(1 x) + ln2 (1 x),
12
4
2
f2 ( ; 0) d
0
14x 3 + 41x 2 + 120x 48 (3) 1

+ Li2 (x) ln(1 x)
96
4
x 2 + 2x + 6
4x 3 31x 2 44x 8 2 + 156
ln(1 x) +
Li2 (x)
+
96
16
1
1
9x 2 + 18x 40 + 24 ln (x) 2
ln (1 x) ln3 (1 x) + Li3 (1 x),
+
96
8
2
398
1
f1 ( ; 0)(1 ) d
x
(1 x)(x 2 + x 8)
1
9x 4 + 22x 3 + 60x 2 96x + 23 +
ln(1 x),
72
6
1
f2 ( ; 0)(1 ) d
x
408 2 369x 4 + 1360x 3 66x 2 + 1380x 2305 x 3 9x 9

+
Li2 (x)
3456
24
3

x2 + x 8 2
9x 61x 2 7x 7
ln(1 x) +
ln (1 x) ,
(1 x)
288
16
1
f1 ( ; 0)(1 )2 d
x

1
144x 5 495x 4 340x 3 + 1650x 2 1020x + 61
1440
(1 x)2 (3x 2 + 2x 17)
ln(1 x),
+
24
1
f2 ( ; 0)(1 )2 d
x
7950 2 + 7272x 5 33435x 4 + 35320x 3 25995x 2 + 91110x 74272

86400
4
3
2
3x + 4x + 18x 36x 36
Li2 (x)
+
96
(1 x)2 (3x 2 + 2x 17) 2
ln (1 x)
48
(1 x)(72x 4 528x 3 + 562x 2 473x + 1957)
ln(1 x).
+
(A.15)
2880
Finally, we provide simple numerical expressions for the first and second moments of
the second order BLM correction with /mb fixed at the value 1/4.6:

1

1
1
1
b1 x;
= B2
+ f2 ;
(1 ) d
4.6
4.6
4.6
x

0.032 0.096y 0.67y 2 + 1.6y 3 3.2y 4
=
0.032 0.11y 0.33y 2 0.47y 3 0.26y 4
y x (1 2/4.6),
for y 0,
for y 0,
(A.16)
399

1

1
1
= f2 ;
(1 )2 d
b2 x;
4.6
4.6
x

0.0225 0.051y 0.066y 2 + 0.040y 3 + 0.24y 4
=
0.0225 0.050y 0.12y 2 0.12y 3 0.04y 4
for y 0,
for y 0.
(A.17)
They are sufficiently accurate in the whole relevant domain of x between 0.65 and 0.93.
The evaluation for different values of mb can easily be recovered from the fact that the
above integrals are functions of the ratio /mb , and at fixed mb the -dependence is well
described by Eqs. (7).
A.2. Non-O7 perturbative corrections
The non-O7 spectrum was given in Ref. [13]. We have interpolated the moments of
mc
these contributions for m
= 1.18
4.61 :
b

C22
0.053 0.81(x 0.87) + 2.93(x 0.87)2
2
C7

C2
+
0.026 + 0.25(x 0.87) + 0.29(x 0.87)2
C7

C8
+
0.56 7.5(x 0.87) + 21(x 0.87)2 ,
C7
2

C
100b1 (x) = 22 0.064 0.86(x 0.87) + 2.25(x 0.87)2
C7

C2
+
0.0049 0.25(x 0.87) + 2.18(x 0.87)2
C7

C8
+
0.72 8.8(x 0.87) + 18(x 0.87)2 ,
C7
2

C
100a 2 (x) = 22 0.0023 0.055(x 0.87) + 0.38(x 0.87)2
C7

C2
+
0.0010 + 0.015(x 0.87) 0.0413(x 0.87)2
C7

C8
+
0.047 1.1(x 0.87) + 6.5(x 0.87)2 ,
C7

C2
100b2 (x) = 22 0.005 0.12(x 0.87) + 0.71(x 0.87)2
C7

C2
+
0.0007 0.042(x 0.87) + 0.42(x 0.87)2
C7

C8
+
0.06 1.2(x 0.87) + 6.75(x 0.87)2 .
C7
100a 1 (x) =
(A.18)
400
Since their effect is small and well below a host of the corrections left out, closer attention
would seem superfluous.
A.3. Tables for biases
Here we tabulate the values for biases obtained with the purely nonperturbative distribution functions, as referred to in Sections 3.1 and 3.3. To make Tables 1 and 2 compact,
we present biases in the scaling form (14), evaluated near the expected value r = 0.75. The
two primary entries are the averages of fs (q) (first moment) or of gs (q) (second moment)
obtained with the two distribution functions F1 and F2 in Eqs. (13). Their interpolating
expressions are given in Eqs. (15). To show and/or to correct for the residual dependence
on r the third and the fourth entries, Df (q) and Dg (q), respectively, linearly approximate
this dependence of f and g:
fs (q) fs (q) + Df (q)(r 0.75),
gs (q) gs (q) + Dg (q)(r 0.75)
(A.19)
(these dependences are approximate and were derived by varying r by 0.25). Finally, the
b and
differences between the two anstze are illustrated by the last two columns, for m
Table 1
The estimated values for the bias corrections around r = 0.75
q
fs
gs
Df
Dg
F12
G12
0
0.5
0.75
1.0
1.25
1.5
1.75
2.0
2.5
3.0
0.413
0.208
0.137
0.0854
0.0507
0.0287
0.0156
0.00829
0.00225
0.000606
0.800
0.577
0.452
0.333
0.231
0.151
0.0938
0.0560
0.0187
0.00597
0.0337
0.00453
0.0141
0.0179
0.0175
0.0146
0.0110
0.00769
0.00329
0.00129
0.0920
0.107
0.112
0.112
0.105
0.0915
0.0743
0.0566
0.0288
0.0132
0.0151
0.00312
0.00222
0.00571
0.00712
0.00684
0.00561
0.00408
0.00168
0.000553
0.0209
0.0325
0.0389
0.0440
0.0460
0.0436
0.0375
0.0294
0.0143
0.0055
Table 2
The similar bias corrections for r around 1.0
q
fs
gs
Df
Dg
F12
G12
0
0.5
0.75
1.0
1.25
1.5
1.75
2.0
2.5
3.0
0.405
0.209
0.140
0.0899
0.0551
0.0324
0.0184
0.0102
0.00307
0.000930
0.823
0.603
0.480
0.361
0.257
0.174
0.112
0.0702
0.0259
0.00927
0.0311
0.00290
0.0116
0.0153
0.0153
0.0131
0.0101
0.00729
0.00332
0.00141
0.0760
0.0932
0.0988
0.0999
0.0951
0.0845
0.0701
0.0548
0.0296
0.0146
0.0165
0.00454
0.00120
0.00522
0.00717
0.00733
0.00633
0.00487
0.00225
0.000840
0.0184
0.0321
0.0399
0.0465
0.0501
0.0492
0.0440
0.0361
0.0195
0.0084
401
2 , respectively,
fs1,2 (q) = fs (q) F12 (q),
gs1,2 (q) = fs (q) G12 (q),
(A.20)
where the upper and lower signs refer to the biases for F1 and F2 , respectively.
We also give the similar table around a different value of r = 2 / 2 = 1.0; in this case
(r 0.75) in Eqs. (A.19) must be replaced by (r 1).
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]
M. Battaglia, et al., Phys. Lett. B 556 (2003) 41.

BaBar Collaboration, B. Aubert, et al., Phys. Rev. Lett. 93 (2004) 011803.
CLEO Collaboration, M.R. Shepherd, hep-ex/0405045.
N. Uraltsev, Phys. Lett. B 603 (2004) 203.
I.I. Bigi, N. Uraltsev, Phys. Lett. B 579 (2004) 340.
M. Shifman, in: M. Shifman (Ed.), Boris Ioffe Festschrift at the Frontier of Particle Physics/Handbook of
QCD, World Scientific, Singapore, 2001.
C. Bauer, Phys. Rev. D 57 (1998) 5611;
C. Bauer, Phys. Rev. D 60 (1999) 099907.
G. Altarelli, et al., Nucl. Phys. B 208 (1982) 365.
A. Kagan, M. Neubert, Eur. Phys. J. C 7 (1999) 5.
N. Uraltsev, in: M. Shifman (Ed.), Boris Ioffe Festschrift at the Frontier of Particle Physics/Handbook of
QCD, vol. 3, World Scientific, Singapore, 2001, p. 1577, hep-ph/0010328.
M. Battaglia, et al., DELPHI 2004-021 CONF 696.
A.J. Buras, A. Czarnecki, M. Misiak, J. Urban, Nucl. Phys. B 631 (2002) 219;
M. Misiak, M. Steinhauser, Nucl. Phys. B 683 (2004) 277.
Z. Ligeti, M. Luke, A. Manohar, M. Wise, Phys. Rev. D 60 (1999) 034019.
M. Neubert, hep-ph/0408208;
M. Neubert, hep-ph/0408179.
T. Mannel, F.J. Tackmann, hep-ph/0408273.
M. Beneke, Y. Kiyo, D.S. Yang, Nucl. Phys. B 692 (2004) 232.
I.I. Bigi, M. Shifman, N.G. Uraltsev, A. Vainshtein, Int. J. Mod. Phys. A 9 (1994) 2467.
R.D. Dikeman, M. Shifman, N.G. Uraltsev, Int. J. Mod. Phys. A 11 (1996) 571.
F.E. Low, Phys. Rev. 110 (1958) 974;
T.H. Burnett, N.M. Kroll, Phys. Rev. Lett. 20 (1968) 86.
V.N. Gribov, Sov. J. Nucl. Phys. 5 (1967) 280, Yad. Fiz. 5 (1967) 399.
I.I. Bigi, N.G. Uraltsev, Int. J. Mod. Phys. A 17 (2002) 4709.
N.G. Uraltsev, Mod. Phys. Lett. A 10 (1995) 1803.
M. Neubert, Phys. Rev. D 46 (1992) 2212;
M. Neubert, Phys. Rep. 245 (1994) 259.
O. Buchmuller, private communication.
Belle Collaboration, P. Koppenburg, et al., Phys. Rev. Lett. 93 (2004) 061803.
CLEO Collaboration, S. Chen, et al., Phys. Rev. Lett. 87 (2001) 251807.
I.I. Bigi, M. Shifman, N.G. Uraltsev, A. Vainshtein, Phys. Rev. D 52 (1995) 196.
D. Pirjol, N. Uraltsev, Phys. Rev. D 59 (1999) 034012.
N. Uraltsev, Talk at Workshop on the CKM Unitarity Triangle, IPPP Durham, April 2003, eConf C0304052
(2003) WG118, hep-ph/0306290;
N. Uraltsev, Talk at Int. Conference Flavor Physics & CP Violation 2003, Paris, 36 June 2003,
eConf C030603 (2003) JEU05, hep-ph/0309081.
D. Benson, I. Bigi, Th. Mannel, N. Uraltsev, Nucl. Phys. B 665 (2003) 367.
P. Gambino, N. Uraltsev, Eur. Phys. J. C 34 (2004) 181.
N. Uraltsev, hep-ph/0403166, Int. J. Mod. Phys. A, in press.
N. Uraltsev, Int. J. Mod. Phys. Lett. A 17 (2002) 2317.
Study of the eightfold degeneracy with a standard

-beam and a super-beam facility
A. Donini a , E. Fernndez-Martnez a , P. Migliozzi b , S. Rigolin a ,
L. Scotto Lavina c
a I.F.T. and Departamento Fsica Terica, U.A.M., E-28049 Madrid, Spain
b I.N.F.N., Sezione di Napoli, I-80126 Napoli, Italy
c Dipartimento Fisica, Universit di Napoli Federico II and I.N.F.N., I-80126 Napoli, Italy
Received 28 September 2004; received in revised form 12 November 2004; accepted 22 December 2004
Abstract
The study of the eightfold degeneracy at a neutrino complex that includes a standard -beam and
a super-beam facility is presented for the first time in this paper. The scenario where the neutrinos are
sent toward a megaton water Cerenkov

detector located at the Frjus laboratory (baseline 130 km)
is exploited. The performance in terms of sensitivity for measuring the continuous (13 and ) and
discrete (sign[m223 ] and sign[tan(223 )]) oscillation parameters for the -beam and super-beam
alone, and for their combination has been studied. A brief review of the present uncertainties on the
neutrino and antineutrino cross-sections is also reported and their impact on the discovery potential
discussed.
PACS: 13.15.+g; 14.60.Pq; 14.60.Lm
1. Introduction
In the past years the hypothesis of neutrino oscillations has been strongly confirmed in
the atmospheric [1], accelerator [2], solar [3] and reactor [4] sectors. If we do not conE-mail address: pasquale.migliozzi@cern.ch (P. Migliozzi).
doi:10.1016/j.nuclphysb.2004.12.029
A. Donini et al. / Nuclear Physics B 710 (2005) 402424
403
sider the claimed evidence from the LSND experiment [5], that must be confirmed or
excluded by the ongoing MiniBooNE experiment [6], oscillations in the leptonic sector
can be accommodated in the three family PontecorvoMakiNakagawaSakata (PMNS)
mixing matrix [7]. Therefore, the next steps on the way of a full understanding of neutrino
oscillations are:
confirm the source of atmospheric neutrino oscillations, i.e., observe directly the
oscillation;
perform precision measurements of the angles 12 and 23 and of the mass differences
|m212 | and |m223 |;
measure the sign of the atmospheric mass difference, m223 ;
measure the remaining parameters of the PMNS mixing matrix: 13 (for which only
an upper limit exists so far [8]) and the leptonic CP-violating phase (that is still
completely unknown).
With the aim to perform the above measurements, over recent years there has been
a marked growth of interest in the development of new neutrino sources: conventional
neutrino beams from pion and kaon decay, but with a more intense flux (super-beams);
neutrino beams from muon decays (neutrino factories); neutrino beams from the decay of
intense beams of -unstable heavy ions (-beams). For a comprehensive review of future
neutrino sources we refer to [9] and references therein.
In this paper we focus on a CERN-based neutrino complex including a -beam, that
could leverage existing facilities at CERN and complement the EURISOL physics program [10], and a super-beam based on an intense proton driver (the SPL). Although the
possibility to exploit higher -beams has been put forward (see for example [11,12]),
we consider here only the configuration of the -beam where e ( e ) are produced by 18 Ne
(6 He) ions [13] that are accelerated by the SPS up to 100 ( 60), respectively (standard -beam). These values have been chosen in order to tune the neutrino/antineutrino
mean energy in such a way that the maximum of the atmospheric neutrino oscillation lies
at a distance of around 100 km, i.e., the distance from CERN to Frjus.
A first estimate of the potentiality of a CERN to Frjus based neutrino complex was
given in Ref. [13]. However, in that study only the intrinsic degeneracy1 was taken into
account and only the very peculiar value = 90 was considered. In Ref. [11] a more
thorough analysis for both experimental and theoretical issues appears, although again
only positive m223 and 23 = 45 were considered, the main interest of the paper being
the study of higher setups.
The super-beam envisaged at the CERN neutrino complex studied in this paper is based
on the planned SPL of 4 MW power described in Ref. [14]. In that paper, a comprehensive analysis of the CERN-based super-beam potential is presented (although based on
old fluxes and considering only the intrinsic degeneracy), with the super-beam average
neutrino energy fixed to approx. 0.25 GeV to match the CERNFrjus distance. Similar
1 I.e. the sign of the atmospheric mass difference m2 was assumed to be positive and = 45 .
23
23
404
projects have been also proposed in Japan and USA, and carefully studied by several authors [15], with neutrinos energies around 12 GeV.
In this paper we study for the first time the complete eightfold degeneracy for the CERN
neutrino complex, i.e., the -beam and the super-beam either separately or together. In
Section 2, the neutrino oscillation formalism is introduced and the eightfold degeneracy
discussed from a theoretical point of view. The main features of the neutrino complex
at CERN are discussed in Section 3, while the set of cross-sections used in this paper are
briefly summarized in Section 4. In particular, we compared these cross-sections with other
calculations and briefly comment the present status of the cross-section knowledge for neutrino energies below 1 GeV. In Section 5 we give our results on the sensitivity calculation
both for the 13 angle and the CP violating phase , and in Section 6 we eventually draw
our conclusions.
2. The eightfold degeneracy

In Ref. [16] it was pointed out that the appearance probability P obtained for neutrinos
has no unique solution. Indeed,
at a fixed energy and baseline with input parameter (13 , )
the equation
= P (13 , )
P (13 , )
(1)
has a continuous number of solutions. The loci of the (13 , ) plane satisfying this equation
are called equiprobability curve. Considering the equiprobability curves for neutrinos
and antineutrinos with the same energy (and the same input parameters), the following
system of equations ( referring to neutrinos and antineutrinos respectively)

= P (13 , )
(13 , )
P
(2)
and a second, energy dependent, point. This

has two intersections: the input pair (13 , )
second intersection introduces an ambiguity in the measurement of the physical values of
13 and : the so-called intrinsic clone solution. Knowing the two probabilities of Eq. (2)
is consequently not enough for solving the intrinsic degeneracy. One needs to add more
information.
Unfortunately, the appearance of the intrinsic degeneracy is only a part of the clone
problem. As it was made clear in [1719], two other sources of ambiguities arise due to the
present (and near future) ignorance of the sign of the atmospheric mass difference, satm =
sign[m223 ], and of the 23 octant, soct = sign[tan(223 )]. These two discrete variables
assume the values 1, depending on the physical assignments of the m223 sign (satm = 1
for m23 > m22 and satm = 1 for m23 < m22 ) and of the 23 -octant (soct = 1 for 23 < /4 and
soct = 1 for 23 > /4). As a consequence, future experiments will have as ultimate goal
the measurement of the two continuous variables 13 and plus the two discrete variables
satm and soct .
Moreover, it should be noticed that experimental results are not given in terms of oscillation probabilities but of number of charged leptons observed in a specific detector. It
has been noticed [20] that clones location calculated starting from the probability or the
405
number of events can be significantly different. We must therefore integrate the oscillation
probability over the neutrino flux, the N cross-section and the detector efficiency. From
these considerations it follows that Eq. (2) should be more correctly replaced by:
satm , soct ) = N (13 , ; satm = satm ; soct = soct )
N (13 , ;
(3)
(with the final lepton). In Eq. (3) we have implicitly assumed to know the right sign
and the right octant for the atmospheric mass difference and angle. As these quantities are
unknown (and presumably they will still be unknown at the time of the neutrino facilities
considered in this paper) the following systems of equations should be considered as well:
satm , soct ) = N (13 , ; satm = satm , soct = soct ),
N (13 , ;
(4)
satm , soct ) = N (13 , ; satm

N (13 , ;
satm , soct ) = N (13 , ; satm

N (13 , ;
= satm , soct = soct ),
(5)
= satm , soct = soct ).
(6)
Solving the four systems of Eqs. (3)(6) will result in obtaining the true solution plus additional clones to form an eightfold degeneracy [19]. These eight solutions are respectively:
the true solution and its intrinsic clone, obtained solving the system in Eq. (3);
the m223 -sign clones (hereafter called sign clones) of the true and intrinsic solution,
obtained solving the system in Eq. (4);
the 23 -octant clones (hereafter called octant clones) of the true and intrinsic solution,
obtained solving the system in Eq. (5);
the m2atm -sign 23 -octant clones (hereafter called mixed clones) of the true and intrinsic solution, obtained solving the system in Eq. (6).
A complete description of the clone location has been done in [20] and we refer to that
article for all the theoretical details. In this paper we are interested in presenting a detailed
analysis of a concrete experimental facility, that we are going to describe in the following
sections.
3. Neutrino beam facilities at CERN

In this section we summarize some of the technical details of the two considered neutrino beams, the standard -beam (Section 3.1) and the super-beam (Section 3.2). Both
beams are directed from CERN toward the underground Frjus laboratory, where it has
been proposed to locate a 1 megaton UNO-like [21] water Cerenkov

detector with a
440 kton fiducial mass. The considered baseline is thus L = 130 km for both beams. Therefore, in order to be at the maximum of the atmospheric neutrino oscillations, the peaks of
the energy spectra have been chosen of order few hundred MeV.
3.1. The -beam
The -beam concept was first introduced in Ref. [22]. It involves producing a beam of
-unstable heavy ions, accelerating them to some reference energy, and allowing them to
406
Fig. 1. -beam fluxes at the Frjus location (130 km baseline) as a function of the neutrino energy for the two
specific values shown in the legend.
decay in the straight section of a storage ring, resulting in a very intense neutrino beam.
Two ions have been identified as ideal candidates: 6 He, to produce a pure e beam, and
18 Ne, to produce a beam. The golden [23] sub-leading transitions and
e
e
e
can be measured through the appearance of muons in a distant detector.
The neutrino beam energy depends on the of the parent ions in the decay ring. For
the scenario considered in this paper the ratio for the two ions has been fixed to be
(6 He)/ (18 Ne) = 3/5 [13]. This constraint comes from the request to accelerate in
the same accelerator at the same time ions with different atomic mass. The optimal
values that match the CERNFrjus distance have been found to be (6 He) = 60 and
(18 Ne) = 100. The mean neutrino energies of the e , e beams corresponding to this configuration are 0.23 and 0.37 GeV, respectively. On the other hand the energy resolution is
very poor at these energies, given the influence of Fermi motion and other nuclear effects.
Therefore, in the following all the sensitivities are computed for a counting experiment
with no energy cuts.
A flux of 2.9 1018 6 He decays/year and 1.1 1018 18 Ne decays/year, as discussed
in Ref. [13], will be assumed. Fig. 1 shows the -beam neutrino fluxes computed at the
130 km baseline, keeping me = 0, following the formul derived in Ref. [11]; in Ref. [13]
the fluxes are calculated in the me = 0 approximation. Be aware of the fact that even if
me effects seem negligible, their inclusion could be sizable due to the dramatic crosssection suppression of low energy neutrinos.2 Furthermore, in our calculations we take
2 For example we checked that, by using the cross-sections discussed in Section 4, the rate for both neutrinos
and antineutrinos computed in the me = 0 approximation is about 20% smaller than in the case me = 0.
407
Table 1
18 Ne and 6 He -decay channels and relative end-point energies from [24]
Element
End-point (MeV)
18 Ne
3.4114
2.3699
1.7106
6 He
3.5078
Decay fraction
92.1%
7.7%
0.2%
100%
into account the fact that the 18 Ne has three different decay modes, each with a different
end-point energy, see Table 1.
3.2. The super-beam
The super-beam is a conventional neutrino beam, but with a higher proton intensity.
Therefore, it has the advantages of a high intensity flux and of a well proved technology.
On the other hand its composition ( main component, if + are focused, plus a small
admixture of , e and e ) is affected by large systematic uncertainties that limit the
sensitivity in searching for neutrino oscillations mainly e .
As baseline for this work we consider a super-beam based on a 2.2 GeV proton beam
of 4 MW power SPL, described in Ref. [14]. The predicted energy spectra and fluxes for
the main components have been computed starting from a full simulation of the neutrino
beamline [25], assuming a decay tunnel length of 60 m. The neutrino fluxes expected at the
Frjus location are shown in Fig. 2. The average energy of the neutrino and antineutrino
beams is 0.27 and 0.25 GeV, respectively.
4. Neutrino cross-sections
The present knowledge of the neutrino and antineutrino cross-sections for energies below 1 GeV is very poor [26]: either there are very few data, as for neutrinos, or there are
no data available at all, as for antineutrinos. On top of that, the few available data have not
been taken on water, the target we are interested in, and the extrapolation from different nuclei is complicated by nuclear effects that at the considered energies play an important role.
For this calculation we adopted the cross-sections on water shown in Fig. 3 [27]. Notice
the difference between the e N and e N cross-sections: the former, being an interaction
between the e and a neutron inside the oxygen nucleus, is affected by nuclear effects and
thus shows a threshold energy. The latter is mainly a e interaction with the protons of the
two hydrogens, approximately free. This effect, although less pronounced, is visible also
for and for . This feature is quite relevant for neutrino/antineutrino of hundreds of
MeV energy, region where different cross-sections can easily differ by a factor 2. Compare
for example our Fig. 3 with Fig. 3 of Ref. [11] where NUANCE cross-sections [28] are
plotted. These differences can explain the (sometimes relevant) discrepancies within the
numbers of charged-current interactions and oscillated events predicted in different analyzes. Be aware that there are other nuclear effects (see [29] and references therein) not
408
Fig. 2. SPL super-beam fluxes at the Frjus location (130 km baseline) as a function of the neutrino energy [25].
Fig. 3. Cross-sections on water as a function of the neutrino energy [27].
409
included yet in any of the available calculations that could play an important effect at the
cross-section threshold energy.
With our calculations the expected number of charged-current events in the absence
of oscillation at the -beam per kton-year is 30.3 and 4.4 for e and e , respectively; in
Ref. [11], they quote 32.8 and 4.7 for e and e , respectively. We verify, that using the
NUANCE cross-sections we are able to reproduce exactly their results as well as those of
Ref. [13]. As far as the super-beam is concerned, the expected number of charged-current
events with a super-beam without oscillations per kton-year is 27.6 and 7.2 for and ,
respectively. We verify, that using the NUANCE cross-sections we are able to reproduce
within 5% the results of Ref. [30], while we still found a discrepancy of more than 30%
with those quoted in Ref. [13].
Besides the absolute value of the cross-sections, another important unknown is their
shape. Indeed, as it will be discussed later, some of the backgrounds have a neutrino energy
threshold. Therefore, the expected background strongly depends on the adopted model.
At the time the neutrino complex would become operational the cross-sections will
be measured precisely. However, nowadays we have the problem to compute the physics
potential of a facility having in mind that the expected number of signal and background
events strongly depend on the adopted calculation.
A facility where the neutrino fluxes are known with great precision is the ideal place
where to measure neutrino cross-sections. At a -beam the neutrino fluxes are completely
defined by the parent ions -decay properties and by the number of ions in the decay
ring. A close detector of 1 kton placed at a distance of about 1 km from the decay ring
could then measure the relevant neutrino cross-sections. Furthermore the factor of the
accelerated ions can be varied. In particular a scan can be initiated below the background
production threshold, allowing a precise measurement of the cross-sections for resonant
processes.
Unfortunately, there are no studies available on the ultimate precision achievable at future facilities in measuring the charged-current cross-sections. A first attempt to estimate
the ultimate systematic error achievable in the cross-section measurements at future facilities was given in [9]. By assuming the -beam complex described before a systematic error
of 2% was estimated.
5. 13 and sensitivity
In this section we present our results for the sensitivity to 13 and at the CERN-based
standard -beam and SPL super-beam. The sensitivity to the 13 and parameters has been
evaluated using the following reference values for the solar and atmospheric parameters:
m212 = 8.2 105 eV2 , 12 = 33 , m223 = 2.5 103 eV2 and 23 = 40 .
We proceed first summarizing the backgrounds for both facilities (Sections 5.1, 5.2) and
then studying the -beam and the super-beam separately and in combination (Section 5.3).
5.1. Signal and background at a standard -beam
The signal in a -beam looking for e ( e ) oscillations would be the appearance of ( ) charged-current events, mainly via quasi-elastic interactions, in a pure
410
Fig. 4. Reconstruction efficiency as a function of the true neutrino energies for 6 He ( e ) and 18 Ne (e ) in water [11].
e (e ) beam. Background rates and signal efficiencies have been studied, by means of a
full simulation based on the NUANCE code [28], in Refs. [11,13]. In this paper, we make
use of those results for the beam and detector fractional background and compute the expected number of oscillated events by using the fluxes of Fig. 1, the cross-section on water
of Fig. 3, the full three-families oscillation probability in matter and the detection efficiency of Fig. 4.
Given its excellent purity (neither nor are in the initial beam) the background in a
detector exploiting a -beam can be generated either by inefficiencies in particle identification, such as mis-identification of pions produced in neutral-current single-pion resonant
interactions, electrons (positrons) mis-identified as muons, or by external sources such as
atmospheric neutrino interactions.
The pion background has a threshold at neutrino energies of about 0.45 GeV, and is
highly suppressed at the -beam energies. The electron background is almost completely
suppressed by the request of the detection of a delayed Michel electron following the
muon track. If a bunch length of 10 ns (which seems feasible) is assumed, this background
becomes negligible [13]. Moreover, out-of-spill neutrino interactions can be used to normalize this background to the 1% accuracy level.
The event rates for a 4400 kton-year exposure are also given in Table 2, both for nonoscillated and oscillated e , e , with e oscillation probability computed for 13 =
10 and = 90 , and 13 = 1 and = 90 , 90 . In Table 2 the expected beam and
411
Table 2
Event rates for a 10 years exposure at a standard -beam. The oscillated charged-current events for different
values of 13 , and sign of the atmospheric mass difference, satm , for both neutrinos and antineutrinos are given.
For comparison with literature we show here the values obtained with the reference parameters but 23 = 45 .
The oscillation probabilities at the mean neutrino/antineutrino energy (in GeV) are also shown
13
satm
Pe (E = 0.37)
P e (E = 0.23)
No osc.
10
10
1
1
1
1
0
0
90
90
90
90
133205
2444
1818
83
79
9
9
19557
463
412
1
1
21
21
5.38 102
4.16 102
2.03 103
1.95 103
1.05 104
1.13 104
5.07 102
6.26 102
2.60 104
2.22 104
3.74 103
3.86 103
0
360
0
1
Beam back.
Detector back.
detector backgrounds (derived from the fractional background in Refs. [11,13]) are also
given.
Finally, some comments on the overall systematic error and on the expected background
are in order. The ultimate precision on cross-sections achievable at future facilities is about
2%. This value has been also assumed in Ref. [13] as an overall systematic error. However,
it does not take into account possible systematic errors on the detection efficiencies and on
the neutrino fluxes. In order to be conservative, we adopted an overall systematic error of
5%. Nonetheless, we also studied the impact on the physics potential going from 5 to 2%.
As far as the background is concerned, we would like to stress that it is due to the
coherent pion production process, with a threshold at 0.45 GeV. Therefore, following the
arguments of Section 4, it strongly depends on the adopted model. We checked that our
results are stable against variations of the background in the neutrino channel, while this is
not the case for the antineutrino channel. In absence of a full Monte Carlo simulation, in
the following calculations we adopt the central values given in Table 2. The impact of an
increase of this background on the sensitivity is discussed in Section 5.4.
5.2. Signal and background at the super-beam
The search for e ( e ) appearance with a super-beam is complicated by
the e (e ) contamination of the beam. Indeed, contrary to the case of the -beam where
the beam-induced background is absent (see Table 2), for the super-beam a significant
background from e (e ) charged-current interactions must be considered, resulting in a loss
of sensitivity. In a water Cerenkov

detector the appearance of a e (e ) signal is detected by
exploiting the high efficiency and purity of the detector in identifying electrons and muons
in low multiplicity interactions.
Besides the e (e ) contamination of the beam, the main sources of background are the
charged-current interactions of ( ) and the production of 0 in neutral-current interactions.
412
Table 3
Event rates for an exposure at a standard super-beam. The oscillated charged-current events for different values of
13 , and sign of the atmospheric mass difference, satm for both neutrinos (2 years data taking) and antineutrinos
(8 years of data taking) are given. For comparison with literature we show here the values obtained with the
reference parameters but 23 = 45 . The oscillation probabilities at the mean neutrino/antineutrino energy (in
GeV) are also shown
13
No osc.
10
10
1
1
1
1
Beam back.
Detector back.
0
0
90
90
90
90
satm
P e (E = 0.27)
P e (E = 0.25)
24245
1148
1015
3
3
55
54
25467
1002
1008
56
57
6
5
6.07 102
5.74 102
5.70 105
6.60 105
3.28 103
3.22 103
5.63 102
6.15 102
3.50 103
3.57 103
1.22 104
1.05 104
92
24
110
56
To compute the expected number of non-oscillated and oscillated events we have used
the fluxes of Fig. 2, the cross-section on water of Fig. 3, the full three-families oscillation
probability in matter and the electron detection efficiency computed in Ref. [14]: 70.7
and 67.1% for e and e , respectively. The corresponding beam and detector
background has been computed making use of the results quoted in Refs. [13,14].
Regarding the signal and background systematic errors, we followed the arguments
given in Ref. [31], but we have considered only 2 and 5%, and finally presented results
for the worst case. The two cases are on the other hand considered when sensitivity plots
are presented.
The event rates are given in Table 3. Note that, given the difference in cross-section
between neutrino and antineutrino and that it is not possible to run a super-beam with both
polarities at the same time, to get comparable statistics the run should be asymmetric in
time: in Ref. [13], 2 years run with main component and 8 years run with main component were assumed. It is however not clear that having a comparable statistics is really
necessary to get a good 13 , signal. In the case of the neutrino factory, for example, it
has been shown that a combination of two different detectors, the first looking for e
oscillations with high statistics and the second for e with low statistics is extremely
useful to solve some of the parameter space degeneration [32]. Indeed, with a 2 + 8 run
the gain in the flux is compensated by a loss in the flux. We have therefore also run
in a symmetric 5 + 5 years configuration: our results indicate that the two choices work
similarly on the average, with one or the other performing slightly better depending on the
particular region of the (13 , ) parameter space. In the rest of the paper, to establish direct
comparison with [13], we adopt the 2 + 8 configuration.
5.3. Extraction of neutrino oscillation parameters in presence of signal
The sensitivity to the 13 and parameters has been evaluated by assuming the signal
and background rates reported in the previous sections and the following input values:
413
m212 = 8.2 105 eV2 , 12 = 33 , m223 = 2.5 103 eV2 , and 23 = 40 . Since
the sign of m223 and the 23 -octant are unknown, fits to both sign[m223 ] = 1 and
sign[tan(223 )] = 1 have been performed. In the particular case 23 = 45 , four out of
eight solutions of the systems of Eqs. (3)(6) disappear and only fits to the wrong assignment of sign(m223 ) must be performed. However, since at a given confidence level the
contours for the allowed regions (around the theoretical location of the true solution, of the
intrinsic clone and of the two sign clones) are not qualitatively different from those we get
for 23 = 40 , we have opted to present results for this last case, only.
Our results for two specific values, 13 = 1 , 7 , are presented in Figs. 57. In each
figure we plot the 90% CL contours for the two considered values of 13 and a fixed value
of : = 90 (Fig. 5), = 0 (Fig. 6) and = 90 (Fig. 7). From top to bottom, the results
for the -beam, for the super-beam and for the combination of the two are presented.
In every separate case, the four possible choices of the discrete variables satm , soct are
reported: continuous lines stand for the true solution and the intrinsic degeneracy (right
sign(m223 ) and right 23 -octant); dashed lines stand for the sign degeneracy (wrong
sign(m223 ) and right 23 -octant); dot-dashed lines stand for the octant degeneracy (right
sign(m223 ) and wrong 23 -octant); dotted lines stand for the mixed degeneracy (wrong
sign(m223 ) and wrong 23 -octant). In all the figures, the light (red in the web version)
circle shows the input value. The dark (black) circles in the plots for the -beam and the
super-beam represent the theoretical clone locations computed as in Ref. [20] for the two
facilities considered in this paper. They have been included to show the rather good agreement between the theoretical computation and the output of the fits. Notice that for 13 = 7
some of the theoretical clones are missing. Indeed, as it was made clear in [20], an analytic
solution of the systems in Eqs. (3)(6) is not always found for large 13 , although due to
statistical uncertainties a low 2 region can still be identified.
Notice first that the impact of the octant-ambiguity in all the considered cases is by far
more relevant for large values of 13 than for small values. For the three values of that
we have analyzed we get the octant clones with a 13 1 shift with respect to the true
solution when 13 = 7 , whereas we get a significantly smaller shift when 13 = 1 . This
comment apply both to the -beam and the super-beam results: indeed, when the octantambiguity is considered, both facilities show pretty similar contours. This is why, as a
general result, the combination of the two facilities does not solve this ambiguity. Even in
the case of 13 = 7 , for which a significant statistics is accumulated for both facilities, the
bottom plot of Figs. 57 shows that at 90% CL the true solution and the octant ambiguity
survive for the three values of considered.
Concerning the sign ambiguity we notice that, for both facilities and for both considered
values of 13 = 1 , 7 , the allowed regions corresponding to the clone solutions overlap
with the true solution in a significant way. As a consequence the combination of the two
facilities, albeit reducing the 90% CL contours, does not solve the sign ambiguity, either.
However, contrary to the case of the octant degeneracy treated above, this does not affect
the measurement of the two continuous unknowns 13 and . For large 13 , a considerably
good measurement of both quantities is achieved, with no clue on the mass hierarchy. An
exception to this statement is the particular case 13 = 7 , = 0, for which the sign clones
are quite definite regions located in different places for the -beam and the super-beam.
414
Fig. 5. Fits to 13 and after a 10 yrs -beam run and a 2 + 8 super-beam run. The 90% CL contours are
shown for the following input values: 13 = 1 , 7 and = 90 . Upper panel: -beam results; middle panel:
super-beam results; lower panel combined results. Continuous lines stand for the intrinsic degeneracy; dashed
lines stand for the sign degeneracy; dot-dashed lines stand for the octant degeneracy; dotted lines stand for the
mixed degeneracy. The light circle (red in the web version) shows the input value. Dark (black in the web version)
dots are the theoretical clone locations computed as in Ref. [20].
415
Fig. 6. Fits to 13 and after a 10 yrs -beam run and a 2 + 8 super-beam run. The 90% CL contours are shown
for the following input values: 13 = 1 , 7 and = 0. Upper panel: -beam results; middle panel: super-beam
results; lower panel: combined results. Continuous lines stand for the intrinsic degeneracy; dashed lines stand for
the sign degeneracy; dot-dashed lines stand for the octant degeneracy; dotted lines stand for the mixed degeneracy.
The light circle (red in the web version) shows the input value. Dark (black in the web version) dots are the
theoretical clone locations computed as in Ref. [20].
416
Fig. 7. Fits to 13 and after a 10 yrs -beam run and a 2 + 8 super-beam run. The 90% CL contours are
shown for the following input values: 13 = 1 , 7 and = 90 . Upper panel: -beam results; middle panel:
super-beam results; lower panel: combined results. Continuous lines stand for the intrinsic degeneracy; dashed
lines stand for the sign degeneracy; dot-dashed lines stand for the octant degeneracy; dotted lines stand for the
mixed degeneracy. The light circle (red in the web version) shows the input value. Dark (black in the web version)
dots are the theoretical clone locations computed as in Ref. [20].
417
This actually allows the cancellation of some of the allowed regions when combining the
two facilities. The ambiguities are however not solved, since clone regions survive near
= 180 .
Discussing the mixed ambiguity, we notice that much of what was said for the sign
degeneracies can be repeated in this case: the only difference being that allowed regions
around mixed clones overlap significantly with the octant degeneracy, and not with the true
solution as it was the case for the sign degeneracy. As a consequence, mixed clones survive
after the combination of the two facilities. This comment applies to all the considered input
pairs, with the exception of the 13 = 7 , = 0 case (as for the sign ambiguity).
We observe that in general four allowed regions are still present after adding data from
the -beam and the super-beam at large 13 . These four regions either overlap in pairs (case
of = 90 , 90 ) or stay well apart (case of = 0), in all cases allowing a measure of the
two continuous parameters but not of the two discrete parameters. On the other hand, for
small 13 we are clearly on the sensitivity limit of these experiments: no clear measure
of is possible (at 90% CL we can just state if is positive or negative) and no discrete
ambiguity is solved.
From the results reported in the top plots of Figs. 57, it is clear that by using the -beam
or the super-beam as isolated experiment it is not possible to solve any of the degeneracies,
although for large enough 13 a first estimate of the two continuous parameters 13 and
can be attempted. Even when combining the two experiments we have found that as a
general result the discrete parameters are not measured. These results are in contrast with
the following statement of Ref. [13]: We stress the fact that an experiment working at very
short baselines has the smallest possible parameter degeneracies and ambiguities and it is
the cleanest possible environment where to look for genuine leptonic CP-violation effects.
The correct statement is that in many cases the discrete ambiguities, although not solved,
does not affect in a significant way the measure of the continuous parameters. Notice,
however, that in general multiple solutions are found with either larger uncertainties in
both parameters when these regions overlap or a proliferation of disconnected regions in
the parameter space.3
5.4. Exclusion plots in absence of signal
We present here exclusion plots in the absence of a signal for the -beam, the SPLsuper-beam and their combination, first in 13 for fixed and after in for fixed 13 . The
different lines represent different choices of the two discrete parameters satm , soct , in the
same notation used in the previous figures.
(1) Sensitivity to 13 at fixed . In Fig. 8 we draw the 90% CL contour defining the
sensitivity limit on 13 in case of absence of a signal, with as a fixed free parameter for
the -beam, the SPL super-beam and their combination. The sensitivity has been computed
applying a 2 analysis including the expected background and a tentative 2% systematic
error.
3 The presence and the locations of the clones as derived in this paper are in rather good agreement with the
analytical calculations quoted in [20].
418
Fig. 8. The 90% CL sensitivity plot to 13 at fixed value of for: (a) A 10 years -beam run; (b) A 2 + 8 years
SPL super-beam run; (c) The combination of the two with a 2% systematic error and (d) with a 5% systematic
error. As in the previous figures, continuous lines stand for the intrinsic degeneracy; dashed lines stand for the
sign degeneracy; dot-dashed lines stand for the octant degeneracy; dotted lines stand for the mixed degeneracy.
The upper left figure is the sensitivity limit after a 10 years standard -beam run. Notice
how the sensitivity limit spans a region from 0.5 to 2.0 , the less stringent limit
being for = 0 and = 180 . It is important to stress that whereas the worse limit is
generally given by the true solution (solid line), for = 0 a significant sensitivity loss is
induced by the unsolved sign and mixed ambiguities. It is also worth noticing that Fig. 8(a)
shows a better sensitivity for = 90 than for = 90 (see also [13], Fig. 4). This is
a consequence of the expected large signal statistics for neutrinos at = 90 and of the
extremely small background for antineutrinos (see Table 2). The upper right figure is the
sensitivity limit after a 2 + 8 years SPL super-beam run. In this case, the effect of the
degeneracies on the sensitivity is much lower than in the case of the -beam: for any value
of the different excluded regions coming from different choices of the discrete parameters
419
Fig. 9. Expected number of signal events in case of e and e as a function of for different values
of 13 in the range 0.2 1.8 (with step 0.2 ). Continuous (dashed) lines correspond to neutrinos (antineutrinos).
The vertical lines show the expected background for neutrinos (continuous) and antineutrinos (dashed). The
thicker line corresponds to the lower value of 13 .
satm and soct overlap significantly. Moreover, at the super-beam we notice no asymmetry
in . This is a consequence of the more symmetric background reported in Table 3 for
both the neutrino and antineutrino beams. In the lower left figure we show the combined
sensitivity for the two facilities. It can be seen that in the combination the strong asymmetry
for = 90 and = 90 in the -beam plot has been reduced. Eventually, in the lower
right figure we have studied the dependence of the sensitivity plot on the systematic error.
For the combined -beam and SPL super-beam setup we present our results using a 5%
systematic error: it can be noticed that the impact of this change is not dramatic.
In Figs. 9 and 10 we will show how the observed asymmetry in for the -beam sensitivity to 13 , Fig. 8(a), is only a consequence of the 2 statistical analysis. Indeed, given
the small expected background in the antineutrino channel at the -beam (see Table 2),
a more suitable approach when the signal is also small is the unified approach discussed
in Ref. [33]. In Fig. 9 we plot the number of signal and background events for neutrinos
(solid lines) and antineutrinos (dashed lines) as a function of for different values of 13 .
It can be noticed how the expected number of signal plus background events at the -beam
in the antineutrino channel is of the order of few units only for positive , whereas in the
neutrino channel this is not the case even at = 90 due to the large expected background. This asymmetry in the signal-to-noise ratio is exploited in the unified approach,
resulting in the exclusion plot for the -beam given in Fig. 10. Within this approach, the
420
Fig. 10. Sensitivity plots at 90% CL to 13 after a 10 yrs -beam run as computed with the unified approach.
Continuous lines stand for the intrinsic degeneracy; dashed lines stand for the sign degeneracy; dot-dashed lines
stand for the octant degeneracy; dotted lines stand for the mixed degeneracy. The thick line is the effective
sensitivity limit.
exclusion plots are similar to the ones obtained with the 2 for all values of but for a
range of values around = +90 . A proper treatment of small event samples shows that
the observed enhancement of the 13 sensitivity for = 90 in Fig. 8(a) is nothing but an
illusion due to a wrong statistical approach. We have applied the unified approach to the
super-beam data, also. In this case, as expected being the number of background events
large both for neutrinos and antineutrinos, the results are pretty similar to those obtained
with the 2 method.
As a final check we have increased the expected number of background events in the
antineutrino sample. This is justified by the fact that the background itself is affected by
large uncertainties. As expected, the sensitivity enhancement at = 90 disappears not
only using the unified approach but the 2 method, also.
(2) Sensitivity to at fixed 13 . In Fig. 11 we draw the 90% CL contour defining the
sensitivity limit on in case of absence of a signal (i.e., = 0 ), for the -beam, the SPL
super-beam and their combination. For any given value of 13 , we compare N (13 , 0 ) with
N(13 , ), with 13 a fixed parameter. The sensitivity has been again computed applying a
2 analysis including the expected background and a tentative 2% systematic error.
The upper left figure again represents the results for a 10 years -beam run. It is clearly
visible how the true solution defines an upper bound on || 30 40 , whereas the sign
degeneracy draw a lower bound for | | 150 170 . This is a consequence of the
fact that N (13 , 0 ; satm ) N (13 , 180 ; satm ): the number of events for = 0 and a
given assignment of the atmospheric mass difference, satm , is indeed identical to that for
421
Fig. 11. The 90% CL sensitivity plot to at fixed value of 13 for: (a) A 10 years -beam run; (b) A 2 + 8 years
SPL super-beam run; (c) The combination of the two with a 2% systematic error and (d) with a 5% systematic
error. As in the previous figures, continuous lines stand for the intrinsic degeneracy; dashed lines stand for the
sign degeneracy; dot-dashed lines stand for the octant degeneracy; dotted lines stand for the mixed degeneracy.
= 180 and the opposite assignment in vacuum and/or in the limit of vanishing 13 .
However, for large values of 13 the difference between the two samples is small but not
zero, thus inducing small asymmetries. On the other hand, the octant and mixed degeneracy
do not play a significant role in this sensitivity plot, apart for excluding small regions of
parameter space around 13 4 and positive due to some specific parameter conspiracy.
The upper right figure shows the results for a 2 + 8 years SPL super-beam run. In this
case, the bound on positive high values of is somewhat improved with respect to the beam case, whereas for negative values of the limit is a bit worse. The lower left figure
represent the combination of the two runs for a 2% systematic error. A sizable improvement
on the sensitivity can be noticed for large 13 : indeed, a lower bound || 10 is achieved
at 90% CL. For large values of ||, || = 180 can be distinguished from = 0 for
422
13 5 . Eventually, the lower right figure represent the combination of the two runs for
a 5% systematic error. As in the case of the sensitivity to 13 , we notice a rather small
difference with respect to the previous case (|| = 180 can be distinguished from = 0
above 13 6 ).
6. Conclusion
Over the past years the possibility to build a CERN neutrino complex, based on the
novel concept of the -beam and on the proposed SPL super-beam, exploiting a 1 megaton
water Cerenkov
detector located at the Frjus underground laboratory (130 km baseline)
has been put forward. In this paper we study for the first time the eightfold degeneracy for
such a scenario.
After a brief theoretical discussion about the eightfold degeneracy and a short description of the standard -beam and SPL super-beam facilities, we focus on the neutrino and
antineutrino cross-sections at low energies. At the time the neutrino complex will become
operational, it would be possible to measure with high accuracy the cross-sections. Unfortunately, we face nowadays the problem of computing the physics potential of a facility
having in mind that the expected number of signal and background events strongly depend
on the adopted model. In particular we pointed out that the background in the antineutrino
channel of a -beam depends on the shape and absolute value of the cross-section and that
this background can significantly affect the sensitivity to 13 and .
From the results reported in this paper it is clear that by using the -beam or the SPL
super-beam alone it is not possible to solve all the degeneracies, although for large enough
13 a first estimate of the two continuous parameters 13 and can be attempted. Even
when combining the two experiments we have found that as a general result the discrete
parameters are not measured. In many cases the discrete ambiguities, although not solved,
do not affect in a significant way the measurement of the continuous parameters. Notice,
however, that in general multiple solutions are found with either larger uncertainties in
both parameters when these regions overlap or a proliferation of disconnected regions in
the parameter space.
In the absence of a signal (Section 5.4) we have found that the -beam and the SPL
super-beam can exclude 13 [1 , 2 ] depending on and that their combination does not
improve a lot on these bounds. Moreover, a significant loss in sensitivity arise for = 0
due to the unsolved sign ambiguity. An apparent asymmetry in the sensitivity to 13 plot
for the standard -beam between = 90 and = 90 has been removed by using the
unified approach [33] statistical treatment. The effect arise when applying a standard 2
analysis to a sample with extremely few signal events and a very low expected background,
such in the case of the antineutrino data for = 90 .
As a final comment we want to stress that being the experiment discussed in this paper a counting experiment, the small differences between the fluxes are averaged-out.
Therefore, being Nevts ( e ) T (Nevts (e )), the combination (synergy) of a beam and of a super-beam only determines an increase of statistics for both neutrinos and
antineutrinos. Therefore, there is not a real synergy.
423
Acknowledgements
We would like to thank B. Gavela, S. Gilardoni, J. Gomez-Cadenas, P. Hernandez, P. Lipari, O. Mena, M. Mezzetto and F. Terranova for useful discussions during the preparation
of this paper. A.D. and S.R. acknowledge the M.C.Y.T. for financial support through the
Ramn y Cajal program.
References
[1] Y. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562, hep-ex/9807003;
M. Ambrosio, et al., MACRO Collaboration, Phys. Lett. B 517 (2001) 59, hep-ex/0106049.
[2] M.H. Ahn, et al., K2K Collaboration, Phys. Rev. Lett. 90 (2003) 041801, hep-ex/0212007.
[3] B.T. Cleveland, et al., Astrophys. J. 496 (1998) 505;
J.N. Abdurashitov, et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801, astro-ph/9907113;
W. Hampel, et al., GALLEX Collaboration, Phys. Lett. B 447 (1999) 127;
S. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 86 (2001) 5651, hep-ex/0103032;
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301, nucl-ex/0106015.
[4] K. Eguchi, et al., KamLAND Collaboration, Phys. Rev. Lett. 90 (2003) 021802, hep-ex/0212021.
[5] C. Athanassopoulos, et al., LSND Collaboration, Phys. Rev. Lett. 81 (1998) 1774, nucl-ex/9709006;
A. Aguilar, et al., LSND Collaboration, Phys. Rev. D 64 (2001) 112007, hep-ex/0104049.
[6] I. Stancu et al., MiniBooNE Collaboration, FERMILAB-TM-2207.
[7] B. Pontecorvo, Sov. Phys. JETP 6 (1957) 429, Zh. Eksp. Teor. Fiz. 33 (1957) 549 (in Russian);
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870;
B. Pontecorvo, Sov. Phys. JETP 26 (1968) 984, Zh. Eksp. Teor. Fiz. 53 (1967) 1717 (in Russian);
V.N. Gribov, B. Pontecorvo, Phys. Lett. B 28 (1969) 493.
[8] M. Apollonio, et al., CHOOZ Collaboration, Phys. Lett. B 466 (1999) 415, hep-ex/9907037;
M. Apollonio, et al., CHOOZ Collaboration, Eur. Phys. J. C 27 (2003) 331, hep-ex/0301017.
[9] M. Apollonio, et al., hep-ph/0210192.
[10] Workshop on Radioactive Beams for Nuclear Physics and Neutrino Physics, 37th Rencontre de Moriond,
Les Arcs, France, 1722 March 2003, http://moriond.in2p3.fr/radio/index.html.
[11] J. Burguet-Castell, D. Casper, J.J. Gomez-Cadenas, P. Hernandez, F. Sanchez, Nucl. Phys. B 695 (2004)
217, hep-ph/0312068.
[12] F. Terranova, A. Marotta, P. Migliozzi, M. Spinetti, Eur. Phys. J. C 38 (2004) 69, hep-ph/0405081.
[13] J. Bouchez, M. Lindroos, M. Mezzetto, AIP Conf. Proc. 721 (2004) 37, hep-ex/0310059.
[14] J.J. Gomez-Cadenas, et al., CERN Working Group on Super Beams Collaboration, hep-ph/0105297.
[15] H. Minakata, H. Nunokawa, Phys. Lett. B 495 (2000) 369, hep-ph/0004114;
V.D. Barger, S. Geer, R. Raja, K. Whisnant, Phys. Rev. D 63 (2001) 113011, hep-ph/0012017;
V.D. Barger, et al., hep-ph/0103052;
H. Minakata, H. Nunokawa, JHEP 0110 (2001) 001, hep-ph/0108085;
P. Huber, M. Lindner, W. Winter, Nucl. Phys. B 645 (2002) 3, hep-ph/0204352;
G. Barenboim, et al., hep-ex/0206025.
[16] J. Burguet-Castell, M.B. Gavela, J.J. Gomez-Cadenas, P. Hernandez, O. Mena, Nucl. Phys. B 608 (2001)
301, hep-ph/0103258;
J. Burguet-Castell, O. Mena, hep-ph/0108109.
[17] H. Minakata, H. Nunokawa, JHEP 0110 (2001) 001, hep-ph/0108085.
[18] G.L. Fogli, E. Lisi, Phys. Rev. D 54 (1996) 3667, hep-ph/9604415.
[19] V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023, hep-ph/0112119.
[20] A. Donini, D. Meloni, S. Rigolin, JHEP 0406 (2004) 011, hep-ph/0312072.
[21] C.K. Jung, hep-ex/0005046.
[22] P. Zucchelli, Phys. Lett. B 532 (2002) 166.
424
[23] A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cadenas, P. Hernandez, O. Mena, S. Rigolin, Nucl. Phys.
B 579 (2000) 17;
A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cadenas, P. Hernandez, O. Mena, S. Rigolin, Nucl. Phys.
B 593 (2001) 731, Erratum, hep-ph/0002108.
[24] L.P. Ekstrm, R.B. Firestone, WWW table of radioactive isotopes, from http://ie.lbl.gov/toi/index.htm.
[25] S. Gilardoni, CERN Thesis, in press;
S. Gilardoni, G. Grawer, G. Maire, J.M. Maugain, S. Rangod, F. Voelker, J. Phys. G 29 (2003) 1801.
[26] G.P. Zeller, hep-ex/0312061.
[27] P. Lipari, private communication;
P. Lipari, M. Lusignoli, F. Sartogo, Phys. Rev. Lett. 74 (1995) 4384, hep-ph/9411341.
[28] D. Casper, Nucl. Phys. B (Proc. Suppl.) 112 (2002) 161, hep-ph/0208030.
[29] J. Serreau, C. Volpe, hep-ph/0403293.
[30] M. Mezzetto, J. Phys. G 29 (2003) 1781, hep-ex/0302005.
[31] Y. Itow, et al., hep-ex/0106019.
[32] A. Donini, D. Meloni, P. Migliozzi, Nucl. Phys. B 646 (2002) 321, hep-ph/0206034;
D. Autiero, et al., Eur. Phys. J. C 33 (2004) 243, hep-ph/0305185.
[33] G.J. Feldman, R.D. Cousins, Phys. Rev. D 57 (1998) 3873, physics/9711021.
3-charge geometries and their CFT duals

Stefano Giusto, Samir D. Mathur, Ashish Saxena
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
Received 22 November 2004; accepted 5 January 2005
Abstract
We consider two families of D1D5P states and find their gravity duals. In each case the geometries are found to cap off smoothly near r = 0; thus there are no horizons or closed timelike curves.
These constructions support the general conjecture that the interior of black holes is nontrivial all the
way up to the horizon.
PACS: 11.25.-w; 11.27.+d
1. Introduction
The traditional picture of a black hole has a horizon, a central singularity, and essentially
empty space in between. This picture leads to contradictions with quantum mechanics
Hawking radiation leads to a loss of unitarity [1]. More recently a different picture of
the black hole interior has been suggested, where the information of the state of the hole
is distributed throughout the interior of the horizon, creating a fuzzball [2]. While the
general state of a Schwarzschild hole is expected to be very nonclassical inside the horizon, we expect that for extremal holes we can find appropriately selected states that will
be represented by classical solutions. In [3] it was found that the generic state of the 2charge extremal D1D5 system could be understood by studying classical solutions of
supergravity, and in [46] classical solutions were constructed for specific families of 3charge extremal D1D5P states. The traditional picture of the 3-charge extremal hole is
E-mail address: giusto@ge.infn.it (S. Giusto).
doi:10.1016/j.nuclphysb.2005.01.009
426
S. Giusto et al. / Nuclear Physics B 710 (2005) 425463
Fig. 1. (a) Naive geometry of 3-charge D1D5P; there is a horizon at r = 0 and a singularity past the horizon.
A = 2 n n n .
(b) Expected geometries for D1D5P; the area at the dashed line will give 4G
1 5 p
pictured in Fig. 1(a). But the 3-charge geometries constructed in [5,6] were of the form
Fig. 1(b); the throat caps off without any horizon or singularity. (All 2-charge extremal
states have a geometry that caps off like Fig. 1(b); the naive geometry in this case has a
zero area horizon coinciding with the singularity at r = 0.)
In this paper we pursue this program further, by finding further sets of 3-charge extremal CFT states and their dual geometries. One can write down a large class of 3-charge
extremal solutions of classical supergravity, and these will in general have pathologies. But
a basic tenet of our conjecture is that the geometries that are dual to actual microstates of
the 3-charge CFT will be regular solutions with no horizons, singularities or closed timelike curves. The solutions in [5,6] were smooth, and we will find that the solutions we now
construct will also be free of horizons and closed timelike curves (though most will have
an orbifold singularity along certain curves). Thus our solutions will be like Fig. 1(b) rather
than 1(a), and will lend support to the general fuzzball picture of the black hole interior.1
In more detail, we do the following.
(a) In [11,12] a family of D1D5 geometries was obtained, by taking extremal limits in
the general family of rotating 3-charge solutions constructed in [13]. This family is labelled
by a parameter 0 < 1. The geometries have no horizons and the only singularity is an
orbifold singularity along an S 1 in the noncompact directions. The corresponding CFT
duals can be identified [3,11,12]. The orbifold singularity vanishes for the special case
= 1.
If we perform a spectral flow on the left sector of the CFT then from a 2-charge D1D5
state we get a 3-charge D1D5P state. In [6] we found the geometries for the 3-charge
states that are related by spectral flow to the 2-charge state with = 1. Here we extend
this computation to find the geometries for 3-charge states starting from 2-charge states for
arbitrary . It is straightforward to identify the corresponding CFT duals.
(b) Given a D1D5P state we can use dualities to interchange any two of its charges.
This leads to a new geometry which must also represent a true state of the 3-charge system,
since these dualities are exact symmetries of the theory. We construct these geometries
1 Additional evidence for this picture comes from a study of the nonzero size of supertubes [710].
427
obtained by S, T dualities. These geometries turn out to have orbifold singularities along
two nonintersecting S 1 curves. These geometries and the ones obtained in (a) all fall into
a general class that we identify; they are rotating extremal solutions with parameters such
that there are no horizons and no closed timelike curves.
(c) It is not immediately obvious what the CFT states corresponding to the geometries
in (b) are. To find information about the state, we study the infall of a quantum down the
throat of this geometry, and study the travel time tSUGRA for a complete bounce. It
was found in [3] that this bounce time exactly equalled the time tCFT for excitations
to travel around the corresponding effective string in the CFT. By computing tSUGRA
for the geometries we find the length of each component of the effective string, and thus
identify the CFT state.
(d) We observe that the result found in (c) supports the picture of spacetime bits arrived at in [3]: Under duality the number of components of the effective string remains the
same, though the total winding number of the effective string changes. We also observe
that the travel time in the geometry and in the CFT are related by a redshift factor which
relates the time coordinate at infinity to the time coordinate in the AdS region ( becomes
unity if the momentum charge P vanishes).
2. D1D5P states from spectral flow of D1D5 states

2.1. The D1D5 CFT
S1
We take IIB string theory compactified to M4,1 S 1 T 4 . Let y be the coordinate along
with
0 y < 2R.
(2.1)
The T 4 is described by 4 coordinates z1 , z2 , z3 , z4 , and the noncompact space is spanned

by t, x1 , x2 , x3 , x4 . We wrap n1 D1 branes on S 1 , and n5 D5 branes on S 1 T 4 . Let
N = n1 n5 . The bound state of these branes is described by a (1 + 1)-dimensional sigma
model, with base space (y, t) and target space a deformation of the orbifold (T 4 )N /SN (the
symmetric product of N copies of T 4 ). The CFT has N = 4 supersymmetry, and a moduli
space which preserves this supersymmetry. It is conjectured that in this moduli space we
have an orbifold point where the target space is just the orbifold (T 4 )N /SN [14].
The rotational symmetry of the noncompact directions x1 , . . . , x4 gives a symmetry
so(4) su(2)L su(2)R , which is the R symmetry group of the CFT.
The CFT with target space just one copy of T 4 is described by 4 real bosons X 1 ,
2
X , X 3 , X 4 (which arise from the 4 directions z1 , z2 , z3 , z4 ), 4 real left moving fermions
1 , 2 , 3 , 4 and 4 right moving fermions 1 , 2 , 3 , 4 . The central charge is c = 6.
The complete theory with target space (T 4 )N /SN has N copies of this c = 6 CFT, with
states that are symmetrized between the N copies. The orbifolding also generates twist
sectors, which are created by twist operators k . A detailed construction of the twist operators is given in [15,16], but we summarise here the properties that will be relevant to
us.
428
The twist operator of order k links together k copies of the c = 6 CFT so that the X i , i
act as free fields living on a circle of length kL (L is the length of the spatial circle of the
CFT). Let us first discuss the NS sector. The left fermions i carry spin 12 under the su(2)L
and the right fermions i carry spin 12 under the su(2)R . The charge of a state is given
by the quantum numbers (j, j) = (jL3 , jR3 ). Adding a suitable charge to the twist operator
we get a chiral primary
k :
h=j =
k1
h = j =
.
2
k1
,
2
(2.2)
+
We can act on this chiral primary with J1
to get another chiral primary
+
k :
k+ J1
h=j =
k+1
,
2
k1
h = j =
.
2
(2.3)
Similarly we also get

+
k :
k+ J1
k1
k+1
,
h = j =
,
2
2
k+1
k+1
,
h = j =
.
h=j =
2
2
h=j =
+ +
J1 k :
k++ J1
(2.4)
(2.5)
(We can get additional chiral primaries by applying for example +1 (which increase h
and j by 12 , but we will not need such states in this paper).)
2.1.1. A subclass of states

In the NS sector we can start with the NS vacuum
|0NS :
h = j = 0,
h = j = 0
(2.6)
and act with k to generate chiral primaries. Consider the subclass of states
N
k |0
k
NS :
h=j =
N (k 1)
,
k
2
N (k 1)
h = j =
.
k
2
(2.7)
All copies of the CFT are linked into long circles which are all of the same length kL,
and the spin orientation (given by the choice () for ) is also the same for each circle.
We therefore expect that the corresponding states exhibit some symmetry; it will turn out
that their gravity duals have axial symmetry around two circles , . Similarly we have
the states
+ N
k |0
k
NS :
+ N
k |0
k
NS :
++ N
k |0
k
NS :
N (k + 1)
,
k
2
N (k 1)
,
h=j =
k
2
N (k + 1)
h=j =
,
k
2
h=j =
N (k 1)
h = j =
,
k
2
N (k + 1)
h = j =
,
k
2
N (k + 1)
h = j =
.
k
2
(2.8)
(2.9)
(2.10)
429
2.1.2. Spectral flow

The NS sector states can be mapped to R sector states by spectral flow [17], under
which the conformal dimensions and charges change as
c
h = h q + 2 ,
(2.11)
24
c
q = q .
(2.12)
12
Setting = 1 gives the flow from the NS sector to the R sector, and we can see that under
this flow chiral primaries of the NS sector (which have h = q) map to Ramond ground
c
states with h = 24
.
The field theory on the D1D5 branes system is in the R sector. This follows from
the fact that the branes are solitons of the gravity theory, and the fermions on the branes
are induced from fermions on the bulk. The latter are periodic around the S 1 ; choosing
antiperiodic boundary conditions would give a nonvanishing vacuum energy and disallow
the flat space solution that we have assumed at infinity.
If we set = 2 in (2.12) then we return to the NS sector, and setting = 3 brings us
again to the R sector. More generally, the choice
= 2n + 1,
nZ
(2.13)
brings us to the R sector.

