Nucl - Phys.B v.635 PDF

Nuclear Physics B 635 (2002) 332
www.elsevier.com/locate/npe
Implications of superconformal symmetry for

interacting (2, 0) tensor multiplets
G. Arutyunov a,1 , E. Sokatchev b,2
a Max-Planck-Institut fr Gravitationsphysik, Albert-Einstein-Institut, Am Mhlenberg 1,
D-14476 Golm, Germany

b CERN Theoretical Division, CH-1211 Geneva 23, Switzerland
Received 20 February 2002; received in revised form 24 April 2002; accepted 2 May 2002
Abstract
We study the structure of the four-point correlation function of the lowest-dimension 1/2
BPS operators (stress-tensor multiplets) in the (2, 0) six-dimensional theory. We first discuss
the superconformal Ward identities and the group-theoretical restrictions on the corresponding
OPE. We show that the general solution of the Ward identities is expressed in terms of a single
function of the two conformal cross-ratios (prepotential). Using the maximally extended gauged
seven-dimensional supergravity, we then compute the four-point amplitude in the supergravity
approximation and identify the corresponding prepotential. We analyze the leading terms in the OPE
by performing a conformal partial wave expansion and show that they are in agreement with the
non-renormalization theorems following from representation theory. The investigation of the (2, 0)
theory is carried out in close parallel with the familiar four-dimensional N = 4 super-YangMills
theory. 2002 Published by Elsevier Science B.V.
PACS: 11.25.Hf; 11.30.Pb; 11.10.Kk
1. Introduction
One theory of particular interest which has emerged on the scene of the AdS/CFT
duality [13] is the superconformal six-dimensional theory of (2, 0) self-dual tensor
multiplets. It is conjectured to describe the world-volume fluctuations of the M-theory
E-mail addresses: agleb@aei-potsdam.mpg.de (G. Arutyunov), emery.sokatchev@cern.ch (E. Sokatchev).
1 On leave of absence from Steklov Mathematical Institute, Gubkin str. 8, 117966 Moscow, Russia.
2 On leave of absence from Laboratoire dAnnecy-le-Vieux de Physique Thorique LAPTH, B.P. 110,
F-74941 Annecy-le-Vieux et lUniversit de Savoie.

0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 9 - 0
G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332
five-branes and this explains the special rle the (2, 0) theory might play in the possible
formulations of M-theory [4].
Our actual knowledge of the (2, 0) theory is very limited, primarily due to the lack of
a field theory formulation. This theory does not allow a dimensionless coupling and hence
a perturbative approach. This is quite different from the superconformal N = 4 Yang
Mills theory in four dimensions, where the coupling is merely a free parameter. On the
other hand, the conformal superalgebras in d = 4 and d = 6 and their unitary irreducible
representations (UIR) have a very similar structure. Moreover, both N = 4 SYM with
a gauge group SU(N) and the (2, 0) theory possess well-defined supergravity duals: in
the large N limit and for a large t Hooft coupling the former is dual to type IIB tendimensional supergravity compactified on AdS5 S 5 , while the latter is dual to elevendimensional supergravity compactified on AdS7 S 4 , provided the number of tensor
multiplets grows like N 3 . Thus, representation theory confronts us with the problem to
understand what precisely makes these two theories so different and whether they have
something in common. The aim of the present paper is to provide a partial answer to this
question. At the same time, we put the AdS/CFT correspondence to another non-trivial
test.
We exploit two different but complementary approaches. The first is to study the
operator product expansion (OPE) of two stress-tensor multiplets. These are the simplest
non-trivial examples of the so-called 1/2 BPS operators. Their conformal dimension is
protected from quantum corrections by conformal supersymmetry, but their OPE has a rich
spectrum of both protected and unprotected multiplets. In Section 2.1 we recall the known
facts about this OPE both in d = 4 and in d = 6 and make a detailed comparison of the OPE
spectra of the two theories. In particular, we point out the different realization of what one
may call Konishi-like multiplets. In d = 4 these are represented by operators bilinear in the
fundamental fields which have canonical dimension and satisfy conservation conditions in
the free theory, but develop anomalous dimension in the presence of interaction. This is
related to the fact that the corresponding superconformal UIRs lie at the unitarity bound of
the continuous series of representations. At the same time, other bilinear operators, also at
the unitarity bound of the continuous series, remain protected. This is due to the kinematics
of the OPE, i.e., to the properties of the three-point functions which these operators may
form with the two 1/2 BPS operators.
In d = 6 the picture is quite different. There the operators at the unitarity bound of the
continuous series of UIRs are trilinear and cannot appear in the OPE. The closest analogs
of the Konishi-like bilinear operators belong to a discrete series of UIRs with quantized
dimension. Thus, they are automatically protected by unitarity. The lowest-dimension
unprotected multiplets in this OPE correspond to UIRs lying above the unitarity bound
of the continuous series and are realized by quadrilinear operators.
The second approach is based on the superconformal Ward identities for the four-point
function of stress-tensor multiplets. In d = 4 they are known to restrict the freedom in
the amplitude to just two functions of conformally invariant variables, one depending on
two such variables and the other on one variable. An additional, dynamical mechanism
which generates the quantum corrections to the amplitude by insertion of the SYM action,
fixes the function of one variable at its free theory value (partial non-renormalization).
Once again, the situation changes in d = 6. In Sections 2.2 and 3 we show that the
analogous Ward identities are solved in terms of a single function of two variables, which
we call prepotential. In other words, in d = 6 the partial non-renormalization is purely
a kinematical effect. We consider this new phenomenon as an indication that there exists
no smooth interpolation between the two fixed conformal points, the free one and the dual
of the supergravity theory.
In the absence of a perturbative formulation of the (2, 0) theory one can only test
the above general predictions via the AdS/CFT correspondence. In Sections 2.3 and 4
we use the maximally extended gauged seven-dimensional supergravity to derive the
corresponding four-point amplitude and verify that it satisfies the constraints found in
Section 3. We provide an explicit simple formula for the six-dimensional gravity-induced
prepotential and show how it is related to its four-dimensional analogue. Finally, in
Section 5 we analyze the leading terms in the conformal partial wave expansion of the
supergravity four-point amplitude and show that they are in complete agreement with
the OPE structure discussed in Section 2.1. Some technical details are gathered in the
appendices.
2. Overview and summary of the results

2.1. OPE of stress-tensor multiplets
One can learn a lot both about the kinematics and the dynamics of a (super)conformal
theory by studying the OPE of various operators. In the context of the AdS/CFT
correspondence the so-called 1/2 BPS short operators3 are of particular interest since,
on the one hand, their conformal dimension is quantized (protected) in the CFT, and
on the other hand, they can be identified with the KaluzaKlein excitations in the AdS
supergravity spectrum [3,5]. They correspond to states which are annihilated by half of the
supercharges. The simplest example of a 1/2 BPS operator is the stress-tensor multiplet
OI whose lowest component is a scalar of dimension = d 2 belonging to the vector
representation of the R symmetry group SO(6) SU(4) or SO(5) USp(4) in the cases
d = 4 or d = 6, respectively.
Before discussing the OPE of 1/2 BPS operators, it is useful to recall some known
facts [6] about the UIRs of the d = 4 N = 4 and the d = 6 (2, 0) superconformal
algebras PSU(2, 2/4) and OSp(8 /4), correspondingly. They are labeled by the quantum numbers of the lowest-weight state D( ; J1 , J2 ; a1, a2 , a3 ) (for PSU(2, 2/4)) or
D( ; J1 , J2 , J3 ; a1 , a2 ) (for OSp(8 /4)). Here is the conformal dimension, Ji label the
Lorentz group SO(3, 1) SL(2, C) or SO(5, 1) SU (4) irrep, and [ai ] are the Dynkin
labels of the SU(4) or USp(4) irrep, correspondingly. We will be interested in superconformal UIRs which can appear in the OPE of two 1/2 BPS operators; since the latter are
Lorentz scalars, the former must be vector-like with spin s, i.e., with J1 = J2 = s/2 for
3 In the AdS/CFT literature the BPS operators are often called Chiral Primary Operators (CPO). This name
does not seem very adequate in the (2, 0) theory, where all spinors are chiral.
d = 4 and J1 = J3 = 0, J2 = s for d = 6. Below we list the relevant UIRs:4

PSU(2, 2/4),
J1 = J2 = s/2:
(A)
2 + s + a1 + a2 + a3 ,
(C)
= a1 + a2 + a3 ,
OSp(8 /4),
s = 0;
(1)
J1 = J3 = 0, J2 = s:
(A)
6 + s + 2(a1 + a2 ),
(B)
= 4 + s + 2(a1 + a2 ),
(C)
= 2 + 2(a1 + a2 ),
(D)
= 2(a1 + a2 ),
s = 0,
s = 0.
(2)
In both cases series (A) is continuous whereas (B), (C) and (D) are isolated and contain
operators with quantized conformal dimension. Series (C) in d = 4 and (D) in d = 6
correspond to the BPS states. In particular, if only a2 = 0 we obtain 1/2 BPS states.
The d = 4 and d = 6 stress-tensors belong to the 1/2 BPS multiplets D(2; 00; 020) and
D(4; 000; 02), respectively.
The OPE of two 1/2 BPS operators is restricted by the kinematics in the sense that it
can contain only the operators whose quantum numbers D( ; Jm ; an ) allow them to form a
non-vanishing three-point function with the two 1/2 BPS operators [7]. The case of interest
for us is the OPE of two stress-tensor multiplets. Then the allowed R-symmetry irreps are
obtained by decomposing the tensor products
SU(4):
[020] [020] = [040]0 + [121]0 + [202]0 + [020]1 + [101]1
(3)
+ [000]2;
USp(4): [02] [02] = [04]0 + [40]0 + [22]0 + [02]1 + [20]1 + [00]2,
(4)
1
where the subscript indicates the value of the integer k = 2 2 ai . In Refs. [711] it
was shown that the existence of a non-vanishing three-point function for k = 0, 1 implies
certain selection rules. In particular, the dimension of the operators appearing in the OPE
becomes quantized. No such selection rules are found for k = 2. More specifically, in terms
of the classifications (1) and (2) the picture is as follows.
k = 0: The three irreps with k = 0 in (3) and (4) belong to the series (C) (d = 4) and
(D) (d = 6), and therefore they are 1/2 or 1/4 BPS states. The corresponding operators
are protected, i.e., their conformal dimension = 2(d 2) cannot be modified by the
interaction.
k = 1: In this case all the operators are protected as well. However, this time, besides
1/2 or 1/4 BPS short operators of dimension = d 2, we encounter a new species
of protected operators, the so-called semishort operators (see, e.g., [12]). They have
different interpretations in four and six dimensions.
In d = 4 the semishort operators with spin s 0 lie at the unitarity bound = s + 4 of
the continuous series (A). They satisfy conservation-like conditions in superspace (which
4 Series (B) of PSU(2, 2/4) contains chiral superfields with J J = 0, so it is not relevant here.
1 2
imply the existence of conserved component tensors in the multiplet). For this reason they
may also be called current-like. We stress that in d = 4 the semishort operators are a
priori not protected, since their dimension can be continuously varied above the unitarity
bound. However, the careful analysis of the corresponding three-point function [711]
shows that the kinematics of the particular OPE under consideration protects the dimension
of the semishort operators with k = 1, so that they remain at the unitarity bound even in
the presence of interaction. A well-known example of a protected semishort operator is the
4 corresponding to the UIR D(4; 00; 020), first discovered in Refs. [13,14].
so-called O20
In d = 6 the semishort operators with k = 1 correspond to UIRs from the isolated
series (B) with quantized dimension = s + 8 [9,15]. Since their conformal dimension
cannot be continuously modified, they are automatically protected by unitarity. In this
respect they resemble the BPS short operators which belong to an isolated series of UIRs
as well. Note that the existence of an isolated series of semishort operators is specific to
the six-dimensional superconformal algebras OSp(8 /2N ).
k = 2: This is the most interesting case since only it involves unprotected operators. As
can be seen from (3) and (4), these are R-symmetry singlets. Here the analysis of the threepoint functions produces no further selection rules. Still, a particular type of operators can
be singled out. Again, the situation is different in four and in six dimensions.
In d = 4 the operators with k = 2 lying at the unitarity bound = s + 2 have twist
s = 2. So, they correspond to bilinears made out of the free N = 4 SYM field-strength
superfields. Still in the free case, these bilinears satisfy conservation conditions which
make them semishort. However, this conservation does not reflect any symmetry of the
interacting theory, therefore such operators develop anomalous dimension and drift away
from the unitarity bound. So, the semishort operators with k = 2 are unprotected. The bestknown example of this type is the Konishi multiplet (a singlet scalar of dimension 2), but
there exists an infinite series of similar operators with spin which we call Konishi-like.
Their anomalous dimension at one loop has been calculated in Refs. [14,1619].
It should be pointed out that the Konishi-like operators are not present in the OPE
derived from gauged N = 8 supergravity. This was demonstrated in Ref. [13] by analyzing
the supergravity four-point function of 1/2 BPS operators found in Ref. [20]. A common
lore to explain their absence in the strongly coupled N = 4 theory is to say that they
2 N)1/4 as the t Hooft coupling g 2 N tends
develop large anomalous dimension (gYM
YM
to infinity, and thus they drop out of the spectrum. Note that the peculiar asymptotic
2 N)1/4 has not yet been obtained by field theory means and it remains
behavior (gYM
a prediction of string theory.
The six-dimensional case is rather different. Here the multiplets with k = 2 lying at the
unitarity bound = 6 + s of the continuous series A have twist 6, so in the free theory
they could be realized only by trilinear operators. Such operators cannot appear in the OPE
of two stress-tensor multiplets (bilinears). Therefore, we should look for analogs of the
Konishi-like multiplets among the bilinear composites with k = 2. In d = 6 they should
have twist 4 and we see that they can only appear in the isolated series (B). Thus, we may
say that in d = 6 the Konishi-like semishort multiplets are protected by unitarity.
Being protected operators in d = 6, the Konishi-like multiplets are not expected
to appear in the supergravity-induced OPE. This can be anticipated on the general
grounds of the AdS/CFT correspondence, because there is no field in the spectrum of the
corresponding supergravity theory dual to any of these currents. In Section 4, by using

gauged seven-dimensional N = 4 supergravity [21], we compute the four-point amplitude
of the lowest dimension 1/2 BPS operators in the (2, 0) theory. Subsequently, in Section 5
we indeed demonstrate the absence of the Konishi-type currents in the supergravityinduced OPE. On the other hand, the Konishi-like multiplets are present in the free OPE of
two 1/2 BPS operators. In our opinion, the fact that they drop out of the spectrum of the
interacting theory clearly demonstrates the absence of a superconformal theory that could
smoothly interpolate between the free CFT and the CFT dual to the eleven-dimensional
supergravity on the AdS7 S 4 background.
Note that the d = 6 OPE under consideration does not contain operators from series (C).
Indeed, since in the k = 2 channel a1 = a2 = 0, they should have the dimension of the
fundamental field = 2.
2.2. Four-point function of stress-tensor multiplets
The complete, i.e., both kinematical and dynamical information about the OPE of
two stress-tensor multiplets is encoded in their four-point correlation function. We have
already seen that the kinematics (or, in other terms, superconformal representation theory)
strongly restricts the content of the OPE. We should expect to see the implications of these
restrictions on the four-point amplitude. The easiest and most economic way to do this is
to use the superconformal Ward identities. Below we summarize the already known results
about this four-point amplitude in d = 4 [19,2226] and compare them to our new results
in the six-dimensional case.
We would like to stress that the four-point amplitude of 1/2 BPS short multiplets
that we consider is rather special in the sense that superconformal symmetry is powerful
enough to restore the complete superspace dependence solely from the knowledge of the
lowest ( = 0) component of the amplitude. Indeed, the 1/2 BPS short superfields depend
on half of the Grassmann variables. Thus, a four-point function of this type depends on
4 (1/2) = 2 full sets of odd variables. At the same time, the superconformal algebra has
two sets of odd shift-like generators (Q and S supersymmetry). This leaves no room for
nilpotent superconformal invariants made out of the odd variables and thus the expansion
is completely fixed.
The lowest component of this amplitude corresponds to the correlator of four scalar
operators OI of dimension = d 2 in the vector representation of the R-symmetry group
SO(6) (for d = 4) or SO(5) (for d = 6):

I
O 1 (x1 ) OI4 (x4 )
= a1 (s, t)
I1 I2 I3 I4
I1 I3 I2 I4
I1 I4 I2 I3
+
a
(s,
t)
+
a
(s,
t)
2
3
2 x 2 )d2
2 x 2 )d2
2 x 2 )d2
(x12
(x13
(x14
34
24
23
+ b1 (s, t)
+ b3 (s, t)
C I1 I3 I2 I4
2 x 2 x 2 x 2 ) d2
(x13
14 23 24 2
C I1 I2 I4 I3
2 x 2 x 2 x 2 ) d2
(x12
13 24 34 2
+ b2 (s, t)
.
C I1 I2 I3 I4
2 x 2 x 2 x 2 ) d2
(x12
14 23 34 2
(5)
Here s, t are the conformal cross-ratios

s=
2 x2
x12
34
2 x2
x13
24
t=
2 x2
x14
23
2 x2
x13
24
The six tensor structures I1 I2 I3 I4 , C I1 I2 I3 I4 (and permutations) are invariant tensors of

SO(6) (or SO(5)) and are related to the six channels in the OPE (3) (or (4)). In general,
we define the invariant tensors C I1 In as tr(C I1 C In ), where the matrices CijI , which are
symmetric and traceless in their lower indices, realize a basis of the corresponding vector
representation of the R symmetry group.
Among the six coefficient functions in (5) only two are independent, for example,
a1 (s, t) and b2 (s, t). The others are obtained from the crossing symmetry relations
a1 (s, t) = a3 (t, s) = a1 (s/t, 1/t),
a2 (s, t) = a2 (t, s) = a3 (s/t, 1/t),
b1 (s, t) = b3 (t, s) = b1 (s/t, 1/t),
b2 (s, t) = b2 (t, s) = b3 (s/t, 1/t).
(6)
The amplitude (5) must obey superconformal Ward identities which follow from the
1/2 BPS nature of the supermultiplets. In the four-dimensional case they take the form of
two first-order PDEs for the independent coefficient functions [24]:
d = 4 Ward identities:
s
s+t 1
t a1 ,
t b2 = s a3 s a1
t
s
t
s+t 1
s a3 .
s b2 = t a1 t a3
(7)
s
t
In six dimensions the corresponding equations look very similar (see Section 3 for the
derivation):
d = 6 Ward identities:
s2
t
t (s + t 1)
s a3 s a1
t a1 ,
2
t
s
s2
t2
s
s(s + t 1)
s b2 = 2 t a1 t a3
(8)
s a3 .
s
t
t2
However, the small change in the coefficients from Eq. (7) to Eq. (8) results in an important
difference when it comes to their general solution.
In Ref. [25] it was found that the general solution5 of the d = 4 Ward identities (7)
is parametrized by two independent functions, one of two variables and the other of a
single variable.6,7 It was further shown in Ref. [25] that the function of one variable can
t b2 =
5 We recall some details in Appendix A.

6 It should be mentioned that a similar picture has been observed many years ago by Fradkin, Palchik and
Zaikin [28]. They have studied the conformal correlator of one conserved current with three scalar operators. By
imposing the Ward identity for the current they have found a differential equation whose solution has exactly the
same functional freedom. However, the difference between their study and ours is in the fact that the functions
determining their correlator are obtained as derivatives of our coefficients ai and bi (when reconstructing
the corresponding component of our superamplitude starting from the lowest one). Thus, the solution of our
constraints lies one level deeper than that of Ref. [28].
7 For a recent discussion in which crossing symmetry is not imposed see Ref. [19].
10
be set to its free-theory value by evoking a dynamical mechanism. It consists in employing

Intriligators insertion procedure [27] which gives the quantum (interacting) part of the
amplitude as the result of the insertion of the SYM action into it. Thus, combining
kinematics with dynamics, the full solution of the d = 4 superconformal Ward identities is
reduced to
a1 (s, t) = A + sF (s, t),
b2 (s, t) = B + (1 s t)F (s, t),
where A and B are constants determined from the free N = 4 theory.

The non-trivial part of the amplitude is therefore encoded in the single function of two
variables F (s, t) satisfying the crossing-symmetry conditions
F (s, t) = F (t, s) = 1/tF (s/t, 1/t).
(9)
It includes all (non-)perturbative corrections to the free field amplitude. This prediction
of the superconformal Ward identities and of the dynamical insertion procedure about
the form of the amplitude, called partial non-renormalization in Ref. [25], has been
confirmed by all the available perturbative [29], instanton [30] and strong coupling [20]
results.
In six dimensions the solution of the Ward identities (8) is directly given in terms of one
unconstrained function of two variables (see Section 3):

1
4
a1 (s, t) = A + s 3 tF (s, t) ,

1
b2 (s, t) = B + s 2 t 2 3 (1 s t)F (s, t) ,
where A and B are additive integration constants, is a second-order differential operator,

= sss + tt t + (s + t 1)st + 3s + 3t ,
(10)

and = (s + t 1)2 4st is its discriminant. Here the function F (s, t) satisfies the
same crossing-symmetry relations as its four-dimensional analogue (recall (9)):
F (s, t) = F (t, s) = 1/tF (s/t, 1/t).
(11)
Again, it encodes the dynamics of the theory and, in particular, it comprises all M-theory
corrections to the leading supergravity result. However, in d = 4 this function itself is
a coefficient function of the amplitude, whereas in d = 6 it plays the rle of a prepotential
in the sense that all the coefficients can be obtained from it by applying derivatives. It
would be very interesting to find out whether this prepotential has a deeper origin.
We would like to underline once more the important difference between the four- and
six-dimensional cases. In d = 4 one can reduce the freedom in the amplitude to just one
unconstrained function by combining kinematics (the superconformal Ward identities)
with dynamics (the insertion formula). The latter relies on the existence of a certain
nilpotent superconformal five-point covariant with rather special properties [25,31]. Our
attempts to find a similar construction in d = 6 were unsuccessful. This again points at
the absence of a Lagrangian formulation of the six-dimensional theory. However, now we
11
see that in d = 6 the kinematics (superconformal symmetry) alone leaves exactly the same
freedom, the single function F (s, t).
Our final remark concerns an alternative explanation of the rle of the function of one
variable in the four-dimensional amplitude, recently discussed by Dolan and Osborn [19].
They relate this function to the possible exchanges only of protected operators in the OPE
(the first five channels in the decomposition (3)). Indeed, it is easy to show that Intriligators
insertion procedure forbids such exchanges [11], and so it is natural to expect that this fixes
the function at its free-theory value. However, in six dimensions similar protected channels
exist, but the insertion procedure cannot be applied. Still, we do not find a function of
one variable here. It would be interesting to understand this phenomenon from the OPE
point of view advocated in Ref. [19]. One might speculate about the different behavior of
the protected semishort operators in d = 4 which lie at the boundary of the continuous
series (A), and of those in d = 6 which belong to the isolated series (B).
2.3. Obtaining the prepotential F from gauged supergravity
Since no field-theory formulation of the interacting (2, 0) six-dimensional theory is
available, the way to check the general predictions we have found here is to compute
the amplitude via the AdS/CFT correspondence and to try to identify the prepotential F .
We perform this program in Section 4 by using the gauged seven-dimensional N = 4
supergravity and find a perfect agreement. In particular, we show that the supergravity
four-point amplitude of the 1/2 BPS operators is generated by the following very simple
prepotential
240 3
D7333 (s, t),
N 3 st 2
together with the integration constants
F (s, t) =
A = 1,
B=
1
.
N3
(12)
(13)
1 2 3 4 (s, t) are defined in Appendix B.

The conformally covariant functions D
It is interesting to compare this supergravity-induced solution with the theory of free
(2, 0) tensor multiplets. For this theory the prepotential F vanishes, while the constants A,
B equal
A = 1,
4
B= .
(14)
If one would try to view the supergravity solution F as being obtained from the free one
F = 0 by some smooth deformation, then one should obviously set = 4N 3 . The factor
4N 3 was found in Ref. [32] by studying the absorption rate of longitudinally polarized
gravitons by M5-branes. The same factor appears as the universal coefficient between
the free and the AdS two- and three-point correlators of the stress tensor [33], as well
as between the free and the AdS type (B) conformal anomaly [34]. Since all the non-trivial
dynamics is encoded in the prepotential, we see that 4N 3 is just what is needed to match the
integration constants of the free and of the supergravity-induced four-point amplitudes. In
12
Section 4 we discuss however that the existence of a smooth superconformal deformation

from the free to the supergravity theory appears to be in conflict with unitarity.6
It is instructive to compare the six-dimensional prepotential (12) with the potential
which generates the strongly-coupled d = 4 N = 4 amplitude found via the AdS/CFT
correspondence. The latter can be extracted from the results of Refs. [20,25] (cf.
Appendix C):7
24 1
4
D4222 (s, t),
A = 1,
B = 2.
(15)
N2 t
N

Since differentiating a D-function
with respect to s or t amounts to raising by unity the
values of two of its indices, we see that the six-dimensional prepotential is obtained from
the four-dimensional potential by dressing it with a certain third-order differential operator
(see Section 4). It would be interesting to find out if this operator has some intrinsic
meaning.
Finally, we observe that we might construct infinite towers of superconformal four-point
amplitudes both in the d = 4 and d = 6 theories as follows:
F (s, t) =
1
3
F (s, t) D
(16)
F (s, t) 2 D
32,,, (s, t),
32,,, (s, t),
t
st
where = 1, 2, . . . . These functions are symmetric and satisfy (11) as a consequence of

the corresponding symmetry properties (C.4) and (C.5) of the D-functions.
One could ask
the question whether any of the amplitudes (16) (or their linear combinations), other then
(12) and (15), together with some appropriate integration constants A and B, has an OPE
free from Konishi-like multiplets. If not, this might explain the distinguished rle of the
amplitudes (12) and (15).
3. General structure of the four-point amplitude

We begin this section by recalling, just in a few words, the procedure which leads to
the d = 4 Ward identities (7). The origin of these constraints can be traced back to the fact
that the N = 4 SYM stress-tensor multiplet is 1/2 BPS short. The natural framework for
describing BPS shortness is harmonic superspace [35]. The constraints (7) can be derived
directly in N = 4 harmonic superspace [36], but it is easier to do this using its N = 2
version (both methods are explained in detail in Ref. [24]; for a recent rederivation of the
same constraints without using harmonic superspace see Ref. [19]). The main point is that
when reconstructing the full harmonic superspace dependence of the four-point amplitude
starting from its lowest component (5), one encounters harmonic singularities already at
the level ( )2 . Their absence, i.e., the requirement of harmonic analyticity, is equivalent
6 One way to smoothly connect the free and the supergravity amplitudes is to multiply the supergravity
prepotential by a function f (g) of some coupling constant g, such that f (0) = 0 and f (1) = 1. This is a trivial
possibility that would explain the decoupling of the Konishi-like multiplets from the supergravity OPE by the
vanishing of the corresponding OPE coefficients. In what follows we discard such deformations.
7 See also Appendix D of Ref. [19], where a similar formula for F (s, t) was obtained from the results of [20].
13
to imposing irreducibility under the R symmetry group. It is precisely this requirement

which leads to the constraints (7). Once these constraints have been imposed, it can be
shown that no new harmonic singularities appear at the higher levels of the expansion of
the amplitude.
In d = 6 one could go through exactly the same steps in order to obtain the new
constraints (8). However, there is a much faster way which consists in simply adapting
the d = 4 constraints (7) to the case d = 6. The key observation is that the coefficient
functions a1 , a3 , b2 appear in (7) only through their first-order derivatives. The origin of
these derivatives is in the completion of the conformal invariant s to a full superconformal
invariant s = s + -terms (and similarly for t). Expanding, e.g., a1 (s , t) up to the level ( )2
gives rise to the terms s a1 , t a1 . It is not hard to show that s , t are exactly the same both
in d = 4 and in d = 6. Next, the rational coefficients in (7) originate from the propagator
2 x 2 )d2 , etc. in (5). These differ in d = 4 and d = 6, as can be seen most
factors 1/(x12
34
clearly by pulling out one of them in front of the amplitude (only the relevant terms are
shown):

d2
s
1
I1 I2 I3 I4
a1 (s, t)
+
a3 (s, t) I1 I4 I2 I3
2
2
d2
t
(x12 x34 )
d2

s 2
+
b2 (s, t)C I1 I2 I3 I4 + .
t
Now, the completion of these propagator factors to full superconformal covariants does
not affect the derivative terms in (7). Therefore, in order to pass from d = 4 to d = 6 it is
sufficient to just redefine the coefficient functions as follows:
s 2 d=4
s
(17)
a
,
b2d=6 b2d=4.
t
t2 3
This redefinition should be done in (7) so that the derivatives do not act on the factors in
(17). The result is precisely the constraints (8).
We remark that due to the crossing symmetry relations (recall (6))
a1d=6 a1d=4 ,
a3d=6
a3 (t, s) = a1 (s, t),
b2 (s, t) = b2 (t, s)
(18)
the second equation both in (7) and in (8) is automatically satisfied.