2.1.3. The states we consider
Suppose we start with a chiral primary in the NS sector. Perform a spectral flow (2.13)
on the right movers with = 1; this brings us to an R ground state for the right movers.
Perform a spectral flow with = 2n + 1 on the left movers. This brings us to the R sector
but not in general to an R ground state. The state thus has a momentum charge
np = h h.
(2.14)
Applying this procedure to the state (2.7) we get an R sector state

N
k |0

(k, n)
k
NS =2n+1, =1
L
with
(2.15)

n 1
2
,
h=N n + +
k 4

N
1
j =
2n +
,
2
k
N
h = ,
4
N
j =
2k
(2.16)
and

1
.
np = h h = N n n +
k
(2.17)
2.1.4. Explicit representations of the states

Let us construct explicitly the above CFT states. Consider one copy of the c = 6 CFT,
i . The 4 real fermions can be grouped into 2
in the R sector. The fermions have modes m
+
complex fermions , which form a representation of su(2). ( + has j = 12 and
430
has j = 12 .) The anti-commutation relations are

+

m , p+ = m+p,0 ,
m , p = m+p,0 .
(2.18)
The su(2) currents are

Jm+ = mp p+ , Jm = + mp p ,

1
(2.19)
mp p + mp p+ .
2
In the full theory with n1 n5 copies of the c = 6 CFT the currents are the sum of the currents
in the individual copies

Jna,total = Jna 1 + + Jna n n .
(2.20)
Jm3 =
1 5
First consider the state

momentum charge zero
N

k |0
k
NS
| (k, 0)
L =1,R =1
((2.15) for n = 0). This gives a D1D5 state with

N
,
2k
N
j = ,
2k
N
,
4
N
h = ,
4
j =
h=
np = 0.
(2.21)
Each set of k copies of the c = 6 CFT which are joined together by k behave like one
copy of the c = 6 CFT but on a circle of length kL.
Note. We will call each such set of linked copies a component string.
Thus in the presence of a twist operator of order k we can apply fractional modes of
currents
Ja m ,
k
Ja m .
(2.22)
Since we are in the R sector the fermions have fractional modes

m . Apart from this
k
fractionation the situation is identical to the one studied in [6] where we had no twist
(k = 1) and we applied currents to find the states arising after spectral flow. In the present
case the lowest dimension current operator that we can apply to lower charge is J 2 to
k
(2.21); this is equivalent to applying ( ) 1 1 . The next operator we can apply is J 4 ,

k
and so on. The orbifold CFT requires that the total momentum on each component string be
an integerwe will discuss this issue (and possible exceptions) in more detail in Section 6.
Thus we can apply
J 2(k1) J 4 J 2
k k
(2.23)
which adds dimension and charge

h = k 1,
j = (k 1)
(2.24)
431
or we can apply

J 2(k1) J 4 J 2
J2
(2.25)
k k
which adds dimension and charge

h = k + 1,
j = k.
(2.26)
To get states with a high symmetry we apply the same set of current operators to all the Nk
twist operators in the state. Applying (2.23) we get a state

J 2(k1) J 4 J 2 (k, 0)
(2.27)
k k
where the product runs over the Nk connected components of the CFT created by the twists.
This state has dimensions and charges
N
N
+ (k 1),
4
k
N
N
h = ,
j =
4
2k
j =
h=
N
N
(k 1),
2k
k
(2.28)
and
N
(k 1).
k
Applying (2.25) instead to each of the component strings we get the state

J2
J 2(k1) J 4 J 2 (k, 0)
np =
k k
(2.29)
(2.30)
with dimensions and charges

N
N
N
N
N
+ (k + 1),
j = N,
h = ,
j = ,
(2.31)
4
k
2k
4
2k
N
np = (k + 1).
(2.32)
k
We now observe that the state (2.30) has the correct dimensions and charges to be the
member of the spectral flow family (2.15) with n = 1. Similarly, the state (2.27) can be
N
identified with the state obtained by spectral flow, with n = 1, from the state (k+ ) k |0NS
in Eq. (2.8).
We can apply further sets of currents to get states with spectral flow by n > 1 units

J 2 (k, 0) .
J2n
(2.33)
h=
These states have

N
N
N
N
N
j = nN,
h = ,
j = , (2.34)
h = + n(nk + 1),
4
k
2k
4
2k
N
np = n(nk + 1).
(2.35)
k
We can similarly get those with n < 0; latter are obtained by applying modes of J + instead
of J .
432
2.2. Gravity duals

2.2.1. Duals of 2-charge states
In [13] a set of D1D5P solutions was given. The solutions had axial symmetry along
two circles , , and angular momenta J , J . In [11,12] an extremal limit was obtained
for solutions with P = 0, getting the geometries

dr 2
1 2
2
2
2
+ d
ds = dt dy + hf 2
h
r + a2 2

a 2 2 Q1 Q5 cos2
cos2 d 2
+ h r2 +
h2 f 2

a 2 2 Q1 Q5 sin2
2
2 2
sin2 d 2
+h r +a
h2 f 2

4

H1 2
2a Q1 Q5 2
cos dy d + sin2 dt d +
dxi ,
(2.36)
hf
H5
i=1
where
Q1 Q5
,
R
Q1
,
H1 = 1 +
f
a=
f = r 2 + a 2 2 cos2 ,
H5 = 1 +
Q5
,
f
h=
H1 H5 .
(2.37)
These metrics have angular momenta

J = j + j = 0,
J = j j = n1 n5
(2.38)
with n1 and n5 the numbers of D1 and D5 branes. j, j are the angular momenta in the L, R
factors of the so(4) su(2)L su(2)R describing the angular directions. For
1
= ,
k
k = 1, 2 . . .
(2.39)
we obtain geometries that are the duals of the states | (k, 0) (Eq. (2.21)) [3,11,12].
2.2.2. Duals of 3-charge states obtained by spectral flow of 2-charge states
We would now like to find the duals of the 3-charge states obtained by spectral flow of
the above 2-charge states. We again start from the 3-charge nonextremal solutions and take
an extremal limit, keeping the charges and angular momenta at the values given by the CFT
state. For the case where all twists were trivial (k = 1 for all twist operators) this procedure
was carried out in [6]. The starting nonextremal solution was derived in [6] by starting
with the neutral rotating hole in 4 + 1 dimensions, and applying a sequence of boosts and
dualities. This solution is reproduced in Appendix A. Taking the limit for P = 0 needs
some care, but the procedure was described in detail in [6] and needs no changes for the
more general case here.
We fix the values of the angular momenta and the momentum charge np to the values we
desire, and then take the extremal limit. For general values of these parameters, we obtain
433
the solution given in Appendix A. These solutions have pathologies in general; for example
they can have closed timelike curves. But we must select only those that correspond to
microstates of the 3-charge system, and these we expect to be free of pathologies.
We set the charges and momentum to equal those of the states (2.33)
J = nn1 n5 ,
J = (n + )n1 n5 ,
np = n(n + )n1 n5 .
(2.40)
The metric is written naturally in terms of the dimensionful quantities

4G(5)
4G(5)
J ,
2 =
J ,
Q1 Q5
Q1 Q5
4G(5)
(2)4 g 3
np ,
n1 ,
Qp =
Q1 =
R
V
1 =
Q5 = g n5 ,
(2.41)
where G(5) is the 5D Newtons constant

G(5) =
4 5 g 2 4
G(10)
=
,
V (2R)
VR
(2.42)
V is the volume of T 4 , R the radius of the S 1 and g the string coupling. We find
Q1 Q5 J
Q1 Q5
Q1 Q5 J
Q1 Q5
1 ,
2
2 =
1 =
R
n1 n5
R
R
n1 n5
R
and thus, for the duals of the states (2.33),
Q1 Q5
Q1 Q5
n,
2 =
(n + ),
1 =
R
R
We observe that
Qp = 1 2 .
Qp =
(2.43)
Q1 Q5
n(n + ). (2.44)
R2
(2.45)
We will see that it is this relation that selects, from the class of all axisymmetric solutions,
the geometries that are free of pathologies.2 Using (2.45) to simplify the extremal solution,
we get

Qp
1 2
dr 2
2
2
2
2
(dt dy) + hf 2
+ d
ds = dt dy +
h
hf
r + (1 + 2 )2

Q1 Q5 (12 22 ) cos2
cos2 d 2
+ h r 2 + 1 (1 + 2 )
h2 f 2

Q1 Q5 (12 22 ) sin2
2
sin2 d 2
+ h r + 2 (1 + 2 ) +
h2 f 2
2
Qp (1 + 2 )2 2 2
cos d + sin2 d
+
hf
2 We certainly have states of the system with Q = 0 and small or vanishing angular momenta, but as seen in
p
the 2-charge case [3] such states will break axial symmetry, and thus not be in the class that we are considering
at present.
434

2 Q1 Q5
1 cos2 d + 2 sin2 d (dt dy)
hf

4

2 Q1 Q5 (1 + 2 ) 2
H1 2
2
cos d + sin d dy +
dxi ,
hf
H5
i=1
Q1 Q5 cos2
(2 dt + 1 dy) d
C2 =
H1 f
Q1 Q5 sin2
(1 dt + 2 dy) d
H1 f

(1 + 2 )Qp
+
(Q1 dt + Q5 dy) cos2 d + sin2 d
Q1 Q5 H1 f

Q5 cos2 2
Q1
dt dy
r + 2 (1 + 2 ) + Q1 d d,
H1 f
H1 f
H1
e2 =
,
H5
(2.46)
(2.47)
(2.48)
where
Q1 Q5
,
Q1 Q5 + Q1 Qp + Q5 Qp

f = r 2 + (1 + 2 ) 1 sin2 + 2 cos2 ,

Q1
Q5
,
H5 = 1 +
,
h = H1 H5 .
H1 = 1 +
f
f
(2.49)
3. Obtaining new solutions by S, T dualities

As mentioned in the introduction, we are interested in making geometries that are dual
to actual bound states of the D1D5P system, and not just formal solutions of supergravity carrying D1, D5, P charges. In the previous section we started with known states in the
CFT and found their gravity duals by looking for solutions with the same symmetries and
quantum numbers. In this section we make D1D5P solutions by a different method: we
start with a D1D5P geometry that we have already constructed and perform S, T dualities to permute the charges. Since these dualities are exact symmetries of the theory, we
know that the resulting geometry represents a true microstate. But it will not be immediately obvious what the dual CFT state is. We will identify the CFT state later, by analyzing
the properties of the supergravity solution, and find that the change of CFT state under
these S, T dualities provides insight into the AdS/CFT duality map.
The metric (2.46) is invariant under the interchange of Q1 , Q5 , so the only nontrivial
duality is the one which interchanges the momentum charge with, say, the D1 charge

P
P
F1
D1
Ty ,Tz1
S
S
(3.1)
P .
D1 F1
P
D5
NS5
NS5
D5
435
The metric which results from this chain of dualities, and the corresponding dilaton are

Q1
1
dr 2
2
ds 2 = dt 2 dy 2 +
+
d
(dt dy)2 + hf
r 2 + (1 + 2 )2
h
hf

Qp Q5 (12 22 ) cos2
2
+ h r + 1 (1 + 2 )
cos2 d 2
h 2 f 2

Qp Q5 (12 22 ) sin2
+ h r 2 + 2 (1 + 2 ) +
sin2 d 2
h 2 f 2
2
Qp (1 + 2 )2 2 2
cos d + sin2 d
+
hf

2 Q1 Q5
hf

4

2Qp (1 + 2 ) Q5 2
Hp 2
cos d + sin2 d dy +
dxi ,
Q1
H5
hf
i=1
Q1 Q5 cos2
Q1 Q5 sin2
(2 dt + 1 dy) d
(1 dt + 2 dy) d
C2 =
Hp f
Hp f

(1 + 2 ) Q1
(Qp dt + Q5 dy) cos2 d + sin2 d
+
Hp f
Q5

Qp
Q5 cos2 2
r + 2 (1 + 2 ) + Qp d d,
dt dy
Hp f
Hp f
Hp
=
,
H5
e2
(3.2)
where
Hp = 1 +
Qp
,
f
h =
Hp H5 .
(3.3)
The solution above is again of the general form (2.46) but with different parameters3
Q1 = Qp ,
Q5 = Q5 ,
Qp = Q1 ,

Q1
Q1
1 =
1 ,
2 =
2 .
Qp
Qp
(3.4)
3 The quantities Q have units of (length)2 , so we have to be careful about the meaning of (3.4). If we start
i
Q
Q
Q
Q
with the FNS5P system and do Ty to get PNS5F then we get 21 = 2p , 2p = 21 , etc., where ls is the
ls
ls
ls
ls
Q
Q
Q
Q
string length. Starting with D1D5P and applying STy Tz1 S to get PD5D1 we get 21 = 2p , 2p = 21
ld
ld
ld
ld
1
where ld = g 2 ls is the D-string length. Since the classical geometry is unchanged by an overall rescaling, we can
write (3.4).
436
4. Conical defect angles

The geometries (2.46) possess, generically, an orbifold singularity along a circle in the
noncompact space directions, just like the subset of 2-charge metrics (2.36). Let us recall
the physics of these singularities, and then study the conical defect angle (created in
the AdS part of the geometry by the orbifolding) for the metrics constructed in the above
sections.
4.1. The physics of conical defects
Let us recall the reason why we get conical defects in 2-charge geometries. In [3] a
method was developed to compute the gravity duals for all states of the 2-charge system
(not just the subclass giving the geometries (2.36)). By S, T dualities we map the D1
D5 system to the FP system, which has a fundamental string (F) wrapped on S 1 carrying
momentum (P) along S 1 . The bound state of the FP system has the strands of the F string
all joined up into one multiwound string, and all the momentum is carried as travelling
waves on the string. Metrics for the vibrating string were constructed, and dualized back
to get D1D5 geometries. The general geometry was thus parametrized by the vibration
profile F (v) of the F string.
The detailed map between D1D5 states and D1D5 geometries is found in the following way. The vibration on the F string is written in terms of harmonics. If we have
a quantum of the kth harmonic on the string then we get a twist operator k acting on
the NS vacuum in the D1D5 CFT. The polarization of the vibration is given by an index i = 1, . . . , 4 labeling the four noncompact directions. The so(4) symmetry group of
the angular directions is su(2)L su(2)R , and writing the vector index i in terms of
the representation ( 12 , 12 ) of su(2)L su(2)R we get the choice of superscripts of the twist
operator k, . The collection of all twist operators (arising from all quanta of vibration
on the F string) give an NS state, which is spectral flowed to get an R sector state in the
D1D5 CFT. The geometry for this state is known, since it is obtained by S, T dualities
from the geometry created by the vibrating F string.
The strands of the F string wrap the y direction, but under the vibration they carry they
separate out from each other, and spread out over a simple closed curve in the transverse
space x 1 , . . . , x 4 . It appears at first that there would be a singularity in the FP and D1
D5 geometries at this curve, but it was found in [3] that all waves reflect trivially off this
singularity. The reason for this was explained in [18] where it was found that for the D1D5
geometries the singularity was just a coordinate singularity similar to that at the origin of a
KK monopole; we have a KK monopole tube (KK monopole S 1 ) centered at the above
curve. As long as this curve does not self-intersect (it generically does not self-intersect
since it is just an S 1 in R4 ) the 2-charge D1D5 geometry is completely smooth, with no
horizon or singularity.
The generic solution has no particular symmetry. If we look at solutions that have axial
symmetry then there are very few possibilities. We must let the F string swing in a uniform
helix in the covering space of the y circle; let this helix have k turns. The vibrations are
N
now all in the kth harmonic, so the D1D5 CFT state is created by (k ) k (the choice
() says that we let the F string swing in the x1 x2 plane; changing this plane changes
437
the superscripts). The KK monopole tube now runs over itself k times, so all points on
the S 1 exhibit the geometry of k KK monopoles coming together. But it is known that this
generates an ALE singularity, which has a conical defect angle of 2(1 k1 ).
4.2. Singularity structure of 3-charge metrics
We thus see that a conical defect is a harmless singularityit arises only if we make
nongeneric states by letting the S 1 run over itself, and in the full quantum theory one can
imagine that quantum fluctuations separate the intersecting strands and smooth out the
singularity. It is important though that the conical defect angle be of the form 2(1 k1 );
if we find an irrational angle for instance then we would not be able to understand how the
given geometry sits in a family of geometries that are generically smooth. We would like
to understand more precisely the nature of these defects. In particular we would also like
to know the values of the parameters 1 and 2 for which such defects might arise. The
arguments we use below are similar to the ones given in [6], where more details can be
found.
Around r = 0 the 6-dimensional part of the 3-charge metric (2.46) have the following
form:

2
hf
2
2
2 d y
dr + r
ds
(1 + 2 )2
R2

2
+ hf d + g cos2 d 2 + g sin2 d 2 + 2g cos2 sin2 d d
+ gtt dt 2 + 2g t cos2 dt d + 2g t sin2 dt d ,

(4.1)
where
= |1 + 2 |,
(4.2)
y
y
y = y,
= 2 ,
= 1 ,
R
R

f (1 + 2 ) 1 sin2 + 2 cos2
(4.3)
(4.4)
and g , g , g , gtt , g t , g t are differentiable functions of with

g (/2) = 1,
g (0) = 1.
(4.5)
The above form of the metric shows that at r = 0 the y cycle shrinks. It is important to
know if, for some particular value of , some other cycle shrinks at the same time. This
can be understood by looking at the determinant of the metric restricted to the t, and
coordinates
det g|t,,
=
(Q1 Q5 )2 2 2 2
sin cos2 .
hf
R2
(4.6)
This determinant only vanishes at = 0, /2: At = 0 the coordinate decouples from

and t and the coefficient of d 2 vanishes, i.e., the cycle shrinks at this point. At = /2
the same happens for the cycle. No other combination of the and cycles vanishes at
any other value of .
To proceed further, we need to look at the actual values of , 1 and 2 .
438
4.2.1. Geometries obtained by spectral flow: orbifold singularities

Let us consider first the case of the 3-charge metrics obtained by spectral flow from the
2-charge metrics in (2.36). For them

1
1
,
= , k N, n Z.
1 = n,
(4.7)
2 = n +
k
k
In this case the coordinates y,
and are

y
1 y
y = ,
= n +
,
k
k R
y
= + n .
R
If y/R,
and were periodic coordinates with period 2 , i.e., if

y
y
, ,
, , + 2(l1 , l2 , l3 )
R
R
(4.8)
(4.9)
with l1 , l2 , l3 integers, then the metric (4.1) would be smooth as can be seen from the
coefficients of dy, d and d in Eqs. (4.1)(4.5). However, it follows from the definition
(4.8) and from the periodicity of the asymptotic coordinates y/R, , , that y/R,
and
are subject to the further identifications:

y
1 1
y
, ,
, , + 2l , , 0
(4.10)
R
R
k
k
with l = 0, . . . , k 1. The identifications above generate a group isomorphic to Zk and
the space characterized by the metric (4.1) is topologically equivalent to an orbifold (R3
S 3 )/Zk . The orbifold action Zk has fixed points where both the y and cycles shrink to
zero size, which happens at r = 0 and = /2. Thus the spectral flow metrics with k > 1
have Zk orbifold singularities of the same kind as the original 2-charge metrics (2.36).
4.2.2. Metrics obtained after STy Tz1 S duality: orbifold singularities
Let us now turn to the case of the metrics obtained from the spectral flow metrics by
S, T dualities. Since these dualities interchange n1 and np and do not change the angular
momenta J and J , we find, using (2.43), (2.44)
1 =
2 =
R
Q1 Q5
R
Q1 Q5
1 =
J
k
,
=
np n5
n5 (kn + 1)
2 =
J
1
.
=
np n5 n5 n
(4.11)
We thus see that the parameter after duality

= 1 + 2 =
1
n5 n(kn + 1)
(4.12)
is again of the form (4.7) with the integer k replaced by the integer
k = n5 n(kn + 1).
(4.13)
439
In this case, then, the y,

and coordinates can be written as
y =
y
,
k
1 y
,
n5 n R
= +
k
y
n5 (kn + 1) R
(4.14)
and the space (4.1) is topologically equivalent to (R3 S 3 )/Zk , where now the Zk group
acts as

y
1
k
1
y

+ 2l
, ,
, ,
,
,
(4.15)
R
R
k
n5 n n5 (kn + 1)
with l = 0, . . . , k 1. Denote by the generator of this Zk group. We notice that, in
contrast with the orbifold action (4.10), acts nontrivially on all the three cycles y,
and
Thus at
However, the group element n5 n only acts by a 2 rotation on the cycle .
.
r = 0 and = 0, where the other two cycles y and shrink, n5 n has a fixed point and the
6-dimensional space (4.1) has an orbifold singularity. The order of this orbifold singularity
is k /(n5 n) = kn + 1. Similarly, if k and n5 have no common factors, the group element
n5 (kn+1) acts trivially on and has a fixed point when y and shrink, which happens
at r = 0 and = /2. Thus at r = 0 and = /2 there is an orbifold singularity of order
then
k /(n5 (kn + 1)) = n. If n5 and k have a common factor m, i.e., n5 = mn 5 and k = mk,
n 5 (kn+1) has a fixed point at r = 0 and = /2 and the order of the orbifold singularity
increases to k /(n 5 (kn + 1)) = mn.
In conclusion, the metrics obtained from the spectral flow metrics by S and T dualities
have orbifold singularities at both (r = 0, = 0), and at (r = 0, = /2). The order of the
orbifold singularity is kn + 1 at (r = 0, = 0) and mn at (r = 0, = /2), where m is the
highest common factor shared by n5 and k.
Note: In the classical limit of the D1D5 system we take n1 , n5 , though BPS
states exist of course for all n1 , n5 . If we start with a large n5 then the orbifold shifts
involving n15 in (4.15) are very close together, and cannot be seen in the classical limit
n5 . But since microstates exist for all n1 , n5 the geometries studied in this section
can be considered for n5 of order unity, and then they give well-defined classical metrics
with the orbifold group (4.15).
4.2.3. Absence of horizons and closed timelike curves
If we write down a generic 3-charge solution with rotation, we find closed timelike
curves (see for example [19]). But we expect that geometries that actually arise as duals to 3-charge states will be free of pathologies. In [6] computations were developed to
show that the geometries constructed there had no horizons and no closed timelike curves.
A similar result holds for the geometries found in this paper; the computations are given in
Appendix B.
5. Wave equation for a scalar

In [3,20] the wave equation for a massless minimally coupled scalar was studied in the
2-charge geometry. Such a scalar arises for instance from fluctuations of the metric on T 4 ,
for example the component hz1 z2 . It was found that the wavepacket spent a time tSUGRA
440
traveling down and back up the throat of the supergravity solution. The time tSUGRA
exactly equalled the time taken for excitations to travel around the effective string in the
dual CFT. We now compute tSUGRA for the 3-charge solutions we have found, and then
use the result to find the dual state in the CFT.
The wave equation for a massless minimally coupled scalar in the 6D geometry is4

1
2 gg = 0.
(5.1)
g
We give g , det g in Appendix C. Writing

y
t
)
(t, y, r, , , ) = exp i + i + im1 + im2 (r,
R
R
we get a wave equation that is separable in r, [21]. We write
) = H (r)( ).
(r,
(5.2)
(5.3)
We introduce the dimensionless radial coordinate

R2
Q1 Q5
and the following convenient quantities
x = r2
(5.4)

1
2

2 Q1 Q5
=
,
R4
1/2

Q1 + Q5

2 Qp
= 1 + 2 2
)
,
R2
R2

Qp (Q1 + Q5 )

m1 1 m2 2 ,
=
Q1 Q5
= ( + m1 2 + m2 1 ).
= |1 + 2 | = ,
(5.5)
Then the radial and angular part of the wave equation become

d
2
2
2 d
2
2
x x +
H + x +1 +
H = 0,
4
(5.6)
dx
dx
x
x + 2

1 d
d
sin 2
+
sin 2 d
d

m2
m2
1 + 2
2
2
+ 21 22 +
(5.7)
sin
cos
= 0.
1
2
cos
2
sin
Reality of the metric implies that the wave equation is real; thus the complex conjugate of
a solution gives another solution. We can thus take
0.
(5.8)
The solutions with < 0 are obtained by complex conjugation.

4 The 6D string metric is obtained by ignoring the T 4 , and the 6D Einstein metric turns out to be the same as
the 6D string metric.
441
5.1. Solving the wave equation

The radial wave equation can be solved in the two regions
outer region:
x 1,
inner region:
x 2.
(5.9)
If one chooses the frequency of the scattering wave to be very low

2 1
(5.10)
the inner and outer regions have a wide overlap, where the two limiting solutions can be
matched. Due to this large overlapping region, we expect that a reliable solution can be
found, in the low frequency limit (5.10), without the need to introduce a further region or
to make any further assumption on the parameters. In particular the quantity
2
Q1 + Q5
Qp
2
(5.11)
+ ( )2 2
R2
R
is not assumed to be small. For large it can be seen from (5.5) that becomes imaginary;
this corresponds to energies where the quantum travels over the potential barrier in the
neck region instead of tunneling through it. Since the CFT is known to describe the low
energy dynamics of the system, we will restrict ourselves to real ; we can choose the sign
of to be positive.
Note also that in the low frequency limit (5.10) the angular part of the wave equation
(5.7) reduces to the angular equation in flat space and thus the eigenvalue is
= l(l + 2).
(5.12)
The technique of matching solutions across the two regions is well known, and details
of the computation are given in Appendix D. Here we outline the main steps and results.
The solution in the outer region is a linear combination of Bessels functions

1
Hout = C1 J 1 x + C2 J 1 x .
x
(5.13)
The coefficients C1 and C2 are fixed by demanding continuity with the inner solution. The
latter is uniquely determined by the requirement of regularity at x = 0 and is given in terms
of the hypergeometric function

x
Hin = x x + 2 F p, q; 1 + 2; 2
(5.14)
with
| |
=
| + m1 2 + m2 1 |,
=
2 2

Qp (Q1 + Q5 )
=
m1 1 m2 2 ,
2
2
Q1 Q5
1
p= ++ + ,
q = ++ .
2
2
2
2
(5.15)
442
As a result of the matching we get

C2 ( + 1)() ( 12 + + + 2 )( 12 + + 2 ) 2
=
.
C1
( + 1)() ( 12 + + 2 )( 12 + 2 ) 4 2
(5.16)
Note that since 2 < 1 2 we have C2 C1 .

5.2. Time of travel and absorption probability
For very large x we get from (5.13)

2 1 i( 1 x )
4
Hout =
e
C1 ei 2 + C2 ei 2
3/4
x
1

1 x )
4
C1 ei 2 + C2 ei 2
1 + O x 2 .
+ ei(
(5.17)
The ratio between the outgoing and the ingoing wave amplitude is (ignoring the constant
phase shift caused the by factors 4 )
R=
C1 ei
C1 e i
+ C2 e i
+ C2 ei
2

C2

C2
= ei + 1 e2i
+O
.
C1
C1
(5.18)
In [22] a procedure was given to compute the travel time in the throat from R; we
summarize the method here. The quantum coming in from infinity tunnels through the
neck region with some probability p 1 and enters the throat. Here it travels freely
down to the cap and bounces back up. We again have the same probability p that it
emerges to infinity, while with probability 1 p it turns back for another trip in the throat.
We thus get emergent waves at times separated by a fixed interval tSUGRA .
To find p and tSUGRA we note that R can be written in the form
R=a+b
e2in R t ,
(5.19)
n=1
where a and b are some real functions of . Let us send in from infinity a wave packet

kr
dkr f (kr )ei R ri R t ,

(5.20)
where kr /R is the radial wave number kr = 2 2 . After scattering from the geometry
the wavepacket will be

kr
kr

dkr f (kr ) ei R ri R t + Rei R ri R t .
(5.21)
443
From the form (5.19) of R we see that the wave packet will have peaks at
kr r = (t 2nt),
n = 0, 1, . . . .
(5.22)
The n 1 peaks represent waves that have travelled n times down the throat and back, and
we identify t = tSUGRA . From (5.19) we also see that the probability to enter the throat
and reemerge is
P = p 2 = |b|2 .
(5.23)
Note. In order to be able to distinguish between successive peaks of the wavepacket, the
separation between the peaks (seen at infinity) should be larger than the width of the
wavepackets. This implies

t
kr
R
1
R
.
kr
(5.24)
Our approximate solution(5.16) is only valid in the low frequency limit kr R 2 / Q1 Q5 ,

so we should have t Q1 Q5 /R. We will find that t R/ so the requirement (5.24)
is

R2
Q1 Q5 .
(5.25)
The above condition can be satisfied either by taking R large or small.

We now list the results found in Appendix E.
5.2.1. Energy threshold for absorption
If the wave frequency is low enough so that
<+
+1
2
(5.26)
then the wave is reflected back at the neck region and the absorption probability vanishes.
We can interpret this threshold in the following way. If the energy of the quantum is
low enough then we cannot fit a complete wavelength in the throat. In the limit of large
R the throat is a large asymptotically AdS region. The spectrum of a scalar in such an
asymptotically AdS geometry was computed in [23]. Adapting those results to the notation
of our paper, we find that the energy levels are given by solutions to the equation
=+
l+2
+ k,
2
k = 0, 1, 2, . . . .
(5.27)
If we take the limit R then we get a large AdS region, and we can compare quantities to those computed in asymptotically AdS space. Taking this limit we see that the
frequencies (5.26) which are not absorbed are those which lie below the AdS3 excitation
spectrum.
444
5.2.2. Travel time in the geometry

If is high enough so that
+1
(5.28)
2
then the reflection amplitude computed in (E.6) has a form similar to (5.19) with the sum
over phase factors
>+
e2in(
1+
2 )
n=1
e2in R t .
(5.29)
n=1
The factor depends linearly on ; the parameter , on the other hand has a nonlinear
dependence on that will distort the wave packet. Note however that

l+2
1 + | | (1 + )
=
=
1 + O() ,

(5.30)
2
2
2
where
Q1 + Q5

Qp
(5.31)
(l + 1)2 2 2
( )2 2 .
R2
R
We take all the Qi to be of the same order. We see from (5.25) that travel time makes good
i
i
1. Thus either 1 or Q
1 (or both). In either case we find that
sense only if Q
R2
R2
1. We then find
(l + 1)
l+2
1+

= + ( independent terms).
2
2
2
(5.32)
In this limit the wave packet is not distorted and it travels up and down the throat in the
time
R
R
t = =
(5.33)
.
5.2.3. Absorption probability

The probability for the wave to be absorbed and reemitted in the throat is found to be
2 4

4 2
P=
2
2 () 2 ( + 1)

( 12 + + + 2 )( 12 + + 2 ) 2
.
(5.34)
r( 12 + + 2 )( 12 + 2 )
The probability for just absorption or just emission is p = P .

5.2.4. The factor as a redshift
In [3] the travel time was computed for 2-charge D1D5 geometries, and was found to
be
R
.
tSUGRA =
(5.35)
445
This is seen to differ by a factor from the 3-charge (5.33). We offer a simple physical
explanation for this factor.
Consider first the D1D5 system. In the dual CFT the absorption of the quantum is
described by the creation of a set of left and a set of right moving excitations, which
travel at the speed of light around the component string in a time tCFT = tSUGRA .
Now suppose we have a P charge as well. This corresponds to the presence of left movers
on the component strings. But the left and right excitations in the CFT travel around the
component without interacting with each other, so one may think that one again gets the
same t as in the 2-charge case and thus the value of the P charge does not enter t.
But this cannot be right, since the D1, D5, P charges can all be permuted by duality.
Indeed the factor makes t in (5.33) invariant under such permutations. To understand
the role of consider the limit of the metric (2.46) for small r and small conical defect

Qi
R2
.
(5.36)
Q1 Q5
In this case we have a large AdS type region which would possess a CFT dual description.
In this limit f Qi and thus one can replace Hi by Qi /f , obtaining an asymptotically
AdS3 S 3 geometry. As is clear from our computation above, the time of travel is dominated by the time spent by the wave in this part of the geometry. Let us look at the form of
the metric (2.46) in the near horizon limit (5.36). We get
r
(i = 1, 5),

ds 2
d 2
+ 2 d y 2 + d 2
n.h. = 2 + 2 ( d t)2 + 2
+2
Q1 Q5
+ cos2 d 2 + sin2 d 2 ,
where we have made the following coordinate redefinitions

Qp (Q1 + Q5 )
r 2 R2
t
1
y
2 =
,
t = ,
t ,
y =
Q1 Q5
R
R
Q1 Q5

Qp (Q1 + Q5 ) t
y
= 1 2
2 ,
Q1 Q5
R
R

Qp (Q1 + Q5 ) t
y
1 .
= 2 1
Q1 Q5
R
R
(5.37)
(5.38)
2 is locally AdS S 3 with curvature radius

As anticipated, the near horizon metric dsn.h.
3
(Q1 Q5 )1/4 , but the time is rescaled by with respect to the time t at asymptotically flat
infinity. Thus the time computed in the CFT will be a factor of times the time between
wavepackets measured at infinity. If we take a limit R keeping other parameters
fixed then Qp 0, 1 and we can directly compare the CFT time to the gravity time.
We will take such a large R limit in the more detailed analysis of the CFT state below.
6. Finding the CFT duals

We started with CFT states (2.15) and found their gravity duals (2.46). We then made
new solutions by applying STy Tz1 S duality to (2.46) and obtained the solutions (3.2). What
446
are the CFT states dual to the geometries (3.2)? We use two closely related tools to identify
the CFT duals: the time of travel tSUGRA down the throat of the supergravity solution, and
the threshold of absorption into this throat. Longer travel times map to longer components
of the effective string in the CFT, and lower absorption thresholds also reflect longer
effective strings.
6.1. Time of travel
In [3] it was found that for the 2-charge D1D5 geometries (2.36) a quantum falling
into the throat emerges after a time
tSUGRA =
(6.1)
or its integer multiples. In the dual CFT the corresponding state is (2.21), where twist
operators have joined together
k=
(6.2)
copies of the CFT together to create effective strings of length 2Rk = 2R

. In the gravity picture a quantum can fall down the throat of the geometry; in the dual CFT description
the energy of the quantum gets converted to a set of left and a set of right moving vibrations on the effective string [24,25]. These vibrations travel at the speed of light and meet
halfway around the effective string, so that the energy can leave the effective string after a
time
1
tCFT = 2Rk = Rk
2
(6.3)
in exact agreement with (6.1). (If the vibrations fail to collide and leave the string, then
they encounter each other again after a time (6.3), etc., this corresponds to the successive
waves emerging at separations tSUGRA in the gravity picture.)
Note that the number of effective strings created by the twists is
m=
n1 n5 N
= .
k
k
(6.4)
Recall that each connected piece of the effective string is termed a component string.
6.1.1. The number of component strings m for the 3-charge states (3.2)
The 3-charge geometries (2.46) had D5, D1, P charges (n5 , n1 , np ). The orbifold CFT
had N = n1 n5 , the winding number of each component string was k, and thus the number
of component strings was
m=
N
n1 n5
=
.
k
k
(6.5)
447
By the duality STy Tz1 S we obtained the geometries (3.2) which had charges5
n5 = n5 ,
n1 = np ,
np = n1 .
(6.6)
The orbifold CFT now has

N = n1 n5 = n5 np .
(6.7)
The travel time for such geometries is derived in (5.33). Take the limit R ; this gives
a large AdS region which will be the dual of the CFT. We get Qp 0, 1 and the
travel time is
tSUGRA =
R
= R n5 n(nk + 1)

(6.8)
where we used (4.12). Thus we expect the winding number of each component string to be
k = k
np
= n5 n(nk + 1).
n1
(6.9)
The number of components of the effective string are then

m =
N n5 np n1
n1 n5
.
=
=
k
k np
k
(6.10)
We thus see that while N, k change to N , k , the number of components of the effective
string remains unchanged
m = m.
(6.11)
To interpret this fact we recall the physical significance of m found in [3]. When we put
a particle in the throat of the gravity solution then we excite left and right movers on one
component string in the dual CFT. Putting another particle in the throat excites another
component string, and so on. Let each particle have the longest possible wavelength which
can still fit in the throat of the geometry. When we have enough particles in the throat so
that we use up all the m component strings in the CFT then we find that we have enough
energy in the supergravity solution to give a backreaction of order unity in the geometry;
the geometry distorts and we can no longer study the particles as independent excitations.
We now see that under S, T dualities this critical number of particles for the geometry
stays unchanged.
5 S, T dualities certainly map a state of the 3-charge system to another state of the system, but after duality
we may not be in a range of parameters where the state is welldescribed by the conformal field theory. The
Q Q
1 5
. If we start with a large circle radius
conformal limit is the low energy limit, and is attained for small
R
R then after dualities we get small R . But since we are dealing with BPS states we can follow the state as we
increase R and get back to a CFT domain. It is this latter CFT state that we will mean when we look at the states
after STy Tz1 S duality.
448
6.1.2. Level of excitation of the component strings

The CFT state for the geometries (2.46) had np units of momentum distributed equally
over m component strings; thus each component had
np
(6.12)
m
units of momentum. It will be helpful to write this in terms of the basic units of momentum
on the component string. Since the string has winding number k the excitations come in
units of k1 . Thus the number of these basic units of momentum on each component string
is
T=
T = kT =
np k np N
n1 n5 np
=
=
.
m
m m
m2
(6.13)
For the state after STy Tz1 S duality we have np = n1 units of momentum, distributed over
m = m component strings, and
T =
n1 n5 np
n1 np
T = k T = k
=
m n1
m2
n1
,
m
(6.14)
so we see that
T = T =
n1 n5 np
.
m2
(6.15)
Thus T is a duality invariant; the charges permute under dualities and we have seen that m
stays unchanged. We note that the number of possible states on each component string depends on the number T : we have to just excite free bosons and fermions on the component
string so that the total level (as measured in units of the basic excitation on the component
string) is T . Thus we see that under STy Tz1 S duality the number of allowed states on the
component strings remains unchanged.
6.1.3. The state after STy Tz1 S duality
The angular momentum of the state does not change under the STy Tz1 S duality (the T
dualities are all along compact directions, while angular momentum reflects the properties
of the state in the noncompact directions). For the entire state we have from (2.16)

n1 n 5
1
n1 n5
j =
(6.16)
2n +
,
j =
.
2
k
2k
Both before and after the STy Tz1 S duality we have m component strings, so the angular
momentum on each component string is
1
j = nk,
2
1
j = .
2
(6.17)
Both before and after the duality we have the same level of excitation T on each component string. The state before the duality was given by (2.33)the charge (j, j) was attained
by the lowest energy possible by having all fermion spins aligned and all fermions in the
lowest levels allowed by the Pauli exclusion principle. We see that the only way to make
the state after STy Tz1 S duality is to make a construction similar to (2.33)

J2n
J k 1 J 2 (k , 0) .
2n
k
k
449
(6.18)
The component strings now have winding k each instead of k, but the rest of the construction is the same.
6.2. Absorption threshold
In Section 5 it was found that an incoming wave was absorbed only if
+1
(6.19)
.
2
Take the CFT limit R which gives Qp 0, 1. The angular momentum components m1 , m2 imply for the two su(2) factors the eigenvalues
>
m1 + m2
m1 m2
,
m
=
.
(6.20)
2
2
The condition (6.19) then gives a pair of conditions from the two possible signs of the
absolute value in

j
j
j
j
2 m +m
+2 m m
> (l + 2),
N
N
N
N

j
j
j
j
2 m +m
(6.21)
+2 m m
> (l + 2).
N
N
N
N
m=
Here (j, j) are the angular momenta of the geometry into which the quantum is falling

j
j
= n+
= .
(6.22)
,
N
2
N
2
Note that
2 are the increments of L0 , L0 caused by the incoming quantum, so we can
write (6.21) as

l
m 2mn,
h > +
(6.23)
2

l
h > +
(6.24)
m
.
2
6.2.1. Absorption of quanta in the CFT description
In [2426] the CFT description of absorption was studied. We have supposed that the
quantum being absorbed is a scalar arising from the component hij of the metric on the
T 4 . The quantum has angular momentum l, which means that it is in the representation
( 2l , 2l ) of su(2)L su(2)R . In the CFT state this quantum creates an excitation that has
X, ,
X,
(6.25)
450
where the X variables carry the indices i, j (we must symmetrize over the two permutations) and there are l fermions on each of the left and right sectors. The fermions are in the
Ramond sector, and thus both fermionic and bosonic excitations come in multiples of the
basic harmonic on the component strings
h = h =
1
= .
k
(6.26)
There are two species of

The (J 3 , J3 ) quantum numbers in su(2)L su(2)R are (m, m).
left moving fermions with m = 12 and two with m = 12 ; similarly for the right movers.
We will simply write , without distinguishing the two species since the difference
is irrelevant for the discussion below. Thus on the left sector we must have 2l + m fermions
+ and 2l m fermions . On the right sector we have 2l + m
fermions + and 2l m
fermions .
6.2.2. Identifying the excitations
Recall the structure of the CFT state (2.33) into which we are absorbing the quantum.
On each component string we have fermion zero modes. For the left movers we started by
choosing the state which is killed by the 0 , and then applied operators J which resulted
in the application of modes 1 n

(for both species of ). For the right movers we
k
also have the vacuum killed by the 0 , but applied no other excitations.
needs a minimum h of 1 = ; this
Consider first the right movers. The excitation X
k
is the first term on the RHS of (6.24). Now consider the fermions. Suppose that m
= 2l .
Then we have l operators + acting on the state of the CFT. But we can choose each of
these operators to be a zero mode 0+ which changes the vacuum on a component string
to one that is killed by 0+ . The fermions then do not contribute to h and we find exact
agreement with (6.24). Note that we could find at most two such zero modes on any given
component string, so we will have to apply the 0+ in general to many different component
strings.6
Now suppose that m
= 2l 1. We have l 1 operators + and one . The + again
but the lowest allowed mode for is 1 and we again
give no contribution to h,
k
get agreement with (6.24). Proceeding this way, we find agreement for all m
for the right
movers.
Now consider the left movers, and let m = 2l . We have to apply l operators + . The
, which
lowest excitation results if we use the operators + to annihilate l modes n
1
gives a total h = k nl (including the boson X) which agrees with (6.23). Next con
and
sider m = 2l 1. Now we have l 1 operators + which again annihilate modes n
,
again
in
exact
agreeone operator which creates the lowest allowed mode
1
(n+ k )
6 We can read off the value of = 1 from the classical geometry, for example by the conical defect angles of
k
n n
the metrics (2.46). But n1 , n5 are infinite in the classical limit, and so m = 1k 5 is also infinite in this limit. Thus
the absorption thresholds computed from the classical geometry have an essentially infinite number of component
strings, and we will not run out of component strings for any value of l that we choose.
451
ment with (6.23). (The operator cannot just fill in one of the empty levels created by
since the overall operator describing the absorption is taken to

annihilation of modes n
be normal ordered.)
We thus see that the absorption threshold seen in the wave equation analysis can be
understood in detail in terms of the occupied levels on the effective strings. This computation supports the conjecture that the state after duality has the form (6.18) analogous to the
states before the STy Tz1 S duality; the absorption computation applies equally to both sets
of geometries.
6.3. A puzzle about the orbifold theory
Consider the CFT state (6.18) dual to the geometries obtained after STy Tz1 S duality.
The momentum T per component string (given in (6.14)) is fractional in general, not an
integer. Interestingly, when this momentum is measured in units of the basic excitation 1k
on the component string then we get an integer T . (If T were nonintegral, we would have
a severe contradiction, since we could not carry the excitation on the component string.)
But here we face a puzzle, since in the orbifold CFT the quantity T is also required to
be integral. The reason for this is as follows. We have to orbifold by the symmetric group
SN which permutes the N copies of the c = 6 CFT. In a given state of the CFT we can
label the copies by how they make up the different component strings, and even inside
each component string the copies can be ordered by the sequence in which they link up to
make the long cycle. But one part of the symmetry group still survives: we can cyclically
permute the copies inside a component string
c1 c2 ck c1 .
(6.27)
This symmetry forces the momenta on the component string to be integral.

Faced with this problem, we first review the steps that led us to our solutions. The
fundamental string (F) is an elementary excitation of the theory on M4,1 S 1 T 4 , so
the 2-charge FP solutions we started with certainly correspond to BPS states of the full
string theory. Since S, T dualities are exact symmetries of the theory the 2-charge D1D5
states are also states of the theory. Spectral flow was just a coordinate change, and so the
3-charge solutions that were obtained by spectral flow must also be valid. Finally, the exact
symmetry STy Tz1 S was used to get the solutions which we are now discussing, so we
conclude that they must be allowed states of the string theory.
Even though one may accept the gravity solutions one may question the identification
of the CFT state. We have made each component string have the same winding number
k , but if we allowed the winding numbers to be different then we could have carried the
momentum np = n1 on the component strings in such a way that each component string
had an integral number T of units of momentum. But let us recall the evidence we have that
the component strings should all be equal: (a) We have learn from the explicit construction
of 2-charge systems that states with axial symmetry have all component strings identical;
if the component strings are different then all symmetries are broken in general. (b) The
return time from the throat gave a precise value that could be put in correspondence with the
length of the component strings; if component strings had different lengths then we would
get distortion of the wavepacket since different parts would be returned after different
452
times. (c) The threshold of absorption worked out exactlyif we change the structure of
the filled levels on the component strings then the allowed levels that could be excited by
the incoming quantum would change. For all these reasons it seems hard to have any other
construction of the CFT state.
It may be that we need changes in our understanding of the CFT dual to the gravity
theory. It is not completely clear where the orbifold point sits in the moduli space, though
there are some leading candidates [27,28]. It is also not clear if the orbifold (T 4 )N /SN does
describe some point in the moduli space, or if we need to consider other related orbifold
theories like the iterated orbifolds [29].
Before concluding we note a point about the winding number of the component strings.
The geometries (2.46) had an integral k, since this number can be traced back to the number
of turns of the helix of the F string in the starting FP solution that led to (2.46). For the
geometries obtained after STy Tz1 S duality we still found in Eq. (6.9) that the winding
number of each component string k was integral. This was important, since if k turned out
fractional we could make no sense of the CFT state. But if we assume that the quantities T
do not need to be integral and only the T need to be integral in general, then after STy Tz1 S
duality we would get fractional k in general. What then are the rules for the allowed CFT
states?
For the D1D5 CFT we can have m component strings with equal winding if
n1 n5
Z.
(6.28)
m
Given that m was found to be a duality invariant, and that n1 , n5 , np permute under duality,
we conjecture that the state can have m equal length component strings if all the following
conditions are true
n5 np
np n1
n1 n5 np
n1 n5
Z,
Z,
Z,
T =
Z,
(6.29)
m
m
m
m2
where we have included the requirement that T be integral. We hope to return to these
issues elsewhere.
7. Discussion
Our basic conjecture states that the black hole interior is not empty space with a central
singularity but a fuzzball of horizon size. If we consider extremal holes, and look at
states where in the dual CFT we have many component strings in the same state, then we
can have a good description of the geometry in classical supergravity. All 2-charge states
could be approached through such classical geometries, and in [5,6] some classes of 3charge extremal states were considered and their dual geometries identified; the geometries
were smoothly capped as in Fig. 1(b). In the present paper we have looked at two families
of states and their dual geometries. The geometries were again found to be capped; there
were no horizons or closed timelike curves.
The first family arose from spectral flow of a subfamily of 2-charge states. These 2charge states generically had an orbifold singularity along a curve; this was understood
as a trivial singularity in the sense that it arose from the coincidence of two or more
453
KK monopoles tubes and was thus arose only as a limit in a family of regular solutions. Since spectral flow is given by a coordinate transformation in AdS3 S 3 [11,12]
it is not surprising that a similar orbifold singularity arose also for the 3-charge states
obtained by spectral flow of the 2-charge states. What was interesting to note was that
the conical defect angle did not get corrected when the asymptotically AdS solution is
modified to become an asymptotically flat solution. If the conical defect parameter had
changed away from the form k1 to say an irrational value then we would not be able
to understand the orbifold singularity in any simple way as a limit of nonsingular solutions.
The second family we considered was found by applying S, T dualities to the first
family of geometries so that the D1 and P charges got interchanged. We identified the CFT
states dual to this second family of geometries by computing the time of travel tSUGRA
for a quantum to fall down the throat and bounce back out. The winding number of the
component strings in the CFT, computed by this method, gave a minimum threshold
energy for excitations of the CFT state. This minimum energy agreed with the threshold of
energy below which the incident quantum was unable to enter the throat of the geometry,
thus confirming the identification of the CFT state. We noted that the CFT states found
this way had a fractional momentum on each component string (the total momentum
was of course an integer). This suggested that we need to go beyond the simple orbifold
CFT to understand all 3-charge bound states; we may need to understand deformations
away from the orbifold point [30] or perhaps we may have to consider iterated orbifolds
[29].
The conical defect in the 2-charge D1D5 geometries could be directly linked to the
winding number k of each component string. For the metrics obtained by spectral flow
from 2-charge geometries we have a similar relation since spectral flow is just a coordinate transformation on the geometry [11,12]. The conical defect is caused by an orbifold
singularity of order k along the S 1 given by (r = 0, = /2) and the component strings
in the dual CFT state have winding number k. For the geometries obtained after S, T
dualities the situation is more complexwe have orbifold singularities along two different S 1 curves, with order kn + 1 at (r = 0, = 0) and order mn at (r = 0, = /2) (m
is the highest common factor shared by n5 , k). There is a suggestive pattern though that
we observe. It was argued in [30] that the D1D5 CFT with charges (n1 , n5 ) could be
mapped to the theory with charges (n1 n5 , 1), and that the orbifold point occurs at this latter value of the charges. If we set n5 = 1 to enable a comparison to the orbifold theory,
then we observe that the product of the orders of the two orbifold groups is (kn + 1)n
which in this case equals k , the winding number of the effective string in the dual CFT
state.
We have noted that the travel time is symmetric between the three charges of the solution, but the factor enforcing this symmetry is responsible for providing a redshift
which relates the time at infinity to the time in the AdS region. The AdS region is the one
that is dual to the CFT description, so understanding such effects in more detail may help
us to identify the deformations in the CFT that correspond to deforming AdS space to flat
space at infinity.
454
Acknowledgements
S.G. was supported by an INFN fellowship. The work of S.D.M. was supported in part
by DOE grant DE-FG02-91ER-40690. We thank Yogesh Srivastava for helpful discussions.
Appendix A. Non-extremal rotating 3-charge metrics and their extremal limits

The metric for general rotating 3-charge solutions was given in [13], and the metric, 2form field and dilaton were given in [6] by generating the solution by a different technique.
We have

M cosh2 p
M sinh2 p
dt 2
dy 2
2
ds = 1
+ 1+
f
f
H1 H5
H1 H5

2
2

M sinh 2p
r dr
2

dt dy + f H1 H5
+
d
f H1 H5
(r 2 + a12 )(r 2 + a22 ) Mr 2

(a 2 a12 )K1 K5 cos2
+ r 2 + a12 H1 H5 + 2
cos2 d 2
H1 H5

(a 2 a22 )K1 K5 sin2
+ r 2 + a22 H1 H5 + 1
sin2 d 2
H1 H5

2
M
a1 cos2 d + a2 sin2 d
+
f H1 H5
2M cos2
(a1 cosh 1 cosh 5 cosh p a2 sinh 1 sinh 5 sinh p ) dt
+
f H1 H5

+ (a2 sinh 1 sinh 5 cosh p a1 cosh 1 cosh 5 sinh p ) dy d
2M sin2
(a2 cosh 1 cosh 5 cosh p a1 sinh 1 sinh 5 sinh p ) dt
f H1 H5

+ (a1 sinh 1 sinh 5 cosh p a2 cosh 1 cosh 5 sinh p ) dy d

4
H1 2
++
dxi ,
H5
i=1
C2 =
M cos2
(a2 cosh 1 sinh 5 cosh p a1 sinh 1 cosh 5 sinh p ) dt

+ (a1 sinh 1 cosh 5 cosh p a2 cosh 1 sinh 5 sinh p ) dy d
f H1
M sin2
(a1 cosh 1 sinh 5 cosh p a2 sinh 1 cosh 5 sinh p ) dt
f H1

+ (a2 sinh 1 cosh 5 cosh p a1 cosh 1 sinh 5 sinh p ) dy d
M sinh 21
dt dy
2f H1

M sinh 25 2
r + a22 + M sinh2 1 cos2 d d,
2f H1
H1
=
.
H5
455
e2
(A.1)
Here
f = r 2 + a12 sin2 + a22 cos2 ,
Hi 1 + Ki = 1 +
M sinh2 i
,
f
i = 1, 5.
(A.2)
In terms of the parameters appearing in the metric, the D1, D5 and momentum charges
and the two angular momenta are
M
M
M
sinh 21 ,
sinh 2p ,
sinh 2p ,
Q5 =
Qp =
2
2
2
J = M(a1 cosh 1 cosh 5 cosh p a2 sinh 1 sinh 5 sinh p ) (5)

4G
Q1 Q5
= 1
,
4G(5)
J = M(a2 cosh 1 cosh 5 cosh p a1 sinh 1 sinh 5 sinh p ) (5)

4G
Q1 Q5
= 2
4G(5)
Q1 =
(A.3)
(A.4)
with G(5) the 5D Newtons constant. The extremal limit is the limit in which M 0 while
Q1 , Q5 , Qp , 1 , 2 are kept finite. We will give the extremal metric for generic values of
the charges and of the angular momenta satisfying
(1 2 )2 4Qp 0.
In this case we find it convenient to parametrize Qp as

1 2 2
1 + 2 2
Qp =
, > 0.
2
2
The extremal metric, Ramond field and dilaton are

Qp
dr 2
1 2
2
dt dy 2 +
(dt dy)2 + hf 2
+
d
h
hf
r + 1 (1 + 2 )2

(1 + )1 + (1 )2
+ h r 2 + (1 + 2 )
2

2
2
2
( 2 )Q1 Q5 cos
cos2 d 2
1
h2 f 2

(1 )1 + (1 + )2
+ h r 2 + (1 + 2 )
2
2
2
2
( 2 )Q1 Q5 sin
sin2 d 2
+ 1
h2 f 2
ds 2 =
(A.5)
(A.6)
456
2
Qp (1 + 2 )2 2 2
cos d + sin2 d
hf

2 Q1 Q5
hf

4

H1 2
2(1 + 2 ) Q1 Q5 2
2
cos d + sin d dy +
(A.7)
dxi ,
hf
H5
i=1
Q1 Q5 cos2
(2 dt + 1 dy) d
C2 =
H1 f
Q1 Q5 sin2
(1 dt + 2 dy) d
H1 f

(1 + 2 )Qp
+
(Q1 dt + Q5 dy) cos2 d + sin2 d
Q1 Q5 H1 f
Q1
dt dy
H1 f

Q5 cos2 2
(1 )1 + (1 + )2
r + (1 + 2 )
+ Q1 d d,
H1 f
2
(A.8)
H1
2
e =
(A.9)
,

H5
+
(1
(1
+
)
1
2
sin2
f = r 2 + (1 + 2 )
2

(1 )1 + (1 + )2
cos2 ,
+
2

Q1
Q5
,
H5 = 1 +
,
h = H1 H5 .
H1 = 1 +
(A.10)
f
f
The metric (A.7) with = 1 can be obtained from the metric (2.46), which corresponds to
the case = 1, via a boost in the y direction:
+
t t cosh + y sinh ,
y y cosh + t sinh
(A.11)
with
e2 =
1
+ 1.
(A.12)
Appendix B. Singularities, closed timelike curves and horizons

The determinant of the metric (A.7)
g = hf sin cos r
(B.1)
only vanishes at r = 0 and = 0, /2. Note that = 0, /2 are the points where spherical
coordinates degenerate, so singularities of (A.7) can only occur at r = 0. In a neighborhood
457
of r = 0 y can be decoupled from all the other coordinates by the coordinate change
t = t ct y,
= c y,
= c y
(B.2)
with
( 1)[2 (1 2 )2 (1 + 2 )2 + 22 Q1 Q5 ]
,
ct =
( 1)[1 (1 + ) + 2 (1 )][1 (1 ) + 2 (1 + )] + 22 (1 + ) Q1 Q5
2 [1 (1 ) + 2 (1 + )]
c = 2
,
( 1)[1 (1 + ) + 2 (1 )][1 (1 ) + 2 (1 + )] + 22 (1 + ) Q1 Q5
2 [1 (1 + ) + 2 (1 )]
.
c = 2
( 1)[1 (1 + ) + 2 (1 )][1 (1 ) + 2 (1 + )] + 22 (1 + ) Q1 Q5
In the new coordinates, the part of the metric that involves y and r is
2

hf
2
dsry
(B.3)
dr + cy2 r 2 dy 2 + O r 4
=
2
(1 + 2 )
with
4(1 + 2 )2
.
( 1)[1 (1 + ) + 2 (1 )][1 (1 ) + 2 (1 + )] + 22 (1 + ) Q1 Q5
(B.4)
From the expressions above we see that, unless = 1, the transformation (B.3) involves a
shift of t by the periodic coordinate y. At r = 0 the y circle shrinks. Thus, as explained in
[6], metrics with = 1 have closed timelike curves.
The geometries (2.46) dual to CFT states have = 1. (The condition = 1 is seen to
give, using (A.6), the relation Qp = 1 2 .) For these metrics one can show that there are
no closed timelike curves by looking at the determinant g of the metric restricted to the
three periodic coordinates y, and :

2

Q1 Q5
r 2 sin2 cos2
2
r + (1 + 2 ) (f + Q1 + Q5 + Qp ) +
.
g =
(Q1 + f )(Q5 + f )
(B.5)
Using an argument given in [6], it is enough to show that the determinant above never
vanishes to prove that the metric is free of closed timelike curves. By explicit inspection,
we know that the zeros of g at r = 0 or = 0, /2 do not signal the presence of closed
timelike curves. So we need to show that g has no other zeros. This follows from the fact
that f + Qi > 0 for i = 1, 5, p. In order to prove this last statement, consider first the
geometries obtained by spectral flow, which have
cy =
1 = an,
2 = a(n + ),
Qp = a 2 n(n + ).
(B.6)
Since n Z and < 1 we have Qp 0 and thus
Q1 Q5
Q1

Q1 Q5 + (Q1 + Q5 )Qp
Qp
(B.7)
458
and
1.
(B.8)
Let us look at f + Q1 and distinguish the two cases n > 0 and n < 0. (It will be very
important here that n is an integer and that 0 < < 1.) If n > 0

a 2 n
2
= Q1 1
> 0.
f + Q1 Q1 a n Q1 1
(B.9)
Qp
n+
If n < 0 (and thus n + < 0)

a 2 (n + )
= Q1 1 +
> 0.
f + Q1 Q1 + a (n + ) Q1 1 +
Qp
n
2
(B.10)
In both cases we used (B.7). The symmetry of the metric under interchange of Q1 and Q5
then also implies f + Q5 > 0. Look now at f + Qp . If n > 0

f + Qp Qp a 2 n = a 2 n n + (1 ) > 0
(B.11)
as 1 > 0. If n < 0 (and thus n + < 0)
f + Qp Qp + a 2 (n + ) = a 2 (n + )(n + ) > 0
(B.12)
as n + n + < 0. In order to prove that the same results hold for the metrics after S,
T dualities it is enough to notice that f is duality invariant, since the transformation of 1
and 2 (3.4) cancel that of :
=
Qp Q5
Qp
=
.
Q1 Q5 + (Q1 + Q5 )Qp
Q1
(B.13)
One can also verify that the metrics (2.46) do not have any horizon. For this purpose
we again refer to an argument given in [6]: there is no horizon if one can find a timelike
vector in the forward light cone which has a nonzero positive component along the radial
direction. In turn, the existence of such a vector follows from the fact that the determinant
of the metric restricted to the t, y, , coordinates

g = r 2 r 2 + (1 + 2 )2 sin2 cos2
(B.14)
is always negative (apart from the points r = 0 and = 0, /2, where we know there is no
horizon by direct analysis7 ).
The form of the metric around r = 0 given in (B.3) shows that the metrics (A.7) generically have orbifold singularities. If R is the radius of the y circle, the conical defect
parameter is given by
= |cy |R.
(B.15)
7 The naive geometry with no rotation [31] has + = 0, and this case has a horizon at r = 0; for the
1
2
geometries we constructed as duals of actual microstates 1 + 2 = 0 and the analysis around r = 0 done in
Section 4.2 shows that there is no horizon at r = 0.
459
Appendix C. Inverse metric

The determinant of the extremal metric (2.46) is
g = hf sin cos r.
(C.1)
The inverse of the metric (2.46) is

Q1 Q5 + Q1 Qp + Q5 Qp
1
f + Q1 + Q5 + Qp +
,
g tt =
hf
r 2 + (1 + 2 )2

Q2p
(Q1 + Q5 )2
1
Q1 Q5
yy
g =
,
f + Q1 + Q5 Qp +
2
hf
r2
r + (1 + 2 )2 Q1 Q5

Qp
Q1 + Q5
1+ 2
g ty =
,
hf
r + (1 + 2 )2

22
12
1
1
g =
+
,
hf cos2
r2
r 2 + (1 + 2 )2

22
12
1
1
g =
,
+ 2 2
hf sin2
r
r + (1 + 2 )2

Qp 1
1
g =
,
hf r 2 r 2 + (1 + 2 )2
1
Q1 Q5
,
g t =
2
hf
r + (1 + 2 )2
2
Q1 Q5
g t =
,
2
hf
r + (1 + 2 )2

12
Q1 + Q5
Q1 Q5 2 1
,
g y =
+
hf
r 2 r 2 + (1 + 2 )2 Q1 Q5

22
Q1 + Q5
Q1 Q5 1 1
y
g =
,
+
hf
r 2 r 2 + (1 + 2 )2 Q1 Q5

H5 ij
r 2 + (1 + 2 )2
1
,
g =
,
g xi xj =
g rr =
(C.2)
.
hf
hf
H1
Appendix D. Solution of the wave equation
D.1. Matching region
In the matching region 1 x 2 the wave equation becomes

d
2 d
x
Hmatch + 1 2 Hmatch = 0.
4
dx
dx
A basis of independent solutions is
(1)
Hmatch = x
1
2
Hmatch = x
(2)
+1
2
(D.1)
(D.2)
460
D.2. Outer region

In the outer region x 1 the radial equation is

d
2 dHout
4
x
+ 2 x + 1 2 Hout = 0
dx
dx
(D.3)
with solution

1
Hout = C1 J 1 x + C2 J 1 x .
x
In the matching region this gives

1 2 x r2
1
1 2 x 2
1
+ C2
.
Hout = C1
4
( + 1)
4
( + 1)
x
x
(D.4)
(D.5)
D.3. Inner region

In the inner region, 2 x 1, the wave equation simplifies to

d
2
2
d
x x + 2
Hin + 1 2 +
Hin = 0.
4
dx
dx
x
x + 2
(D.6)
By defining

Hin = x x + 2 F
(D.7)
with
| |
,
=
(D.8)
2
2
the equation above reduces to an hypergeometric equation for F . Of the two independent
solutions, only one is regular at x = 0:

x
Hin = x x + 2 F p, q; 1 + 2; 2 ,
(D.9)
where
1
++ + ,
q = ++ .
2
2
2
2
To get the large x behavior we write

x
F p, q; 1 + 2, 2
p

x
2
(1 + 2)()
F
p,
p
2;
+
1;
=
x
(q)( 12 + 2 ) 2

q
(1 + 2)()
2
x
+
F
q,
q
2;
+
1;
x
(p)( 12 + + 2 ) 2
p=
(D.10)
(D.11)
461
and note that

F (p, q; ; z) =

(p + n)(q + n)
n=0
(p)(q)
() zn
.
( + n) n!
(D.12)
One thus gets (dropping an overall constant)

x 2
()
1
Hin =
1 + O 2 x 1
1
2
x (p)( 2 + + 2 )

2
2 1
x
()
1
+
O
.
x
+
(q)( 12 + 2 ) 2
(D.13)
D.4. Matching the solutions

Matching the coefficients of x (1)/2 between Hout and Hin we find the ratio C2 /C1
given in (5.16).
Appendix E. Computation of travel time and absorption probability

In order to write the reflection amplitude R given in (5.16)(5.18) as in (5.19) it is
important to know the sign of the arguments
+
+1
2
(E.1)
of the gamma functions appearing in (5.16). As we chose 0, + +

positive and thus we are left with only two possibilities.
+1
2
is always
Case 1
+1
> 0.
2
In this case one can use the series expansion
+
(E.2)
(1)k (b a)k B(k, a b + 1, a)zk

(z + a)
= zab
(z + b)
k!
(E.3)
k=0
((a)k is the Pochhammer symbol and B(k, a, b) the Bernoulli polynomial) which is valid
for z + a > 0, to show that R has no oscillating factor like exp(2in/Rt), and thus
there is no absorption.
Case 2
+1
< 0.
(E.4)
2
One can rewrite the ratio of gamma functions in (5.16) with negative arguments as
+
( 12 + + 2 )
( 12 + 2 )
462
( 12 + + 2 ) sin(( +
1
2 ))
1+
2 ))
( 12 + 2 ) sin(( +

( 12 + + 2 ) i i
i
2in( 1+
)
2
=
+ e
e
e
e
( 12 + 2 )
n=1
(E.5)
so that R is brought into the form (5.19)

R = ei +
2
ei
4 2
2 () 2 ( + 1) e2i 1 2
( 12 + + + 2 )( 12 + + 2 )
( 12 + + 2 )( 12 + 2 )
2 1
( 2 + + + 2 )( 12 + + 2 )
4 2 ei
2
() 2 ( + 1) 2
( 12 + + 2 )( 12 + 2 )
e2in(
1+
2 )
(E.6)
n=1
From the equation above we can read off the probability of absorption/emission (5.34)
and the time of travel (5.33).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
S.W. Hawking, Commun. Math. Phys. 43 (1975) 199.

S.D. Mathur, Nucl. Phys. B 529 (1998) 295, hep-th/9706151.
O. Lunin, S.D. Mathur, Nucl. Phys. B 623 (2002) 342, hep-th/0109154.
O. Lunin, JHEP 0404 (2004) 054, hep-th/0404006.
S.D. Mathur, A. Saxena, Y.K. Srivastava, Nucl. Phys. B 680 (2004) 415, hep-th/0311092.
S. Giusto, S.D. Mathur, A. Saxena, hep-th/0405017.
D. Mateos, P.K. Townsend, Phys. Rev. Lett. 87 (2001) 011602, hep-th/0103030;
R. Emparan, D. Mateos, P.K. Townsend, JHEP 0107 (2001) 011, hep-th/0106012.
I. Bena, P. Kraus, hep-th/0402144.
I. Bena, hep-th/0404073.
B.C. Palmer, D. Marolf, hep-th/0403025.
V. Balasubramanian, J. de Boer, E. Keski-Vakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hepth/0011217.
J.M. Maldacena, L. Maoz, JHEP 0212 (2002) 055, hep-th/0012025.
M. Cvetic, D. Youm, Nucl. Phys. B 476 (1996) 118, hep-th/9603100;
D. Youm, Phys. Rep. 316 (1999) 1, hep-th/9710046;
M. Cvetic, F. Larsen, Nucl. Phys. B 531 (1998) 239, hep-th/9805097.
N. Seiberg, E. Witten, JHEP 9904 (1999) 017, hep-th/9903224;
F. Larsen, E.J. Martinec, JHEP 9906 (1999) 019, hep-th/9905064;
J. de Boer, Nucl. Phys. B 548 (1999) 139, hep-th/9806104.
O. Lunin, S.D. Mathur, Commun. Math. Phys. 219 (2001) 399, hep-th/0006196.
O. Lunin, S.D. Mathur, Commun. Math. Phys. 227 (2002) 385, hep-th/0103169.
A. Schwimmer, N. Seiberg, Phys. Lett. B 184 (1987) 191.
O. Lunin, J. Maldacena, L. Maoz, hep-th/0212210.
[19] C.A.R. Herdeiro, Nucl. Phys. B 582 (2000) 363, hep-th/0003063;

C.A.R. Herdeiro, Nucl. Phys. B 665 (2003) 189, hep-th/0212002;
L. Dyson, JHEP 0403 (2004) 024, hep-th/0302052;
N. Drukker, hep-th/0404239;
E.G. Gimon, P. Horava, hep-th/0405019.
[20] I.Y. Park, hep-th/0403218.
[21] M. Cvetic, F. Larsen, Phys. Rev. D 56 (1997) 4994, hep-th/9705192.
[22] O. Lunin, S.D. Mathur, Nucl. Phys. B 615 (2001) 285, hep-th/0107113.
[23] O. Lunin, S.D. Mathur, Nucl. Phys. B 642 (2002) 91, hep-th/0206107.
[24] C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 (1996) 591, hep-th/9602043.
[25] S.R. Das, S.D. Mathur, Nucl. Phys. B 478 (1996) 561, hep-th/9606185.
[26] S.D. Mathur, Nucl. Phys. B 514 (1998) 204, hep-th/9704156;
S.S. Gubser, Phys. Rev. D 56 (1997) 4984, hep-th/9704195;
J.M. Maldacena, A. Strominger, Phys. Rev. D 56 (1997) 4975, hep-th/9702015.
[27] F. Larsen, E.J. Martinec, JHEP 9906 (1999) 019, hep-th/9905064.
[28] J. Gomis, L. Motl, A. Strominger, JHEP 0211 (2002) 016, hep-th/0206166.
[29] S. Gukov, E. Martinec, G. Moore, A. Strominger, hep-th/0404023.
[30] R. Dijkgraaf, Nucl. Phys. B 543 (1999) 545, hep-th/9810210.
[31] J.C. Breckenridge, R.C. Myers, A.W. Peet, C. Vafa, Phys. Lett. B 391 (1997) 93, hep-th/9602065.
463
The analysis of spacetime structure in QCD

vacuum, I: localization vs global behavior in local
observables and Dirac eigenmodes
Ivan Horvth
University of Kentucky, Lexington, KY 40506, USA
Received 26 November 2004; accepted 28 December 2004
Abstract
The structure of QCD vacuum can be studied from first principles using lattice-regularized theory.
This line of research entered a qualitatively new phase recently, wherein the spacetime structure (at
least for some quantities) can be directly observed in configurations dominating the QCD path integral, i.e., without any subjective processing of typical configurations. This approach to QCD vacuum
structure does not rely on any proposed picture of QCD vacuum but rather attempts to characterize
this structure in a model-independent manner, so that a coherent physical picture of the vacuum can
emerge when such unbiased numerical information accumulates to a sufficient degree. An important
part of this program is to develop a set of suitable quantitative characteristics describing the space
time structure in a meaningful and physically relevant manner. One of the basic pertinent issues
here is whether QCD vacuum dynamics can be understood in terms of localized vacuum objects, or
whether such objects behave as inherently global entities. The first direct studies of vacuum structure
strongly support the latter. In this paper, we develop a formal framework which allows to answer this
question in a quantitative manner. We discuss in detail how to apply this approach to Dirac eigenmodes and to basic scalar and pseudoscalar composites of gauge fields (action density and topological
charge density). The approach is illustrated numerically on overlap Dirac zero modes and near-zero
modes. This illustrative data provides direct quantitative evidence supporting our earlier arguments
for the global nature of QCD Dirac eigenmodes.
E-mail address: horvath@pa.uky.edu (I. Horvth).

doi:10.1016/j.nuclphysb.2004.12.038
I. Horvth / Nuclear Physics B 710 (2005) 464484
465
1. Introduction
The studies of QCD vacuum structure using lattice regularization have a long history.
However, only recently did this research took on the challenge to approach the problem independently, without the guidance from a particular idealized model of the QCD vacuum
[15]. In particular, there was a significant progress in the quest to avoid any subjective
processing of configurations dominating the regularized QCD path integral. This progress
happened at two qualitatively different levels. First, it was recognized that the local behavior of Dirac eigenmodes can serve as an ideal unbiased probe of the underlying vacuum
structure [35], which lead to an increased level of activity in this direction [6,7]. The
eigenmode approach, though unbiased, is still indirect since certain theoretical framework
or model is needed to connect the structure of eigenmodes to the structure of underlying
fields [1]. A purely direct approach, wherein one would inspect the structure of local fields
directly in regularized QCD ensemble is not possible yet in general. However, in a rather
surprising turn of events, it was demonstrated [2] that using topological charge density
operator associated with the overlap Dirac operator [8], the fundamental spacetime structure in topological charge density appears, and the direct approach is thus possible (see
also [9]). Moreover, in this framework it is possible to study topological charge fluctuations at arbitrary low-energy scale by employing the effective topological field constructed
via Dirac eigenmode expansion [1].
The drive for model-independent approach to QCD vacuum structure requires the introduction of model-independent concepts in order to characterize this structure. The goal of
this paper is to propose a sufficiently general framework designed for systematic studies focusing on the questions of spacetime localization and/or global behavior in QCD vacuum.
Since the seminal work of Anderson [10], the issues of localization played an important
role in solid state physics. Here one usually means the exponential localization in the wave
function describing an electron in the solid. Localized wave-function is pinned to a point x0
and behaves so that |(x)| ce(xx0 )/0 , where the maximal 0 is the localization scale.
In the localized regime the pinning points are dilute in 3d space and the wave-function effectively has a support on non-overlapping localized regions. When increasing the amount
of disorder the localized regions start to overlap, and the modes can become delocalized
if the linear size of (at least some) path-connected subsets of the support becomes comparable to the size of the sample. The physics of the system changes drastically when
that happens. In the case of QCD vacuum structure the situation is unfortunately more
complicated for several reasons. (1) We cannot assume the existence of exponentially localized structures neither in local composite fields nor in Dirac eigenmodes. Indeed, there
is no direct evidence that exponential localization takes place in QCD vacuum. (2) In case
of certain proposed topological structures the standard nomenclature already violates the
above classification. For example, a single instanton in arbitrary large volume and the associated t Hooft mode would both be classified as global, and yet they are referred to as
a localized structure which can become delocalized for interacting multi-instanton case.
In fact, the nomenclature (especially in lattice QCD vacuum literature) is quite loose and it
is a common practice to refer to almost any peak (or just a lower-dimensional section that
resembles a peak) freely as localized structure. (3) There is no obvious tunable parameter
in QCD (analogous to the amount of disorder in 3d solids) which could be used to iden-
466
tify the potential candidates for localized objects, which in turn could be used for at least
approximate physical analysis of possible delocalization.1 (4) It has been suggested that
vacuum structure in topological density [2] (and the structure reflected in Dirac near-zero
modes [5]) is not localized and cannot even be viewed as delocalized. Rather this structure
is inherently global (super-long-distance) in the sense that one cannot identify any local
parts contained in bounded regions of spacetime that would be physically dominant over
the rest of the global structure.
Given the above, the challenge we face in this work is how to meaningfully decide
the question whether QCD vacuum structure is localized, delocalized or inherently global
without making any a priori assumptions about the nature of the structure. To answer this
question unambiguously is clearly of great importance for our views of QCD vacuum and
for the possible directions of future research. For example, if the topological structure
is inherently global as argued in [2], then the attempts to understand topological charge
fluctuations in terms of lumps concentrated in bounded regions of spacetime will not be
productive. One will have to approach the problem in terms of some fundamental global
structure instead.
The general approach described here is the result of two considerations. (A) The concept of localization relates to the geometric properties of the support. Here support is
the subset of spacetime which, from the physics point of view, dominates the behavior
of some local observable. Assuming that the support can be identified unambiguously, the
question becomes of purely geometric nature. Indeed, consider the three particular cases
schematically depicted in Fig. 1. In case (a) the support 1 consists of disconnected
pieces with the characteristic linear size much smaller than the linear size of spacetime .
We associate this with the case of localized physical structure. In part (b) the support exhibits connected pieces whose linear size is comparable to the size of the spacetime. Such
connected pieces are geometrically global and could mediate physics over very large distances. However, the localized parts can still be recognized within the geometrically global
component, and there might also be isolated pieces that still retain their localized identity.
We associate this with delocalized physical structure. Finally, in part (c) the behavior is
dominated by geometrically global pieces without any distinguishable localized parts that
Fig. 1. The examples of localized (a), delocalized (b), and inherently global (c) structure.
1 Instantons could in principle play the role of such building blocks, but there is a considerable numerical
evidence that they cannot be identified neither in Dirac eigenmodes [5] nor in effective densities [1]. Instantons
are thus unlikely to provide a suitable starting point (basis) for the description of global behavior.
467
could be used for valid physical analysis. The structure is inherently global and can mediate
physics over arbitrarily large distances.2 In Section 2 we will define geometric characteristics that will help to identify these typical situations. (B) Without the existence of a strict
support3 and without exponential localization, the very notion of support is ambiguous. Indeed, it is possible to give many generic definitions with very different results for what the
support is. Therefore, rather than fixing the definition, we advocate the use of spacetime
cumulative function for the local quantity studied. Ranking the spacetime points by the
magnitude of local quantity in question, this function embodies the notion of support and
contains detailed information on how much different parts of spacetime contribute to the
associated physics. In this way we can study the issues of localization as described in (A)
for different levels of saturation. In Section 3 we describe this idea in detail and set up a
general framework that will allow us to decide unambiguously the issue of localization in
QCD vacuum.
In Section 4 we discuss the application of the above ideas specifically to Dirac eigenmodes (both zero modes and near-zero modes), and the basic scalar and pseudoscalar
composites of the gauge fields (action density and topological charge density). We illustrate
our approach explicitly (including some numerical results) on zero modes and near-zero
modes of the overlap operator [8]. These examples show, within the formal framework
developed here, that low-lying Dirac eigenmodes are inherently global in QCD as first
observed in [5]. The full description of these results will be given in the upcoming publication [11].
2. The case of well-defined support

In this section we will assume that for every typical gauge background we were given
a subset S of spacetime which dominates the behavior of some local physical quantity (x) defined on . Such support S could have been determined via exponential
localization or by using some physical input which helped to select it uniquely for typical
configurations contributing to regularized QCD path integral. It should be noted that when
we talk about spacetime we implicitly have in mind a finite hypercubic discretization
of the torus with inherited Euclidean metric. However, the considerations of this section
are independent of this assumption, and one can equally well think of the Euclidean torus
itself. To proceed, it will be useful to first define the notion of spacetime geometric structure as it will be used here and in some other studies of the subject.
2.1. Geometric structure and its global behavior
We adopt a very general definition of geometric structure which can be easily refined
when used in more specific situations. It should be emphasized that the discussion in this
2 For this reason we sometimes refer to such structure as super-long-distance [2].
3 The function f on domain is said to have a strict support on if f (x) = 0 for x and f (x) = 0
1
1
for x 1 .
468
subsection is purely geometrical and independent from the physics context in which it will
be applied.
Definition (Geometric structure). Let be the set of spacetime points. The collection
S = {Si , i = 1, . . . , N } of path-connected subsets Si such that Si Sj is disconnected
for all i = j will be referred to as a geometric spacetime structure.
In case of hypercubic lattice the term path-connectivity means the usual link connectivity. The reason why the above definition focuses on path-connected sets is the underlying
expectation that in physically relevant situations such sets will be associated with mediation of long-distance physics.
We wish to characterize the degree of global behavior in geometric structure S via that
of its connected components Si . To do that, we need a notion of linear size for connected components regardless of their shapes. This is standard for arbitrary subsets of
metric spaces and is defined as a maximum (supremum) over distances between any two
points of the set.
Definition (Linear size). Let be the subset of the Euclidean spacetime . The linear
size of is defined as ( ) max{|x y|: x, y }.
The subset of spacetime will be called global if its linear size is comparable to the
size of the underlying space, i.e., ( ) (). Conversely, will be called localized if
( ) ().
To quantify the degree of global behavior in geometric structure S, it is useful to define the following two characteristics.4 The first is the maximal linear size max (S) of the
connected component, i.e.,

max (S) max (Si ) .
(1)
Si S
The second is the point-wise average

of the linear size over the points of the structure.
More precisely, to every point x S i Si we assign the linear size of the corresponding
connected component,5 and take the average over S

S (x) (Si ), x Si ,
(2)
pta (S) S (x) S .
Note that S (x) is a piece-wise constant function on S. For continuum spacetime manifold
a precise definition of the integrals entering the mean value has to be given for different
classes of subsets Si . However, for the regularized case of discretized torus the above
expression has the usual meaning of the point-wise average.
The purpose of introducing max (S) and pta (S) is that their values can signal different
regimes of global behavior in the structure. Indeed, while max (S) indicates whether there
4 The characteristic of this type has already been introduced in Ref. [2], where its use lead to the proposal that
the newly discovered low-dimensional structure in topological charge density is super-long-distance.
5 Note that we distinguish between the structure S representing the collection of path-connected sets and the

set of all points belonging to the structure S i Si .
469
is any global behavior within the structure at all, pta (S) measures to what extent is such
behavior prevalent in the structure as a whole. Obviously, pta (S) max (S) for arbitrary S.
Let us distinguish three typical situations:
(a) pta (S) max (S) () (localized structure). In this case all connected components behave as localized subsets. Fig. 1(a) represents an example of such geometric
behavior.
(b) pta (S) < max (S) () (partially global structure). Here only parts of the structure behave as strictly global, while parts of it can still have localized character. Fig. 1(b)
illustrates this case.
(c) pta (S) max (S) () (global structure). In this case the structure has to be
viewed as global since it is completely dominated by global connected parts, as exemplified
in Fig. 1(c).
In essence, we are proposing to associate with any geometric structure two non-negative
numbers less than one, namely pta (S)/() and max (S)/(). These values quantify
the global behavior of the structure. We emphasize that the above classification is purely
geometrical and will be useful only to the extent to which the characteristic situations
described above turn out to be typical for spacetime structures occurring in the QCD
vacuum.
2.2. Support from physics considerations
We now come back to the discussion of the situation where the spacetime support
S corresponding to local observable (x) is assumed to be uniquely determined on
physical grounds. It is easy to see that S defines a spacetime structure S as defined in
the previous subsection. Indeed,
S can be uniquely decomposed into maximal
arbitrary
connected subsets, i.e., S = i Si such that Si is connected, and that Si Sj is dis

connected for all i = j . Then S = {Si , i = 1, . . . , N}. We can thus use the geometric
concepts of the previous subsection to characterize the degree of its global behavior. However, since the geometry of S has been determined by the underlying physics, the global
or localized geometric nature of S are expected to be manifestations of qualitatively very
different types of underlying dynamics which we try to understand. We are implicitly making the following associations:
(a) If S behaves as localized geometric structure, then we expect the vacuum objects
driving the dynamics of the gauge field also to be localized, with typical size approximately
given by pta (S ). From the physics point of view we are thus dealing with localized structure.6
6 It is worth emphasizing here that there is a relatively widely-spread confusion in lattice QCD literature about
the use of the term localized. In particular, the notion of localization is frequently interchanged with the potency
of the support. In reality, these two concepts are entirely independent. The support can be a set of measure zero
and yet the structure can still be global. Some global characteristics (such as pta and max ) have to be computed
in order to determine localized versus global nature of the structure.
470
(b) If S behaves as partially global geometric structure, it is an indication that we are

dealing with delocalized physical structure. Indeed, the presence of global connected subsets signals that the dynamics produces correlations over very large distances but, at the
same time, the presence of localized subsets points to the existence of underlying localized objects that mediate this correlation. Such objects can still serve as a useful tool for
theoretical analysis of the situation.
(c) If S behaves as global geometric structure, then this indicates that the underlying
dynamics produces correlations that universally span the whole system. This could happen
as a result of two qualitatively different physical situations. The first is the case of extreme
delocalization, wherein the underlying localized objects are essentially never seen individually, but can still be distinguished within the global subsets and are relevant for physics.
The second possibility is that the global subsets do not contain any privileged parts and behave as a single whole. Here the notion of a localized object is no longer useful to analyze
the physics, and the structure is inherently global (super-long-distance).
To avoid the confusion in nomenclature in what follows, let us summarize. Our goal
is to be able to distinguish three physically distinct situations: localized, delocalized and
inherently global (super-long-distance) behavior. At the same time, we introduced the concepts that distinguish three geometric spacetime arrangements: geometrically localized,
partially global and global structure. If there is a well-defined support S dominating the
physical behavior of , we associate the localized geometric nature of corresponding S
with the underlying localized physical behavior. Similarly, the partially global nature of S
is associated with delocalization. However, if S is a global geometric structure, we could
be dealing with two qualitatively different physical cases, namely extreme delocalization
or super-long-distance behavior. To distinguish the two, one could introduce additional
geometric concepts. However, we will take a different approach in what follows.
3. The use of fraction supports

While the discussion of the previous section provides a partial basis for analyzing the issues of localization, it is not quite satisfactory. Indeed, the main problem is that frequently
it is not possible to use it because there is no unique way to determine the support for
given local quantity (x) from physics considerations or otherwise. Even though there are
many ways to assign a support to a function in a generic way (i.e., regardless of the functional form), the arbitrariness of such procedures makes it dangerous for making definite
conclusions. We discuss the related issues in Appendix A.
In order to overcome this shortcoming, we propose to work with many potential
supports and monitor both the degree of global behavior, and the degree to which (x)
is concentrated in such possible supports. In this way, much more complete information
enters the consideration. We will argue and demonstrate (both here and in upcoming publications) that this strategy leads to definite conclusions in the specific case of QCD vacuum
structure. The procedure can be described in few steps and, to exemplify the discussion, we
will frequently refer to the case of Schrdinger wave function, i.e., (x) (x), where x
471
is a space coordinate. In what follows, we will assume (both in definitions and formulas)
that is a hypercubic discretization of the torus.
(1) Choose a positive semi-definite measure (x) that characterizes the physically
relevant strength (intensity) of (x). For Schrdinger wave function this is obviously
(x) = (x)(x) because this represents the probability density for the particle to be
in the vicinity of x. The role of (x) is to introduce an importance ranking for all space
time points: high-intensity points are physically more relevant than low-intensity points.
This allows us to define the family of nested subsets S (f ) (fraction supports) respecting
this ranking.
Definition (Fraction support). Let (x) be the local observable on spacetime with
intensity (x). The collection S (f ) of highest intensity-ranked spacetime points
that occupy fraction f of available volume, i.e., N (S (f ))/N () = f , will be referred to
as a fraction support corresponding to f and (x).
In the above definition the volume N ( ) for arbitrary represents the number of
lattice points contained in . The sets S (f ) form a family of potential supports whose
global behavior will be examined.
(2) While the volume fraction f is a convenient label for fraction supports, we also
need a measure expressing how good a support S (f ) would be, i.e., to which degree the
behavior of (x) within S (f ) dominates relative to the whole . In case of Schrdinger
particle the appropriate measure is the total probability of the particle to be within the
support, i.e., the sum of (x) = (x)(x) over S (f ). In general, this information is
contained in the normalized cumulative function of intensity7

C (f )
(x)
(x).
(3)
S (f )
At the regularized level considered here (before taking the continuum limit) f is a discrete
variable chosen from the set {n/N (), n = 0, 1, 2, . . . , N ()} [0, 1]. However, we will
keep the continuous notation and associate C (f ) with a continuous piece-wise linear
function connecting the points of the discrete map. In the continuum limit (and as a regularized approximation), C (f ) is a non-decreasing concave function from interval [0, 1]
to itself, such that C (1) = 1. If (x) is non-zero only in a single point or on the set of
measure zero, then C (f ) = 1 identically. In the other extreme situation when (x) is
constant through spacetime, we have C (f ) = f . In case of arbitrary strict support, i.e.,
when there exists such that (x) = 0 for x and (x) = 0 for x
/ , the
function C (f ) reaches value 1 at the corresponding fraction f and remains constant on
the interval [f , 1].
7 To arbitrary and (x) we can assign a coefficient (x)/ (x) expressing its quality as a
potential support. C (f ) represents a maximum of this coefficient over all subsets occupying the volume fraction
f of the underlying space.
472
We emphasize that C (f ) is a basic characteristic of the spacetime structure induced by

(x). It has a specific behavior for particular classes of spacetime distributions and contains detailed quantitative information about how much is (x) concentrated in fraction
supports S (f ). We propose that this information is necessary in order to judge the potential global character of the physically relevant structure.
(3) Every fraction support S (f ) generates a corresponding geometric structure S (f )
via decomposition into maximal connected subsets. To monitor the global behavior of these
structures, we compute the corresponding characteristics introduced in Section 2.1, and
thus define the functions

max (f ) max S (f ) ,
(4)
pta (f ) pta S (f ) .
When measuring the length in the units of maximal linear distance () (which we will
assume from now on), then max (f ) and pta (f ) are functions from interval [0, 1] to itself
and max (1) = pta (1) = 1 similarly to C (f ) (see however Section 3.1). Contrary to C (f )
though, these functions are not necessarily concave. Also, max (f ) is a non-decreasing
function, while the monotonic properties of pta (f ) are indefinite in general.
We have thus defined three characteristic functions associated with local physical observable (x) defined on spacetime . While C (f ) describes and quantifies the tendency of
(x) to be concentrated in preferred regions of spacetime, max (f ) and pta (f ) describe

and quantify the degree of global behavior for such preferred regions. The knowledge of
these functions allows us to judge the degree of global behavior for arbitrary degree of
saturation in physically relevant intensity (x). We propose that this detailed information has to be considered in conjunction in order to make definite conclusions on whether
the physical structure inducing the behavior of (x) is localized, delocalized, inherently
global or indefinite. Following the discussion in Section 2, we distinguish three typical and
qualitatively different types of behavior shown schematically in Fig. 2.
(i) C (f ) saturates considerably faster than max (f ) as exemplified in Fig. 2(a).8
This means that at fractions where C (f ) starts to saturate (smallest fractions such that
C (f ) 1) the corresponding fraction supports form a localized geometric structure, i.e.,
Fig. 2. The signatures of localized (a), delocalized (b), and inherently global (c) behavior.
8 Note that it is not essential that we have drawn the concave graphs for (f ) and (f ) in Fig. 2(a).
max
pta
These functions could in principle be convex or mixed as well.
473
max (f ) C (f ). In this case we classify the underlying physical structure as localized

since the dominating fields are concentrated on such geometric structure.
(ii) C (f ) saturates at similar fractions as either max (f ) or pta (f ), or in between the

two, as exemplified in Fig. 2(b). This situation indicates that the underlying physical struc
ture is delocalized. Indeed, if C (f ) saturates faster than pta (f ) then the fraction support
at the point of saturation forms a partially global geometric set, i.e., max (f ) C (f ) >
pta (f ), which is typical of delocalization (see Section 2.2). On the other hand, if C (f )
and pta (f ) saturate at approximately the same rate, then we are dealing with geometrically
global structure at the point of saturation, which is common both to extremely delocalized
and inherently global cases. However, the signature of extremely delocalized structure is
that a small decrease in fraction changes global geometric behavior to partially global,
where some individual localized objects starts to appear in isolation. This is precisely what
happens in this case.
(iii) C (f ) saturates considerably slower than pta (f ), as shown in Fig. 2(c). This
means that at fractions where C (f ) starts to saturate the corresponding fraction supports
form a geometrically global structure, i.e., pta (f ) C (f ). Moreover, this arrangement

is very stable with respect to the change of the fraction. This indicates that the underlying
physical structure is inherently global (super-long-distance).
We emphasize that a generic spacetime arrangement does not necessarily have to fall
into any of the typical cases described above. If that happens, than we have an indefinite
situation which has to be analyzed in more detail by other means.
(4) While the fraction f of spacetime is a convenient variable to introduce the func
tions C (f ), max (f ) and pta (f ), there is a more economic way to convey the physics
information on global behavior. Indeed, the above characteristics provide answer to the
following relevant question. Given arbitrary degree of saturation C (0 C 1), what
are the values of max , pta characterizing the global behavior of the corresponding fraction
support where fields are concentrated? In case of a Schrdinger wave function we could
ask what are the global characteristics of the minimal region of space comprising, e.g.,
90% chance to find the particle. We can express this information directly by eliminating f
in favor of C . In generic case this can be done since f C (f ) is a one-to-one map of
the interval [0, 1] to itself. Formally, we are introducing the change of variable

max (f ) max f C ,
pta (f ) pta f C .
(5)
While the functional forms are obviously different after the change of variable, we will
keep the same notation, namely max (C) and pta (C), where it is implicitly assumed that C
is related to local field (x). We also note that for many relevant questions C is indeed a
more physical variable in the sense that C (rather than f ) should be fixed when approaching
a continuum limit.
It is straightforward to see how the typical localized, delocalized and global physical be
havior reflects in the functions max (C) and pta (C). In Fig. 3 we plot these functions
corresponding to cases shown in Fig. 2.
474
Fig. 3. The signatures of localized (a), delocalized (b), and inherently global (c) behavior.
3.1. Partitioning of the fraction support

In our discussion so far, the spacetime structure was introduced solely via the behavior of (f ), and the resulting geometric properties of fraction supports. However, there
could be additional physics aspects of local observable (x), which should be taken into
account when judging the global properties. A typical example is represented by topological charge density q(x), whose sign has important physical consequences. Indeed, the
quarks of preferential chirality are attracted to regions of definite sign. This is expected to
be very relevant for understanding various aspects of hadron structure and hadron propagation. In such a case it is then appropriate to partition the fraction support accordingly, and
to consider global properties of the physically relevant partitioned sets
In general
[1,2,5].
case, we will thus consider the sets Sr (f ), r = 1, . . . , R, such that r Sr (f ) = S (f ),
and Sr (f ) Ss (f ) = for all r, s. There are several things that should be pointed out
with regard to this situation.
(1) The cumulative function C (f ) is not affected by partitioning and is defined in the
usual way relative to S (f ).
,r
,r
(2) The global characteristics max (f ), pta (f ) will characterize the partitioned sets
and the only generic property that will change in this regard is that they do not necessarily
,r
,r
approach unity as f 1. In principle then the values max (1) and pta (1) are relevant characteristics that enter the classification of global properties. It is straightforward to extend
our discussion to include them.9
(3) If there is a symmetry in the theory which relates the average properties of partitioned fraction supports (which is usually the case), then the global characteristics for
different r just represent samples (possibly correlated) for the unique functions max (f )
and pta (f ).
9 We note that in cases relevant to QCD vacuum structure that we have studied so far, we have always observed
that 1 = rmax (1) rpta (1), and so it is not necessary to do this explicitly here.
475
3.2. Loose ends

Our discussion in the main part of this section concentrated on clear exposition of basic
ideas in our approach, but left open certain issues that arise in special cases. In what follows
we will consider some of them.
The first is the definition of the fraction support (at the regularized level) if some
functional values of (x) are degenerate.10 The extreme case is that of a constant
field (x). The problem is how to assign the ranking within the degenerate subset, so
that the associated fraction supports lead to unique definition of C (f ), pta (f ) and
max (f ). Note that this is not a problem for definition of C (f ) and arbitrary ad hoc
rule leads to the same function. However, the functions pta (f ) and max (f ) can in principle depend on the rule chosen.11 It is straightforward to complete the definition of
these functions in arbitrary degenerate case, so that the notion of localization, delocalization and global behavior (as defined here) will correspond to its intended meaning. To
do that, note that the process of ranking the spacetime points corresponds to specifying a non-repeating sequence x1 , x2 , . . . , xN () . Denoting the sets of points contained in
partial sequences containing n points as P n = {x1 , x2 , . . . , xn }, assume that this partial sequence has been assembled. Then the rules for continuing the sequence recursively are
the following: (1) If the next value of intensity, i.e., max
max{ (x), x P n }
n
max
is k-fold degenis non-degenerate, then xn+1 = y such that (y) = n . (2) If max
n
= (y1 ) = = (yk ),
erate, i.e., if there are points y1 , y2 , . . . , yk such that max
n
then we need to specify the permutation p(i), i = 1, . . . , k such that xn+i = yp(i) . We
require this permutation to be chosen in such a way that max (f ) grows fastest for arguments fm m/N (), m {n + 1, n + 2, . . . , n + k}. More precisely, we require that (i)
max (fn+1 ) max (fn ) is maximal over all permutations. (ii) If this does not specify permutation p uniquely, we consider the restricted degenerate set of permutations and further
require that max (fn+2 ) max (fn+1 ) is maximal. (iii) This recursion continues and stops
when p is uniquely determined. (iv) If p is not uniquely determined at the end of the finite recursion, we choose it from the remaining degenerate set by arbitrary fixed rule. One
can easily check that the resulting max (f ) and pta (f ) are invariant under this remaining
freedom. Note also that the above rules lead to classification of the constant field (x) as
global configuration which is appropriate.
The second class of situations which needs to be mentioned is exemplified by the following. Assume the existence of a strict support which becomes the set of measure zero
in the continuum limit. For example, a strict support could form a finite low-dimensional
manifold. In that case C (f ) 1 in the continuum limit, and thus C (f ) will not carry any
information about the behavior within the support. In that case one has to define the cumulative function and the corresponding global characteristics for the support itself. In the
other extreme case the global characteristics max (f ) or pta (f ) could become identically
one in the continuum limit, indicating a global behavior on sets of lower dimension. These
10 This will not be encountered in practice at all, but it is still appropriate to complete the framework by
discussing this situation.
11 In fact, the continuum limit of these functions will not depend on the rule for ranking within degenerate
subsets if these subsets become of measure zero in the continuum limit.
476
situations are of particular interest for QCD vacuum case [2], and will be addressed in a
separate publication.
4. Cases of physical interest

In this section we will discuss in more detail few basic examples of local observables
(x) which are of interest for studies of QCD vacuum structure, and which will be analyzed
in the upcoming publications. In particular, we will mention observables related to gauge
fields U (x) and the Dirac eigenmodes (x). To ease the notation, from now on we will
skip labels of (x) on various characteristics discussed. The appropriate association will
be obvious from the context. In this section we will also give examples (in case of Dirac
eigenmodes) of how the methods developed here work in QCD.
4.1. Gauge fields
In pure glue lattice gauge theory all local observables are constructed from fundamental gauge fields U (x) eigA (x) . The intensity of gauge fields is physically associated
with the corresponding field-strength tensor U (x) F (x) describing chromo-electric
and chromo-magnetic fields. Two basic physically relevant composite fields characterizing the intensity of F are the scalar density s(x) tr F F (x) (action density) and the
pseudoscalar density q(x) tr F F (x) (topological density). Understanding the space
time structure of s(x) and q(x) in configurations dominating the QCD path integral is one
of the most basic ingredients for understanding the QCD vacuum structure.
To characterize the global properties of this structure according to the scheme developed in previous sections, we first have to specify the corresponding physically relevant
intensities. Since s(x) is positive semidefinite, it can serve as its own intensity (x). One
might be tempted to consider other positive-semidefinite functions of s(x), e.g., the powers (x) = s (x) for > 0. The cumulative functions C(f ) would obviously be different
for different values of . However, while the intensity corresponding to = 1 has a welldefined physical meaning (we study the spacetime distribution of the action), this is not so
for other values of . Similarly, for q(x) the appropriate intensity to use is (x) |q(x)|,
since it allows for relevant comparison to the behavior of s(x).
In the case of q(x) there is a more physically-motivated approach suggested by the
fact that global fluctuations of topological charge are directly related to the mass via
WittenVeneziano relation [12]. One would thus like to judge the global properties of the
relevant spacetime structure relative to the saturation of topological susceptibility. While
conceptually this conforms to the general strategy developed here, it requires some technical modifications which we now describe.
The main change one has to deal with is the definition of the cumulative function C(f )
because there is no intensity (x), local in the gauge fields, that one could use to define
this function directly using Eq. (3). Nevertheless, assuming that a well-defined topological
density operator (such as the operator associated with GinspargWilson fermions [13,14])
is at our disposal, it is still possible to construct the appropriate cumulative function which
477
will play the corresponding role in classifying the global behavior of the spacetime structure. Indeed, defining the fraction supports S(f ) relative to (x) = |q(x)|, let us consider
the corresponding cumulative charge Q(f ), and the associated fluctuation (f ) which is
only defined in the ensemble average
(f )
Q2 (f ) Q(f )2

,
V (f )
Q(f )
q(x),
(6)
xS (f )
where V (f ) = a 4 N (S(f )) is the corresponding physical volume. We then define the normalized cumulative function of global fluctuations via
C(f ) (f )/(1).
(7)
There are two points that need to be emphasized here. (1) While we call the C(f ) defined above a cumulative function in analogy with definition in Eq. (3), it is not a priori
obvious that it has the specific properties of the cumulative function. In particular, it is not
obvious that it is a concave non-decreasing function. Indeed, we are making an assumption
that the ordering induced by |q(x)| will ensure these properties in the ensemble average.
This is reasonable since it is expected that regions with most intense fields will contribute
most to global fluctuations. Our numerical data a posteriori confirm that this is indeed the
case in QCD [15]. (2) The precise determination of C(f ) does not necessarily require the
use of very large ensembles to compute the physical susceptibility (1) to comparable
accuracy. Indeed, computing C(f ) directly via jackknife samples of (f )/(1) leads to
much smaller fluctuations over the whole range of f [15].
4.2. QCD Dirac zero modes
For Dirac eigenmodes (x) (i.e., D (x) = (x)) the basic scalar and pseudoscalar
composites are the density d (x) (x) (x) and chirality c (x) (x)5 (x).
However, the zero modes are globally chiral and this is exactly true even on the lattice if
overlap fermions are used (which we assume). We thus only have one independent bilinear
to study, which we choose to be the density (x) d(x). In direct analogy to the case of
Schrdinger particle, the associated intensity (x) for determining the fraction supports
and for computing the cumulative functions will be the density d(x) itself. Note that the
choice of the bilinear form (rather than higher powers) is further physically motivated by
the fact that the fermionic action is bilinear in fermion fields.
To illustrate how the global characteristics of the spacetime structure introduced here
behave in QCD, we have analyzed the low-lying eigenmodes of the overlap operator. In
particular, we have considered an ensemble of 20 Wilson gauge configurations on 164 lattice at = 6.07. This corresponds to the lattice spacing of a = 0.082 fm as determined
from the string tension [16], and the linear extent L 1.3 fm. The eigenmodes of the overlap operator with negative mass = 1.368 ( = 0.19) were computed using the Zolotarev
approximation to implement the operator numerically. The details of this numerical implementation are given elsewhere (see, e.g., [1,17]).
We found the total of 26 zero modes in the above ensemble and computed the average C(f ), max (f ) and pta (f ). The results are shown in Fig. 4 (top left) together with the
478
Fig. 4. Characteristics of global behavior for Dirac zero modes (top), Dirac near-zero modes via density (middle)
and Dirac near-zero modes via chirality (bottom). The structure is clearly super-long-distance in all cases. See
the discussion in the text.
associated max (C) and pta (C) (top right). One can clearly observe that the structure is geometrically global at the fractions of just few percent (e.g., pta (f = 0.036) = 0.9). At such
fractions the cumulative function only achieves values C 0.2. Following our discussion
479
in Section 3 we thus conclude that we clearly observe a super-long-distance spacetime

structure in QCD Dirac zero modes. Note also that max (f ) and pta (f ) are quite close to
one another, indicating that there is very little fragmentation at arbitrary fraction. In fact,
the structure typically comes in a single dominating piece as it is grown from very small
fractions up.
4.3. QCD Dirac near-zero modes
In the case of near-zero modes we can study the global behavior in both d(x) and
c(x). Due to 5 -hermiticity, the near-zero modes of the overlap operator come in complexconjugate pairs with identical d(x) and c(x). We will thus consider only a single representative of the pair. For our calculation, we have considered two lowest-lying pairs for every
configuration from the ensemble. This gave us altogether 40 modes to analyze.
We start with the density for which the appropriate intensity is (x) d(x) as in the
case of zero modes. The resulting global characteristics are shown in Fig. 4 (middle). From
this data it is very clear again that the scalar structure in QCD near-zero modes is superlong-distance. For example, pta (f = 0.038) = 0.9, while C 0.16 at the same fraction.
Similarly to zero modes, the functions max (f ) and pta (f ) are quite close one to another
and the structure is dominated by a single connected piece. As a side remark, note also that
the cumulative function C(f ) has steeper behavior in case of zero modes than in case of
near-zero modes. This results in steeper behavior of functions max (C), pta (C) for nearzero modes. We will study such details in a dedicated publication [11].
In case of chirality the appropriate intensity to study is (x) = |c(x)|. Since the sign of
local chirality carries a relevant physical information (e.g., on the preferred duality of the
underlying gauge field and the qualitative features of the quark propagation), we partition
the fraction supports into subsets of definite sign (see Section 3.1). Since QCD dynamics
has no preference for a particular chirality, we average over connected subsets of both signs
in calculation of pta (f ). Also, in calculation of max (f ) we consider the maximum over
connected subsets of both signs. The results are shown in Fig. 4. Again, we find the space
time characteristics corresponding to clear super-long-distance structure in c(x). Note that
the functions pta (f ), max (f ) saturate slightly slower than in the case of d(x) which is due
to the partition and the fact that the connected subsets of opposite chirality obstruct one
another in their global behavior. At the same time the cumulative function is steeper than
in the case of d(x) (it is in fact similar to cumulative function for zero modes) resulting in
the slower saturation of max (C) and pta (C).
5. Conclusions
In this paper we have proposed the framework for studying the issues of localization
in QCD vacuum in a model-independent manner.12 This framework was set up in such
a way as to inquire into three hypothetical scenarios underlying the QCD dynamics. The
12 The methods introduced here are by no means special to QCD or to four dimensions.
480
first scenario (localization) involves the existence of fundamental localized objects in the
gauge field populating the QCD vacuum and driving the dynamics. The second scenario
(delocalization) involves localized entities which however clump into global structures (via
interaction) and can support physical correlations over very large distances. A distinctive
property of delocalization is that the localized objects still preserve their identity and can
serve as a basis for physical analysis of vacuum properties. The third scenario (inherently
global or super-long-distance behavior) is qualitatively different in that there are fundamental entities that themselves exhibit global behavior. The geometric structure of these
entities can facilitate correlations over arbitrarily large distances and the vacuum properties can be fully understood only in terms of these global objects.
The approach proposed in this work is based on the idea that the existence of gauge field
objects with above properties will lead to a corresponding geometric behavior in local composite fields, and possibly also in Dirac eigenmodes. In particular, the spacetime regions
containing dominating fields (the support of the field) will behave in the specific geometric manner, forming localized, partially localized or global geometric structure. However,
since the definition of field support on physical grounds is not generically available, we
proposed the use of cumulative functions with simultaneous monitoring of both the degree
of saturation of the field, and of the global characteristics of the underlying fraction supports. It was then proposed that the specific functional behavior of these dependencies can
be associated with the three physical situations described above.
We have discussed in detail the application of these ideas to special cases of action
density, topological density, and to the scalar and pseudoscalar density of Dirac low-lying
modes. Our approach was illustrated via the study of overlap zero modes and near-zero
modes. The results clearly demonstrate the super-long distance nature of the underlying
gauge structure as reflected in the Dirac eigenmodes. They also provide the quantitative
validation for the original suggestion by the Kentucky group that the Dirac near-zero modes
form global structure (ridges rather than separate peaks) [5].13
Finally, as emphasized in the introduction, the aim of this paper (first in the series) is
to introduce techniques suitable for analysis of QCD spacetime structure obtained in a
model-independent manner. In this approach, gauge fields are not processed or altered in
any way, and the structure is uncovered as it appears directly in the configurations dominating the QCD path integral. The culmination of these attempts so far is the discovery of the
strictly low-dimensional, inherently global sheet/skeleton structure in the overlap topological charge density [2].14 The methods developed in this work (see Section 4.1) provide a
formal framework for a detailed quantitative confirmation of the super-long-distance nature
of this structure [15].
13 We emphasize that in the case of Dirac modes we talk about the eigenmodes of the proper chirally symmetric
lattice Dirac operator such as the overlap operator. Exponentially localized modes could be observed, e.g., in
hermitian WilsonDirac operator for unphysical values of the hopping parameter [18].
14 It is interesting that the notion of strictly low-dimensional structure is also emerging in an indirect way,
using the monopole and center-vortex detection (projection) techniques [19]. If there is a relation between these
findings it is yet to be found and understood.
481
Acknowledgements
Numerous discussions with members of the Kentucky group on the topics of this work
are gratefully acknowledged. Thanks to Andrei Alexandru and Hank Thacker for feedback
on the manuscript. The author is indebted to Andrei Alexandru and Nilmani Mathur for
help with the figures, and to Jianbo Zhang for the help with the data, part of which was used
to illustrate the methods proposed in this manuscript. The Kentucky group is acknowledged
for making this data available. The author benefited from communications with Ganpathy
Murthy, tefan Olejnk and Joe Straley.
Appendix A. Generic definitions of the support

Assume that we are given a generic function (x) on spacetime, and the associated
real-valued positive-semidefinite intensity (x). If the specific properties of (x) are not
known, it is tempting to use some generic way of determining the spacetime region where
the function is effectively concentrated, i.e., its effective support. This support is implicitly
associated with spacetime regions occupied by strongest fields, such that fields in the
rest of the spacetime can either be neglected or behave in a qualitatively different (low
intensity) manner. Thus, from the geometric point of view, effective support is one of the
fraction supports defined in Section 3, and the goal of a given generic procedure is to fix
the corresponding fraction (effective fraction feff ). However, as emphasized in Section 3,
the generic definitions of this type are highly ambiguous.
To illustrate this, let us concentrate on the scalar structure of Dirac eigenmodes. The
same discussion can be extended straightforwardly to the case of arbitrary local observable (x). Given a (normalized) Dirac eigenmode (x), the physically relevant intensity
characterizing its (scalar) strength is (x) d(x) = (x), and the definition of the support will thus be given in terms of this associated function feff = feff [d(x)]. Consider, for
example, the following definitions of feff .
(1) A natural possibility is to base the definition on the behavior of the corresponding
cumulative function C(f ). In Fig. 5 (left) we show C(f ) for the zero mode #1 from the set
analyzed in Section 4.2. Even though we work on the latticized torus with N sites, we will
keep the continuous notation and treat C(f ) as a piece-wise linear continuous function
connecting the points of the discrete map (see Section 3). C(f ) is a concave function on
1
the interval [0, 1] such that C(0) = 0 and C(1) = 1, and hence 1 I 0 df C(f )
1/2. Clearly, if d(x) is sharply concentrated on small fraction of spacetime, then I 1,
and when d(x) is spread out over the whole lattice then I 1/2. In an extreme case of
localization on single site we have I = 1 1/2N (I = 1 in the continuum limit), and
I = 1/2 for constant d(x). We can thus in principle base the definition of feff on the value
of I . The simplest possibility is to use the linear expression in I which interpolates between
the above extreme values, i.e.,
feff 2(1 I ) f1 .
(A.1)
482
Fig. 5. Left: the cumulative function C(f ) for zero mode #1 and the associated effective fractions f1 (with C1 (f )),
f2 (with C2 (f )), and fIPR are shown. Right: three generic definitions of effective fraction discussed in the text
for all 26 zero modes.
This definition gives f1 = 1/N and f1 = 1 for d(x) = x,y and d(x) = 1/N , respectively.
Also, 0 f1 1 for arbitrary d(x), and thus f1 has the intended meaning of the effective
fraction. It is tailored for the functions d(x) that are constant on fraction f1 of sites and
vanish on the rest of the sites. In effect then, the above procedure associates with arbitrary
d(x) a unique function of this type, such that the corresponding integrals I are the same.
In Fig. 5 (left) we show the value f1 together with the cumulative function C1 for the
associated constant function on the strict support.
(2) Other than its simplicity, there is no particular reason to choose f1 to be the definition of effective fraction in a generic situation. For example, instead of relating d(x) to
the constant function on strict support, we could relate it to the function which is constant
and non-zero both on the support and (with lower intensity) on the complement of the support. To uniquely specify such function (and with it the corresponding fraction feff f2 ),
we can, e.g., require simultaneously that (a) the integrals I of the cumulative functions
are the same and that (b) the corresponding C2 (f ) is the best approximation of C(f ), i.e.,
1
that 0 df (C(f ) C2 (f ))2 is minimal. With this choice the resulting f2 and C2 (f ) for
zero mode #1 are shown in Fig 5 (left). Note that this definition could be useful if it is a
priori known that the actual spacetime distributions in question have a well-defined, approximately constant core, but also an approximately constant background. Again, in the
extreme cases of d(x) = x,y and d(x) = 1/N , we obtain f2 = 1/N and f2 = 1 respectively, while 0 f2 1 for arbitrary d(x).
(3) Effective fraction can also be based on the inverse participation ratio (IPR) [20].
This technique was developed in condensed matter physics where both its virtues and
limitations are well known. Its original intended use was to monitor the departures from
exponential localization.15 In this case the effective fraction can be assigned generically
15 In lattice QCD literature the IPRs started to be used in the study of Dirac eigenmodes only relatively re-
cently [21]. (See also a new conference proceedings preprint [22], where they are employed in an attempt to
estimate the dimensionality of structures in the eigenmodes of the Asqtad operator.)
483
via

1
1
N
d 2 (x).
feff fIPR
x
(A.2)
Needless to say, in this case we have also that fIPR = 1/N and fIPR = 1 for d(x) = x,y
and d(x) = 1/N , respectively. Also, 0 fIPR 1 in generic case.
As can be seen clearly on the example of the overlap Dirac zero mode in Fig. 5 (left),
the above generic definitions, while quite sensible for their designed purpose, give wildly
different estimates for what the effective fraction is. It is thus potentially dangerous to use
generic methods to determine the effective fraction in situations where sufficiently detailed
characteristics of typical spacetime distributions are not known, while the quantitative
aspects of the problem are important. In particular, it would not be possible to seriously
address the issues of global behavior in QCD vacuum at this stage by using a particular
definition from the infinite set of generic possibilities available. At the same time, using a
more complete information stored in the cumulative function makes it possible to resolve
this issue.
Finally, we should emphasize that generic definitions, such as the above, are rather robust for comparative purposes, i.e., to decide if the support of mode 1 is larger (more
potent) than the support of mode 2 . Indeed, in Fig. 5 (right) we show the effective fractions for all 26 modes using the above three definitions, and the comparisons are largely
consistent.
References
[1] I. Horvth, S.J. Dong, T. Draper, F.X. Lee, K.F. Liu, H.B. Thacker, J.B. Zhang, Phys. Rev. D 67 (2003)
011501(R);
I. Horvth, et al., Nucl. Phys. B (Proc. Suppl.) 119 (2003) 688.
[2] I. Horvth, S.J. Dong, T. Draper, F.X. Lee, K.F. Liu, N. Mathur, H.B. Thacker, J.B. Zhang, Phys. Rev. D 68
(2003) 114505;
I. Horvth, et al., Nucl. Phys. B (Proc. Suppl.) 129130 (2004) 677;
I. Horvth, et al., in: N. Brambilla, G. Prosperi (Eds.), Proceedings of the Quark Confinement and the
Hadron Spectrum V, Gargnano, Italy, 1014 September 2002, World Scientific, Singapore, 2003, p. 312,
hep-lat/0212013.
[3] I. Horvth, N. Isgur, J. McCune, H.B. Thacker, Phys. Rev. D 65 (2002) 014502.
[4] T. DeGrand, A. Hasenfratz, Phys. Rev. D 64 (2001) 034512.
[5] I. Horvth, S.J. Dong, T. Draper, N. Isgur, F.X. Lee, K.F. Liu, J. McCune, H.B. Thacker, J.B. Zhang, Phys.
Rev. D 66 (2002) 034501.
[6] T. DeGrand, A. Hasenfratz, Phys. Rev. D 65 (2002) 014503;
I. Hip, et al., Phys. Rev. D 65 (2002) 014506;
R. Edwards, U. Heller, Phys. Rev. D 65 (2002) 014505;
T. Blum, et al., Phys. Rev. D 65 (2002) 014504;
C. Gattringer, et al., Nucl. Phys. B 618 (2001) 205.
[7] N. Cundy, M. Teper, U. Wenger, Phys. Rev. D 66 (2002) 094505;
C. Gattringer, Phys. Rev. Lett. 88 (2002) 221601;
P. Hasenfratz, et al., Nucl. Phys. B 643 (2002) 280;
C. Gattringer, Phys. Rev. D 67 (2003) 034507;
C. Gattringer, R. Pullirsch, Phys. Rev. D 69 (2004) 094510;
484
T. Draper, et al., hep-lat/0408006.

[8] H. Neuberger, Phys. Lett. B 417 (1998) 141;
H. Neuberger, Phys. Lett. B 427 (1998) 353.
[9] H. Thacker, S. Ahmad, J. Lenaghan, hep-lat/0409079.
[10] P. Anderson, Phys. Rev. 109 (1958) 1492.
[11] I. Horvth, et al., in preparation.
[12] E. Witten, Nucl. Phys. B 156 (1979) 269;
G. Veneziano, Nucl. Phys. B 159 (1979) 213.
[13] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125.
[14] R. Narayanan, H. Neuberger, Nucl. Phys. B 443 (1995) 305.
[15] I. Horvth, Talk at the workshop The QCD Vacuum from a Lattice Perspective, Regensburg, 2931 July
2004;
I. Horvth, et al., in preparation.
[16] M. Guagnelli, R. Sommer, H. Wittig, Nucl. Phys. B 535 (1998) 389.
[17] Y. Chen, et al., Phys. Rev. D 70 (2004) 034502.
[18] M. Golterman, Y. Shamir, Phys. Rev. D 68 (2003) 074501.
[19] F.V. Gubarev, A.V. Kovalenko, M.I. Polikarpov, S.N. Syritsyn, V.I. Zakharov, Phys. Lett. B 574 (2003) 136;
A.V. Kovalenko, M.I. Polikarpov, S.N. Syritsyn, V.I. Zakharov, hep-lat/0408014.
[20] D.J. Thouless, Phys. Rep. 13 (1974) 93.
[21] C. Gattringer, Nucl. Phys. B 617 (2001) 101.
[22] MILC Collaboration, C. Aubin, et al., hep-lat/0410024.
The four-loop QCD -function

and anomalous dimensions
M. Czakon
DESY, Platanenallee 6, D-15738 Zeuthen, Germany
Institute of Physics, University of Silesia, Uniwersytecka 4, PL-40-007 Katowice, Poland
Abstract
The four-loop -function of quantum chromodynamics is calculated and agreement is found with
the previous result. The anomalous dimensions of the quarkgluon vertex, and quark, gluon and
ghost fields are given for a general compact simple Lie group.
PACS: 11.10.Gh; 12.38.Bx
1. Introduction
The renormalization group properties of quantum chromodynamics were the reason of
acceptance of this theory as the theory of strong interactions. The central rle played by the
QCD -function, calculated at the one- [1], two- [2], three- [3] and finally at the four-loop
[4] level, cannot be overestimated in this respect.
The calculation of the last known, four-loop, term in the expansion of the -function,
was performed by only one group [4]. The authors evaluated . . . of the order of 50 000
4-loop diagrams. These two facts lead to the conclusion, stressed many times (in, e.g.,
[5]) that it is not only justified, but also necessary to independently evaluate this quantity.
The present work is intended to fulfill this need.
E-mail address: mczakon@physik.uni-wuerzburg.de (M. Czakon).
doi:10.1016/j.nuclphysb.2005.01.012
486
M. Czakon / Nuclear Physics B 710 (2005) 485498
Apart from the -function itself, we present two new results. The first one is the complete set of MS anomalous dimensions at the four-loop level in the linear gauge and with
color structures of a general compact simple Lie group. These give, in a compact form,
the complete set of four-loop QCD renormalization constants. We notice that the case of
the SU(N ) group has been solved in [5] with the assumption, however, that the four-loop
-function is correct. Our second new result is of a more technical nature, but should simplify future renormalization group calculations at this level of the perturbative expansion.
To this end, we derived the necessary set of divergent parts of the four-loop fully massive
tadpole master integrals. A purely numeric result was available from [6], but could not be
used for our purpose.
This paper is organized as follows. The next section presents the methods used in the
calculation, as well as some efficiency considerations. Subsequently, our results for the
anomalous dimensions are listed. Conclusions and remarks close the main part of the paper.
The expansions of the tadpole master integrals are contained in Appendix A.
2. Calculation
The calculation of renormalization group parameters, i.e., anomalous dimensions and
beta functions allows for vast simplifications in comparison to the actual evaluation of
Green functions with kinematic invariants in the physical region. Dimensional regularization and the MS scheme are particularly well suited for this kind of problems, since they
make it possible to manipulate dimensionful parameters of the theory. In fact, ever since
the introduction of the infrared rearrangement [7], it is known how to set most of them to
zero and avoid spurious infrared poles. With the advent of the R operation [8], infrared divergences are even allowed in individual diagrams and only compensated by counterterms
afterward.
At the four-loop level two techniques seem to be most promising. One is a global R
operation [5], where one would set all of the external momenta to zero and also almost
all of the masses, keeping just one massive line. The spurious infrared divergences are
then compensated by adding a global counterterm. The advantage of this approach is that
the whole problem can be reduced to the calculation of three-loop massless propagators,
for which there exists a well tested and efficient FORM [9] package, MINCER [10]. The
disadvantage is that the construction of the global counterterm is not trivial. In fact, up to
now it has not been possible for the gluon propagator.
A second technique consists in setting all of the external momenta to zero, but keeping a
common non-zero mass for all the internal lines [4,11,12]. The advantage of this approach
is that one never encounters any infrared divergences. One minor part of the price for this
convenience is the necessity of a gluon mass counterterm. The more problematic part is,
of course, the calculation of the divergent parts of four-loop tadpole diagrams that occur.
One way to do this is to generalize the algorithms of [12] to the four-loop level. We, however, decided to use integration-by-parts identities to reduce all of the integrals to a set of
master integrals depicted in Fig. 1. The divergent parts of the latter were then calculated as
described in Appendix A.
487
Fig. 1. Completely massive four-loop tadpole master integrals. Dashed lines mark those integrals which are
needed to one order higher in the expansion, i.e., up to finite parts for this calculation.
Instead of developing a dedicated software for the reduction of tadpole integrals1 we

used our own implementation of the Laporta algorithm [14] in the form of the C++ library
DiaGen/IdSolver [15]. We found it also a good opportunity to study the efficiency of this
approach on a large scale problem.
Since the calculation was performed in the linear gauge, the gluon propagator had the
form

k k
i
iD (k) = 2 g + 2 .
(1)
k
k
It is clear that this implies that every power of the gauge parameter will lead to more
powers of the denominators and irreducible numerators in the integrals. A minimal way to
test gauge invariance at the end is to keep at most a single power of , which corresponds
to a first order expansion of the result around the Feynman gauge. Under this restriction,
the calculation involved a little less than 210 000 independent integrals. The distribution of
1 Such a software has been developed in [13].
488
Table 1
Distribution between the tadpole prototypes of the 203 580 different integrals occurring in the calculation of the
gluon propagator in linear gauge limited to at most one power of . Denominator powers denote in fact the total
number of dots on the lines
Prototype
Number of lines
Denominator powers
Numerator powers
PR1
6764
PR2
PR3
PR4
8575
19402
2659
6
7
5
3
4
2
8944
26614
15058
21528
5
7
6
7
3
5
4
4
15906
28244
6988
16157
5
7
6
7
4
5
4
5
PR5
PR6
PR7
PR8
PR9
PR10
PR13
PR14
Number of integrals
PR11
PR15
11654
10973
5
6
5
5
PR12
PR0
2720
1394
5
5
5
5
them among the four-loop tadpole prototypes in the hardest case of the gluon propagator
is given in Table 1. The remaining cases of the quarkgluon vertex, and quark and ghost
propagators involved about one-third of these integrals and one power of denominator and
irreducible numerator less for each of the prototypes.
It turned out that to reduce all of the integrals to masters, it was sufficient to generate integration-by-parts identities with up to six additional powers of the denominators and
four of the numerators for all of the integrals up to seven lines, and with five additional powers of the denominators and four of the numerators for the eight- and nine-liners. The total
number of solved integrals was then about 2 000 000, meaning a 10% efficiency. About
37% percent of the integrals turned out to be finite. These could have been eliminated from
the very beginning by a careful study of divergences. We convinced ourselves, however,
that this would not allow for lower powers of denominators and numerators in the reduction
process, unless some very involved procedure were used.
3. Results
Since the -function is, up to normalization, the anomalous dimension of the coupling
constant, it is necessary to perform the complete renormalization of some vertex. To this
end we chose the quarkgluon interaction, mostly because the calculation of [4] involved
the ghost and gluon instead.
The renormalization constant of the quarkgluon vertex will subsequently be denoted
by Z1 , whereas the renormalization constants of the quark, gluon and ghost fields by Z2 ,
489
Z3 and Z3c , respectively. Even though Z3c is not necessary for the present calculation, we
derived it in order to have the complete set of renormalization constants at the four-loop
level.
We will not give the renormalization constants explicitly, but instead we will limit ourselves to the anomalous dimensions, which are defined by
= 2
d log Z
,
d2
(2)
where is the t Hooft unit of mass introduced to keep the renormalized coupling constant
dimensionless. Since we use dimensional regularization and the MS scheme, the renormalization constants can be expanded as
Z=1+
(i)

z (as , )
i=1
i
=1+

i

i=1 j =1
asi
z(i,j ) ( )
,
j
(3)
where as is connected to the QCD coupling constant g by as = s /(4) = g 2 /(16 2 ).

Using the fact that the dependence of Z on enters only through as and , and that the
renormalization constants of the gluon field and of the gauge parameter are equal, one
obtains
log Z
log Z
= ,
3 (1 )
( + )as
(4)
as
where the -function is simply equal to the anomalous dimension of s , i.e., = s . This
implies that
= as

z(1)
=
asi+1 (i) ,
as
(5)
i=0
where now
(i) = (i + 1)z(i+1,1) .
(6)
Eq. (4) can be used in turn to reconstruct the original renormalization constant from a given
anomalous dimension.
Since the quarkgluon vertex, as well as the quark and gluon anomalous dimensions
have already been given up to the three-loop level in the linear gauge in [3], we will not
reproduce them here2 but only give our result for the four-loop anomalous dimensions with
color structures of a general compact simple Lie group. At this point we stress once more,
that the results have been obtained in a first order expansion around the Feynman gauge

8
(3)
2 2 7870
3 + 244
1 = CA CF T F n f
243
3

797
2
+ 1183 + 364 + 405
+ CA C F T F n f
18
2 During the course of our calculation we found full agreement with [3]. Contrary to the four-loop case, our
three-loop calculation was performed without expansion in the gauge parameter.
490

5131
166 32
+ 8483 14405 + CA TF3 n3f
+ 3
12
81
9

104542
187
+
3 884 205
+ CA2 CF TF nf
243
3

304
23777
323 + CA2 CF2
2143 664 + 7905
+ CF2 TF2 n2f
9
36

6307
94
+ 3 184
+ CA2 TF2 n2f
972
3

10059589
280
1489
173
1865
3 +
4
5 +
CF TF3 n3f
+ CA3 CF
15552
24
4
12
81

473903 3311
387
3 +
4 + 555
+ CA3 TF nf
7776
24
8

76
350227
2959
111
5125
3
4
643 + CA
+
3
4
5
+ CF T F n f
3
3888
72
32
96

d abcd dAabcd
1027
4003 + 6405 + nf F
(483 + 605 )
+ CF4
8
NA

dAabcd dAabcd
d abcd dFabcd
21 367
335
+
+
3
5 + 128nf F
NA
8
4
4
NF

dFabcd dAabcd
1076 16
+
(66 + 1903 1705 ) + CA CF TF2 n2f
3
NF
243
3

767
7423
76
9
443 124 + CA2 CF TF nf
+ 3 + 4
+ CA CF2 TF nf
12
243
3
2

7
1229
4
3
+ CA2 CF2 3 + 3 55 + CA2 TF2 n2f
2
972
3

2127929
1013
87
665
3 + 4 +
5
+ CA3 CF
31104
24
16
24

35345 37
13
+ 3 + 4
+ CA3 TF nf
7776
3
8

dAabcd dAabcd 9
1539403 389
139
73
55
4
3 + 4 + 5 +
3
+ CA
62208
32
32
8
NA
16
8

d abcd dAabcd
(1 483 + 705 ) ,
+ F
(7)
NF
+ CA CF3
(3)

6835 112
+
3
243
3

2407
+ 443 + 364 + 1605
+ CA CF2 TF nf
36

5131
+ 8483 14405
+ CA CF3
12
2 = CA CF TF2 n2f
491

23777
2143 664 + 7905
+ CA2 CF2
36

1365691 119
280
+ CA2 CF TF nf
3 254 805 +
CF TF3 n3f
3888
3
81

10059589 1489
173
1865
+ CA3 CF
3 +
4
5
15552
24
4
12

304
76
323 + CF3 TF nf
643
+ CF2 TF2 n2f
9
3

d abcd dFabcd
1027
4003 + 6405 + 128nf F
+ CF4
8
NF
dFabcd dAabcd
(66 + 1903 1705 )
N
F

767
16
2 2 1076
2
3 + C A C F T F n f
443 124
+ CA C F T F n f
243
3
12

48865 118
7
15
2
2 2
+ CA C F T F n f
+
3 + 4 + CA CF 3 + 3 55
3888
3
2
2

2127929 1013
87
665
3 + 4 +
5
+ CA3 CF
31104
24
16
24

d abcd dAabcd
(1 483 + 705 ) ,
+ F
(8)
NF

15082 1168
(3)
3 = CA CF TF2 n2f
3 + 484
243
9

10847 980
1420 64
+
3 2405 + CA TF3 n3f
+ 3
+ CA CF2 TF nf
54
9
243
9

363565 2492
+
3 1264 + 1205
+ CA2 CF TF nf
1944
9

41273 340
1352 704
+
3 364 + CF2 TF2 n2f
+
3
+ CA2 TF2 n2f
486
9
27
9

1404961 1285
387
3 +
4 + 1105 46CF3 TF nf
+ CA3 TF nf
3888
4
4

252385 1045
1232
111
5125
+ CA4
+
3
4
5
CF TF3 n3f
1944
12
16
48
243

dFabcd dAabcd
512 1376
+
3 + 1205
+ nf
NA
9
3

d abcd dAabcd 131 307
d abcd dFabcd 704 512
335
3 + A
3
5
+ n2f F
NA
9
3
NA
36
6
2

863
1229
8
283 64 + CA2 TF2 n2f
3
+ CA2 CF TF nf
24
486
3
492

35345 74
13
+ 3 + 4
+ CA3 TF nf
3888
3
4

1539403 389
73
55
+ CA4
3 + 4 + 5
31104
16
16
4

abcd
abcd
dA
d
9 139
3 .
+ A
NA
8
4
(9)
The first three terms of the expansion of the anomalous dimension of the ghost field
in the linear gauge can be found in [5]. Even though the results there are restricted to the
SU(N ) group, the values for a general compact simple Lie group can be derived on the
basis of the fact that only quadratic Casimir operators occur.
As before our four-loop result is obtained in a first order expansion around the Feynman
gauge

115
c(3)
2 2
3 = CA CF T F n f
+ 403 244
27

271
743 + 1205
+ CA CF2 TF nf
12

166 32
8315 86
3 + CA2 TF2 n2f
3 + 184
+ CA TF3 n3f
81
9
972
3

22517
863 + 694 605
+ CA2 CF TF nf
432

449239 2983
423
3
+
3
4 555
+ CA T F n f
7776
24
8

123
4505
256337 2485
4
+ CA
3 +
4 +
5
3888
72
32
96

dAabcd dAabcd 21 299
dFabcd dAabcd
265
+ nf
3 +
5
(483 605 ) +
NA
NA
8
4
4

425
779 4
23 34 + CA2 TF2 n2f
3
+ CA2 CF TF nf
48
972 3

2527 7
23
+ 3 + 4
+ CA3 TF nf
7776 4
8

256273
47
25
199
+ CA4
+
3 4 + 5
62208
48
16
48

abcd
abcd
d
dA
27
9
85
3 + 5 .
+ A
(10)
NA
16
8
4
The notation is similar to the one used in [4]. Besides Riemann functions, the result
contains the quadratic Casimir operators of the fundamental and adjoint representations,
CF and CA , as well as the normalization of the trace of the fundamental representation
TF ab = Tr(T a T b ), where T a are the representation generators, and the number of fermion
493
families nf . The higher order invariants are constructed from the symmetric tensors
dFabcd =
1 a b c d
Tr T T T T + T a T b T d T c + T a T c T b T d
6

+ T a T cT d T b + T a T d T bT c + T a T d T cT b
(11)
(and similarly for the adjoint representation) and of the dimensions of the fundamental and
adjoint representations, NF and NA , respectively. Using the specific values for the SU(N )
group
1
N2 1
TF = ,
CF =
,
CA = N,
2
2N
dFabcd dFabcd
N 4 6N 2 + 18
=
,
NA
96N 2
dFabcd dAabcd
N (N 2 + 6)
,
=
NA
48
NA = N 2 1,
(12)
dAabcd dAabcd
N 2 (N 2 + 36)
,
=
NA
24
(13)
we checked that Eqs. (7)(10) are in perfect agreement with [5].