From here on we concentrate on the solution of the d = 6 constraints (8). The
integrability condition for this system of PDEs is

a3
a1
t 2 = s 2 ,
(19)
s
t
where was defined in (10). To solve this second-order PDE we make the substitution
a1
s
a3
t
(20)
= 3 G(s, t),
= 3 G(t, s)
2
2
s
and change the variables s, t to the normal coordinates x, y for the hyperbolic
operator :
x = s,
y = t,
(21)
14

where = 2(1 s t + )1 and = (1 s t)2 4st . After this change Eq. (19)
becomes

y y + x 2 x + (1 xy)xy G(x, y) = x x + y 2 y + (1 xy)xy G(y, x).
Then we set
G(x, y) = (x, y) + (x, y),
where
(x, y) = (y, x),
(x, y) = (y, x)
(22)
and perform one more change of variables

x
= ln .
y
= ln(xy),
In these new variables the symmetry conditions (22) become

(, ) = (, ),
(, ) = (, )
(23)
and Eq. (19) takes the form

= coth(/2) .
(24)
This equation can be integrated to give

= coth(/2)

, d + c( ),
where c( ) is an integration constant depending on . Let us introduce the function

F (, ) =
, d .
(25)
It is even, F (, ) = F (, ) and is supposed to obey the boundary condition

F (, 0) = 0. Without loss of generality c( ) is absorbed into F , which results only in
a change of the boundary condition for F . Thus,

+ = + coth(/2) F
(26)
and the function F plays the rle of a prepotential.
Switching back to the original variables s, t we find

s4t
s +t 1
s + sss + (s + t 1)st + tt t F (s, t),
a1 = 3 2t +
t

t 4s
s +t 1
t + sss + (s + t 1)st + tt t F (s, t),
a3 = 3 2s +
s
where the prepotential F (s, t) is an arbitrary symmetric function.
(27)
(28)
15
Some comments are due here. First of all, it is obvious that a1 and a3 admit the constant
solution a1 = a3 = A. Computing (, ) for this trivial solution and further integrating
over we find the corresponding prepotential F0 (s, t):
F0 (s, t) =
A s3 + t 3 3
.
3 s3t 3
(29)
Secondly, the prepotential is not uniquely defined. The freedom in redefining F without
changing the amplitude can be easily found by solving the homogeneous equation implied
by (26). This is done by separating the variables. The resulting freedom is
F (s, t) F (s, t) + h(s, t),
(30)
where

2
1
2
2 ln st
h(s, t) = h1 st
+ h2
st 2
(31)
and h1 and h2 are arbitrary constants.

Since the integrability condition (19) is already satisfied, we can now integrate, e.g., the
first of Eq. (8) for b2 . We obtain

s2t 2
2s
b2 (s, t) = B 3 (s + t 1) sss + (s + t 1)st + tt t + 1 +
s
s+t 1

2t
+ 1+
(32)
t F (s, t),
s +t 1
where B is a new integration constant. Under the replacement (30) the coefficients a1 and
a3 and, therefore, Eq. (8) remain unchanged. However, the solution (32) is allowed to pick
an additive constant. Indeed, we find that (30) leads to the shift B B + h1 . Finally, note
that the constant B remains unchanged under the replacement F (s, t) F (s, t)+F0 (s, t).
Now we are in a position to find the implications of the global crossing symmetry
conditions. Under the change s s/t, t 1/t we find
a1 (s/t, 1/t) =

s4
sss + (s + t 1)st + tt t F (s/t, 1/t).
3
(33)
Clearly, by choosing
F (s/t, 1/t) = tF (s, t)
(34)
we are able to satisfy the crossing symmetry relation a1 (s/t, 1/t) = a1 (s, t). Note that
neither F0 (s, t) nor h(s, t) obey Eq. (34). Therefore, in the following we represent the
general solution of the superconformal Ward identities in the form
F0 (s, t) + F (s, t),
(35)
where F (s, t) satisfies the crossing symmetry relation (34). This requirement also fixes the
freedom (30) in the prepotential.
16
Formulae (27), (28) and (32) can be further simplified to give

s4
sss + (s + t 1)st + tt t tF (s, t) ,
3
t4
a3 = 3 sss + (s + t 1)st + tt t sF (s, t) ,
s2t 2
b2 = 3 sss + (s + t 1)st + tt t (1 s t)F (s, t) .
(36)
Now we note that the operator has the following property. For any function w(s, t),
a1 =

1
w(s, t)
= 3 tt t + (s + t 1)st + sss w(s, t).
3
This allows us to move the factor 1/3 to the right through the differential operator in
Eqs. (36).
We thus obtain the complete solution for the coefficients of the four-point amplitude in
terms of the prepotential F (s, t):

1
a1 (s, t) = A + s 4 3 tF (s, t) ,

1
a2 (s, t) = A + 3 stF (s, t) ,

1
a3 (s, t) = A + t 4 3 sF (s, t) ,

1
b1 (s, t) = B + t 2 3 s(s t 1)F (s, t) ,

1
b2 (s, t) = B + s 2 t 2 3 (1 s t)F (s, t) ,

1
b3 (s, t) = B + s 2 3 t (t s 1)F (s, t) .
(37)
In this form the crossing symmetry relation is most transparent, given that the operator
transforms as

s/t,1/t t 2 w(s, t) = t 4 s,t w(s, t)

(38)
for any function w(s, t). Finally, note that is a symmetric function of s, t, however under
s s/t, t 1/t it transforms as /t. Thus, if we redefine F as F 3 F , the
new function obeys the crossing symmetry relation F (s/t, 1/t) = t 4 F (s, t).
4. Four-point amplitude from gauged supergravity

According to the duality conjecture for the (2, 0) theory, in the supergravity regime the
correlation functions of any 1/2 BPS operators and of their supersymmetry descendants
can be computed from eleven-dimensional supergravity on an AdS7 S 4 background. In
17
this way many two- and three-point correlation functions have already been found [37].
Below we present the first example of a four-point amplitude of 1/2 BPS operators in this
theory and subsequently analyze the leading terms of the underlying OPE. The operators
whose amplitude we are going to find, have the lowest scaling dimension = 4 and
their dual supergravity scalars belong to the massless graviton multiplet of the AdS7 S 4
compactification. Thus, for our present purposes it is enough to consider only the sector of
the theory described by gauged seven-dimensional supergravity.
The gauged seven-dimensional N = 4 supergravity was constructed in Ref. [21] by
gauging Poincar supergravity. Alternatively, it can be obtained by compactifying elevendimensional supergravity on AdS7 S 4 with a further KaluzaKlein truncation to the
massless graviton multiplet [38].
The bosonic sector of the theory consists of the metric g , fourteen scalars parametrizing the coset space SL(5, R)/SO(5)c , the SO(5)g YangMills gauge fields AIJ and a fiveI , I, J = 1, . . . , 5. The relevant part of the Lagrangian is
plet of antisymmetric tensors S
(the metric is assumed to have Minkowskian signature)
1
g2 2
1 I J
(39)
T 2Tij T ij Pij P ij F
FI J .
8
4
2
is the field strength for AIJ . To describe the scalar manifold one introduces the
e1 L = R +
IJ
Here F
vielbein (S 1 )iI SL(5, R), where i = 1, . . . , 5 is an SO(5)c index. Then Ti = (SS t )i and
T = Tr(SS t ). The kinetic term is given by the matrix P :

t
P = S S 1 gSA S 1 + S 1 S t + g S 1 A S t ,
j
where g is the YangMills coupling. Note that in writing Eq. (39) we omitted the part of
the action depending on the antisymmetric fields since, as can be easily shown, the latter
do not propagate in the AdS exchange graphs involving four external scalar fields.
To proceed, we choose the following natural parametrization for S: S = e where is
a traceless symmetric 5 5 matrix. The scalar fields parametrizing are dual to the 1/2
BPS operators OI of dimension = 4 with the index I transforming under the irrep [02]
of the R-symmetry group SO(5).
Since we are interested in the four-point amplitude of the operators OI , we decompose
the Lagrangian in power series in and then truncate it at the fourth order. The resulting
expression reads

2
2
e1 L = R + 2 + 2g [, A ]
3
3

2
2 44

g
1 I J
tr 4 F
FI J . (40)
+
15 + 4 tr 2 8 tr 3 + 4 tr 2
8
3
2
Obviously, to obtain the correct value 2 = (d 1)(d 2) = 30 of the cosmological
constant in d = 7, one has to set g 2 = 16.
To simplify the resulting expression, we perform the field redefinition 23 3 .
It is also convenient to introduce another parametrization for the matrix of scalars and for
the vector field:
1
I
(A )ij = Ci;j
AI ,
ij = CijI s I ,
(41)
2
18
I are traceless symmetric and antisymmetric matrices providing bases

where CijI and Ci;j
(an upper index) for the irreps [02] and [20], respectively. The normalization properties of
these matrices are discussed in Appendix B. We set out to work with the Euclidean version
of the AdS metric which results in changing the overall sign of the Lagrangian.
Let us mention the issue of the overall normalization of the gravity action. We normalize
the action of eleven-dimensional supergravity as

1
g R + ,
S= 2
2k11
where k11 is the eleven-dimensional Newton constant:

2N 3
1
=
.
2
5
2k11
For the AdS7 S 4 solution with the radii RAdS7 = 1 and RS 4 = 1/2 the reduction to seven
dimensions yields
N3
1
=
.
2k72 3 3
Thus, in the sequel we will work with the action

N3
g L,
S(s) =
3
3
(42)
AdS7
where, after the manipulations described above the Lagrangian acquires the form
1
1
L = s I s I 8s I s I + 2CI1 I2 I3 s I1 s I2 s I3 T L2 ( )
4
4
I I
I I
1 I ;I
1
;I
+ 4J;I A + F F
4 CI1 I2 I3 I4 s 1 s 2 s 3 s 4
2
2
1 I1 I1 I2 I2 5
s s s s + CI1 I2 I3 I4 s I1 s I2 s I3 s I4 .
2
2
Here we introduced the currents

1
T = s I s I g s I s I 8s I s I ,
2
JI = T I1 I2 I s I1 s I2
(43)
obeying an on-shell conservation law. The tensors C I1 In were already introduced in

I3
Section 2.2, and T I1 I2 I3 = CijI1 CjI2k Ck;i
is antisymmetric in the indices I1 , I2 . In what
follows we use a more concise notation, e.g., CI1 I2 I3 I4 C1234 . In Eq. (43) L2 ( ) stands
for the standard quadratic Lagrangian of the graviton .
Now it is straightforward, although rather tedious to compute the on-shell value of the
above action subject to the Dirichlet boundary conditions. Like in the case of gauged
d = 5 supergravity, we have to evaluate the exchange graphs8 describing quartic scalar
8 The exchange AdS graphs are reduced to the contact ones by using the technique developed in Ref. [39].
19
interactions [20,40]. We omit the details of the computation since they are similar to those
of Ref. [20]. We present only the final result for the four-point amplitude of the canonically
normalized 1/2 BPS operators which is found by varying the on-shell action with respect
to the boundary data for the scalars:
OI1 (x1 ) OI4 (x4 )

=
12 34
8 x8
x12
34

25 33 +
A+ + 12 34 A01234 + C1234
A
C
1234 + t + u.
3 N 3 1234 1234
(44)
= 12 (C1234 C2134). We exhibit explicitly only the expression in the s-channel,

Here C1234
the t-channel is obtained by replacing 1 4, and the u-channel by 1 3. The first
term in (44) and its t- and u-counterparts represent the contributions of the disconnected
AdS graphs. The coefficients A,0 are obtained in terms of the D-functions defined in
Appendix B and read
1
1
7
2
D4422 + 2 D4433 D4444 + 4x34
D4455 ,
4
2
2x34
x34
2 2

2 x2
x24 x14
1
1 x13
25 1
23
0
+ 2 2
D4433
A1234 = 4 D4422 +
2 x2
2
18 x12
4 x34
6x34
x12 x34
34
2 2

2 x2
x24 x14
1 x13
31
23
+
+
+
D4444
2 x2
2 x2
8 x12
4
x12
34
34

2 2
2 x2
x24 x14
1 x13
23
2
+
+ 2 2 1 x34
D4455 ,
2 x2
2 x12
x
x
34
12 34

2 2
2 2
x23 x13
x24 D4444
A1234 = 2 2 x14
x12 x34

5 2
2
2
2
+ x24
D3434 x14
D4334 x23
D3443 + x13
D4343
6

1 2 2
1
2 2
x24 D3333
+ 4 4
x14x23 x13
x12 x34 9

1 2
2
2
2
x24 D2323 x14
+
D3223 x23
D2332 + x13
D3232 .
18
A+
1234 =
(45)
Having found the four-point amplitude, we first check if it obeys the superconformal
Ward identities (8) derived in Section 3. Rewriting the amplitude (44) in the form (5), we
make the following identification:
25 33 8 8 0
x x A
,
3 N 3 12 34 1234
24 33 4 4 4 4 +
x14x23 x34 A1234 + A+

b2 (s, t) = 3 3 x12
3214 + A1234 + A3214 .
N
a1 (s, t) = 1 +
(46)
20
Now we recall that all D-functions appearing in (45) can be expressed as derivatives of
D2222 with respect to xij2 . On the other hand, the function D2222 itself is given by
D2222 =
3
2 x2 x2 x2
2x12
13 24 34
(ss )2 (s, t),
(47)
where the function (s, t), introduced in Ref. [41], admits the following explicit
representation in terms of logarithms and dilogarithms:

1
2
t 1 + t
(s, t) =
2 Li2 (s) + Li2 (t) + ln ln
+ ln(s) ln(t) +
.
s 1 + s
3
(48)
In this way we therefore obtain a representation for the coefficients A,0 in terms of
certain differential operators in the variables s, t acting on (s, t), which is given in
Appendix C. Such a representation proves useful, since the derivatives s (s, t) and
t (s, t) are again expressed in a simple manner via (s, t). Using the formulae (71)(73)
together with (46), we have verified that the supergravity amplitude we found does indeed
obey the superconformal Ward identities (8). According to our general considerations from
Section 3, this means that a prepotential of the type (37) should exist.
At first sight, the problem of finding the prepotential corresponding to the supergravity
solution (45) looks extremely complicated, because one needs to perform the integral (25)
whose integrand involves (s, t). To solve this problem we make the assumption that F
(25), written in terms of the variables s, t, has the structure q1 (s, t)(s, t) + q2 (s, t), where
q1 and q2 are two unknown symmetric functions. Then using the fact that the derivatives
of are again expressed via and by trial and error we were able to find these unknown
functions. The final answer is surprisingly simple
F (s, t) =
3
(1 ss )(1 tt )(2 + ss + tt )(1 + ss + tt )(stst )
2N 3 st
(s, t).
(49)
We can now directly verify that substituting (49) in Eq. (37) reproduces exactly the
coefficients ai of the four-point amplitude. To reproduce bi from the prepotential (49)
as well, we found that a particular value of the integration constant B is required, namely,
B = N13 . Since (s, t) obeys the crossing symmetry relation (s/t, 1/t) = t(s, t), one
can prove that the same relation holds for the prepotential F (s, t), in accord with our
previous considerations.

The form (49) suggests that it can be recast in terms of the so-called D-functions
that
are defined in Appendix C. Indeed, it is easy to see (cf. Appendix C) that the following
formula holds
240 3
D7333 (s, t).
(50)
N 3 st 2
This time the crossing symmetry relations for the prepotential follow from the ones for the

corresponding D-function,
Eqs. (C.4) and (C.5) from Appendix C. Such an elegant form
of the prepotential suggests that there may exist a much simpler way of extracting it from
the supergravity solution.
F (s, t) =
21
Thus we have completely unraveled the structure of the supergravity solution. It consists
of the prepotential (50) supplemented with the following integration constants A, B:
A = 1,
B=
1
.
N3
(51)
The comparison of this result with the one provided by the free theory has already been
discussed in Section 2.3.
Finally, we establish a non-trivial relation between the d = 6 prepotential (49) and its
four-dimensional analogue (15). The function F (s, t) from the solution (15) can be written
in a form similar to (49):
F (s, t) =
4
(1 + ss + tt )(stst )(s, t).
N2
(52)
Therefore, comparing with (49) we obtain the following relation

F (s, t) = Ds,t F (s, t),
(53)
where
Ds,t =
1 3
(1 ss )(1 tt )(2 + ss + tt )
8N st
is a symmetric third-order differential operator. One can easily check that it satisfies the
commutation relation Ds/t,1/t t = t Ds,t which makes the crossing symmetry relation
for F (s, t) obvious. Thus, we conclude that at large N the dynamical properties of the
stress-tensor multiplet in (2, 0) theory are inherited from those of the d = 4 N = 4 theory.
The last observation suggests another non-trivial test of the original AdS/CFT duality
conjecture for the d = 4 N = 4 theory. In perturbation theory the d = 4 function F (s, t)
appears as a series

1
1
,
F (s, t) = 2 F1 (s, t; g) + O
N
N4
2 N is the t Hooft coupling. If we assume that F (s, t; g) interpolates
where g = gYM
1
smoothly between the free theory (g = 0) and the theory dual to the corresponding
supergravity (g = ), then formula (53) provides a smooth deformation connecting the
free (2, 0) theory and its supergravity dual. However the OPE of the 1/2 BPS operators
in the free (2, 0) theory contains the Konishi-like multiplets while the corresponding
supergravity OPE does not. Therefore, the decoupling of the protected Konishi-type
multiplets along this particular deformation flow induced by the N = 4 theory should take
place only due to the vanishing of their OPE coefficients. Here we should recall the d = 4
case, where the Konishi-like multiplets decouple because their conformal dimensions tends
to infinity. Thus, the known one- and two-loop results for F1 (s, t; g) could be analyzed to
see whether under (53) the protected Konishi-type multiplets emerge or not. Actually, it
would be very interesting to understand how the supermultiplets arising in the OPE of the
= 2 1/2 BPS operators of the N = 4 d = 4 theory are rearranged under (53), as well as
to clarify the meaning of the operator Ds,t both on the AdS and the CFT sides.
22
5. Operator product expansion

In Section 2.1 we have already presented the general kinematical restrictions on the OPE
content of two 1/2 BPS operators D(4; 000; 02). At the level of the four-point amplitude
these restrictions are encoded in the solution (37) of the superconformal Ward identities.
In principle, one should be able to obtain the conformal partial wave expansion of the fourpoint amplitude (37) with an arbitrary function F (s, t) and to restore all the information
about the OPE which was obtained from solving the kinematical constraints.9 However,
the presence of the second-order differential operator in (37) complicates the analysis
considerably and we have not yet found an easy way to do it. Therefore, in this paper we
confine ourselves to the study of the conformal partial wave expansion of the particular
supergravity amplitude (44).
The leading terms in the double OPE arising in the short-distance limit x1 x2 ,
x3 x4 can be found as follows. First we project the four-point amplitude (44) on the
different R-symmetry channels (the necessary projectors are given in Appendix B). Then
we replace the D-functions by their series representation (with powers and logs). The series
representation of an arbitrary D-function was worked out in detail Ref. [14]. In particular,
the short distance limit under consideration is naturally described in terms of the variables:
1
s
Y =1 ,
v= ,
t
t
such that v 0, Y 0. The leading term
v /2 F (Y )
in the conformal partial wave amplitude (CPWA) expansion of the four-point amplitude
corresponds to the contributions of all operators of twist = s. A logarithmic term of
the form v /2 Y s ln v signals an anomalous dimensions for an operator of twist and spin s.
In the sequel we will work out in detail only the leading terms of the conformal partial wave
expansion for (44) for the singlet (unprotected) R-symmetry channel and briefly comment
on what we have found in the remaining (protected) channels.
5.1. Projection on the singlet
The projection of the connected part of the four-point supergravity amplitude on the
R-symmetry singlet channel can be schematically written in the form

12 34
12 3
12 2
4
v
v
F
(Y
)
+
F
(Y
)
+
v
log
vG
(Y
)
,
O O[00] =
(54)
2
3
4
8 8
175
N 3 x12
x34 175
where the functions F2 (Y ) and F3 (Y ) coincide with the canonically normalized CPWA of
a second-rank tensor with = 6. In particular, the corresponding power series expansions
start as follows:
1
1
1 1
F3 (Y ) = Y + .
F2 (Y ) = Y 2 + Y 3 + ,
4
2
6 4
9 For the d = 4 N = 4 theory the corresponding conformal partial wave analysis was performed in Ref. [17].
23
This leading operator is nothing but the stress tensor. Since the log v term appears only at
order v 4 , we conclude that the stress tensor keeps its canonical dimension. The function
G4 (Y ) = 12
7 + O(Y ) which implies that the first operator receiving anomalous dimension
a of order N 3 is a scalar of approximate dimension = 8. Taking into account the
disconnected part of the supergravity amplitude, one finds
24
.
N3
This perfectly agrees with the classifications of the UIRs presented in Section 2.1: the
superconformal primary operator of canonical dimension = 8 lies beyond the unitarity
bound of the continuous series (A) and is allowed to acquire an anomalous dimension in
a non-trivial interacting theory. With the help of the techniques developed in Refs. [19,42]
the calculation of the anomalous dimension could be extended to the higher supermultiplets
occurring in the R symmetry singlet channel. It is then of interest to see how these
anomalous dimensions are related to those of the N = 4 theory.
It is also instructive to make the comparison with the free theory of (2, 0) tensor
multiplets. One tensor multiplet comprises five scalars, two Weyl fermions and a two-form
ij
with a self-dual field strength. Assuming the free form of the propagator 3 ab
4 for the
a =
x12
scalars ai , where i = 1, . . . , 5 and a = 1, . . . , , the canonically normalized BPS operator

j
under consideration is of the form OI = (2)1/2 3 CijI : ai a : . The leading terms of the
corresponding free OPE in the R symmetry singlet channel are
OI1 (x1 )OI2 (x2 )

1/2

s
8
4 3 x12 x12
1
I1 I2
=
T + ,
+
[K]
8
4
5
5 x12
x12
where K = 3 (10)1/2 : ai ai : is a canonically normalized bilinear operator which we
can call a Konishi-type scalar and

1
1
s
T
= ai ai ai ai ai ai
5
10
(55)
is the stress-tensor of 5 free scalars normalized as T s T s = 66 . The brackets [ ]

denote the contribution of the conformal block of the corresponding primary operator. It
is easy to see, however, that in the free theory the operator T s can be written as a sum of
three operators T , K and which belong to different supermultiplets:
1
25
1
T + K + .
(56)
14
42
3
The operators on the right-hand side are orthogonal to each other, i.e., the mixed two-point
functions vanish. T is the stress tensor of copies of the (2, 0) theory, K belongs to
the Konishi-type multiplet and is the leading component of a new current multiplet.
In fact, the operators K and are the first two operators from an infinite tower of
Konishi-type currents arising in the singlet channel of the free OPE, all of them having twist
= 4. However, as we have shown above, the only operator of = 4 contributing to the
CPWA expansion of the supergravity four-point amplitude and thus to the OPE, is the stress
s
=
T
24
tensor. Therefore, all the Konishi-type currents are absent in the supergravity OPE. Unlike
the d = 4 N = 4 theory, in d = 6 unitarity puts all of these currents in the isolated series B
of UIRs, so they cannot develop an anomalous dimension in the interacting theory. We
therefore arrive at our conclusion about the absence of a superconformal theory smoothly
interpolating between the free theory and the one described by the supergravity dual. Apart
from this, the free and the supergravity-dual (2, 0) theories have exactly the same features
as their counterparts in d = 4. In particular, the same type of splitting (56) for T s occurs in
d = 4 [13,14,16], which merely reflects the similar structure of the supersymmetry algebras
in d = 4 and d = 6.
Finally, we comment once more on the relationship of our results with those obtained in
Refs. [3234]. If we substitute the splitting (56) into the free OPE, then the coefficient in
3
front of the stress tensor, which equals COOT /CT , becomes 2
35 . Here CT = T T is the
coefficient of the two-point function of the stress tensor and COOT is the normalization
constant of the three-point function of two scalars O with the stress tensor. According to
24
Ref. [34], one has CT = 846 and, therefore, we get COOT = 5
3 . The same value also
follows from the conformal Ward identity relating the three-point function OOT to the
two-point function OO. Hence, the coefficient in front of the canonically normalized
CPWA of the stress tensor in the CPWA expansion of the four-point amplitude turns out
48
2
to be COO
T /CT = 175 . If we want to match it with the supergravity result (54), i.e.,
12
3
with the value 175N
3 , we should choose the number of free multiplets to be = 4N .
This is of course a manifestation at the OPE level of the equality between the free and the
supergravity dual integration constants A, B.
5.2. Projection on [02]

This projection gives
O O[02] =
12 C 34
CJ
[02] J[02]
8 x8
N 3 x12
34

27 2
27
v F2 (v) + v 3 F3 (Y ) + v 4 F4 (Y ) + v 4 log vG4 (Y ) .
10
10
(57)
12 denote the orthonormal ClebshGordon coefficient for an

Here and in what follows CJ
irrep J appearing in the tensor product [02] [02] (recall (4)). In (57) the functions
2 3
+ Y +
5 5
represent the contribution of the canonical CPWA of the = 4 scalar which is nothing but
97
+ O(Y ), where the first term
the 1/2 BPS primary operator O. We also find F4 (Y ) = 175
receives, in particular, a contribution from a scalar of free field dimension = 8. On the
other hand, the function G4 (Y ) has the form

27
G4 (Y ) = Y 2 + O Y 3 .
7
Since G4 does not contain a constant term, we conclude that the scalar of free field
dimension 8 transforming in the [02] does not receive corrections to its free field
F2 (Y ) = 1 + Y + ,
F3 (Y ) =
25
dimension. According to the classification of UIRs, this scalar gives rise to a semishort
multiplet from the isolated series (B). Recall that in the d = 4 case the corresponding
operator is also a protected semishort multiplet, however, there unitarity puts it at the bound
of the continuous series (A). Its protection can be understood as a consequence of the threepoint function selection rules. The first operator in (57) receiving an anomalous dimension
is a second-rank tensor of approximate dimension 10.
O O[20] =

12 C 34
CJ
7 2
7 3
[20] J[20]
4
v
v
F
(v)
+
F
(Y
)
+
v
log
vG
(Y
)
,
2
3
4
8 x8
10
10
N 3 x12
34
where
6
3
3
F2 (Y ) = Y + Y 2 + ,
F3 (Y ) = Y + Y 2 +
2
7
7
are precisely the contributions of the CPWA of the = 5 R symmetry current. The function
G4 (Y ) = 8Y + O(Y 2 ), therefore the first operator in the [20] receiving an anomalous
dimension is a vector of approximate dimension 9.
12 C 34
CJ
4

[40] J[40]
O O[40] =
v F4 (v) + v 5 F5 (Y ) + v 5 log vG5 (Y ) .
8
8
3
N x12x34
Therefore, all traceless symmetric rank-2k tensors of twist 8 transforming in the [40]
are non-renormalized. The explicit form of the function G5 shows that the first operator
acquiring an anomalous dimension is a scalar of approximate dimension 10.
12 C 34
CJ

4
[04] J[04]
O O[04] =
v F4 (v) + v 5 F5 (Y ) + v 6 log vG6 (Y ) .
8
8
3
N x12x34
Such a structure shows that all rank-2k tensors of twists 8 and 10 are non-renormalized.10
The first operator receiving an anomalous dimension is a scalar of approximate dimension 12.
10 Note that the channel [04] contains two towers of protected tensors. The same behavior occurs for the irrep
[040] in the d = 4 theory [13].
26

O O[22] =
12 C 34
CJ

4
[22] J[22]
v F4 (v) + v 5 F5 (Y ) + v 5 log vG5 (Y ) .
8
8
3
N x12x34
The function F4 (v) comprises contributions of rank-(2k + 1) tensors of = 9. Since the

term v 4 log v is absent, all these tensors are non-renormalized. The first operator with
anomalous dimension turns out to be a vector of approximate dimension 11.
This completes our OPE analysis. We see that there are several towers of traceless
symmetric tensors in the irreps [40], [04] and [22] with vanishing anomalous dimensions.
This is again in complete agreement with the classification of the UIRs, because the
corresponding superconformal primary operators belong to the isolated series (D), i.e., they
are BPS short. Operators with anomalous dimensions in the first five channels of (4) are
supersymmetry descendants of the superconformal primary operators in the R symmetry
singlet. As to the OPEs in the supergravity regime, we conclude that the d = 4 and d = 6
theories have an identical structure though their relation with the free field limit is of
completely different nature.
Acknowledgements
We are grateful to F. Bastianelli, S. Ferrara, P. Fr, S. Frolov, S. Minwalla, A. Petkou,
K. Skenderis, S. Theisen and A. Tseytlin for useful discussions. G.A. would like to thank
the CERN Theory Division, where part of this work was done, for the kind hospitality. G.A.
was supported by the DFG and by the European Commission RTN programme HPRN-CT2000-00131, and in part by RFBI grant N99-01-00166 and by INTAS-99-1782.
Appendix A
For completeness here we recall the general solution of the d = 4 Ward identities [25].
The integrability condition for the system (7) reads

a1 a3

= 0.
sss + tt t + (s + t 1)st + 2s + 2t
(A.1)
s
t
This equation can be integrated to give

s
a1 (s, t) = h(s) h(t) + sF (s, t),
(A.2)

t
a3 (s, t) = h(t) h(s) + tF (s, t).
(A.3)
Here F (s, t) is an a priori arbitrary symmetric function and h is a function of a single

variable. In particular, the constant (free) solution a1 = a3 = A corresponds to

1
A
As+t
s +
,
F0 (s, t) =
.
h0 (s) =
(A.4)
2
s
2 st
Integrating the equation for b2 (s, t) one gets

b2 (s, t) = B h(t) + h(s) + (1 s t)F (s, t).
27
(A.5)
Shifting the solution by the free values, h h + h0 , F F + F0 leaves b2 (s, t)

unchanged. Separating the trivial solution and regarding the remaining h and F as
independent functions, the crossing symmetry relation for the four-point amplitude takes
the form (9) for F together with the following condition on h:

h(s) + h(t) ss/t,t 1/t = const.
(A.6)
This completes our discussion of the d = 4 Ward identities.
Appendix B
I
The matrices CijI and Cm;l
introduced in Eq. (41) are subject to the following
normalization condition:

1
1
1
I
CijI Ckl
= ik j l + il j k ij kl ,
2
2
5
I

1
I
I
Ci;j
Ck;l
= (ik j l il kj ).
(B.1)
2
I
J
The projectors P1234
projecting the four-point amplitude (5) on the contributions of
different R-symmetry irreps J are constructed by using the technique of Ref. [13]. We
find the following formulae:
1
12 34 ,
196

10
1
[02]
+
C1234
=
12 34 ,
P1234
189
5
2
[20]
P1234
= C1234
,
35

1
1
4 +
[40]
P1234 =
13 24 + 14 23 + 12 34 2C1324 C1234 ,
105
6
3

1
8
16 +
[04]
P1234 =
13 24 + 14 23 + 12 34 + 4C1324 C1234 ,
330
63
9

1
8
[22]
13 24 14 23 + C1234
.
P1234
=
162
7
[00]
=
P1234
In particular, the projectors are normalized to obey the condition (PD )2 = 1/J , where J
is the dimension of the representation J :
dim[00] = 1,
dim[22] = 81.
dim[02] = 14,
dim[20] = 10,
dim[04] = 55,
dim[40] = 35,
28
When working out the action of the projection operators on the four-point amplitude
(5), the following contractions prove helpful:
3199
6671
+
+
,
C1234
,
C1234
=
50
200
273
245
,
C1234
,
C2134 C1234 =
C1234
=
100
8
196
21
,
C1212 = .
C1122 =
5
5
C1234 C1234 =
+
C1234
C1324 =
273
,
100
C1234
C1324 = 0,
(B.2)
Appendix C
The D-functions related to the space AdSd+1 can be defined by the formula
d
d w dw0
D1 2 3 4 (x1 , x2 , x3 , x4 ) =
Ki (xi , w),
w0d+1 i
where
K (x, w) =
w0
2
w0 + ( x)2
(C.1)
and the integral is taken over the space parametrized by w = (w0 , ), being a ddimensional vector. From this definition one can deduce the following Feynman parameter
representation
D1 2 3 4 (x1 , x2 , x3 , x4 )

( j j 1)
d/2 ( d
) ( 2 )
j 1
2

=
dj j
,

2 )/2
2 i (i )
( k<l k l xkl
(C.2)

where = i i .

We also define the D-functions
[13] which are viewed here as functions of the conformal
cross-ratios s, t
D1 2 3 4 (x1 , x2 , x3 , x4 )
1 2 3 4 (s, t)
2D
d
2 )
(x12
1 +2 3 4
2
2 )
(x13
1 +3 2 4
2
2 )
(x23
2 +3 +4 1
2
(C.3)
2 )4
(x14
These functions have the following transformation properties

1 (1 2 3 4 )
2
1 2 3 4 (t, s) = t
1 4 3 2 (s, t),
D
D
s

1
1 2 3 4 s , 1 = t 2 (1 2 3 4 ) D
1 2 4 3 (s, t),
D
t t
which can be easily proven by using, e.g., the Schwinger parameter representation.
(C.4)
(C.5)
29
We find the following representations for the coefficients A,0 of the four-point
amplitude in terms of differential operators acting on (s, t):
3
A01234 =
2 x 2 )3 x 2 x 2
(x12
34 13 24
1
3 3 (2 ss )(1 ss )(ss )2
2 3

25
1
1 t
+
+ 5 3
(2 ss )(1 ss )2 (ss )2
2 3 s
s
4

31
1 t
1
+ +
(2 ss )2 (1 ss )2 (ss )2
+ 5 3
s
4
2 3 s

5
1 t
+ 1 (3 ss )(2 ss )2 (1 ss )2 (ss )2
+ 7 3
2 3 s
s
(s, t),
(C.6)
3
A+
1234 =
2 x 2 )3 x 2 x 2
(x12
34 13 24
1
3 2 (2 ss )(1 ss )(ss )2
2 3
1
7
+ 4 (2 ss )(1 ss )2 (ss )2 3 3 (2 ss )2 (1 ss )2 (ss )2
2 3
2 3
5
(C.7)
+ 4 3 (3 ss )(2 ss )2 (1 ss )2 (ss )2 (s, t),
2 3
A
1234 =
3
2 x 2 )3 x 2 x 2
(x12
34
13 24

1
1
t
2 3
(2 ss )2 (1 ss )2 (ss )2
2 3 s
s
5
+ 4 2 (1 + ss + 2tt )(1 ss )2 (ss )2
2 3

1
1
1
t
2
2
2
+ 4 2
(1 ss ) (ss ) + 3 2 (1 + ss + 2tt )(ss )
2 3 s
s
2 3
(s, t).
(C.8)
Further simplification is achieved by successive use of the identities [25]

1
s+t 1
ln t ,
s (s, t) = 2 (s, t)(1 s + t) + 2 ln s
s

1
s+t 1
t (s, t) = 2 (s, t)(1 t + s) + 2 ln t
ln s .
t
We recall the following Feynman parameter representation for (s, t) [41]:

( j j 1)
1
(s, t) =
dj
.
2 x2
2 )2
x13
( k<l k l xkl
24
j
(C.9)
30
This formula, together with (C.2) and (C.3) can be used to derive differential relations

between the D-functions
and (s, t). For instance, we apply to both sides of (C.9) the
3
differential operator 2 2 2 and use the definition of the variables s, t to express
x14 x13 x12
1
= 2 ss ,
2
x12
x12
1
= 2 tt ,
2
x14
x14
1
= 2 (ss + tt ).
2
x13
x13
(C.10)
Thus we find
1
(1 + ss + tt )stst (s, t)
2 x2 x2 x4
x12
24 14 13

13 2 3 4 ( j j 1)
= 24
dj
.

2 )5
( k<l k l xkl
j
(C.11)
4222 .
We then use (C.2) and (C.3) for d = 4 to rewrite the r.h.s. of this identity in terms of D
In this way we obtain the relation
4222 = 1 t (1 + ss + tt )stst (s, t),
D
(C.12)
6
which is used to recast the strongly-coupled d = 4, N = 4 potential, obtained in Ref. [25]
from the results of Ref. [20], into the form (15). Similar considerations are used to establish
the equivalence of the representations (49) and (50) for F (s, t).
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231.
[2] G.G. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] E. Witten, Some comments on string dynamics, Contributed to STRINGS 95: Future Perspectives in String
Theory, Los Angeles, CA, 1318 March 1995, In Los Angeles 1995, Future perspectives in string theory,
501523, hep-th/9507121;
A. Strominger, Open p-branes, Phys. Lett. B 383 (1996) 44, hep-th/9512059;
N. Seiberg, Notes on theories with 16 supercharges, Nucl. Phys. Proc. Suppl. 67 (1998) 158, hep-th/9705117.
[5] L. Andrianopoli, S. Ferrara, KK excitations on AdS(5) S(5) as N = 4 primary superfields, Phys. Lett.
B 430 (1998) 248, hep-th/9803171.
[6] V.K. Dobrev, V.B. Petkova, Phys. Lett. B 162 (1985) 127;
V.K. Dobrev, V.B. Petkova, Fortschr. Phys. 35 (1987) 7;
V.K. Dobrev, V.B. Petkova, Fortschr. Phys. 35 (1987) 537;
S. Minwalla, Restrictions imposed by superconformal invariance on quantum field theories, Adv. Theor.
Math. Phys. 2 (1998) 781, hep-th/9712074;
V.K. Dobrev, Positive energy unitary irreducible representations of D = 6 conformal supersymmetry, hepth/0201076.
[7] G. Arutyunov, B. Eden, E. Sokatchev, On non-renormalization and OPE in superconformal field theories,
Nucl. Phys. B 619 (2001) 359, hep-th/0105254.
[8] B. Eden, E. Sokatchev, On the OPE of 1/2 BPS short operators in N = 4 SCFT(4), Nucl. Phys. B 618 (2001)
259, hep-th/0106249.
31
[9] B. Eden, S. Ferrara, E. Sokatchev, (2, 0) superconformal OPEs in D = 6, selection rules and nonrenormalization theorems, JHEP 0111 (2001) 020, hep-th/0107084.
[10] P.J. Heslop, P.S. Howe, A note on composite operators in N = 4 SYM, Phys. Lett. B 516 (2001) 367, hepth/0106238.
[11] P.J. Heslop, P.S. Howe, OPEs and three-point correlators of protected operators in N = 4 SYM, hepth/0107212.
[12] S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor.
Phys. 40 (2001) 935, hep-th/0005151.
[13] G. Arutyunov, S. Frolov, A.C. Petkou, Operator product expansion of the lowest weight CPOs in N = 4
SYM(4) at strong coupling, Nucl. Phys. B 586 (2000) 547, hep-th/0005182.
[14] G. Arutyunov, S. Frolov, A. Petkou, Perturbative and instanton corrections to the OPE of CPOs in N = 4
SYM(4), Nucl. Phys. B 602 (2001) 238, hep-th/0010137.
[15] S. Ferrara, E. Sokatchev, Universal properties of superconformal OPEs for 1/2 BPS operators in 3 D 6,
New J. Phys. 4 (2002) 1, hep-th/0110174.
[16] D. Anselmi, The N = 4 quantum conformal algebra, Nucl. Phys. B 541 (1999) 369, hep-th/9809192.
[17] G. Arutyunov, B. Eden, A.C. Petkou, E. Sokatchev, Exceptional non-renormalization properties and OPE
analysis of chiral four-point functions in N = 4 SYM(4), Nucl. Phys. B 620 (2002) 380, hep-th/0103230.
[18] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Properties of the Konishi multiplet in N = 4 SYM theory,
JHEP 0105 (2001) 042, hep-th/0104016;
S. Penati, A. Santambrogio, Superspace approach to anomalous dimensions in N = 4 SYM, Nucl. Phys.
B 614 (2001) 367, hep-th/0107071.
[19] F.A. Dolan, H. Osborn, Superconformal symmetry, correlation functions and the operator product expansion,
hep-th/0112251.
[20] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM(4) in supergravity
approximation, Phys. Rev. D 62 (2000) 064016, hep-th/0002170.
[21] M. Pernici, K. Pilch, P. Nieuwenhuizen, Gauged maximally extended supergravity in seven dimensions,
Phys. Lett. B 143 (1984) 103.
[22] P.S. Howe, P.C. West, Int. J. Mod. Phys. A 14 (1999) 2659, hep-th/9509140;
P.S. Howe, P.C. West, Phys. Lett. B 389 (1996) 273, hep-th/9607060;
P.S. Howe, P.C. West, Is N = 4 YangMills soluble? in Moscow, 1996, hep-th/9611074;
P.S. Howe, P.C. West, Phys. Lett. B 400 (1997) 307, hep-th/9611075.
[23] P.S. Howe, P.C. West, Phys. Lett. B 400 (1997) 307, hep-th/9611075.
[24] B. Eden, P.S. Howe, A. Pickering, E. Sokatchev, P.C. West, Nucl. Phys. B 581 (2000) 523, hep-th/0001138.
[25] B. Eden, A.C. Petkou, C. Schubert, E. Sokatchev, Partial non-renormalisation of the stress-tensor four-point
function in N = 4 SYM and AdS/CFT, Nucl. Phys. B 607 (2001) 191, hep-th/0009106.
[26] F.A. Dolan, H. Osborn, Conformal four point functions and the operator product expansion, Nucl. Phys.
B 599 (2001) 459, hep-th/0011040.
[27] K.A. Intriligator, Bonus symmetries of N = 4 super-YangMills correlation functions via AdS duality, Nucl.
Phys. B 551 (1999) 575, hep-th/9811047.
[28] E.S. Fradkin, M.Y. Palchik, V.N. Zaikin, Conformally invariant Greens functions with the current and
energymomentum tensor, Phys. Lett. B 57 (1975) 364.
[29] F. Gonzalez-Rey, I. Park, K. Schalm, Phys. Lett. B 448 (1999) 37, hep-th/9811155;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 557 (1999) 355, hep-th/9811172;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 466 (1999) 20, hep-th/9906051;
B. Eden, C. Schubert, E. Sokatchev, Phys. Lett. B 482 (2000) 309, hep-th/0003096;
B. Eden, C. Schubert, E. Sokatchev, Four-point functions of chiral primary operators in N = 4 SYM, Talk
given by C. Schubert at Quantization, Gauge Theory and Strings dedicated to the memory of E.S. Fradkin,
Moscow, 510 June, 2000, hep-th/0010005;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On the logarithmic behavior in N = 4 SYM theory, JHEP 9908
(1999) 020, hep-th/9906188;
M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM theory at order g**4,
Nucl. Phys. B 584 (2000) 216, hep-th/0003203.
[30] M. Bianchi, M.B. Green, S. Kovacs, G. Rossi, Instantons in supersymmetric YangMills and D-instantons
in IIB superstring theory, JHEP 9808 (1998) 013, hep-th/9807033.
32
[31] B. Eden, P.S. Howe, P.C. West, Nilpotent invariants in N = 4 SYM, Phys. Lett. B 463 (1999) 19, hepth/9905085.
[32] I.R. Klebanov, World-volume approach to absorption by non-dilatonic branes, Nucl. Phys. B 496 (1997)
231, hep-th/9702076;
S.S. Gubser, I.R. Klebanov, A.A. Tseytlin, String theory and classical absorption by three-branes, Nucl.
Phys. B 499 (1997) 217, hep-th/9703040.
[33] F. Bastianelli, S. Frolov, A.A. Tseytlin, Three-point correlators of stress tensors in maximallysupersymmetric conformal theories in d = 3 and d = 6, Nucl. Phys. B 578 (2000) 139, hep-th/9911135.
[34] F. Bastianelli, S. Frolov, A.A. Tseytlin, Conformal anomaly of (2, 0) tensor multiplet in six dimensions and
AdS/CFT correspondence, JHEP 0002 (2000) 013, hep-th/0001041;
R. Manvelian, A.C. Petkou, A note on R-currents and trace anomalies in the (2, 0) tensor multiplet in d = 6
and AdS/CFT correspondence, Phys. Lett. B 483 (2000) 264, hep-th/0003017;
R. Manvelian, A.C. Petkou, The trace anomaly of the (2, 0) tensor multiplet in background gauge fields,
JHEP 0006 (2000) 003, hep-th/0005256.
[35] A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 1 (1984) 469.
[36] G.G. Hartwell, P.S. Howe, Int. J. Mod. Phys. A 10 (1995) 39013920, hep-th/9412147;
G.G. Hartwell, P.S. Howe, Class. Quantum Grav. 12 (1995) 18231880.
[37] G. Arutyunov, S. Frolov, Three-point Green function of the stress-energy tensor in the AdS/CFT
correspondence, Phys. Rev. D 60 (1999) 026004, hep-th/9901121;
R. Corrado, B. Florea, R. McNees, Correlation functions of operators and Wilson surfaces in the d = 6,
(0, 2) theory in the large N limit, Phys. Rev. D 60 (1999) 085011, hep-th/9902153;
F. Bastianelli, R. Zucchini, Three-point functions of chiral primary operators in d = 3, N = 8 and d = 6,
N = (2, 0) SCFT at large N , Phys. Lett. B 467 (1999) 61, hep-th/9907047;
F. Bastianelli, R. Zucchini, Three-point functions for a class of chiral operators in maximally supersymmetric
CFT at large N , Nucl. Phys. B 574 (2000) 107, hep-th/9909179;
F. Bastianelli, R. Zucchini, Three-point functions of universal scalars in maximal SCFTs at large N ,
JHEP 0005 (2000) 047, hep-th/0003230;
E. DHoker, B. Pioline, Near-extremal correlators and generalized consistent truncation for AdS(4|7)
S(7|4), JHEP 0007 (2000) 021, hep-th/0006103.
[38] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS(7) S(4) reduction and the origin of
self-duality in odd dimensions, Nucl. Phys. B 581 (2000) 179, hep-th/9911238.
[39] E. DHoker, D.Z. Freedman, L. Rastelli, AdS/CFT four-point functions: How to succeed at z-integrals
without really trying, Nucl. Phys. B 562 (1999) 395, hep-th/9905049.
[40] G. Arutyunov, S. Frolov, Scalar quartic couplings in type IIB supergravity on AdS(5) S(5), Nucl. Phys.
B 579 (2000) 117, hep-th/9912210.
[41] N.I. Ussyukina, A.I. Davydychev, Phys. Lett. B 298 (1993) 363.
[42] L. Hoffmann, L. Mesref, W. Rhl, Conformal partial wave analysis of AdS amplitudes for dilaton axion
four-point functions, Nucl. Phys. B 608 (2001) 177, hep-th/0012153;
L. Hoffmann, L. Mesref, A. Meziane, W. Rhl, Multi-trace quasi-primary fields of N = 4 SYM(4) from
AdS n-point functions, hep-th/0112191.

Open string models with ScherkSchwarz SUSY

breaking and localized anomalies
C.A. Scrucca a , M. Serone b , M. Trapletti b
a CERN, 1211 Geneva 23, Switzerland
b ISAS-SISSA, Via Beirut 2-4, 34013 Trieste, and INFN, Trieste, Italy
Received 12 April 2002; received in revised form 8 May 2002; accepted 15 May 2002
Abstract
We study examples of chiral four-dimensional IIB orientifolds with ScherkSchwarz supersymmetry breaking, based on freely acting orbifolds. We construct a new Z3 Z3 model, containing only
D9-branes, and rederive from a more geometric perspective the known Z6 Z2 model, containing
D9, D5 and
D5 branes. The cancellation of anomalies in these models is then studied locally in the
internal space. These are found to cancel through an interesting generalization of the GreenSchwarz
mechanism involving twisted RamondRamond axions and 4-forms. The effect of the latter amounts
to local counterterms from a low-energy effective field theory point of view. We also point out that
the number of spontaneously broken U (1) gauge fields is in general greater than what expected from
a four-dimensional analysis of anomalies. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Supersymmetry (SUSY) is certainly one of the key ideas to understand how to embed
the Standard Model (SM) into a more fundamental microscopic theory. It provides, among
other things, an elegant solution to the hierarchy problem (stabilizing the electroweak
scale) if broken at a sufficiently low scale Msusy TeV. In string theory, SUSY plays
an even more important role and apparently represents a crucial ingredient in defining
absolutely consistent models. In fact, the construction of truly stable non-SUSY string
vacua is tremendously hard and no such model has been found so far. While waiting for new
principles or breakthroughs that hopefully will shed light on this fundamental problem, it is
E-mail address: serone@sissa.it (M. Serone).