We can now combine the anomalous dimensions to reach the goal of our calculation,
i.e., the four-loop -function. Since, Zs = Zg2 = Z12 Z22 Z31 , we have = 21 22 3 ,
and

4204 352
448
2 2 17152
2
+
3 + CA CF TF nf
+
3
3 = CA CF TF nf
243
9
27
9

7073 656
424
+
CA TF3 n3f + CA2 CF TF nf
3
243
243
9

7930 224
1232
+
3 +
CF TF3 n3f
+ CA2 TF2 n2f
81
9
243

39143 136
150653 44
+
3 + CA4
3
+ CA3 TF nf
81
3
486
9

1352 704
3 + 46CF3 TF nf
+ CF2 TF2 n2f
27
9

abcd d abcd
dFabcd dAabcd 512 1664
704 512
2 dF
F
+ nf
3 + n f
+
3
NA
9
3
NA
9
3

d abcd dAabcd
80 704
+
3 .
+ A
(14)
NA
9
3
This result, manifestly gauge invariant, confirms [4]. For completeness, we reproduce the
lower order values, which we have also calculated
2 =
2857 3 1415 2
158
44
C
C TF n f +
CA TF2 n2f + CF TF2 n2f
54 A
27 A
27
9
205
CF CA TF nf + 2CF2 TF nf ,
494
1 =
34 2 20
C CA TF nf 4CF TF nf ,
3 A
3
0 =
11
4
CA T F n f .
3
3
(15)
4. Conclusions
We have completed the four-loop MS renormalization program of an unbroken gauge
theory with fermions and a general compact simple Lie group in the linear gauge. The fact
that we performed an expansion in the gauge parameter still allows for gauge invariance
tests in practical calculations, although special gauges such as the Landau gauge cannot be
chosen. If such need would occur, one may use the SU(N ) results from [5]. The correctness
of the latter relies, however, on the four-loop -function. Therefore, the present work was
not only indispensable to confirm the value of the four-loop -function itself, but also the
correctness of [5]. This goal has been reached.
It should be stressed that with the accumulated solved integrals, one could easily obtain
one more term in the expansion of the anomalous dimensions. This is due to the fact
that the complexity of the integrals occurring in the gluon propagator with at most a single
power of is comparable to the complexity of the integrals for the other functions with
at most two powers of . Having solved the latter, the gauge invariant -function enables
then to recover the third term in the expansion of the gluon propagator as well. We did
not perform such a calculation, since from the purely pragmatical point of view, only the
complete dependence would be an improvement.
A second comment concerns the independence of our calculation. The reader should
notice that besides basic software as FORM [9], Fermat [16], etc., and our own programs
[15], the only external package that we used was the Color package [17] of the FORM
distribution. On a diagram per diagram basis, we made sure that the color factors produced
by this package agree with the SU(N ) values obtained with the algorithm of [18]. Our
confidence was strengthened by the fact that we agreed with [5] on the final results.
Acknowledgements
The author would like to thank K. Chetyrkin, M. Misiak and Y. Schrder for stimulating
discussions, and the Institute of Theoretical Physics and Astrophysics of the University of
Wrzburg for warm hospitality during the period when part of this work was completed.
This work was supported in part by TMR, European Community Human Potential Programme under contract HPRN-CT-2002-00311 (EURIDICE), by Deutsche Forschungsgemeinschaft under contract SFB/TR 903, and by the Polish State Committee for Scientific
Research (KBN) under contract No. 2P03B01025.
495
Appendix A. Master integral expansions

Here we give the expansions of the master integrals depicted in Fig. 1 up to the necessary order. The notation is similar to that used in MATAD [19], i.e., we use the integration
measure
4
e E

4
d d ki
,
i d/2
(A.1)
i=1
with d = 4 2 and the constants which occur in the divergent parts of the integrals are

4
S2 = Cl2
,
3
9 3
45 3 log2 3 35 3 3 9
T1ep =
2 + 3
2
8
216
2

i
e 6
,
6 3 Im Li3
+ 6 3 Cl2
3
3

2
1
2
4 Cl2
D6 = 63 174 42 log2 2 + log4 2 + 16 Li4
(A.2)
,
3
2
3
where Cl2 (x) = Im(Li2 (eix )) is the Clausen function. Notice, however, that we use the
Minkowski space metric and Minkowski type propagators, i.e., 1/(k 2 m2 ):

PR0 = O 0 ,

PR12 = O 0 ,

0
3
1
3
1 3
+
O
,
D6
+
3
3
4

2
4
2 2

5
PR11 = 5 + O 0 ,

PR11d = O 0 ,

1 25
1 137 81
3
1 3
+ 2
S2 + 2
PR14 = 4 + 3
4
2
2
4 4

1 363
27
162 S2 3 T1ep 2 3 + O 0 ,
+
4
2

2
1
PR13 = 2 3 + (D6 + 23 ) + O 0 ,

1 53
1 51
1 1
+ 2
27 S2 + 2
PR10 = 4 + 3
2
2 12

1
1 937 243
32
S2 2 T1ep 2 3 + O 0 ,
+
12
2
6
3
PR15 =
496

1 7
1 169 27
1
1 1
+
+
S2
+
2
3
2
2
4 4 3 3 2 12

1
1 143 135
4
3
S2 T1ep + 2 3 + 4 + O 0 ,
+
3
2
6
3
2

1 1
1 7
9
1
1 1
+
+
S2 + 2
PR9d = 4
6
12 3 3 2 12 2

26 27
1
5
1
29
+
S2 T1ep 2 + 3 + O 0 ,
+

3
2
3
6
9

1 9
1 27
1 207 81
9
PR8 = 4 + 3
+ 2
S2 + 2
4
2
2
4 2

1 189 243
27
+
S2 3 T1ep + 2 + O 0 ,
2
2
2

1 491
1 3719 81
13
1 13
+ 3
+ 2
S2 + 2
PR7 = 4
96
4
4
8
48

3
329
1 13741 459
2
S2 T1ep +
2 3
+
192
8
2
24
3
381313 1593
17
3
5
+
S2
T1ep T1ep2 PR4dfin
10368
16
4
2
4
1
6805
81
61
153
PR4fin +
2
S2 2 + 3 +
4 + O(),
4
144
4
2
16
1
1 20
1
+ 2 (29 27 S2 + 22 )
PR6 = 4 + 3
3

13
1 181
16
81 S2 2 T1ep + 2 3 + O 0 ,
+
3
3
3

1 127
1 3
1 19
1 67
PR5 = 4 + 3
+ 2
+ 32 +
+ 192 53
2
2
2 2

0
+O ,

1 35
1 4565
1 5
+ 2
+ 52
PR4 = 4 + 3
144
2 3

1 58345 70
10
+ 2 3 + PR4fin + O(),
+
864
3
3

1 7
1 215 1
965 7
PR4d = 3 2
+
2 + PR4dfin + O(),

96
3
6 48

1 3
27
1 15
1
PR3 = 4 + 3
+ 2 24
S2 + 32
2
2 2

81
21
1 81
27 S2 T1ep + 2 3 + 57
S2 2 T1ep
+
2
2
2
57
27
51
T1ep2 + 2
S22 53 + 4 + O(),
2
2
8
PR9 =
497

1 601 58
1 163
2
1 29
8
+
+
+
4
+
2
2
3
12
3
3
4 3 3
2 6
220
635 163
+
+
2 +
3 + 124 + O(),
24
3
9

1
4
1
1
4
20 + 82 3
PR1 = 4 + 3 + 2 (10 + 22 ) +

3

16
+ 35 + 202 3 + 64 + O().
3
The expansions of the irreducible four-loop integrals have been obtained either from
the finiteness of the same integral with higher powers of denominators or with the help
of the method described in [12]. After taking into account the different normalization and
translating master integrals with dots to integrals with suitable irreducible numerators, the
above results for the divergent parts are in agreement with the numerical values obtained
in [6].
The integrals PR1PR4, PR4d and PR7 are needed up to constant parts. It is, however,
sufficient to express the finite part of PR7 through the finite parts of PR4 and PR4d and
keep them as symbols, since they always cancel from the divergent part of any four-loop
vacuum integral.
The reader should notice that we also needed expansions of two- and three-loop tadpoles. These have been obtained in [20,21]. The values can be read off from our results on
the reducible four-loop integrals above.
PR2 =
References
[1] D.J. Gross, F. Wilczek, Phys. Rev. Lett. 30 (1973) 1343;
H.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346;
G. t Hooft, Report at the Marseille Conference on YangMills Fields, 1972.
[2] W.E. Caswell, Phys. Rev. Lett. 33 (1974) 244;
D.R.T. Jones, Nucl. Phys. B 75 (1974) 531;
E. Egorian, O.V. Tarasov, Theor. Math. Phys. 41 (1979) 863, Teor. Mat. Fiz. 41 (1979) 26.
[3] O.V. Tarasov, A.A. Vladimirov, A.Y. Zharkov, Phys. Lett. B 93 (1980) 429;
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 (1993) 334.
[4] T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379.
[5] K.G. Chetyrkin, hep-ph/0405193.
[6] S. Laporta, Phys. Lett. B 549 (2002) 115.
[7] A.A. Vladimirov, Theor. Math. Phys. 43 (1980) 417, Teor. Mat. Fiz. 43 (1980) 210.
[8] K.G. Chetyrkin, V.A. Smirnov, Phys. Lett. B 144 (1984) 419;
K.G. Chetyrkin, Phys. Lett. B 390 (1997) 309;
K.G. Chetyrkin, Phys. Lett. B 391 (1997) 402.
[9] J.A.M. Vermaseren, math-ph/0010025.
[10] S.A. Larin, F.V. Tkachov, J.A.M. Vermaseren, NIKHEF-H-91-18.
[11] M. Misiak, M. Munz, Phys. Lett. B 344 (1995) 308.
[12] K.G. Chetyrkin, M. Misiak, M. Munz, Nucl. Phys. B 518 (1998) 473.
[13] Y. Schrder, Nucl. Phys. B (Proc. Suppl.) 116 (2003) 402.
[14] S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087.
[15] M. Czakon, DiaGen/IdSolver, unpublished.
[16] R.H. Lewis, Fermat, http://www.bway.net/~lewis/.
498
[17]
[18]
[19]
[20]
T. van Ritbergen, A.N. Schellekens, J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 41.
P. Cvitanovic, Phys. Rev. D 14 (1976) 1536.
M. Steinhauser, Comput. Phys. Commun. 134 (2001) 335.
A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123;
A.I. Davydychev, J.B. Tausk, Phys. Rev. D 53 (1996) 7381;
A.I. Davydychev, Phys. Rev. D 61 (2000) 087701.
[21] D.J. Broadhurst, Z. Phys. C 54 (1992) 599;
D.J. Broadhurst, Eur. Phys. J. C 8 (1999) 311;
J. Fleischer, M.Y. Kalmykov, Phys. Lett. B 470 (1999) 168;
K.G. Chetyrkin, M. Steinhauser, Nucl. Phys. B 573 (2000) 617.
Four-loop renormalization of QCD: full set

of renormalization constants and anomalous
dimensions
K.G. Chetyrkin 1
Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
Abstract
The anomalous dimensions of the gluon and ghost fields as well as those of the ghostghost
gluon and quarkquarkgluon vertexes are analytically computed at four loops in pQCD. Taken
together with already available anomalous dimensions of the coupling constant, the quark field and
the mass the results lead to complete knowledge of all renormalization constant entering into the
renormalization of the QCD Lagrangian at the four-loop level. As a by-product we get scale and
scheme invariant gluon and ghost propagators at NNNLO. Using a theorem due to Dudal, Verschelde
and Sorella, we also construct the four-loop anomalous dimension of the gluon mass operator, A2 ,
in the Landau gauge.
PACS: 12.38.-t; 12.38.Bx
1. Introduction
The renormalization group equation is a powerful tool in investigating the properties of
the Green functions of a renormalizable field theory. Its crucial ingredients are the anomalous dimensions of quantum fields as well as those of mass and coupling constant(s).
E-mail address: chet@particle.uni-karlsruhe.de (K.G. Chetyrkin).
1 On leave from Institute for Nuclear Research of the Russian Academy of Sciences, Moscow 117312, Russia.
doi:10.1016/j.nuclphysb.2005.01.011
500
K.G. Chetyrkin / Nuclear Physics B 710 (2005) 499510
In recent years there has been achieved a significant progress in perturbative calculation
of higher orders corrections to renormalization group functions For example, the most
physically important RG functions of QCDthe -function and the quark mass and field
anomalous dimensionshave been computed at a record-setting four-loop level [14].
In the same time the anomalous dimensions of the gluon and ghost fields are available
in literature only at three-loop level [4,5]. It is unfortunate for at least two reasons. First,
the gluon and ghost field anomalous dimensions are important in comparison of the nonperturbative results for the momentum dependence of the corresponding propagators with
perturbative predictions [617]. Second, only the knowledge of anomalous dimensions of
all fields of the QCD Lagrangian leads, together with the -function, to complete reconstruction of all renormalization constants (RCs) entering into the renormalization of the
QCD Lagrangian (see below).
In the present paper we fill the gap by analytically computing the anomalous dimensions
of the gluon and ghost fields as well as that of the ghostghostgluon vertex at four loops.
All calculations have been done in the general covariant gauge.
We apply our results to find the scheme and scale invariant gluon and ghost propagators
at next-next-next-leading order (NNNLO) as well as the four-loop anomalous dimension
of the composite operator A2 .
2. Notations and generalities

The QCD Lagrangian with nf quark flavors in the covariant gauge reads:
nf

1
f (iD
/ mf ) f
L = Ga Ga +
4
f =1

1 a 2
A + ca ca gf abc cb Ac ,
2L
(1)
where
Ga = Aa Aa + g(A A )a ,
D = igAa T a ,
f
D
/ = D .
(A B)a = f abc Ab B c ,
(2)
(3)
The quark field i has a mass mf and transforms as the fundamental representation
and the gluon fields Aa as the adjoint representation of the gauge group SU(3). Tija and
f abc are the generators of the fundamental and adjoint representation of the corresponding
Lie algebra. The ca are the ghost fields and L is the gauge parameter (L = 0 corresponds
to the Landau gauge).
By adding to (1) all counterterms necessary to remove UV divergences from Green
functions, one arrives at the bare QCD Lagrangian written in terms of the renormalized
501
fields2

1
1 3g
L0 = Z3 ( A A )2 gZ1 Aa Aa (A A )a
4
2
1 2 4g
2g 1
2
g Z1 (A A ) Z
( A )2 + Z3c c(
c)
4
2L
ccg
+ gZ1 c(A
c) + Z2
nf

g
f i/
+ gZ1 Z21A
/ Zm mf f .
(4)
f =1
2g
Here Z is expressed through the RC of the gauge fixing parameter L as follows

L,0 = Z L ,
2g
Z = Z3 /Z .
Z3 , Z2 , Z3c are the wave-function RCs appearing in the relations between the renormalized
and bare gluon, quark and ghosts fields, viz.

f
f
a
A0 = Z3 Aa ,
(5)
0 = Z2 0 ,
c0a = Z3c ca .
The full set of the vertex RCs
Z1V ,
V {3g, 4g, ccg, g}
(6)
serve to renormalize 3-gluon, 4-gluon, ghostghostgluon, quarkquarkgluon vertex

functions, respectively.
The SlavnovTaylor identities allows one to express all four vertex RCs in terms of an

independent one, Zg = g0 g , and the above listed wave function RCs. The corresponding
relations are:
Z = Z3 ,

4g
Zg = Z1 (Z3 )1 ,
(7)
Zg = Z1 (Z3 )3/2 ,
1
ccg
Zg = Z1 (Z3 )1/2 Z3c ,
(9)
(10)
g
Zg = Z1 (Z3 )1/2 (Z2 )1 .
(11)
(8)
3g
Within the MS scheme each RC does not depend on dimensional parameters (masses
and momenta) and can be represented as follows
Z(h) = 1 +
(n)

z (h)
n=1
n
(12)
where h = g 2 /(16 2 ) and the parameter is related to the running spacetime dimension
D via D = 4 2. Given a RC Z(h), the corresponding anomalous dimension is defined
2 For simplicity we set the t Hooft mass = 1 in Eq. (4).
502
as
(h) = 2

d log Z(h)
z(1) (h)
=
h
( )n h(n+1) .
=
2
h
d
(13)
n=0
Customarily, one also defines Zh = Zg2 and refers to the corresponding anomalous dimension as the QCD -function:

zg (h)
n h(n+1) .
=
h
(1)
(h) = 2g (h) = 2h
(14)
n=0
Eqs. (8)(11) imply that

4g
= 1 23 ,
3g
= 21 33 ,
ccg
= 21 23c 3 ,
g
= 21
22 3 .
(15)
(16)
(17)
(18)
As is well known there is a one-to-one correspondence between an anomalous dimension and the corresponding RC. For instance, Zh obeys an equation

log Zh
+ (h) h
= (h)
h
which leads to

dh
log Zh =
.
h
(19)
In general case Z depends on through both h and L and an analog of Eq. (19) assumes
the form

log Z
log Z
+ 3 (h)L
= (h).
+ (h) h
(20)
h
L
Eq. (20) can be easily utilized to reconstruct Z from and . As anomalous dimensions
are more compact than corresponding RCs in what follows we will write explicitly only
the former.
3. Calculations and results

Relations (8)(11) demonstrate that a minimal set of the RCs necessary to reconstruct
all coefficients of the bare QCD Lagrangian (4) consists of Zm , all three wave-function RCs
Z3 , Z2 , Z3c , and the coupling constant RC Zg or, instead, at least one from the collection
Z1V , V {3g, 4g, ccg, g}. Taking into account that Zm , Z2 and Zg are known with four
loop accuracy from the works [14], one is left with just two specific RC to compute, say,
Z3 and Z3c .
503
At present there are basically two different ways to perform RG calculations at the
four-loop level. Both approaches make use of the method of infrared rearrangement (IRR)
Ref. [18] in order to set zero (possibly after a proper Taylor expansion) masses and external
momenta. Both eventually employ the traditional integration by parts method to compute
the resulting Feynman integrals.3
The first one, pioneered in the yearly works of Dubna group [2426], amounts to adding
an artificial mass or an external momentum to a properly chosen propagator of a given
Feynman diagram before the expansion in masses and true external momenta is made.
The artificial external momentum has to be introduced in such a way that all spurious
infrared divergences are removed and the obtained Feynman integral is calculable. In practical multi-loop calculations the condition of absence of the infrared divergences leads to
unnecessary complications and, in some cases, even prevents from reduction to the simplest integrals. The problem was solved with elaborating a special technique of subtraction
of IR divergencesthe R -operation [2729]. This technique succeeds in expressing the
UV counterterm of every (L + 1)-loop Feynman integral in terms of divergent and finite
parts of some L-loop massless propagators.
In the second approach the infrared rearrangement is performed by introducing a single
auxiliary mass to all propagators in each Feynman diagram at hand [1,30,31]. No IR divergences can ever appear due to absence of any massless propagators. Next, after a proper
expansion in all the particle masses (except the auxiliary one) and external momenta is
performed. The resulting integrals are completely massive tadpoles, i.e., Feynman integrals without external momenta and with only a single mass inserted in all the propagators.
In our calculation of Z3 and Z3c we have used the first, massless approach. It proved
ccg
also to be more convenient to compute the RC Z1 instead of Z3 and then to find Z3 from
Eq. (10).
The four-loop diagrams contributing to the ghost propagator and to the ghostghost
gluon vertex to order s4 (altogether about 35 000) have been generated with the program
QGRAF [32], then globally rearranged to a product of some three-loop p-integrals with
a trivial (essentially one-loop) massive Feynman integral and, finally, computed with the
program MINCER [33,34]. The total amount of CPU time needed to compute RC Z3c and
ccg
Z1 was about a month of work of a standard PC with an Athlon XP 2000 + processor.
qqg
For testing purposes we have also computed the RC Z1 , which has required an almost
double amount of calculational time.4
ccg
Our results for the anomalous dimensions 1 and 3c read
ccg
3
1 0 = L ,
2
ccg
45
9
1 1 = L + L2 ,
8
8
ccg
5427
1053 2 135 3 135
1 2 =
L +
+

L n f ,
64
64 L
32 L
16
(21)
(22)
(23)
3 Though the technical implementations could be quite different, cf. works [1923].
4 These figures should be understood as effective ones; that is in reality we have used a sort of trivial paral-
lelization by distributing diagrams between a few PCs.
504
ccg
635749
21519
729
91125
29547 2
L +
3 L +
4 L
5 L +
1 3 =
384
32
64
128
128 L
6993
243
8505
7371 3 729
3 L2 +
4 L2
5 L2 +
+
3 L3
+
64
16
128
128 L
16
243
6075
1539 4 459
945
4 3
5 L3 +

3 L4 +
5 4
+
64 L
128
128 L
64
128 L
54623
663
99
453 2 45
+ nf
L
3 L 4 L
3 L2
288
8
4
32 L
8

251
L + 63 L ,
+ n2f
54
c
9 3
3 0 = + L ,
4 4
c
9
285
5
L + n f ,
3 1 =
16
16
4
c
15817 243
459
81
81
81
81
3 2 =
3 +
L 3 L + L2 3 L2 + L3
64
32
32 8
32
32
64
637 33
35 2
63
+ 3 L + nf ,
+ nf
24
2
16
36
c
2857419 1924407
8019
40905
368231
3 3 =
3 +
4 +
5 +
L
512
512
64
8
1536
75573
17901
12015
17613 2 4131
3 L +
4 L
5 L +

3 L2
256
128
256
512 L
128
81
21465
4185 3 9045
729
5 L2 +
+
3 L3 +
4 3
+ 4 L2
2
256
512 L
256
128 L
17145
81
3213
945
5 L3 + L4
3 L4 +
5 L4
256
32
512
256

1239661 48857
8955
3355
11107
+
3
4
5
L
+ nf
1152
48
32
4
288

39
153
651 2 27
27
4 L
L 3 L2 4 L2
3 L
8
8
128
8
32

586 55
83
4
33
779
3 + 4
L + 33 L + n3f
3 .
+ n2f
27
2
2
432
108 3
(24)
(25)
(26)
(27)
(28)
Finally, a use of Eq. (17) immediately leads us to
2
13 3
(29)
+ L + n f ,
2
2
3
531 99
9
61
(3 )1 =
(30)
+ L + L2 + nf ,
8
8
4
6
29895 243
4509
81
891 2 81
189 3
(3 )2 =
+
3 +
L + 3 L +
L + 3 L2 +
32
16
32
4
32
16
32 L
8155
215 2
333 9L
n ,
+ nf
(31)
36
27 f
(3 )0 =
505
10596127 1012023
8019
40905
2174765
+
3
4
5 +
L
768
256
32
4
768
247725
16443
170235
100575 2 18117
3 L
4 L
5 L +
+
3 L2
+
128
64
128
256 L
64
405
4455
25299 3 2619
243
4 L2 +
5 L2 +
+
3 L3
4 L3
8
128
256 L
128
64
4995
1215 4 459
945
5 L3 +
L
3 L4 +
5 4
+
128
64
256
128 L

23350603 387649
8955
3355
10879
3 +
4 +
5
L
+ nf
5184
216
16
2
36

45
1161 2 9
27
L 3 L2 + 4 L2
1563 L 4 L
4
64
2
16

43033 2017
1229
+ n2f
3 334
L + 63 L
162
81
216

4427 8
+ n3f
(32)
+ 3 .
1458 3
(3 )3 =
For an important particular case of the Landau gauge L = 0 we get

2
61
13
531
+ nf ,
(3 )1 =
+ nf ,
2
3
8
6

8155
215
29895 243
(3 )2 =
+
3 + n f
333 n2f
,
32
16
36
27
10596127 1012023
8019
40905
+
3
4
5
(3 )3 =
768
256
32
4

23350603 387649
8955
3355
+ nf
3 +
4 +
5
5184
216
16
2

43033 2017
4427 8
3 334 + n3f
+ 3
+ n2f
162
81
1458 3
(3 )0 =
(33)
(34)
(35)
or, numerically,
3 = h(6.5 0.666667nf ) + h2 (66.375 10.1667nf )

+ h3 915.963 186.86nf + 7.96296n2f

+ h4 19920.2 4692.27nf + 331.285n2f 0.169134n3f .
(36)
4. Applications
In this section we consider some applications of our results. The case of the massless
QCD with the Landau gauge fixing is understood in both subsections.
506
4.1. Scheme-invariant gluon and ghost propagators in NNNLO

In general case a (multiplicatively renormalizable) Green function G depends on both
a renormalization prescription (scheme) and the choice of the normalization scale . In
many cases it is more convenient to deal with the scheme and scale invariant version of G
Given the RG equation for G
which we will denote as G.
G = (h)G(h, ),
2 2 G(h, ) 2 2 + (h)h
(37)
h
d
reads
then a formal solution for G
= G(h, )/f (h),
G
h
f (h) = exp

f (h) = (h) 1 + (1 1 0 )h

dx (x)
,
x (x)
(38)

1
(1 1 0 )2 + 2 + 12 0 1 1 2 0 h2
2

1
1
+ (1 1 0 )3 + (1 1 0 ) 2 + 12 0 1 1 2 0
6
2

1
+ 3 13 0 + 21 2 0 3 0 + 12 1 2 1 1 2 h3
3

+ O h4 .
+
(39)
Here i = i /0 , i = i /0 (i = 1, 2, 3), the coefficients i , i are defined in Eqs. (13),

(14).
Now, after writing the gluon and ghost propagators in the form

2
q q
ab
ab 2
ab
ab
g
D
q
,
D (q) =
(40)
+
(q)
=
q
q 2
q2
q 2
and a use of explicit expressions for the very propagators from [35] we arrive at the following NNNLO predictions for the asymptotic behavior of the scheme and scale invariant
functions D and at large Euclidean Q2 = q 2 in the MS scheme. First, for the case of
pure gluodynamics (nf = 0)

13
h 22 D 1 q 2 n
f =0

412485993 9747
25085
2
=1
h+h
+
3
2904
1874048
352

141629801206331 80968605
477315
3
+h
+
3 +
5
16326706176
42592
704
= 1 8.63809h 186.819h2 5686.55h3 ,
(41)
507

9
h 44 1 q 2 n
f =0

5271
615512003 5697
2
=1
h+h
+
3
1936
7496192
704

430343889400537 674654895
73845
3
+
3 +
5
+h
130613649408
1362944
352
= 1 2.72262h 72.3825h2 2482.24h3 .
(42)
In order to illustrate the nf dependence we give below the results for nf = 3 and nf = 6
(to save space only in the numerical form)

1
h 2 D 1 q 2 n =3 = 1 5.18056h 85.0853h2 2178.1h3 ,
(43)
f

1
h 4 1 q 2 n =3 = 1 2.78472h 52.591h2 1359.11h3 ,
(44)
f

5
h 14 D 1 q 2 n =6 = 1 0.857993h + 16.6153h2 + 220.455h3 ,
(45)
f

9
h 28 1 q 2 n =6 = 1 3.27934h 31.5908h2 372.071h3 .
(46)
f
Note that in Eqs. (41)(46) the coupling constant h should be understood as h(2 =
q 2 ).
4.2. Four-loop anomalous dimension of the composite operator A2
Recently there has been a lot of activity in studying the possibility of a condensate in
YangMills theory of mass dimension two (see, e.g., recent works [36,37] and references
therein). The relevance of the operator A2 Aa Aa in the Landau gauge in that context has
been widely discussed. In this connection a thorough investigation of the renormalization
properties of the composite operator A2 Aa Aa has been carried out in [38,39]. In particular, the author of [38] has discovered by explicit three loop computation a remarkable
relation5
2A2 |L =0 = 3
(47)
A2
expressing the anomalous dimension A2 of the operator

in terms of the -function
and the gluon field anomalous dimension 3 . An all orders proof of (47) have been later
constructed in [40] using algebraic renormalization methods. As both ingredients of (47)
are known now at four loops one arrives at

35 2
137
2 1347
nf + h
nf
A2 |L =0 = h
4
3
16
12

18221 33
755 2
243
3 75607
+h
3 + n f
+ 3 +
n
64
32
72
2
108 f

29764511 99639
8019
40905
+ h4
3 +
4 +
5
1536
512
64
8
5 We have adjusted the coefficients in (47) to our notations.
508
57858155 335585
3355
8955
+
3
5
4
+ nf
10368
432
4
32

46549
6613
33
8489
4
+ n2f
+
3 + 4 + n3f
3 .
162
162
2
2916 3
(48)
5. Discussion and conclusion

Calculation of the two missing RCs Z3 and Z2c completes renormalization of the QCD
Lagrangian at four loops. An important issue relevant for any calculation of such complexity is the correctness of the obtained anomalous dimensions. The following comments are
in order.
The FORM program MINCER used by us to compute three loop massless propagators was developed more than a decade ago and has been since heavily cross-checked in a
number of various multi-loop calculations.
At three loop level we have full agreement with the results of [4,5]. This checks our
way to use R -operation because the three loop results of [4,5] have been obtained with
direct application of MINCER to three-loop propagators. In the present work the three loop
contributions come exclusively from two loop propagators.
The leading nf behaviour of the quark, gluon and ghost anomalous dimensions as
well as the anomalous dimensions of the quarkquarkgluon and ghostghostgluon vertices was investigated in all orders of perturbation theory in the work [41]. At the four loop
level the predictions of [41] are in full agreement to the corresponding leading nf pieces
of our results.6
We have also computed the O(s4 ) anomalous dimension of the quarkquarkgluon
vertex and found that it satisfies Eq. (18) as it should.
As is known from [42,43] the ghostghostgluon vertex is unrenormalized in the
Landau gauge, that is
ccg
1 |L =0 = 0.
(49)
Eq. (49) is in obvious agreement to Eqs. (21)(24).

We have not performed a direct calculation of the gluon wave function RC Z3 but
rather extract the result from the -function of [1] and the anomalous dimensions 3c and
ccg
1 . Thus, an independent reevaluation of either 3 or/and the -function at four loops is
highly desirable.
An interesting feature of the scheme and scale invariant functions D and is the full
absence of the irrational constant 4 as illustrated in Eqs. (41), (42). It stems from a nontrivial mutual cancellation of the terms proportional to 4 which enter into the ingredients
(see Eq. (38)). We
the Green function and its anomalous dimensionof the definition of G
have checked that this is true also for the SU(N ) color group and a generic value of nf .
6 We thank John Gracey for this comment.
509
Finally, a comment about the gauge group dependence of our results. Preferring compact and readable formulas to huge and general ones, the author has deliberately formatted all results in this paper for the most practically important case of the SU(3) color
group. In reality all calculations have been carried out for a little bit more general case
of the SU(N ) color group. Full expressions of RCs (and the corresponding anomalous
dimensions) describing the renormalization of the Lagrangian (4) in the general covariant gauge and for the SU(N ) color group are available (in a computer-readable form) in
http://www-ttp.physik.uni-karlsruhe.de/Progdata/ttp04/ttp04-08/.
Note added in proof

Just before the submission of the manuscript for publication we have been informed
that our main assumptionthe validity of the result for four-loop QCD beta-function first
obtained in [1]is confirmed by a completely independent calculation [44].
Acknowledgements
I would like to thank Michael Czakon and York Schrder for helpful discussions which
have motivated me to finish the old project. I also want to thank Mboyo Esole for information about the work [37].
This work was supported by the SFB/TR 9 (Computational Particle Physics), by Volkswagen Foundation and by the European Union under contract HPRN-CT-2000-00149.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379, hep-ph/9701390.
K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161, hep-ph/9703278.
J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327, hep-ph/9703284.
K.G. Chetyrkin, A. Rtey, Nucl. Phys. B 583 (2000) 3, hep-ph/9910332.
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 (1993) 334, hep-ph/9302208.
H. Suman, K. Schilling, Phys. Lett. B 373 (1996) 314, hep-lat/9512003.
D. Becirevic, et al., Phys. Rev. D 60 (1999) 094509, hep-ph/9903364.
D. Becirevic, et al., Nucl. Phys. B (Proc. Suppl.) 83 (2000) 159, hep-lat/9908056.
D. Becirevic, et al., Phys. Rev. D 61 (2000) 114508, hep-ph/9910204.
H. Nakajima, S. Furui, A. Yamaguchi, Nucl. Phys. B (Proc. Suppl.) 94 (2001) 558, hep-lat/0010083.
P. Boucaud, et al., Phys. Lett. B 493 (2000) 315, hep-ph/0008043.
P. Boucaud, et al., Phys. Rev. D 63 (2001) 114003, hep-ph/0101302.
P. Boucaud, et al., Nucl. Phys. B (Proc. Suppl.) 106 (2002) 266, hep-ph/0110171.
F.D.R. Bonnet, et al., Phys. Rev. D 64 (2001) 034501, hep-lat/0101013.
A. Cucchieri, T. Mendes, D. Zwanziger, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 697, hep-lat/0110188.
K. Van Acoleyen, H. Verschelde, Phys. Rev. D 66 (2002) 125012, hep-ph/0203211.
S. Furui, H. Nakajima, Phys. Rev. D 69 (2004) 074505, hep-lat/0305010.
A.A. Vladimirov, Theor. Math. Phys. 43 (1980) 417.
S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417.
K.G. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 518 (1998) 473, hep-ph/9711266.
510
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
P. Mastrolia, E. Remiddi, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 76.

S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
Y. Schrder, Nucl. Phys. B (Proc. Suppl.) 116 (2003) 402, hep-ph/0211288.
O.V. Tarasov, A.A. Vladimirov, Sov. J. Nucl. Phys. 25 (1977) 585.
D.I. Kazakov, O.V. Tarasov, A.A. Vladimirov, Sov. Phys. JETP 50 (1979) 521.
O.V. Tarasov, A.A. Vladimirov, A.Y. Zharkov, Phys. Lett. B 93 (1980) 429.
K.G. Chetyrkin, V.A. Smirnov, Phys. Lett. B 144 (1984) 419.
M. Misiak, M. Mnz, Phys. Lett. B 344 (1995) 308, hep-ph/9409454.
K.G. Chetyrkin, M. Misiak, M. Mnz, Phys. Lett. B 400 (1997) 206, hep-ph/9612313.
P. Nogueira, J. Comput. Phys. 105 (1993) 279.
J.A.M. Vermaseren, Symbolic Manipulation with FORM, Version 2, CAN, Amsterdam, 1991.
S.A. Larin, F.V. Tkachov, J.A.M. Vermaseren, NIKHEF-H-91-18.
K.G. Chetyrkin, A. Retey, hep-ph/0007088.
D. Dudal, et al., JHEP 0401 (2004) 044, hep-th/0311194.
M. Esole, F. Freire, hep-th/0401055.
J.A. Gracey, Phys. Lett. B 552 (2003) 101, hep-th/0211144.
D. Dudal, et al., Phys. Lett. B 574 (2003) 325, hep-th/0308181.
D. Dudal, H. Verschelde, S.P. Sorella, Phys. Lett. B 555 (2003) 126, hep-th/0212182.
J.A. Gracey, Phys. Lett. B 318 (1993) 177, hep-th/9310063.
J. Taylor, Nucl. Phys. B 33 (1971) 436, 1.
A. Blasi, O. Piguet, S.P. Sorella, Nucl. Phys. B 356 (1991) 154.
M. Czakon, Preprint DESY 04-223, SFB/CPP-04-62, 2004.
Heavy particles from inflation

G.F. Giudice a , A. Riotto b , A. Zaffaroni c
a Theory Division, CERN, Geneva, Switzerland
b INFN Sezione di Padova, Padua, Italy
c Universit di Milano-Bicocca, Milan, Italy
Received 13 September 2004; received in revised form 1 December 2004; accepted 27 December 2004
Abstract
We describe a simple and efficient mechanism by which very heavy particles are copiously created
from a primordial inflationary epoch. It works for scalar fields which are massless or very light during
inflation and acquire a large mass right after the end of inflation. Such particles can exist in realistic
scenarios, as we illustrate with several examples in the context of both field and string theory. Longwavelength fluctuations of these fields are generated during inflation with an almost scale-invariant
spectrum and may give the dominant contribution to the energy density of the heavy fields at late
times. Applications of our results to superheavy dark matter and leptogenesis are discussed.
1. Introduction
Inflation has become the standard paradigm for explaining the homogeneity and the
isotropy of our observed universe [1] (for a recent review, see Ref. [2]). If at some primordial epoch the energy density of the universe is dominated by some scalar fieldthe
inflatonwhose kinetic energy is negligible, the corresponding vacuum energy gives rise
to an exponential growth of the scale factor. During this phase a small, smooth region of
size of the order of the Hubble radius grew so large that it easily encompasses the comoving
volume of the entire presently observed universe and one can understand why the observed
universe is homogeneous and isotropic to such high accuracy. Furthermore, it is now clear
E-mail address: gian.giudice@cern.ch (G.F. Giudice).
doi:10.1016/j.nuclphysb.2004.12.037
512
G.F. Giudice et al. / Nuclear Physics B 710 (2005) 511525
that structure in the universe comes primarily from an almost scale-invariant superhorizon
curvature perturbation. This perturbation originates presumably from the vacuum fluctuation, during the almost-exponential inflation, of some field with mass much less than the
Hubble parameter H during inflation. Indeed, any scalar field whose mass is lighter than
H suffers, in a (quasi) de Sitter epoch, quantum fluctuations whose power spectrum is independent of the length scales, if they are superhorizon [35]. On the contrary, quantum
fluctuations of scalar fields whose mass M is larger than the Hubble rate during inflation
2
2
are not efficiently excited, and their power spectrum is suppressed as e2M /H [6].
While massive fields may be produced after inflation, e.g., during the process of preheating [7], in this paper we show that there are interesting cases where fluctuations generated
during inflation rather than quantum fluctuations produced after inflation give the dominant
contribution to heavy particle production. The mechanism of heavy particle production
from inflation we have in mind is based on a simple observation: particles which are massive in the present-day vacuum could have been very light during inflation. This implies
that fluctuations of a generic scalar field, with mass M H during inflation and M H
right after inflation, are copiously generated during inflation with an almost scale-invariant
spectrum and become heavy and non-relativistic right at the end of inflation. This provides
an efficient mechanism to create massive particles from inflation without any exponential
suppression. We will show that the number density of heavy particles at late times comes
from the long-wavelength modes which are far outside the horizon at the end of inflation.
As long as they are outside the horizon, these modes do not have a truly particle nature,
but nevertheless may provide the dominant contribution to the energy density stored in the
heavy field.
In Section 2 we study the quantum creation of heavy particles from inflation. Next, we
describe some interesting cases in which scalar fields can be massive in the present vacuum,
but indeed massless during inflation: field-theoretical models are discussed in Section 3 and
D-brane models in Section 4. We devote Section 5 to possible applications of our findings.
Finally, Section 6 contains our conclusions.
2. Heavy particles from inflation

We start by considering a field which is nearly massless (M < H ) during inflation, but
very massive (M H ) right at the end of inflation. Our goal is to compute its abundance
at the end of the inflationary stage and to show that the production of such massive field is
very efficient.
Since the field is nearly massless during the inflationary epoch, its vacuum fluctuations in momentum space k are generated during the almost-exponential inflation.
Indeed, the perturbations k acquire a nearly scale-invariant spectrum [25]. The fluctuations of the field have exponentially large wavelengths andas long as their wavelength
is larger than the horizonfor practical purposes they behave as a homogeneous classical
field.
During inflation, the fluctuations of the field obey the equation

2
k
k + 3H k +
k = 0,
a
513
(2.1)
where a is the scale factor and H is the Hubble rate. They can be conveniently described
in terms of the variance

2
d 3k
=
(2.2)
|k |2 ,
(2)3
which, during inflation, obeys the equation [3]
H3
d 2
=
(2.3)
.
dt
4 2
In slow-roll models of inflation [25], the Hubble rate is not exactly constant, but slowly
decreases with time according to the relation
H = H 2 ,
(2.4)
where = [V ()/V ()]2 /(16G) is one of the slow-roll parameters. Combining

Eqs. (2.3) and (2.4) and assuming that does not change appreciably with time (in slowroll, time derivatives of slow-roll parameters are proportional to the squares of slow-roll
parameters and can be legally neglected), one finds
1 dH 2
d 2
=
.
(2.5)
dt
8 2 dt
If we indicate by H the initial value of the Hubble rate at the beginning of inflation, the
value of the fluctuations of the massless field depends on such an initial value as

H2
2 = 2 .
(2.6)
8
This means that the largest contribution to the variance of the field is given by the time
when the Hubble constant took its largest value or, equivalently, that the number density of
particles at late times comes predominantly from those infra-red long-wavelength modes
which are far outside the horizon after inflation and crossed the horizon much before the
end of inflation. As long as they remain outside the horizon these modes do not manifest a
truly particle-like behaviour, but are indistinguishable from a classical homogeneous field
whose amplitude at the end of inflation is 2 1/2 101 (H / 1/2 ). This amplitude depends upon the model of inflation, but one can perform a simple estimate of it by assuming
that the universe expanded for at least 60 e-folds before the end of inflation in order to
solve the horizon and the flatness problems. In many models of inflation (but not in all)
the slow-roll parameter 1/N , when there are N e-folds to go till the end of inflation.
Taking N 60, one finds that the amplitude of the variance of the field at the end of
inflation is given by
2 1/2
(2.7)
0.9H ,
514
though inflationary models typically provide a number of e-folds much larger than 60 and
the fluctuations of the field at the end of inflation are expected to be greater.
In models of inflation where the inflaton field is well-anchored at its false vacuum, such
as the model of stringy old inflation proposed in Ref. [8], the Hubble rate is constant and
H2
the variance of the -fluctuations is given by 2 = 42 N thus providing a quantitative

estimate similar to Eq. (2.7).
Suppose now that the field acquires a large mass M right at the end of inflation. This
large mass is supposed to be greater than the Hubble rate at the end of inflation and at the
beginning of the reheating stage. The long-wavelength modes satisfy the equation
k + 3H k + M 2 k 0,
(2.8)
which tells us that modes beyond the horizon scale like k a 3/2 cos Mt. This means
that the long-wavelength modes of become non-relativistic and act like a classical homogeneous field. Their associated number density will be simply given by
M 2
(2.9)
=
.
M
2
The ratio at the end of inflation between the energy density stored in the heavy particles
and the energy density stored in the inflaton field is

M 2
1

,
(2.10)
6 Mp
n =
where Mp = 1.2 1019 GeV is the Planck mass.

The energy density in the heavy non-relativistic particles scales as a 3 , where a is the
scale factor. Meanwhile, during reheating the energy density stored in the inflaton field is
redshifted away as a 3 as well. This means that the initial ratio, given by Eq. (2.10), does
not change with time until the time when reheating occurs. At the time of reheating, the
energy density stored in the inflaton oscillations gets converted into radiation with entropy
density s = 2 2 g Tr3 /45, where g is the number of light degrees of freedom and Tr is the
reheating temperature. This yields
n
3Tr
MTr
=

.
s
4M
8Mp2
(2.11)
This ratio can be sizeable for large values of masses M and small values of the slowroll parameter . The production of heavy particles which are massless during inflation and
acquire a mass right after inflation can therefore be rather efficient.
Eq. (2.11) is approximately valid also in the case of a particle with the same mass
M ( H ) before and after inflation (see also Ref. [9]). In this case, using the condition
that is not thermally produced (Tr M), we obtain an upper bound on its abundance,
n /s H 2 /(8Mp2 ). The present constraint on tensor perturbations (H < 2 1014 GeV
[10]), therefore requires n /s 1 1011 /. In the case under consideration in this paper,
this limit can be violated, making use of the property that the mass M after inflation can
actually be much larger than H .
515
It is also interesting to compare the result of Eq. (2.11), obtained for minimally-coupled
scalars massless during inflation, with the case of conformally-coupled scalar particles
or of fermions , which have the same mass before and after inflation. Now the mass
term is the only source of conformal-symmetry breaking, and such particles can be
generated during the non-adiabatic transition between the end of inflation and the beginning of the radiation phase. Numerical simulations in chaotic inflation show that [11]
/ 103 H M /Mp2 when M H , while / is exponentially suppressed for
M H . This gives
n
H Tr
103 2 .
s
Mp
(2.12)
Again, the requirement that is not thermally produced (Tr M) gives a stringent bound
on its relic density n /s 3 1013 (H /2 1014 GeV)2 .
3. Massless states during inflation which become massive after inflation

In this section we want to show that it is indeed conceivable that states which are massive in the present-day vacuum could have been massless, or at least very light, during
inflation. Let us suppose that inflation is driven by a scalar field with potential V ().
There are various possibilities for which a given field is massless during inflation and
very massive right at the end of inflation. We describe some of them, but the list is certainly
not exhaustive.
The first option one can envisage is that the mass M of the field is provided by an
interaction term with the inflaton field itself of the form g 2 2 2 , where g is some coupling
constant. If during inflation the vacuum expectation value of the inflaton field is close to
the origin in field space, 0, the field is effectively massless. However, as soon as
inflation ends and the inflaton field rolls down the potential and acquires a large vacuum
expectation v, the field becomes massive with mass squared M 2 = g 2 v 2 H 2 . This
may happen, for instance, in the so-called small-field models [2,5] where the potential of
the inflaton field, expanded around the origin, is given by
m2 2
+ .
(3.1)
2
Here V0 represents the vacuum energy density driving inflation and the inflaton field slowly
rolls down during inflation with a mass squared m2 H 2 V0 /Mp2 . During inflation, the
V () = V0
value of the inflaton field is exponentially close to the origin [2], f eN (V0 /m Mp ) ,
where N is the number of e-folds to go till the end of inflation and f is the value of
the inflaton field at the end of inflation. This means that the field is practically massless
during inflation and its vacuum fluctuations can be generated during the almost-exponential
expansion epoch, as described in Section 2.
Another possibility may arise in hybrid models of inflation [12]. The potential relevant
for the inflationary trajectory is given by
2
2 m2 2

+ h2 |S|2 2 .
V (, S) = |S|2 2 +
2
(3.2)
516
The true vacuum of such a potential is given at |S| = and = 0. However, if h2 2

22 , the curvature of the potential along the S-direction becomes positive and the po2
tential reduces to V = 4 + m2 2 . Inflation is driven by the vacuum energy density
V0 = 4 and ends when the inflaton field reaches the critical point c2 = 2(/ h2 )2 .
If the particle acquires mass only through couplings with S, it remains massless during
inflation, when the vacuum expectation value of S vanishes, but it becomes massive after
the field S has relaxed to its true minimum.
The mechanisms described above rely on tree-level potentials, but large masses for the
scalar particles can be induced by radiative corrections. This is just another incarnation
of the usual hierarchy problem of the Higgs potential, whose solution is still unknown.
Supersymmetry or non-linearly realized symmetries could protect the scalar mass during
inflation. A more exotic possibility is that is the extra component of gauge fields in
theories with more than four dimensions. If the corresponding gauge symmetry is unbroken
during inflation, a mass term for (which is a scalar field under the 4-dimensional Lorentz
group) is forbidden. Once gauge symmetry is broken, after inflation, the field becomes
massive.
A second source of concern comes from induced interaction terms of the form
(/2) gR(g)||2 , where R(g) is the scalar curvature, function of the metrics g, and
is a coupling constant. During inflation, this term generates a squared mass for equal
to 6 H 2 , spoiling the condition M H . This problem is actually generic in supergravity, since supersymmetric scalars are conformally coupled ( = 1/6). The H 2 terms in the
mass can be made small, but at the price of a certain conspiracy between the different
supersymmetry-breaking contributions. On the other hand, Goldstone bosons are minimally coupled ( = 0), and the non-linearly realized symmetry can simultaneously protect
the mass from both quantum and H 2 corrections. This can be realized in our scenario if
the inflaton background field breaks an exact (or approximate) global symmetry: a Goldstone (or pseudo-Goldstone) boson appears during inflation. If the inflaton field vanishes
after inflation, the symmetry is restored and acquires a large mass.
Let us close this section with a final comment. In Section 2 we have assumed that the
classical field vanishes during inflation. There is no strong justification for this assumption in theories where the mass of the field is smaller than H during inflation. However,
the main purpose of the paper is to demonstrate that a large contribution to the energy
stored in the heavy field may come from the quantum fluctuations created during inflation. In this respect, the computation of the energy density stored in the field presented
in Section 2 should be regarded as conservative.
4. Massless particles in D-brane models

Other set-ups where massless fields during inflation become massive after the inflationary epochs may arise in models of inflation inspired by supergravity and (super)string
theories. In scenarios with D-brane inflation, light fields may arise as pseudo-Goldstone
bosons associated to broken translational invariance [13]. In these models, the distance
between a D-brane and an anti-D-brane plays the role of the inflaton. If the branes are
embedded in a string compactification where the background geometry has exact or ap-
517
proximate isometries, brane fluctuations transverse to the mutual distance may give rise to
almost massless fields. Inflation ends when the branes come close together and annihilate.
In this process the light fields become very massive and decay. Indeed, after the end of the
inflation the branes have disappeared and certainly cannot fluctuate anymore.
As a specific example, let us discuss the KKLMMT model [14] for D-brane inflation.1
The KKLLMT model is based on a compactification of the type IIB string theory on a
CalabiYau manifold in the presence of orientifolds and fluxes for the antisymmetric tensors. As advocated in Refs. [16,17], the fluxes can stabilize all the complex structure moduli
of the compactification (the shape of the CalabiYau). The Khler moduli (the volume) can
be then stabilized by non-perturbative superpotentials generated by additional D7-branes
(or Euclidean instantonic D3-branes) naturally present in such compactifications [17,18].
The fluxes also induce a warp factor in the metric. The KKLMMT scenario corresponds
to a CalabiYau geometry that develops a long warped AdS-like throat. The model may
be thought as a stringy realization of the RandallSundrum model (RSI) [19], where the
infra-red brane has been effectively regularized by an infra-red geometry and the ultraviolet brane has been replaced by the CalabiYau geometry (see Fig. 1).
The entire physics of inflation takes place in the throat region and it is mostly insensitive
to the details of the ultra-violet CalabiYau region. In the KKLMMT an anti-D3-brane is
sitting at the infra-red end of the throat and a D3-brane is moving toward it. In the absence
of anti-branes, the D3 is a supersymmetric BPS object and it feels no force [20]. The only
source of supersymmetry breaking in the compactification is the inclusion of the anti-D3brane. The insertion of a distant anti-D3-brane just produces a very flat potential [14] for
the scalar field corresponding to the position of the D3-brane, which plays the role of the
Fig. 1. The CalabiYau compactification with a throat and its corresponding interpretation in terms of a simplified
RSI model.
1 The KKLMMT model was proposed in order to overcome some of the problems of the D-brane inflation
scenario in flat space but it is still plagued by the familiar problem of F-term inflation in supergravity (the brane
position appears in the Khler potential and the stabilization of the volume produces a mass for the inflaton of
order H ). Various ways to solve this problem, with or without fine tuning, have been proposed in Ref. [14] and
in subsequent works on the subject [15]. Here we take the conservative attitude of assuming that this problem has
been solved and look for the presence of massless fields during the inflationary period.
518
inflaton field. The potential for the canonically normalized inflaton is

C
V () = T3 a04 1 4 ,
(4.1)
where T3 is the tension of a D3-brane, a0 is the redshift factor at the bottom of the throat
and C is a numerical constant whose value can be found in Ref. [14]. If we ignore the
order H mass that the inflaton will acquire due to the Khler potential [14], the condition of
slow-roll and the constraints from the density perturbations can be satisfied with reasonable
values of the parameters.2
For our purposes, we notice that the KS solution [20] has an SU(2) SU(2) isometry group. During its motion toward the infra-red, the D3-brane can fluctuate in the
internal angular directions. Since during the inflationary period, the two brane set-up preserves the SU(2) SU(2) symmetry, the scalar fields associated with the angular positions
are almost massless and lighter than the Hubble rate. Furthermore, being angles, they
do not get masses from the volume stabilization mechanism. In the Khler potential of
four-dimensional supergravity the volume modulus always appears in the combination
+ k(i , i ) [21], where i collectively denote the position of the branes in the six internal directions, and k is the Khler potential for the geometry. This coupling generates a
does not
mass for the fluctuations i . However, since the geometry has isometries, k(, )
depend on some of the angles. In this way, the form of potential for the angular fluctuations
is not affected by the stabilization mechanism and leaves some angles much lighter than
the Hubble rate during inflation. At the end of inflation, the D3-brane comes close to the
anti-D3-brane. At stringy distances, we cannot ignore the fields coming from open strings
connecting the branes. In particular, in a generic braneantibrane system, the open string
ground state is a tachyon T with mass of order Ms , the string scale. Since the annihilation
takes place at the bottom of the throat, the tachyonic mass is redshifted to m2T a02 Ms2 .
We can write a toy-model potential for the inflaton , the tachyon T and the angular positions i which captures the salient features of the tachyonic condensation when the brane
and antibrane are close to each other

C
V (, i , T ) = T3 a04 1 4 + T 2 2 a02 Ms2 + V (i ) + ,
(4.2)
where the ellipses stand for higher-order power terms in the field T . In particular, the form
of the potential accounts for the SU(2) SU(2) sources of breaking when the two branes
are close. Notice that this scenario is similar to that discussed in Section 3 in the context
of hybrid inflation. As long as is much larger that the (redshifted) string scale, T has
a large positive mass, its vacuum expectation value T vanishes and all the terms in the
potential involving T can be ignored. The inflationary potential is then independent of the
angles i . At stringy distances, T becomes tachyonic and triggers the braneantibrane
annihilation. When T condenses, all the fluctuation fields on the branes acquire masses
roughly of order of the redshifted string scale. Taking into account that a0 104 at the
2 In Ref. [14] the integer fluxes M, K are taken O(10), T /M 4 103 and the warping is mild a 104 .
3
0
p
The Hubble scale is consequently of typical order 109 GeV.
519
bottom of the throat, these masses are O(a0 Ms ) and much larger than H a02 Ms . All these
massive fields will eventually decay.
There are two effects that break the SU(2) SU(2) invariance and may prevent the
angular fluctuations of the brane from being exactly massless during inflation. There is a
small source of breaking of the SU(2) SU(2) symmetry due to the presence of the antiD3-brane in the infra-red, which sits at a specific point on the three-sphere. This explicit
breaking is though suppressed when the distance between the two branes during the inflationary period is sizeable. The second effect is caused by the moduli stabilization that
requires fluxes in the ultra-violet region typically breaking SU(2) SU(2); it is well known
indeed that a CalabiYau manifold has no isometries at all. Even this second effect is suppressed by the distance. An estimate based on the arguments in Appendix A.2 of Ref. [22]
shows that the typical mass induces by the SU(2) SU(2) breaking scales as a 8 (),3
where a() is the warp factor evaluated at the position of the brane. On the other hand,
2/3
using formulae in Appendix C of Ref. [14], we can easily evaluate a() a0 when there
are about 60 e-folds till the end of inflation. Therefore, the mass of the angular variables
16/3
m2 a0 is suppressed compared to H a04 by powers of the redshift factor a0 104 .
We finish this section by commenting on the dual interpretation of these massless
fields. As familiar from the AdS/CFT correspondence and the holographic interpretation
of the RS model, the throat part of the compactification can be effectively replaced by a
strongly interacting gauge theory coupled to the four-dimensional gravity. In this picture,
the D3-brane position corresponds to a flat direction in the moduli space of vacua of the
gauge theory. The global symmetry SU(2) SU(2) is spontaneously broken along this flat
direction.4 The angular positions of the brane are identified with the massless Goldstone
bosons of the spontaneously broken global symmetry.
5. Applications of heavy particle production from inflation

In Section 2 we have estimated the number density of heavy particles generated during
inflation in the case in which the large masses are acquired after inflation. The generation
mechanism does not directly involve the particle mass and it can be very efficient. We now
proceed to describe some applications of our previous findings.
5.1. Superheavy dark matter
The case for dark non-baryonic matter in the universe is today stronger than ever. The
observed large-scale structure suggests that dark matter (DM) accounts for about 30% of
3 Actually, the smallest breaking effect could be of order a 5 () with a generic CalabiYau. As in Ref. [22] we
can however choose a manifold with a discrete Z2 symmetry that forbids this effect.
4 The KS model with K and M fluxes is dual to a SU(KM + M) SU(KM) N = 1 supersymmetric theory
with pairs of bifundamental fields Ai and Bi , i = 1, 2, each one rotated by a global SU(2) symmetry, with an
SU(2) SU(2) invariant superpotential. The same compactification with a D3-brane inserted can be represented
by a gauge group SU(KM + M + 1) SU(KM + 1) broken to SU(KM + M) SU(KM) along a flat direction
where A, B acquire a VEV breaking the global symmetry SU(2) SU(2).
520
the critical mass density of the universe C = 1.88 1029 g cm3 [23]. Despite this compelling evidence, the nature of the DM is still unknown. Some fundamental physics beyond
the Standard Model (SM) is certainly required to account for the cold and slowly moving
particles, , composing the bulk of the non-baryonic dark matter. The most familiar assumption is that dark matter is a thermal relic, i.e., it was initially in chemical equilibrium
in the early universe and its present-day abundance is inversely proportional to the annihilation cross section. Since the largest possible annihilation cross section is roughly M 2 ,
one expects a maximum mass for a thermal DM particle, which turns out to be a few hundred TeV [24]. While a thermal origin for DM particles is the most common assumption,
it is not the only possibility. DM particles may be produced independently of any microphysics process. It has been pointed out that DM particles might have never experienced
local chemical equilibrium during the evolution of the universe and that their mass may be
in the range 1012 to 1019 GeV, much larger than the mass of thermal relics [2529]. These
superheavy DM particles have been called WIMPZILLAs [28]. Independently of the production mechanism, the fraction of the total energy density of the universe in superheavy
DM particles today is given by

= R

Tr
.
T0
(5.1)
Here, R h2 = 4.3 105 is the fraction of critical energy density in radiation today, T0 is
the present temperature of radiation, and the subscript indicates the epoch at which heavy
particles are generated. The present abundance of non-thermal DM particles is, as expected,
independent of the cross section [25,26], and one can easily verify that, if it amounts to
1, non-equilibrium during the evolution of the universe is automatic.
Several scenarios for superheavy DM particles have been developed, which involve
production during different stages of the evolution of the universe. They may be generated
in the transition between an inflationary and a matter-dominated (or radiation-dominated)
universe due to the non-adiabatic expansion of the background spacetime acting on the
vacuum quantum fluctuations. This mechanism was studied in details in Refs. [25,30], in
the case of chaotic inflation. The distinguishing feature of this mechanism is the capability
of generating particles with mass of the order of the inflaton mass (usually much larger than
the reheating temperature, but at most of the order of the Hubble rate during inflation) even
when the particles only interact extremely weakly (or not at all) with other particles, and
do not couple to the inflaton. Superheavy DM particles may be also created during bubble
collisions if inflation is completed through a first-order phase transition [31]; at the preheating stage after the end of inflation with masses up to 1015 GeV if they are bosons [7],
and up to the Planck scale if they are fermions [32]; or during the reheating stage after inflation [27] with masses which may be as large as 2 103 times the reheat temperature. In
addition, particles as heavy as the inflaton can be efficiently produced perturbatively, both
in inflaton decay [33] and during thermalization of energetic inflaton decay products [34].
We can use the production mechanism described in Section 2 and imagine that superheavy DM particles are massless during inflation and get their large mass right after
inflation. In such a case, the present-day abundance is readily computed using Eqs. (2.10)
and (5.1) and we find h2 = 0.1 for

1/2

TeV 1/2
M
4 1012 GeV.
Tr
102
521
(5.2)
The request that the DM particles are not produced by thermal processes (M Tr ), implies
that
1/3

3 109 GeV.
Tr
(5.3)
102
Therefore the proposed mechanism can account for DM in a large range of Tr which, in
particular, can be compatible with an acceptable regeneration of thermal gravitinos. The
upper bound on Tr reflects the extreme efficiency of the mechanism described in this paper
in creating heavy particles from inflation.
5.2. Baryogenesis
Another natural application for the proposed mechanism of heavy-particle production
is baryogenesis. Since baryon-number violation (or lepton-number violation, if sphalerons
are active) is a necessary ingredient of baryogenesis, the relevant dynamics generally takes
place at very short distances and involves very heavy particles. This is often at odds with
the request of sufficiently low Tr , coming from cosmological considerations on unwanted
relics like, for instance, gravitinos or moduli (for a recent analysis, see Ref. [35]). A mechanism for producing particles with masses much larger than Tr could resolve this conflict.
The situation is well illustrated by leptogenesis [36], one of the most concrete mechanisms to successfully reproduce the observed cosmic baryon asymmetry. Therefore we
will focus on this case, although our discussion can be extended also to the case in which
baryogenesis occurs through the decay of some GUT field, nearly massless during inflation.
Thermal leptogenesis implies a lower bound on Tr , which depends on the right-handed
neutrino mass, but it is always larger than about 109 GeV [37]. If we want to consider
lower values of Tr , we should rely either on variations of leptogenesis (soft leptogenesis [38], resonant leptogenesis [39], leptogenesis with sneutrino dominance [40,41]) or on
non-thermal production of the right-handed neutrinos N . One possibility is to generate N
at reheating by direct inflaton decay [42]. Alternatively, the right-handed neutrinos could be
produced during the reheating stage when temperatures much larger than Tr are actually
achieved [27]. If a preheating stage occurs, when large inflaton oscillations decay nonperturbatively, than a non-thermal density of heavy fermions can be generated [32]. Here
we want to study if the mechanism proposed in this paper could also create a sufficiently
large population of right-handed neutrinos.
Since N are fermions, we cannot take direct advantage of fluctuations generated at inflation. The estimate given in Eq. (2.12) shows that their density is too small to account for the
baryon asymmetry. However, there are still some interesting possibilities for leptogenesis.
First, we can consider the scalar field whose lepton-number violating VEV gives mass
to the right-handed neutrinos though the coupling N cT N . If is massless during inflation
and acquires a mass M right after, it can play the role of the field described in Section 2.
522
In particular, this could happen if is the dominant source of lepton violation during
inflation; then the associated pseudo-Goldstone boson could obtain large fluctuations. The
large number density given by Eq. (2.11), after decay, will be transferred to N . Assuming
that dominantly decays into N , the baryon asymmetry is given by
M Tr
nB
= aCP
.
s
4Mp2
(5.4)
Here a = 0.35 (both in the Standard Model and in minimal supersymmetry) relates the
BL to the B asymmetry through sphaleron interactions. For hierarchical neutrino masses,
the CP asymmetry in N decays (CP ) has the upper bound [43]
CP <
MN
3MN matm
,
= 3 107 10
2
16v
10 GeV
(5.5)
where MN is the mass of the lightest right-handed neutrino.5 Requiring that Eq. (5.4)
reproduces the observed value nB /s = 8.7 1011 and using Eq. (5.5), we obtain the
constraint
Tr
>
MN
1015 GeV
MN
2
1018 GeV
M

102
104 .
(5.6)
This shows that, for heavy N , it is possible to generate the correct baryon asymmetry even
if Tr is several orders of magnitude smaller than MN . This region of parameters (Tr MN )
is not accessible to the standard thermal leptogenesis. The proposed mechanism is most
efficient when MN is large, close to the GUT scale. Indeed, absence of thermal processes
(MN Tr ) gives a lower bound on the right-handed neutrino mass

MN
1018 GeV
M
1/2

102
1/2
1013 GeV.
(5.7)
Supersymmetry offers another possibility. The right-handed sneutrino could be the

source of leptogenesis and, if its mass term is suppressed during inflation, its density before
decay will be given by Eq. (2.11). The same argument that lead us to Eqs. (5.6) and (5.7)
now gives us the constraints
15

10 GeV 2

Tr > 2
2 1014 GeV,
MN
10
1/3

MN
6 1014 GeV.
102
(5.8)
(5.9)
These bounds are more stringent than in the previous case, where a larger density of righthanded neutrinos is achieved by considering very large masses.
5 This bound can be relaxed if the neutrino-mass hierarchy is reduced [44].
523
6. Conclusions and discussion

In this paper we have demonstrated how a primordial epoch of inflation might be associated to the generation of very heavy particles. The mechanism is rather simple once
it is realized that heavy states might have been either exactly or nearly massless during
inflation. Long-wavelength fluctuations of such fields are generated during inflation with
an almost scale-invariant spectrum and might provide the largest contribution to their energy density at late times. This mechanism is present not only in effective quantum field
theories, but also in brane-world scenarios inspired by string theory where the lightness of
would-be heavy states during inflation is guaranteed by exact or approximate isometries of
the background geometry. We have described applications of our findings to superheavy
dark matter and leptogenesis.
There are other interesting issues which would deserve further and careful investigation.
First of all, the mechanism discussed in this paper might provide a natural way to implement the curvaton mechanism to generate cosmological perturbations. It has been proposed
that the field responsible for the observed cosmological perturbations is some curvaton
field different from the inflaton [45]. During inflation, the curvaton energy density is negligible and isocurvature perturbations with a flat spectrum are produced in the curvaton field.
After the end of inflation, the curvaton field oscillates during some radiation-dominated
era, causing its energy density to grow and thereby converting the initial isocurvature into
curvature perturbation. To be operative the curvaton mechanism requires the mass of the
curvaton to be lighter than the Hubble rate during inflation and decay must occur after inflation. As we have shown, these requirements are satisfied, for instance, by brane models
where the distance between a D-brane and an anti-D-brane plays the role of the inflaton.
Brane fluctuations transverse to the mutual distance give rise to almost massless fields
during inflation, which become heavy and decay once inflation is terminated. The curvaton acts as a pseudo-Goldstone boson whose dynamics has been thoroughly studied in
Ref. [46].
Our findings might also be relevant for the dynamics of reheating after inflation. One can
envisage two extreme cases. Suppose that the decay rate of the inflaton field into ordinary
matter is highly suppressed. If so, populating the universe with known particles might occur
through -states which are efficiently generated during inflation and very massive after
inflation. On the contrary, if the inflaton couplings to ordinary matter are of order unity,
defrosting the universe after inflation is very efficient and might lead to high reheating
temperatures Tr , possibly in conflict with the gravitino bound [35]. This drawback could
be avoided if there is a subsequent release of entropy caused by the decay of heavy -states
abundantly produced during inflation.
References
[1]
[2]
[3]
[4]
A.H. Guth, Phys. Rev. D 23 (1981) 347.