0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 0 0 - 5
34
C.A. Scrucca et al. / Nuclear Physics B 635 (2002) 3356
nevertheless very important to explore the structure and main properties of SUSY breaking
in string theory.
One of the most interesting and promising mechanisms of symmetry breaking in
theories with compact extra-dimensions, such as string theory, is the so-called Scherk
Schwarz (SS) symmetry-breaking mechanism [1], which consists in suitably twisting the
periodicity conditions of each field along some compact directions. In this way, one obtains
a non-local, perturbative and calculable symmetry-breaking mechanism. String models
of this type can be constructed by deforming supersymmetric orbifold [2] models, and a
variety of four-dimensional (4D) closed string models, mainly based on Z2 orbifolds, have
been constructed in this way [3]. More in general, SS symmetry breaking can be achieved
through freely acting orbifold projections [4]. This fact has recently been exploited in [5]
to construct a novel class of closed string examples, including a model based on the Z3
orbifold. Unfortunately, a low compactification scale is quite unnatural for closed string
models, where the fundamental string scale Ms is tied to the Planck scale, and can be
achieved only in very specific situations [6] (see also [7]). The situation is different for
open strings, where Ms can be very low [8], and interesting open string models with SS
SUSY breaking have been derived in [911]. Recently, the SS mechanism has been object
of renewed interest also from a more phenomenological bottom-up viewpoint, where it
has been used in combination with orbifold projections to construct realistic 5D non-SUSY
extensions of the SM [12,13].
The main aim of this paper is to exploit the general ideas proposed in [5] to construct
chiral IIB compact orientifold models with SS supersymmetry breaking. We derive a new
Z3 Z3 orientifold by applying a freely acting Z3 projection defined as a translation of
order 3 and a non-SUSY twist to the known SUSY Z3 orientifold [14,15]. The model turns
out to be chiral and extremely simple, since only D9-branes are present. It exhibits SS
SUSY breaking in both the closed and open string sectors. All the gauginos are massive,
but there is an anomalous spectrum of massless charginos. The model is classically stable,
since all massless NeveuSchwarzNeveuSchwarz (NSNS) and RamondRamond (RR)
tadpoles vanish, and potential tachyons can be avoided by taking a sufficiently large volume
for the SS torus, i.e., the torus where the translation acts. We also rederive from a more
geometrical perspective the Z6 Z2 model of [10] (see also [16]), by applying to the
SUSY Z6 model of [15] a freely acting Z2 projection generated by a translation of order 2
along a circle combined with a ()F operation, where F is the 4D spacetime fermion
number operator. We then discuss in some detail its rich structure involving D9, D5 and

D5 branes.
An other important goal of this work is to perform a detailed study of local anomaly
cancellation (i.e., point-by-point in the compact space) for this kind of models. This study
is motivated by the results of [17] where it has been pointed out that orbifold field theories
can have anomalies localized at fixed points that vanish when integrated over the internal
space, and originate from loops of heavy KaluzaKlein (KK) modes. To this aim, we will
extend the approach that has been followed in [18] for 4D SUSY orientifolds to distinguish
between different points in the internal space. We find that all anomalies cancel locally,
thanks to an interesting GreenSchwarz (GS) mechanism [19] involving twisted RR axions
belonging to 4D sectors localized at fixed points or 6D sectors localized at fixed-planes,
as found in [18,20], but also 4-forms coming from 6D sectors localized at fixed planes.
35
The latter effect arises whenever RR tadpoles are cancelled globally but not locally,1
and involves only heavy KK modes of the 4-forms. In non-compact string vacua, such
as intersecting branes, this kind of effect is already included in the usual anomaly inflow
of [22]. Global irreducible anomalies can arise in this case, since there is no constraint on
the global RR flux; they are cancelled thanks to RR forms propagating in more than 4D.
This shows once again the very close relation between the GS mechanism and the inflow
mechanism of [23], even for irreducible terms.
Our results reveal an important distinction between anomalies appearing through a 6form in the anomaly polynomial and anomalies appearing through the product of a 2-form
and a 4-form. In the former case, the GS mechanism is mediated by twisted RR 4-forms
and the corresponding symmetry is linearly realized. In the latter, instead, anomalies are
cancelled by a GS mechanism mediated by twisted RR axions, and the symmetry is realized
only non-linearly. When applied to a U (1) factor with an anomaly that is globally but not
locally vanishing, these two situations lead respectively to a massless and massive 4D
photon.2 This leads to the important conclusion that the number of spontaneously broken
U (1) gauge factors is in general greater than what is expected from a global analysis
of anomalies. This fact, which has not been appreciated so far in the literature, could
have an important impact in the context of open string phenomenology. The difference
between the two mechanisms involving axions and 4-forms is particularly striking from a
4D low-energy effective field theory point of view, where heavy KK modes are integrated
out. The axions remain dynamical, but the 4-forms must be integrated out, and we will
show that their net effect then amounts to a local 6D ChernSimons counterterm with a
discontinuous coefficient, jumping at the fixed points; this counterterm thus occurs in a
way that is manifestly compatible with local supersymmetry and falls in the category of
terms discussed in [25] (see also [26]). This realizes a 6D version of the possibility of
cancelling globally vanishing anomalies through a dynamically generated ChernSimons
term [17]. It also confirms in a string context that operators that are odd under the orbifold
projection can and do in general occur in the 4D effective theory with odd coefficients, as
emphasized in [27].
The paper is organized as follows. In Section 2 we rederive the model presented in [10]
as a Z6 Z2 orientifold, emphasizing geometrical aspects. In Section 3, the novel Z3 Z3
model is constructed and described in some detail. Section 4 is devoted to the study of local
anomaly cancellation, and contains both a general discussion and a detailed analysis for
the two models at hand. In Section 5 we state our conclusions. Finally, some more details
concerning lattice sums and anomalous couplings are reported in Appendices A and B,
where we also clarify an issue left partially unsolved in [18], regarding the factorization of
anomalous couplings in twisted sectors with fixed planes.
1 The global cancellation of RR tadpoles ensures only the global cancellation of cubic irreducible anomalies;
see, e.g., [21].
2 As in the standard case [24], a pseudo-anomalous photon can become massive by eating an axion through a
Higgs mechanism.
36
2. The Z6 Z2 model

The Z6 Z2 orientifold of [10] is obtained by applying a SUSY-breaking Z2 projection
to the SUSY Z6 model of [15]. The Z6 group is generated by , acting as rotations of angles
2vi in the three internal tori Ti2 (i = 1, 2, 3), with vi = 1/6(1, 3, 2). The Z2 group is
instead generated by , acting as a translation of length R along one of the radii of T22
(that we shall call SS direction in the following), combined with a sign ()F , where F is
the 4D spacetime fermion number. Beside the O9-plane, the model contains O5-planes
at y = 0 and y = R along the SS direction (as the corresponding SUSY model [15])
and
O5-planes at y = R/2 and y = 3R/2 along the SS direction (see Figs. 1 and 2),
corresponding to the two elements of order 2, 3 and 3 . In order to cancel both NSNS
and RR massless tadpoles, D9, D5 and
D5-branes must be introduced.
2.1. Closed string spectrum
The main features of the closed string spectrum of the Z6 Z2 model can be deduced
from those of the Z6 model, which can be found in [15]. The only SUSY-breaking
generators are , 2 and 4 ; all the other elements preserve some SUSY (generically
Fig. 1. The fixed-points structure in the Z6 Z2 model. We label the 12 fixed points with P1bc and the 12
fixed points with P1bc , each index referring to a T 2 , ordered as in the figure. Similarly, we denote with Pac the
9 2 fixed planes filling the second T 2 , and respectively with Pa b and Pa b the 16 3 fixed and 3 fixed
D5-branes are located at point 1 in the first T 2 , fill the
planes filling the third T 2 . The 32 D5-branes and the 32
third T 2 , and sit at the points 1 and 1 respectively in the second T 2 .
Fig. 2. Brane positions along the SS direction for the Z6 Z2 model. The different supersymmetries left unbroken
at the massless level in the 55 and 5 5 sectors are also indicated.
37
Table 1
Massless closed string spectrum for Z6 Z2 and Z3 Z3 models. We used as the generator of Z6 (Z3 ) and
as the generator of Z2 (Z3 ). Hypermultiplets are multiplets of N = 1 SUSY in 6D, while chiral multiplets are
multiplets of N = 1 SUSY in 4D. The SUSY generators are different in the different sectors, as explained in the
text. The two spinors in the untwisted sector of Z3 Z3 have opposite chirality
Sector
Untwisted
twisted
2 twisted
3 twisted
twisted
3 twisted
2 twisted
Z6 Z2
Z3 Z3
1 graviton, 5 scalars
6 chiral multiplets
18 scalars
6 chiral multiplets
6 chiral multiplets
6 chiral multiplets
1 graviton, 11 scalars, 1 + 1 spinors

6 hypermultiplets
9 chiral multiplets
9 chiral multiplets
different from sector to sector). The Z2 projection acts therefore in a SUSY-breaking way
in the untwisted and 2,4 twisted sectors, and in a SUSY-preserving way in the remaining
1,5 and 3 twisted sectors of the Z6 model. In addition, we must consider the new k
twisted sectors.
Consider first the k sectors already present in the Z6 model. In the untwisted sector, one
gets a gravitational multiplet and 5 chiral multiplets of N = 1 SUSY, and the Z2 projection
eliminates all the fermions. In the 2,4 twisted sectors, one gets 9 hypermultiplets of N = 2
SUSY, and the Z2 projection again eliminates all the fermions. Finally, the 1,5 and 3
twisted sectors give each 12 chiral multiplets of N = 1 SUSY, and the Z2 action reduces
this number to 6, since it identifies sectors at fixed points that differ by a R shift in the
position along the SS direction.
Consider next the new k sectors emerging in the Z6 Z2 model. The twisted
sector yields one real would-be tachyon of mass m2 = 2 + R 2 /(2 ). Similarly, the 2
sectors yield 6 complex would-be tachyons of mass m2 = 2/3 + R 2 /(2 ).3 Finally,
the 1,5 twisted and 3 twisted sectors each give 6 chiral multiplets of N = 1 SUSY,
which have opposite chirality and a different unbroken SUSY compared to those arising in
the 1,5 and 3 -sectors, because changes the GSO projection due to the (1)F operation
that it involves.
The closed string spectrum that we have just derived is summarized for convenience in
Table 1.
2.2. Tadpole cancellation
As mentioned above, the -projection in the closed string sector introduces O9, O5,

D5 branes is
O5 planes and hence a non-vanishing number n9 , n5 and n5 of D9, D5 and
needed to cancel all massless tadpoles.
The computation of the partition functions on the annulus (A), Mbius strip (M) and
Klein bottle (K) surfaces and the extraction of the tadpoles is standard, although lengthy,
3 These are clearly the lightest would-be tachyons in both the and 2 twisted sectors, but it should be
recalled that there is actually an infinite tower of such states, with increasing winding mode.
38
and we do not report all the details. The world-sheet parity operator is defined in such
a way that ( ) 1 = (2 ) for a generic world-sheet field ( ), and its action
on the RR and NSNS vacua is given by |0NSNS = |0NSNS and |0RR = |0RR
in the closed string sector and by |0NS = i|0NS and |0R = |0R in the open
string sector. The modular parameter for the A, M and K surfaces is taken to be tA = it,
tM = it 1/2 and tK = 2it. The modular transformation needed to switch from the direct
to the transverse channel is simply S : 1/ for the A and K surfaces. For the M
surface, the appropriate transformation is instead P = T ST 2 ST , where T : + 1. The
corresponding modular parameters in the transverse channel are lA = 1/(2t), lM = 1/(8t)
and lK = 1/(4t).
The only novelty with respect to the Z6 model are the non-SUSY sectors that arise when
the Z2 generator enters as twist or insertion in the trace defining the partition function.
The corresponding contributions to the partition functions can be easily deduced from their
analogues in the Z6 model. In the K amplitude, owing to the presence of , the insertion
of acts only in the lattice contribution, as reported in Eq. (A.13). As a twist, inverts the
GSO projection and acts in the lattice. This implies that the twisted contribution, after
the S modular transformation, will be the same as the SUSY untwisted sector contribution,
but proportional to (1NSNS + 1RR ) instead of the usual (1NSNS 1RR ), and with some terms
dropped due to the vanishing of the lattice contribution as in (A.13). This represents a nonvanishing tadpole for the untwisted RR six-form, and reflects the presence of
O5-planes
(beside O5-planes) in this model. On the A and M surfaces, the insertion of acts in the
lattices as discussed in Appendix A. Apart from that, it simply reverts the R contribution to
the partition function. This simple sign flip has, however, different consequences in the two
surfaces when analyzing the closed string channel, because of the two different modular
transformations (S and P ) that are involved. For the M amplitude, the result is obtained
from its SUSY analogue by replacing the factor (1NSNS 1RR ) by (1NSNS + 1RR ), and
has a clear interpretation as D-branes/
O-planes and
D-branes/O-planes interactions. For
the A amplitude, the action of in the closed string channel reverses the GSO projection
and, depending on which Z6 generator is inserted (and which boundary conditions are
considered), this can lead to an exchange of would-be tachyons.
The group action on the ChanPaton degrees of freedom is encoded in the twist matrices
and , respectively for the Z6 and Z2 generators. The group algebra, as usual, allows us
to write the ChanPaton contribution of a M amplitude with the insertion of n m as
Tr( n m )2 , the freedom of sign being fixed by tadpole cancellation and by the relative
action of on 5 and 9 branes, as studied by Gimon and Polchinski (see for example [15,
28]). Tadpole cancellation and the action fix 6 = I in the 9, 5 and 5 sector, as in [15],
and 2 = I in the 5 and 5 sectors, with the further condition { , } = 0. We also impose
2 = I in the 9 sector; the case 2 = I will be considered later on.
To be fully general, we will use, for the twist matrices in the 5 and 5 sectors, an extra
index that distinguishes between distinct k fixed points (or fixed planes). Similarly, an
other extra index is needed also for the matrices in the 5 and 5 sectors, running over the
k fixed points.
The final form of the massless tadpoles is most conveniently presented by distinguishing
the two closed string sectors with a sign equal to +1 for the NSNS sector and 1 for
39

the RR sector. The result is given by v4 /12 dl times
I:
:
2
2
v3 6
v1 v2 v3 5
2 ,1 n5 n5 ,
2 n9 +
8
8v1 v2
3 4
2
3 1
2 Tr 9 Tr 5b Tr 5b
,
6
c=1 b=1
3 4
2
3 1
2 Tr 9 9 Tr 5b 5 Tr 5b
5 ,
6
(1)
(2)
(3)
c=1 b =1
2:
3:
3
3

2
1 4
v2 3
2
2 2
2 a,1 ,1 + Tr 5ac
+ Tr 5ac
+
2 + Tr 92 ,
4v2
12
a,c=1
a,b=1
(4)
4

3
3 2
v3
22 Tr 93 + Tr 5b
+ Tr 5b
,
(5)
a ,b=1
3 : v3
4

a ,b =1
2

3
3
22 Tr 93 9 + Tr 5b
,
5 + Tr 5
5b
(6)
where we denoted by n m the tadpole contribution of the n m twisted closed string

states, summed over the various fixed points or planes; for convenience we have taken
the sums in Eqs. (2), (3), (5) and (6) to run over closed string twisted states and their
images under some orbifold elements. Moreover, v4 = V4 /(4 2 )2 , vi = Vi /(4 2 )
(i = 1, 2, 3), with V4 being the volume of the four-dimensional spacetime and Vi the
volume of the Ti2 .
The NSNS and RR tadpoles differ mainly through relative signs between the contribution from D5 and
D5 branes. In addition, there are cross-cap contributions to the NSNS
tadpoles in the I and 2 sectors that have no analogue in the RR sector (the terms involving
,1 ).
We also report the lightest massive NSNS tadpoles, where would-be tachyons can
develop:

v1 v2 v3 1 1
:
(7)
q 2
[ Tr 9 ]2 ,
64
2
3
v2 1 1
2 :
(8)
q 6
[ Tr 92 9 ]2 .
24
2
a,b=1
2.3. Open string spectrum

We now turn to the determination of a solution for the ChanPaton matrices satisfying
the above conditions for the global cancellation of massless tadpoles, Eqs. (1)(6). For
simplicity, we consider the case of maximal unbroken gauge symmetry where all D5 and

D5 branes are located respectively at P11 and P11 (see Fig. 1 and its caption). The Z2
40
projection then requires that an equal number of image branes be located respectively at
P12 and P12 . We do not consider the case in which branes and antibranes coincide also
along the SS direction, since this configuration is unstable even classically, because of the
presence of open string tachyons. On the other hand, fixing the branes at antipodal points
along the SS direction allows a metastable configuration without open string tachyons for
sufficiently large SS radius.
The untwisted tadpoles imply n9 = n5 = n5 = 32, whereas a definite solution of the
twisted tadpoles is given by

0
16
9 = 5 = 5 5 =
(9)
,
0 16

0
I
0
I16
,
5 = 5 =
,
9 = 16
(10)
0 I16
I16 0
where ( = exp(i/6)):

2 , 5 I2 , 3 I4 .
16 = diag I2 , 5 I2 , 3 I4 , I
(11)
It is easy to verify that with such a choice all massless tadpoles cancel (although (7) and
(8) do not vanish). Notice that the above choice for 9,5 coincides with that of [15]. The
structure of the twist matrices given in (9) and (10) reflects our choice for brane positions;
in particular, the matrix implements the translation in the ChanPaton degrees of
freedom. Hence, as far as the massless spectrum is concerned, we can effectively restrict
our attention to the 16 branes and antibranes at P111 and P11 1 respectively, and work with
16 16 ChanPaton matrices.
The massless open string spectrum can now be easily derived, and is summarized
in Table 2. In the 99 sector, the bosonic spectrum is unaffected by the Z2 element
and therefore coincides with that of the Z6 orbifold;4 all fermions (both gauginos and
charginos) are instead massive. The 55 and 5 5 sectors are supersymmetric at the massless
level, but with respect to different supersymmetries: N = 1 and N = 1 . The 55 and 5 5
gauge groups G5 and G5 are reduced by the non-trivial action of the translation in these
sectors, and the corresponding states are in conjugate representations. A similar reasoning
also applies for the 95 and 95 sectors. Finally, the 55 sector does not contain massless
states, thanks to the separations between D5-branes and
D5-branes. There are massive
scalars and fermions in the bifundamental of G5 G5 , and charged would-be tachyons of
mass m2 = 1/2 + R 2 /(16 ).
Notice that the above solution of the tadpole cancellation conditions is not unique. In
fact, another interesting and more symmetric solution is obtained by choosing 9 of the
same form as 5 = 5 in (10). This solution is not maximal in the sense that the resulting
G9 gauge group is reduced and equal to G9 = G5 G5 = U (4)2 U (2)4 . However, it
has the nice feature that now also the tadpoles (7) and (8) do vanish. Clearly, there exist
other solutions, which we do not report here. Note for instance that a non-vanishing twist
matrix in the 9 sector can be considered as a Z2 Wilson line along the SS radius. Since
4 These are as in [18], but differ slightly from [15] and [16].
41
Table 2
Massless open string spectrum for Z6 Z2 and Z3 Z3 models. In the 55 sector, chiral multiplets in the
representation of G5 are reported. The matter content of the 5 5 sector is the same as in the 55 sector, but
in conjugate representations of G5 = G5 . In the 95 sector, chiral multiplets are present in representations of
G9 G5 . Again, the matter content in the 95 sector is obtained from that in the 95 sector by conjugation
Z6 Z2
Z3 Z3
G9 :
G5 = G5 :
U (4)2 U (8)
U (2)2 U (4)
SO(8) U (8) U (4)
99 scalars
4,
1), (1, 1, 28),
(4, 4, 1), (4,
(6, 1, 1), (1, 4, 8),
(1, 1, 28),
1, 8), (1, 6,
1), (4,
4, 1),
(4,
8)
(4, 1, 8), (1, 4,
1),
2(8, 8, 1), 2(1, 28,
(8, 1, 4), (1, 1, 6)
2(8, 1, 4), 2(1, 1, 6),

4), (1, 8,
4)
(1, 8,
55 chiral mult.
2,
1), (1, 1, 6),
(2, 2, 1), (2,
(1A , 1, 1), (1, 2, 4),
(1, 1, 6),
2, 1),
1, 4), (1, 1 A , 1), (2,
(2,
4)
(2, 1, 4), (1, 2,
95 chiral mult.
1, 1; 2,
1, 1), (1, 4, 1; 1, 2, 1),
(4,
(1, 1, 8;
2, 1, 1),
(4, 1, 1; 1, 1, 4),
1; 1, 1, 4), (1, 1, 8; 1, 2,
1)
(1, 4,
Model
99 fermions
implements a SS gauge symmetry breaking, this reflects the close interplay between
Wilson line symmetry breaking [29] and SS gauge symmetry breaking.
Let us now comment on the brane content of this orbifold. From the tadpoles, we learn
that there is no local Z6 and Z2 twisted RR charge at all in the model (Tr = Tr 3 = 0),
but there is a Z3 -charge, since Tr 2 = 0, that globally cancels between D9 and O9 planes,
and D5,
D5, O5 and
O5 planes. On a Z6 orbifold, a regular D-brane must have 5 images.
Since we start with 32 branes, it is clear that the branes in this model cannot all be regular.
In fact, the presence of a non-vanishing Z3 RR (and NSNS) charge suggests that fractional
D5 and
D5 branes are present at Z3 fixed planes5 of the orbifold. The configuration is then
the following. We have 2 regular D5 and
D5 branes (and 5 images for each) and 2 fractional
Z3 D5 and
D5 branes (and one Z2 image for each). In our maximal configuration, they are
D5). Clearly, there are the additional Z2 images located
all located at P11 (D5) and P11 (
at P12 and P12 . Regular branes can move around freely, whereas fractional branes are
stuck at the fixed points. However, one can still shift a fractional brane from one fixed point
to another, suggesting that this freedom represents the T -dual of discrete Wilson lines in
orbifolds. Notice that also D9-branes have Z3 RR charge. Although it is not appropriate
to speak about fractional D9-branes, this kind of object represents the T -dual version of
the usual lower-dimensional fractional branes. In some sense, they are stuck in the gauge
bundle, and do not admit continuous Wilson lines, but only discrete ones.
5 In our case, the Z fixed plane is at the origin. However, seen as D7-branes wrapped on vanishing two-cycles
3
[30], these branes wrap only the Z3 vanishing cycles.
42
The Z2 twist acts trivially in the gauge-bundle of the D9-branes, whereas in the 5 sector
it is T -dual to a discrete Wilson line given by the matrix . More precisely, the breaking
of the gauge group in the 5 and 5 sector is the T -dual version of a Wilson line symmetry
breaking. The additional ()F action is on the other hand responsible for the D5
D5
flip for half of the branes.
3. A Z3 Z3 model
It has been shown in [5] that SS symmetry breaking can be obtained also in Z3 models
through a suitable freely acting and SUSY-breaking Z3 projection. In this section, we will
construct a new Z3 Z3 model, based on this structure, that will prove to be much simpler
than the Z6 Z2 model.
The Z3 Z3 orbifold group is defined in the following way [5]. The Z3 generator acts
as a SUSY-preserving rotation with twist vi = 1/3(1, 1, 0), while the Z3 generator acts
as a SUSY-breaking rotation with vi = 1/3(0, 0, 2) and an order-three diagonal translation

in T12 .
3.1. Closed string spectrum
It is convenient to consider first the massless closed string spectrum in the parent
Type IIB orbifold, before the projection. In this case, we get an untwisted sector, and
both SUSY-preserving and SUSY-breaking twisted sectors.
The untwisted sector contains the 4D spacetime part of the NSNS spectrum, i.e., the
graviton, the axion and the dilaton; furthermore, there are 10 scalars arising from fields
with internal indices in the NSNS sector, 12 scalars from the RR sector and 2 spinors for
each chirality from the NSR + RNS sectors.
The twists and 2 act only on two of the three tori, and their action preserves N = 2
and QL,R
in the 4D
SUSY in 6D. More precisely, they preserve the supercharges QL,R
2
3
notation of [5]. Each twisted sector contains a 6D N = 2 tensor multiplet. The states are
located at the Z3 fixed points, and are Z3 -invariant, while Z3 acts exchanging states from
one fixed point to the other, so that in the first torus the three Z3 fixed points are identified.
The twists and ()2 act instead on all the compact space, preserving two
supercharges, QL,R
4 . The twisted spectrum contains a 4D N = 2 hypermultiplet. In these
sectors, the elements and 2 act by exchanging states from one fixed point to the other
in the first torus, so that, as before, there is only one physical fixed point in this torus.
The 2 and ( 2 )2 twisted sectors can be treated similarly, the only difference being the
position of the fixed points and the unbroken supercharges QL,R
1 .
The twists and 2 are SUSY-breaking, and the corresponding
twisted sectors yield
each a real would-be tachyon of mass m2 = 4/3 + 2T2 /(3 3 ) (where T2 is the
imaginary part of the Khler structure of the SS torus) and 16 massive RR 16 scalars. These
states are -invariant and located at fixed points, and again the remaining elements only
switch fixed points.
43
It is now easy to understand the effect of the projection. In the untwisted sector,
removes the axion, half of the NSNS and RR scalars, and half of the fermions. In the
twisted sectors, relates QL to QR and projects away half of the supersymmetries, so
that the surviving states fill supermultiplets of N = 1 SUSY in 4D or 6D. Furthermore,
relates the twist i to ( i )2 , and only half of the corresponding states survive the
projection. The spectrum is therefore reduced to 2 hypermultiplets of 6D N = 1 SUSY
from twists, for each fixed point; 1 chiral multiplet of 4D N = 1 SUSY from
twists, for each fixed point, and the same for 2 twists; 1 real would-be tachyon and
16 massive scalars from twists. The massless closed string spectrum is summarized in
Table 1.
3.2. Tadpole cancellation
The computation of the partition functions on the A, M and K amplitudes and the
extraction of the tadpoles is again standard. The only novelty occurs in the untwisted sector,
with n (n = 1, 2) inserted in the trace. In these sectors, the oscillator contribution to the
partition function is given by
n ( ) =
1/2
ab
a
(2
sin
2n/3)
(
)
b+2n/3 ( )
b
,

1/2
9 ( )
( )
a 3
a,b=0
(12)
1/2+2n/3
and the corresponding partition function on each surface reads:

ZA

1
v4
dt 2in(m1 )
=
e
1 [m]2 [m]n (it)(Tr n )2 ,
n

2NN
64t 3 m
0
1
v4
=
n
8NN
ZM
ZK
1
v4
=
n
2NN
dt 2in(m1 )
e
1 [m]2 [m]n (it 1/2) Tr 2n ,
4t 3 m
dt 2in(m1 )
e
1 m/ 2 2 m/ 2 2n (2it),
4t 3 m
(13)
where NN is the total order of the group (i.e., 9 in our case) and i [m] is the 2D lattice
of the ith torus as defined in (A.3).
The tadpoles for massless closed string modes are easily computed. We skip the explicit
form of the usual 10D tadpole, arising in all orientifold models, that fixes to 32 the
number of D9-branes and requires t = . All other tadpoles are associated to twisted
states occurring only at fixed points or fixed planes. We list them here using the already
introduced notation. We denote the twist matrices associated to the Z3 and Z3 actions by
and , and we assume 3 = I , 3 = I , where , = 1. The tadpoles are at the
9 fixed planes, the 27 fixed points and the 27 2 fixed points; they are given by
44

(1NSNS 1RR )v4 /72 dl times:

2
v3
(8 Tr )2 + 8 Tr 2 ,
:
3
a,b
:
2 :
2
2
1
4 + Tr + 4 + Tr 2 2 ,
3 3 a ,b,c
2
2
1
4 + Tr 2 + 4 + Tr 2 .
3 3 a ,b,c
(14)
We wrote explicitly the contributions from the k and Nk sectors, arising from the same
physical closed string state.
By taking the transverse channel expressions of the amplitudes (13) through S and
P modular transformations, the additional tadpoles for the non-SUSY twisted sectors
arising at the 3 fixed planes Pc can be derived. They yield the following result for the
massive would-be tachyonic NSNS states:

2
1
1
v1 v2 1
2
q 3
1 2m +
(Tr ) + 1 2m
Tr 2
3
3
4 3
c m=

2
2
1 2m
16 Tr 2
3

2
(16 Tr )2 ,
+ 1 2m +
(15)
3
where we have retained the lattice sum along the SS directions. These tadpoles are
associated to massive states for sufficiently large radii along the SS torus, and are therefore
irrelevant in that case. They imply that would-be tachyons and massive RR 7-forms are
exchanged between D9-branes and/or O9-planes. Contrarily to the Z6 Z2 model, there
is no choice for the twist matrix that makes Eq. (15) vanish.
3.3. Open string spectrum
In the following, we take = = 1, because all the other choices lead to equivalent
theories. It is then easy to see that the twisted tadpoles (14) are cancelled by choosing
( = exp 2i/3):

= diag I16 , I8 , 1 I8 ,
(16)
= diag I4 , 1 I4 , I24 .
Notice that the order of the entry in (16) is crucial to cancel the tadpoles; the above choice
is such that 2 = , where is the twist matrix of the 4D N = 1 Z3 model constructed
in [15].
The massless open string spectrum is easily determined. The maximal gauge group is
SO(8) U (8) U (4). The U (8) U (4) factor comes from the U (12) gauge factor of
the 4D N = 1 Z3 model, which is further broken by the Z3 projection. As in the previous
model, this can be interpreted as a Wilson line symmetry breaking. In this perspective,
45
= I and as above, and the tadpoles in (14) are cancelled thanks to a (discrete) Wilson
line W equal to along the first torus in (16). Notice that all the gauginos are massive. The
spectrum of charged massless states is easily obtained and reported in Table 2.
4. Local anomaly cancellation

Chiral string models have generically an anomalous spectrum of massless states, but
it is well known that this does not represent a problem, provided that cubic irreducible
anomalies vanish. Reducible U (1) anomalies are instead cancelled through a 4D version of
the GS mechanism [19], and the corresponding U (1) symmetries are spontaneously broken
[24]. For Type IIB orientifold models, the absence of irreducible anomalies is ensured by
the cancellation of RR tadpoles [21]; the GS mechanism taking care of reducible anomalies
is mediated by twisted RR axions [18,20]. For the models constructed in the previous
sections, the same situation occurs.
The above considerations apply to 4D anomalies, which are determined by the massless
fields. In theories with compact extra-dimensions, however, also massive KK modes can
contribute non-vanishing anomalies, which are localized at the orbifold fixed points and
vanish globally, when integrated over the internal manifold [17]. In fact, global anomaly
cancellation is not sufficient to guarantee the consistency of the theory and anomalies must
therefore cancel locally, i.e., point by point in the internal space. From a 4D effective
field theory point of view, this is due to the fact that localized anomalies vanishing only
globally lead to effective operators that, although suppressed by the compactification scale,
violate 4D gauge invariance. In the light of this observation, we will study below how local
anomaly cancellation is achieved in generic orientifold models, with a detailed analysis for
the particular models discussed in Sections 2 and 3. Interestingly, this will also allow us
to clarify a point that was left partially unsolved in [18], concerning the factorization of
anomalous couplings in twisted sectors with fixed planes; see Appendix B.
As expected, we find that all anomalies are cancelled through a GS mechanism. An
important novelty occurs however for globally vanishing anomalies, that correspond
to an

anomaly polynomial6 I that vanishes when integrated over the orbifold: I = 0. In this
case, the GS mechanism can be mediated not only by RR axions (or their dual 2-forms),
as for globally non-vanishing anomalies, but also by KK modes of RR 4-forms. The
occurrence of one or the other mechanism depends on the way the anomaly is factorized
in terms of forms Xn (F, R) of definite even degree n, constructed out of the gauge and
gravitational curvature 2-forms F and R. If it has the form I X2 X4 , the GS mechanism
will be mediated by twisted RR axions, arising at the fixed points (or fixed-planes) where
the anomaly is distributed. If it has instead the form I X6 , the relevant fields are twisted
RR 4-forms, arising at fixed planes that contain all the fixed points where the anomaly
is distributed. Notice that localized irreducible anomalies are always of the second type,
whereas mixed U (1) anomalies can be of both types. As we shall now illustrate with
6 We use here and in the following the standard characterization of anomalies in D dimensions through a
polynomial of curvatures with total degree D + 2. The anomaly itself is given by the WessZumino descent of I
(see, e.g., [22]).
46
simple and general examples, the fate of the symmetry suffering from a globally vanishing
anomaly is radically different in the two alternative mechanisms.
Consider first anomalies of the type I X2 X4 . In this case, the relevant GS mechanism
can be easily understood by distinguishing anomalous couplings localized at different
points in the internal space. The qualitative novelty can be illustrated by focusing on
the case of a U (1) gauge anomaly distributed at two distinct fixed points z = z1,2 ,
corresponding to a term of the type I = X2 X4 |z1 X2 X4 |z2 in the anomaly polynomial.
This anomaly is cancelled through a GS mechanism mediated by two axions, C01,2 , living
at the two fixed points z1,2 .7 The action is:

2
2
1 1
1 2
4
1
4
2

dC0 + X1 + C0 X4
dC0 + X1 C0 X4 ,
SGS = d x
+ d x
2
2
z1
z2
(17)
where the 1-form X1 denotes the U (1) gauge field associated to the curvature 2-form X2 ,
such that X2 = dX1 . The modified kinetic terms in (17) require that C01,2 = X0 (z1,2 )
under a gauge transformation8 with parameter X0 (x, z), under which X1 = dX0 . The
variation of the X4 terms in (17) then provides the required inflow of anomaly that restores
gauge invariance. The form of the action (17) is fixed by the requirement of having full
gauge invariance, and implies that the U (1) field becomes massive, independently of
whether the anomaly vanishes or not globally. Indeed, one can choose a gauge in which
C01 = C02 , where the kinetic terms in (17) are diagonalized and mass terms for the 4D
gauge field are generated. This fact has not been appreciated so far in the literature, where
only integrated anomalies were studied.
Consider next the case of anomalies of the type I X6 . A globally non-vanishing
anomaly of this kind, associated to a global tadpole for a RR 4-form, would lead to
an inconsistency, because it cannot be cancelled by a standard GS mechanism; indeed,
the latter should be mediated by a RR 4-form, that in 4D is a non-propagating field,
whose dual in 4D would be a manifestly non-physical and meaningless (2)-form [32].
Instead, if the anomaly is globally vanishing, and therefore associated to a local tadpole
for a RR 4-form, the situation is different. The crucial observation is that this type of
anomaly always appears in conjunction with twisted RR states living on fixed planes
rather than fixed points in the internal space. Such states propagate in 6D rather than
4D, and this opens up new possibilities, since a 4-form is now a physical propagating
field and can mediate a GS mechanism. Moreover a 4-form in 6D is dual to a 0-form,
and not to a meaningless (2)-form. However, internal derivatives will play a role and
the corresponding states will thus be massive KK modes from the 4D point of view. The
situation is most conveniently illustrated with a simple example consisting of an irreducible
term in the anomaly polynomial of the form I = X6 |z1 X6 |z2 , where the points z1 and
z2 differ only in the fixed-plane direction. The relevant 6D action for the RR 4-form C4
7 The same basic mechanism works for anomalies localized on fixed-planes, that will thus be cancelled by RR
axions propagating in 4 or 6 dimensions.
8 In our setup the gauge fields that can have anomalies localized at distinct fixed points are in general linear
combinations of fields coming from D9, D5 and
D5-branes.
responsible for the inflow is:

6 1
2
4
4

SGS = d x |dC4 + X5 | + d x C4 d x C4 ,
2
z1
z2
47
(18)
where X5 is the ChernSimons 5-form associated to X6 , such that X6 = dX5 . The kinetic
term in (18) requires that C4 = X4 under a 10D gauge transformation, where X4 is
defined as usual from the gauge variation of X5 : X5 = dX4 . The variation of the second
and third terms in (18) then provides the required inflow of anomaly.9 Contrarily to the
previous case, no U (1) gauge factor is broken by (18). Since C4 enters in (18) only through
massive KK states, it is interesting to understand its effect in the 4D low-energy effective
field theory. In order to do that, notice that the 2-form (z z1 ) (z z2 ) can be written
locally as d(z) for some 1-form (z). Eq. (18) can then be interpreted as a 6D action with
Lagrangian LGS = 12 |dC4 + X5 |2 dC4 . We can now integrate out the massive modes
of C4 and evaluate their action on-shell. This is easily done by substituting back into the
Lagrangian the equations of motion for C4 , that imply dC4 + X5 = (where denotes
1
2
the 6D Hodge operator); it yields Leff
GS = 2 || + X5 . Finally, we obtain the local 6D
ChernSimons term

eff
SGS = d 6 x X5 .
(19)
Note that this gives, at it should, the same gauge variation as the original action, since
(X5 ) = dX4 , which gives d X4 = ((z z1 ) (z z2 ))X4 after integration by
parts. Moreover, the discontinuous coefficient is achieved exactly as proposed in [25],
the only difference being that the involved 4-form is a dynamical field in the full 10D
theory, which behaves like an auxiliary field only in the 4D effective theory. Importantly,
the results of [25] ensure that the term (19) is compatible with local supersymmetry at the
fixed points.
Summarizing, it is clear that there is an important qualitative difference between
anomalies that vanish globally and other
that do not. From a purely 4D effective field

theory point of view, the condition I = 0 on the anomaly polynomial I guarantees
that the corresponding anomaly can be cancelled through the addition of a local Chern
Simons counterterm with a discontinuous coefficient.10 In open string models, however,
anomalies with I X2 X4 always lead to a spontaneous symmetry breaking, and only
those with I X6 are cancelled through a local counterterm. It would be interesting to
understand whether there is some deeper physical principle determining this distinction,
besides factorization properties.
All the above considerations apply qualitatively to any orientifold model. For 6D SUSY
models, for instance, part of the GS mechanism is mediated by untwisted RR forms, and
these can play the same role as 6D twisted sectors in 4D models. In particular, we have
9 In short, localized irreducible 6-form terms in the 4D anomaly polynomial look like reducible terms in a 6D
anomaly polynomial, given by the product of the 6-form term and a field-independent -function 2-form.
10 For example, an orbifold field theory that is globally free of anomalies can be regulated in a gaugeinvariant way by adding heavy PauliVillars fields with mass terms that also have a discontinuous coefficient;
the appropriate ChernSimons term is then automatically generated when integrating out the regulator [27,33].
48
Fig. 3. The fixed-point structure in the Z3 Z3 model. We label the 9 fixed planes with Pab , the 27 fixed
points with Pa bc , the 27 2 fixed points with Pa bc , and the 3 fixed planes with Pc .
verified local anomaly cancellation in the SUSY 6D Z2 model of [28,34]. In the case of
maximal unbroken gauge group with all D5-branes at a same fixed point, irreducible Tr F94
and Tr R 4 terms in the anomaly polynomial do not vanish locally and are indeed cancelled
by a local GS mechanism similar to that described in (18), but mediated by untwisted RR
6-forms propagating in 10D. Again, from a 6D effective theory point of view, these amount
to a local ChernSimons term.
Let us now be more concrete and apply the general arguments outlined above to the
Z6 Z2 and Z3 Z3 models. We will begin with the Z3 Z3 model, which does not have
irreducible anomalies at all, and then analyze the more complicated Z6 Z2 model, where
some are present. Fortunately, the techniques of [18] can be easily generalized to study the
local structure of anomalies, and some details of the analysis are reported in Appendix B.
It is convenient to define abc as a 6D Dirac -function in the internal orbifold, localized at
the fixed point with positions labelled a, b and c in the three T 2 s respectively, as reported
in Figs. 1 and 3. We also define ab as a 4D -function in the internal space, localized at
the fixed planes with positions a and b in the first two T 2 s, and similarly for ac and bc .
Moreover, we will denote by Fi the field strength of the ith factor of the gauge group,
D-brane sector, and with
ordered as in Table 2 (i = 1, 2, 3 in all cases), in the (9, 5 or 5)
tr the traces in fundamental representations of the gauge groups.
4.1. Z3 Z3 model
The anomaly polynomial for the Z3 Z3 model is easily computed, and is encoded in
Eqs. (B.4) and (B.5). Its explicit form is given by ( = 37/4 ):

I = 2
(20)
a bc X29 X49 + 4Z4 a bc Y29 Y49 + 4Z4 .
a,b,c
The quantities Xn , Yn and Zn are combinations of curvatures with total degree n,

and are obtained by expanding the topological charges of D-branes and fixed points as
defined in Eqs. (B.1)(B.3). Their explicit expressions are (mi = (1, 1, 0), ni = (1, 1, 0),
si = (1, 1, 1)):
Y29 = 3 ni tr Fi ,
X29 = 3 mi tr Fi ,
(21)

1
1
1
tr R 2 ,
tr R 2 .
Z4 =
X49 = Y49 = si tr Fi2 +
(22)
2
12
192
49
There are two anomalous combinations of U (1) factors, X19 and Y19 , defined by X29 = dX19 ,
Y29 = dY19 . These have opposite anomalies at the two types of fixed points. This means that
the combination X19 Y19 has true 4D anomalies, whereas X19 + Y19 suffers only from a
globally vanishing anomaly of the type corresponding to (17). It can easily be verified that
the integrated anomaly coincides with the contribution of the massless chiral fermions in
the representations reported in Table 2.
The leading order couplings (arising from the disk and cross-cap surfaces) that are
responsible for the cancellation of these anomalies are easily obtained from the anomalous
couplings for D9-branes and fixed points, Egs. (B.6) and (B.7). One finds:

L=
a bc da bc X19 a bc X49 + 4Z4
a,b,c

a bc da bc Y19 + a bc Y49 + 4Z4 .
(23)
a,b,c
The first term in each row corresponds to the cross-term in a mixed kinetic term of the
form (17) for the two axions.
4.2. Z6 Z2 model
The complete anomaly polynomial of the Z6 Z2 model is encoded in a compact form
in Eqs. (B.8) and (B.9), which also distinguish between the various D-brane sectors. It can
be written explicitly as ( = (12)3/4 ):
2
3

2
I =
2
1bc X25 2X45 + X49 + X29 X45 + Y25 Y49 + 8Z4 + Y29 Y45 4Y69
c=1
b=1
2

1b c X25 2X45 + X49 + X29 X45 + Y25 Y49 + 8Z4 + Y29 Y45
b=1
4Y69

4

+
(1bc 1b c ) X29 X49 + 4Y29 Z4 + 4Y69 ,
(24)
b=1
in terms of the components of the charges (B.1)(B.3), which read in this case (pi =
(1, 1, 2), qi = (1, 1, 2), ri = (1, 1, 0)):
Y2 = 3ri tr Fi ,
X2 = pi tr Fi ,
(25)

1
1
3
ri tr Fi2 ,
tr R 2 ,
X4 =
Y4 = qi tr Fi2 +
2
2
48
1
tr R 2 ,
Z4 =
(26)
192

ri
1
tr Fi tr R 2 .
Y6 = tr Fi3 +
(27)
36
2 3
50
When integrated over the internal space, Eq. (24) is in agreement11 with the contribution
of the massless chiral fermions in the representations listed in Table 2. There are however
additional anomalies (of all types, including irreducible terms) that do not involve the
gauge fields associated to the D5 or
D5 branes, which are distributed with opposite signs
at different fixed points and are therefore not detectable in the 4D effective theory. These
anomalies are generated by KK modes of charged fields in the 99 sector. In total, there are
4 truly anomalous U (1)s, two U (1)s that have only localized anomalies, and localized
irreducible anomalies.
The anomalous couplings for the two kinds of twisted axions and and the twisted
4-form c are easily deduced from Eqs. (B.10)(B.12). Defining for convenience the
combination of Kronecker -functions b = b,1 + b,2 , we find:
L=
3
4

c=1

1bc d1bc X19 2b X15 + 1bc X49 2b X45

b=1
d 1c 2b Y15 + 1c 4Z4 + 2b Y45 + c1c 4 8b
3
4

c=1

1b c d1b c X19 2b X15 1b c X49 2b X45

b=1

+ d 1c 2b Y15 1c 4Z4 + 2b Y45 c1c 4 8b
3
3

ac dac Y19 ac Y49 + 8Z4 dcac Y59 .
(28)
a=1 c=1
The terms relevant to the cancellation of localized irreducible anomalies are the last
terms of each square bracket. The other terms, instead, are relevant to the cancellation
of reducible U (1) anomalies.
5. Discussion and conclusions

In this paper, two chiral 4D open string models with SS SUSY breaking have been
constructed as geometric freely acting orbifolds. In this setting, we derived the known
Z6 Z2 model and constructed a new and very simple Z3 Z3 model. Both are
classically stable, since all massless NSNS and RR tadpoles vanish. The compactification
backgrounds are non-SUSY deformations of usual CalabiYau orbifolds. In the Z3 Z3
model, the deformation is induced by the Z3 element, which is a diagonal translation in a
torus together with a non-SUSY rotation along another torus. This deformation is very
similar to the one that gives rise to Melvin spacetime backgrounds, where a generic
rotation along a non-compact plane is performed together with a 2R translation along
a circle [35].12 It would be interesting to analyze this analogy better. The quantum stability
of both orientifolds remains an open question that deserves further analysis.
A detailed study of local anomaly cancellation in the two models has been performed.
All pure gauge and mixed gauge-gravitational anomalies cancel, thanks to a generalized GS
11 In particular, it reproduces the results of [16] for gauge anomalies, apart from irrelevant chirality conventions.
12 See [36] for a discussion of D-branes on Melvin backgrounds.
51
mechanism that involves also twisted RR 4-forms, necessary to cancel localized irreducible
6-form terms in the anomaly polynomial, which vanish only globally. The 4D remnant
of this mechanism is a local ChernSimons term. The local (and global) cancellation
of reducible anomalies is instead ensured by twisted RR axions. In the latter case, even
U (1) gauge fields affected by anomalies that vanish only globally in 4D are spontaneously
broken by the GS mechanism.
Although we have not performed any detailed analysis of local anomaly cancellation in
closed string models, we believe that irreducible anomalies should be absent in that case,
whereas reducible ones might present some new feature.
Acknowledgements
We would like to thank L. Alvarez-Gaum, I. Antoniadis, C. Angelantonj, P. Creminelli,
R. Contino, R. Rabadn, R. Rattazzi, L. Silvestrini, A. Uranga and F. Zwirner for useful
discussions. This work was partially supported by the EC through the RTN network The
quantum structure of spacetime and the geometric nature of fundamental interactions,
contract HPRN-CT-2000-00131.
Appendix A. Lattice sums

We denote the 2D lattice sum over the ith torus by:
i ( ) =
i [m, n]( ) =
n,m
(i) 2
|
q 2 |PL
1
(i) 2
|
q 2 |PR
1
(A.1)
m,n
where q = exp[2i ] and the lattice momenta are given by

1
(i)
PL =
m1 Ui + m2 + Ti (n1 + n2 Ui ) ,
2 Im Ti Im Ui

1
(i)
m1 Ui + m2 + Ti (n1 + n2 Ui ) ,
PR =
2 Im Ti Im Ui
(A.2)
in terms of the standard dimensionless moduli Ti and Ui , parametrizing respectively the

Khler and complex structure of the torus. We also define:
i [m] i [m, 0](it),
i [w] i [0, w](it),
(A.3)
and denote respectively by i [m] and i [w] the corresponding Poisson resummed lattice
sums, where the dependence on the transformed modular parameter l is understood.
In the following, we show in some detail how a translation affects the toroidal lattice
sums defined over the annulus, Mbius strip and Klein bottle world-sheet surfaces. In the
Z3 Z3 model, the translation acts diagonally on the torus, as in [5], whereas in the Z6 Z2
model it actually acts non-trivially only along a circle. The torus case has already been
analyzed (see for instance [4,5]).
52
Annulus. It is convenient to define [N, D|g] as the annulus lattice sum for Neumann (N)
and Dirichlet (D) boundary conditions (b.c.) with the insertion of the operator g. The only
non-trivial case to be considered is when g = I, . The relevant Poisson resummed lattice
sums are found to be (omitting the index i in ):

(j ) 1
(i)
,
[m]W
[N|I ] =
(A.4)
m Wm
m
[D|I ] =

w
[N|] =
(j ) 1
(i)
,
[w]W
w Ww
(A.5)

+ ]Wm(i) Wm(j ) 1 ,
[m
(A.6)
[D|] = 0,
(A.7)
(i)
where Ww encodes the position Xi of the ith brane along the corresponding torus and
(i)
Wm is a generic Wilson line along the torus, parametrized by the i phase factors:
Ww(i) = exp[iw Xi /R],
Wm(i) = exp[im i ].
(A.8)
The sum (A.7) vanishes because a translation has no fixed points and hence the operator
is not diagonal on the states. The action of the translation in (A.6) produces a phase in
the KK modes that, in the Poisson resummed lattice sums, gives a shift on m. Notice that
D-branes couple to all KK and winding modes.
Mbius strip. In this case, the N b.c. give lattice sums similar to those in the annulus,
since does not act on KK modes. For D b.c., the non-trivial cases are obtained when
g = R and g = R, where R and are respectively a rotation and a translation of order 2
on the torus (actually only on a circle). Indicating with [N, D|g] the Mbius strip lattice
sum contribution, we therefore get:

(i)
[2m]W
[N|I ] =
(A.9)
2m ,
m
[N|] =
(i)
[2m
+ 2]W2m
,
(A.10)
[D|R] =
(i)
[2w]W
2w ,
(A.11)
[D|R] =
(i)
e2iw [2w]W
2w .
(A.12)
The fact that only even KK and winding mode appear in the above equations implies
that O-planes couple only to even KK momenta and winding modes. Notice, moreover,
that Eq. (A.11) represents the interaction of a D5- or
D5-brane with O5-planes in the R
fixed points, i.e., y = 0 and y = R along the SS direction, whereas Eq. (A.12) represents
the interaction of a D5- or
D5-brane with the O5-planes (actually
O5-planes due to the
()F action that comes together with ) located at the R fixed points, i.e., y = R/2 and
y = 3R/2 along the SS direction. Similarly, Eqs. (A.9) and (A.10) represent respectively
the D9 (or
D9) interactions with O9 and
O9-planes.
53
Klein bottle. Define i [h|g] as the Klein bottle lattice sum in the h twisted sector with
the insertion of the operator g in the trace. Since lattice sums can only appear for the
usual untwisted sector or for sectors twisted by a translation of order 2, h = I, , where
is the translation. On the other hand, non-trivial lattice contributions are obtained when
g is a generic translation, as well as a Z2 reflection R (aside the identity). As in the
analogue annulus case, the insertion of a translation gives rise to KK-dependent phases
exp (2i m), whereas the twisted sector presents half-integer winding modes for i .
Therefore, the relevant Poisson resummed lattice sums are given by:

|] =
[I
[2m],
m
|] =
[I
[2m + 2] ,
|R] = [I
|R] =
[I
[2w],
[|]
= [|]
= 0,
[|R]
= [|R]
=
e2iw [2w].
(A.13)
Notice that (A.13) confirms that O-planes couple only to even KK momenta or even
winding modes, differently from D-branes.
Appendix B. Anomalous couplings

In this appendix, we discuss the computation of anomalies for the Z6 Z2 and Z3 Z3
models, and the deduction of anomalous couplings by factorization. We proceed along the
lines of [18]. We use a compact differential form notation where Cabc

denotes the formal
sum/difference of a RR axion (0-form) abc and its 4D dual 2-form babc arising at a generic
fixed point Pabc : Cabc

= abc babc . The inflows mediated by these fields can then be
schematically written as Cabc

Cabc
= 1. A similar notation is adopted also for twisted
states associated to a fixed plane, say Pab , which consist of an axion ab and its 6D dual
4-form cab , and a self-dual 2-form bab ; we define in this case Dab = ab + bab + cab .
Since these fields live in 6D, D5-branes or fixed-points and D9-branes or fixed-planes
9,5
couple to different 4D components Dab
. In particular, the 6D 2- and 4-form fields bab
and cab give rise in 4D to 2- and 4-forms bab and cab when no index is in the fixed-plane
direction, but also to 0- and 2-forms ab and bab when 2 indices are in the fixed-plane
9
5
direction. In this notation, Dab
= ab + ab + bab and Dab
= ab + bab + cab .
Since ab and bab , as well as ab and bab , are dual from the 4D point of view, whereas
ab and cab are dual from the 6D point of view, the only non-vanishing inflows mediated
9 D 5 = 1. This setting allows us to
by these fields can be formally summarized in Dab
ab
understand the form of the anomalous couplings in sectors with fixed planes, including
those left unexplained in [18].
54
As for standard D-branes [22] and O-planes [31], it is very useful to define the following
field-dependent topological charges for D-branes and fixed points:

X (F , R) = Tr X eiF A(R),
(B.1)
iF
Y (F , R) = Tr Y e
(B.2)
A(R),

Z(R) = L(R/4).
(B.3)
The labels X and Y distinguish between the two different sectors contributing to the
anomaly in each of the models under analysis. These charges must be intended as sums
of components with growing degree n, which we shall denote by Xn , Yn and Zn .
Z3 Z3 model In the Z3 Z3 model, X refers to the twisted sector, whereas Y refers
to the 2 twisted sector, so that X = and Y = 2 . The anomaly polynomial is
9
easily computed and is given by I = IA99 + IM
, where

2
a bc X9 X9 a bc Y 9 Y 9 ,
2
a,b,c

2
a bc X9 Z a bc Y 9 Z ,
= 2
IA99 =
9
IM
(B.4)
(B.5)
a,b,c
are the contributions from the annulus and Mbius strip surfaces respectively, and
= 37/4 . The anomaly polynomial can be easily factorized, and yields the following
anomalous couplings:

a bc Ca bc X9 + a bc Ca+ bc Y 9 ,
SD9 =
(B.6)
a,b,c

SF = 4
a bc Ca bc Z + a bc Ca+ bc Z .
(B.7)
a,b,c
Z6 Z2 model In the Z6 Z2 model, X refers to the and sectors, whereas Y refers
to the 2 and 2 sectors; X is defined as 9 in the 9 sector and 16 in the 5 and 5 sectors,
2 in the 5 and 5
sectors. The anomaly is given by
and Y as 92 in the 9 sector and 16

I = IA + IM in terms of the contributions from each sector of the annulus and

Mbius strip, which are given by
IA
n
3

2

=
1bc X X + D Y Y 1b c X X + D Y Y ,
2
b=1 c=1
(B.8)
IM
= 42
n
3

1bc Y Z 1b c Y Z ,
(B.9)
b=1 c=1
where = 123/4 ; for = 99, 55, 95, 59, the coefficient is equal to 1, 4, 2, 2, D is
0, 0, 1, 1, and n is 4, 2, 2, 2. Written in this form, the anomaly can be easily factorized,
55
and we find the following anomalous couplings:

4

3
3

+
9
1bc C1bc
ac Dac Y 9 ,
X9 + 1b c C1b
SD9 =
c X
c=1
SD5 = 2
b=1
3
2

c=1
SF = 4
3

c=1
(B.10)
a=1
+ 5

5
5
X D1c Y 5 + 1b c C1b
1bc C1bc
,
c X + D1c Y
b=1
4

b=1
1bc D1c Z 1b c D1c Z 2
3

ac Dac Z
(B.11)
(B.12)
a=1
References
[1] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
[2] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[3] R. Rohm, Nucl. Phys. B 237 (1984) 553;
C. Kounnas, M. Porrati, Nucl. Phys. B 310 (1988) 355;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
C. Kounnas, B. Rostand, Nucl. Phys. B 341 (1990) 641;
I. Antoniadis, C. Kounnas, Phys. Lett. B 261 (1991) 369.
[4] E. Kiritsis, C. Kounnas, Nucl. Phys. B 503 (1997) 117;
E. Kiritsis, C. Kounnas, hep-th/9703059.
[5] C.A. Scrucca, M. Serone, JHEP 0110 (2001) 017, hep-th/0107159.
[6] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[7] I. Antoniadis, C. Muoz, M. Quirs, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69, hep-th/9310151.
[8] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[9] I. Antoniadis, E. Dudas, A. Sagnotti, Nucl. Phys. B 544 (1999) 469, hep-th/9807011.
[10] I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 553 (1999) 133, hep-th/9812118;
I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 565 (2000) 123, hep-th/9907184.
[11] A.L. Cotrone, Mod. Phys. Lett. A 14 (1999) 2487, hep-th/9909116.
[12] A. Pomarol, M. Quirs, Phys. Lett. B 438 (1998) 255, hep-ph/9806263;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quirs, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410;
A. Delgado, A. Pomarol, M. Quirs, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[13] R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[14] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 385 (1996) 96, hepth/9606169.
[15] G. Aldazabal, A. Font, L.E. Ibez, G. Violero, Nucl. Phys. B 536 (1998) 29, hep-th/9804026.
[16] I. Antoniadis, K. Benakli, A. Laugier, hep-th/0111209.
[17] C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073.
[18] C.A. Scrucca, M. Serone, JHEP 9912 (1999) 024, hep-th/9912108.
[19] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
[20] L.E. Ibez, R. Rabadan, A.M. Uranga, Nucl. Phys. B 542 (1999) 112, hep-th/9808139.
[21] G. Aldazabal, D. Badagnani, L.E. Ibez, A.M. Uranga, JHEP 9906 (1999) 031, hep-th/9904071;
M. Bianchi, J.F. Morales, JHEP 0003 (2000) 030, hep-th/0002149.
[22] M.B. Green, J.A. Harvey, G.W. Moore, Class. Quantum Grav. 14 (1997) 47, hep-th/9605033.
56
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
C.G. Callan, J.A. Harvey, Nucl. Phys. B 250 (1985) 427.

M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589.
E. Bergshoeff, R. Kallosh, A. Van Proeyen, JHEP 0010 (2000) 033, hep-th/0007044.
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142.
R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, hep-th/0203039.
E.G. Gimon, J. Polchinski, Phys. Rev. D 54 (1996) 16671676, hep-th/9601038.
Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193;
E. Witten, Nucl. Phys. B 258 (1985) 75;
L.E. Ibez, H.P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25.
M.R. Douglas, JHEP 9707 (1997) 004, hep-th/9612126.
J.F. Morales, C.A. Scrucca, M. Serone, Nucl. Phys. B 552 (1999) 291, hep-th/9812071;
C.A. Scrucca, M. Serone, Nucl. Phys. B 556 (1999) 197, hep-th/9903145;
C.A. Scrucca, M. Serone, hep-th/9911223.
J. Polchinski, Y. Cai, Nucl. Phys. B 296 (1988) 91.
L. Pilo, A. Riotto, hep-th/0202144.
G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59.
F. Dowker, J.P. Gauntlett, D.A. Kastor, J. Traschen, Phys. Rev. D 49 (1994) 2909, hep-th/9309075.
E. Dudas, J. Mourad, Nucl. Phys. B 622 (2002) 46, hep-th/0110186;
T. Takayanagi, T. Uesugi, JHEP 0111 (2001) 036, hep-th/0110200;
T. Takayanagi, T. Uesugi, Phys. Lett. B 528 (2002) 156, hep-th/0112199.

Moduli space metric of N = 2 supersymmetric

SU(N) gauge theory and the enhancon
Gian Luigi Alberghi, Steven Corley, David A. Lowe
Department of Physics, Brown University, Providence, RI 02912, USA
Received 19 April 2002; accepted 3 May 2002
Abstract
We compute the moduli space metric of SU (N) YangMills theory with N = 2 supersymmetry
in the vicinity of the point where the classical moduli vanish. This gauge theory may be realized
as a set of N D7-branes wrapping a K3 surface, near the enhancon locus. The moduli space metric
determines the low-energy worldvolume dynamics of the D7-branes near this point, including stringy
corrections. Non-abelian gauge symmetry is not restored on the worldvolume at the enhancon point,
but rather the gauge group remains U (1)N1 and light electric and magnetically charged particles
coexist. We also study the moduli space metric for a single probe brane in the background of N 1
branes near the enhancon point. We find quantum corrections to the supergravity probe metric that
are not suppressed at large separations, but are down by 1/N factors, due to the response of the
N 1 enhancon branes to the probe. A singularity appears before the probe reaches the enhancon
point where a dyon becomes massless. We compute the masses of W-bosons and monopoles in a
large N limit near this critical point. 2002 Published by Elsevier Science B.V.
PACS: 11.15.-q; 12.60.Jv; 11.25.Sq
1. Introduction
Some time ago, Argyres and Faraggi [1], and Klemm, Lerche, Theisen and Yankielowicz [2] obtained the exact solution of the low-energy effective action of N = 2 supersymmetric SU(N) gauge theory, generalizing the work of Seiberg and Witten for SU(2) [3].
The answer is written as a prepotential for the N = 2 theory, expressed as the period matrix
of a certain hyperelliptic Riemann surface. The period matrix becomes the metric for the
E-mail addresses: gigi@het.brown.edu (G.L. Alberghi), scorley@het.brown.edu (S. Corley),
lowe@het.brown.edu (D.A. Lowe).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 4 - 1
58
G.L. Alberghi et al. / Nuclear Physics B 635 (2002) 5774
(N 1)-dimensional moduli space of this gauge theory. This period matrix was evaluated
in an instanton expansion, in the limit of well-separated moduli, by DHoker, Kirchever and
Phong [4]. PicardFuchs equations were set up by [5]. Many of these results are described
in the review article [6].
The large N limit of this theory was considered by Douglas and Shenker [7]. This limit
is rather subtle, because if the moduli have finite separations as N the instanton
corrections vanish exponentially fast. To obtain a nontrivial limit, one must consider points
on moduli space where separations vanish as 1/N . Douglas and Shenker focused on the
case where the moduli eigenvalues are located on the real axis. This is the case that is
smoothly connected to vacua of the N = 1 gauge theory with massive adjoint matter. In
this paper we will generalize the analysis of [7], to the case when the moduli sit near a
circle (or circles) in the complex plane, as well as to the case where a pair of eigenvalues
are separated from such a circle.
One of the main motivations for studying this problem is given by recent attempts to
understand the resolution of certain naked singularities in string theory. For the case of
D-branes wrapping a K3 surface, it was argued [8] that the timelike repulson singularity
present in the naive continuation of the geometry, should be replaced by a shell of D-brane
source sitting outside this pointthe so-called enhancon locus. Inside this locus, it was
argued the geometry should be flat. This picture received further support from a detailed
analysis of the junction conditions [9], and from black hole entropy considerations [10,11].
A detailed understanding of this phenomena requires a full string theory treatment of
the problem. From the point of view of the supergravity analysis, the D-brane source
boundary conditions are put in by hand. In this paper we present the nonperturbative
moduli space metric that describes the low-energy worldvolume dynamics of D7-branes
wrapping K3, near the enhancon locus. This is an important step toward a better string
theory understanding of the enhancon phenomenon.
Our results support the picture advocated in [8]. A wrapped D7-brane probe approaching N 1 branes at the enhancon point, smoothly melts into the other branes, and then
emerges from the other side. The probe brane does not see the interior of the geometry at
all. The enhancon locus is simply a non-singular point from the viewpoint of the probe.
Interestingly, we find nontrivial quantum corrections to the probe metric that do not fall off
at large separations, however these appear at subleading order in 1/N .
The plan of this paper is as follows: in Section 2 we briefly review the solution of
SU(N) N = 2 supersymmetric gauge theory of [1,2] to fix notation; in Section 3 we
compute the moduli space metric along a one-complex parameter slice of the space in
two different situations: for a pair of enhancon-like shells, and for a probe brane in an
enhancon background. An analytic form of the metric along this slice is presented for all
values of N . The metric includes nontrivial quantum corrections. For the case of the probe
brane, the moduli space metric has a singularity as the probe starts to get close to the
enhancon shell, when two eigenvalues collide and a dyon becomes massless. We obtain
analytic expressions for the masses of dyons and monopoles in the large N limit, near
this critical point. In Section 4 we comment on the implications of these results for the
resolution of spacetime singularities in string theory, and discuss future prospects.
59
2. Review and notation

Recall that the solution to the low-energy effective action of N = 2 supersymmetric
SU(N) YangMills theory is given in terms of an auxiliary Riemann surface C, that consists
of a genus N 1 hyperelliptic curve C defined by
Y2 =
N

2
(X j )2 2N = P (X) 2N
(2.1)
j =1
along with the 1-form

X dP
(2.2)
2iY
whose j derivatives are holomorphic. The i denote the classical expectation values of the
gauge theory moduli. For SU(N) gauge theory the moduli space coordinates must satisfy
the tracelessness condition
=
N
j = 0.
(2.3)
j =1
The periods aj and aDk are then given by the integrals

aj = ,
aDm =
(2.4)
for some choice of cycles m and k where only N 1 of each of these sets of cycles is
independent.
To construct the moduli space metric one requires derivatives of the periods with respect
to the moduli space coordinates k . If the periods (2.4) can be evaluated as functions of
the coordinates j then this is easy to compute. Finding the periods at a generic point in
moduli space however is difficult and it is easier instead to compute the derivatives at a
fixed point in moduli space directly by noting that

1 P
X P
.
=
dX + d
(2.5)
k
i2Y k
Y k
The term inside the total derivative is periodic around a closed contour, so the -derivatives
of the period integrals can be evaluated using only the first term of (2.5).
Using this information one can then compute the period matrix mk via the prescription
mk =

N1
aDm daDm da 1
=
,
ak
dl d lk
(2.6)
l=1
where the derivatives on the far right-hand side are total derivatives. That is, we have
removed the N dependence using the tracelessness property (2.3) so that
daDm aDm aDm
=
.
dl
l
N
(2.7)
60
It is important to note however that this prescription has an apparent singularity if j = k

for any indices j = k. The singularity can be easily seen from (2.5) by noting that
al /j = al /k when evaluated at the point j = k . The (N 1) (N 1) matrix
[a/] is then not invertible. If the branch points of the curve (2.1) are all distinct, then we
do not expect this point in moduli space to be a real singularity, but rather just a coordinate
singularity. In the cases that we discuss below we will indeed see that this is true.
3. Moduli space metric

It is rather difficult to obtain explicit expressions for the metric throughout the full
moduli space. To simply matters we will focus on two distinct one complex parameter
slices through the full (N 1)-dimensional moduli space, each of which contains the
enhancon point (all i = 0).
3.1. Two enhancon shells
The first slice we consider retains a ZN rotation symmetry in the curve C, and
corresponds to a pair of concentric enhancon shells, which coalesce at the enhancon
point. The corresponding brane configurations have been analyzed from the supergravity
viewpoint in [911]. We parametrized the moduli space coordinates in terms of a complex
parameter v as
j = v ei2j/N ,
(3.1)
where 1 j N . Note that this choice of moduli space coordinates does satisfy the
tracelessness condition (2.3). The enhancon point sits at v = 0, which leads to a coordinate
singularity in the period matrix (2.6) as discussed above. One can get around this problem
as we shall see by computing the period matrix for nonzero v and then taking the limit
v 0, in which case a nonsingular answer is obtained. For nonzero v this subspace
corresponds to two separate enhancon shells.
To see this in more detail we must solve for the branch points of the curve given by
Y = 0. For the simple subspace in moduli space that we have taken these can be found
exactly. The point is that the product in (2.1) simplifies to
N
(X j ) = XN v N ,
(3.2)
j =1
which follows easily from the identity

N
ei2j k/N = Nk,0 mod N .
(3.3)
The branch points are then found to be

1/N i2k/N

X2k = N + v N
e
,
N

1/N i2(k+1/2)/N
X2k+1 = v N
e
(3.4)
j =1
61
Fig. 1. Positions of branch cuts, and choice of 1-cycles for double enhancon configuration.
for 1 k N . For real v we get the structure shown in Fig. 1 with a pair of concentric
circles of branch points. We have also shown some of the cycles m and j . We take the
m cycle to enclose the X2m and X2m+1 branch points, with 1 m < N . The j cycle
encloses the X2j 1 and X2j branch points, with 1 j N . The n cycles are then given
in terms of the cycles by
n =
n
(3.5)
j =1
and we take 1 n < N . The and cycles comprise a canonical basis of homology
1-cycles.
The holomorphic 1-forms (2.5) on this slice of moduli space are given by
(XN v N )dX
1

=
.
k
i2 (X v ei2k/N ) (XN v N )2 2N
(3.6)
While one could try to compute the period matrix using this basis of holomorphic 1-forms,
it turns out that it is not the most convenient. The computation of the period matrix is
simpler if we change basis to
i2kn/N
=
e
tn
k
N
k=1
ei2kn/N
k N
k=1
Nn
v
dXN
1

=
,
i2N X
(XN v N )2 2N
N1

(3.7)
62
where the second line follows from the identity (3.3).

To compute the period matrix we now compute the derivatives of the periods

aj
aDm
=
,
=
.
tn
tn
tn
tn
(3.8)
For the aD -periods we make the change of variables

1/N i2j/N

e
X = N ei + v N
(3.9)
which reduces (3.7) to
1
ei d
v Nn i2j n/N
e
.
=
tn
2N
(N ei + v N )1n/N ei2 1
(3.10)
Defining

1
f v N , n/N
N

d
0
ei
1
(N ei + v N )1n/N ei2 1
(3.11)
we then have

aDm
= v Nn f v N , n/N ei2mn/N .
tn
(3.12)
Similarly for the a-periods we find

a k
= v Nn f v N , n/N ei2(k1/2)n/N .
tn
(3.13)
The tilde notation here denotes a period evaluated along a contour. To obtain the a-period
we must sum as in (3.5), or in matrix notation
N1

aj
a k
=
qj k
,
tn
tn
(3.14)
k=1
where the matrix entries of qj k are given by ones for j k and zeroes everywhere else.
What makes the coordinate transformation in (3.7) so convenient is that (3.12) and
(3.13) depend on the aD and a indices, respectively, only through a phase factor.
Consequently either matrix is easy to invert provided that f (v N , n/N) does not vanish
for any n and v. Specifically the (N 1) (N 1) matrix
mj n ei2j n/N
(3.15)
has the inverse

m1

nk

1 i2nk/N
e
1
N
(3.16)
from which it follows that

1
1
aD 1
m np ,
= Nn
t np
v
f (v N , n/N)
1
1
a
ein/N
m nj .
= Nn
t nj
v
f (v N , n/N)
63
(3.17)
Using (3.12) and (3.17) it is now straightforward to compute the period matrix (2.6), and
we find
N1
2i f (v N , n/N)
sin(n/N)ei2n(mk)/N .
mk =
N
f (v N , n/N)
(3.18)
n=1
In obtaining this result we have multiplied the inverse a matrix in (3.17) on the right by the
matrix qj1
k = j,k j 1,k which converts the -period a matrix to a -period a matrix.
While we cannot simplify the period matrix any further for generic values of v, at v = 0
the sum is straightforward to do,

i

cot (m k + 1/2)/N cot (m k 1/2)/N .
mk =
(3.19)
N
There is a theorem in Riemann surface theory that says that mk is symmetric and that
its imaginary part is positive definite. By replacing the summation index n by N n one
easily shows that symmetry implies the condition that
f (v N , 1 n/N)
f (v N , n/N)
=
.
f (v N , n/N) f (v N , 1 n/N)
(3.20)
For v = 0 this condition holds trivially as both sides are one, and moreover from the explicit
expression given in (3.19). For generic values of v we have checked numerically that it does
hold for various values of v and N , which provides a check on our explicit expression for
the period matrix.
For positivity we again do not have a direct analytic proof that this condition holds
for the explicit form for given above. Even for v = 0 we have not been able to show
analytically that all eigenvalues are positive. Curiously though it is straightforward to
check that sin(2kj/N) for 1 j [N/2] are eigenvectors of Im[mk ] with eigenvalues
2 sin(j/N), which are indeed positive. For the remaining eigenvalues and for more
general values of v, we have checked numerically that the eigenvalues are indeed positive.
3.2. Probe in enhancon background
In this section we compute the period matrix in various limits along a one-dimensional
slice of moduli space corresponding to a wrapped D7-brane probe of the enhancon
background. For the most part we are able to derive explicit expressions for finite N , and
consider the large N expansion only when necessary.
The moduli in this case are given by
v
j = , 1 j (N 1),
N
64
v
,
(3.21)
N
where v is the complex coordinate which parametrizes the subspace that we will be
investigating. The associated curve C is given by
N = v
Y 2 = (X v + v/N)2 (X + v/N)2(N1) 2N .
(3.22)
v/ 1 limit
For v/ 1 the branch points of the curve are given by the series expansion in v

2

1
1
v
v 3
Xk = eik/N 1 +
(3.23)
ei2k/N 1
.
+O
2N
N
This form of the branch points suggests a more convenient basis for the holomorphic
1-forms than those given in (2.5) (which, as discussed above, are not independent when
more than one of the moduli space coordinates k are equal). Specifically we shall take the
basis of holomorphic forms given by
1 Xn1 dX
, 1 n (N 1).
(3.24)
2i
Y
The advantage of this basis is that the linear in v correction to the period matrix vanishes
identically, as we now explain in more detail. We shall take the same basis of cycles as in
the previous example for v = 0. That is, the k cycle encircles the X2k and X2k+1 branch
points, while the j cycle encircles the X2j 1 and X2j branch points. The k cycles are
then defined as in (3.5).
To evaluate a period integral of any of the holomorphic forms n over an k cycle, it is
convenient to parametrize the integral in terms of an angular variable as

2

1
1
v
v 3
X = ei2k/N ei/N 1 +
ei4k/N ei2/N 1
,
+O
2N
N
n =
(3.25)
a result which follows naturally from the form of the branch points (3.23). Similarly for
a period integral about a k cycle, one simply replaces k (k 1/2) in (3.25). In this
parametrization Xn1 dX clearly has no linear in v piece. Furthermore Y has no linear in v
piece as follows by substituting (3.25) into (3.22) and expanding, or more simply by noting
that dX(v)/dv vanishes at v = 0 for X(v) a branch point of the curve which follows from
differentiating (3.22) with respect to v.
The periods are now straightforward to write down in an expansion in v. Specifically
we find

aDm
1
(Nn)
F (n)mmn
= n =
n
2N
m

2
v
v 3
(N 1)(n 2)
+
F (n 2)mm,(n2) + O
,
2
2N
a k
=
n

n =
k

aDm
,
n m=k1/2
65
(3.26)
where we have defined

F (n)
2e
i/4
ei(n/N1/2)
d
sin()

N ,(1 + n/(2N)) in/(2N)
e
sin n/(2N)
= i4
n ,(1/2 + n/(2N))
(3.27)
and the matrix

mkn ei2kn/N .
(3.28)
The ak /n periods are integrals over k -cycles defined in terms of the cycles in (3.5).
To compute the period matrix we need the inverse

1
2 N1

a
iNNn in/N 1
v
e
=
m1 nl 2 (l, k)mlk m1 kj
m nj
nj
F (n)
l,k=1

3
v
,
+O
(3.29)

(N 1)(n 2) F (n 2) i4(k+1/2)/N
e
2 (k, n)
.
F (n)
2N 2
(3.30)
The period matrix is then given by

2 N1

v
double
mk = mk
+
D2 (m, n)mmn m1 nj
v=0
n,j =1

N1

1
1
v 3
mmn m nl 2 (l, p)mlp m pj (j,k j 1,k ) + O

,
l,p=1
(3.31)
where the first term is the period matrix of the double enhancon evaluated at v = 0 found
in the previous subsection and we have defined
D2 (m, n)
(N 1)(n 2) F (n 2) i4m/N
e
.
2N 2
F (n)
(3.32)
The last two Kronecker delta functions convert the a periods to a periods as discussed
above.
To connect this result with the supergravity description, we compute the induced metric
on the subspace of moduli space parametrized by v. This induced metric corresponds to
the moduli space metric of a probe D7-brane in the enhancon background, which may be
computed using the BornInfeld action plus ChernSimons couplings. To construct this
66
induced metric we need the periods ak as functions of v. These are given by the contour
integrals (2.4). The computation is similar to what we have done before. Namely one
parametrizes the integrals as (3.25) and expands the integrand in powers of v. The end
result is

ak (v) =
F (N + 1) 1 ei2k/N
i4 sin(/N)

2

(N 1)
v
v 3
i2k/N
.
F
(N
1)
1
e
+
(3.33)
+
O
2N 2
The induced metric is now given by
2
dsprobe
N1

m,k=1
dam
Im[mk ]
dv
dak
|dv|2
dv
1 |F (N 1)|2
|v dv|2
8 2 N 2 sin(/N)

1 |F (N 1)|2 2
d + 4 2 d 2 ,
=
(3.34)
2
2
8 N sin(/N)
where the (, ) coordinates are defined as v 2 exp(i). The last form shows in
particular that the induced metric has a conical singularity at = 0 with negative deficit
angle.
One may use this result to compute the trajectory of a probe brane approaching the
enhancon shell. At the enhancon point, the probe hits the conical singularity in the induced
metric. The full metric is smooth at this point, and may be used to continue the motion
of the probe past the conical singularity. On the subspace parametrized by v, the i
respect a ZN1 symmetry. Since the periods a vary continuously (and with continuous
first derivatives) through the enhancon point, the same symmetry will be present after
passing through the enhancon point. The unique vacuum expectation value that preserves
this symmetry corresponds to the subspace parametrized by v, so the probe remains on the
v-slice. From the point of view of the spacetime coordinates, the probe merges with the
enhancon, and then smoothly reemerges from the other side.
=
v/ 1 limit
In this section we consider the opposite extreme for the probe, v/ 1. In this limit
the probe is located far from the remaining moduli as is clear from the polynomial (3.22).
In particular one finds for the branch points in this limit

N
v
=
X
,
1
v

2

xj 3
1
1
2
j = xj 1 +
X
(3.35)
(xj ) + O
,
xj
N v
2N v
v
67
Fig. 2. Branch cuts and 1-cycles for the probe configuration.
where
xj
1/(N1)
ei(j 1)/(N1)
(3.36)

for 1 j 2(N 1). We have here introduced the scaled and shifted variable X
(X + v/N)/.
So far our choice of electric versus magnetic cycles has been relatively arbitrary.
However in this case we must be careful to identify the N -cycle which encircles the
branch points, see Fig. 2, with an electric contour and, e.g.,
branch cut connecting the X
and X
1 branch points, with a magnetic contour. For the
N1 , the cycle encircling the X
remaining cycles we take j to encircle X2j 1 and X2j for 1 j (N 1) and k to
encircle X2k and X2k+1 for 1 k (N 2). We further define -cycles as
j N +
j
k =
N1
k ,
1 j (N 2)
k=j +1
k=1
N =
N1
k ,
j = (N 1).
(3.37)
k=1
The periods aj and aDm are then defined as in (2.4).

To compute the period matrix we again need the matrices of integrals of some basis of
holomorphic forms over the - and -cycles. We shall take a slightly different basis for the
holomorphic forms than in the previous two cases, namely let

n1 d X
1 X
, 1 n (N 1).
(3.38)
2i
Y
so clearly the basis (3.38)
The only difference as compared to (3.24) is to replace X by X,
is just a linear transformation of (3.24).
n
68
The periods encircling pairs of branch points on the circle are straightforward to
compute in an expansion in powers of (/v) and basically follow from the previous two
sections. Specifically one finds

a j
1
n
= n =
(x2j 1 ) F (n) + O
,
N
2(N 1)
v
n
j
1 j, n (N 1),

a k
aDm
= n =
,
n
n k=m+1/2
1 n (N 1), 1 m (N 2),
(3.39)

(n) F (n)|N(N1) ,
F
(3.40)
where F (n) was given in (3.27).