D.H. Lyth, A. Riotto, Phys. Rep. 314 (1999) 1, hep-ph/9807278.
A.D. Linde, Physics and Inflationary Cosmology, Harwood, Chur, 1990.
A.R. Liddle, D.H. Lyth, Cosmological Inflation and Large-Scale Structure, Cambridge Univ. Press, Cambridge, 2000.
524
[5] A. Riotto, Inflation and the theory of cosmological perturbations, in: ICTP Summer School on Astroparticle
Physics and Cosmology, Trieste, Italy, 17 June5 July 2002, Astroparticle Physics and Cosmology, in: ICTP
Lecture Notes Series, vol. XIV, Abdus Salam International Centre for Theoretical Physics, Trieste, 2003,
pp. 317413, hep-ph/0210162.
[6] N.D. Birrell, P.C.W. Davies, Quantum Fields in Curved Space, Cambridge Univ. Press, Cambridge, 1982.
[7] E.W. Kolb, A. Linde, A. Riotto, Phys. Rev. Lett. 77 (1996) 4290, hep-ph/9606260;
E.W. Kolb, A. Riotto, I.I. Tkachev, Phys. Lett. B 423 (1998) 348, hep-ph/9801306.
[8] L. Pilo, A. Riotto, A. Zaffaroni, hep-th/0401004.
[9] J. McDonald, Phys. Lett. B 511 (2001) 1, hep-ph/0011198.
[10] U. Seljak, et al., astro-ph/0407372.
[11] V. Kuzmin, I. Tkachev, Phys. Rev. D 59 (1999) 123006, hep-ph/9809547.
[12] A.D. Linde, Phys. Rev. D 49 (1994) 748, astro-ph/9307002.
[13] S. Alexander, Phys. Rev. D 65 (2002) 023507, hep-th/0105032;
G. Dvali, Q. Shafi, S. Solganik, hep-th/0105203;
C.P. Burgess, M. Majumdar, D. Nolte, F. Quevedo, G. Rajesh, R.J. Zhang, JHEP 0107 (2001) 047, hepth/0105204;
G. Shiu, S.H. Tye, Phys. Lett. B 516 (2001) 421, hep-th/0106274;
D. Choudhury, D. Ghoshal, D.P. Jatkar, S. Panda, JCAP 0307 (2003) 009, hep-th/0305104.
[14] S. Kachru, R. Kallosh, A. Linde, J. Maldacena, L. McAllister, S.P. Trivedi, JCAP 0310 (2003) 013, hepth/0308055.
[15] J.P. Hsu, R. Kallosh, S. Prokushkin, JCAP 0312 (2003) 009, hep-th/0311077;
F. Koyama, Y. Tachikawa, T. Watari, Phys. Rev. D 69 (2004) 106001, hep-th/0311191;
H. Firouzjahi, S.H.H. Tye, Phys. Lett. B 584 (2004) 147, hep-th/0312020;
J.P. Hsu, R. Kallosh, JHEP 0404 (2004) 042, hep-th/0402047;
C.P. Burgess, J.M. Cline, H. Stoica, F. Quevedo, hep-th/0403119;
O. DeWolfe, S. Kachru, H. Verlinde, JHEP 0405 (2004) 017, hep-th/0403123;
N. Iizuka, S.P. Trivedi, hep-th/0403203.
[16] S. Giddings, S. Kachru, J. Polchinski, Phys. Rev. D 66 (2002) 106006, hep-th/0105097.
[17] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, Phys. Rev. D 68 (2003) 046005, hep-th/0301240.
[18] F. Denef, M.R. Douglas, B. Florea, JHEP 0406 (2004) 034, hep-th/0404257.
[19] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[20] I. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191.
[21] O. DeWolfe, S.B. Giddings, Phys. Rev. D 67 (2003) 066008, hep-th/0208123.
[22] O. DeWolfe, S. Kachru, H. Verlinde, JHEP 0405 (2004) 017, hep-th/0403123.
[23] For a recent review, see M.S. Turner, Phys. Rep. 333 (2000) 619.
[24] K. Griest, M. Kamionkowski, Phys. Rev. Lett. 64 (1990) 615.
[25] D.J. Chung, E.W. Kolb, A. Riotto, Phys. Rev. D 59 (1999) 023501, hep-ph/9802238.
[26] D.J. Chung, E.W. Kolb, A. Riotto, Phys. Rev. Lett. 81 (1998) 4048, hep-ph/9805473.
[27] D.J. Chung, E.W. Kolb, A. Riotto, Phys. Rev. D 60 (1999) 063504, hep-ph/9809453.
[28] E.W. Kolb, D.J. Chung, A. Riotto, hep-ph/9810361.
[29] D.J.H. Chung, P. Crotty, E.W. Kolb, A. Riotto, Phys. Rev. D 64 (2001) 043503, hep-ph/0104100.
[30] V. Kuzmin, I. Tkachev, JETP Lett. 68 (1998) 271, hep-ph/9802304.
[31] A. Masiero, A. Riotto, Phys. Lett. B 289 (1992) 73, hep-ph/9206212.
[32] G.F. Giudice, M. Peloso, A. Riotto, I. Tkachev, JHEP 9908 (1999) 014, hep-ph/9905242.
[33] T. Asaka, K. Hamaguchi, M. Kawasaki, T. Yanagida, Phys. Lett. B 464 (1999) 12, hep-ph/9906366;
R. Allahverdi, M. Drees, Phys. Rev. Lett. 89 (2002) 091302, hep-ph/0203118.
[34] R. Allahverdi, M. Drees, Phys. Rev. D 66 (2002) 063513, hep-ph/0205246.
[35] M. Kawasaki, K. Kohri, T. Moroi, astro-ph/0402490.
[36] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[37] G.F. Giudice, A. Notari, M. Raidal, A. Riotto, A. Strumia, Nucl. Phys. B 685 (2004) 89, hep-ph/0310123.
[38] Y. Grossman, T. Kashti, Y. Nir, E. Roulet, Phys. Rev. Lett. 91 (2003) 251801, hep-ph/0307081;
G. DAmbrosio, G.F. Giudice, M. Raidal, Phys. Lett. B 575 (2003) 75, hep-ph/0308031.
[39] A. Pilaftsis, T.E.J. Underwood, Nucl. Phys. B 692 (2004) 303, hep-ph/0309342.
[40] K. Hamaguchi, H. Murayama, T. Yanagida, Phys. Rev. D 65 (2002) 043512, hep-ph/0109030.
[41]
[42]
[43]
[44]
[45]
R. Allahverdi, M. Drees, Phys. Rev. D 69 (2004) 103522, hep-ph/0401054.

K. Kumekawa, T. Moroi, T. Yanagida, Prog. Theor. Phys. 92 (1994) 437, hep-ph/9405337.
S. Davidson, A. Ibarra, Phys. Lett. B 535 (2002) 25, hep-ph/0202239.
T. Hambye, Y. Lin, A. Notari, M. Papucci, A. Strumia, Nucl. Phys. B 695 (2004) 169, hep-ph/0312203.
S. Mollerach, Phys. Rev. D 42 (1990) 313;
A.D. Linde, V. Mukhanov, Phys. Rev. D 56 (1997) 535;
K. Enqvist, M.S. Sloth, Nucl. Phys. B 626 (2002) 395, hep-ph/0109214;
T. Moroi, T. Takahashi, Phys. Lett. B 522 (2001) 215, hep-ph/0110096;
T. Moroi, T. Takahashi, Phys. Lett. B 539 (2002) 303, Erratum;
D.H. Lyth, D. Wands, Phys. Lett. B 524 (2002) 5, hep-ph/0110002.
[46] K. Dimopoulos, D.H. Lyth, A. Notari, A. Riotto, JHEP 0307 (2003) 053, hep-ph/0304050.
525
Nuclear Physics B 710 [PM] (2005) 529576
D-branes in N = 2 Liouville theory and its mirror

Dan Isral a , Ari Pakman b,c , Jan Troost a
a Laboratoire de Physique Thorique de lcole Normale Suprieure, 1
24 Rue Lhomond, F-75231 Paris Cedex 05, France

b Racah Institute of Physics, The Hebrew University, Jerusalem 91904, Israel
c Erwin Schrdinger International Institute for Mathematical Physics,
Boltzmanngasse 9, A-1090 Vienna, Austria

Received 4 October 2004; accepted 11 January 2005
Abstract
We study D-branes in the mirror pair N = 2 Liouville/supersymmetric SL(2, R)/U (1) coset superconformal field theories. We build D0-, D1- and D2-branes, on the basis of the boundary state
construction for the H3+ conformal field theory. We also construct D0-branes in an orbifold that rotates the angular direction of the cigar. We show how the poles of correlators associated to localized
states and bulk interactions naturally decouple in the one-point functions of localized and extended
branes. We stress the role played in the analysis of D-brane spectra by primaries in SL(2, R)/U (1)
which are descendents of the parent theory.
PACS: 11.25.-w; 11.25.Hf; 11.25.Pm
1. Introduction
We have learned that it is very useful to study non-perturbative objects in string theory, especially when they are related to an implementation of holography. This study has
proved instrumental both for understanding gauge theory physics, and for getting to grips
E-mail addresses: israel@lpt.ens.fr (D. Isral), pakman@phys.huji.ac.il (A. Pakman), troost@lpt.ens.fr
(J. Troost).
1 Unit mixte du CNRS et de lcole Normale Suprieure, UMR 8549.
doi:10.1016/j.nuclphysb.2005.01.014
530
D. Isral et al. / Nuclear Physics B 710 [PM] (2005) 529576
with aspects of quantum gravity. In this paper, we will concentrate on constructing boundary conformal field theories for theories with N = 2 supersymmetry and central charge
c > 3. These boundary conformal field theories are arguably the most important missing
ingredient in the construction of D-branes in non-compact curved string backgrounds with
bulk supersymmetry.
The models we will study in detail are the N = 2 Liouville theory [1,2], and the
SL(2, R)/U (1) supercoset [3] theory. These two theories are known to be dual [4,5] (see
also [6,7]), and are mapped to each other by mirror symmetry.
Constructing D-branes in these superconformal field theories will not only increase our
knowledge of D-branes in non-minimal N = 2 superconformal field theories (see also [10
12]), but it will also enable us to study holographic dualities in closer detail. There are two
conjectured instances of holography where these D-branes are expected to be relevant. The
N = 2 Liouville/SL(2, R)/U (1) background is related to a conjectured calculable version
of holography [5] which involves a bulk closed superstring background [8] dual to a nongravitational non-local little string theory [9]. Note, for instance, that D1-branes stretching
between NS5-branes (which can be interpreted as the W -bosons of the little string theory)
can be constructed using these boundary conformal field theories.
Another natural area where the D-branes built in this model are relevant is the study of
matrix models for two-dimensional non-critical superstrings [11] and type 0 strings in a
2D black hole. For the latter case, it was conjectured in [13], following the ideas of [14],
that the decoupled theory of N D0-branes of ZZ type in N = 2 Liouville, leads to a
version of the matrix model of [15] with the matrix eigenvalues filling symmetrically both
sides of the inverted harmonic oscillator potential. This matrix model would be dual to 2D
type 0A string theory in the supersymmetric 2D black hole background.
Our paper is structured as follows. We first review in Section 2 the bulk theories, and the
duality between N = 2 Liouville theory and the supersymmetric coset. In Section 3 we go
on to discuss how to relate H3+ (i.e., Euclidean AdS3 ) boundary conformal field theories to
the N = 2 theories under consideration. In the following sections, we then explicitly construct and study D0-, D1- and D2-branes. We pause in Section 5 to explain how to extend
our result to an angular orbifold of the cigar. In Appendices AE, we collect several important remarks. One concerns the fact that the SL(2, R)/U (1) supercoset characters are equal
to the characters of the N = 2, c > 3 Virasoro algebra. Another treats the Fourier transformation of the one-point functions, while an Appendix C analyzes the SL(2, R) symmetry
of general N = 2 c > 3 conformal field theories.
2. The bulk
In this section we will study aspects of the bulk theory in which the D-branes studied in
this paper will be embedded. We give original points of view on some of the topics treated
in the literature. We first discuss the spectrum of the supersymmetric SL(2, R)/U (1) conformal field theory for concreteness. We then study the nature of the duality between the
susy coset and N = 2 Liouville (as well as its bosonic counterpart). We next review the
pole structure of the bulk correlators and comment upon the pole structure of the one-point
functions. We finish by suggesting that from the perspective of recent developments in
531
non-rational conformal field theories, the duality of the two theories is a case of dynamics
being completely determined by chiral symmetries.
2.1. The bulk spectrum of the supersymmetric coset
We consider a bulk theory with N = 2 supersymmetry, namely, the axial SL(2, R)/U (1)
supercoset conformal field theory [3]. We review the algebraic details of this N = 2 algebra
in Appendix A. The central charge of the conformal field theory is
6
c=3+ ,
k
(2.1)
j
where k is the level of the parent (total) SL(2, R) current algebra. The primaries m,m in
the bulk, coming from SL(2, R) primaries with spin j , can belong either to the NS or R
sector of the N = 2 algebra. They have conformal dimensions and N = 2 R-charges [8,13]
R )2
j (j 1) m2 (
+
+
,
k
k
8
R )2
2 (
j (j 1) m
j,w,n =
+
+
,
k
k
8
where
j,w,n =
m=
n + kw
,
2
m
=
R
2m
+
,
k
2
R
j,w,n = 2m
Q
,
k
2
Qj,w,n =
(2.2)
n kw
,
2
(2.3)
R is zero in the NS sector and 1 in the R sector. The Ramond

with n, w Z. The constant
primaries appear always in pairs due to the double degeneracy of the Ramond vacuum. The
numbers m, m
are the eigenvalues of the left and right elliptic generators of the SL(2, R)
Lie algebra. Notice that the conformal dimension j,w,n is invariant under j j + 1.
The two corresponding primaries are related through
j +1
m,m
m,m = R NS/R (j + 1, m, m)
=
R NS/R (j, m, m)
m,m ,
(2.4)
where the reflection coefficients R NS/R (j, m, m)

are given by [13]
R NS (j, m, m)
=
12j
12j
k )
12j
k )
(2j + 1)(1 +
(2j 1)(1
(j + m)(j m)
,
(j + 1 + m)(j + 1 m)
(2.5)
R R (j, m, m)
= 12j
12j
k )
12j
k )
(2j + 1)(1 +
12 )
(j + m 12 )(j m
(2j 1)(1
(j + 1 + m 12 )(j + 1 m
12 )
,
(2.6)
for primaries of the NS and R sectors, respectively.
532
The two-dimensional geometry of the axial coset theory

is a cigar [16,17], or Euclidean
2D black hole [18,19]. It has an asymptotic radius equal to k , and the quantum number
n is interpreted in the axial SL(2, R)/U (1) coset as the momentum around the compact
circle at infinity. The number w denotes the winding number of the closed string states
[20], but can be viewed also as the spectral flow parameter of the affine SL(2, R) algebra
[21,22].
The spectrum of the supercoset includes both continuous representations with j =
+
1/2 + iP (and P R+
0 ), and discrete lowest-weight representations Dj [20] with values
of j R satisfying
k+1
1
<j <
,
2
2
(2.7)
and
j + r = m,
j + r = m,
(2.8)
where r, r are integers (half-integers) for the NeveuSchwarz (Ramond) sector. The bound
for j (2.7) is stricter than the unitarity bound on the coset representations [23] as well as
than the bound corresponding to normalizable operators [22]. The improved bound has
been shown to apply in all physical settings [5,8,22,2426]. Although coset primaries for
discrete representations appear for all values of r, r , only those with r, r 0 correspond
to coset primaries inherited from SL(2, R) primaries, and only for them expressions (2.2)
hold. The coset primaries with r, r < 0 arise from descendants of SL(2, R) discrete lowestweight representations Dj+ , but they can be interpreted as primaries coming from discrete
highest-weight representations D(k+2)/2j

with spin (k + 2)/2 j . We discuss the details
of the r, r < 0 primaries in this section, in Section 4.1 and in Appendix A.2
The spectrum of primaries that we have reviewed above follows from studying the representation of the affine SL(2, R) algebra and from the analysis of the modular invariant
partition function of the model [8,24,26]. The reflection coefficients (2.5), (2.6) are related
to the dynamics of the theory, i.e., the correlation functions, which we consider now.
2.2. Towards the duality with N = 2 Liouville: a bosonic ancestor
Before discussing the duality between the susy coset SL(2, R)/U (1) and N = 2 Liouville, we will look at its bosonic counterpart. The correlators of the bosonic SL(2, R)/U (1)
theory can be obtained by free field computations in SL(2, R). For this one can use the
Wakimoto free field representation of the algebra3
j + = ,
j 3 =
1
,
Q
2 Note that when we take into account all primaries including those with r, r < 0, there is no need to consider
both D+ and D representations, the D+ representations being enough to cover all the spectrum.
3 We will take the SL(2, R) level k + 2, which is more convenient in order to move to the susy case later. We
take = 2.
j = 2 +
where
Q=
2
,
k
(z) (w)
2
+ k ,
Q
533
(2.9)
(, ) = (1, 0),
1
,
zw
(w)(z) log(z w)
(2.10)
and the energymomentum tensor of the theory is:

Q
1
T = ()2 2 ,
(2.11)
2
2
with the central charge given by (2.1). The SL(2, R) vertex operators are represented by
m,m = j +m1 j +m1

e(j 1)Q .
j
(2.12)
Correlators were computed in this formalism first in [27], by using the screening charge4
Q .
L1 = 1 e
(2.13)
The two-point function is

j

j
m,m m,m = |z12 |4j n+n w+w (j + j 1) + R NS (j, m, m)(j
j ) ,
(2.14)
where R NS (j, m, m)
is given in (2.5). Similar free field computations can be performed
using a dual screening charge [2830]
k e Q ,
L2 = 2 ( )
2
(2.15)
and the correlators agree under the identification [29]

k
(k)
(k 1 )
= 1
2
.
(1 k)
(1 k 1 )
(2.16)
The vertex operators (2.12) correspond to the SL(2, R) theory. To obtain the primaries
of the coset one multiplies them by the exponential of a free boson, which represents
the gauged coordinate. But the non-trivial part of any correlator computed is the same
in SL(2, R) or SL(2, R)/U (1) since both interaction terms L1 and L2 commute with the
gauged current j 3 .
The free field formalism actually allows to compute (2.14) by inserting both L1 and
L2 , and only for those values of j where the anomalous momentum conservation for is
satisfied as
2(j 1)Q n1 Q n2
2
= Q,
Q
4 Whenever we say screening charge we imply the integrated form
(2.17)
2
d z L.
534
where n1 , n2 = 0, 1, . . . are the number of insertions of L1 and L2 , respectively.5 The

results are then analytically continued to arbitrary j .
In the limit the potential drops exponentially and the theory becomes weakly
coupled. In the cigar geometry, the dilaton becomes linear in , which has the interpretation
of the radial coordinate away from the tip. Following [15], it is easy to see that the relation
between L1 and L2 is that of a strong-weak coupling duality. For k , L2 is supported
in the strong coupling region ( ) and L1 has support in the weak coupling region.
This is consistent with the fact the L1 screening can be obtained from the geometry of the
parent theory, by parameterizing AdS3 with Poincar coordinates, and the geometry of the
cigar becomes weakly curved at k since the radius tends to infinity. The relation is
inverted at the opposite limit k 0. Notice that all this is consistent with the fact that 1
and 2 are interchanged under k k 1 as follows from (2.16).
It was first observed in [4] that correlators computed in the bosonic SL(2, R)/U (1)
model coincide with those
of the sine-Liouville model, whose fields are a compact boson
X at the radius R = 2(k + 2) of the cigar, and a non-compact one with background
charge. The stress tensor of the theory is
1
1
Q
T = (X)2 ()2 2 ,
2
2
2
(2.18)
and the central charge is6 c = 2 + 6/k. The SL(2, R)/U (1) primaries are mapped to the
following primaries of sine-Liouville:

2
2
j
XL + i m
XR exp (j 1)Q .
m,m = exp im
(2.19)
k+2
k+2
The correlation functions are computed in the Coulomb formalism by using the screening
charges

k+2
1
Lsl = sl exp i
(2.20)
(XL XR ) exp ,
2
Q
which are primaries of 1 winding in the compact boson. The relevant computations can
be found in [31] for the two-point functions and in [32] for some three point functions.7
Using L
sl as screening charges, the anomalous momentum conservation for a two-point
function is

1
2(j 1)Q n + n+
= Q,
Q
(2.21)
where n is the number of insertions of L

sl .
5 Note that the values of j selected by (2.17) are nothing but 2j = 1 + n + n k, which correspond to degen1
2
erate representations of the affine SL(2, R) algebra.
6 This central charge differs by 1 from the central charge in (2.1). This corresponds to the addition of a trivial
boson as mentioned before.
7 The analytical structure of correlators computed in [32], was recently shown in [33] to agree with that
obtained in the SL(2, R) approach for winding-violating processes.
535
An important result of [4] is that in the computation of N -point functions, with N 3,

the correlators can violate winding number by up to N 2 units. This result is easily
obtained in the sine-Liouville side, as shown in [32], since the integrals to which the correlators reduce in the Coulomb formalism vanish when the difference n n+ does not have
the correct value. The same result, of course, appears in the SL(2, R)/U (1) side, though
through some hard work [29,34]. Note that as a consequence, the perturbative computation
of a two-point function requires n = n+ , and thus an even amount of insertions of L
sl .
interaction
belongs
to
In terms of strong-weak coupling duality, the sine-Liouville L
sl
the same side of the duality as the interaction Lagrangian L2 . The relation between the
coupling constants sl and 1 was shown in [35] to be

sl 2/k
(k 1 )
(2.22)
,
= 1
k
(1 k 1 )
from which, using (2.16) it follows that

(k)
sl 2
.
= 2
k
(1 k)
(2.23)
This quadratic relation between 2 and sl is consistent with KPZ scaling, since it is clear
from (2.17) that any value of j which is screened with n2 insertions of L2 , can be equally
screened with twice as much insertions of L
sl .
2.3. The supersymmetric case
The supersymmetric version of the equivalence between the bosonic coset SL(2, R)/
U (1) and the sine-Liouville theory, is the celebrated mirror duality between the supersymmetric coset SL(2, R)/U (1) and the N = 2 Liouville theory. This duality was conjectured
in [5]. It was then shown in [6] that the equivalence between the actions of this two N = 2
theories follows from mirror symmetry, using the techniques of [36]. The same equivalence
was shown in [7] to follow from an analysis of the dynamics of domain walls.
In both analyses of [6] and [7], crucial use is made of the explicit N = 2 supersymmetry in the action. On the other hand, it is easy to see that the computational content of
the duality, i.e., the identity of the correlators, is exactly the same as that of the bosonic
version. In the supersymmetric coset SL(2, R)/U (1) side, the primaries are obtained from
the parent susy SL(2, R)k model. In the latter, one can decouple the fermions and shift
the level k k + 2 (see Appendix B), so that the NS primary states are the product of a
bosonic SL(2, R)k+2 primary and the fermionic vacuum. Descending to the coset involves
removing a free U (1) boson, so the correlators reduce to those of a purely bosonic model
at level k + 2, with the screening charges given by (2.13) and (2.15). In the Ramond sector,
a coset primary is obtained by extracting the contribution of the gauged U (1) from the
product of an SL(2, R)k+2 primary and a spin field of the decoupled fermions (see, e.g.,
[13]). The computation reduces also to that of the bosonic case, but the dependence on the
quantum number m is now shifted to m 1/2, as can be seen in (2.6).
The N = 2 Liouville is an interacting theory for a complex chiral super-field , with
a Liouville potential. For some previous works on N = 2 Liouville see [1,2,37,38]. The
536
super-field has a non-compact real component with background charge, a compact

imaginary component Y , and the corresponding fermions. The stress tensor is
1
Q
1
1
1
T = (Y )2 ()2 2 Y Y
2
2
2
2
2
(2.24)
with a central charge given by (2.1). The compactness of Y actually allows to define the following chiral/anti-chiral superfields in N = (2, 2) superspace of coordinates (z, ; z , ):
= i(YL YR ) + i ( i Y ) i ( iY ) + i F .
(2.25)
Correspondingly we have the following two N = 2 Liouville chiral interactions:

1
LN =2 = 2 d 2 e Q
=
2 Q1 [i(YL YR )]
e
( iY )( i Y ).
Q2
(2.26)
In the expansion we have set to zero the auxiliary field F . Its presence only contributes
contact terms in the correlators, so it can be ignored [39]. At this point it is convenient to
bosonize the fermions as
i Y
= eiHR ,
iY
= eiHL ,
(2.27)
so that the interaction terms become8

Q i[ Q YL +HL ] i[ Q YR +HR ]
L
e
e
.
N =2 = k2 e
1
We will now rotate the two bosons YL and HL as

k+2
k+2
1
1
XL = YL + HL ,
ZL = YL + HL
2
Q
2
Q
(2.28)
(2.29)
and similarly for XR , ZR . The two bosons XL , ZL commute and are canonically normalized. Making this change of variables in the interaction (2.28), we see that ZL completely
decouples from the interaction term, and L

N =2 becomes identical to the bosonic Lsl , as
announced. This change of variables is actually the N = 2 Liouville equivalent of the chiral rotation in susy SL(2, R) that allows to decouple the fermions by shifting the level k
to k + 2. We again arrived to a form of the screening charge without fermions,and the
coefficient
of the compact boson in the interaction has been shifted from 1/Q = k/2 to
(k + 2)/2. The NS primaries of the model are given by (2.19). The Ramond sector is
treated as in the SL(2, R)/U (1) case with the same result.
By bosonizing the fermions, we have lost explicit N = 2 supersymmetry at the level
of the action, which was so important in the approaches of [6,7]. This is not uncommon
in conformal field theories, where fermions are typically bosonized in order to compute
8 The interaction (2.28) is the N = 2 point in a continuous family of theories studied in [31].
537
correlators.9 In our case it is a signal that the reason for the identity of the correlators may
not lie in the form of the action of the theory (see later).
A dual, non-chiral interaction term is also allowed by the N = 2 symmetry of the N = 2
Liouville theory:

Q
+
LN =2 = 2 d 4 e 2 ( + ) .
(2.30)
We now observe [38] that this screening charge coincides with the screening charge of the
SL(2, R)/U (1) supercoset, i.e., the supersymmetric equivalent of (2.13). Using the same
steps of bosonization and field redefinitions one can reduce this equivalence to the bosonic
one already discussed. Thus this circle of ideas clarifies the fact that the N = 2 Liouville
duality proposed in [37] is nothing but the bosonic duality of [4] discussed in the previous
section.
2.4. Bulk versus localized poles and self-duality
Poles in the correlators of our model can be either of bulk or localized type. Bulk
poles correspond to interactions taking place along the infinite direction of the (asymptotic)
linear dilaton. On the other hand, localized poles are associated to discrete normalizable
states living near the tip of the cigar.10
In particular, in the two-point function (2.5), the first two gamma functions of the numerator have bulk poles, and the last two gamma functions, with m, m
dependence, have
localized poles. We analyze here the NS two-point function, the R case being similar.
Let us consider first the bulk poles. The first gamma function has single poles at values
2j = 1, 2, . . . . From (2.17), we see that these values of j are screened by n1 = 2j 1
charges of type L1 , and no L2 charges. In the Coulomb formalism, one first separates the
and then the integral of the zero mode 0 over its
non-compact field into = 0 + ,
infinite volume gives the pole [41]. The second gamma function has poles at (2j 1)/k =
1, 2, . . . . These values of j are screened in SL(2, R)/U (1) by n2 = (2j 1)/k charges
of type L2 , and no L1 charges. In the sine-Liouville theory, these corresponds to having
+
n+
sl = (2j 1)/k charges of Lsl , and the same amount of Lsl . This follows from (2.21)
+
and the condition nsl = nsl for two-point functions.
The same phenomenon of having two families of bulk poles occurs in bosonic Liouville
theory, where the two-point function has two sets of poles, their semi-classical origin being
2
the insertion of either the Liouville interaction L e2b or its dual L e b , with a relation
between L and L similar to (2.16) [42].
The bulk poles in (2.5) are simple poles, except at the level k = 1. At this level, corresponding to c = 9, both the first and second gamma functions in (2.5) have each a pole
9 Explicit N = 2 supersymmetry at the level of the action is not necessary for a conformal field theory to
have N = 2 supersymmetry in its spectrum and in its chiral algebra. For example, an N = 2 minimal model with
central charge c = 1 can be realized through a free compact boson. We thank A. Giveon for comments on this
point.
10 See [40] for a recent discussion on this double nature of poles in the context of holographic descriptions of
little string theories.
538
at 2j = 2, 3, . . . . This is signaled by the fact that the two dual charges L1 and L2 become
equal for k = 1, including 1 = 2 , so this the self-dual point of the theory. There is only
one single bulk pole left at k = 1, that comes from the first gamma function of (2.5) at
2j = 1. But this is the expected result since for this j no screening charges are needed to
satisfy the anomalous momentum conservation (2.17), and the single pole comes from the
infinite volume of the zero mode of .11
Notice that although one can interpret L1 as being a strong-weak dual to both L2 and
L
sl , the sine-Liouville interaction remains outside the liaison between the Wakimoto pair
L1 and L2 , and this becomes more manifest at the self-dual point k = 1.
As mentioned, localized poles occur in the third and fourth gamma functions of the
numerator of (2.5). They signal the presence of normalizable bound states in the strong
coupling region associated with the residue of the singularity. We will consider here the
case m = m
= kw/2, which is the relevant case for the D-brane analysis. At values of
r = kw/2 j 0, corresponding to primaries of SL(2, R)/U (1) coming from primaries
of SL(2, R), the fourth gamma function in the numerator of (2.5) has a localized pole. On
the other hand, at values of r = kw/2j < 0 corresponding to primaries of SL(2, R)/U (1)
which are descendants of SL(2, R), the two-point function has a zero from a pole in the
third gamma function in the denominator of (2.5). We will return to this issue in Section 4.1.
A qualitative feature that we wish to stress is that, for the theories with boundary that we
will construct, the two kinds of poles naturally decouple. One-point functions of localized
D0-branes will have only poles associated to the discrete normalizable states, and those
of extended D1-branes will have only bulk poles associated to the zero-mode of the
radial coordinate. Intuitively, on the one hand, the D0-branes are localized near the tip
of the cigar, as are the normalizable bound states, and on the other hand, the D1-branes
stretch along the radial direction and only couple to momentum modes, thus forbidding the
coupling to discrete bound states that all carry a non-trivial winding charge. In the last case
of D2-branes, we have both type of poles since these non-compact branes have a induced
D0-brane charge localized at the tip of the cigar [43].
A similar decoupling phenomenon can be observed for the one-point functions of
bosonic (and N = 1 [44]) Liouville theory. In this case, the ZZ one-point functions [45]
have no poles and the FZZT one-point functions [46] have the bulk Liouville poles mentioned above.
2.5. The conformal bootstrap approach
We have reviewed above how perturbative calculations with different screening charges
lead to the same correlators in both theories. The perturbative approach leads to physical
insights related to the nature of the poles, strong-weak coupling regimes, etc. Also, in the
11 The importance of the self-dual point at k = 1 has recently been stressed in [33]. The same self-duality phenomenon, with single poles becoming double poles, occurs in Liouville at c = 25 (b = 1). It would be interesting
to see if these features of the k = 1 point are captured in the matrix model for the 2D black hole proposed in [15],
which is an extension of a matrix model for c = 25 Liouville (coupled to c = 1 matter).
539
last years a new powerful approach to compute correlators in non-rational conformal field
theories has appeared [47]. In non-rational conformal field theories the normalizable states
have a continuous spectrum (in H3+ they correspond to the continuous representations of
SL(2, R)) and appear in the intermediate channels of the correlators. The new approach
consists in assuming that a general property of the conformal bootstrap, namely the factorization constraints, can be analytically continued to primary states corresponding to
non-normalizable degenerate operators with discrete spectrum. This assumption, together
with assuming that a strong-weak coupling duality is present in the theory (of the type
between L1 and L2 above), leads to constraints for two and three point functions which
have a unique solution. We will not review this method here, and we refer the reader to
[42,46,47] for details.
A natural question is what new light can be shed on the equivalence between N = 2
Liouville and the susy SL(2, R)/U (1) through these methods. This question is related to
the more general issue of what role the action or the perturbative screening charges play
in this approach. The method essentially reduces to a minimum the dynamical information
needed to solve the theory. It asks as an input two pieces of information: (i) the quantum
numbers of a degenerate operator, which are given by the chiral symmetries of the theory,
and (ii) the value of certain fixed correlators involving one or two screenings. As an output
we get the value of arbitrary correlators. In the sine-Liouville/SL(2, R)/U (1) context, this
method has been exploited in [35] to obtain the relation (2.22) between coupling constants
1 and sl .
Now, at least in some cases, it turns out that the second piece of input, namely, the
perturbative fixed correlators, can be obtained by considering factorization constraints implied by higher level degenerate operators.12 This means that the chiral algebra of the
model would fully fix the correlators, through the family of its degenerate primaries. The
argument holds both for bulk and boundary theories. In this way, for example, in [48],
factorization constraints for the boundary H3+ theory were obtained without introducing a
boundary action. In other words, under certain analyticity assumptions one could in principle achieve for non-rational conformal field theories, what is known to hold for rational
ones, namely, that the chiral symmetries of the theory completely fix the correlation functions (under a certain prescription as to how left and right fields are glued). Notice that in
our theory many factors of the two-point function (2.5) can be seen as the result of Fouriertransforming the same object from a basis of primaries where the SL(2, R) chiral symmetry
is realized through differential operators (see Appendix B). So the idea would be to push
these symmetry constraints further to their very end.
Carrying this program in our case, the SL(2, R)/U (1) and N = 2 Liouville theories
would appear just as different realizations of the same chiral structure. The latter would be
nothing but the common core of their respective chiral algebras, affine SL(2, R) and N = 2,
c > 3 Virasoro, which are [23] related by a free boson which does not affect the correlators.
In Appendix D we show how an N = 2, c > 3 algebra yields always an SL(2, R) algebra
by adding a free boson. Concerning the possible additional geometrical information,
namely, the way left and right chiral fields are glued, there is only one known modular
12 We thank V. Schomerus for this crucial comment.
540
invariant with N = 2, c > 3 spectrum [8,26], up to discrete orbifolds acting on the N = 2

charges.
That is the idea behind a central assumption of our paper, namely, that boundary CFT
quantities computed in H3+ , which descend naturally to SL(2, R)/U (1), describe also the
boundary theory of N = 2 Liouville.
3. The boundary
3.1. The bosonic coset
Recall that for the bosonic coset we have that the one-point functions [43] are given
in terms of the product of one-point functions for H3+ and the one-point functions for an
auxiliary boson X (at radius R = k) which we can give the geometric interpretation of
being the angular variable in the coset. The left and right chiral U (1) quantum numbers of
H3+ (labeled nH and pH ) and X are related as:

nH ipH
n + kw n kw
nH + ipH
,
=
,
,
(3.1)
2
2
2
2
where the relative minus sign arises because we gauge axially. We then use the factorization
of the one-point function:
j,coset j,H3 X
n,w
(3.2)
= n,w Vn,w
to obtain the one-point function in the coset [43]. Since X has the geometric interpretation
of being the angular variable, Dirichlet conditions on X (i.e., Neumann conditions on the
H3+ gauged current) have the interpretation of branes localized in the angular direction.13
Note that we have assumed the ghost contribution to the one-point function to be trivial.
(We can detect non-trivial renormalizations through the Cardy check.) Note also that our
derivation is only valid for coset primaries that are associated to primaries in the parent
theory. Coset primaries associated to descendents in the original model require special
care.
3.2. The supercoset
For the supercoset, we can tell an analogous story. We have that the one-point functions
are given by a product of one-point functions for H3+ at level k + 2, for a U (1) associated to
the fermions, and for an auxiliary boson X that has again the interpretation of the angular
direction. The one-point functions factorize, and we have:
j,supercoset j,H3 X
n,w
(3.3)
= n,w Vn,w VF
.
We can be more precise about the relationship between the boundary condition for the
various currents. The left and right N = 2 R-currents are given in terms of the total currents
13 See next paragraph for a more rigorous argument based on BRST symmetry.
541
J 3 , J3 as follows (see Appendix A for conventions):

2J 3
2J3
and JR = +
.
(3.4)
k
k
The A-type boundary conditions of the N = 2 algebra [49] are defined through the
, = 1
twisted gluing conditions (in the closed string channel): J R = JR , G = iG
being the choice of spin structure. Thus we see from (3.4)and the expressions for the
supercurrentsthat the total currents of the SL(2, R) algebra has to satisfy untwisted
boundary conditions: J 3, = J3, . As we know from [48,50] this corresponds to AdS2
branes in AdS3 . The axial supercoset has a BRST charge corresponding to the gauge-fixed
local symmetry, whose expression is (see, e.g., [51]):

dz 3
c J + iX + 3 + X
QBRST =
2i

d z 3
+ 3 X .
c J i X
(3.5)
2i
J R = + +
Thus the preservation of the BRST current will impose that the extra boson X has the
in the closed string channel, i.e., Dirichlet conditions. The
boundary conditions X = X
net effect of the (, ) superghosts will be to remove the contributions of the fermions 3
associated to J 3 and X associated to X, leaving the fermions with (relative) A-type
boundary conditions. In the cigar these are the D1-branes, extending to infinity [43,52].
The B-type boundary conditions of the N = 2 algebra [49] are defined through the
. Using
untwisted gluing conditions (in the closed string channel): J R = JR , G = iG
the same lines of reasoning we find that these boundary conditions corresponds to twisted
boundary conditions for the SL(2, R) currents, J 3 = J3 , J = J . These are either H2
branes or S 2 of imaginary radius. In the former case we obtain D2-branes in the cigar, and
in the latter D0-branes localized at the tip of the cigar [43,52]. In both cases the extra field
X has to satisfy Neumann boundary conditions.
To summarize, the one-point functions for the supercoset are as in [43] (but, importantly,
the basis of Ishibashi states to which they correspond is a basis of Ishibashi states that
preserves N = 2 superconformal symmetry and the level is shifted by two units), with an
additional factor corresponding to two real fermions.
We move on to apply the dictionary above to the particular cases of D0-, D1- and
D2-branes in the supersymmetric N = 2 theories. We will be using the quantum number notations traditional for the SL(2, R)/U (1) supercoset, for convenience of comparison
with (technically similar) results in the bosonic coset [43]. But it should always be kept
in mind that the construction equally well applies to N = 2 Liouville theory, since the
primaries in one theory can be associated to unique primaries in its dual and since the
characters in both theories are identical.
4. D0-branes
The first type of branes we discuss are the B-type branes localized at the tip of the cigar
SL(2, R)/U (1) conformal field theory.
542
4.1. One-point function for the localized branes

These one point functions for the D0-branes of N = 2 Liouville are similar to those of
the ZZ branes [45] for bosonic Liouville theory. Their expression is (see Appendix B):
nw
D0
u
= n,0
u (j, w)
,
|z z |j,w
(4.1)
where
uNS (j, w) = k 1/2 (1)uw 1/2j
(j +
kw
2 )(j
(2j 1)(1
kw
2 )
12j
k )
sin k u(2j 1)
,
sin k (2j 1)
uNS (j, w) = i w uNS (j, w),
uR (j, w) = k 1/2 (1)uw 1/2j
(j +
kw
2
12 )(j
(2j 1)(1
kw
1
2 2)
12j
k )
sin k u(2j 1)
.
sin k (2j 1)
(4.2)
The normalization constant k 1/2 (1)uw (and i w ) is fixed to satisfy the Cardy condition
(see next section).
The numbers u = 1, 2, . . . are inherited from the one-point functions of localized
D-branes in H3+ , and they correspond to finite-dimensional representations of SL(2, R) of
spin j = (u 1)/2. The latter are only a subset of the degenerate representations of the
SL(2, R) affine algebra.14 Remember that in the bosonic and N = 1 Liouville theory, the
localized ZZ branes [44,45] have two quantum numbers, associated to all the degenerate
representations of the (N = 1) Virasoro algebra. This suggests that more general solutions
of the factorization constraints in [48] are expected, which in turn would imply a bigger
family of D-branes for our N = 2 model.
For a discrete state with pure winding, the numbers j and w are correlated as [8]
kw
(4.3)
2
with r Z for the NS sector and r Z + 1/2 for the R sector. Moreover, the unitarity
bound (2.7) for j implies that for every r there is a unique pair of j, w such that (4.3)
holds.
Given a j in the unitary bound (2.7) and satisfying (4.3), the one point functions (4.2)
have poles at values of r 0, w > 0. The case r < 0 should be treated differently, since
from the point of view of the parent H3+ or SL(2, R)theory, the states with w > 0 and
w 0 are of very different origin. Indeed in the former case, Eq. (4.3) can be solved with
r 0, hence those states descend from flowed primaries of a lowest weight representation
j,w
D+ of the SL(2, R) affine algebra, see [21,22]. Accordingly the formulae (4.2) are obtained from descent of the H3+ onesas in [43]valid for primaries of H3+ . The one-point
function has simple poles for all these states, coming form the second gamma function in
j +r =
14 They correspond to n = 0 in footnote 5.

2
543
the numerator of (4.2). On the contrary, in the latter case w 0, the solution of (4.3) is
solved with r < 0. Those states, while primaries of the coset,15 are descendents of the
SL(2, R) flowed algebra. To use nevertheless the formulas of the H3+ branes, we have to
, with j = (k + 2)/2 j . Applyuse the isomorphism of representations: Dj+,w Dj,w1

ing this mapping on the one-point functions, we find poles coming form the first Gamma
function in the numerators of (4.2).
4.2. Cardy computation for the D0-branes
In this section we will verify that the one-point functions of localized D-branes obtained
in Section 4.1 satisfy the Cardy condition relating the open and closed string channels. We
As usual [53],
distinguish between three cases for the open string spectrum: NS, R and NS.

they correspond to NS, NS and R sectors in the closed string channel, respectively. We will
show below the computation in detail for the NS/NS case. For the other cases, we state the
result, and defer details to Appendix C.
For the annulus partition function of a NS open string stretching between a brane labeled
by u = 1, 2, . . . and the basic brane with u = 1 we consider

NS
Zu,1
(, ) =
chNS
f (u, r; , )
rZ
3 (, )
( )3

sZ+1/2
s 2 +su 2s+u
1 s 2 su 2su
k y k
k y k
q
.
q
1 + yq s
(4.4)
We have taken the N = 2 NS characters (A.37) associated to the u-dimensional representations of SL(2, R) of spin j = (u 1)/2, and we have summed over the whole spectral
flow orbit. For an open string stretching between two branes with general boundary conditions u and u , we expect, as in [43,48,54], that the partition function is obtained by
NS over the irreducible representations appearing in the fusion of the represensumming Zu,1
tations associated to u and u . Calling the latter j = |j | and j = |j |, we have
j j = |j j | |j j | + 1 j + j .
(4.5)
In terms of the u, u indices, this decomposition implies

NS
Zu,u
(, ) =

min(u,u
)1
NS
Zu+u
2n1,1 (, ).
(4.6)
n=0
In order to verify the Cardy condition on these branes, we will perform a modular transNS to the closed string channel.16 Let us start with the Z NS partition
formation of Zu,u

u,1
function (4.4). We will parameterize as: = 1 2 , 1,2 R. The modular transform
15 In fact these diagonal states are obtained from the lowest weight state of the representation. In a bosonic
r
model they are: (J1
) |j, j
. For the supersymmetric case, see Appendix A.
16 The result for the NS/NS case follows as a particular case of an identity proved in [55,56], and the other

and Zp orbifold) are variations thereof. We perform the computation here
cases considered below (NS/R,
R/NS
for completeness and in order to adapt the notation to our present purposes.
544
Fig. 1. Choice of contour of integration (for 1 > 0).
NS (, ) can be expressed as
of Zu,1
c 2
ei 3

2i 2
1
3 (, ) 1
NS
,
dZ (i)eZ+ k Z
=
Zu,1
+
3

( ) 2i
C
C+
sinh(2 Zk u) ei(i Z)
,
cosh(Z) cos (i Z )
(4.7)
where the parameter is infinitesimal and positive. The contour encircles the line C: Z =
iy/ + i2 , y R (see Fig. 1). This identity is valid because we pick up all the poles of
the integrand inside the contour. These are the zeroes of cos (i Z ) which occur at
i Z = y + 1 = s Z + 1/2. At these poles, the contour integral yields the residues
2 ei(s)/ sin(2 uk (s )/ )ei2(s)
ei(s)/ + ei(s)/
2 /(k )
(4.8)
which lead to the identity (4.7). This identity is valid as long as there are no poles coming
from the cosh Z factor on the contour of integration. These special cases occur only for
2 Z + 1/2 and we assume that 2 does not take these values. We now note that we have
the expansions:

ei(i Z)
=
(1)w e2iw(i Z) ,
2 cos (i Z )

for ei2(i Z) < 1
(4.9)
w=0
and
1

ei(i Z)
(1)w e2iw(i Z) ,
=
2 cos (i Z )
w=

for e+i2(i Z) < 1.
(4.10)
545
Fig. 2. Change of contour of integration (case w > k2 /2).
The expansion (4.9) is valid in C+ and (4.10) is valid in C . Plugging these expansions
into the right-hand side of Eq. (4.7) we get

2
1
i 3c
NS
e
Zu,1 ,

Z

2
3 (, )
w sinh(2 k u) Z w iwZ+ Zk
e
(4.11)
=
dZ
(1)
y
q
.
cosh(Z)
( )3
wZ C
When taking 0 we added a minus sign to the contour C+ and switched its direction.
NS , we use the
To obtain the corresponding expression for the general case (4.6) of Zu,u

identity

min(u,u
)1
n=0

sinh(2 Zk u) sinh(2 Zk u )
Z

,
sinh 2 (u + u 2n 1) =
k
sinh(2 Zk )
(4.12)
so we get

1
,
e

Z
Z

2
3 (, )
w sinh(2 k u) sinh(2 k u ) Z w iwZ+ Zk
=
e
dZ
(1)
y
q
.
( )3
cosh(Z) sinh(2 Zk )
i 3c
NS
Zu,u

wZ C
(4.13)
Note that the exponent of q is complex. In order to get a real exponent for q we will
(i) shift the C contour of integration by iwk/2 i2 , for each term indexed by w and
(ii) tilt it parallel to the real axis (see Fig. 2). The resulting exponent of q will be that
needed to get the characters of the continuous representations. As we shift the contour we
pick poles which will make up the contributions of the N = 2 discrete representations to
the closed string channel amplitude. No poles are picked when tilting to the real axis. So
NS is decomposed into
the modular transform of Zu,u

546

c 2
ei 3

1
NS,c
NS,d
NS
= Zu,u
Zu,u
,
+ Zu,u .

(4.14)
The first term comes from the continuous integral, which after shifting and tilting the contour, becomes
NS,c
Zu,u

+

=
dP
wZ
e(
iwk
2 +P )
sinh(2 Pk u) sinh(2 Pk u )
P
(1)w(uu ) cosh((P + i wk
2 )) sinh(2 k )
P 2 +(wk/2)2
k
yw
3 (, )
.
( )3
(4.15)
In the last factor we get a continuous character chNS

c (P , wk/2; , ) (see (A.20)), corresponding to a pure winding state with m = kw/2, as expected. And since chc (P , wk/2;, )
is even in P we can rewrite this as
+

2 sinh(2P ) sinh(2 Pk u) sinh(2 Pk u )
NS,c
dP
Zu,u =

(1)w(uu ) [cosh(2P ) + cos(kw)] sinh(2 Pk )
wZ 0

wk
P
,
chNS
(4.16)
;
,
.
c
2
This expression is equal to
NS,c
Zu,u

+

1
1
wk
NS

iP , w u
+ iP , w chc P ,
; , ,
=
dP u
2
2
2
wZ 0
(4.17)
so we get the expected continuous spectrum from the overlap between D0 boundary
states.17
The poles picked up arise from the zeroes of the factor cosh Z in (4.13), at values of
Z such that Z = is with s Z + 1/2. Note that for irrational k the shifted contour will
never fall on a pole, and we will assume this to be the case. The poles will contribute with
positive sign for s > 2 and with negative sign for s < 2 . Let us consider the former case.
A given s Z + 1/2 will give a pole contribution for all the terms in (4.11) with w ws ,
where ws is the only integer satisfying
2s
> ws 1.
(4.18)
k
For poles corresponding to s < 2 , we get contributions for all the terms in (4.11) with
w ws 1, where ws is also defined by (4.18). Summing all these residues we get

ws 1
sin(2 ks u) sin(2 ks u ) w sws 2 /k

3 (, )
NS,d
.
Zu,u = 2
y q
(1)w sin(2 ks )
( )3 s> w=w s<
s
2
2
(4.19)
ws >
17 Note that in the out boundary state we take the opposite U (1) charge, as follows from CPT conjugation
[57].
547
Noting that |yq s | = e22 (s2 ) is smaller (bigger) than 1 for s > 2 (s < 2 ), we can sum
on w for every s to get

NS,d
Zu,u
= 2
(1)ws
sZ+1/2
sin(2 ks u) sin(2 ks u ) y ws q sws s

sin(2 ks )
1 + yq s
2 /k
3 (, )
.
( )3
(4.20)
This result can be recast in a more transparent fashion as follows. Calling s = r + 1/2
with r Z, a character of the N = 2 NS discrete representations corresponding to a pure
winding mode J03 = j + r = kw/2 is given by (see (A.31))
2
chNS
d (j, r; , ) =
y w q sws /k 3 (, )
.
1 + yq s
( )3
(4.21)
Now, for every r Z there is only one value of w, such that j = r + kw/2 lies in the
improved unitary bound (2.7). This is exactly the value of w fixed by the condition (4.18).
Moreover, all possible representations corresponding to pure winding states and such that
j lies inside the unitary bound are covered by taking all r Z and fixing ws as in (4.18).
Calling jr the spin of the representation associated to each r, then (4.20) is equal to
NS,d
Zu,u
=

2 sin( k (2jr 1)u) sin( k (2jr 1)u )
(1)wr (uu )
sin( k (2jr 1))
rZ
chNS
d (jr , r; , ),
(4.22)
where ws has been renamed wr .

As shown in Section 4.1, the product u (j + 1, w)u (j, w) has a single pole for
every discrete, pure-winding state. It is natural to expect the discrete part of the annulus
amplitude (in the closed string channel) to be given by the residue of this pole. This is
indeed the case, and one can check that (4.22) is equal to

NS
NS,d
Zu,u
(4.23)
Res uNS (jr + 1, wr )uNS
(jr , wr ) chd (jr , r; , ),
= 2
rZ
where the residue is computed when considering the bracketed expression as an analytic
function of j . More details about this last step are given in Section 7 that treats D2-branes.
We have thus verified the Cardy consistency condition for the D0-branes in the NS/NS
sector.
and NS/R

The computation for the R/NS
cases is basically the same, mutatis mutandi.
We state here the results and provide the details of the computation in Appendix C for the
interested reader. For the open string sector we take the partition functions

R
Zu,1
(4.24)
(, ) =
chR
f (u, r; , ),
rZ+1/2

NS
(, ) =
chNS
Zu,1
f (u, r; , ),
rZ
(4.25)
548

R/NS
and Zu,u (, ) is given by a sum as in (4.6). The modular transforms are

c 2
1
R
ei 3 Zu,u
,

+

wk
1
1

P
,
iP , w uNS
+ iP , w chNS
;
,
=
dP uNS

c
2
2
2
wZ 0

NS

(4.26)
+ 2
(jr , wr ) chd (jr , r; , ),
rZ

1

NS
Zu,u
,

+

wk
R 1
R 1
R
iP , w u
+ iP , w chc P ,
; ,
dP u
=
2
2
2
ei
c 2
3
wZ 0
+ 2

Res uR (jr + 1, wr )uR (jr , wr ) chR
d (jr , r; , ),
(4.27)
rZ+1/2
where jr and wr are defined in the same way as for the NS/NS case, but notice that in the R
case r is half-integer. In both (4.26) and (4.27) the residues are again computed considering
the bracketed expression as a function of j .
Some comments are in order.
Although the Cardy condition holds for arbitrary boundary conditions u, u , there are
reasons to argue that only the u = u = 1 case corresponds to physical D-branes.
Firstly, for u, u = 1 the open string partition function is built from non-unitary N = 2
representations. An alternative argument was given in [43], based on comparing higher
values of u to expectations for the physics of coinciding single D0-branes.
From the case of the D0-branes we draw an important lesson, which will remain valid
for the D1- and D2-branes. Although the open string partition function is built out of
N = 2 characters, the product of the one point functions in the closed string channel
is the same as that of the bosonic 2D black hole studied in [43] at level k + 2. Using identities developed in [56] (see also [58]) to relate sums of characters of N = 2
representations to characters of a bosonic SL(2, R)/U (1) model, one can expand the
partition function (4.4) of a string stretching between a u-brane and a basic brane as
NS
(, ) =
Zu,1
+r)
k/2 (j +r)
1 (jk/2
+n k+2
( k/2 +n)2
z
q
( )
n,rZ

j,rn ( ) j +1,rnu ( ) ,
(4.28)
where j = (u 1)/2 and

2
j,r ( ) = ( )
(j 12 )2 (j +r)2
+ k+2
k

1
(1)s q 2 s(s+2r+1) ,
s=0
(4.29)
549
are the characters of the bosonic coset SL(2, R)/U (1) descending from bosonic
SL(2, R) primaries with J03 = j + r. On the other hand, the open string partition function of a bosonic open string stretching between similar D-branes in the bosonic cigar
background is given by [43]

bosonic
Zu,1
(4.30)
j,r ( ) j +1,r ( ) .
(, ) =
rZ
Notice that the partition functions (4.28) and (4.30) differ by characters of a U (1)
boson. This is the U (1) R-current of N = 2, who is the responsible for the extension
of the bosonic SL(2, R)/U (1) algebra into N = 2 [23]. It is a free boson, and it is
coupled in a way that is of mild consequence to the modular matrix and to the onepoint functions.
5. D0-branes in a Zp orbifold
Before proceeding to the D1-branes, we will discuss a natural and interesting generalization of the D0-branes considered in the previous section.
In the annulus partition function (4.4) we summed over the whole infinite spectral flow
orbit, and this was essential in order to obtain a discrete spectrum of U (1) charges in
the closed string channel. But a consistent result is also obtained if the sum is taken with
jumps of p units of spectral flow. As we will see, this correspond to D0-branes living in a
Zp orbifold of the original background, and in this framework we will be able to connect
with previous works on boundary N = 2 Liouville theory [1012].
The orbifold acts freely as a shift of 2/p on the angular direction of the cigar. As for
a compact free boson, we expect that the orbifold theory is equivalent to the original one
with the radius divided by p. This implies a spectrum of charges given by
kw
J03 + J03 = m + m
=
,
p
= pn.
J03 J03 = m m
(5.1)
Moreover, since the (infinite) volume of the target space decreases

now by a factor 1/p,
we expect that the one point functions should be renormalized by 1/p. For k integer one
can mod out the theory by Zk (see below), and we get the spectrum of single cover of the
vector coset SL(2, R)/U (1) [25], which has the target space geometry of the trumpet, at
first order in 1/k.
By applying the same method of descent from H3+ we can compute the one-point function for this orbifold. For simplicity, we will study here the NS/NS case, with u = u = 1,
and we omit the NS/u/u labels. The other sectors are similar.
We obtain the following one-point function for the D0-branes:
kw
kw
1 2ia wp 1 j (j + 2p )(j 2p )
,
2
p,a (j, w) = e
pk
(2j 1)(1 12j )
(5.2)
where a Zp is an additional quantum number of these D-branes and the phase e

is fixed by the Cardy condition.
2ia w
p
550
Let us start with the open string partition function for an open string stretching between
two such D0-branes, for which we consider
Zp;a,a (, )

=
chf (1, r; , )
rZp+aa
3 (, )
=
( )3
sZp+aa +1/2
s 2 s
k
2s1
y k
1 + yq s

sZp1+aa +1/2
s 2 +s
k
2s+1
y k
1 + yq s

.
(5.3)
We have summed over all the representations of the spectral flow with p-jumps, with the
starting point r = a a coming from the additional parameters of these D0-branes.
In order to check the Cardy condition, we will perform a modular transformation of
Zp;a,a to the closed string channel. The computation is very similar to that of the previous
section, so we will indicate only the major steps. We start with

c 2
1
ei 3 Zp;a,a ,

Z 2i 2

(i) eZ+2 k + k Z
3 (, ) 1
dZ
=
+
p
cosh(Z)
( )3 2i
C
C+
) 1
i( i Z1/2(aa
+2)
p

)
2 cos ( i Z1/2(aa
+ 12 )
p

Z 2i 2

(i) eZ2 k + k Z
1
dZ
+
2i
p
cosh(Z)
C
C+
) 1
i( i Z+1/2(aa
+2)
p

)
2 cos ( i Z+1/2(aa
+ 12 )
p
(5.4)
The contour of integration is the same as in the previous section, but now the last factor
of the integrand has poles at values i Z = pZ + a a + 1/2 in the first term, and at
i Z = pZ + a a 12 for the second. Expanding the last factor of (5.4) as in (4.9),
(4.10), and taking 0 in the contours C , yields,

2
1
i 3c
Zp;a,a ,
e

iw
sinh(( 2Z
wZ Z 2
3 (, )
k p )) Z w
2i(aa ) w
p
p q i p + k .
e
=
(5.5)
dZ
e
y
p cosh(Z)
( )3
wZ C
In order to express (5.5) as a sum over characters we shift the contour in each term by
an amount of iwk/(2p) i2 . As in the previous section, we pick poles for Z = is, with
551
s Z + 1/2, corresponding to the discrete representations. Defining ws Z as

2ps
> ws 1,
(5.6)
k
the poles with s > 2 (s < 2 ) are picked with positive (negative) sign, by all the values
w ws (w < ws ) in (5.5). The sum of all the poles is

ws 1
w sw s 2
3 (, )
s w
2i(aa ) w
p sin 2
2
ypq p k
3
k
p
( )
s>2 w=ws
s<2

3 (, )
s
(ws + n)
2i(aa ) wsp+n
= 2
e
sin 2
k
p
( )3
ws >
sZ+1/2 nZp
ws +n
p
s(ws +n) s 2
k
p
q
1 + yq s
(5.7)
Collecting the continuous and discrete contributions, we get finally

2
1
i 3c
e
Zp;a,a ,

+

1
1
kw
NS
iP , w p,a
+ iP , w chc P ,
; ,
=
dP p,a
2
2
2p
wZ 0

+ 2
Res p,a (jr,n + 1, ws n)p,a (jr,n , ws + n)
rZ nZp
chNS
d (jr,n , r; , ),
(5.8)
where as usual the residue is computed when considering the bracketed expression as a
function of j . The spins jr,n of the discrete representations are defined through
jr,n + r =
k(ws + n)
,
2p
(5.9)
where s = r +1/2, and jr,n satisfies the unitary bound (2.7). Moreover, from (5.6) it follows
that the width k/2 of the unitary bound (2.7) is sliced into p intervals, and for each n Zp
we have
k(n + 1) 1
1 kn
+
< jr,n <
+ .
2 2p
2p
2
(5.10)
Finally, note that the choice of a in the open string channel, which is a statement about the
spectrum, becomes a phase in the closed string channel.
5.1. The Zk and Z cases
In this context, we can now construe previous studies of D-branes in N = 2 Liouville
theory as particular cases of the orbifold discussed here.
552
In [10] D-branes were built in a N = 2 model with rational central charge c = 3 +

6K/N , with K, N positive integers, and the D0-brane open string partition function was
taken with jumps of N spectral flow units (for u = u = 1). This corresponds to level k =
N/Kand p = N in our case. In particular, for K = 1, the asymptotic radius shrinks from
R = k to R/k = /R, which is the T-dual radius.18 Indeed, notice that for p = k, the
spectrum of pure winding modes in (5.1) becomes pure momentum in the T-dual picture.
Note that we strengthen some of the results in [10], since our construction is explicitly
based on branes which have been checked to satisfy factorization constraints in the parent
theory [48]. Moreover, our results our based on a systematic analysis of a solution to the
branes preserving the full chiral algebra in the parent theory, and do not depend on the
central charge being rational. Apart from these bonuses, we also clarified the connection
to the semi-classical geometrical (mirror) cigar-picture.
Another interesting case to consider is the limit p . This corresponds to the cigar
radius shrinking to zero, so that the spectrum (5.1) of m for pure winding states becomes
continuous, thus yielding a continuum of R-charges in the closed string channel. As shown
in [25] it corresponds to the vector gauging of the universal cover of SL(2, R).
At the level of the Cardy computation, naively taking the limit p in (5.8) gives
zero in the closed string channel, since p,a (j, w) goes to zero, see (5.2). The correct way
of proceeding is to notice that Eq. (5.5) becomes a Riemann sum, which leads to an integral
over t = w/p R. We can moreover express t = x + g, g Z, x [0, 1). The sum over w
in (5.5) becomes
1
p w
+
1
dt dx
.
(5.11)
gZ
One can show that the resulting expression for the closed string channel is equivalent to
starting with
Z1,1 (x; , ) =
e2ixr chNS
f (1, r; , ),
(5.12)
rZ
and computing
1
dx e
2ixd i 3c

Z1,1

1
x, ,
,

(5.13)
where d Z depends on the precise way the limit is taken in (5.11). The computation of
(5.13) itself can be performed with an identity proved in [55,56], to which we refer for
details. The resulting expression can be found in [12].
18 In the case of rational k there will be additional discrete terms in the closed string channel partition function,
coming from poles falling exactly on the displaced contour of Section 4. But taking them into account does not
change the orbifold picture.
553
6. D1-branes
In this section, we check the relative Cardy condition for D1-branes in the N = 2 Liouville/supersymmetric coset conformal field theory. In this and the next sections, we will
work in the NS sector. The other sectors are obtained straightforwardly. Following the
general logic, we assume that the one-point functions for closed string primaries in the
presence of a D1-brane labeled by the parameters (r, 0 ) is given by (see Appendix B):
D1
r,0 (j, n)
j
nw (z, z ) r, = w,0
0
|z z |j,n
(6.1)
with

r,0 (j, n) = N1 ein0 er(2j +1) + (1)n er(2j +1) 1/2j
(2j + 1)(1 + 12j

k )
.
n
(j + 1 + 2 )(j + 1 n2 )
(6.2)
We put a normalization factor N1 up front that will be fixed during the Cardy computation.
Note that the only poles in the one-point function originate from infinite volume divergences (i.e., they are bulk poles). Thus the only contributions to the closed string channel
amplitude between two D1-branes originate from continuous representations (unlike the
D0- and D2-brane computations where poles associated to discrete representations require
special attention). The closed string channel amplitude will only contain contributions proportional to the continuous character:

1
P 2 n2
n 3 ( , )
n
1

k + 4k y
k
chNS
P
,
;
,
=
q
c
2

3 ( 1 )
(6.3)
We compute the closed string channel amplitude between two branes with identical boundary conditions both labeled by (r, 0 ), for simplicity. The computation can then easily be
generalized to include the case of differing boundary conditions, following [59]. For our
case, we find the partition function:

D1
Zr,
0
dP

nZ
4N12
=
k

dP
0

1
n
r,0 (P , n)r,0 (P , n)chc P , ; ,
2

n2Z+1

1
sinh 2P sinh 2P
k
2
sin 2rP sinh
P chNS
c
n2Z

n
P
,
cos2 2rP cosh2 P chNS
c
2
n
P,
2

.
(6.4)
We modular transform the characters to obtain the annulus amplitude suitable for interpretation in the open string channel:
554
ci 2
3
D1
Zr,
0

2

8N1
4P P
=
dP dP
cos
k
k
Z
0
w 2
cos2 2rP
cosh P + (1)2w sin2 2rP sinh2 P

2
sinh 2P sinh
2P
k

chNS
c (P , kw; , ).
(6.5)
Since the transformation properties of continuous N = 2 characters are analogous to the

transformation properties of purely bosonic coset characters, we can discuss the results
similarly as for the bosonic coset [43]. To make the discussion more explicit, and in particular to match on to the boundary reflection amplitude, we follow [43,48] to check the
relative Cardy condition. We compare the regularized density of states obtained in the open
string channel with the one we expect from a reflection amplitude which is the natural
N = 2 generalization of the reflection amplitude in [43,48,59] (which satisfies factorization). To this end, we subtract a reference amplitude labeled by r , and use trigonometric
identities and the change of variables t = 2P to obtain:
D1

Zr,0 ZrD1
,
0

8N12
dt k cosh2 2t (1)2w sinh2
NS

=
dP chc (P , kw)
k
P
t 16
sinh t sinh kt
0
0

2t
rk
rk
2t

.
sin
P +
+ sin
P
r r
k
ci 2
3
t
2
(6.6)
To link the relative density of continuous states in the open string channel to the boundary
reflection amplitude it is convenient to define the special functions:
(0)
log Sk (x) = i
dt
t
0
(1)
log Sk (x) = i
dt
t

x
,
t
2 sinh kt sinh t
sin 2tx
k
cosh t sin 2tx

k
2 sinh kt sinh t

x
.
t
(6.7)
We can then write our ansatz for the boundary reflection amplitudes as:
(0)