Computation of the remaining periods over the N1 cycle is slightly more involved.
As it turns out we have not been able to evaluate all of the remaining periods directly.
However, as we discuss in the appendix, one can evaluate them indirectly and still compute
the period matrix. We record the results here and relegate the details to the appendix. For
the remaining periods we find
i n
n
(n)
aDN1
e N1 1 v N1
F
i2 n
2(N 1)N e N1
n
1
for 1 n (N 2) while the n = (N 1) case is given by

aDN1
v
(N + 1)
.
ln
i
N
N1
(3.41)
(3.42)
The period matrix can now be expressed as

mj =
N1

n,k=1
daDm
d n
d a
d
1
nk
(q)
1
kj .
(3.43)
The qj k matrix was essentially defined in (3.37) as the matrix which converts the -cycles
(or a periods) into -cycles (or a periods), analogously to the earlier discussion around
(3.14). The inverse matrix does the reverse, and is given by
1
q kj (k,j k,j +1 ), 1 k (N 1), 1 j (N 2)
k,1 ,
1 k (N 1), j = (N 1).
(3.44)
The period matrix is now straightforward to evaluate and we find to leading order at large v

m j + 1/2
m j 1/2
i
cot
cot
,
mj =
(N 1)
N 1
N 1
1 m, j (N 2)
69
ei(m1/2)/(N1)
i
, 1 m (N 2), j = (N 1)
(N 1) sin((m 1/2)/(N 1))

i
v
, m = (N 1), j = (N 1).
= (N + 1) ln
(3.45)
The row (N1),j for 1 j (N 2) then follows by the fact that the period matrix is
symmetric.
The general form of this result was to be expected. Indeed, one might think that when
(v/) 1 the probe is located far from the enhancon and the physics of the probe reduces
to that of a weakly coupled SU(2) gauge theory, which is given by the tree-level plus oneloop result. In the large N limit, this accounts for the N1,N1 entry, which agrees with
the supergravity probe brane metric [8]. However, already at order 1/N we see a shift in the
coefficient of the log term. The one-loop beta function would give a coefficient of N 1,
rather than the N + 1 that emerges from our detailed calculation. This arises from the
response of the N 1 background branes to the position of the probe, which can generate
such corrections due to the close spacing O(/N) of the eigenvalues of the associated
moduli. We also see instanton corrections in the other entries, which do not fall off as v
becomes large, but are also suppressed by 1/N factors. These have a similar interpretation.
v/ 1 limit
We have so far discussed the large and small v/ limits. Starting from v/ 1 and
decreasing v, the probe branch points move toward the enhancon branch points distributed
on (an approximately) unit circle. On the real axis there is a critical point
v
N 1/N
ln N
=
,
1+
(11/N)
(1 1/N)
N
(3.46)
where one of the probe branch points collides with an enhancon branch point. At this
point the period matrix becomes singular, and a dyon becomes massless. For finite N , the
singularity is of the same form as the singularities of the SU(2) theory [3]. Near, but not
at, this point however the period matrix is nonsingular. To fully understand the motion
of the probe in this region we must compute the period matrix. We will obtain analytic
expressions for the masses of charged particles near this point, but so far have been unable
to obtain an analytic expression for the period matrix.
The roots of the SeibergWitten curve for v/ < 1 are given approximately by

2

k = eik/N 1 1 ln 1 v eik/N 1 v eik/N + O 1/N 3
X
N
2 N
(3.47)
was defined after (3.36).
for 1 k 2N where X
To compute the periods we use the same set of holomorphic one forms (3.38) as given
in the previous section. Furthermore we shall take the same set of cycles as in the small
v/ case. Parameterizing the integrals in terms of an angular coordinate as

k |kk+/ ,
X()
X
(3.48)
70
the expansion of the holomorphic one forms (3.38) produces to leading order in 1/N the
expression
ei2kn/N
d ein/N
(3.49)
+ .
2N ei2 1 (1 (v/)ei2k/N ei/N )(n1)/N

Expanding the exponential exp(i/N) in powers of N , one is left with integrals that we
have seen before, (3.27). In particular one finds
n =
ei2kn/N
aDk
1
=
2N (1 (v/)ei2k/N )(n1)/N
n

v (n 1) ln(1 (v/)ei2k/N )
1 + 2F (n)
N2
(ei2k/N v/)

1
1
v
ln 1 (v/)ei2k/N
i2k/N
N
N (e
v/)

v n (F (n) F (n + 1))
1
,
+2
+O
N (ei2k/N v/)
N2

a m aDk
=
.
n
n k(m1/2)
The periods are given in terms of the derivatives of the periods by
2

v
aDk
aDk
v aDk
1
aDk = N
+
1

N+1
N
N
N1
(3.50)
(3.51)
with similar expressions for the as and as.
To compute the period matrix we must invert the matrix a/ . Unlike in our previous
cases however the k-dependence of ak / n is no longer contained only in a phase factor,
but rather something more complicated. We do not know of any nice analytic expression
for the inverse of this matrix and therefore have not been able to compute an analytic form
for the period matrix. Nevertheless the expressions for the periods given above generate the
masses of charged particles including instanton corrections, at leading orders in the large
N expansion.
4. Conclusions
In this paper we have computed the moduli space metric for N = 2 supersymmetric
SU(N) gauge theory along some distinguished one-complex dimensional slices of
relevance for resolution of singularities in string theory via the enhancon mechanism. One
unexpected result of this analysis is the presence of corrections suppressed by powers of
1/N that do not fall off at large separations, in the regime where one of the classical
moduli is taken to be large. When all moduli are well-separated, one expects the tree-level
plus one-loop contributions to only receive corrections due to instantons, which should be
exponentially suppressed in the large N limit. The reason other corrections survive here is
71
because the moduli associated with the branes on the enhancon shell have separations of
order 1/N , so we are never in a purely semiclassical regime. In [12], similar effects have
been attributed to fractional instantons.
From the supergravity perspective, this limit of moduli space corresponds to a probe
brane moving in the background of N 1 branes on an enhancon shell. The corrections,
which are suppressed by powers of 1/N , contain information about stringy and quantum
corrections to the classical supergravity description of the enhancon.
We have also explicitly computed the enhancon metric near the point where all the
classical moduli vanish. As expected, the moduli space metric is smooth in this limit.
Gauge symmetry is not enhanced at this pointthe gauge symmetry remains U (1)N1 .
The light particles include both electrically and magnetically charged states which are
mutually non-local. The lightest masses go like /N , which sets the cutoff scale for our
description of the low-energy effective field theory.
We computed the induced metric on a one-complex dimensional subspace corresponding to a probe brane near the enhancon locus, and find a conical singularity at the center.
In [13,14] it was argued a similar conical singularity appears on the induced metric on a
subspace of a higher-dimensional analog of the AtiyahHitchin space, relevant for a probe
brane in the background of an enhancon shell arising from wrapped D6-branes. As mentioned above, the full moduli space metric at this point is smooth. This supports the picture
of a probe brane approaching the enhancon shell advocated in [8]. The probe brane melts
into the enhancon branes and becomes indistinguishable from them at the point where the
classical moduli vanish. In particular, a probe brane corresponding to a wrapped D7-brane
is unable to probe the region inside the enhancon shell. We are currently completing a computation of the moduli space metric for SU(N) with fundamental matter. This will allow
us to consider instead a probe D3-brane which is able to probe inside the enhancon shell.
This will provide a useful check on the idea that the region inside the enhancon shell is
simply flat space.
Finally, it is interesting to address the question of to what extent the enhancon
mechanism is a general phenomenon. In the highly symmetric situations we have studied
in this paper, the presence of the enhancon shell corresponds to the fact that the branch
points of the Riemann curve are not coincident, but rather become closely spaced points
around a circle. However nothing stops us considering other regions in moduli space where
branch points collide. In particular, we could consider subspaces of moduli space where
large numbers of branch points collide and give rise to generalizations of the Argyres
Douglas point [15]. These presumably give rise to timelike singularities in the supergravity
description for which the enhancon mechanism will not work. We hope to return to a study
of this regime in the future.
Acknowledgements
We thank F. Ferrari, A. Jevicki, R. Myers, J. Polchinski, and S. Ramgoolam for helpful
comments. This research is supported in part by DOE grant DE-FE0291ER40688-Task A.
72
Appendix A. Period matrix for v/ 1

We discuss the details of the computation of the periods aDN1 / n for the (v/) 1
case in this appendix. Some of these integrals have essentially been done in [12]. The
1 1 + ln(/v)/N and
basic point is the following. The integral is between the limits X
2(N1)
(v/)(1 (/v)N ). It follows that over most of the integration region the X
X
term in Y will dominate so that the 1 can be dropped and one has the simple integral
aDN1
=
n
1
n
iN
N1

X
1
X

dX
v/)X
Nn
(X
v
1
i
N N 1 n

v
(N + 1)
,
ln
i
N
n/(N1)
,
1 n (N 2)
n = N 1.
(A.1)
Actually the integral on the far right-hand side of the top line in (A.1) can be done exactly
and one would get O(N n) terms. We have written only the dominant term above.
However for this approximation to the full integral to be valid one needs to check if the
contribution of both the other O(N n) terms as well as the contribution to the integral
coming from the integration region near the limits where Y vanishes is subdominant.
Actually the former contribution can be removed easily by taking a different basis of
holomorphic one-forms. For example, if we take the basis

n n n+1
(A.2)
v

for 1 n < (N 2) and N1
= N1 instead, the periods in the above approximation
would reduce to just one term.
The real problem is that the contribution to the periods coming from the region of
integration near the limits becomes important once n = O(1). To see this consider the
following correction to the aDN1 /n period (we are working in the basis instead of
the basis here simply because it decouples the two effects mentioned above),
m

X
n1

X
v N+n

1 (/v)X
+ X
d X,
Y
(A.3)
1
X
1 < X
m < X
, but otherwise X
m is arbitrary. The integrand here is just the
where X
difference of the exact holomorphic one-form (A.2) and its approximate form given by
=X
1 (1 + x/(N 1)) one obtains
dropping the 1 in Y . Expanding near the lower limit as X
the integral

xm x(n1)/(N1)
1 n/(N1) 1
e
ex(Nn)/(N1) dx.
i v
N 1
e2x 1
0
(A.4)
73
The upper limit xm can be taken to infinity at the cost of an exponentially small correction
to the integral. This integral can now be evaluated and one finds

Nn
,( 2(N1) )
1 n/(N1) 1
N 1
+
.
(A.5)
i v
N 1
N n
2 ,( 2Nn1
2(N1) )
This should be much less than the leading order contribution to the period integral, which
in this case simply coincides with the contribution coming from the first term in brackets
of the above expression. When n = N O(1) then the term in brackets vanishes to leading
order and indeed this correction is subdominant. However when n = O(1) the term in
brackets does not vanish to leading order and the correction term is of the same order as
the leading order piece, so the approximations made in this regime are not valid.
As we commented on in the text, we do not actually need to compute all of these
period integrals directly to construct the period matrix. We only need to evaluate, e.g.,
aDN1 / N1 (we are working once again with the basis) directly, for which the
approximations described above are valid, and the remaining periods can be obtained
indirectly as we now discuss. Basically one notes that given just the periods in (3.39), and
the fact that the period matrix is symmetric, is enough to construct the entire period matrix
except for the element (N1),(N1) . To compute this element we need the values of the
remaining periods. We can however solve for them by viewing the symmetry requirement
(N1),k = k,(N1) for 1 k (N 2) as a set of equations to determine the periods
aDN1 / k over the same range of k, where one must remember that aDN1 / N1
has been
given above. In more detail, in terms of the tilded period matrix defined as
mj = k mk (q 1 )kj , the equations determining the remaining periods are

N1
aDN1 a 1 a 1
(A.6)
= k,1 .
n
nk
n,(k+1)
n=1
It is a simple exercise to solve these equations for the remaining periods to find (3.41).
References
[1] P.C. Argyres, A.E. Faraggi, The vacuum structure and spectrum of N = 2 supersymmetric SU(n) gauge
theory, Phys. Rev. Lett. 74 (1995) 3931, hep-th/9411057.
[2] A. Klemm, W. Lerche, S. Yankielowicz, S. Theisen, Simple singularities and N = 2 supersymmetric Yang
Mills theory, Phys. Lett. B 344 (1995) 169, hep-th/9411048.
[3] N. Seiberg, E. Witten, Electricmagnetic duality, monopole condensation, and confinement in N = 2
supersymmetric YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum.
[4] E. DHoker, I.M. Krichever, D.H. Phong, The effective prepotential of N = 2 supersymmetric SU(N (c))
gauge theories, Nucl. Phys. B 489 (1997) 179, hep-th/9609041.
[5] J.M. Isidro, A. Mukherjee, J.P. Nunes, H.J. Schnitzer, A new derivation of the PicardFuchs equations for
effective N = 2 super YangMills theories, Nucl. Phys. B 492 (1997) 647, hep-th/9609116.
[6] E. DHoker, D.H. Phong, Lectures on supersymmetric YangMills theory and integrable systems, hepth/9912271.
[7] M.R. Douglas, S.H. Shenker, Dynamics of SU(N ) supersymmetric gauge theory, Nucl. Phys. B 447 (1995)
271, hep-th/9503163.
74
[8] C.V. Johnson, A.W. Peet, J. Polchinski, Gauge theory and the excision of repulson singularities, Phys. Rev.
D 61 (2000) 086001, hep-th/9911161.
[9] C.V. Johnson, R.C. Myers, A.W. Peet, S.F. Ross, The enhancon and the consistency of excision, Phys. Rev.
D 64 (2001) 106001, hep-th/0105077.
[10] C.V. Johnson, R.C. Myers, The enhancon, black holes, and the second law, Phys. Rev. D 64 (2001) 106002,
hep-th/0105159.
[11] N.R. Constable, The entropy of 4D black holes and the enhancon, Phys. Rev. D 64 (2001) 104004, hepth/0106038.
[12] F. Ferrari, The large N limit of N = 2 super YangMills, fractional instantons and infrared divergences,
Nucl. Phys. B 612 (2001) 151, hep-th/0106192.
[13] C.V. Johnson, Enhancons, fuzzy spheres and multi-monopoles, Phys. Rev. D 63 (2001) 065004, hepth/0004068.
[14] C.V. Johnson, The enhancon, multimonopoles and fuzzy geometry, Int. J. Mod. Phys. A 16 (2001) 990,
hep-th/0011008.
[15] P.C. Argyres, M.R. Douglas, New phenomena in SU(3) supersymmetric gauge theory, Nucl. Phys. B 448
(1995) 93, hep-th/9505062.

Ten-dimensional supergravity constraints from the

pure spinor formalism for the superstring
Nathan Berkovits a , Paul Howe b
a Instituto de Fsica Terica, Universidade Estadual Paulista, So Paulo, Brazil
b Department of Mathematics, Kings College, London, UK
Received 27 March 2002; accepted 3 May 2002
Abstract
It has recently
been shown that the ten-dimensional superstring can be quantized using the BRST
operator Q = d , where is a pure spinor satisfying m = 0 and d is the fermionic
supersymmetric derivative. In this paper, the pure spinor version of superstring theory is formulated
in a curved supergravity background and it is shown that nilpotency and holomorphicity of the pure
spinor BRST operator imply the on-shell superspace constraints of the supergravity background. This
is shown to lowest order in for the heterotic and Type II superstrings, thus providing a compact
pure spinor version of the ten-dimensional superspace constraints for N = 1 Type IIA and Type IIB
supergravities. Since quantization is straightforward using the pure spinor version of the superstring,
it is expected that these methods can also be used to compute higher-order corrections to the
ten-dimensional superspace constraints. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
For many purposes, superstring theory is most conveniently expressed as an effective
field theory of its massless modes consisting of supergravity theory together with
corrections arising order by order in . In principle these higher order corrections
can be obtained by computing scattering amplitudes or by demanding consistency
of the superstring sigma model in a curved background. However, neither of these
procedures is easy to carry out in superstring theory in a way in which spacetime
supersymmetry is guaranteed. In the RamondNeveuSchwarz (RNS) formalism, it is
difficult to introduce fermionic or RamondRamond background fields, while the Green
Schwarz (GS) formalism ensures spacetime supersymmetry but is difficult to quantize.
E-mail address: phowe@mth.kcl.ac.uk (P. Howe).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 2 - 8
76
N. Berkovits, P. Howe / Nuclear Physics B 635 (2002) 75105
Although the hybrid formalism for the superstring can be used to compute corrections
in a manner which manifestly preserves D = 2 [1], D = 4 [2], or D = 6 [3] superPoincar covariance, one needs a D = 10 covariant formalism if one wants to describe
the superstring in arbitrary supergravity backgrounds.
In this situation, one might try to study the constraints that ten-dimensional supersymmetry imposes on higher-order contributions to the effective action. One difficulty here is
that, with the exception of the supergravity sector of the heterotic string, it is not known
how to construct any superspace actions due to the absence of any known sets of auxiliary fields. Even in the heterotic case, the auxiliary fields are rather complicated [4] and
it is not clear how to construct higher order actions which correspond to superstring corrections, although the R 4 invariant was discussed from this point of view in [5]. It seems
that additional input apart from supersymmetry is required. On the other hand, it has been
possible to obtain information about some particular terms, for example, in the work of [6
12]. Other approaches to the problem have involved supersymmetrization of bosonic sigma
model terms [13] for the heterotic string [14,15], and studying corrections to heterotic superspace constraints directly [16,17], which has at least been successful in incorporating
anomaly terms. This work has been reviewed in [18] where string results were used to partially construct R 4 corrections in M-theory. Other recent approaches to supersymmetrizing
the R 4 term in M-theory are described in [19] and [20].
The fact that one is forced to look at the equations of motion rather than Lagrangians
suggests that a way forward might be to understand the geometry behind these equations.
Many years ago, Witten showed how the N = 1 D = 10 superspace YangMills equations
can be understood in terms of integrability along light-like lines and how this is related
to -symmetry of the superparticle action [21,22]. This sort of analysis was subsequently
carried out for the heterotic [23] and IIB strings [24], and reinterpreted in terms of light-like
integrability in loop superspace in [25,26], at least for the heterotic case.
In some related work, one of the present authors showed that light-like integrability
could be replaced by integrability along pure spinor lines, and that this can also be
employed in eleven dimensions in the context of the supermembrane [27]. A virtue
of this approach is that it is simpler than light-like integrability, but, at the time, it
was not entirely clear how it was related to particle or string actions. More recently,
the other author has shown that ten-dimensional superparticles and superstrings can be
effectively quantized using pure spinor variables [28,29]. These pure spinor variables
can be interpreted as bosonic ghosts for a fermionic symmetry, although it is not
currently fully understood how this can be implemented in a worldsheet reparameterization
invariant fashion. Nevertheless, the final gauge-fixed action does have manifest spacetime
supersymmetry and correctly fixes the central charge to be zero. Unlike the GS formalism,
however, the pure spinor formalism has the tremendous advantage that it can be quantized
straightforwardly since the action is free in a flat background.
In this pure spinor formalism [29], the left-moving BRST operator for the heterotic
superstring is

Q=
d ,
(1)
77
where is a bosonic pure spinor variable satisfying1

m

=0
(2)
for m = 0 to 9, and d is the worldsheet variable corresponding to the N = 1 D = 10

spacetime supersymmetric derivative. In a flat background, and d are holomorphic and
m , where = x + i
d satisfies the OPE d (y)d (z) i (y z)1
m
m
m
m
2
m
is the supersymmetric momentum. So = 0 implies that Q is nilpotent. A natural
conjecture is that in a curved supergravity/super-YangMills background for the heterotic
superstring, nilpotency and holomorphicity of d implies the superspace equations of
motion for the background superfields.
Similarly, in the pure spinor formalism for the Type II superstring, the left- and rightmoving BRST operators are2

Q = d ,
Q = d ,
(3)
where and are independent pure spinor variables satisfying
m
= 0,

m = 0,

(4)
for m = 0 to 9, and d and d are worldsheet variables corresponding to the N = 2 D = 10

spacetime supersymmetric derivatives. In a flat background, d is holomorphic and
nilpotent whereas d is antiholomorphic and nilpotent. So it is natural to conjecture
that in a curved N = 2 D = 10 supergravity background for the Type II superstring, the
superspace equations of motion for the background are implied by the condition that these
properties of d and d are preserved.
In this paper, we shall verify the above conjectures to lowest order in for the
heterotic and Type II superstrings in N = 1 and N = 2 supergravity backgrounds. This
verification will lead to new pure spinor versions of the superspace constraints for tendimensional N = 1 Type IIA and Type IIB supergravity. These have the property that they
are remarkably compact and may be useful for studying other aspects of ten-dimensional
supersymmetric theories such as harmonic superspace. Furthermore, since the superstring
action is quantizable, this conjecture can be used in principle to compute the superspace
equations of motion to arbitrary order in .
For the N = 1 D = 10 supergravity/super-YangMills background of the heterotic
superstring, nilpotency of d will imply
I
T C = HC = F
= 0,
(5)
I are the superspace torsion, three-form field strength and

where TAB C , HABC and FAB
super-YangMills field strength. These equations are identical to those derived from pure
1 We will use the notation where m and m are 16 16 symmetric matrices which form the off-diagonal
blocks of the 32 32 ten-dimensional -matrices in the Weyl representation.

2 Throughout this paper, we will use spinor notation simultaneously for the Type IIA and Type IIB superstring
by imposing that and denote D = 10 spinors of opposite chirality for the IIA superstring and denote spinors
of the same chirality for the IIB superstring.
78
spinor integrability in [27]. Since (5) must be satisfied for an arbitrary pure spinor
satisfying (2), (5) implies that
I
(mnpqr ) T C = (mnpqr ) HC = (mnpqr ) F
=0
(6)
for any self-dual five-form direction mnpqr. Up to conventional constraints (which will
be implied by holomorphicity of d ), the constraints of (6) will be shown to imply the
standard N = 1 supergravity/super-YangMills equations of motion.
For the N = 2 D = 10 supergravity background of the Type II superstring, nilpotency
of d and d will imply
T C = T C = T C = 0,

HC = H C
= H C
= 0.
(7)
Since and are arbitrary pure spinors satisfying (4), (7) implies that
(mnpqr ) T C = (mnpqr ) T C = T C = 0,
(mnpqr ) HC = (mnpqr ) H C
= H C
=0
(8)
for any self-dual five-form direction mnpqr. Up to conventional constraints (which will be
implied by holomorphicity and antiholomorphicity of d and d ), the constraints of
(8) will be shown to imply the standard Type II supergravity equations of motion.3
In Section 2 of this paper we shall use the heterotic superstring sigma model to derive
a pure spinor version of the N = 1 supergravity/super-YangMills constraints, and in
Section 3 we shall use the Type II superstring sigma model to derive a pure spinor
version of the Type IIA and Type IIB supergravity constraints. In Section 4 the pure spinor
description of Type IIB supergravity will be shown to agree with the standard HoweWest
(HW) superspace description of [30]. In Section 5 we shall briefly discuss the procedure
for extending to higher order in these computations of the ten-dimensional superspace
constraints.
2. Heterotic superstring sigma model

In this section, the pure spinor version of the heterotic superstring will be reviewed
in flat and curved backgrounds. Nilpotency and holomorphicity of d will then be
shown to imply the superspace equations of motion for the supergravity/super-YangMills
background.
C
3 As will be explained in Section 3, the superspace torsion T
AB appearing in (7) and (8) is not the usual one
since some of its components depend on a left-moving spin connection and some of its components depend on
a right-moving spin connection.
79
2.1. Heterotic superstring in a flat background

In the pure spinor version of the heterotic superstring, the worldsheet variables consist
of the N = 1 D = 10 superspace variables (x m , , p ) for m = 09 and = 116, where
p is the conjugate momentum to , as well as the left-moving pure spinor ghost variable
and its conjugate momentum w , the E8 E8 or SO(32) right-moving currents JI ,
c)
and (b,
right-moving Virasoro ghosts. Because is defined to satisfy (2), it has only
eleven independent degrees of freedom and its conjugate momentum w is only defined up
to the gauge transformation w = m ( m ) for any m . This gauge transformation can
be used to eliminate five components of w , so both and w have eleven independent
components.
The action and stress-tensor in a flat background is

1
1 m
2
c + S + SJ ,
+ b
S=
(9)
x
z
+
p
d
x
2
2

1 m
1
T = x xm p + T ,
2

1
1
m x
m b c (
b c)
+ TJ ,
T = x
(10)
2
where S and SJ are the actions for and J I , and T and TJ are the c = 22 and c = 16
stress tensors for and J I . As described in [29], one can write explicit expression for S
and T by solving the constraint of (2) in terms of eleven chiral bosons ( , uab ) and their
conjugate momenta (, v ab ) where a = 1 to 5 and uab = uba . However, these explicit
expressions will not be necessary here. We will only need to know that S is defined such
that has no singular OPE with itself and Lorentz currents N mn can be constructed out
of and its conjugate momentum w as N mn = 21 mn w which satisfy the OPEs
N mn (y) (z)
1 mn (z)
,
yz
2
N kl (y)N mn (z)
kn lm km ln
m[l N k]n (z) n[l N k]m (z)
3
.
y z
(y z)2
(11)
(12)
Similarly, the explicit expression for SJ will not be necessary and we will only need to
know the OPE
J I (y)J K (z)
L
I K
I K J (z)
,
+
f
L
(y z)
(y z)2
(13)
where fLI K are the E8 E8 or SO(32) structure constants.

Physical states of the superstring are defined as vertex operators in the cohomology of
the left- and right-moving BRST operators

=
Q = d ,
(14)
Q
cT + c cb ,
80
where
i m
1 m
xm +
m ,
d = p
2
8
i
m = x m + m
2
are spacetime supersymmetric and satisfy the OPEs [32]
(15)
m
m (z),
d (y)d (z) i (y z)1
m
(z).
d (y) m (z) i (y z)1
(16)
To construct the sigma model for the heterotic superstring, it will be useful to know the
integrated form of the massless supergravity and super-YangMills vertex operators, VSG
and VsYM , which are

2
VSG = d z Am (x, ) + n Anm (x, ) + d Em

(x, )

1
np
m,
+ Nnp m (x, ) x
(17)
2

VsYM = d 2 z AI (x, ) + n AnI (x, ) + d WI (x, )

1
np
+ Nnp UI (x, ) JI ,
(18)
2
where N np are the Lorentz currents for the pure spinor. Note that the first two terms in VSG
and VsYM are the same as in the GreenSchwarz heterotic superstring vertex operators,
but the third and fourth terms are needed for the vertex operators to be BRST invariant.
These two vertex operators can be obtained by taking the product of a massless open
superstring vertex operator,

1
Vopen = dz A (x, ) + n An (x, ) + d W (x, ) + Nnp U np (x, ) ,
2
(19)

m
I
with either d z x or d z J .
Using the fact that is proportional to mnpqr ( mnpqr ) and the OPEs of (11) and
SG = 0 implies that
(16), one can check that QVSG = QV
D Am = 0,
npqrs
m (m An n Am ) = 0,
i
i
Em
= n (D Anm n Am ),
Anm = D n Am ,
8
10
1 np
np
E = [n Ap]m ,
m = D
m
8
(20)
(21)
m is the N = 1 D = 10 supersymmetric derivative. Similarly,

where D = + 2i
m
sYM = 0 implies that
QVsYM = QV
D AI = 0,
mnpqr
(22)
81
i
i
AnI = D m AI ,
WI = n (D AnI n AI ),
8
10
1
UnpI = D (np ) WI = [n Ap]I .