R P , w Z +
k2 ( 12 iP + k)k (2iP + k)Sk (P +
rk
)
,
(0)
1
2
k ( 2 + iP + k)k (2iP + k)Sk (P + rk
)

(1)
1
rk
2
1
iP k ( 2 iP + k)k (2iP + k)Sk (P + )
r
=
.
k
(1)
2
k2 ( 12 + iP + k)k (2iP + k)Sk (P + rk
)
R(P , w Z|r) = iP
(6.8)
For the definition of the generalized gamma-functions, we refer to, e.g., [43]they immediately drop out of the computation of the relative partition function. Using the reflection
amplitudes (and the fact that they are parity odd) we can show that the relative Cardy
555
condition holds:
e
ci 2
3
D1

Zr,0 ZrD1
,
0
N2
= 1
i
dP

wZ
R(P , w|r)

chNS
log
c (P , kw)
P
R(P , w|r )

, w|r) NS
R(P
chc (P , kw) .
+
log
, w|r )
P
R(P
wZ+1/2
(6.9)
To obtain agreement with the density of states expected on the basis of the boundary reflection amplitude, we can fix N12 = 1/2. Note that in the course of our computation, we
have made use of the simple fact that the standard definition of mass in the open string
channel differs by a factor 4 from the mass in the closed string channel which leads us
to interpret the open strings as having half-integer or integer winding number which is
natural in terms of the D1-brane semi-classical geometry [43]. We have thus verified that
the relative Cardy condition is satisfied by our D1-branes. In summary, in this section we
have argued that the relative Cardy condition holds, given the one-point functions for the
D1-branes we started out with, and the extension of the boundary reflection amplitudes of
[48,59] to N = 2 theories. The computation follows the lines of [43] due to the close connection between (continuous) N = 2 characters and those of the bosonic coset, and their
modular properties (see comment at the end of Section 4). Note that the D1-branes couple
to continuous bulk states with zero winding only.
7. D2-branes
In this section, we analyze the Cardy condition for the D2-branes which we can construct from the H2 branes in Euclidean AdS3 . They correspond to type B branes w.r.t. the
N = 2 superconformal algebra. We find some puzzling features when trying to perform
the Cardy check. Following [43] and the general logic outlined before, we propose the
following one-point function for the D2-branes parameterized by , in the NS sector:
D2
(j, w)
j
nw (z, z ) = n,0
,
|z z |j,w
1
kw
2 )(j
kw
2 )
12j
(2j 1)(1 k )
i
(12j
) sin (j kw ) + ei (12j ) sin (j
e
2
sin (1 2j ) sin 12j
k
(j, w) = N2 2 j
(j +
with
kw
2 )
(7.1)
This one point function has poles corresponding to the discrete representations, and therefore will couple both to localized and asymptotic states. This is expected on general
grounds since these D2-branes carry D0-brane charge.
556
The annulus partition function in the closed string channel, for general Casimir labeled
by j , in the NS sector, becomes:

D2 1
Z
,

chNS (j, kw ; 1/, / )
2
= kN22 dj
(7.2a)
)
sin
(1
2j ) sin (12j
wZ
k

2 cos( + )(1 2j ) 2 cos( )(1 2j ) cos 2j
(7.2b)
2 cos( )(1 2j ) sin2 2j 2i sin( )(1 2j ) sin 2j sin kw
.
+
cos kw cos 2j
cos kw cos 2j
(7.2c)
We can read from this expression that the different terms will contribute in a very different
fashion.
7.1. D1-like contribution
The two terms of the third line (7.2b) will give a contribution similar to the D1-branes
(which is an expected contribution, on the basis of the fact that the D2-branes also stretch
along the radial direction), with imaginary parameter though. Explicitely, we have in the
closed channel:

kw
chNS
D2,(b) 1
c (P , 2 ; 1/, / )
2
dP
,
= 2kN2
Z

sinh 2P sinh 2P /k
wZ
0

cosh 2P ( + ) + cosh 2P ( ) cosh 2P .
(7.3)
As for the D1-branes, we consider the relative partition function w.r.t. the annulus amplitude for reference branes of parameters (0 , 0 ). Going through the same steps as in the
previous sections, we obtain the open string channel amplitude:

R(P |i +
D2,(b)
2

2 )R(P |i 2 )
Z (, ) = 4kN2
log
dP

+
2iP
|i 0 0 )
R(P |i 0 2 0 )R(P
2

(7.4)
chNS
c (P , n; , )
nZ
in terms of reflections amplitudes similar as before, see (6.8), but with imaginary parame1
ters i( )/2. We fix the normalization constant to N22 = 4k
.
7.2. D0-like contribution
Now we concentrate on the last two terms (7.2c) of the annulus amplitude. As we will
see this will give a contribution similar to those of D0-branes, hence with both discrete and
continuous contributions. Let us first concentrate on the latter. Since the last term is odd in
w, it will cancel from the amplitude and we are left with:

1
D2,(c)
,
Z ,cont

kw
cosh 2P ( ) sinh2 2P chNS
1
c (P , 2 ; 1/, / )
=
dP
.
2
(cosh 2P + cos kw) sinh 2P sinh 2P /k
0
557
(7.5)
wZ
Assuming that = 2m/k, m Z, we recognize a D0 amplitude for two branes of

same parameter m:

D2,(c)
Z ,cont
1
,

(2 sinh2 (2P m/k) + 1) sinh 2P

(cosh 2P + cos kw) sinh 2P /k
wZ
0

kw 1
P
,
;
,
,
chNS
c
2
1
=
2
dP
(7.6)
up to the constant term in the bracketed expression, that will drop from the relative partition function. The normalization 1/2 of this expression has to be compared with the
normalization ()w(mm) = 1 of the D0 computation (4.16).
7.2.1. Discrete representations
We can also make the identification with a D0-like contribution as follows. First we
consider the more straightforward case w > 0. Then we pick the poles of the discrete
representations in the domain: j D = (kw/2 N) ]1/2; (k + 1)/2[. For each pole we
will have a contribution of 2 times the residue:
2j 1
sin (2j 1) i sin 2m 2jk1 sin kw
kw cos 2m k
1
()j 2
2j 1
2
sin (j + kw
j D w>0
2 ) sin k

kw
j,
j
.
chNS
d
2
(7.7)
We write j = kw/2 r, with r N. Then for every r, there is only one value of w, that we
shall call wr such that j is in the correct range. Explicitely wr is given by: wr = 2r+1
k +1.
We call also jr the value of j that has been picked. With this procedure we get:
2r+1
cos 2m 2r+1
1
k i sin 2m k
()wr
chNS
d (jr , r).
2
sin 2r+1
k
(7.8)
rN
Let us now consider the case w 0. In this case we have to use the isomorphism of reprek

sentations: Dj+,w Dj,w1
, with j = k+2

2 j . We have then j = 2 (w 1)r , r N,

and the value of w is fixed to: wr = 2r k+1 . This leads to the following contribution to
the annulus amplitude:

cos 2m 2r k+1 + i sin 2m 2r k+1 NS
1

()wr 1
chd jr , r .
+1
2r
2
sin k
r N
(7.9)
558
Then it is possible to add the two contributions, which cancels the imaginary part, and
leaves us with:

1
D2,(c)
Z ,disc ,

r+1/2
2
1 NS
2r+1 2 sin 2m
1
1
k
j
,
.
() k
ch
,
r;
=
(7.10)
r
d
2

sin 2 r+1/2
rZ
This is again 1/2 of the expression of the amplitude for two D0s of parameter m,
Eq. (4.22), up to the irrelevant constant term.
7.2.2. Some comments on D2-branes physics
The following comments are in order:
In the computation we assumed that the difference of the parameters of the D2 in the
annulus amplitude is quantized: = 2m/k, m Z. This relative quantization
condition is discussed in [43] for the bosonic coset. Indeed the difference of the D2branes parameters is the net induced D0-charge, hence it should be quantized. To be
more precise it is believed that a D2-brane with parameter is a bound state of a
brane of parameter with m D0-branes (for the > case, otherwise one has to
reverse the picture).
As a corollary, the one-point functions for the D2-branes have poles both of the localized type and of the bulk type.
The computation of the annulus amplitude gives a continuous spectrum of open strings
attached to the D2-branes, with a sensible density of states, and also a contribution
similar to the D0see the previous remark about the induced D0 chargebut with a
normalization (1/2) which complicates the task of making sense of the open string
spectrum, and blurs the physical picture of bound states. The only open string spectrum
leading to a good physical picture would seem to correspond to two D2-branes with
the same parameter .
Clearly further study of the physics of D2-branes is needed to clarify their interpretation.
8. Conclusions
We constructed D-branes in N = 2 Liouville theory and the SL(2, R)/U (1) supercoset
conformal field theory, and checked the (relative) Cardy condition as well as consistency
with the proposed boundary reflection amplitudes, which were proven earlier to satisfy the
factorization constraints. The one-point functions that we constructed remarkably decouple
the poles in bulk amplitudes (notably the reflection amplitude) into poles associated to infinite volume and the ones associated to normalizable discrete states. It would be interesting
to investigate whether the boundary decoupling phenomenon aids in understanding more
aspects of the conjectured holographic duality in the (doubly scaled) little string theory.
559
We have shown (for the particular but extendable) case of the D0-branes how to generalize the results to other sectors of the supersymmetric Hilbert space, and to orbifolds of
the cigar conformal field theory.
We have pointed out throughout this work some similarities of the one-point functions
for N = 2 Liouville with those of the N = 0, 1 cases, like the decoupling of the bulk poles.
It would be interesting to see also whether the interesting relations uncovered in [60,61]
between the boundary states associated to localized and extended branes have also any
manifestation in the N = 2 theory.
In appendices, we argued for the general use of the SL(2, R) symmetry that can be
obtained by enhancing N = 2 theories with an orthogonal free scalar, and we suggested
that this tool provides some technical support for an attempt to interpret the variables
(x, x)parameterizing
the SL(2, R) quantum numbers in position space on the boundary

of AdS3 as new worldsheet variables. We showed in great detail the fact that supercoset
characters agree with N = 2 characters, and the N = 2 spectral flow has a natural interpretation in terms of SL(2, R) quantum numbers.
Note that we have constructed D-branes for generic values of the level k. It is known that
the conformal field theory correlators depend strongly on whether k is rational or irrational
(for instance, via the shift equations, or the structure of the poles in the bulk correlators). It
seems important to further clarify the relation between the construction of branes at rational
and irrational values of k (for instance, by further comparing the techniques developed in
[56] to the results of [43]), as well as the dependence of bulk correlators on this most
intriguing distinguishing feature.
The branes we constructed form an integral part of the construction of D-branes in
non-compact non-trivially curved supersymmetric string theory backgrounds. One particular application amongst these is the construction of D1- and D3-branes in the little string
theory background in the double scaling limit (see also, e.g., [6264]). Indeed, when we
consider the doubly scaled limit for NS5-branes which are evenly distributed over a topologically trivial circle, we obtain a closed string background which is supposed to represent
a holographic dual of the Higgs phase of the little string theory living on the NS5-branes
[5], whose partition function has been studied in [8,26]. The role of the W -bosons in the
LST is played by the D1-branes stretching between the NS5-branes. One can construct at
least the one-point functions corresponding to these D-branes by combining the D0-brane
in the SL(2, R)/U (1) theory and a D1-brane in the supersymmetric SU (2)/U (1) coset
theory, properly taking into account the discrete orbifold operation. The construction of
these one-point functions is now straightforward. We will return to some of these issues in
the near future [65].
Acknowledgements
We thank Shmuel Elitzur, Angelos Fotopoulos, Stefan Fredenhagen, Gastn Giribet,
Amit Giveon, Anton Kapustin, Elias Kiritsis, Costas Kounnas, David Kutasov, Marios
Petropoulos, Boris Pioline, Sylvain Ribault, Adam Schwimmer and especially Volker
Schomerus for comments and discussions. A.P. thanks the Laboratoire de Physique
Thorique de lcole Normale Suprieure for hospitality during the first part of this work,
560
and the organizers of the ESI Workshop String Theory in Curved Backgrounds and
Boundary Conformal Field Theory. A.P. is supported by the Horowitz Foundation and
by an ESI Junior Fellowhip. This work is supported in part by the Israeli Science Foundation and by the EEC under the contracts HPRN-CT-2000-00122, HPRN-CT-2000-00131.
Appendix A. Computing N = 2, c > 3 characters
In this appendix we will compute the N = 2, c > 3 characters appearing in the spectrum
of the D-branes studied in the paper. We will obtain them as characters of the supersymmetric coset SL(2, R)/U (1), through a rather standard procedure. The same characters can
be computed by subtracting the null-vectors modules from the free action of the N = 2
generators, as has been done in [6668]. The N = 2 representations that we consider are
both unitary and non-unitary. The characters of the former have been computed in those
papers and coincide with our results, while for non-unitary representations we present the
characters for the first time. The fact that the coset yields irreducible N = 2 characters
can be traced back to the fact that the SL(2, R) characters we start with correspond to
irreducible representations of the SL(2, R) algebra.
Notice that we have here a non-minimal version of a similar situation for (N = 1, 2
supersymmetric) unitary minimal models, where the (supersymmetric) Virasoro characters
coincide with the characters of certain cosets involving SU (2) factors which realize the
minimal models [69].
Moreover, this coincidence of the two ways of computing the characters, which was
independently noticed in [26] and [8], is central to the fact that the D-branes that we build
in this paper belong to the class of objects, such as the correlation functions [4,32,35] or the
modular-invariant partition function [8,26], for which there is no distinction as to whether
we are in the N = 2 Liouville or in the supersymmetric cigar.
Let us review first how the N = 2 algebra arises in the susy SL(2, R)/U (1) coset [3].
The supersymmetric SL(2, R) model at level k has currents J a , b (a, b = 1, 2, 3), with
OPEs
g ab k/2
f ab c J c (w)
,
+
zw
(z w)2
f ab c c (w)
g ab
,
a (z) b (w)
,
J a (z) b (w)
zw
zw
J a (z)J b (w)
(A.1)
where g ab = diag(+, +, ), f 123 = 1 and indices in the antisymmetric f abc are raised and
lowered with g ab .
We first define
i
(A.2)
Ja = f a bc b c .
2
The currents j a commute with the three fermions and generate a bosonic SL(2, R) model
at level k + 2. The currents Ja , a form a supersymmetric SL(2, R) model at level 2.
The Hilbert space of the original supersymmetric SL(2, R)k theory is the direct product of
the Hilbert space of the bosonic SL(2, R)k+2 and that of the three free fermions.
j a = J a Ja ,
561
We are interested in the coset obtained by gauging the U (1) symmetry generated by
J 3 , 3 . This coset has an N = 2 algebra generated by

2
2
k + 2 3 2 3
j ,
JR = j3 +
G =
J = J + + ,
k
k
k
k
T = TSL(2,R) TU (1) ,
(A.3)

where 2 = 1 i 2 and
1
1
TU (1) = J 3 J 3 + 3 3 .
k
2
(A.4)
The currents (A.3) commute with J 3 , 3 and satisfy the N = 2 superconformal algebra
J R (z)J R (w)
c/3
,
(z w)2
G+ (z)G (w)
G (w)
,
(z w)

2T (w) + J R (w)
J R (z)G (w)
2c/3
2J R (w)
1
+
+
3
(z w)
(z w)2 (z w)
(A.5)
with the central charge

6
c=3+ .
(A.6)
k
A highest-weight representation of the N = 2 algebra with a given central charge c is
determined by the conformal dimension h and the U (1) charge Q of the highest weight
state. By building the N = 2 representations through the SL(2, R)/U (1) coset, we can
parameterize h, Q by means of SL(2, R) quantum numbers as follows from the zero modes
of J 3 , T in (A.3).
The states of the coset are all those states in the parent SL(2, R) theory annihilated by
3 , 3 , and a primary state of the coset is a coset state which is also a
the modes Jn>0
n>0
primary of the N = 2 algebra. Every primary state of the parent theory is clearly a primary
of the coset. In addition, for the discrete and finite-dimensional representations of j a , there
are also descendent states of the parent theory which are primaries of the coset.
The Hilbert space of the parent theory can be decomposed into subspaces with definite
J03 eigenvalue m. Since the N = 2 algebra commutes with J03 , from a given SL(2, R)
representation with spin j , we will obtain a different N = 2 representation for each value
of m. The use of j, m labels is most convenient, because performing integer spectral flow
in N = 2 just amounts to an integer shift in m, as we will see below.
For a given SL(2, R) representation, we are interested in the characters
24
chNS
TrNS q L0 y J0 ,
j,m (q, y) = q
c
(A.7)
24
chNS
TrNS (1)F q L0 y J0 ,
j,m (q, y) = q
c
24
TrR q L0 y J0 ,
chR
j,m (q, y) = q
R
(A.8)
(A.9)
where the trace is taken on the Hilbert space of the coset. We will compute in all the cases
the NS character first. The Ramond characters will be obtained by half-spectral flow [70]
562
as

6 1 2 c 1
chR (, ) = q c ( 2 ) y 3 2 chNS , +
,
2
(A.10)
characters, it is easy to see that they are given

where q = ei2 , y = ei2 . As for the NS
by

NS
chNS
j,m (q, y) = chj,m (q, y).
(A.11)
The NS characters can be obtained from

j (q, x, y) = Tr q L0
c+3/2
24
x J0 y J0 =
x m m (q)chNS
j,m (q, y),
(A.12)
where
m (q) = q 3/48m
2 /k

(1 + q n1/2 )
.
(1 q n )
(A.13)
n=1
The trace in (A.12) is taken on the whole NS Hilbert space of the supersymmetric SL(2, R)k
model and J0R is well-defined even before going to the coset. We have expanded j (q, x, y)
into terms with definite J03 charge m and factored in each such term a supersymmetric U (1)
character m (q) with highest weight = m2 /k, generated by the modes of J 3 , 3 .
The computation of j (q, x, y) itself goes as follows. In the factorized SL(2, R)k+2
{ a } theory we can easily compute
j (q, z, w) = q
c+3/2
24
Tr q L0 zj0 Tr q L0 w J0 ,
(A.14)
and then from (A.2) and (A.3) it follows that j (q, x, y) is given from j (q, z, w) by the
replacements
z xy 2/k ,
w xy (k+2)/k .
(A.15)
In (A.14), the trace over the fermions NS Hilbert space is

3
Tr q L0 w J0 =
=

1 + q n1/2 1 + wq n1/2 1 + w 1 q n1/2
n=1

n=1
(1 + q n1/2 ) p p2 /2
w q
.
(1 q n )
(A.16)
pZ
For the first factor in (A.14), we should consider the different possible representations of
the j a algebra. We take q = ei2 , y = ei2 .
A.1. Continuous representations
a
The j a representations are built by acting with jn<0
on representations of the zero
a
modes j0 . For the continuous representations we have j = 1/2 + iP , P R+ . The
SL(2, R) primaries are |P , m
, with m = r + , r Z, [0, 1) and conformal weight
563
j (j 1) 1/4 + P 2
=
.
(A.17)
k
k
These primaries of j a are multiplied by the fermionic NS vacuum to get the full primaries,
so j03 = J03 = m on these states. Considered as primaries of N = 2, they give rise to unitary
N = 2 representations [23] with (see (A.3))
1/4 + P 2 + m2
2m
,
Qm =
.
k
k
The character of the susy SL(2, R) representation is
hP ,m =
j (q, z, w) = q
c+3/2
24
(A.18)
Tr q L0 zj0 Tr q L0 w J0
c+3/2 1/4+P 2
p 2 (1 + q n1/2 )
= q 24 q k
z+r w p q 2
,
(1 q n )4
(A.19)
n=1
r,pZ
and applying the steps described above it is immediate to obtain the N = 2 characters
L0 c/24 J0
y =q
chNS
c (P , m; , ) = Tr q
R
P 2 +m2
k
2m
k
3 (, )
,
( )3
4 (, )
,
( )3
P 2 +m 2 2m 2 (, )

k
chR
y k
,
c (P , m ; , ) = q
( )3

chNS
c (P , m; , ) = q
P 2 +m2
k
2m
k
(A.20)
(A.21)
(A.22)
where m = m + 1/2 when the Ramond character is obtained by half-spectral flow from
(A.20).
A.2. Discrete representations
Let us consider discrete lowest-weight representations Dj+ of j a , with primaries |j, r
,
with m = j + r and r Z, r > 0, and with j R, j > 0. For 0 < j < (k + 2)/2 they
give rise to N = 2 unitary representations [23,71,72]. This bound is further constrained in
physical settings to be 1/2 < j < (k + 1)/2. For each value of m = j + r, r Z the N = 2
primaries are
r 0,
|j, j + r
,
r1
1/2 |j, j
.
r < 0, j1
The quantum numbers are:

r 0,
j (j 1) + (j + r)2
2(j + r)
,
Qj,r =
,
k
k
j (j 1) + (j + r)2
1
=
r
k
2
k+2
k+2
2
( k+2
j
)(
1)
+
(
2
2
2 j r 1)
,
=
k
hj,r =
r < 0, hj,r
(A.23)
564
2(j + r)
1
k
2( k+2
2 j r 1)
=
.
k
Qj,r =
(A.24)
The second and fourth lines in the r < 0 case shows that these states are similar to the r 0
. Note that r = 0, 1 correspond to chiral and antichiral

states when built from D(k+2)/2j
primaries respectively, as follows from hj,0 = Qj,0 /2 and hj,1 = Qj,1 /2. These states
are mapped to fermionic null states of relative charge 1 along the spectral flow orbit.
To compute the characters we start with
c
j (q, z, w) = q 24 Tr q L0 zj0 Tr q L0 w J0
3
j (j 1)
q 24 k zj
3
Tr q L0 w J0 .
n1 )(1 z1 q n )(1 q n )
(1
zq
n=1
=
(A.25)
Using19

1

zt St (q),
=
n1 z)(1 q n z1 )
n 2
n=1 (1 q
n=1 (1 q ) t=
(A.26)
where
St (q) =

1
(1)s q 2 s(s+2t+1) ,
(A.27)
s=0
the character (A.25) can be expanded into

2
j (q, z, w) = q
c
24
j (jk1)
j +t w p

1 St (q)q 2 z

1 + q n 2
.
n 4
n=1 (1 q )
n=1
(A.28)
p,tZ
After the replacement (A.15), and defining r = t + p, we get the decomposition as in

(A.12)
j (q, x, y) =
j +r
j +r (q)
(j 1/2)2 +(j +r)2

k
2(j +r)
k
( )3
rZ
Srp (q)y p q
p2
2
pZ
(A.29)
Using

pZ
Srp (q)y q
p2
2

=
p2
p 2
pZ y q
1 + yq r+1/2
19 See [73] for a proof of this identity.
3 (, )
,
1 + yq r+1/2
(A.30)
565
we get finally
c
L0 24 J0
chNS
y
d (j, r; , ) = Tr q
= q
(j 1/2)2 (j +r)2
+ k
k
= q
(j 1/2)2 (j +r)2
+ k r 12
k
2(j +r)
k
1
3 (, )
1/2+r
1 + yq
( )3
2(j +r)
1
k
1
1 + y 1 q 1/2r
3 (, )
. (A.31)
( )3
We wrote the characters in two forms, each form reflecting the structure of the representation for a different range of r (see (A.23)). For the other sectors we get

chNS
d (j, r; , ) = q

chR
d (j, r ; , ) = q
(j 1/2)2 (j +r)2
+ k
k
2(j +r)
k
(j 1/2)2 (j +r )2
+ k
k
2(j +r )
k
1
4 (, )
,
1/2+r
1 yq
( )3
(A.32)
1
2 (, )
,

1/2+r
( )3
1 + yq
(A.33)
where r = r + 1/2 when the Ramond character is obtained by half-spectral flow from
(A.31).
A.3. Finite-dimensional representations
In this case the spin takes the values j = (u 1)/2, u = 1, 2, . . . . The highest weights
are u-dimensional representations of j0a given by |j, m
, with m = j, j + 1, . . . , j . Only
for u = 1 the induced N = 2 representation is unitary. For every m = j + r, r Z the
N = 2 primaries are
r < 0,
r1
j1
1/2 |j, j
,
0 r u 1, |j, j + r
,
+ ru +
1/2 |j, j
r > u 1,
j1
(A.34)
with quantum numbers

r < 0,
hj,r =
0 r u 1, hj,r =
r > u 1,
1
j (j 1) + (r + j )2
r ,
k
2
j (j 1) + (r + j )2
,
k
Qj,r =
1
j (j 1) + (r + j )2
+r u+ ,
k
2
2(r + j )
+ 1.
=
k
Qj,r =
2(r + j )
1,
k
2(r + j )
,
k
hj,r =
Qj,r
In order to compute the characters, we start with
(A.35)
566

3
j (q, z, w) = q 24 Tr q L0 zj0 Tr q L0 w J0
c
j (j 1)
(u1)
q 24 k (z 2 z 2 )
=
n1 )(1 z1 q n )(1 q n )
n=1 (1 zq
u+1

(1 + q n 2 ) p p2 /2
w q
.
(1 q n )
n=1
(A.36)
pZ
The expansion goes along similar lines to the discrete representations and is left as an
exercise. The resulting N = 2 characters are
chNS
f (u, r; , )
c
= Tr q L0 24 zJ0
3 (, )
(1 q u )
1
1/2r
1
u1/2r
(1 + y q
)(1 + y q
) ( )3
u
2
2
(j 1/2)
(r+j )
2(r+j )
3 (, )
(1 q )
= q k + k y k
+1/2+r
1
u1/2r
(1 + yq
)(1 + y q
) ( )3
u
2
2
(j 1/2)
(r+j )
1 2(r+j )
(1 q )
3 (, )
.
= q k + k +r+2j 2 y k +1
(1 + yq 1/2+r )(1 + yq ru+1/2 ) ( )3
(A.37)
Again, we expressed the characters in forms reflecting the structure of the representation
for different ranges of r. These characters can also be expressed as
= q
(j 1/2)2 (r+j )2
+ k r1/2
k
2(r+j )
1
k
NS
NS
chNS
f (u, r; , ) = chd (j, r; , ) chd (j + 1, r u; , )

1
1
3 (, ) s 2 su 2su
k y k
q
=
1 + yq s
1 + yq su
( )3
(A.38)
with s = r + 1/2. For the other sectors we get

chNS
f (u, r; , ) = q
chR
f (u, r ; , ) = q
(j 1/2)2 (r+j )2
+ k
k
2(r+j )
k
4 (, )
(1 q u )
,
+1/2+r
1
u1/2r
(1 yq
)(1 y q
) ( )3
(j 1/2)2 (r +j )2
+ k
k
2(r +j )
k
(1 q u )

(1 + yq +1/2+r )(1 + y 1 q u1/2r )
2 (, )
,
( )3
(A.39)
where r = r + 1/2 when the Ramond character is obtained by half-spectral flow from
(A.37).
A.4. Spectral flow
The N = 2 algebra has the spectral flow automorphism [70]
c
c
Ln Ln + wJn + w 2 n,0 ,
Jn Jn + n,o ,
6
3
G
m Gmw
567
(A.40)
which maps a representation into another one. The spectrum of the flowed representations
is obtained by measuring the flowed L0 , J0 on the original representation. The character of
the spectrally flowed representation is then given by
c
ch(, ) q 6 w y 3 w ch(, + w ).
(A.41)
Using (E.5), it is immediate to verify that in the characters considered above the spectral
flow by w integer units is equivalent to a shift m m + w, thus verifying our claim that
m keeps track of the N = 2 spectral flow orbit.
Finally, notice that in the discrete and finite cases, SL(2, R) representations in which the
generators act freely give rise to an N = 2 representations with null descendents due to the
semi-infinite/finite base in SL(2, R). To get some insight into the form of the characters,
note that the free action of the modes Ln , Jn , G
n+1/2 on the highest weight state gives
the contribution (in the NS sector),
1
1

(1 + q n 2 y)(1 + q n 2 y 1 )
3 (, )
= q 1/8
.
n
2
(1 q )
( )3
(A.42)
n=1
In the discrete and finite representations the character is further modded out by the null
descendents.
Appendix B. Changing basis in SL(2, R)
In this appendix we recall how to Fourier transform the one-point functions of [48] into
the form in which they were used in [43], and in the bulk of our paper, in our conventions.
B.1. Reflection amplitude
We use a quadratic Casimir for SL(2, R) representation of the form:
c2 = j (j 1)
(B.1)
and we adopt conventions in which the bulk two-point functions are:

j 1 1 j 2 2
x , x x , x

B(j )
4j
2 1
2

(j j ) ,
= |z12 |
x x (j + j 1) +
|x12 |4j
2j 1
k 12j (1 k )
,
B(j ) =
( 2j 1 )
(B.2)
which leads after Fourier transformation

1
j
nw =
d 2 x j 1+m x j 1+m j (x, x),

4 2
(B.3)
568
to

j j
nw n w = |z12 |4j n+n w+w (j + j 1) + R(j, n, w)(j j ) ,
R(j, n, w) = 12j
12j
k )
.
+ 1 m)(1
12j
k )
(2j + 1)(j + m)(j m)(1
+
(2j 1)(j + 1 + m)(j
(B.4)
In the supercoset, we identify the elliptic eigenvalues of SL(2, R) with the geometric angular momentum and winding by:
m=
n + kw
,
2
m
=
n + kw
.
2
(B.5)
B.2. Localized branes

Starting from the one-point function for spherical branes in H3 [48]:

(1 + (2j + 1)/k) sin s(2j + 1)
1
2j
j (x|z) s = (1 + x x)
2
(1 1/k)
sin s
j +1 |z z |2j ,
(B.6)
we find its Fourier transform:

j

np (z) s = d 2 x ein arg(x) |x|2j 2ip j (x|z) s
(j ip/2)(j + ip/2)(1 + 1/k) sin /k sin s(2j + 1)
1
,
= n,0
2
(2j 1)(1 (2j + 1)/k) sin s sin (2j + 1)/k
(B.7)
where we made use of the integral:

2j 1 (j ip/2)(j + ip/2)
.
dr r 2j 1ip 1 + r 2
=
2
(2j )
(B.8)
Up to normalization, this corresponds to the one-point function in the bosonic and supersymmetric coset.
B.3. Extended AdS2 branes
Starting from the one-point function for extended branes in H3 (where = sgn(x + x)):

j

Nk
2j + 1 (12j )r
e
(x|z) r =
|z z |2j ,
|x x|
2j j +1/2 1 +
k
(B.9)
we find the transformed coset one-point function:

j
np (z) r =
569

dx 2 ein arg(x) |x|2j 2ip j (x|z) r
(2j + 1)(1 + (2j + 1)/k)

= 2k(p)N j +1/2
(j + 1 + n/2)(j + 1 n/2)

(1)n er(2j +1) + er(2j +1) .
(B.10)
We used the integrals:

/2
d ein (2 cos )2j =
/2
(2j + 1)
,
(1 j + n/2)(1 j n/2)
dr r 1ip = 2(p).
(B.11)
Thus we have reviewed the connection between the one-point functions in [48] and the
one-point functions for the coset [43] in our conventions.
sectors
Appendix C. Cardy condition for D0-branes in R/NS
In this appendix we will provide some more details of the Cardy computation for D0 sectors, along the lines of Section 4.
branes in the R and NS
Let us start Ramond sector in the open string channel. For the partition function in the
(u, 1) case we take

R
Zu,1
(, ) =
chR
f (u, r; , )
rZ+1/2
s 2 +su 2s+u
2 (, )
1 s 2 su 2su
k y k
k y k
q
q
,
1 + yq s
( )3
(C.1)
sZ
where we have summed over the whole spectral flow orbit of the N = 2 Ramond characters
associated with the u-dimensional representation of SL(2, R). For the general (u, u ) case
we make a sum as in (4.6). The computation is very similar to the NS/NS case, so we will
only indicate the major steps.
R (, ), given by
We start with the modular transform of Zu,1

2
1
i 3c
R
Zu,1 ,
e

2i 2
4 (, ) 1
dZ (i)eZ+ k Z
+
=
3
( ) 2i
C
sinh(2 Zk u)
C+
ei(i Z+ 2 )
cosh(Z) cos (i Z + 12 )
(C.2)
570
The contour of integration is the same as in Section 4, and the integrand has poles at
i Z = s Z. Expanding the last factor of the integrand as a sum over w as in (4.9),
(4.10), taking 0 and shifting the contour by iwk
2 i2 in each term, we arrive, for
general (u, u ), to a decomposition

c 2
1
R,c
R,d
R
ei 3 Zu,u
(C.3)
= Zu,u
,
+ Zu,u ,

where the first term is an integral over the shifted contour and the second is the sum of
poles picked during the shift. These terms are
R,c
Zu,u

+

=
dP
wZ 0

(1)w(u+u +1) [cosh(2P )+cos(kw)] sinh(2 Pk )
P 2 +(kw/2)2
k
4 (, )
( )3
+

wk
1
1

NS
NS
NS
dP u
=
iP , w u
+ iP , w chc P ,
; ,
2
2
2
q
2(kw/2)
k
wZ 0
(C.4)
and
2
R,d
Zu,u

2 sin( 2 (2jr 1)u) sin( 2 (2jr 1)u ) 4 (, ) y wr q swr sk

k
k
=
wr (u+u +1) sin( 2 (2j 1))
1 yq s
( )3
(1)
r
k
rZ

NS

= 2
(jr , wr ) chd (jr , r; , ),
(C.5)
rZ
where jr and wr are defined in the same way as in Section 4 and s = r + 12 as usual. We
see thus that the Cardy condition is verified.
sector, which computes the open string
For the open string partition function in the NS
Witten index [74], we start with

NS
Zu,1
(, ) =
chNS
f (u, r; , )
rZ
4 (, )
( )3

sZ+1/2
s 2 +su 2s+u
1 s 2 su 2su
q k y k q k y k .
s
1 yq
(C.6)
Its modular transform is

2
1

NS
i 3c
e
Zu,1 ,

Z
ei(i Z)
2 (, ) 1
2 sinh(2 k u)
Z+ 2i
k Z
dZ
(i)e
+
=
3
sinh(Z) cos (i Z )
( ) 2i
C
C+
(C.7)
571
with the same contour as before, but now the integrand has poles at i Z = s Z +
1
2 . Expanding again the last factor of the integrand as in (4.9), (4.10), taking 0 and
shifting the contour by iwk/2 i2 in each term, we arrive, for general (u, u ), to

2
1

NS,c
NS,d
NS
i 3c
= Zu,u
Zu,u ,
e
(C.8)
+ Zu,u ,

where

NS,c
Zu,u

+

=
dP

(1)w(u+u ) [cosh(2P )cos(kw)] sinh(2 Pk )
wZ 0
P 2 +(kw/2)2
k
2(kw/2)
k
2 (, )
( )3
+

wk
1
+ 1
P
,
=
dP uR
iP , w uR
+ iP , w chR
;
,
c
2
2
2
wZ 0
(C.9)
and

NS,d
Zu,u

(1)
rZ+ 12
s2
1)u) sin( k (2jr 1)u ) y ws q sws k

sin( k (2jr 1))
1 + yq s
wr (u+u ) 2 sin( k (2jr
2 (, )
( )3

+
= 2
Res uR (jr + 1, wr )uR (jr , wr ) chR
d (jr , r; , ),
(C.10)
rZ+ 12
with jr and wr defined as in Section 4, but now taking r Z + 12 .

NS,d
R,d
In both Zu,u
and Zu,u the residues are computed when considering the bracketed
expressions as analytical functions of j .
Appendix D. Embedding N = 2 into SL(2, R)

Our goal in this appendix is to elaborate on the relation between the N = 2, c > 3 chiral
algebra and the SL(2, R) algebra, showing how the former always lead to the latter. These
ideas go back to [23].
Consider an N = 2 algebra with c > 3. We can write the supercurrents and the U (1)
R-current in the following form:

2c i 3c
c
e
,
G =
(D.1)
,
JR = i
3
3
where we made use of a canonically normalized scalar and the fact that the supercurrents carry U (1) R-charge 1. The fields are the first parafermionic currents of
572
SL(2, R)/U (1). We now observe that if we add a trivial auxiliary U (1) to the theory, parameterized by an anti-Hermitean scalar field T , we can define the following currents:

(k + 2)
k+2
(k + 2)
3
+
j =i
T = J R +
T ,
k
2k
2k

2
2
k+2
i
j = k + 2 exp
(D.2)
+
T
k+2
k
k
which satisfy an SL(2, R) algebra at level k + 2:20
j 3 (z)j 3 (w)
j + (z)j (w)
k+2
,
2(z w)2
j 3 (z)j (w)
k+2
2j 3 (w)
.
zw
(z w)2
j (w)
,
zw
(D.4)
To make use of the purely bosonic current algebra when solving the theory, we will need
to introduce primary fields for the N = 2 algebra combined with the U (1) that transform as
standard representations of the SL(2, R) current algebra. This can be achieved as follows.
When we denote Zj ;r,r a primary of the N = 2 algebra with conformal weight j,r =
j (j 1)/k + m2 /k and U (1) R-charge m = j + r, we can introduce the new primaries
Z j,r,r :
Z j,r,r = Zj,r,r ej,r TL +j,r TR ,
(D.5)
where we choose j,r such that

2
j,r
j (j 1)
(D.6)
= j,r
k
2
(i.e., the momentum of the boson T is coupled to the N = 2 U (1) R-charge). The new
primaries can be checked to transform in a standard fashion as operators of SL(2, R), in a
basis labeled by the spectrum of an elliptic generator. To go to the hyperbolic basis we can
Fourier transform:

j (z, z , x, x)
(D.7)
Z j,r,r x j 1+m x j 1+m .
=
(Z j,r,r ) =
m,m
We can now compute correlators of these fields, which are linear combinations of primaries of the N = 2 algebra dressed with an orthogonal and free U (1), using a full
20 If we wish, we can add a superpartner for the boson T , and view the linear combination of and T
orthogonal to I 3 as bosonized complex fermion:

i
k+2
+
k
2
T = : + :.
k
(D.3)
The three fermions then complete an N = 1 supersymmetric SL(2, R)k theory, containing a purely bosonic
SL(2, R)k+2 .
573
chiral SL(2, R). Thus, if the solution of the H3+ theory is purely based on symmetries, this
holds for any N = 2 theory with c > 3.21
We note also that the construction above shows that the x-variables basis will remain
an efficient formalism for computation as long as N = 2 supersymmetry is preserved. This
complements the remarks in [40] where it was suggested that these variables should be
interpreted as parameterizing a new worldsheet, at any point in the moduli space.
Appendix E. Conventions
We take q = ei2 , y = ei2 , and we define:
1
( ) = q 24

1 qn ,
(E.1)
n=1
1
2 (, ) = q 8 y 2

n2

1 q n 1 + q n y 1 + q n1 y 1 =
q 2 yn,
n=1
3 (, ) =
4 (, ) =

n2 n
1
1
1 q n 1 + q n 2 y 1 + q n 2 y 1 =
q2y ,
n=1
(E.2)
nZ+ 12
(E.3)
nZ

1
1
n2
1 q n 1 q n 2 y 1 q n 2 y 1 =
(1)n q 2 y n .
n=1
(E.4)
nZ
A shift in the second argument of the third theta-function can be compensated for as follows
(w Z):
3 (, + w ) = q
w2
2
y w 3 (, ).
(E.5)
References
[1] L. Girardello, A. Pasquinucci, M. Porrati, N = 2 MorseLiouville theory and nonminimal superconformal
theories, Nucl. Phys. B 352 (1991) 769.
[2] D. Kutasov, N. Seiberg, Noncritical superstrings, Phys. Lett. B 251 (1990) 67.
[3] Y. Kazama, H. Suzuki, New N = 2 superconformal field theories and superstring compactification, Nucl.
Phys. B 321 (1989) 232.
[4] V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, unpublished notes.
[5] A. Giveon, D. Kutasov, Little string theory in a double scaling limit, JHEP 9910 (1999) 034, hep-th/9909110.
[6] K. Hori, A. Kapustin, Duality of the fermionic 2d black hole and N = 2 Liouville theory as mirror symmetry,
JHEP 0108 (2001) 045, hep-th/0104202.
21 We assume that the N = 2 theory does not split into subtheories with c < 3. This assumes the irreducibility
of the representations space of the N = 2, c > 3 algebra. Otherwise, a similar SU(2) algebra would be a more
appropriate tool to solve the theory.
574
[7] D. Tong, Mirror mirror on the wall: on two-dimensional black holes and Liouville theory, JHEP 0304 (2003)
031, hep-th/0303151.
[8] D. Israel, C. Kounnas, A. Pakman, J. Troost, The partition function of the supersymmetric two-dimensional
black hole and little string theory, hep-th/0403237.
[9] N. Seiberg, New theories in six dimensions and matrix description of M-theory on T 5 and T 5/Z(2),
Phys. Lett. B 408 (1997) 98, hep-th/9705221;
M. Berkooz, M. Rozali, N. Seiberg, On transverse fivebranes in M(atrix) theory on T**5, Phys. Lett. B 408
(1997) 105, hep-th/9704089.
[10] T. Eguchi, Y. Sugawara, Modular bootstrap for boundary N = 2 Liouville theory, JHEP 0401 (2004) 025,
hep-th/0311141.
[11] J. McGreevy, S. Murthy, H. Verlinde, Two-dimensional superstrings and the supersymmetric matrix model,
JHEP 0404 (2004) 015, hep-th/0308105.
[12] C. Ahn, M. Stanishkov, M. Yamamoto, One-point functions of N = 2 super-Liouville theory with boundary,
Nucl. Phys. B 683 (2004) 177, hep-th/0311169.
[13] A. Giveon, A. Konechny, A. Pakman, A. Sever, Type 0 strings in a 2d black hole, JHEP 0310 (2003) 025,
hep-th/0309056.
[14] M.R. Douglas, I.R. Klebanov, D. Kutasov, J. Maldacena, E. Martinec, N. Seiberg, A new hat for the c = 1
matrix model, hep-th/0307195.
[15] V. Kazakov, I.K. Kostov, D. Kutasov, A matrix model for the two-dimensional black hole, Nucl. Phys. B 622
(2002) 141, hep-th/0101011.
[16] S. Elitzur, A. Forge, E. Rabinovici, Some global aspects of string compactifications, Nucl. Phys. B 359
(1991) 581.
[17] G. Mandal, A.M. Sengupta, S.R. Wadia, Classical solutions of two-dimensional string theory, Mod. Phys.
Lett. A 6 (1991) 1685.
[18] E. Witten, On string theory and black holes, Phys. Rev. D 44 (1991) 314.
[19] A. Giveon, Target space duality and stringy black holes, Mod. Phys. Lett. A 6 (1991) 2843.
[20] R. Dijkgraaf, H. Verlinde, E. Verlinde, String propagation in a black hole geometry, Nucl. Phys. B 371
(1992) 269.
[21] M. Henningson, S. Hwang, P. Roberts, B. Sundborg, Modular invariance of SU(1, 1) strings, Phys. Lett.
B 267 (1991) 350.
[22] J.M. Maldacena, H. Ooguri, Strings in AdS(3) and SL(2, R) WZW model. I, J. Math. Phys. 42 (2001) 2929,
hep-th/0001053.
[23] L.J. Dixon, M.E. Peskin, J. Lykken, N = 2 superconformal symmetry and SO(2, 1) current algebra, Nucl.
Phys. B 325 (1989) 329.
[24] A. Hanany, N. Prezas, J. Troost, The partition function of the two-dimensional black hole conformal field
theory, JHEP 0204 (2002) 014, hep-th/0202129.
[25] D. Israel, C. Kounnas, M.P. Petropoulos, Superstrings on NS5 backgrounds, deformed AdS(3) and holography, JHEP 0310 (2003) 028, hep-th/0306053.
[26] T. Eguchi, Y. Sugawara, SL(2, R)/U (1) supercoset and elliptic genera of non-compact CalabiYau manifolds, JHEP 0405 (2004) 014, hep-th/0403193.
[27] K. Becker, M. Becker, Interactions in the SL(2, R)/U (1) black hole background, Nucl. Phys. B 418 (1994)
206, hep-th/9310046.
[28] V.S. Dotsenko, The free field representation of the SU(2) conformal field theory, Nucl. Phys. B 338 (1990)
747;
V.S. Dotsenko, Solving the SU(2) Conformal field theory with the Wakimoto free field representation, Nucl.
Phys. B 358 (1991) 547.
[29] G. Giribet, C. Nunez, Correlators in AdS(3) string theory, JHEP 0106 (2001) 010, hep-th/0105200.
[30] D.M. Hofman, C.A. Nunez, Free field realization of superstring theory on AdS(3), hep-th/0404214.
[31] P. Baseilhac, V.A. Fateev, Expectation values of local fields for a two-parameter family of integrable models
and related perturbed conformal field theories, Nucl. Phys. B 532 (1998) 567, hep-th/9906010.
[32] T. Fukuda, K. Hosomichi, Three-point functions in sine-Liouville theory, JHEP 0109 (2001) 003, hepth/0105217.
[33] G. Giribet, D. Lopez-Fogliani, Remarks on free field realization of SL(2, R)/U (1) U (1) WZNW model,
hep-th/0404231.
575
[34] J.M. Maldacena, H. Ooguri, Strings in AdS(3) and the SL(2, R) WZW model. III: correlation functions,
Phys. Rev. D 65 (2002) 106006, hep-th/0111180.
[35] A. Giveon, D. Kutasov, Notes on AdS(3), Nucl. Phys. B 621 (2002) 303, hep-th/0106004.
[36] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
[37] C. Ahn, C. Kim, C. Rim, M. Stanishkov, Duality in N = 2 super-Liouville theory, hep-th/0210208.
[38] Y. Nakayama, Liouville field theory: a decade after the revolution, hep-th/0402009.
[39] M.B. Green, N. Seiberg, Contact interactions in superstring theory, Nucl. Phys. B 299 (1988) 559.
[40] O. Aharony, A. Giveon, D. Kutasov, LSZ in LST, hep-th/0404016.
[41] M. Goulian, M. Li, Correlation functions in Liouville theory, Phys. Rev. Lett. 66 (1991) 2051;
P. Di Francesco, D. Kutasov, World sheet and spacetime physics in two-dimensional (super)string theory,
Nucl. Phys. B 375 (1992) 119, hep-th/9109005.
[42] J. Teschner, Liouville theory revisited, Class. Quantum Grav. 18 (2001) R153, hep-th/0104158.
[43] S. Ribault, V. Schomerus, Branes in the 2D black hole, JHEP 0402 (2004) 019, hep-th/0310024.
[44] T. Fukuda, K. Hosomichi, Super-Liouville theory with boundary, Nucl. Phys. B 635 (2002) 215, hepth/0202032.
[45] A.B. Zamolodchikov, Al.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[46] V. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, Boundary Liouville field theory. I: boundary state
and boundary two-point function, hep-th/0001012;
J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138.
[47] J. Teschner, On the Liouville three point function, Phys. Lett. B 363 (1995) 65, hep-th/9507109;
J. Teschner, On structure constants and fusion rules in the SL(2, C)/SU(2) WZNW model, Nucl. Phys.
B 546 (1999) 390, hep-th/9712256.
[48] B. Ponsot, V. Schomerus, J. Teschner, Branes in the Euclidean AdS(3), JHEP 0202 (2002) 016, hepth/0112198.
[49] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996) 407,
hep-th/9606112.
[50] C. Bachas, M. Petropoulos, Anti-de Sitter D-branes, JHEP 0102 (2001) 025, hep-th/0012234.
[51] H. Rhedin, Gauged supersymmetric WZNW model using the BRST approach, Phys. Lett. B 373 (1996) 76,
hep-th/9511143.
[52] A. Fotopoulos, Semiclassical description of D-branes in SL(2)/U (1) gauged WZW model, Class. Quantum
Grav. 20 (2003) S465, hep-th/0304015.
[53] Y. Matsuo, S. Yahikozawa, Superconformal field theory with modular invariance on a torus, Phys. Lett.
B 178 (1986) 211.
[54] A. Giveon, D. Kutasov, A. Schwimmer, Comments on D-branes in AdS(3), Nucl. Phys. B 615 (2001) 133,
hep-th/0106005.
[55] K. Miki, The representation theory of the SO(3) invariant superconformal algebra, Int. J. Mod. Phys. A 5
(1990) 1293.
[56] D. Israel, A. Pakman, J. Troost, Extended SL(2, R)/U (1) characters, or modular properties of a simple
non-rational conformal field theory, JHEP 0404 (2004) 045, hep-th/0402085.
[57] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hep-th/9712186.
[58] I. Bakas, E. Kiritsis, Beyond the large N limit: nonlinear W (infinity) as symmetry of the SL(2, R)/U (1)
coset model, Int. J. Mod. Phys. A 7 (1992) 55, hep-th/9109029.
[59] S. Ribault, Two AdS(2) branes in the Euclidean AdS(3), JHEP 0305 (2003) 003, hep-th/0210248.
[60] E.J. Martinec, The annular report on non-critical string theory, hep-th/0305148.
[61] N. Seiberg, D. Shih, Branes, rings and matrix models in minimal (super)string theory, JHEP 0402 (2004)
021, hep-th/0312170.
[62] S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici, G. Sarkisian, D-branes in the background of NS5-branes,
Int. J. Mod. Phys. A 16 (2001) 880.
[63] O. Pelc, On the quantization constraints for a D3-brane in the geometry of NS5-branes, JHEP 0008 (2000)
030, hep-th/0007100.
[64] S. Ribault, D3-branes in NS5-branes backgrounds, JHEP 0302 (2003) 044, hep-th/0301092.
[65] D. Israel, A. Pakman, J. Troost, D-branes in little string theory, in preparation.
[66] V.K. Dobrev, Structure of Verma modules and characters of irreducible highest weight modules over N = 2
superconformal algebras, in: H.D. Doebner, et al. (Eds.), Proceedings of the XV International Conference of
576
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
Differential Geometric Methods in Theoretical Physics, Clausthal, July 1986, World Scientific, Singapore,
1987, pp. 289307.
V.K. Dobrev, Characters of the unitarizable highest weight modules over the N = 2 superconformal algebras,
Phys. Lett. B 186 (1987) 43.
E. Kiritsis, Character formulae and the structure of the representations of the N = 1, N = 2 superconformal
algebras, Int. J. Mod. Phys. A 3 (1988) 1871.
P. Goddard, A. Kent, D.I. Olive, Virasoro algebras and coset space models, Phys. Lett. B 152 (1985) 88;
P. Goddard, A. Kent, D.I. Olive, Unitary representations of the Virasoro and super-Virasoro algebras, Commun. Math. Phys. 103 (1986) 105;
Z.a. Qiu, Modular invariant partition functions for N = 2 superconformal field theories, Phys. Lett. B 198
(1987) 497.
A. Schwimmer, N. Seiberg, Comments on the N = 2, N = 3, N = 4 superconformal algebras in twodimensions, Phys. Lett. B 184 (1987) 191.
W. Boucher, D. Friedan, A. Kent, Determinant formulae and unitarity for the N = 2 superconformal algebras
in two-dimensions or exact results on string compactification, Phys. Lett. B 172 (1986) 316.
A. Pakman, Unitarity of supersymmetric SL(2, R)/U (1) and no-ghost theorem for fermionic strings in
AdS(3) N , JHEP 0301 (2003) 077, hep-th/0301110.
A. Pakman, BRST quantization of string theory in AdS(3), JHEP 0306 (2003) 053, hep-th/0304230.
M.R. Douglas, B. Fiol, D-branes and discrete torsion. II, hep-th/9903031.
Towards p-adic string in constant B-field

Debashis Ghoshal 1 , Teruhiko Kawano
Department of Physics, University of Tokyo, Hongo, Tokyo 113-0033, Japan
Received 13 October 2004; accepted 17 December 2004
Abstract
Spacetime properties of the tachyon of the p-adic string theory can be derived from a (non-local)
action on the p-adic line Qp , thought of as the boundary of the worldsheet. We show that a term
corresponding to the background of the antisymmetric second rank tensor field B can be added to this
action. We examine the consequences of this term, in particular, its relation to a non-commutative
deformation of the effective theory of the p-adic tachyon in spacetime.
1. Introduction
The tachyon sector of the D-branes of the open p-adic string theory is amenable to exact
analytic calculation. It turns out that in many ways these theories behave like the ordinary
bosonic string. The history of the p-adic string is also not unlike the ordinary string theory. It was introduced in Ref. [1], which proposed the KobaNielsen formula for the tree
amplitude of N tachyons of the p-adic theory by generalizing the formula from the real
numbers to the p-adic numbers. Namely, the integral was taken over (a quadratic extension
of) the p-adic number field Qp and the absolute values in the integrands were replaced by
the non-archimedian norm. It was shown that this prescription defines a consistent string
theory. Soon afterwards it was realized [25] that these integrals can be computed exactly.
E-mail addresses: ghoshal@hep-th.phys.s.u-tokyo.ac.jp (D. Ghoshal),
kawano@hep-th.phys.s.u-tokyo.ac.jp (T. Kawano).
1 On sabbatical leave from the Harish-Chandra Research Institute, Chhatnag Road, Allahabad 211 019, India.
doi:10.1016/j.nuclphysb.2004.12.025
578
D. Ghoshal, T. Kawano / Nuclear Physics B 710 [PM] (2005) 577598
The spacetime effective field theory of the tachyons is, therefore, known exactly. This was
written down in Refs. [3,4]. (Let us notice that in this approach, spacetime is the usual
Minkowski one. A more exotic type of p-adic string living in p-adic spacetime was studied in [6], but we will not consider it here.)
Although the prescription to define the p-adic string was well motivated mathematically
and paid rich dividends, it did not shed much light on the nature of the p-adic string itself.
The development in this direction came from the works of [711]. In Refs. [79], a nonlocal action on the p-adic field Qp was proposed for X ( ) ( Qp ), the target space
coordinates. Finally, a worldsheet theory was given in [10,11]. The bulk of the open
string worldsheet turns out to be an infinite Bethe lattice, also called a BruhatTits tree,
with coordination number p +1; the boundary of which is isomorphic to Qp . The Polyakov
action on the worldsheet is the usual lattice action for the free massless fields X . It was
shown in [10] that starting with a finite lattice and inserting the tachyon vertex operators
on the boundary, one recovers the prescription of [15] in the thermodynamic limit. For
related works on the p-adic string theory, see [1214], as well as the review [15].
More recently, the p-adic string theory have come to focus through the realization that
the exact spacetime theory of its tachyon allows one to study non-perturbative aspects
of string dynamics, like the process of tachyon condensation. In [16], the solitons of the
effective theory of the p-adic tachyon [3,4,12] were identified with the D-branes and shown
that the dynamics is according to the behaviour conjectured by Sen [17]. Moreover, it turns
out that in the p 1 limit2 [18], the theory provides an approximation to the boundary
string field theory (BSFT) [19,20] of ordinary strings, the formalism of which was useful in
proving the Sen conjectures [18,21,22]. Various properties of the p-adic string have been
explored recently in [2330].
Despite these progress, much remains to be understood about the p-adic strings. We
know some properties of its D-branes in flat space, though restricted mostly to the tachyon
sector. Recently a first step towards understanding the behaviour of p-adic strings in a
non-trivial background was taken in [31,32]. The motivation comes from the fact that a
constant value of the second rank antisymmetric tensor field B in ordinary string theory
has the effect of providing a non-commutative deformation of the open string fields in the
target space. Thus the effective spacetime theory of the p-adic tachyon was deformed by
introducing a non-commutative parameter . Gaussian solitons corresponding to D-branes
were obtained for all values of and shown to interpolate smoothly from the p-adic soliton
[3] to the non-commutative GMS soliton [33]. It was shown that this continues to be the
case down to p 1, i.e., in the BSFT of the ordinary string theory; thus providing a new
insight to the effectiveness of the approach of Ref. [34] in understanding tachyon dynamics.
In this paper, we would like to examine a possible worldsheet origin of the noncommutativity introduced in [31,32] to the effective action of the p-adic tachyon. We
will modify the non-local action [79] on the boundary of the worldsheet in analogy
with the ordinary string theory in a constant B-field.3 We show that the correlation func2 See also the prescient comments in [7].
3 It turns out that this can be done consistently for a subset of primes p, namely p = 3 (mod 4), which roughly
are half of all the prime numbers. However, once we are in the domain of the effective theory in spacetime, there
is no reason why the results cannot be continued beyond this set, indeed to all integers.
579
tions with any number of the tachyon vertex operators can be solved in the presence of
the coupling to the B-field (Section 3). Unfortunately, however, the deformation does not
respect the complete GL(2, Qp ) symmetryit is only invariant under the infinitesimal
transformationthe underlying reason being the lack of a natural order in Qp . As a result,
the four-tachyon scattering amplitude does not quite agree with that computed from the (deformed) effective field theory (Section 4). However, we discuss some formal time ordered
Green functions and related commutators of the tachyon field. As an alternative approach,
we explore the possibilities of deforming the KobaNielsen amplitudes (Section 5) and
end with some comments (Section 6). For completeness, the worldsheet approach to the
p-adic string, developed in Refs. [10] and [79], in flat spacetime with no background
field, is briefly recapitulated (Section 1). Two appendices provide a collection of results
from p-adic analysis useful for our purpose (Appendix A); and a brief derivation of an
infrared regulated integral (Appendix B).
2. Worldsheet action of the p-adic string in the trivial background

In this section we will review the worldsheet approach to the p-adic open string in flat
spacetime with no other background field, as proposed in Refs. [711].
Let us start by recalling the worldsheet construction of the p-adic string given in
Ref. [10]. The interior of the worldsheet, analogous to the unit disc or the upper halfplane of the usual theory is an infinite lattice with no closed loops, i.e., a tree Tp in
which p + 1 edges meet at each vertex (see Fig. 1). This is known to mathematicians
as the BruhatTits tree and is familiar to physicists as the Bethe lattice. The boundary of
the tree Tp , defined as the union of all infinitely remote vertices, can be identified with
the p-adic field Qp . On the other hand, the tree Tp is the (discrete) homogeneous space
Fig. 1. A finite part of the worldsheet of the 3-adic string: the tree Tp for p = 3.
580
PGL(2, Qp )/PGL(2, Zp ): the coset obtained by modding PGL(2, Qp ) by its maximal compact subgroup4 PGL(2, Zp ). The action of PGL(2, Qp ) on Qp extends naturally to Tp .
Let z label the vertices of Tp and e its edges. We will use to label the points on the
boundary Tp Qp . The spacetime coordinates X (z) are functions on the lattice. The
difference e X = X (z ) X (z), where z and z are the end-points of the edge e, is the
finite lattice analogue of the directional derivative. Likewise, the Polyakov action on the
p-adic worldsheet is the free action
2
1
e X (z) .
Sp [X] = p
(1)
2
{e}
The variation of the above gives rise to

(p)
Sp = p
X (z) 2 X (z) + (p 1)
d X ( )Dn X ( ) ,
{z}
(2)
Tp
where
2 X (z) =
p+1
X (zi ) (p + 1)X (z)
(3)
i=1
(p)
(zi are the p + 1 nearest neighbours of z), is the lattice Laplacian, Dn X ( ) is an appropriately defined normal derivative5 at the boundary point and d is a measure on the
boundary Tp (see [10] for the precise definitions). The tension of the p-adic string is p .
(p)
We see that either the Neumann (Dn X ( ) = 0) or the Dirichlet (X ( ) = const) boundary condition can be imposed on a spacetime coordinate X . The solutions of the classical
equations of motion 2 X (z) = 0 are harmonic functions on the tree Tp .
We are interested in the scattering amplitude of N tachyonic scalars. This requires us
to compute the correlation function of the N vertex operators exp(ik I X), (k I )2 = 2,
corresponding to the tachyon with momentum k I (I = 1, . . . , N ), inserted at the boundary
points I Tp = Qp . To this end, we work with a finite part of the lattice of radius R
(N )
centred at the point C (see Fig. 1) and define the scattering amplitude Ap as the limit
R of the finite lattice correlator:

N

1
(N )
I
k X(I ) ,
DX exp Sp [X] + i
Ap = lim
(4)
R Zp
I =1

where I are points on the boundary at radius R. The measure DX = ,z dX (z) is the
usual measure and the normalization in (4) is by the partition function

Zp = DX eSp [X] .
(5)
4 This construction parallels the case of the usual string theory, in which the UHP is the homogeneous coset
PSL(2, R) modulo its maximal compact subgroup SO(2).
5 This is defined in terms of the limit of difference of f (z) and f ( ) as the interior point z of the tree T
p
approach the boundary Qp = Tp .
581
The regularity of the thermodynamic limit R determines the string tension to be

p = 1/ ln p, which also happens to be the condition for the projective GL(2, Qp ) invariance of the tachyon scattering amplitude. In Ref. [10], it is shown that the expression (4) is
the p-adic analogue of the KobaNielsen amplitude
N
I N
I N1
I J
)
=
dI |I |kp k |1 I |kp k
|I J |kp k ,
A(N
(6)
p
Qp I =1
1I <J N 3
where the projective invariance is used to fix the boundary points N 2 = 0, N 1 = 1 and
N = .
It is also possible to integrate over the degrees of freedom in the interior of the tree Tp
to arrive at a non-local action on the boundary (see Appendix C of Ref. [10] for details):

p(p 1)p
(X ( ) X ( ))2
Sp =
(7)
d d
.
4(p + 1)
| |2p
Qp
This action was proposed in [79] and shown to yield the correlators of the tachyon vertex
operators inserted on the boundary. In the following, this will be our starting point for
coupling the B-field. Let us note that in the case of the usual bosonic string theory too, there
is an analogue of this non-local actionit is the one obtained by the obvious substitutions
in Eq. (7).
3. A constant B-field background in the boundary effective action

In ordinary string theory, a constant background B-field contributes a total derivative in
the worldsheet action. Hence its only effect is in the boundary

1
d 2 z X (z) X (z)B = d B X ( )t X ( ),
(8)
2
where t is the tangential derivative along the boundary , which we label by . (Formally, this is the same as the insertion of the vertex operator for a background gauge field
A [X( )] B X ( ).) In principle therefore, it is a straightforward problem to generalize the effect of a constant B-field in the p-adic string theory: add the p-adic analogue of
the boundary term (8) to the action (7). However, one is immediately faced with a problem:
there is no natural notion of a (tangential) derivative along the p-adic line Qp (see [3538]
for aspects of p-adic analysis).
The same problem was encountered by the authors of [13] in their computation of the
correlation function involving the vertex operator of a vector field (k)t X eikX ( ) in the
p-adic string theory. The solution proposed there is to use the CauchyRiemann relations
for a pair of harmonic functions in the usual case, and write the tangential derivative as
[13,35]

sgn ( )
(p)
X ,
t X ( ) X = d
(9)
| |2p
Qp
582
where sgn ( ) is an analogue of the sign function over R. More precisely, it is a socalled
branching character on Qp corresponding to the quadratic extension6
multiplicative
Qp ( ) by . It turns out that, up to equivalence, there are three choices = , p and p

(where is a (p 1)th root of unity). However, if we demand the antisymmetry property
of the sign function, as in R:
sgn ( ) = sgn ( )
(10)
holds only for the last two choices (namely, = p, p) with an additional restriction on
the value of the prime p = 3 (mod 4), which is satisfied by roughly half the prime numbers
[35]. The antisymmetry property (10) is of importance to us, so we will restrict to these
values of p. For a Qp , sgn ( ) is defined as follows:

2
2
sgn ( ) = +1, if = 1 2 for some 1 , 2 Qp ,
(11)
1, otherwise.
Let us recall that the function sgn was used in defining p-adic string amplitudes with
ChanPaton factors [3,15,39].
We are now in a position to put forward our proposal for the boundary effective action
in the presence of a constant background B-field. It is:

(X ( ) X ( ))(X ( ) X ( ))
T0
d d
Sp (B) =
2
| |2p
Qp
+i
1+p
p 2 (1)

B X ( )
Qp

sgn ( )

,
d
d
X
| |2p
(12)
where ( ) is a generalized gamma function over the p-adic field7 discussed in

Appendix A and
p(p 1) 1
(13)
2(p + 1) ln p
is the string tension. The antisymmetry of the B-field requires that at least two directions
are involved, thus the rank of B is always even. In the following, although we will not
be precise about it: if a (spatial) coordinate X is involved, it is assumed that B has a
component in that direction.
The normal derivative [710] on Qp

f ( ) f ( )
(p)
n f ( ) = d
(14)
| |2p
T0 =
Qp
6 The subscript Q refers to a p-adic number such that

/ Qp . One obtains a quadratic extension
p
Qp ( ) of the p-adic field by adjoining ; i.e., by considering elements of the form + for , Qp .
Unlike R the quadratic extension of Qp is not unique. In other words, there are several choices for and the
function sgn depends on this choice. It is worth emphasizing, however, that sgn is a (real valued) function on
Qp , and does not require that we extend it.
7 The generalized gamma functions (s) associated with sgn , is not singular at zero or negative integer
arguments, or indeed anywhere on the complex s-plane.
583
appears in the first line of Eq. (12). Let us note in parenthesis that a normal derivative
(p)
Dn was introduced in Eq. (2). Neither the Haar measure d nor the normal derivative
(p)
(14), match the corresponding quantities in Eq. (2). However, the combination d n f ( )
(p)
equals d Dn f (see [10] for details).
3.1. Correlators in the presence of B-field
In the rest of the section, we will analyze various consequences of action (12) using
standard field theoretic techniques. We recall that the N -point correlation function of the
tachyon vertex operator eikX ( ) in a constant B-field background is given by the correlator

I
ik X

DX exp(Sp (B) + i N
I =1 k X(I ))
ikN X
1

e
(15)
.
(1 ) e
(N ) B =
DX exp(Sp (B))
Let us introduce the generating function

Z[J ] = DX exp Sp (B) + i d J X( ) ,
(16)
Qp

I
which, when the external current J ( ) = N
I =1 k ( I ), yields the N -point correlation function (15). The generating function Z[J ] satisfies the DysonSchwinger equation

ln Z[J ]
= J ( ),
d
(17)
J ( )
Qp
where the operator

(p)
= T0 n + i
1+p
(p)
B t
p 2 (1)
(p)
(18)
(p)
is a combination of the normal derivative n and tangential derivative t defined in

Eqs. (14) and (9), respectively.
It remains to solve the Green function G ( ) which is a kernel to the differential
operator (18)

d G = .
(19)
Qp
To this end, we solve the equation of motion:

d X = J ( ),
(20)
Qp
by taking a Fourier transformation on the p-adic number Qp , as we do for ordinary strings

(see Appendix A and [3538] for details). In the Fourier space Eq. (20) takes the form
T0

1 +p
iB sgn () ||p X () = J ().
2
p
(21)
584
Introducing the open string metric G and the theta parameter by the relation

p1
1
i
1
G +
=
,
2 p ln p (0)
iB
we find that
X () =
(22)

p1
p2
i
G +
sgn ()
T0 (1 + p)
2 p ln p (0)
N
1 I
k p (I ),
||p
(23)
I =1
where p () is the p-adic analogue of the function ei (see Appendix A). We have also

I
used the expression of the external current J = N
I =1 k ( I ), appropriate for (15).
Finally, after an inverse Fourier transform, we arrive at the desired expression
X ( ) =
N

I =1

i
kI G ln | I |p sgn ( I ) .
2
(24)
To be precise, the expression above is obtained after an infrared regularization. It is similar to the usual string theory and the details are given in Appendix B.
The Greens function is therefore

i
G = G ln p + sgn ,
(25)
2
in terms of which we have a complete solution

N
1 I J
k k G (I J )
Z[J ] = exp
(26)
2
I,J =1
up to an irrelevant constant of integration. In particular, the N -point correlation function

(15) is:

eik
1 X
(1 ) eik
N X
exp( 2i kI kJ sgn (I J ))

(N ) B =
,
I kJ
G k
|I J |p
I,J =1
(27)
I <J
where we have omitted the momentum conserving delta function. This is the p-adic analogue of the result in Ref. [40] for the usual bosonic string and has the same form as the
latter.
3.2. Formal consequences of the B-field
Notwithstanding its apparently nice expression and the similarity with the real case,
there are problems with (27). It is not invariant under projective invariance of Qp , i.e., under GL(2, Qp ) transformations. Looking back, we find that this is indeed a problem with
(12). As a matter of fact, even in the real case, the amplitudes are invariant under those
585
SL(2, R) transformations which preserve the cyclic ordering of the tachyon vertex operators. It is here that we face a fundamental problem: one cannot define an order in Qp that
is compatible with its algebraic properties. The function sgn is therefore not associated
with any ordering. The best we can do is to prove invariance under infinitesimal SL(2, Qp )
transformation. However, before we sketch the proof, let us provide some formal argument, which show that the B-field background gives rise to spacetime non-commutativity,
at least formally.
We start by defining an ordering (time ordering) in Qp . This can done by ordering the
coefficients in the power series expansion of a p-adic number: = p N (0 + 1 p + )
(given in Appendix A), but let us add a further ingredient to it. The first coefficient 0
can be any of the non-zero elements of the finite field Fp = Zp /pZp Z/pZ. Moreover,
a primitive (p 1)th root of unity exists [35] in Fp . In other words, there exists Fp ,
such that p1 = 1 and powers of generates Fp . This implies that out of the (p 1) values
of 0 , exactly half are odd/even powers of . It is now possible to show that for a fixed N ,
i.e., for | |p = p N , exactly half have sgn ( ) = +1 and the other half have sgn ( ) = +1.
In terms of the worldsheet in Fig. 1, at each node along the dashed line from zero to
infinity, (p 1) branches labelled by 0 originate. Of these, (p 1)/2 correspond to a
p-adic number with sgn ( ) = +1 (respectively 1). By convention, we can say that
the positive (negative) set is the one above (below) the dashed line, and each half can be
ordered by the norm and the coefficients in the power series. The purpose of this exercise
is to show that we can approach = 0 through a sequence of points I such that |I |p 0
and sgn (I ) fixed to be either of 1, exactly as in the case of the real string.
We can now follow standard procedure [41,42] to write

X (0), X (0) = T X (0)X (0) X (0)X (0+)

=
lim
X (0)X ( )
X (0)X ( )
lim
sgn ( )=1
| |p 0
= i
sgn ( )=+1
| |p 0
(28)
where, we have made use of the Green function (25). It is possible to rephrase it in another
way. The exact expression for the correlators (27), can be written equivalently as the formal
operator product expansion between the tachyon vertex operators on Qp :

exp( 2i k k sgn ( )) i(k+k )X

:eikX ( )::eik X : =
:e
( ): + .
G k k
| |
p
Let us introduce, following Ref. [41],

1
ikX
X( ) =
d D k (k)e
( ),
(2)D
(29)
(30)
RD
where (k)
is the Fourier transform of the function (x). Notice that the object (30)
is similar to a string field. More precisely, in the usual string theory in which labels
the real line, (x) = lim 0 (X( ))|0 is the string field for the tachyon. Remaining
in the real case, we recall that the operator product expansion of the vertex operators is
586
equivalent to the statement about multiplication of the objects in (30). Without a B-field,
(X(1)) (X(0)) = ( )(X(0)) + ; this gets deformed to a non-commutative product
in the presence of a constant B-field.
Returning to the case of the p-adic string, we will evaluate the product (X(1)) (X(0)).
(Parenthetically, evaluating this with = 1 and = 0 assumes that the GL(2, Qp ) invariance can be used to find a more general expression [43].)

1

X(1) X(0) =
d D k (
)(k)eikX (0),
(2)D
RD

i

1

d D k k k k e 2 k k .
(
)(k) =
(31)
D
(2)
RD
Thus, it seems that as a result of the B-field in (12), ordinary pointwise multiplication of
functions of spacetime is deformed to the non-commutative Moyal product.
It would, however, be prudent to interpret the above results with some caution. The
arguments we have used are rather formal. The B-field does not respect GL(2, Qp ) invariance. And the order that we have defined is not compatible with the algebraic properties
of the field Qp . In particular, it is not invariant under the projective symmetry. In order to
demonstrate that the B-field leads to spacetime non-commutativity, one needs to calculate
all the tachyon N -point functions, in the presence of the constant B-field. We will discuss
the difficulties with this in the next section.
3.3. Infinitesimal projective invariance
Let us study the p-adic analog of the projective (Mbius) transformation GL(2, Qp ) on
the correlation functions (27). The coordinates I (I = 1, . . . , N ) of the vertex operators
are transformed as
aI + b
,
ad bc = 0,
I I =
(32)
cI + d
where a, b, c, d Qp . It is easy to see that the integrated N -point correlation function
)
A(N
p (k1 , . . . , kN ) =
1

N
dI eik X (1 ) eik X (N ) B
(33)
Qp i=1
is invariant if

sgn I J = sgn (I J ),
(34)
is true. In the usual case of the real strings, this is true under SL(2, R) transformations
which preserve the cyclic ordering of {1 , 2 , . . . , N }. In the p-adic case, the absence of
an order compatible with the algebraic properties of Qp , prevents us from making an analogous restriction. We will therefore have to be content with a limited invariance. Namely,
we will prove Eq. (34) for an infinitesimal transformation:
I I = I + 1 + 0 I + 1 I2 ,
(35)
587
which we require to satisfy

|0 |p < 1,
|1 |p |I |p < 1 (I = 1, 2, . . . , N ).
(36)
Therefore, sgn (I J ) = sgn (I J ) sgn (1 + 0 + 1 (I + J )). Thanks to the nonarchimedian property: |i + j |p max(|i |p , |j |p ) and the condition (36),
1 + 0 + 1 (I + J ) 1 + pZp .
It can be shown that the sign function (11) of p-adic numbers of this form is positive [35].
This proves the invariance of the integrated correlation function (33) under infinitesimal
projective transformation. Unfortunately, however, this is not sufficient to fix the positions
of three of the vertex operators.
4. Tachyon scattering amplitudes

In this section, we evaluate the N tachyon scattering amplitudes, specifically for N =
3, 4. The lack of GL(2, Qp ) invariance due to the B-field does not, strictly speaking, allow
us to fix three of the positions I . We will, nevertheless, fix 1 , 2 and N by hand to 0, 1 and
. The last one requires some care, as we can approach infinity through a sequence of
positive or negative p-adic numbers (see the discussion in Section 3.2). We can average
over the two, however, it turns out that the final result is not sensitive to the details as long
as the sequence of points have the same value of sgn .
Let us consider the three-tachyon amplitude to begin with. Ignoring the momentum
i
(3)
conserving delta function and an infinite factor, the result is Ap (k1 , k2 , k3 ) = e 2 (k1 k2 ) .
Since there is no natural notion of a cyclic (or indeed any) ordering in Qp , we need to add
by hand a term with the role of k1 and k2 interchanged and average over the two. The final
result is:
1
A(3)
p (k1 , k2 , k3 ) = cos (k1 k2 ).
2
(37)
This expression is the same as that of the usual bosonic string.

Consider the case N = 4 next. Once again, fixing 1 , 2 and 4 by hand, we arrive at the
integral:
e
i
2 k1 k2

d
| |kp1 k3 |1 |kp2 k3

i
i
exp (k1 k3 ) sgn ( ) (k2 k3 ) sgn (1 ) .
2
2
Qp
(38)
This can be evaluated in two ways: either we can expand the exponential and use the
expression of generalized p-adic beta-/gamma-functions (given in Appendix A), or we
can divide the integral in four domains as in [3]. Following the first route, we get, after
averaging with a term with (k1 k2 ):
588

A(4)
p = c12 c13 c23 (s) (t) (u) + s12 s13 c23 (s) (t) (u)

s12 c13 s23 (s) (t) (u)

+ c12 s13 s23 (s) (t) (u)

p1
p1
1
1
1
1
+
c
= c12 c34
c
13 24
p p (s) 1 p
p p (t) 1 p

p1
1
p+1
1
+ c12 c13 c23
,
+ c14 c23
(39)
p p (u) 1 p
p
where, we have used the abbreviations cI J = cos 12 kI kJ and sI J = sin 12 kI kJ
and as usual, (s) = 2s = k1 k2 + 1, etc. Alternatively, following [3], we divide Qp into
Zp and its complement D1 ; then divide Zp further into three domains: D2 , in which the
first coefficient 0 in the expansion of is 0; D3 corresponding to 0 = 1 and D4 correi
sponding to the union of 0 = 2, 3, . . . , p 1. In D1,2,3 the integral gives

i
p1 e 2 k2 k4 c13
p
p (t) 1
2i k1 k4
p1 e 2 k1 k2 c34
,
p
p (s) 1
c23
e
and p1
respectively. In the region D4 , we are unable to do the
p
p (u) 1
calculation for an arbitrary prime p. However, explicit evaluation for the first few relevant
primes leads to the result
i

i
i
e 2 (k1 k2 )
(p 3)c13 c23 + e 2 (k1 k3 )+ 2 (k2 k3 ) ,
p
(40)
which, when averaged with the term with (k1 k2 ), agrees with (39). Recall that in the
commutative case, this region (and likewise its analogues for higher N ) gives the fourtachyon (respectively N -tachyon) vertex in the spacetime effective field theory, as the
integrand here is independent of the spacetime momenta in that case. With the B-field
turned on, there is always momentum dependence, but in this region, it is only in the
phase factors. Let us also note that the presence of the -function in the four-tachyon
amplitudes makes it impossible for these to have any adelic property.
Unfortunately, the four-point function is not symmetric under a permutation of the four
momenta. Even if we consider summing up the terms with different permutations by hand,
the resulting expression does not match the results of the non-commutative deformation of
the tachyon effective field theory considered in [32]. The latter is a field theory with the
action:
p
LNC (T ) =

N
p+1

g
(p 1)!
p
1
12 1
N
T+
(T ) ,
T p
p1
2
N !(p N + 1)! p
N =3
(41)
where T is the tachyon field on the D-brane and g is the open string coupling. The threeand four-tachyon amplitudes from this field theory are:
p1
c12 ,
p

c12 c34
c13 c24
c14 c23
p1
p(4)
+
+
ANC
p
p (s) 1 p (t) 1 p (u) 1
p2
(c12 c34 + c13 c24 + c14 c23 ).
+
p
589
p(3)
ANC
(42)
Finally, let us contrast the result (39) with the four-tachyon amplitude in the usual noncommutative bosonic string theory:

1
1
(4)
AB = cos (k1 k3 k2 k4 )B (s), (t) + cos (k1 k4 + k2 k3 )B (t), (u)
2
2

1
+ cos (k1 k2 + k3 k4 )B (u), (s) ,
(43)
2
where we have averaged over cyclic and anticyclic permutations and B(, ) is the beta
function.
5. Alternative proposal for the four-tachyon amplitude

In this section, we will explore the possibility of defining four-tachyon amplitudes of
p-adic string in a constant B-field so as to reproduce the non-commutative deformation
(41) of the effective field theory [32]. To this end, let us recall the situation for the usual
bosonic string theory. Either in the presence of the B-field or with the ChanPaton factor,
(4)
the coefficient of the non-commuting factors contributing to AB in the different channels,
can be summed up to write a single integral over the real numbers [44]. In [45], which
considers introducing the ChanPaton factor in p-adic strings, it was suggested that this
property be abandoned. Following this idea, we will consider different channels separately.
Consider the tu channel first: in the ordinary string theory, this is given by:
(4)
AB,tu
1
=
d | |k1 k3 |1 |k2 k3 cos

1
k1 k2 + ( )k1 k3 + (1 )k2 k3 ,
2
(44)
where, ( ) is the sign function on R. We will attempt to write an appropriate p-adic
analogue of the above. However, there does not exist a unique prescription for this, a fact
that was already realized in [39,45] in the context of the ChanPaton factors. In order to
generalize (44) to the p-adic case, we need to insert a suitable projector to restrict the
range of integration. There are several possibilities8 to achieve this, as has been listed by
the authors of [39]:
1 ( ) =

1
1 + sgn ( ) 1 + sgn (1 ) H ( )H (1 ),
4
8 Our choice, however, is somewhat more restricted by the fact that we consider only = p, p with prime
p 3 (mod 4), so as to ensure sgn ( ) = sgn ( ). This limits our options compared to [39].
590

1
1 + sgn ( ) sgn (1 ) ,
2

1
3 ( ) = sgn ( ) + sgn (1 ) ,
2
2 ( ) =
(45)
where, H ( ) = 12 (1 + sgn ( )) is the p-adic analogue of the Heaviside function. Inserting

one of the step functions above, we can thus define:

(4)
Atu, = d ( )| |kp1 k3 |1 |kp2 k3
Qp

1
sgn (1)k1 k2 + sgn ( )k1 k3 + sgn (1 )k2 k3 .
(46)
2
In the following, we will study the different choices in turn and examine whether a field
theoretic interpretation is possible in each case.
Let us consider 1 first.
cos
(4)
k 1 k4 + k2 k3
1
cos
4

2

(s) (t) (u) (s) (t) (u)

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

k 1 k4 + k2 k3 p 3 p 1
1
1
1
+
= cos
+
.
2
2
2p
p
p (u) 1 p (t) 1
Atu,1 =
(47)
Upon summing over the three, namely, tu, us and st channels, we find that the resulting Veneziano amplitude agrees qualitatively with the calculation from the effective field
theory (41).
For the other choices of the step function in (46), the results are:
(4)
(4)
(4)
(4)
(4)
(4)
Atu,2 = c12 (c13 c23 + s13 s23 )Btu,2 s12 (s13 c23 c13 s23 )Btu,3 ,
Atu,3 = c12 (c13 c23 + s13 s23 )Btu,3 s12 (s13 c23 c13 s23 )Btu,2 ,
where,
(4)
Btu,

=
d ( )| |kp1 k3 |1 |kp2 k3 .
(48)
(49)
Qp
Explicitly,

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

1
1
p1
1 + (t)
+
,
=
2p
p
1 p (u) 1

1
= (s) (t) (u) + (s) (t) (u)
2

1
1
1 p1
+
.
= +
p
2p
p (t) 1 p (u) 1
(4)
Btu,
=
2
(4)
Btu,3
(50)
Hence:
(4)
Atu,2,3
591

1
1
1
p3 p1
+
+
= cos (k1 k4 + k2 k3 )
2
4p
2p
p (t) 1 p (u) 1
p+1
1
,
cos (k1 k2 + k2 k3 + k3 k1 )
(51)
2
4p
the two choices differ by the sign of the term in the second line. When we sum over the
three channels, the result does not match that obtained from a field theoryit is the second
term which is not compatible. Thus, among the step functions in (45), only 1 ( ) allows
for a field-theoretical interpretation. There is one drawback of this prescription, however.
Upon turning off the background B-field, the amplitude does not reduce to the original
amplitude. The difficulty here is reminiscent of the same problem encountered in introducing ChanPaton factors. It is likely that the source of the problem and possibly its solution
have the same origin.
Formally, this prescription can be extended to higher-point amplitudes in one of the
channels by insertion of the Heaviside functions as:

d3 dN 1
N
1
|I |kp1 kI |1 I |kp2 kI H (I )H (1 I )
I =3
Qp
N
1
I,J =3
I <J
|I J |kpJ kJ H (J

N 1
i
I ) exp
(kI kJ ) sgn (I J ) . (52)
2
I,J =1
I <J
Summing over all the permutations under the exchange of momenta of the above amplitude
is expected to yield full N -point amplitude. However, we will not attempt to check if this
agrees with the corresponding amplitude derived from the effective field theory.
6. Discussions
In summary, we have proposed a coupling for the antisymmetric tensor field B in
the p-adic string theory. Specifically, we have added a term corresponding to a constant
B-field to the non-local action on the boundary Qp of the p-adic worldsheet. The exact Greens function for the fields X ( ) is computed in the presence of the constant
B-field. The results parallel the case of the usual string theory in a constant B-field [41,
42]: there is a non-commutative factor in the correlation functions of the tachyons and the
flat closed string metric is replaced by the open string metric G . However, the action
with the B-field is not invariant under Mbius transformation, a problem that is intimately
connected with the fact that there is no natural order among the p-adic numbers. An unfortunate consequence is that the resulting tachyon amplitudes do not quite match those
obtained from the non-commutative deformation of the tachyon effective field theory. We
also examined the possibility of defining KobaNielsen amplitudes so as to derive the field
theory results.
592
Our objective was to derive the non-commutative deformation of the spacetime effective
action of the p-adic tachyon, proposed in Refs. [31,32], from a worldsheet point of view.
In spite of having some encouraging results, this problem remains unresolved. We would
expect that for a proper understanding of the B-field, one has to deal with closed p-adic
strings of which virtually nothing is known. In the early days, in analogy with the usual
strings, it was thought that the closed strings have to do with the quadratic extension of
Qp . Unlike R, however, neither is the quadratic extension unique, nor is it closed. Indeed,
there are an infinite number of finite extensions of Qp and none of these is a closed field.
Moreover, each finite extension is isomorphic to the boundary of a tree (a Bethe lattice)
with appropriate coordination number [10]. Thus, they would correspond to some kind of
generalized open strings. It is possible to define a closed field by taking the union of these
extensions and augmenting the set by putting in limit points of sequences. The resulting
field Cp , surprisingly, is isomorphic to the set of complex numbers [37], although their
topologies are very different. It is likely that closed p-adic strings are to be based on this
field Cp .
Be that as it may, a constant B-field is certainly simpler. First, it only affects the boundary terms. Secondly, one can think of the open p-adic worldsheet, at least for p = 2, as
an exotic discretization of the upper half-plane with its Poincar metric [46]. Thus, one
should be able to capture the constant flux of the B-field at the vertices of the tree. Finally,
let us comment on how one may proceed to couple the B-field to the bulk worldsheet
action. To this end, we note that the effect of the B-field is topological. We add the pullback of the 2-form: X (B) = B X X to the worldsheet action. However, the
worldsheet of the p-adic string, being a tree, has no closed loop. Hence naively, there is
no 2-cycle over which to integrate a 2-form. Nevertheless, we notice that for a prime p (or
for any odd integer for that matter), the tree is a bipartite one. We can consistently divide
the edges E = {e} into two disjoint sets E1 = {e1 } and E2 = {e2 }. It is now possible to
antisymmetrize derivatives along the edges belonging to these two subsets:
B [e1 X (z)e2 ] X (z).
(53)
Note that this division does not correspond to the sgn function. While the proposal above
is unlikely to work as it is, it may be interesting to ponder along these lines.
Note added
A paper [47] [hep-th/0409305] suggesting the same way to couple the B-field in the
p-adic string as that considered here, appeared on the arXiv as this manuscript was being finalized. The conclusions there are formal and suffer from subtleties that we have
discussed in Section 3.
Acknowledgements
We are grateful to Chandan Singh Dalawat and Seiji Terashima for useful discussion.
D.G. would like to thank Tohru Eguchi and the Japan Society for the Promotion of Science
593
(JSPS) for an Invitation Fellowship which partially supports this work. The research of
T.K. is supported in part by the Grants-in-Aid (#16740133) and (#16081206) from the
Ministry of Education, Science, Sports and Culture of Japan.
Appendix A. Materia p-adica
In this appendix, we provide a compendium of formulas related to the p-adic number
field Qp and functions over it. It is not intended as a review, but only to collect in one place
most of what is used in this article. For details, we refer the reader to Refs. [3538].
A.1. p-adic numbers and the BruhatTits tree
An element Qp may be written as a power series in p:

= p N 0 + 1 p + 2 p 2 + = p N
n p n ,
(A.1)
n=0
where N Z and n {0, 1, . . . , p 1}, 0 = 0. (There is, however, nothing special about
this choiceone may work with other representative elements.) Notice the similarity with
the Laurent series of meromorphic functions. Addition and multiplication in Qp , defined
in terms of the series (A.1), give Qp the structure of a ring. The subset Zp consisting
of elements with N 0 in (A.1) is a subring known as the p-adic integers. The nonarchimedian p-adic norm of is defined to be
| |p = p N ,
|0|p = 0;
(A.2)
which satisfies the inequality | |p max(| |p , | |p ), which is stronger than the usual
triangle inequality. Notice that the norm of a p-adic integer is at most 1.
The series representation (A.1) of a p-adic number also provides us with an isomorphism between Qp and the boundary Tp . To see this, let us consider the case p = 3 for
definiteness and refer to Fig. 1, in which the dashed line denotes the (unique) path through
the tree connecting = to = 0. Let us label the vertices along this path by CN , with
N = corresponding to zero and infinity respectively and N = 0 being the (arbitrarily
chosen) point C = C0 at the centre of the tree. In order to find a point on the boundary
Tp corresponding to , we start at CN and then choose the left branch if 0 = 1 or the
right one if 0 = 2. At the next step, we choose one of the branches depending on the value
of 1 , and so on. Continuing this way, we arrive at a point on the boundary. This procedure
can obviously be generalized for any p.
1
nN , consisting of negative
Coming back to the series (A.1), the finite part N
n=0 n p
powers of p, is called the fractional part { }p of . The rest, an infinite series in general, is
the integer part [ ]p . Thinking of the fractional part { }p as real, let us define the complex
valued function p : Qp C, as

p ( ) = exp 2i{ }p = exp(2i ),
(A.3)
where, in writing the second expression, we have allowed for a little imprecision to regard
(A.1) as a formal power series in real. As in the real case, the contribution to (A.3) from
594
the integer part [ ]p is trivial and only the fractional part matters. Since p ( + ) =
p ( )p ( ), it is called an additive character of Qp .
A.2. Fourier transformation
The Fourier transform of a complex valued function9 is defined by:

f () = d p ( )f ( ),
Qp
where d is the translationally invariant Haar measure on Qp normalized as

The inverse Fourier transformation is given by

f ( ) = d p ( )f().
(A.4)

Zp
d = 1.
(A.5)
Qp
This can be proven by use of the formula

N

p , for ||p p N ,
p ( ) d =
0,
for ||p > p N +1 ,
(A.6)
BN
however, we will omit the proof. It is useful to have the following representation of the
delta function ( ):

( ) = d p ( ).
(A.7)
It gives
Qp

Qp
d f ( )( ) = f (0) as expected.
A.3. Generalized p-adic gamma and beta functions

The usual p-adic gamma function, called the GelfandGraevTate gamma function, is
defined as:

1 p s1
(s) = d p ( )| |s1
(A.8)
.
p =
1 p s
Qp
9 Strictly speaking, some of the proofs hold directly for a locally constant function f ( ), which is a function
such that for any Qp , f ( ) = f ( + ), whenever | |p pl( ) , for some integer l( ) Z. The results are
then extended to the generalized functions, which are defined as continuous functionals on the linear vector space
of locally constant functions.
For any locally constant function f ( ) with compact support, there exists a maximum value l(f ) =
l be the set of locally constant functions with supmax Qp l( ), called the parameter of constancy. Let Dm
l = {f D |
port contained in Bm = { Qp | | |p pm } and with the parameter of constancy at least l: Dm
m
l
supp f Bm , l(f ) l}. The Fourier transform f () of f ( ) Dm turns out to be in Dl .
595
It has only one singular point at s = 0, a simple pole with residue (p 1)/p ln p. The
generalized gamma function associated with sgn is:

s1/2
, for p = p, p. (A.9)
(s) = d p ( )| |s1
p sgn ( ) = sgn (1)p
Qp
This is an entire function with no singularity on the s-plane. Using the above, we can define
the following generalized p-adic Beta functions:

y1
d | |x1
= (x)(y)(1 x y),
p |1 |p
Qp
y1
d sgn ( )| |x1
p |1 |p
= sgn (1)(x) (y) (1 x y),
Qp
y1
d sgn (1 )| |x1
p |1 |p
= sgn (1) (x)(y)(1 x y),
Qp
y1
d sgn ( ) sgn (1 )| |x1

p |1 |p
= (x) (y)(1 x y).
(A.10)
Qp
A.4. Fourier transform of derivatives

We will now discuss the Fourier transform of the derivatives of a (complex valued)
function on Qp or its extension to the interior of the worldsheet Tp . Recall the expressions
for the normal and tangential derivatives in Eqs. (14) and (9). More generally, the th
normal derivative is given by

(p)
f ( ) f ( )
n
(A.11)
f ( ) = d
.
| |p+1
Qp
After a Fourier transform, we obtain

(p)

(p)
n
f ( ) = d p ( ) n f (),
(A.12)
Qp
where

(p)
n f () = f()
Qp
p ( ) 1
| |+1
p
(A.13)
The integral above is divergent. Therefore, in order to evaluate it, we need to regularize by an infrared cutoff. This is done as follows. Let Sm = { Qp | | |p = p m }, the
subset of p-adic numbers with a fixed norm p m . Clearly, Sm Sn = for m = n, and
596
Qp =
mZ Sm .

Qp
d
| |p+1
= lim

m=NS
d
| |p+1

d
Qp
We define
p ( )
| |p+1
= lim

m=NS

= lim
N

p1
p m ,
lim
p N m=
p ( )
| |+1
p

= lim
N

m=k
p m
N

p m1
m=k1

N
p 1 m
(k1)1
p p
,
p
(A.14)
m=k
where k Z is an integer such that ||p = p k . In the above, we have used the fact that in
our normalization Sm d = (p 1)p m1 . Thus we obtain

p ( ) 1
1 p 1
(A.15)
d
= ||p
,
+1
1 p
| |p
Qp
specifically for = 1:

1 p 2
(p)
||p f().
n f () =
1p
(A.16)
Finally let us turn to the Fourier transform of the tangential derivative. After a change
of the variable in (A.9), one finds

sgn ()
(A.17)

(s)
=
d p ( )| |s1
p sgn ( ).
||sp
Qp
Using the above, it is easy to find the Fourier transform of the tangential derivative:

sgn ( )
(p)
p ( )f ( )
t f () = d d
| |2p
Qp
= (1) sgn ()||p f().
(A.18)
Let us reiterate that (1) is not singular.

Appendix B. A regularized integral
In this appendix, we provide a brief derivation of an integral that was used in the main
part of the text in finding the solution (24) of the equation of motion, or equivalently, the
Greens function (25). The integral in question is

p ( ) p ( )
p 1 | |p
ln
d
=
,
(B.1)
||p
p ln p | |p
Qp
which follows from the infrared regularized expression

p ( ) p 1
1
lim
ln | |p .
d
=
||p
p ln p 0
597
(B.2)
Qp
In order to obtain this, we consider the following more general integral, well-defined for
> 0, and evaluate it as:

||1
(
)
d
=
||1
p
p
p p ( ) d
Qp
m=S
m

1 p 1
1
,
=
||p 1 p
(B.3)
where k Z is an integer such that ||p = p k . The reguralized expression (B.2) is given in
the limit 0. The procedure here exactly parallels the case of standard two-dimensional
theories, such as the one for the worldsheet theory of the usual strings.
References
[1] P.G.O. Freund, M. Olson, Nonarchimedean strings, Phys. Lett. B 199 (1987) 186.
[2] P.G.O. Freund, E. Witten, Adelic string amplitudes, Phys. Lett. B 199 (1987) 191.
[3] L. Brekke, P.G.O. Freund, M. Olson, E. Witten, Non-archimedean string dynamics, Nucl. Phys. B 302 (1988)
365.
[4] P.H. Frampton, Y. Okada, Effective scalar field theory of p-adic string, Phys. Rev. D 37 (1988) 3077.
[5] P.H. Frampton, Y. Okada, The p-adic string N point function, Phys. Rev. Lett. 60 (1988) 484;
P.H. Frampton, Y. Okada, M.R. Ubriaco, On adelic formulas for the p-adic string, Phys. Lett. B 213 (1988)
260.
[6] I.V. Volovich, p-adic string, Class. Quantum Grav. 4 (1987) L83;
B. Grossman, p-adic strings, the Weyl conjectures and anomalies, Phys. Lett. B 197 (1987) 101.
[7] B.L. Spokoiny, Quantum geometry of non-archimedean particles and strings, Phys. Lett. B 207 (1988) 401.
[8] G. Parisi, On p-adic functional integrals, Mod. Phys. Lett. A 3 (1988) 639.
[9] R.B. Zhang, Lagrangian formulation of open and closed p-adic strings, Phys. Lett. B 209 (1988) 229.
[10] A.V. Zabrodin, Non-archimedean string action and BruhatTits trees, Mod. Phys. Lett. A 4 (1989) 367;
A.V. Zabrodin, Non-archimedean strings and BruhatTits trees, Commun. Math. Phys. 123 (1989) 463.
[11] L.O. Chekhov, A.D. Mironov, A.V. Zabrodin, Multiloop calculations in p-adic string theory and BruhatTits
trees, Commun. Math. Phys. 125 (1989) 675.
[12] P.H. Frampton, H. Nishino, Stability analysis of p-adic string solitons, Phys. Lett. B 242 (1990) 354.
[13] A.V. Marshakov, A.V. Zabrodin, New p-adic string amplitudes, Mod. Phys. Lett. A 5 (1990) 265.
[14] L.O. Chekhov, Y.M. Zinoviev, p-adic string compactified on a torus, Commun. Math. Phys. 130 (1990) 130.
[15] L. Brekke, P.G.O. Freund, p-adic numbers in physics, Phys. Rep. 133 (1993) 1, and references therein.
[16] D. Ghoshal, A. Sen, Tachyon condensation and brane descent relations in p-adic string theory, Nucl. Phys.
B 584 (2000) 300, hep-th/0003278.
[17] A. Sen, Non-BPS states and branes in string theory, hep-th/9904207.
[18] A.A. Gerasimov, S.L. Shatashvili, On exact tachyon potential in open string field theory, JHEP 0010 (2000)
034, hep-th/0009103.
[19] E. Witten, On background independent open string field theory, Phys. Rev. D 46 (1992) 5467, hepth/9208027;
E. Witten, Some computations in background independent off-shell string theory, Phys. Rev. D 47 (1993)
3405, hep-th/9210065.
598
[20] S.L. Shatashvili, Comment on the background independent open string theory, Phys. Lett. B 311 (1993) 83,
hep-th/9303143;
S.L. Shatashvili, On the problems with background independence in string theory, hep-th/9311177.
[21] D. Kutasov, M. Marino, G. Moore, Some exact results on tachyon condensation in string field theory,
JHEP 0010 (2000) 045, hep-th/0009148.
[22] D. Ghoshal, A. Sen, Normalisation of the background independent open string field theory action,
JHEP 0011 (2000) 021, hep-th/0009191;
D. Ghoshal, Normalization of the boundary superstring field theory action, in: A. Dabholkar, et al. (Eds.),
Strings 2001, American Mathematical Society, Providence, 2002, hep-th/0106231.
[23] J.A. Minahan, Mode interactions of the tachyon condensate in p-adic string theory, JHEP 0103 (2001) 028,
hep-th/0102071.
[24] J.A. Minahan, Quantum corrections in p-adic string theory, hep-th/0105312.
[25] N. Moeller, B. Zwiebach, Dynamics with infinitely many time derivatives and rolling tachyons, JHEP 0210
(2002) 034, hep-th/0207107.
[26] H.-t. Yang, Stress tensors in p-adic string theory and truncated OSFT, JHEP 0211 (2002) 007, hepth/0209197.
[27] Y. Volovich, Numerical study of nonlinear equations with infinite number of derivatives, J. Phys. A 36 (2003)
8685, math-ph/0301028.
[28] N. Moeller, M. Schnabl, Tachyon condensation in openclosed p-adic string theory, JHEP 0401 (2004) 011,
hep-th/0304213.
[29] V.S. Vladimirov, Y.I. Volovich, On the nonlinear dynamical equation in the p-adic string theory, Theor.
Math. Phys. 138 (2004) 297, math-ph/0306018.
[30] J. Gomis, K. Kamimura, T. Ramirez, Physical degrees of freedom of non-local theories, Nucl. Phys. B 696
(2004) 263, hep-th/0311184.
[31] D. Ghoshal, Noncommutative p-tachyon, in: Proceedings of p-Adic MathPhys. 2003, 15 October 2003,
Moscow, in press.
[32] D. Ghoshal, Exact noncommutative solitons in p-adic strings and BSFT, hep-th/0406259.
[33] R. Gopakumar, S. Minwalla, A. Strominger, Noncommutative solitons, JHEP 0005 (2000) 020, hepth/0003160.
[34] K. Dasgupta, S. Mukhi, G. Rajesh, Noncommutative tachyons, JHEP 0006 (2000) 022, hep-th/0005006;
J. Harvey, P. Kraus, F. Larsen, E. Martinec, D-branes and strings as non-commutative solitons, JHEP 0007
(2000) 042, hep-th/0005031.
[35] I.M. Gelfand, M.I. Graev, I.I. Pitaetskii-Shapiro, Representation Theory and Automorphic Functions, Saunders, Philadelphia, 1969.
[36] N. Koblitz, p-Adic Numbers, p-Adic Analysis and Zeta Functions, Graduate Texts in Mathematics, vol. 58,
Springer-Verlag, Berlin, 1977.
[37] A.M. Robert, A Course in p-Adic Analysis, Graduate Texts in Mathematics, vol. 198, Springer-Verlag,
Berlin, 2000.
[38] V.S. Vladimirov, I.V. Volovich, E.I. Zelenov, p-Adic Analysis and Mathematical Physics, Series on Soviet
and East European Mathematics, vol. 1, World Scientific, Singapore, 1994.
[39] I.Y. Arefeva, B.G. Dragovic, I.V. Volovich, Open and closed p-adic strings and quadratic extensions of
number fields, Phys. Lett. B 212 (1988) 283.
[40] E.S. Fradkin, A.A. Tseytlin, Nonlinear electrodynamics from quantized strings, Phys. Lett. B 163 (1985)
123;
A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields, Nucl. Phys.
B 280 (1987) 599.
[41] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/9903205.
[42] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[43] E. Melzer, Nonarchimedean conformal field theories, Int. J. Mod. Phys. A 4 (1989) 4877.
[44] D.B. Fairlie, K. Jones, Integral representations for the complete four- and five-point Veneziano amplitudes,
Nucl. Phys. B 15 (1970) 323.
[45] A. Barnett, p-adic amplitudes, IMPERIAL-TP-90-91-02.
[46] S. Nechaev, O. Vasilyev, On metric structure of ultrametric spaces, J. Phys. A 37 (2004) 3783.
[47] P. Grange, Deformation of p-adic string amplitudes in a magnetic field, hep-th/0409305.
A matrix model for branes on WZW

simple current orbifolds
Jacek Paweczyk, Rafa R. Suszek 1
Institute of Theoretical Physics, Warsaw University, ul. Hoza 69, PL-00-681 Warsaw, Poland
Received 9 December 2004; accepted 18 January 2005
Abstract
Non-commutative geometry of D-brane worldvolumes on simple current orbifolds of WZW models of type AN is analysed within the framework of reflection equation algebras. Our study of a class
of symmetries of the latter is shown to lead to an explicit construction of (fractional) quantum orbifold branes. The orbifolding procedure itself is given a straightforward algebraic realisation in the
formalism developed. The ensuing orbifold matrix models are then demonstrated to yield results on
brane tensions and the algebra of functions in agreement with the exact BCFT data.
PACS: 02.40.Gh; 11.25.Hf; 11.25.Uv
Keywords: WZW models; Simple current orbifolds; Non-commutative geometry of D-branes; Reflection
equation algebras
1. Introduction
Physics of D-branes has long been a subject of intense study, driven by the motivation
to obtain new insights into the structure of the moduli space of string theory proper and
Work supported by Polish State Committee for Scientific Research (KBN) under contract 2 P03B 001 25
(20032005).
E-mail addresses: pawelc@fuw.edu.pl (J. Paweczyk), suszek@fuw.edu.pl (R.R. Suszek).
1 Present address: Kings College, University of London, Department of Mathematics, Strand, London WC2R
2LS, UK (Marie Curie Fellow).
doi:10.1016/j.nuclphysb.2005.01.023
600
J. Paweczyk, R.R. Suszek / Nuclear Physics B 710 [PM] (2005) 599613
to better understand the emergence of an essentially stringy geometry and gauge dynamics
within any field-theoretic or matrix model approach to the propagation of strings in curved
gravitational backgrounds with fluxes.2 An example of such a background is a compact
Lie group G [2,3], or a quotient thereof, known to support a nontrivial KalbRamond field
in a conformally invariant theory. Maximally symmetric (untwisted) D-branes on G have
been shown to localise stably [4] around a discrete set of conjugacy classes [5] and are
enumerated3 by dominant integral affine weights from the fundamental affine alcove
P+ (g) of the KacMoody algebra g .
In [7], a (quantum) matrix model of branes in WZW models was constructed, based
on certain quantum algebras called Reflection Equation Algebras (REA). The latter have a
representation theory closely related to that of the DrinfeldJimbo algebra Uq (g), which
in turnis known to have irreducible highest weight representations related directly to
integrable highest weight representations of g .
In this paper, we extend the former construction to a new class of WZW backgrounds,
namely, the so-called simple current orbifolds of the WZW geometries of type AN . Simple currents (analysed in great detail in the context of the fixed point resolution problem
in [8]) are known to form an Abelian group, contained as a proper subgroup in the group
(1)
Out(A(1)
N ) of outer automorphisms of AN . Thus an orbifolding of the corresponding
matrix model based on the relevant REA, REAq (AN ), requires that the latter algebra
has the same group of outer automorphisms as the affine algebra. Indeed, we find out
(1)
that Out(REAq (AN ))
= Out(AN ) and, moreover, that the action of both sets of automorphisms on branes is identical. As a result, Out(REAq (AN )) acquires a geometrical
meaning necessary to perform the construction of the orbifold. Furthermore, it is straightforward to identify the CFT monodromy charge in REAq (AN ) so that the orbifolding itself
admits a purely algebraic realisation in the framework developed, reminiscent of the original approach to orbifold models advanced in [9].
At this stage, we may already outline and assess the content of the paper. In Section 2
we briefly describe the REAs, their automorphisms and their relation to branes, we also
introduce the necessary BCFT background. In the following section, we apply our results
in a construction of quantum orbifolds and discuss their properties in some detail, whereby
we also arrive at a particularly straightforward interpretation of the monodromy projection
(known from the BCFT) accompanying simple current orbifolding. These two sections
form the core of the paper. In the closing Section 4, we present an explicit example of
branes on the SU(2)/Z2 orbifold.
2. Reflection equation algebras, branes and WZW orbifolds

In this section, we recall basic facts concerning the so-called reflection equation algebras
[10,11] and their application to branes in WZW models, as proposed in [7].
2 For a comprehensive review, see [1].
3 Here, as anywhere else in this paper, we restrict our analysis of the pre-orbifolding CFT to the diagonal case
in which Cardys classification of boundary states applies [6].
601
The reflection equation algebra REAq (AN ) is the algebra generated by the operator
entries of the matrix M determined by the reflection equation [10]:
R12 M1 R21 M2 = M2 R12 M1 R21 ,
(1)
where R12 is the universal R-matrix R of Uq (AN ) in the fundamental representation RV .

REAq (AN ) is closely related to Uq (AN ) [12] so that, in particular, representations of the
latter give rise to representations of the former. It was representations of REAq (AN ) of the
form:
M = (RV R )R21 R12 ,
P+ (AN ),
(2)
that were proposed in [7] to describe maximally symmetric branes of the associated WZW
0 . For P (A ), the representations
model. We shall denote such representations by R
N
+
0 are mutually inequivalent irreducible highest weight representations of U (A ), of a
R
q
N
non-vanishing quantum dimension.
It was further claimed in [7] that the operators (M )ij generate a (deformed) algebra
of functions on a WZW brane labelled by . Arguments in favour of this interpretation
are numerous. The set of the (untwisted) branes thus obtained is enumerated by integrable
weights P+ (AN ), as required. The algebra of functions on a given brane reproduces
the OPE algebra of horizontal descendants of primary fields of the corresponding WZW
model. Exact values of brane tensions, as dictated by the BCFT, can straightforwardly be
encoded in the ensuing (Uq (AN )-covariant) algebraic structure. Central operators of the
REA provide geometrical information on the (quantised) localisation of the brane inside
the group G manifold. Independent such central elements are given by [13]

ck := trq Mk , k 1, N,
(3)

N
2i
K := MNN+1,N +1 = exp
(4)
nHn ,
N +1
n=1
where on the right-hand side of the last formula we used the map REAq (AN ) Uq (AN )
implicit in (2). Amazingly, the last Casimir (omitted in [7]), K, is related to the BCFT monodromy charge, as one can easily check by comparing the above with (16). Finally, there
is one more scalar, the quantum determinant, to be set to some specific value (independent
of ):
detq M I.
(5)
(1)
As we already remarked in the introduction, we need Out(REAq (AN ))
= Out(AN ) to
form an appropriate orbifold. An immediate consequence of (1) and (5) is the emergence
of
2iL
M e N+1 M,
(6)
as outer automorphisms of REAq (AN ). In fact, it can readily be established thatjust as

requiredwe have Out(REAq (AN )) = ZN +1 Z2 , with the ZN +1 factor realised as above
and the remaining Z2 being the standard mirror symmetry of the Dynkin diagram of AN .
602
Accordingly, the action of the generator of the ZN +1 factor is

2in
: cn e N+1 cn ,
2i
K e N+1 K.
(7)
It turns out to coincide with the left action of the inverse of the Casimir operator (4),

2i
Mij ik K1 Mkj = e N+1 Mij (Mij ).
(8)
We emphasise that the representations induced through (6) and (2) are pairwise inequivL . Notice that R L = R L+1 .
alent. They shall be denoted by R
Next, we turn to WZW models and their simple currents. The latter generate (through
fusion) a discrete Abelian group isomorphic to the group of strictly affine outer automor(1)
phisms4 sOut(AN ) = ZN +1 . It can be identified with the cyclic symmetry of the extended
Dynkin diagram of A(1)
N and so also with a (discrete rotational) symmetry of the fundamental affine alcove P+ (AN ). At the level of the horizontal algebra AN , the generating
(1)
automorphism of sOut(AN ) yields

N

i =: .
: := [1 , 2 , . . . , N ] 2 , 3 , . . . , N ,
(9)
i=1
It results in a
rescaling5
2in
cn e N+1 cn ,
of the eigenvalues of the scalars (3)(4):

2i
K e N+1 K,
(10)
(1)
ZN +1 = sOut(AN )
on cn and K is the same

which matches (7) exactly. Thus the action of
as that of ZN +1 Out(REAq (AN )).
It seems worth noticing at this point that (8) is just an algebraic analog of the ordinary
geometrical left regular action:

g g,
g SU(N + 1), Z SU(N + 1)
(11)
= ZN +1 ,
in which Z(SU(N + 1)) is the centre of the group SU(N + 1). The ZN +1 is isomorphic to
the latter andconsequentlydividing out its action is tantamount to taking the quotient
SU(N + 1)/ZN +1 with respect to the natural action (11).
In order to directly relate both sets of automorphisms in a physically meaningful manner, we need to modify the former definition of the brane in the algebraic setup. In [7],
string theory branes in the SU(N + 1) group manifold were associated with irreducible
0 . Here, we propose to take instead the
representations of REAq (AN ) of the kind R
(N + 1)-fold direct sum of irreducible representations:
B :=
N

RLL
(12)
L=0
4 We give the name to those outer automorphisms of g which are not automorphisms of the horizontal
algebra g.
5 The remark can be verified using the explicit realisation of the Casimir operators given in [13].
603
as describing the brane assigned to the weight . Notice that all the summands in (12)
return the same Casimir eigenvalues, which is necessary for the correct geometrical picture.
L = R L+1 and R 0 R N +1 )
The advantage of working with (12) is that (recall that R
B = B ,
(13)
i.e., the -conjugate branes are isomorphic (the same Casimir eigenvalues and the same
algebra of functions). In this setting, is indistinguishable from and so it acquires an
analogous geometrical meaning, which shall be crucial in our subsequent considerations.
In particular, it allows us to define the orbifold by dividing out the action of ZN +1 generated
by . Orbifolding will then glue branes with -conjugate Casimir eigenvalues.
For some non-generic weights (fixed point weights), we may have D FP = FP
(with ND+1 =: n N), i.e., D BFP = BFP . When evaluated on these fixed point
weights FP , some of the Casimirs vanish. This will lead to certain new interesting phenomena discussed in Section 3.2.
As an aside, we note here that as we introduce (12) the requirement of consistency with
the BCFT data for the pre-orbifolding WZW models necessitates a reformulation of the
original quantum matrix model, which we now take in the general form (cf. [7]):

T0
Trq I + covUq (AN ) (M) .
(14)
N +1
Here covUq (AN ) (M) denotes an unspecified function of the matrix variable M, covariant
under the action of Uq (AN ) described in [7]. The action (14) is to be evaluated on representations of the kind (12).
Equipped with the above results, we can now try to perform constructions of branes in
simple current WZW orbifold models, e.g., on the manifold SU(N + 1)/ZN +1 .
eff
:=
Sq,A
N
2.1. Simple current WZW orbifolds

The orbifolding procedure for WZW models was laid out in [14], which we follow
closely in this preparatory part. It leads to a formulation of string theory on a quotient
(1)
space G/ , with (here) a subgroup of the group sOut(AN ) of strictly affine automor(1)
phisms of AN .
The action of on the set of primaries, labelled by weights P+ (g), decomposes
this set into orbits and we take [] to label all boundary conditions associated to by the
action. A consistent orbifolding procedure [15] leaves only orbits whose associated weight
labels satisfy
Q () = 0,
(15)
where Q () is the monodromy charge of defined with respect to the generator of

the simple current group . In the case of interest, i.e., for G = SU(N + 1), and for
(1)
generating ZN +1 = sOut(AN ), one obtains
1
nn mod 1.
Q () =
N +1
N
n=1
(16)
604
Recall that Q () appears also in the REA through the central term K, defined in (4).
Some of the orbits may have fixed points, i.e., there may be weights stabilised by subgroups of . We reserve the symbol S for the latter. As indicated by the BCFT,
there are n = |S | distinct branes (sometimes called resolved or fractional branes) for a
given . They shall be labelled by []x , x 1, n. An inspection of the (resolved) cylinder
partition functions subject to boundary conditions defined by resolved branes reveals that
their tensions are reduced by the factor 1/n with respect to those of parent branes. We shall
discuss the algebraic picture of this reduction in Section 3.3.
The set of boundary fields considered in [14] consisted of multiplets of horizontal de
scendants L, of boundary primaries, each carrying a pair of weight labels (, )
associated with the respective boundary conditions, and a group-theoretic (multi-)index L
identifying the field within the multiplet. In what follows, we limit our discussion to the
case = for the sake of simplicity. This means we shall consider only those fields
which, after orbifolding, give rise to the untwisted sector of the theory and describe open
strings with both ends on the same parent brane of label . Thus we shall not discuss
operators corresponding to strings stretching between non-coincident branes on the covering space. For the quantum matrix model description of the latter in the pure group case,
see [16].
Prior to closing the section, we provide one more piece of the BCFT data essential to
orbifolding: the action of the elements of sOut on the boundary fields L, . It was defined
in [14] as
L, := (1)Q (L) L, ,
(17)
with Q (L) the so-called simple current charge, a quantity closely related to the mon (L) mod 1. In the matrix model, this action is mimicked by
odromy charge: Q (L) = Q
(8) accompanied by (13).
We may now further analyse the structure of REAq (AN ) in search of the patterns listed
above.
3. The quantum orbifold

The present section contains an explicit construction of the orbifold. We also demonstrate the appearance of new quantum geometries. The latter describe fixed point D-branes
of the simple current ZN +1 -orbifold of the WZW model of type AN .
In orbifolding, the Casimir K given by (4) proves to play a prominent role. It was used
in (8) to implement the appropriate action of the ZN +1 on the coordinate variables Mij .
Upon restricting to the set of the corresponding K-invariant monomials within REAq (AN ),
we recover the orbifold algebra in the form:6
REAq (AN )/ZN +1 = span
Mi1 j1 Mi2 j2 MiN+1 jN+1 /IRE,detq ,
(18)
6 The discussion below is confined to the maximal Z

N +1 -orbifold for the sake of concreteness only but the
conclusions drawn are of general validity.
605
where IRE,detq is the ideal generated by (1) and (5).

Determining the geometry of D-branes in the quantum-algebraic setup requires two
pieces of data: the algebra of functions and its representation theory. Hence it is legitimate to considerin addition to (8)the action of K on the representation space (module)
V ( v ) of the brane B (see (12)):
K v = R (K)v = e2iQ () v .
(19)
In this case, projecting onto the subset of K-invariant representations is tantamount to

imposing the condition:
Q () = 0 mod 1.
(20)
The latter means that the monodromy charge (16) of the representation considered should
vanish, in accordance with the BCFT results (cf. Section 2.1).
Altogether, the Casimir K is seen to realise the action of ZN +1 Out(REAq (AN )) both
at the level of the algebra and that of the associated representation theory, sending us from
the original modelas given by REAq (AN ) and Rep(REAq (AN ))to the orbifold one
upon dividing out its left action on the two components.
In order to be able to assess the structure encoded in (18), we distinguish within the set
of -invariants the following subsets:7

N +1
, c1 M N , . . . , cN Mij , cN 1 c1 Mij , . . . , cN
M
(21)
1 Mij , {I},
where M p symbolically denotes an arbitrary monomial in Mij s of degree p as in (18).
The above decomposition shows that the orbifold geometry associated to a generic
weight (on which some cn = 0) is as in the pre-orbifolding case, i.e., it is generated by
Mij s. For these weights, we may take the definition of the brane on the orbifold to be as in
(12), supplemented by the constraint (20). The fixed point geometries, on the other hand,
display an altogether different structure, analysed at some length in the next section.
It ought to be remarked at this point that the orbifolding described cuts the range of
admissible affine weight labels to P+ (AN )/ZN +1 , with the division determined by the
action of (equivalent to in this case) together with the monodromy projection (20). The
resulting set spans an N -dimensional solid D within P+ (AN ) with vertices defined by the
vectors (i are the fundamental weights of AN ):
0,
(1 + N ),
3

k ,
N +1
N
k=1
i ,
2
(j + j +1 )
3
for i 1, N and j 1, N 1.
3.1. New geometries in REAq (AN )/ZN +1
The non-generic features of orbifold geometries come to the fore whenever there is a
fixed point of the action of D on P+ (AN ) for some 1 D < N + 1. In this case some
7 The different subsets in (21) consist of multiplets of tensor operators of various spins (from 0 through N + 1)
under the action of the brane symmetry algebra: Uq (AN )V Uq (AN AN )R .
606
of the cn s vanish, as follows from (10): since the N + 1 independent Casimir eigenvalues pick up the N + 1 independent phase factors under the simple current rotation the
corresponding weight FP will be stabilised by D iff the Casimirs whose eigenvalues do
change under the latter vanish. The same conclusion holds for due to (13). Among the
nontrivially stabilised weights, there might be a distinguished central one, N+1 [1, 1, . . . , 1]
(Dynkin label notation), sitting in the geometric centre of the N -simplex of the fundamental affine alcove. It is an actual element of P+ (AN ) whenever N + 1 | (N + 1 divides )
and has a vanishing monodromy iff N + 1 | N2 .
We want to gain some insight into the structure of fixed point algebras. To these
ends, consider a restriction of the algebra of ZN +1 -invariants to the representation (12)
of REAq (AN ) associated with a given weight and characterised by the following set of
Casimir eigenvalues:
Ck1 , Ck2 , . . . , CkK = 0,
Ci = 0 for all i
/ {k1 , k2 , . . . , kK },
(22)
where Cj := cj |B , j 1, N . We then make the following

Claim 1. The coordinate subalgebra for the locus of (22) is generated by independent
monomials of degree n such that
n = gcd(N + 1, kj )j 1,K .
(23)
Here is the proof of our claim.

Proof. Let n be as in (23). Given (22), one can form polynomials of degree N + 1 (ZN +1 pj
p
p
invariants), K
j =1 ckj M , with M an arbitrary monomial of degree p and n|p. Thus one
can effectively use the monomials M p to generate the orbifold algebra. Next, one takes
products (M p )l with l > 0 the smallest number such that (M p )l = M p1 detq M for some
non-zero p1 . Then p1 < p and n|p1 . Continuing the procedure, one concludes, after a
finite number S of steps, that it is the monomials M pS that generate the algebra, with
pS | N + 1. One can subsequently perform analogous reduction with respect to each of the
non-vanishing Casimirs, whereby one finally reaches the desired conclusion (23). 2
The monomials M n are invariant under Zn ZN +1 generated by D . The subgroup
Zn is the stabiliser of within ZN +1 . In the distinguished case of the central weight (and
exclusively in that case), all cn = 0 and we recover the full ZN +1 as the stabiliser, with the
algebra of monomials of degree N + 1 as the generating ones. We expect the emergence of
essentially new quantum geometries whenever the number (23) is greater than one. These
quotient geometries are described by proper subalgebras of REAq (AN ) andas argued
belowhave a reduced representation theory. We plan to dwell on this subject in a separate
publication.
607
3.2. The fixed point resolution

Here, we shall deal with one of the fixed point weights = FP which defines the
group Zn as above. Following the definition (12), we consider
n1 D1

B =
RL+xD
L ,
(24)
x=0 L=0
where the equality holds due to D = (nD = N + 1). The space of functions on
L+lD
the fixed point brane is generated by the monomials M n . Recalling that R
=
0 e N+1 + n , we therefore obtain, symbolically, (R L+lD )n = (R L )n . From this point of

R
view, the x-dependence in (24) drops out completely on passing to the orbifold geometry.
Consequently, we could choose
2iL
b =
2il
D1

RLL
(25)
L=0
as a possible definition of the (fractional) brane. The latter satisfies ( b )n = (b )n (see

(13)), necessary for the geometrical meaning of , but cannot account for the emergence
of fractional branes required by the BCFT. Fortunately, the naive definition (25) is not the
only possibility: one can construct branes through cross products [14] discussed below.
This leads to perfect agreement with BCFT results.
Thus, we introduce a certain cross product extension of the algebra generated by the
L+(x+1)D
M n . Given the action of the Zn -generator: D RL+xD
= RL
, we note that the
L
y
D
original module (24) splits into -eigenspaces and so we define the B -brane:
y
B := Py B ,
(26)
with the projectors Py given by

Py :=
n1
1 2iym/n Dm
e
,
n
Py Pz = y,z Py ,
y, z 0, n 1.
(27)
m=0
The crucial property of B is B = B , from which D B = B follows. We may

equivalently rewrite (26) as
y
B = b py ,
(28)
with py the yth irreducible, one-dimensional representation of Zn . Consequently, we define the space of functions on the yth brane to be generated by
M n Py (K),
where Py (K) is as in (27) but with replaced by K, and py (KD ) = e2iy/n .
(29)
608
The crossed product structure is completed by defining the product of the generators
(29). The latter is inherited from the following crossed product8

Mij KDm Mi j KDm = Mij KDm L Mi j KD(m+m ) ,
consistent with the group-like character of K, and reads (symbolically)

n
M1 Px (K) M2n Py (K) = x,y M1n M2n Px (K).
(30)
The algebra thus defined, together with its representation theory, describes the so-called
fractional orbifold branes, discussed in the present context, e.g., in [14]. Their existence
was inferred, in particular, from the structure of the fixed point boundary OPE algebra,
encoding the crossed product extension of the relevant algebra of functions, as argued
in [14]. Finally, let us add that the emergence of the crossed product extension of the
algebra of functions at fixed points of the orbifold action is a general feature of matrix
models of orbifold geometries, reflecting the existence of fractional branes [18].
3.3. Orbifold brane tensions
An important and nontrivial test of the construction presented above is the computation
of tensions of the orbifold branes within the framework developed. In so doing, we follow the scheme advanced in the original papers [7], that is we compute the tension of a
brane labelled by the weight [] D P+ (AN ) by evaluating the first term of the effecy
tive action (14) on the associated representation: B[FP ] for a fixed point weight = FP ,
off for an off-fixed-point one (clearly, the orbifolding of off-fixed-point modor B =: B[]
ules produces modules isomorphic to the original (pre-orbifolding) ones). Beside giving
rise to the aforementioned crossed product extension of the fixed point algebra of ZN +1 invariants, the former yields the correct result for the tension of resolved fixed point branes
y
B[FP ] :

T0
1
T0
Trq By I =
D Trq R 0 I = EFP , y Zn , n = |SFP |,
(31)
FP
[FP ]
N +1
N +1
n
falling in perfect agreement with the BCFT data (cf. Section 2.1). It is supplemented by an
off :
equally satisfactory result for generic (off-fixed-point) branes B[]

T0
T0
Trq Boff I =
(N + 1) Trq R 0 I = E ,
(32)
[]
N +1
N +1
an immediate consequence of the lack of an internal mechanism of spectrum reduction in
this last case.
4. An example: branes on RPq3
Below, we detail the particularly simple example of the antipodal Z2 -orbifold of the
quantum matrix model for su (2), 4N . The model is exemplary in that it develops
8 Cf. [17].
609
precisely along the lines discussed in the previous sections, hence we restrict ourselves
here to its q-geometric interpretation, relating our results to some well-established mathematical constructs of [19] and [20]. The BCFT properties of the corresponding branes were
discussed in [14,21].
4.1. The setup
Our starting point shall be the RE algebra suggested in [7] as giving a plausible compact
description of the quantum D-brane geometry on WZW group manifolds. Since we aim at
describing the stringy geometry of the antipodal orbifold of SU(2) we will focus on the
REA generated by the operator entries of a matrix M parametrised as

iq 1/2 [2]q M1
M4 iq 2 M0

,
M=
(33)
iq 3/2 [2]q M1
M4 + iM0
0 R 1 , consistently with (12). Here,
where M s are to be realised in the representation R
= . Given the parametrisation, we may write down the additional q-determinant

constraint:
!
r 2 I = detq M = M42 + M02 q 1 M1 M1 qM1 M1 ,
(34)
in a natural way, with r 2 interpreted as the radius squared of the group manifold, S3q ,
which
we set roughly proportional to the level of the underlying WZW model (cf. [7]),
r ( + 2). The RE (1) now takes a manageable (component) form (i {0, 1, 1}):
[M4 , Mi ] = 0,
M0 M1 = qM1 (qM0 iM4 ),
M1 M0 = q(qM0 iM4 )M1 ,
M1 M1 M1 M1 = M0 (M0 iM4 ).
(35)
The above algebra, further constrained by (34), is easily seen to reproduce (after diagonalising the central element M4 and some trivial rescalings) the celebrated Podles spheres,
S2q,c , with the parameter c essentially determined by the value of M4 (cf. [19]).
Upon embedding the REA in Uq (su(2)), we arrive at K = eiH and the quantum Z2 automorphism (8) reads
: M M ,
{4, 0, 1, 1},
(36)
just as required for the automorphism to have the interpretation of an antipodal map on S3q .
The monodromy operator (4) acts trivially only on integer-spinned irreducible representations.
We are now in a position to explicitly construct the building blocks of a q-matrix model
antipod
-orbifold of the su (2) WZW model at 4N .
for the Z2
4.2. The orbifold
antipod
In the light of the general results of Section 3.2, the coordinate algebra of the Z2
orbifold of the stringy manifold S3q is the algebra of quadratic monomials in the generators
M of REAq (su(2)), Z2 -extendedin the case of 4N at the unique central fixed
610
point, FP = /2. The algebra of Z2 -invariants, defining a quantum manifold which could
be called the real quantum projective 3-plane, RPq3 , is easily verified to be generated by
the mutually independent operators:

2
, M0 M 1 ,
O RPq3 = span I, M02 , M12 , M1

2
M1 M0 , M4 , M4 M0 , M4 M1 , M4 M1 /IRE,detq ,
(37)
where IRE,detq is the ideal defined by a set of relations deriving directly from (34) and (35).
The relations are not very illuminating and shall therefore be left out.
An important feature of the ensuing algebra is the central character of M42 which shall
consequently be used to label inequivalent irreducible representations. As a consequence,
for M42 = 0 (or = /2) the algebra (37) is generated by {M0 , M1 , M1 } so that we
recover quantum 2-spheres, as discussed in Section 3.1 and indicated by the BCFT. At the
fixed point, on the other hand, where the original RE and (34) simplify,
M0 M1 = q 2 M1 M0 ,
M1 M0 = q 2 M0 M1 ,
M1 M1 M1 M1 = M02 ,
M02 q 1 M1 M1 qM1 M1 = r 2 I,
(38)
the orbifold algebra reducesupon choosing the rescaled generators:

T , T )
(P , R, R,

[2]q
1
q
q
2 2 2 2 2
:= 2 2
M , q M1 , q M1 ,
M0 M1 ,
M1 M0 ,
q r [2]q 0
[2]q
[2]q
(39)
to the RPq2 algebra found in [20], generalised here to q a phase. According to the general discussion of Section 3.2, we have two inequivalent RPq2 -branes corresponding to
the two inequivalent irreducible representations of the stabiliser group Z2 (with D = 1 of
Section 3.2 here).
We remark at this point thatas follows from the restriction to irreducible representations of vanishing monodromy (i.e.in the case at handinteger spin), effected by
(19)it is only the integer spin branes that compose RPq3 .
4.2.1. Space of functions
Here, we perform a more scrupulous comparison of the two descriptions of the fixed
point geometry: the BCFT and the algebraic one. The resolved fixed point boundary fields
(with label, say, [FP ]+ ) can be grouped as [14]
/4

[ ] ,[ ]
(2K + 1),
span L FP + FP +
4N .
(40)
K/2=0
In the matrix model, on the other hand, we are free to take arbitrary monomials in the
T , T , with the (even integer) spin of any such monomial determined
generators P , R, R,
by its overall degree in the original coordinates Mj , j {0, 1, 1}. Restricting to the basis
monomials [20], ordered according to their spin 2s 2N,
P m R sm , P n R sn , P l T R sn1 , P n T R sn1 ,
m 0, s, n 0, s 1,
(41)
611
we should obtain as many independent operators as there are even-spinned primaries,

which clearly is not the case9 unless we truncate the spin of the monomial generators of
the equatorial RPq2 algebra as s /4. The dimension of the space of monomials reduces
accordingly,
S>/4

s=0

/4
1
+1
+2 ,
(4s + 1)
(4s + 1) =
2 2
2
(42)
s=0
to yield precisely the BCFT-dictated number of geometric degrees of freedom on the side
of the matrix model. With this last observation we conclude our exposition of the simple
example of quantum orbifold geometry.
5. Summary
In the present paper, we advanced a quantum-algebraic model for curved noncommutative D-brane geometries on simple current orbifolds of the SU(N + 1) WZW
manifolds. Following the original ideas of [7], we were able to consistently encode the
BCFT data, regarded here as the ultimate foundation of all our constructions, in a simple
algebraic framework based on the central idea of quantum group symmetry, as suggested
by the structure of the CFT itself. Our study of the relevant reflection equation algebras revealed significant structural similarities between the latter and the BCFT of the corresponding WZW models. The similarities, present both at the level of the respective representation
(1)
theories (Out(AN ) versus Out(REAq (AN ))) and at the level of the algebra (the (resolved)
orbifold boundary OPE algebra versus the (cross-product-extended) REAq (AN )/ZN +1 ),
provide nontrivial evidence for a close relationship between the Uq (AN )-related algebras
REAq (AN ) and the stringy geometries defined by the BCFT. An important feature of this
relationship is its naturalness. It follows, in particular, from the clear geometric meaning
assigned to REAq (AN ), which, in turn, induces a simple realisation of classical-type symmetries (whether continuous or discrete as in the present context) of the non-commutative
manifolds defined by REAq (AN ). One of the important manifestations of the relationship
is a straightforward identification of the CFT monodromy charge in the quantum-algebraic
setup. Last but not least, the explicit physical results on tensions of orbifold branes, falling
in perfect agreement with the BCFT data, lend further support to our choice of the algebraic
structure deforming the underlying BCFT.
It is perhaps worth mentioning thatin addition to the calculations carried out explicitly
in the paperwe are able within the present framework to test quantum stability of the
brane configurations (12) as well as examine the ensuing inter-brane excitations using the
techniques developed in [16]. So far our results seem to indicate towards stability against
9 The origin of the mismatch traces back to the failure of the original Drinfeld twisting procedure to render
the ensuing q-deformed OPE algebra associative in the entire range of spin parameters of the theory. It turns out
that, as was remarked already in [7], the associativity breaks down as early as = E(/4), half-way between the
north pole of S3q and its equator.
612
decay and reproduce an asymptotically correct picture of the lightest open strings stretched
between distant branes.
We believe that our work, corroborating the choice of quantum algebras, REAq (g),
used to encode essential BCFT data, is itself a step towards a better understanding of
the non-classical structure of curved brane worldvolumes within the framework of noncommutative geometry. Clearly, it ought to be regarded as a starting point for further
investigation aimed at incorporating, e.g., gauge fluctuations in the present rigid geometric picture. Attempts in this direction [16] yield encouraging results, althoughclearly
a fully satisfactory algebraic description of quantum WZW branes is still missing. We hope
to return to these issues in near future.
Acknowledgements
We gratefully acknowledge useful discussions with Harald Steinacker and Satoshi
Watamura. J.P. would like to thank Satoshi Watamura for his kind hospitality during J.P.s
visit at the Tohoku University, and the String Theory Group at Queen Mary London for
their interest in this project. R.R.S. would also wish to thank the Theoretical Physics
Group at Kings College London, and in particular Andreas Recknagel, for a stimulating
atmosphere and interest in the project during R.R.S.s stay at Kings College London.
References
[1] V. Schomerus, Lectures on branes in curved backgrounds, Class. Quantum Grav. 19 (2002) 57815847,
hep-th/0209241.
[2] E. Witten, Nonabelian bosonization in two-dimensions, Commun. Math. Phys. 92 (1984) 455.
[3] D. Gepner, E. Witten, String theory on group manifolds, Nucl. Phys. B 278 (1986) 493549.
[4] C. Bachas, M.R. Douglas, C. Schweigert, Flux stabilization of D-branes, JHEP 0005 (2000) 048, hepth/0003037;
J. Paweczyk, SU(2) WZW D-branes and their noncommutative geometry from DBI action, JHEP 0008
(2000) 006, hep-th/0003057;
J.M. Maldacena, G.W. Moore, N. Seiberg, D-brane instantons and K-theory charges, JHEP 0111 (2001) 062,
hep-th/0108100;
P. Bordalo, S. Ribault, C. Schweigert, Flux stabilization in compact groups, JHEP 0110 (2001) 036, hepth/0108201.
[5] A.Yu. Alekseev, V. Schomerus, D-branes in the WZW model, Phys. Rev. D 60 (1999) 061901, hepth/9812193;
G. Felder, J. Frhlich, J. Fuchs, C. Schweigert, The geometry of WZW branes, J. Geom. Phys. 34 (2000)
162190, hep-th/9909030.
[6] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4 (1989) 251;
J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581;
J.L. Cardy, D.C. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett. B 259 (1991)
274278;
D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries, Nucl. Phys.
B 372 (1992) 654682.
[7] J. Paweczyk, H. Steinacker, Matrix description of D-branes on 3-spheres, JHEP 0112 (2001) 018, hepth/0107265;
J. Paweczyk, H. Steinacker, A quantum algebraic description of D-branes on group manifolds, Nucl. Phys.
B 638 (2002) 433458, hep-th/0203110.
613
[8] J. Fuchs, B. Schellekens, C. Schweigert, From Dynkin diagram symmetries to fixed point structures, Commun. Math. Phys. 180 (1996) 3998, hep-th/9506135.
[9] L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Strings on orbifolds, Nucl. Phys. B 261 (1985) 678686;
L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Strings on orbifolds (II), Nucl. Phys. B 274 (1986) 285314.
[10] E. Sklyanin, Boundary conditions for integrable quantum systems, J. Phys. A 21 (1988) 2375.
[11] P.P. Kulish, R. Sasaki, C. Schwiebert, Constant solutions of reflection equations and quantum groups,
J. Math. Phys. 34 (1993) 286304, hep-th/9205039;
P.P. Kulish, E.K. Sklyanin, Algebraic structures related to reflection equations, J. Phys. A 25 (1992) 5963
5976, hep-th/9209054;
P.P. Kulish, R. Sasaki, Covariance properties of reflection equation algebras, Prog. Theor. Phys. 89 (1993)
741762, hep-th/9212007.
[12] L.D. Faddeev, N.Yu. Reshetikhin, L.A. Takhtajan, Quantization of Lie groups and Lie algebras, Leningrad
Math. J. 1 (1990) 193225 (in Russian);
L.D. Faddeev, N.Yu. Reshetikhin, L.A. Takhtajan, Algebra Anal. 1 (1989) 178206 (in Russian).
[13] D. Arnaudon, M. Bauer, Polynomial relations in the centre of Uq (sl(N )), hep-th/9310030.
[14] K. Matsubara, V. Schomerus, M. Smedbck, Open strings in simple current orbifolds, Nucl. Phys. B 626
(2002) 5372, hep-th/0108126.
[15] A.N. Schellekens, S. Yankielowicz, Extended chiral algebras and modular invariant partition functions, Nucl.
Phys. B 327 (1989) 673;
A.N. Schellekens, S. Yankielowicz, Simple currents, modular invariants and fixed points, Int. J. Mod. Phys.
A 5 (1990) 29032952.
[16] J. Paweczyk, H. Steinacker, Algebraic brane dynamics on SU(2): excitation spectra, JHEP 0312 (2003)
010, hep-th/0305226.
[17] S. Majid, Examples of braided groups and braided matrices, J. Math. Phys. 32 (1991) 3246;
S. Majid, Foundations of Quantum Group Theory, Cambridge Univ. Press, Cambridge, 1995.
[18] N. Kim, S.-J. Rey, M(atrix) theory on an orbifold and twisted membrane, Nucl. Phys. B 504 (1997) 189213,
hep-th/9701139;
A. Konechny, A. Schwarz, Compactification of M(atrix) theory on noncommutative toroidal orbifolds, Nucl.
Phys. B 591 (2000) 667684, hep-th/9912185;
A. Konechny, A. Schwarz, Moduli spaces of maximally supersymmetric solutions on noncommutative tori
and noncommutative orbifolds, JHEP 0009 (2000) 005, hep-th/0005174.
[19] P. Podles, Quantum spheres, Lett. Math. Phys. 14 (1987) 193202.
[20] P.M. Hajac, R. Matthes, W. Szymanski, Quantum real projective space, disc and sphere, math.QA/0009185.
[21] N. Couchoud, D-branes and orientifolds of SO(3), JHEP 0203 (2002) 026, hep-th/0201089.
Central elements of the elliptic Zn monodromy

matrix algebra at roots of unity
Wen-Li Yang a,b , Alexander Belavin c , Ryu Sasaki d
a Institute of Modern Physics, Northwest University, Xian 710069, PR China
b Department of Mathematics, University of Queensland, Brisbane, QLD 4072, Australia
c Landau Institute for Theoretical Physics, Chernogolovka 142432, Russia
d Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
Received 29 July 2004; accepted 22 December 2004

Abstract
The central elements of the algebra of monodromy matrices associated with the Zn R-matrix are
n , where N and n are
studied. When the crossing parameter w takes a special rational value w = N
positive coprime integers, the center is substantially larger than that in the generic case for which
the quantum determinant provides the center. In the trigonometric limit, the situation corresponds
to the quantum group at roots of unity. This is a higher rank generalization of the recent results by
Belavin and Jimbo.
PACS: 03.65.Fd; 05.30.-d
Keywords: Integrable models; Elliptical quantum group and quantum group; Central elements
1. Introduction
The algebra of the monodromy matrix (or YangBaxter algebra) is generated by noncommutative matrix elements of the L-operator L(u) satisfying a quadratic RLL relation
R12 (u v)L1 (u)L2 (v) = L2 (v)L1 (u)R12 (u v),
E-mail address: wenli@maths.uq.edu.au (W.-L. Yang).
doi:10.1016/j.nuclphysb.2004.12.033
(1.1)
W.-L. Yang et al. / Nuclear Physics B 710 [PM] (2005) 614628
615
where L1 (u) = L(u) 1, L2 (u) = 1 L(u) and R(u) is the R-matrix, a solution of quantum YangBaxter equation
R12 (u v)R13 (u)R23 (v) = R23 (v)R13 (u)R12 (u v).
(1.2)
We adopt the standard notation: Rij (u) is an embedding operator of R-matrix in the tensor
space Cn Cn , which acts as identity on the factor spaces except for the ith and j th
ones. If R(u) is a rational R-matrix, the algebra after a proper specialization is a Yangian
algebra [1]. For a trigonometric R-matrix, the algebra is closely connected with a quantum
group [2,3]. If R(u) is an elliptic R-matrix, the algebra gives rise to a Sklyanin algebra
[4,5].
In this paper, we will address the issue of the center of an elliptic algebra associated
with the Zn R-matrix [6] when the crossing parameter takes rational values. In the trigonometric limit, it corresponds to a quantum group at roots of unity. So here we call this case
the roots of unity case as its trigonometric limit. As is well known, for the trigonometric algebra (as a degenerate algebra of the elliptic algebra) when the crossing parameter
(deformation parameter) is in roots of unity case the center is much larger than that of
the generic case [7,8]. Recently for the simplest case of the Z2 R-matrix (or eight-vertex
R-matrix) [9], the structure of the center in the roots of unity case was clarified [10].
Here, we will consider the question with a generic case n 2.
The paper is organized as follows. In Section 2 we give the definition of the elliptic algebra and introduce appropriate notation. We formulate the fundamental Boltzmann weights
of the An1 type interaction-around-a-face (IRF) model (or the SOS model) and give the
corresponding face-vertex correspondence relation in Section 3. In Section 4, we construct
the symmetric fusion procedure for the R-matrices, Boltzmann weights and intertwining
vectors, which are useful to handle the center of the algebra. Finally, we give the center of
the elliptic algebra with the Zn R-matrix when the crossing parameter is in the roots of
unity case in Section 5. Section 6 is for conclusions.
2. The algebra of monodromy matrices

Let us fix an integer n (n 2) and a complex number such that Im( ) > 0. Introduce
the following elliptic functions

a
(u, ) =
exp 1 (m + a)2 + 2(m + a)(u + b) ,
b
(2.1)
m=
1
(j ) (u) =
1
2
j
n
1
(u, n ),
(j ) (u) = exp{ 1 u} (j ) (u).
(u) =
2
1
2
(u, ),
(2.2)
(2.3)
616
Among them the -function1 satisfies the following identity

(u + x) (u x) (v + y) (v y) (u + y) (u y) (v + x) (v x)
= (u + v) (u v) (x + y) (x y).
Let {ei | i = 1, 2, . . . , n} be the orthonormal basis of the vector space Cn such that
ei , ej = ij . Let R(u) End(Cn Cn ) be the Zn R-matrix given by

R(u) =
(2.4)
Rijkl (u)Eik Elj ,
i,j,k,l
in which Eij is the matrix with elements (Eij )lk = j k il . The coefficient functions are
[12,13]

h(u) (ij ) (u+w)
if i + j = k + l mod n,
kl
Rij (u) = (ik) (w) (kj ) (u)
(2.5)
0
otherwise,
where a complex parameter w is called the crossing parameter. We have set
n1 (j )
j =0 (u)
.
h(u) = n1
(j )
j =1 (0)
The R-matrix satisfies the quantum YangBaxter equation (1.2), unitarity and crossingunitarity relations [12]. It should be remarked that our R-matrix given in (2.4) and (2.5) is
the same as that of Ref. [13], and that our
R-matrix R(u) is related to the R-matrix S(u) of
e 1 u t1 t2
Richey and Tracy [12] by R12 (u) = n S12 (u), where ti denotes the transposition in
the ith space. Consequently, the R-matrix enjoys the following property
Lemma 1. The R-matrix given in (2.4) and (2.5) satisfies
(+)
R(w) = M P2 ,
(2.6)
(+)
where M is a nondegenerate matrix and P2 is the two-body symmetrization operator

(+)
given by P2 = 12 (1 + P12 ) in terms of the permutation operator P12 : P12 (ei ej ) =
ej e i .
This lemma tells that

Ker R(w) = the antisymmetric subspace of Cn Cn .
(2.7)
The above property enables us to construct symmetric fusion of the R-matrix in higher
tensor space [14] and the intertwining vector in Section 4.
1 Our -function is the -function (u) [11]. It has the following relation with the Weierstrassian -function
1

2
nq 2n
1 .
if denoted it by w (u): w (u) e1 u (u), 1 = 2 ( 16 4
n=1 1q 2n ) and q = e
617
Definition 1. The algebra of monodromy matrices A is an associative algebra generated

by the matrix elements of the L-operator Lij (u), i, j = 1, . . . , n with the defining relations (1.1).
One can introduce a natural coproduct : A A A and counit : A C

Lij (u) =
Lik (u) Lkj (u),
(Lij ) = ij ,
k
making A a bi-algebra. Certain realization of the algebra A can be expressed in terms

of the Zn Sklyanin algebra [4,5] which is an elliptical generalization of quantum group
Uq (gln ).
In this paper, we investigate the center of the algebra A, i.e., the set of the elements
commuting with all the elements of A. For a generic value of the crossing parameter w,
the so-called quantum determinant of the L-operator (detq L(u)) generates the center of the
algebra A [15]. However, it is known [10] that A for the eight-vertex model case (n = 2)
the situation is quite different when w = N2 , N is a positive odd integer. In this case, in
addition to the quantum determinant detq L(u), there exist much more central elements
in A. For the generic case of n 2, the same phenomenon has been observed for A in
the trigonometric limit [7,8]. We shall study the extra central elements of A with a generic
n 2 in the roots of unity case: w = Nn , where N and n are positive coprime integers.
Hereafter, we restrict the crossing parameter to the roots of unity case w = Nn , unless
otherwise stated.
3. A(1)
n1 face model and face-vertex correspondence
In this section, we formulate the fundamental Boltzmann weights of the An1 type SOS
face model and give the face-vertex correspondence relation.
An ordered pair (a, b) (a, b Cn ) is called admissible if
b a = ei ,
i = 1, . . . , n,
(3.1)
and denoted by a b. A path p = (a (0) , a (1) , . . . , a (l) ) (Cn )l+1 is called an l-path
from a to b if a (0) = a, a (l) = b, and all pairs (a (i1) , a (i) ) for i = 1, . . . , l are admissible.
Definition 2. An ordered pair (a, b) (a, b Cn ) is called l-admissible if there exists an
l
l-path from a to b, and denoted by a b.
Let W ac db u denote the Boltzmann weight corresponding to a configuration a, b, d,

c Cn around a face, ordered clockwise from the NW corner. We call that W ac db u is

admissible if (a, b), (b, d), (a, c), (c, d) are all admissible. By definition nonadmissible
weights are set to 0. The nonvanishing weights are given as follows

a
a + e
(u + w)
u =
W
(3.2)
,

a + e a + 2e
(w)
618

a + e
u = (a w u) , = ,
a + e + e
(a w)

(u)
(a w + w)
a
a + e
u =
, = ,
W

a + e a + e + e
(w) (a w)
a = a, e e , , = 1, . . . , n.

a
a + e
(3.3)
(3.4)
(3.5)
The weights given in (3.2)(3.4) satisfy the star-triangle relation (or dynamical Yang
Baxter equation) [13]

a b

u + v W f g u W b c v
W

f g
e d
g d
g

a g a b g c

W
v W
u W
u+v .
=
(3.6)
f e
g c
e d
g
For a generic Cn , define

i = , ei ,
i = , ei ,
ei = ei
1
ek ,
n
(3.7)
ij = i j = i j ,
i, j = 1, . . . , n.
(3.8)
Let us introduce an intertwining vectoran n-component column vector ,+ej (u) whose
kth element is
,+ej (u) = (k) (u nw j ).
(k)
(3.9)
One can easily verify the following properties of the intertwining vector from the definition (3.9) and Eq. (3.7).
Lemma 2. The intertwining vector defined by (3.9) satisfies the following relation
+ej ,+ei +ej (u) = ,+ei (u + w nij w),
i, j = 1, . . . , n,
,+ei (u + n) = ,+ei (u).
(3.10)
(3.11)
The intertwining vector (3.9) is different from that of Jimbo et al. [13] by the definition
of the -function given in (2.3). The present definition is preferred because of its periodicity
(3.11) which will play an important role in Section 5. From the definition of the Boltzmann
weights (3.2)(3.4) and using the face-vertex correspondence obtained in [13], we have the
following relation.
Theorem 1. The intertwining vector satisfies
R12 (u v),+ei (u) +ei ,+ei +ej (v)

+ ek

u v +ek ,+ei +ej (u) ,+ek (v).
W
=
+ ei + ei + ej
(3.12)
Then the YangBaxter equation of the Zn R-matrix R(u) (1.2) is equivalent to the startriangle relation (3.6).
619
4. Fusion procedure
The fusion procedure of the R-matrix and the intertwining vector is essential in handling
the central elements of the algebra A.
4.1. The generic w case
In this subsection, we consider the case that the crossing parameter w takes a generic
value. Let us introduce operators L1...N (u) and R1...N (u) by

R1...N (u) R1...N;0 (u) = R10 u + (N 1)w RN 0 (u),
(4.1)

L1...N (u) = L1 u + (N 1)w LN (u).
(4.2)
We consider L1...N (u) (R1...N (u)) as an operator acting on the so-called auxiliary space
V1 VN , Vi = Cn with the entries belonging to A (with the entries of the operator
R1...N (u) acting on an additional n-dimensional space denoted by 0th tensor space). Define
an operator SN acting on V1 VN as follows
SN =
N
1

N

Rij (j i)w ,
(4.3)
i=1 j =i+1
where both indices grow from left to right. For example, if N = 3, then
S3 = R12 (w)R13 (2w)R23 (w).
Proposition 1. Ker(SN ) is an invariant subspace in V1 VN of the operators
L1...N (u) and R1...N (u).
Proof. Using the YangBaxter equation (1.2) and RLL relation (1.1), one can derive
SN R1...N (u) = RN ...1 (u)SN ,
SN L1...N (u) = LN ...1 (u)SN ,
where RN ...1 (u) and LN ...1 (u) differ from R1...N (u) and L1...N (u) by permutation of all
factors to the opposite order. Hence, R1...N (u) Ker(SN ) Ker(SN ), L1...N (u) Ker(SN )
Ker(SN ). 2
Now, let us consider the subspace Ker(SN ). Define
W0 =
N
1

Ker Ri,i+1 (w) V1 VN .
(4.4)
i=1
Using the YangBaxter equation (1.2), one can show that the factors of SN can be permuted
into the order such that Ri,i+1 (w) for a given i = 1, . . . , N 1 is in the rightmost position.
For example, if N = 3 and i = 1,
S3 = R12 (w)R13 (2w)R23 (w) = R23 (w)R13 (2w)R12 (w).
620
So, Ker(SN ) W0 . Comparing the dimensions of the subspaces in the trigonometric limit,
one can derive that Ker(SN ) = W0 for the generic crossing parameter w. However, when
the crossing parameter w is in the roots of unity case w = Nn , the subspace Ker(SN ) will
be larger than W0 .
It follows from Proposition 1 that W0 is invariant under L1...N (u) and R1...N (u). Thus
these operators can be defined in the quotient space W0 which is defined as follows
W0 = V1 VN /W0 .
(4.5)
This also enables one to construct the symmetric fusion of the R-matrix and the L-operator
in higher tensor space [14]. Moreover, the property (2.7) and the definitions of W0 (4.4)
and W0 (4.5) imply that
W0 = the symmetric subspace of V1 VN ,
(N + n 1)!
(N + n 1)!
dim W0 =
,
dim(W0 ) = nN
.
N !(n 1)!
N !(n 1)!
(4.6)
(4.7)
(+)
Define a complete symmetrizer PN which is given in terms of the element of N -body

permutation group denoted by PN as follows

1
(+)
PN =
(4.8)
P .
N!
P PN
Proposition 1 and Eq. (4.6) lead to the following fusion properties.

Lemma 3.
(+)
(+)
(+)
PN R1...N;0 (u) = PN R1...N;0 (u)PN ,
(4.9)
PN(+) L1...N (u) = PN(+) L1...N (u)PN(+) .
(4.10)
(s)
(s)
Hence, one can define the symmetric fused R-matrix R1...N and L-operator L1...N acting
in W0 as follows
(s)
(+)
(+)
R(s)
1...N (u) R1...N;0 (u) = PN R1...N;0 (u)PN ,
(s)
(+)
(+)
L1...N (u) = PN L1...N (u)PN .
(4.11)
(4.12)
For an N-admissible pair

of its N -paths from a to b: p = (a, a +
(a, b), we choose
one
N
ei 1 , a + ei 1 + ei 2 , . . . , a + N
e
),
b
=
a
+
e
i
l=1 l
l=1 il . Let us introduce a fused intertwining
vector p;a,b (u) as follows

i ,...,i
N1 ;a,b N (u) = a,a+ei1 u + (N 1)w a+N1 ei ,a+N ei (u),
l=1 l
l=1
l

(+) i1 ,...,iN
p;a,b (u) = PN N ;a,b (u) .
Noting Lemma 2, we obtain
(4.13)
(4.14)
Proposition 2. The fused intertwining vector given in (4.14) satisfies

ip ,...,ip
1
(+)
N
1
p;a,b (u) =
N ;a,b
(u) ,
P
N! N
621
(4.15)
P PN
where P is an element of PN :

1
2 N
.
P=
p1 p 2 p N
(4.16)
The proof of this proposition is relegated to Appendix A.

Then, we have
Corollary 1. The fused intertwining vector given in (4.14) is independent of the choice of
the N -path from a to b for any N -admissible pair (a, b).
Because of the path independence, we hereafter denote the fused intertwining vector
given in (4.14) by N ;a,b (u) for any N -admissible pair (a, b) instead of p;a,b (u). From
the construction of the fused intertwining vector, one easily has

Span N ;a,b (u) | for all N -admissible pairs (a, b) = W0 .
(4.17)
Now we establish an equivalence between the symmetric fused R-matrix (4.11) and
Boltzmann weights considered in [16] through the fused intertwining vector.
Proposition 3. The fused intertwining vector satisfies the following fused face-vertex correspondence relation
(s)
R1...N;0 (u v)N ;a,b (u) b,b+ej (v)

a a + ek
u
v
N ;a+ek ,b+ej (u) a,a+ek (v).
WN 1
=
b b + ej
(4.18)
Here the fused Boltzmann weights WN 1

Cn
a b
u
d c
N
are nonvanishing only for a configuration

1
a, b, c, d
such that a b, b c, a d, d c and the nonvanishing weights are
given as follows

a b
u
WN 1
d c
N 1

(u + j w) (u + c w b w) = (c w b w + w)

=
(4.19)
,
(w)
(a w + w)
j =1
when b a = e and c d = e .
Proof. Suppose (a, a (1) , . . . , a (N 1) , b) is an N -path from a to b. By the fundamental
face-vertex correspondence relation of Theorem 1, one can derive
622

R1...N;0 (u v) a,a (1) u + (N 1)w a (N1) ,b (u) b,b+ej (v)

a
a (1)
u
v
+
(N
1)w
=
W
a (1) a (2)
a (1) ,...,a (N1) ,b

(1)
a (2)
a
u v + (N 2)w
W
a (2) a (3)

(N 1)

(N 2)

a
a (N 1)
a
b
u v
u
v
+
w
W
W

b
b + ej
a (N 1) b

a (1) ,a (2) u + (N 1)w b ,b+ej (u) a,a (1) (v).
(4.20)
The definition of the fused intertwining vector (4.14) leads to

a,a (1) u + (N 1)w a (N1) ,b (u) = N ;a,b (u) mod W0 .
(4.21)
(+)
Multiplying Eq. (4.20) by PN
from the left and using Proposition 1, we have

(+)
LHS = PN R1...N;0 (u v) N ;a,b (u) + W0 b, b+ej (v)
= R(s)
1...N ;0 (u v)N ;a,b (u) b, b+ej (v),

(4.22)

a (1)
u v + (N 1)w
RHS =
a (2)
a (1) a (2) ,...,a (N1) ,b

(N 1)

b
a
u v
W
b
b + ej

(+)
PN a (1) ,a (2) u + (N 1)w b ,b+ej (u) a,a (1) (v)

a (1)
a (1)
a
W
a (1)

a (2) ,...,a (N1) ,b
W

WN 1
a
b
a
W
a (1)
a (N 1)
b
a (1)
b + ej

a (1)
u v + (N 1)w
a (2)

u v (1)
N ;a ,b+ej (u) a,a (1) (v)

b
b + ej

u v (1)
N ;a ,b+ej (u) a,a (1) (v).

(4.23)
We have used Corollary 1 in the second last equality and Lemma 2.2 of Ref. [16] in the

last equality. The expression of the fused Boltzmann weight WN 1 da bc u was given in [16].
So, we complete the proof. 2
4.2. The roots of unity case
From now on, we come back to the case that the crossing parameter takes value of roots
of unity w = Nn . It was shown in [10] that for the eight-vertex (or Z2 ) case the subspace
623
Ker(SN ) is larger than W0 by an additional 2-dimensional subspace in roots of unity

case. We shall show that Ker(SN ) is also larger than W0 by an additional n-dimensional
subspace for the generic Zn case.
Let us introduce n vectors ai (u) V1 VN , i = 1, . . . , n,
ai (u) = N ;a,a+N ei (u),
for a generic vector a Cn .
We denote the n-dimensional subspace spanned by the vectors {ai (u) | i
(4.24)
= 1, . . . , n} by W
which is independent of the choice of a.

Proposition 4. When the crossing parameter w =
n
N,
Ker(SN ) = W W0 .
Proof. Using the YangBaxter equation (1.2), one can permute the order of the factors
of SN in such way

SN = R1N (N 1)w R1N 1 (N 2)w R12 (w).
ii
Acting it on the vector N
;a,a+N ei (u) defined in (4.13) (there exists a unique N -path for
the N -admissible pair (a, a + N ei ) such as (a, a + ei , a + 2ei , . . . , a + N ei )) and using
Theorem 1, one can derive
i...i
SN N
;a,a+ei (u)

i...i
= R1N (N 1)w R1N 1 (N 2)w R12 (w)N
;a,a+N ei (u)

= R1N (N 1)w R12 (w)a,a+ei u + (N 1)w
a+(N 1)ei ,a+N ei (u)

(N w) ((N 1)w) (w)
=
a+(N 1)ei ,a+N ei u + (N 1)w
N
( (w))

a,a+ei u + (N 2)w a+(N 2)ei ,a+(N 1)ei (u).
(4.25)
i...i
The fact that N
;a,a+N ei (u) Ker(SN ) follows from the identity (N w) = (n) = 0. Noting that
i...i
ai (u) = N
;a,a+N ei (u) mod W0 ,
and the fact W0 Ker(SN ), one can show W Ker(SN ). Evidently W W0 = {0}. It
follows that W W0 Ker(SN ). Then it remains to prove that
dim Ker(SN ) = dim W0 + dim W .
(4.26)
However, the dimensions of these subspaces

do not depend on the modulus of the torus.
Taking the trigonometric limit + 1 , Eq. (4.26) has been already proved in [8].
So we complete the proof. 2
Propositions 1 and 4 show that W is invariant under the fused R-matrix R1...N (u) de(s)
End(W Cn )
fined in (4.11) and the L-operator L1...N (u) defined in (4.12). Let R(u)
(s)
(s)
and (u) End(W ) A be the restrictions of the fused operators R1...N (u) and L1...N (u)
(s)
624
on W , explicitly,
(s)
n
(s)
j =1
n
R1...N (u)ai (u) = R(u)

a (u) =
L1...N (u)ai (u) = (u)ai (u) =
j
R ji (u)a (u),
ij (u)a (u),

R ji (u) End Cn ,
j
i (u) A.
(4.27)
(4.28)
j =1
5. The center
Let us compute some of the fused Boltzmann weights with a specially chosen configuration a, b, c, d Cn . By using Proposition 3 and the identity (N wi ) = (ni ) = 0,
we can derive for the crossing parameter w = Nn

a
a + e
u
WN 1
a + N ei a + N ei + e
N 1

(u + j w) (u + N wi ) = (a w + N wi )

=
,
(w)
(a w + w)
j =1
in which the quasi-periodicity of the -function

(u + N w) = (u + n) = (1)n (u),
N w = n,
is used. This leads to a simple expression of the fused Boltzmann weight

N

(u + j w)
a
a + e

u =
WN 1
.
a + N ei a + N ei + e
(w)
(5.1)
(5.2)
j =1
In order to compute the restricted R-matrix R(u)

defined in (4.27), we evaluate the action
(s)
of the fused R-matrix R1...N defined in (4.11) on the vector ai
(s)
R1...N;0 (u v)ai (u) a+N ei ,a+N ei +ej (v)

N

(u v + kw)
i
(u) a,a+ej (v).
=
a+e
j
(w)
(5.3)
k=1
Using the shift property of the fundamental intertwining vector ,+ei (u) (3.10), one can
write the fused intertwining vector explicitly as

(+)
ai (u) = PN a,a+ei u + (N 1)w a,a+ei u + (N 1)w nw

a,a+ei u + (N 1)w (N 1)nw .
(5.4)
The periodicity (3.11) enables one to further derive the following relation
a,a+ej (u + N w) = a,a+ej (u + n) = a,a+ej (u).
(5.5)
625
The above equation and (5.4) imply the following relations

ai (u + nw) = ai (u),
(5.6)
i
(u) = ai (u + w).
a+e
j
(5.7)
Then
Proposition 5. When the crossing parameter is in roots of unity case w =
ai (u) satisfies the following relation
i
(u) = ai (u + w) = ai (u).
a+e
j
n
N,
the vector
(5.8)
Proof. The first equality is just Eq. (5.7). We are to prove the second equality ai (u + w) =
ai (u). The expression of ai (u) given in (5.4) and Eq. (5.5) imply that
ai (u + N w) = ai (u + n) = ai (u).
(5.9)
Due to the fact that N and n are positive coprime integers, one can choose an integer l such
that
1 + lN = 0 mod n.
(5.10)
The above property and relations (5.6) and (5.9) allow one to derive that

ai (u + w) = ai (u + w + lN w) = ai u + (1 + lN )w = ai (u).
Hence, we complete the proof.
(5.11)
2
(s)
Using Proposition 5, the action of the fused R-matrix R1...N on the vector ai can be
given explicitly as
Proposition 6.
(s)
R1...N;0 (u v)ai (u) a+N ei ,a+N ei +ej (v)

N

(u v + kw)
=
ai (u) a,a+ej (v).
(w)
(5.12)
k=1
Now, we calculate the exchange relation among the fused L-operators {lk (u) | k, l =
j
1, . . . , n} and the generators of A {Li (u) | i, j = 1, . . . , n}. Firstly, one can derive the
following relation by the RLL relation (1.1)
R1...N ;0 (u v)L1...N (u)L0 (v) = L0 (v)L1...N (u)R1...N;0 (u v).
(5.13)
Proposition 1 allows one to restrict the above equation on W0 , namely in terms of the
fused R-matrix (4.11) and L-operator (4.12)
(s)
(s)
(s)
(s)
R1...N;0 (u v)L1...N (u)L0 (v) = L0 (v)L1...N (u)R1...N;0 (u v).
(5.14)
Proposition 4 and the fact that W W0 allow one further to restrict the equation on W .
Finally, we have our main result.
626
Theorem 2. When the crossing parameter is in roots of unity case w = Nn , the elements
{ij (u) | i, j = 1, . . . , n} defined in (4.28) are the central elements of A, namely,
i

j (u), Lkl (v) = 0.
(5.15)
Proof. For a generic a Cn , we can introduce an n n matrix A with matrix elements Aij :
(i)
Aij = a,a+ej (u),
i, j = 1, . . . , n.
One can verify that det(A) = 0. It means that

Span a,a+ej (u) | j = 1, . . . , n = Cn .
(5.16)
j
This fact allows one to define matrix elements L i (u) A from the L-operator L(u) as
follows
L(u)a,a+ei (u) =
n
j
L i (u)a,a+ej (u).
(5.17)
j =1
The conditions w =
n
N
and (3.11) allow one to derive the following relation
a+N ei ,a+N ei +ej (u) = a,a+ej (u + N w) = a,a+ej (u).
(5.18)
Acting both sides of Eq. (5.14) on the vector al (u) a,a+ej (v) and using the definitions
(4.28) and (5.17), we have
LHS =
n

(s)
R1...N;0 (u v)ak (u) a,a+ei (v) lk (u)L ij (v) ,
k,i=1
(s)
(s)
RHS = L0 (v)L1...N (v)R1...N;0 (u v)al (u) a,a+ej (v).

Proposition 6 and the relation (5.18) imply that we can further derive that
N
n
(u v + kw)

LHS =
ak (u) a,a+ei (v) lk (u)L ij (v) ,
(w)
k=1
k,i=1
N
n
(u v + kw)

RHS =
ak (u) a,a+ei (v) L ij (v)lk (u) .
(w)
k=1
k,i=1
Hence [lk (u), L ij (v)] = 0. Eq. (5.17) and nondegeneracy of the matrix (Aij ) enable us to
derive that [lk (u), Lij (v)] = 0. Therefore, we complete the proof. 2
6. Conclusions
In this paper we have investigated the extra central elements (differing from the socalled quantum determinant detq (L(u))) of the algebra A of monodromy matrices associated with the Zn R-matrix when the crossing parameter takes special value w = Nn . In the
627
trigonometric limit, this case corresponds to the quantum group Uq (gln ) at roots of unity.
Our results show that in even elliptic case the center of A is also extended and is generated
by n2 elements {ij (u)} in addition to detq (L(u)). For the special case of n = 2, our result
recovers that of [10].
Acknowledgements
W. -L. Yang is supported from the Japan Society for the Promotion of Science and Australian Research Council. A. Belavin is partially supported by grant RFBR-04-02-16027
and NATO grant PST.CLG.979008. He also would like to thank Yukawa Institute for Theoretical Physics, Kyoto University and Professor R. Sasaki for kind hospitality during his
visit to Kyoto.
Appendix A. Proof of Proposition 2

The property (3.10) allows one to derive the following relation
+ej ,+ei +ej (u) = ,+ei (u + w),
if i = j.
(A.1)
For a vector B V1 Vm , the corresponding symmetrized vector B (s) is given as

1
(s)
(+)
B = Pm (B) =
P (B) .
m!
P Pm
The terms obtained by nontrivial permutations of B are called descendants. It is easy to

show that the symmetrized vector satisfies
B (s) = Pm(+) (P B),
P Pm .
(A.2)
Hereafter we shall always keep Eqs. (A.1) and (A.2) in mind, because they will play an
important role to proof the proposition.
ip1 ...ipm
i1 ...im
m
Let us denote the vector m;a,a+
by i1 . . . im , and the set of {m;a,a+
|P
m
l=1 eil
l=1 eil

Pm } by i1 . . . im . We shall prove the proposition by induction.
Using the relation (A.1), we can prove that Eq. (4.15) holds for the case of N = 2.
Suppose that (4.15) holds for the case of 2 < N . We are to prove it is satisfied for
N + 1.
(1) In the case of iN +1
/ {i1 , . . . , iN },

i1 ...iN+1
(+) i1 . . . iN iN +1

(A.3)
,
PN +1
..
N +1;a,a+ N+1
l=1 eil
.
628
.
where .. stands for the descendants. Using the induction hypothesis and (A.1) we
can derive that (A.3) is equal to

i1 . . . iN +1 iN 1
i1 . . . iN iN +1 i1 . . . iN +1 iN
(+)
PN +1
+
+
+
..
..
..
.
.
.

(+)
= PN +1 i1 . . . iN +1 .
So, (4.15) holds for N + 1.
(2) If iN +1 {i1 , . . . , iN }, without loss of generality, let us suppose that iN +1 = iN .
Let [iN ] denote the set {il = iN | l = 1, . . . N 1}. Let M be the number of elements in the set. Let [iN ]
l , l = 1, . . . , M, denote the element of the set. Similarly
to the first case, we have

i1 ...iN iN
(+) i1 . . . iN iN
N+1
PN +1
..
N +1;a,a+ l=1 eil
.

i1 . . . iN [iN ]
i1 . . . iN [iN ]
M
1
(+) i1 . . . iN iN
= PN +1
+
+ +
..
..
..
.
.
.

(+)
= PN +1 i1 . . . iN iN .
Then (4.15) holds for N + 1.
Therefore (4.15) holds for any positive integer N 2. Finally we complete the proof of
Proposition 2.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
V.G. Drinfeld, Sov. Math. Dokl. 32 (1985) 254.

V.G. Drinfeld, Proc. ICM Berkeley (1986) 798.
M. Jimbo, Lett. Math. Phys. 10 (1985) 63.
E.K. Sklyanin, Functional Anal. Appl. 17 (1983) 273.
B.Y. Hou, H. Wei, J. Math. Phys. 30 (1989) 2750.
A.A. Belavin, Nucl. Phys. B 180 (1981) 189.
C. De Concini, V.G. Kac, Colloque Dixmier (1989) 471.
V. Tarasov, Commun. Math. Phys. 158 (1993) 459.
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1986.
A. Belavin, M. Jimbo, hep-th/0208224.
E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, fourth ed., Cambridge Univ. Press, Cambridge,
2002.
M.P. Richey, C.A. Tracy, J. Stat. Phys. 42 (1986) 311.
M. Jimbo, T. Miwa, M. Okado, Lett. Math. Phys. 14 (1987) 123;
M. Jimbo, T. Miwa, M. Okado, Nucl. Phys. B 300 (1988) 74.
I.V. Cherednik, Functional Anal. Appl. 19 (1985) 77.
B.Y. Hou, Y.K. Zhou, J. Phys. A 23 (1990) 1147.
M. Jimbo, A. Kuniba, T. Miwa, M. Okado, Commun. Math. Phys. 119 (1988) 543.
CUMULATIVE AUTHOR INDEX B701B710
Abrashkin, A.
Accomando, E.
Actis, S.
Agam, O.
Aguilar-Saavedra, J.A.
Ahl Laamara, R.
Ait Ben Haddou, M.
Alexander, S.
Aliev, T.M.
Altschul, B.
lvarez, E.
Andersson, T.
Anguelova, L.
Antoniadis, I.
Antusch, S.
Arakawa, G.
Arikawa, M.
Arkani-Hamed, N.
Armoni, A.
Arnaudon, D.
Asano, M.
Auzzi, R.
B703 (2004) 320

B706 (2005) 325
B703 (2004) 3
B704 (2005) 407
B706 (2005) 204
B702 (2004) 163
B702 (2004) 163
B704 (2005) 231
B709 (2005) 115
B705 (2005) 593
B709 (2005) 381
B707 (2005) 509
B702 (2004) 269
B707 (2005) 215
B705 (2005) 239
B709 (2005) 296
B702 (2004) 380
B709 (2005) 3
B702 (2004) 37
B704 (2005) 490
B705 (2005) 33
B701 (2004) 207
Babu, K.S.
Bajnok, Z.
Balog, J.
Bandelloni, G.
Banerjee, R.
Barbieri, A.
Barbieri, R.
Barger, V.
Barnes, E.
Bartels, J.
Baseilhac, P.
Baseilhac, P.
Bashir, A.
B708 (2005) 511

B702 (2004) 448
B709 (2005) 329
B703 (2004) 499
B708 (2005) 434
B703 (2004) 391
B703 (2004) 127
B705 (2005) 71
B702 (2004) 131
B706 (2005) 391
B705 (2005) 605
B709 (2005) 491
B709 (2005) 307
0550-3213/2005 Published by Elsevier B.V.

doi:10.1016/S0550-3213(05)00120-3
Bashiry, V.
Battye, R.A.
Becirevic, D.
Becker, K.
Becker, K.
Becker, M.
Becker, M.
Bedford, J.
Behrndt, K.
Blanger, G.
Belavin, A.
Belhaj, A.
Belitsky, A.V.
Bellucci, S.
Benson, D.
Berenstein, D.
Bernabu, J.
Bernreuther, W.
Bertoldi, G.
Bertoldi, G.
Bettelheim, E.
Bialas, P.
Biggio, C.
Bigi, I.I.
Biswas, T.
Blmlein, J.
Bolognesi, S.
Bolognesi, S.
Bonciani, R.
Bonciani, R.
Bonciani, R.
Bonciani, R.
Bordalo, P.
Botella, F.J.
Boudjema, F.
Branco, G.C.
Brandhuber, A.
B709 (2005) 115

B705 (2005) 384
B705 (2005) 339
B704 (2005) 231
B707 (2005) 43
B702 (2004) 207
B704 (2005) 231
B706 (2005) 100
B708 (2005) 45
B706 (2005) 411
B710 (2005) 614
B702 (2004) 163
B708 (2005) 115
B707 (2005) 303
B710 (2005) 371
B702 (2004) 49
B701 (2004) 87
B706 (2005) 245
B701 (2004) 481
B709 (2005) 522
B704 (2005) 407
B704 (2005) 208
B703 (2004) 199
B710 (2005) 371
B708 (2005) 317
B708 (2005) 467
B701 (2004) 207
B709 (2005) 522
B701 (2004) 121
B702 (2004) 359
B702 (2004) 364
B706 (2005) 245
B710 (2005) 189
B706 (2005) 204
B706 (2005) 411
B706 (2005) 204
B706 (2005) 100
630
Brandhuber, A.
Brignole, A.
Brodsky, S.J.
Broncano, A.
Brugues, J.
Brustein, R.
Buchel, A.
Buchel, A.
Burgess, C.P.
Butti, A.
B706 (2005) 150

B701 (2004) 3
B703 (2004) 333
B705 (2005) 269
B710 (2005) 117
B709 (2005) 391
B707 (2005) 56
B708 (2005) 451
B706 (2005) 71
B704 (2005) 527
Calabrese, P.
Callan Jr., C.G.
Cmara, P.G.
Campbell, B.A.
Campostrini, M.
Caracciolo, S.
Casteill, P.-Y.
Castro-Villarreal, P.
Chacko, Z.
Chan, C.-T.
Cheng, J.-F.
Cherkis, S.A.
Chesterman, M.
Chetyrkin, K.G.
Chiang, C.-W.
Choi, K.-S.
Cirelli, M.
Constantin, D.
Cook, P.P.
Corcella, G.
Cornalba, L.
Cottrant, A.
Cvetic, M.
Cvetic, M.
Cvetic, M.
Czakon, M.
B709 (2005) 550

B701 (2004) 180
B708 (2005) 268
B709 (2005) 419
B703 (2004) 454
B707 (2005) 458
B707 (2005) 303
B704 (2005) 111
B706 (2005) 53
B708 (2005) 99
B701 (2004) 54
B702 (2004) 49
B703 (2004) 400
B710 (2005) 499
B705 (2005) 71
B708 (2005) 194
B708 (2005) 215
B706 (2005) 221
B705 (2005) 111
B705 (2005) 363
B710 (2005) 189
B706 (2005) 411
B708 (2005) 45
B708 (2005) 381
B709 (2005) 241
B710 (2005) 485
DAdda, A.
Dai, J.
Damgaard, P.H.
Dasgupta, K.
Dasgupta, K.
de Azcrraga, J.A.
de Florian, D.
de Holanda, P.C.
Delfino, G.
Delgadillo-Blando, R.
Denner, A.
Dereli, T.
Derkachov, S..
de Vega, H.J.
DHoker, E.
DHoker, E.
B707 (2005) 100

B708 (2005) 72
B707 (2005) 509
B702 (2004) 207
B704 (2005) 231
B706 (2005) 181
B704 (2005) 387
B702 (2004) 307
B707 (2005) 381
B704 (2005) 111
B706 (2005) 325
B705 (2005) 60
B708 (2005) 115
B707 (2005) 529
B710 (2005) 58
B710 (2005) 83
Dijkstra, T.P.T.
Dimopoulos, S.
Donini, A.
Drissi, L.B.
Dudas, E.
Dukelsky, J.
Dunning, C.
Dutta, S.
B710 (2005) 3
B709 (2005) 3
B710 (2005) 402
B702 (2004) 163
B708 (2005) 3
B707 (2005) 421
B702 (2004) 481
B704 (2005) 75
Eilers, J.
Elias, V.
Enciso, A.
Enkhbat, Ts.
Eto, M.
Evslin, J.
B704 (2005) 279

B703 (2004) 413
B707 (2005) 553
B708 (2005) 511
B701 (2004) 247
B701 (2004) 207
Farakos, K.
Feinberg, J.
Feng, B.
Feoli, A.
Fernndez-Martnez, E.
Ferrara, S.
Ferroglia, A.
Ferroglia, A.
Ferroglia, A.
Feverati, G.
Finkel, F.
Flournoy, A.
Font, A.
Fotopoulos, A.
Fox, P.J.
Franke, V.A.
Frappat, L.
Frishman, Y.
Fucito, F.
Fugel, F.
Fukuma, M.
B705 (2005) 92
B705 (2005) 403
B701 (2004) 334
B705 (2005) 577
B710 (2005) 402
B704 (2005) 154
B701 (2004) 121
B702 (2004) 364
B703 (2004) 3
B702 (2004) 495
B707 (2005) 553
B706 (2005) 127
B703 (2004) 177
B710 (2005) 309
B706 (2005) 53
B703 (2004) 333
B704 (2005) 490
B703 (2004) 320
B703 (2004) 518
B702 (2004) 333
B703 (2004) 293
Gabrielli, E.
Gardner, C.L.
Garousi, M.R.
Gavela, M.B.
Gehrmann, T.
Gehrmann, T.
Georgiev, L.S.
Geyer, B.
Ghoshal, D.
Gibbons, G.W.
Girelli, F.
Giribet, G.
Giribet, G.
Giudice, G.F.
Giudice, G.F.
Giudice, G.F.
B710 (2005) 139

B707 (2005) 278
B705 (2005) 212
B705 (2005) 269
B703 (2004) 147
B706 (2005) 245
B707 (2005) 347
B704 (2005) 279
B710 (2005) 577
B708 (2005) 381
B708 (2005) 411
B701 (2004) 481
B709 (2005) 522
B706 (2005) 455
B706 (2005) 487
B709 (2005) 3
Giudice, G.F.
Giusto, S.
Giusto, S.
Giusto, S.
Gonzlez-Lpez, A.
Gonzlez-Sprinberg, G.A.
Gorsky, A.
Graham, N.
Grassi, P.A.
Grassi, P.A.
Grassi, P.A.
Grazzini, M.
Grinza, P.
Grisaru, M.
Grosse, H.
Guadagnini, E.
Gutjahr, P.
B710 (2005) 511

B701 (2004) 357
B704 (2005) 181
B710 (2005) 425
B707 (2005) 553
B701 (2004) 87
B702 (2004) 37
B707 (2005) 233
B702 (2004) 189
B702 (2004) 269
B706 (2005) 549
B704 (2005) 387
B702 (2004) 495
B708 (2005) 317
B707 (2005) 145
B703 (2004) 391
B704 (2005) 583
Haan, O.
Hagiwara, K.
Hagiwara, K.
Hall, L.J.
Hamazaki, T.
Harnik, R.
Hasebe, K.
Hatsuda, M.
He, Y.-H.
Hebecker, A.
Heckman, J.
Heinesch, R.
Hiller, J.R.
Ho, P.-M.
Honkonen, J.
Horvth, I.
Huang, C.-S.
Huang, C.-S.
Huiszoon, L.R.
Huitu, K.
Hurth, T.
B705 (2005) 563

B701 (2004) 405
B704 (2005) 75
B703 (2004) 217
B706 (2005) 488
B706 (2005) 372
B709 (2005) 94
B703 (2004) 277
B701 (2004) 334
B701 (2004) 273
B701 (2004) 180
B706 (2005) 245
B703 (2004) 333
B708 (2005) 99
B707 (2005) 493
B710 (2005) 464
B701 (2004) 54
B702 (2004) 109
B710 (2005) 3
B710 (2005) 139
B704 (2005) 56
Ibez, L.E.
Ichinose, I.
Ichinose, S.
Imbimbo, C.
Intriligator, K.
Isidori, G.
Isral, D.
Ivanov, E.
Ivanov, E.
Izquierdo, J.M.
B708 (2005) 268

B709 (2005) 296
B710 (2005) 255
B704 (2005) 181
B702 (2004) 131
B705 (2005) 339
B710 (2005) 529
B704 (2005) 154
B707 (2005) 69
B706 (2005) 181
Jaffe, R.L.
Jenkins, E.
Jiang, J.
B704 (2005) 552

B705 (2005) 269
B702 (2004) 109
631
Jiang, J.
Jin, L.G.
Jockers, H.
B705 (2005) 71
B705 (2005) 3
B705 (2005) 167
Kaiser, A.
Kamimura, K.
Kanamori, I.
Kawamoto, N.
Kawamura, H.
Kawano, T.
Khalil, S.
Khemani, V.
Kim, S.-Y.
Kimura, Y.
King, S.F.
Kishimoto, I.
Klinkhamer, F.R.
Knauf, A.
Knauf, A.
Kniehl, B.A.
Kobayashi, T.
Komarova, M.V.
Konishi, K.
Kono, Y.
Korchemsky, G.P.
Kulish, P.P.
Kurak, V.
Kuramoto, Y.
Kutasov, D.
B706 (2005) 325

B703 (2004) 277
B707 (2005) 100
B707 (2005) 100
B708 (2005) 467
B710 (2005) 577
B710 (2005) 139
B707 (2005) 233
B705 (2005) 504
B709 (2005) 94
B705 (2005) 239
B707 (2005) 3
B709 (2005) 171
B702 (2004) 207
B704 (2005) 231
B702 (2004) 333
B704 (2005) 3
B707 (2005) 493
B701 (2004) 207
B703 (2004) 293
B708 (2005) 115
B709 (2005) 578
B701 (2004) 497
B702 (2004) 380
B702 (2004) 369
Lam, F.
Langacker, P.
Laporta, S.
Larson, D.T.
Lazzarini, S.
Lechtenfeld, O.
Lechtenfeld, O.
Lechtenfeld, O.
Le Doussal, P.
Lee, H.M.
Lee, J.-C.
Lee, S.
Leineweber, T.
Li, C.S.
Li, T.
Li, T.
Li, T.
Liao, W.
Liivat, H.
Lima-Santos, A.
Links, J.
Lipatov, L.N.
Liu, J.J.
Liu, J.T.
B701 (2004) 334

B709 (2005) 241
B704 (2005) 349
B706 (2005) 372
B703 (2004) 499
B704 (2005) 154
B705 (2005) 477
B707 (2005) 69
B701 (2004) 409
B705 (2005) 152
B708 (2005) 99
B705 (2005) 296
B706 (2005) 245
B705 (2005) 3
B702 (2004) 109
B705 (2005) 71
B709 (2005) 241
B702 (2004) 307
B702 (2004) 346
B701 (2004) 497
B702 (2004) 481
B706 (2005) 391
B705 (2005) 3
B707 (2005) 56
632
Liu, T.
Liu, X.
Livine, E.R.
Loide, R.-K.
Lpez, J.A.
Louis, J.
L, H.
Lubicz, V.
Lugo, A.R.
Lunghi, E.
Lst, D.
Lyakhovich, S.L.
B709 (2005) 241

B704 (2005) 510
B708 (2005) 411
B702 (2004) 346
B703 (2004) 177
B705 (2005) 167
B709 (2005) 47
B705 (2005) 339
B701 (2004) 299
B704 (2005) 56
B706 (2005) 3
B703 (2004) 419
Matre, D.
Manashov, A.N.
Mann, R.B.
Marandella, G.
Markov, V.
Marquet, C.
Marshakov, A.
Martinelli, G.
Martins, M.J.
Mastrolia, P.
Mastrolia, P.
Mastrolia, P.
Mastrolia, P.
Mathur, S.D.
Mathur, S.D.
Matias, J.
Matone, M.
Matsui, T.
Matsuo, Y.
Mawatari, K.
Maybury, D.W.
Mazzanti, L.
McCartor, G.
McInnes, B.
McKeon, D.G.C.
McLoughlin, T.
McLoughlin, T.
Meggiolaro, E.
Menotti, P.
Menotti, P.
Mescia, F.
Meusburger, C.
Migliozzi, P.
Miquel-Espanya, C.
Miramontes, J.L.
Miwa, A.
Mognetti, B.M.
Morales, J.F.
Morales, J.F.
Morel, A.
Movahed, M.S.
B703 (2004) 147

B708 (2005) 115
B703 (2004) 413
B708 (2005) 215
B709 (2005) 267
B705 (2005) 319
B709 (2005) 267
B705 (2005) 339
B705 (2005) 521
B701 (2004) 121
B702 (2004) 359
B702 (2004) 364
B706 (2005) 245
B701 (2004) 357
B710 (2005) 425
B706 (2005) 71
B709 (2005) 522
B709 (2005) 296
B707 (2005) 3
B701 (2004) 405
B709 (2005) 419
B705 (2005) 477
B703 (2004) 333
B709 (2005) 213
B703 (2004) 413
B701 (2004) 180
B702 (2004) 86
B707 (2005) 199
B707 (2005) 321
B709 (2005) 465
B705 (2005) 339
B706 (2005) 569
B710 (2005) 402
B706 (2005) 181
B702 (2004) 419
B703 (2004) 293
B707 (2005) 458
B703 (2004) 518
B707 (2005) 303
B704 (2005) 208
B707 (2005) 405
Murayama, A.
Murayama, H.
Murayama, H.
Murayama, H.
Mussardo, G.
B710 (2005) 255

B701 (2004) 207
B706 (2005) 53
B706 (2005) 372
B705 (2005) 548
Nagata, K.
Nagi, J.
Nakamura, S.
Nakayama, Y.
Nakayama, Y.
Nakazawa, N.
Nalimov, M.Yu.
Nebot, M.
Nesterenko, V.V.
Niarchos, V.
Niccoli, G.
Nicolosi, M.
Nishimura, M.
Nitta, M.
Nomura, Y.
B707 (2005) 100

B704 (2005) 475
B709 (2005) 192
B708 (2005) 345
B709 (2005) 522
B702 (2004) 3
B707 (2005) 493
B706 (2005) 204
B705 (2005) 577
B710 (2005) 309
B707 (2005) 381
B708 (2005) 3
B701 (2004) 103
B701 (2004) 247
B703 (2004) 217
Oriti, D.
Ortiz, G.
Oshimo, N.
Ots, I.
Ovchinnikov, A.A.
B708 (2005) 411

B707 (2005) 421
B701 (2004) 395
B702 (2004) 346
B703 (2004) 363
Paccetti Correia, F.
Pakman, A.
Palla, L.
Pankiewicz, A.
Papazoglou, A.
Parisen Toldin, F.
Pasipoularides, P.
Passarino, G.
Passera, M.
Paston, S.A.
Paweczyk, J.
Pelissetto, A.
Pelissetto, A.
Penati, S.
Petersson, B.
Phong, D.H.
Phong, D.H.
Plehn, T.
Poghossian, R.
Polychronakos, A.P.
Pomarol, A.
Pons, J.M.
Pope, C.N.
Pope, C.N.
Popov, A.D.
Porod, W.
B709 (2005) 141

B710 (2005) 529
B702 (2004) 448
B704 (2005) 583
B705 (2005) 152
B704 (2005) 527
B705 (2005) 92
B703 (2004) 3
B703 (2004) 3
B703 (2004) 333
B710 (2005) 599
B707 (2005) 458
B709 (2005) 550
B705 (2005) 477
B704 (2005) 208
B710 (2005) 58
B710 (2005) 83
B706 (2005) 455
B703 (2004) 518
B705 (2005) 457
B703 (2004) 127
B703 (2004) 537
B708 (2005) 381
B709 (2005) 47
B705 (2005) 477
B704 (2005) 56
Pradisi, G.
Prezas, N.
Prokhvatilov, E.V.
Pukhov, A.
B708 (2005) 3
B710 (2005) 309
B703 (2004) 333
B706 (2005) 411
Quadri, A.
Quandt, M.
Quevedo, F.
Quirs, M.
B706 (2005) 549

B707 (2005) 233
B706 (2005) 71
B703 (2004) 199
Raby, S.
Ragoucy, E.
Rahimi Tabar, M.R.
Ramgoolam, S.
Rasmussen, J.
Rasmussen, J.
Rattazzi, R.
Ravindran, V.
Raya, A.
Reffert, S.
Remiddi, E.
Remiddi, E.
Remiddi, E.
Remiddi, E.
Remiddi, E.
Ribeiro, G.A.P.
Rigolin, S.
Riotto, A.
Riva, V.
Robaschik, D.
Rodrguez, M.A.
Rojo, J.
Romanino, A.
Romanino, A.
Rombouts, S.
Rossi, A.
Rhl, W.
Rupp, C.
Russo, J.G.
Russo, R.
B704 (2005) 3
B704 (2005) 490
B707 (2005) 405
B703 (2004) 236
B701 (2004) 516
B704 (2005) 475
B703 (2004) 127
B704 (2005) 332
B709 (2005) 307
B706 (2005) 3
B701 (2004) 121
B702 (2004) 359
B702 (2004) 364
B704 (2005) 349
B706 (2005) 245
B705 (2005) 521
B710 (2005) 402
B710 (2005) 511
B705 (2005) 548
B704 (2005) 279
B707 (2005) 553
B710 (2005) 117
B706 (2005) 487
B709 (2005) 3
B707 (2005) 421
B701 (2004) 3
B705 (2005) 437
B709 (2005) 171
B710 (2005) 117
B705 (2005) 296
Saadat, M.
Saar, R.
Sagnotti, A.
Saidi, E.H.
Saiga, Y.
Sakai, N.
Sakamoto, R.
Sasaki, R.
Saulina, N.
Savc, M.
Saxena, A.
Saxena, A.
Scardicchio, A.
B707 (2005) 405

B702 (2004) 346
B708 (2005) 3
B702 (2004) 163
B702 (2004) 380
B701 (2004) 247
B704 (2005) 490
B710 (2005) 614
B706 (2005) 491
B709 (2005) 115
B701 (2004) 357
B710 (2005) 425
B704 (2005) 552
633
Scarpetta, G.
Schfer-Nameki, S.
Schellekens, A.N.
Schiappa, R.
Schmidt, M.G.
Schrder, O.
Schroers, B.J.
Schwimmer, A.
Scotto Lavina, L.
Sekino, Y.
Semenov, A.
Sharapov, A.A.
Shifman, M.
Shigechi, K.
Shin, H.
Shin, H.
Shiraishi, J.
Siebert, J.A.
Siegel, W.
Simula, S.
Smilga, A.
Smilga, A.V.
Smirnov, A.Yu.
Smith, J.
Sochichiu, C.
Sokatchev, E.
Somma, R.
Sonnenschein, J.
Sotkov, G.
Spence, B.
Spence, B.
Spence, B.
Splittorff, K.
Starinets, A.O.
Steele, T.G.
Steinacker, H.
Steinhauser, M.
Stieberger, S.
Strumia, A.
Strumia, A.
Strumia, A.
Suszek, R.R.
Sutcliffe, P.M.
Swanson, I.
Swanson, I.
Swanson, I.
B705 (2005) 577

B706 (2005) 531
B710 (2005) 3
B710 (2005) 189
B709 (2005) 141
B707 (2005) 233
B706 (2005) 569
B702 (2004) 369
B710 (2005) 402
B705 (2005) 33
B706 (2005) 411
B703 (2004) 419
B702 (2004) 37
B706 (2005) 518
B701 (2004) 380
B709 (2005) 69
B704 (2005) 490
B707 (2005) 529
B708 (2005) 317
B705 (2005) 339
B704 (2005) 445
B706 (2005) 598
B702 (2004) 307
B704 (2005) 332
B707 (2005) 303
B704 (2005) 154
B707 (2005) 421
B703 (2004) 320
B705 (2005) 548
B703 (2004) 236
B706 (2005) 100
B706 (2005) 150
B707 (2005) 509
B707 (2005) 56
B703 (2004) 413
B707 (2005) 145
B702 (2004) 333
B706 (2005) 3
B703 (2004) 127
B706 (2005) 455
B708 (2005) 215
B710 (2005) 599
B705 (2005) 384
B701 (2004) 180
B702 (2004) 86
B709 (2005) 443
Takcs, G.
Takagi, T.
Takeda, K.
Talavera, P.
Tamassia, L.
Tanii, Y.
Tarantino, C.
B702 (2004) 448

B707 (2005) 577
B709 (2005) 296
B703 (2004) 537
B705 (2005) 477
B701 (2004) 103
B705 (2005) 339
634
Tatar, R.
Tatar, R.
Tavartkiladze, Z.
Teodorescu, R.
Thomas, S.
Thormeier, M.
Tonni, E.
Travaglini, G.
Travaglini, G.
Troost, J.
Tseng, L.-S.
Tuckmantel, M.
B702 (2004) 207

B704 (2005) 231
B709 (2005) 141
B704 (2005) 407
B703 (2004) 236
B706 (2005) 372
B707 (2005) 321
B706 (2005) 100
B706 (2005) 150
B710 (2005) 529
B707 (2005) 43
B707 (2005) 215
Weigel, H.
Weisz, P.
West, P.C.
Westphal, A.
Wiegmann, P.
Wiese, K.J.
Williams, B.
Wosiek, J.
Wright, J.
Wu, X.-H.
Wu, Y.-S.
Wyler, D.
B707 (2005) 233

B709 (2005) 329
B705 (2005) 111
B701 (2004) 273
B704 (2005) 407
B701 (2004) 409
B706 (2005) 127
B703 (2004) 454
B702 (2004) 131
B701 (2004) 54
B708 (2005) 72
B703 (2004) 147
Uccirati, S.
Uibo, H.
Uraltsev, N.
Uranga, A.M.
B703 (2004) 3
B702 (2004) 346
B710 (2005) 371
B708 (2005) 268
Vacca, G.P.
Vainshtein, A.
Vajente, G.
van der Bij, J.J.
van der Bij, J.J.
Vanhove, P.
van Neerven, W.L.
van Nieuwenhuizen, P.
Vzquez-Poritz, J.F.
Vicari, E.
Vidal, J.
Villadoro, G.
Vissani, F.
B706 (2005) 391

B704 (2005) 445
B709 (2005) 465
B701 (2004) 121
B702 (2004) 364
B702 (2004) 269
B704 (2005) 332
B702 (2004) 189
B709 (2005) 47
B709 (2005) 550
B701 (2004) 87
B705 (2005) 339
B708 (2005) 215
Yamamoto, T.
Yan, Q.-S.
Yang, H.S.
Yang, J.M.
Yang, L.L.
Yang, W.-L.
Yang, W.-L.
Yang, W.-M.
Yarom, A.
Ydri, B.
Ylmaz, N.T.
Yokoya, H.
Yoshida, K.
Yoshida, K.
Yung, A.
B702 (2004) 380

B704 (2005) 75
B708 (2005) 434
B709 (2005) 409
B705 (2005) 3
B704 (2005) 510
B710 (2005) 614
B707 (2005) 87
B709 (2005) 391
B704 (2005) 111
B705 (2005) 60
B701 (2004) 405
B701 (2004) 380
B709 (2005) 69
B709 (2005) 267
Wadati, M.
Wang, F.
Wang, N.
Wang, Z.-G.
Wecht, B.
Wecht, B.
B706 (2005) 518

B709 (2005) 409
B706 (2005) 518
B707 (2005) 87
B702 (2004) 131
B706 (2005) 127
Zabrodin, A.
Zaffaroni, A.
Zeitlin, A.M.
Zhang, R.-J.
Zhang, Y.-Z.
Zupnik, B.
Zupnik, B.
Zwirner, F.
B704 (2005) 407

B710 (2005) 511
B709 (2005) 578
B704 (2005) 3
B704 (2005) 510
B704 (2005) 154
B707 (2005) 69
B707 (2005) 215

Nucl - Phys.B v.710

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Nucl - Phys.B v.710

Transféré par

Droits d'auteur :

Formats disponibles

Nuclear Physics B 710 (2005) 357

Supersymmetric standard model spectra

E-mail addresses: tdykstra@nikhef.nl (T.P.T. Dijkstra), lennaert_janita@hotmail.com (L.R. Huiszoon),

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

For a variety of reasons we will require the string spectrum to be supersymmetric.

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

U (3) Sp(2) U (1) U (1),

Type 4: U (3) Sp(2) Sp(2) U (1),

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

A set of phases K (J ) for all J H that satisfy

with K (J ) = ei(hKL hK ) (K, L), with (K, L) = 1.

The following objects are introduced:

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

The aforementioned central stabilizer is defined in terms of this quantity as

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

 S i m R[a,a ](m,J ) R[b,

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

To every hi = 1 primary we can associate a Witten index

3.1. The oriented closed string spectrum

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

h21 + h11 . In a geometrical setting the numbers h+

3.4. The unoriented open string spectrum

2 This follows from

which can be derived from Ai ab = Ai a b , unitarity of S and completeness.

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

. The net chirality of fermions transforming as (Va , Vb ) is measured by

Ma c ,A . The net chirality is

3 We now also need

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

4. Tadpoles and anomalies

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

must be cancelled by a generalized GreenSchwarz mechanism. This mechanism involves

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

where we sum over pairs (a, a c ). The GS-diagram is proportional

is massless if and only if

degenerate basis {C(2) } that is one-to-one to the Ishibashi labels (m, J )

(m,J ),wm =0

for all Ishibashi labels (m, J ).

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

(8, 1, 0)0,0 + (1, 1, 0)0,0

(1, 3, 0)0,0 + (1, 1, 0)0,0

5.1. The numbers

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

(1, 5, 41, 1804)

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

T.P.T. Dijkstra et al. / Nuclear Physics B 710 (2005) 357

(1, 6, 28, 118)

(24, 84, 429)

S i m R[a,a ](m,J ) R[b,