(23)
8
Eqs. (20) and (22) are the linearized N = 1 supergravity and super-YangMills equations
of motion written in terms of the superfields Am and AI , and Eqs. (21) and (23) define
the linearized supergravity and super-YangMills connections and field-strengths in terms
of Am and AI . For example, the on-shell graviton hnm and gluon anI are contained
in the i( n ) hnm (x) and i( n ) anI (x) components of Am (x, ) and AI (x, ). The
linearized equations of (20)(23) will be generalized to covariant non-linear equations in
the following subsections.
2.2. Heterotic superstring in a curved background
The heterotic sigma model action in a curved background can be constructed by
adding the massless vertex operators of (17) and (18) to the flat action of (9), and
then covariantizing with respect to N = 1 D = 10 super-reparameterization invariance.
Alternatively, one can consider the most general action constructed from the worldsheet
variables which is classically invariant under worldsheet conformal transformations. In
addition, for quantum worldsheet conformal invariance, one needs to include a Fradkin
Tseytlin term which couples the spacetime dilaton to the worldsheet curvature.
Using the worldsheet variables defined in the previous subsection, we can write the
heterotic sigma model action in the form

1
1
2
N + EM
M
z
(Z)d Z
d
GMN (Z) + BMN (Z) Z M Z
S=

2
2
M + AMI (Z)Z M JI + W (Z)d JI
+ M (Z) w Z
I
1
1
I

c
+ UI w J + (Z)r + b
2
2
+ S + SJ ,
(24)
where M = (m, ) are curved superspace indices, Z M = (x m , ), A = (a, ) are tangent
superspace indices, S and SJ are the same as in the flat action of (9), r is the worldsheet
curvature, and [GMN , BMN , EM , M , AMI , WI , UI , ] are the background superfields. The metric GMN is defined in terms of the vectorial part of the super-vielbein by
M
A
GMN = EN b EM a ab , and we shall define
A to be the inverse of EM .
E
2
Ignoring the FradkinTseytlin term d z (Z)r, (24) is the most general action with
classical worldsheet conformal invariance and zero ghost number which can be constructed
from the heterotic worldsheet variables. Note that d carries conformal weight (1, 0),
carries ghost number + 1 and conformal weight (0, 0), and w carries ghost number 1
and conformal weight (1, 0). Since the conjugate momentum ghost variable w can only
appear in combinations which preserve the gauge invariance w = a (a ) , the back
ground superfields M and UI must satisfy ( bcde ) M = ( bcde ) UI = 0,

i.e.,
1
1
(s)
+ M cd (cd ) ,
UI = UI(s) + UIcd (cd ) .
M = M
(25)
2
2
82
It is worthwhile to pause here to say a few words about the geometry of the target space
which is implied by this action. Clearly, we identify EM A as the usual super-vielbein
matrix, BMN as the two-form potential and as the dilaton. The superfield AMI is
the super-YangMills potential while the superfields WI and UI will turn out to be
related to the spinor and vector super-YangMills field strengths. The way in which the
supervielbein enters into the action indicates that the tangent space should be a direct sum
of bosonic and fermionic subspaces. This is different from the structure of the tangent
space in the GreenSchwarz formalism since the EM components of the super-vielbein
do not appear in the GS action. So one only needs to specify the fermionic subspace of
the GS tangent space (or, dually, the bosonic subspace of the GS cotangent space). The
form of the metric GMN = EN b EM a ab shows that the structure group in the bosonic
sector is the Lorentz group while the existence of pure spinors implies that the fermionic
structure group is the spin group times scale transformations. At this stage, the two Lorentz
groups (in the spinor and vector sectors) are independent, although later on we shall choose
a gauge with respect to one of them after which they will become identified. Note also that
the spin connection M appears explicitly in the action. This implies that conventional
constraints corresponding to tensorial shifts of the connection are restricted by the demand
that the BRST operator and action be unchanged.
Taking all this into account we find that, in addition to being invariant under targetspace super-reparameterizations, the action of (24) is invariant under the local gauge
transformations
b
d
= cd bc EM
,
EM
EM = EM ,
M = M + M M ,
WI = WI ,
= ,
UI = UI UI ,
w = w ,
(26)
where = (s) + 12 bc (bc ) , bc and bc parameterize independent local Lorentz

transformations on the vector and spinor indices, and (s) parameterizes local scale
transformations on the spinor indices. Furthermore, the action of (24) and the BRST
operator d are invariant under the local shift transformations

(s) = (c ) hc ,
bc = 2 [b hc] ,
d = w ,
= EM M ,
hc
UI = WI ,
(27)
where
is a local gauge parameter, and the transformation of
has been chosen such that d = 0. Note that d can be treated as an independent
variable in the action of (24) since p does not appear explicitly.
The first term in the first and second line of (24) is the standard heterotic GS action,
but the other terms will be needed for BRST invariance, just as in the linearized vertex
operators of (17) and (18). As will now be shown to lowest order in , nilpotency and
holomorphicity of d implies the equations of motion for the background superfields
in (24). Note that nilpotency and antiholomorphicity of the right-moving BRST current,
does not impose any conditions to lowest order in because the action of (24)
cT + c cb,
is classically conformally invariant.
83
2.3. Heterotic nilpotency constraints

We shall first derive the constraints coming from nilpotency of Q = d . Defining
the canonical momentum PM in the usual manner as PM = L/(0 Z M ), one finds that

N

1

M
N
I
d = E PM + BMN Z Z M w AMI J .
(28)
2
Using the canonical commutation relations

N
,
w , = ,
PM , Z N = M
I J
J , J = fKI J JK ,
(29)
one computes that

1
M R w FI JI ,
{Q, Q} = T C DC + HM Z M Z
2
(30)
C
M
I
where DC = E (PM M w AMI J ). The torsions TAB , three-form HABC ,
C
curvatures RAB , and field strengths FABI in (30) are defined by

(s)
[A , B ] = TAB C C + RAB S + RAB ab Mab + FABI Y I ,

M N P
EB EC [M BNP ] ,
HABC = 3EA
(31)
(s)
M ( + S + ab M
I
where A = EA
M
ab + AMI Y ), S is a scale generator which
M
M
M
M
(s)
transforms E = E , Mab is the Lorentz generator, Y I is the gauge group
(s)
generator and RAB = RAB + 12 RAB cd (cd ) . Note that f[AB] signifies the graded
commutator, i.e., f[AB] = 12 (fAB + fBA ) when both indices are fermionic and f[AB] =
1
2 (fAB fBA ) otherwise.
So nilpotency of Q implies the constraints
T C = HB = R = FI = 0
(32)
for any
satisfying the pure spinor constraint of (2). Note that the
=0
constraint is implied by T C = 0 through Bianchi identities.
As shown in [27], the constraints (32) follow from pure spinor integrability in loop
superspace and imply all the essential N = 1 supergravity/super-YangMills constraints.
Indeed, the chirality operator introduced in [27] in pure spinor loop superspace precisely
coincides with the BRST operator Q. So (32) implies all but the conventional constraints
which define the vector components of superfields in terms of their spinor components and
define the spin connection in terms of the super-vielbein. As will be shown below, these
conventional constraints (up to gauge invariances) are implied by the holomorphicity of
d .
2.4. Heterotic holomorphicity constraints
We shall now derive the constraints coming from holomorphicity of d . Varying

and its conjugate momentum in (24) and ignoring the contribution from the Fradkin
84
Tseytlin term which is higher order in , one obtains the equations

= M Z
M + UI JI ,

= M Z
M + UI JI w ,
w
(33)
and varying the right-moving variables, one obtains the equations

JK = fJI K AMI Z M + WI d + UI w JJ ,
(34)
where fJI K are the Lie algebra structure constants. And by varying d , one obtains the
equation of motion
M = W JI .
EM Z
I
Finally, by computing EP (S/Z P ), one obtains the equation of motion

d = EP [P E a E b ab + [P E a E b ab + 1 HP MN Z M Z
N
M] N
N] M
2
N

+ 2 [P EN] d + [P N] w Z
P w AP I JI
I

+ 2[P AM]I Z + P WI d + P UI w J .
(35)
(36)
Putting these equations together, one finds

d = 1 (Tbc + Tcb + Hbc ) b + 1 (Tc + Hc )
2
2
c
+ Tc d + Rc w

1
1
FbI WI (Tb + Hb ) b + F I WI H JI
2
2

+ WI T WI UI d + U I R WI w JI ,
(37)
A Z M ,
M and TABc = TAB d cd .
A = E A Z
where
= EM
M
d ) = 0 implies the constraints
So from (37), (
Rc = 0,
T(ab) = Hab = Tc + Hc = Tc = 0,
1
FbI = WI Tb ,
FI = WI H ,
2

UI R WI = 0,
WI T WI = UI ,
(38)
where is any spinor satisfying (2).

The constraints of (32) and (38) will now be shown to imply the correct supergravity
and super-YangMills equations of motion.
85
2.5. N = 1 supergravity/super-YangMills constraints

It will be useful to first consider the supergravity constraints of (32) and (38) which
have lowest scaling dimension since the higher-dimensional constraints will be implied
by these constraints through Bianchi identities. At dimension 12 , the only constraint is
H = 0 which implies that H = 0 since there is no non-zero symmetric H
satisfying H = 0.
At dimension 0, the constraints T c = Hc = 0 and T c = cd Hd
imply that T c = cd Hd = i( d ) fdc for some fdc . The dimension-zero Bianchi
c
D H
identity D( H ) = T(
)D then tells us that fd is an SO(9, 1) matrix times a scale
factor. So using the local spinor Lorentz and scale transformations of (26), fdc can be
gauge-fixed to dc so that

T c = cd Hd = i c .
(39)
Note that at this point we still have one local Lorentz symmetry, acting now on both
spinor and vector indices. The connection for this symmetry is M ab . On the other hand,
the fermionic scale invariance has been fixed and so it need not be the case that other
components of the torsion should respect this symmetry.
At dimension 12 , the constraint T = 0 implies that T = fc ( c ) for some
fc . Using the shift symmetry of (27), fc can be gauge-fixed to zero so that T = 0.

The other dimension 12 constraints, Hcd = T(cd) = 0, imply through the Bianchi identity
( T ) c = T( D T )D c that T b c = 2cd (bd ) (s).
At dimension one, the constraint Tc = 0 decomposes into
defg
Tc = Tc
(defg ) + Tcde (de ) + Tc = 0.
(40)
The constraints Tc = 0 and Tcde = 0 determine the vector components of the spin
defg
= 0 is implied by the Bianchi
connections c(s) and c de , whereas the constraint Tc
bdefg
identity (H + T H )bc (
) = 0. Similarly, the constraints involving the curvature
tensor in (32) and (38) are implied by the Bianchi identity R[ABC] D = [A TBC] D +
T[AB E TC]E F .
To extract the supergravity equations unambiguously from the above constraints it is
convenient to reduce the structure group from Lorentz group times fermionic scale to just
the Lorentz group. The dimension-zero torsions are unchanged but the dimension one-half
torsion T gets amended to
(s)
(s)
T T 2( ) = 2( ) .
(41)
There are corresponding changes at higher dimensions. The leading component of

(s) is the dilatino and to show that there are no unwanted fields one must show that
this superfield is proportional to the spinorial derivative of a scalar superfield whose
leading component is the dilaton. It is straightforward to verify that this is the case,
although it is necessary to go to dimension three-halves to do so. As discussed in Section 5,
holomorphicity of d to the next order in will imply that this scalar superfield is
the same superfield that appears in the FradkinTseytlin term of (24).
86
The above supergravity constraints therefore imply that all of the supergravity
superfields can be expressed in terms of the spinor supervielbein E M , and the equation
T c = i( c ) puts E M on-shell. Similarly, the super-YangMills constraints in (32)
and (38) imply that the super-YangMills superfields AcI , WI and UI can be
expressed in terms of the spinor superfield AI , and the equation FI = 0 puts AI onshell. So nilpotency and holomorphicity of d has been shown to imply the N = 1
supergravity/super-YangMills equations of motion to lowest order in .
3. Type II superstring sigma model

In this section, the pure spinor version of the Types IIA and IIB superstring will be
reviewed in flat and curved backgrounds. Nilpotency and holomorphicity of d and
nilpotency and antiholomorphicity of d will then be shown to imply the superspace
equations of motion for the N = 2 supergravity background.
3.1. Type II superstring in a flat background
In the pure spinor version of the Type II superstring, the worldsheet variables consist of
the N = 2 D = 10 superspace variables (x m , , p , , p ) for m = 09 and , = 116
where p is the conjugate momentum to and p is the conjugate momentum to . For
the Type IIA superstring, and denote SO(9, 1) spinors of opposite chirality while for
the Type IIB superstring, and denote SO(9, 1) spinors of the same chirality. The pure
spinor formalism also contains the worldsheet variables and , and their conjugate
momenta w and w , which are constrained to satisfy the pure spinor conditions
m = 0,
m = 0
(42)
for m = 09. In a flat background, , p , and w are left-moving while , p ,
and w are right-moving.
The action and stress-tensor in a flat background is

1 m
1
2
z
+
p
+
p
d
x
+ S + S ,
S=
(43)
m
2
2

1
1
T = x m xm p + T ,
2

1
1
m
T = x xm p a + T ,
(44)
2
where S and S are the actions for and , and T and T are the c = 22 left- and
right-moving stress tensors for and . As in the heterotic case, the explicit form of
S and S will not be needed. We will only need to know that one can construct left- and
mn = 1
mn w,
right-moving Lorentz currents, N mn = 21 mn w and N
which satisfy
2
the OPEs
N mn (y) (z)
1 mn (z)
,
y z
2
(z)
1
mn (y)
N
,
(z) mn
2
y z
(45)
kn lm km ln
m[l N k]n (z) n[l N k]m (z)
3
,
y z
(y z)2
m[l k]n
n[l k]m
kn lm
km ln
kl (y)
mn (z) N (z) N (z) 3 .
N
N
y z
(y z )2
N kl (y)N mn (z)
87
(46)
(47)
Physical states of the superstring are defined as vertex operators in the cohomology of
the left- and right-moving BRST operators

= d ,
Q
Q = d ,
(48)
where
i m
1 m
d = p
xm +
m ,
2
8
i
m + 1 m ,
d = p m x
2
8 m
are spacetime supersymmetric and satisfy the OPEs
i
m = x m + m ,
2
m = x
m + i m ,
(49)
(50)
m
m (z),
d (y)d (z) i (y z)1
m
d (y) m (z) i (y z)1
(z),
(51)
m (z),
d (y)
d (z) i (y z )1 m

m (z) i (y z )1 m (z).
d (y)

(52)
To construct the sigma model for the Type II superstring, it will be useful to know the
integrated form of the massless Type II supergravity vertex operator

2
m Am (x, , )
+
+ m Am (x, , )
VSG = d z A (x, , )

+ d E (x, , )
+
m Em
n Amn (x, , )
+ m
(x, , )

+ m Em
+ d E (x, , )
(x, , )

1
+
p pmn (x, , )
+ Nmn mn
(x, , )
2

1 mn
(x, , )
pmn (x, , )
+ p
+ d d P (x, , )
+ N
mn
2
mn C
mn (x, , )
+ d N
+ Nmn d C mn (x, , )
pq S mnpq (x, , )
.
+ Nmn N
(53)
Note that the first line of VSG is the same as in the GreenSchwarz Type II superstring
vertex operator, but the other lines are needed for the vertex operator to be BRST invariant.
The Type II superstring vertex operator of (53) can be understood as the square of the
open superstring vertex operator of (19).
88
SG = 0
Using (42) and the OPEs of (45) and (51), one can check that QVSG = QV
implies that
mnpqr
D A = 0,

A = 0,
mnpqr
D
i
i
An = D n A ,
A n = D
n A ,
8
8
1
m n A ,
Amn = D D
64
S) in terms of A
P , C, C,
and similar equations for the superfields (E, , ,
(54)
(55)
.
Note that
= + i m m
D
2

are the N = 2 D = 10 supersymmetric derivatives. Eqs. (54) are the linearized N = 2
supergravity equations of motion written in terms of the superfield A , and equations
(55) define the linearized supergravity connections in terms of A . For example, the
These
on-shell graviton hnm is contained in the ( n ) ( m ) hnm (x) of A (x, , ).
linearized equations will be generalized to covariant non-linear equations in the following
subsections.
D =
i
+ m m ,
2
3.2. Type II superstring in a curved background

The Type II sigma model action in a curved background (except for the Fradkin
Tseytlin term) can be constructed by adding the massless vertex operator of (53) to
the flat action of (43), and then covariantizing with respect to N = 2 D = 10 superreparameterization invariance. Alternatively, one can consider the most general action
constructed from the worldsheet variables which is classically invariant under worldsheet
conformal transformations.
Using the worldsheet variables defined in the previous subsection, we can write the Type
II sigma model action as

1
1
2
N
z
S=
d
GMN (Z) + BMN (Z) Z M Z

2
2
M + E (Z)d Z M + M (Z) w Z
M
+ E (Z)d Z
M
M
(Z) w Z + P (Z)d d
+
+ C (Z) w d
+ S + S ,

+C
(Z) w d + S (Z) w w + (Z)r
(56)
where M = (m, , )
are curved superspace indices, Z M = (x m , , ), A = (a, , )
are tangent superspace indices, S and S are the same as in the flat action of (43),

c
d
M ,
EN
, BMN , EM
, EM
, M ,
r is the worldsheet curvature, and GMN = cd EM

P , C , C
, S , are the background superfields.
89

If the FradkinTseytlin term, d 2 z (Z)r, is omitted (56) is the most general action
with classical worldsheet conformal invariance and zero (left, right)-moving ghost number
which can be constructed from the Type II worldsheet variables. Note that d carries
conformal weight (1, 0), d carries conformal weight (0, 1), carries ghost number (1, 0)
and conformal weight (0, 0), carries ghost number (0, 1) and conformal weight (0, 0),
w carries ghost number (1, 0) and conformal weight (1, 0), and w carries ghost number
(0, 1) and conformal weight (0, 1). Since w and w can only appear in combinations
which commute with the pure spinor constraints of (42), the background superfields must
satisfy

bcde

M = bcde C = bcde C
= bcde

= bcde S = bcde S = 0,
(57)
and the different components of the spin connections will be defined as

1 cd
(s)
M =
(s) + 1
cd (cd ) .
(cd ) ,
M = M + M
(58)
M
2
2 M
Although the background superfields appearing in (56) look unconventional, they all
have physical interpretations. The superfields EM A , BMN and are the supervielbein,
two-form potential and dilaton superfields, P is the superfield whose lowest components

are the Type II RamondRamond field strengths, and the superfields C = C +
1 ab
=C
+ 1 C
ab (ab ) are related to the N = 2 D = 10 dilatino
(ab ) and C
2C
2
and gravitino field strengths. As in the heterotic sigma model, the form of the metric
in the Type II sigma model implies that the structure group in the bosonic sector is
the Lorentz group. But there are now two independent pure spinors, so one has two
independent fermionic structure groups, each consisting of the spin group times scale
transformations. One therefore has two independent sets of spin connections and scale
(s)
ab ) and (
(s),
ab ), which appear explicitly in the Type II sigma
, M
connections, (M
M
M

model action. Finally, the background superfields S appearing in (56) will be related to
curvatures constructed from these spin and scale connections. Note that a similar relation
occurs in the RNS sigma model action which contains the terms

ab
1
ab
m+
b x m + Sabcd (x) a b si
c si
d ,
a si
m
(x)a b x
(x)si
d 2 z m

4
(59)
a
a
m
a
a
m
a
where = em (x) , si = em (x)si , and em (x) is the target-space vielbein.

In addition to being target-space super-reparameterization invariant, the action of (56)
is invariant under the local gauge transformations
b
d
= cd bc EM
,
EM
E ,
EM
=
M
EM
= EM ,
M = M + M M ,
+

,

M = M
M
M
= ,
w = w ,
,
=
w , (60)
w =
90
where
1
= (s) + bc (bc ) ,
2
bc bc bc

, ,
1 bc
(bc ) ,
=
(s) +
parameterize independent local Lorentz transformations on the [vector, unhatted spinor,

(s) parameterize independent local scale transforhatted spinor] indices, (s) and
mations on the unhatted and hatted spinor indices, and the background superfields

, S transform according to their spinor indices.
P , C , C
Furthermore, the action of (56) and the BRST operators d and d are invariant
under the local shift transformations

(s) = (c ) hc ,
bc = 2 [b hc] ,
d = w ,

(s) = (c ) h c ,
bc = 2 [b h c] ,
w ,
d =
= P
,
C
C = P ,

S = C
+ C ,
(61)
where hc and h c are independent local gauge parameters and the transformations of
have been chosen such that d = d = 0. Note that d and d can be
and

treated as independent variables in (56) since p and p do not appear explicitly.

The first line of (56) is the standard Type II GS action, but the other lines are
needed for BRST invariance. As will now be shown to lowest order in , nilpotency
and holomorphicity of d and nilpotency and antiholomorphicity of d imply the
equations of motion for the background superfields in (56).
3.3. Type II nilpotency constraints

= d , it is
To analyze the conditions implied by nilpotency of Q = d and Q
convenient to use the canonical momenta PM = L/(0 Z M ) to write

1
N M w
w ,
d = EM PM + BMN Z N Z
M
2

N

1

N

d = EM
(62)
B
Z
M
MN
M
M
2
Using the canonical commutation relations

N
,
PM , Z N = M

w , = ,
w , = ,
one finds that

1
N HN R w R
w ,
{Q, Q} = T C DC + Z N Z
91

1
N H R w R
w ,
T C DC + Z N Z
N

2

= T C DC + 1 Z N Z
N H R w R
w ,
{Q, Q}
2

M w , TAB and RAB are defined using
where DC = ECM PM M w
are defined using the

spin
the M spin connection, and TAB and R
AB
M
connection.
implies that
So nilpotency of Q and Q
Q}
=
{Q,
= R = 0,
T C = HB = R
= R = 0,
T C = H B
= R

T C = H B
= R = R = 0,
(63)
for any pure spinors and satisfying (42). As in the heterotic case, the nilpotency
constraints on RABC D are implied through Bianchi identities by the nilpotency constraints
on TAB C .
As will be discussed in Section 4, the constraints of (63) can be interpreted as Type
II pure spinor integrability conditions and imply all the essential Type II supergravity
constraints. The remaining conventional Type II supergravity constraints will be implied
by the holomorphicity and antiholomorphicity of d and d .
3.4. Type II holomorphicity constraints
To derive the constraints coming from holomorphicity of d and antiholomorphicity
of d , first vary , w , and w in (56) to obtain the equations

= M Z
M + C d + S w ,

= M Z
M + C d + S w w ,
w

Z M + C
=
d + S w ,
M

M Z M + C
w =
d + S w w .
(64)
And by computing EP (S/Z P ), one obtains the equation of motion

d = EP [P E a E b ab + [P E a E b ab + 1 HP MN Z M Z
N
M] N
N] M
2

N
+ 2 [P EN] d + [P N] w Z

w Z N P w
P w
+ 2 [P EN] d + [P
N]
(Z) w d
+ P P d d + P C (Z) w d + P C
+ P S (Z) w w .
(65)
92
Putting these equations together, one finds

1
d = 1 TBc B
c + c
B + HBC B
C
2
2

B + TB B d + P + C d d
+ TB d
B B w
B + R
+ RB w

d w + S w w
+ C w d + C

+ S d w ,
+ C
(66)
A Z M ,
M , TABc = cd TAB d , and all superspace derivatives
A = E A Z
where A = EM
M
acting on unhatted spinor indices are covariantized using the M connection while all
M
superspace derivatives acting on hatted spinor indices are covariantized using the
connection. Furthermore, the torsion TAB and curvature RAB are defined as in (31)
AB are defined
using the M connection whereas the torsion TAB and curvature R
M connection. Note that TAbc appears only in the combination T(bc) . This
using the
(s)
(s) only act on spinor
and
combination is independent of the spin connections since M
M
ab
ab
are antisymmetric in their vector indices.
indices and since M and
M
Plugging into (66) the equations of motion which come from varying d and d ,
w ,
= P d C
= P d C w ,
(67)
one finds that holomorphicity of d implies that

T(bc) = Hcd = H
= Tc + Hc = T c
H c
= 0,
1
Tc + T c P = Tc T c P = T H P = T = 0,
2
+ Tc C
1 H C = 0,
= R
C + P T P = R
c

T C
= 0,
P + C
S +R

Rc + T c
= R
= 0,
C

C R C
C R P = S R
= 0, (68)
where the last two lines of equations must be satisfied for any pure spinor .
Antiholomorphicity of d implies the hatted version of the above equations. The only
+ Hc
= 0, which together with the
subtle point is that it implies T c
H c
= Tc
above equations implies that

Tc + Hc = T c
H c
= T c
= H c
= 0.
(69)
The constraints of (63) and (68) will now be shown to imply the correct Type II
supergravity equations of motion.
93
3.5. Type II supergravity constraints

The analysis of the Type II constraints of (63) and (68) will closely resemble the analysis
of the heterotic constraints in Subsection 2.5. At scaling dimension 12 , the constraints of
(63) imply that
H = H = H = H = 0
(70)
since there is no non-zero symmetric H and H satisfying H = 0 and
H = 0.

At dimension 0, the constraints T c = T c = 0 imply that T c =

i( d ) fdc and T c = i( d ) fdc for some fdc and fcd . Using the dimension-zero Bianchi
identities and the local Lorentz and scale transformations of (60) for the unhatted and hatted
spinor indices independently, both fdc and fdc can be gauge-fixed to dc . After this gaugefixing, the only remaining gauge invariance is a single local Lorentz invariance which
acts on all spinor and vector indices in the standard fashion. Combining with the other
dimension-0 constraints of (63) and (68), one has

c
T c = cd H d
T c = cd Hd = i c ,
= i ,
T c = H c
= 0.
(71)
At dimension 12 , the constraints T = 0 and T = 0 imply that T =
fc ( c ) and T = fc ( c ) for some fc and fc . Using the shift symmetries of (27),
both fc and fc can be gauge-fixed to zero so that T = T = 0. Other dimension- 21

constraints,
Hcd = T(cd) = T = T = 0,
Hcd
= T(cd)
= T = T
= 0,
imply through the Bianchi identities (T

T b c = 2cd (bd ) (s),
(72)
+ T T )c
= 0 and (T
(s)
T b c = 2cd (bd )
,
+ T T )c

= 0 that
(73)
bc spin connection and T

b c is defined using the
bc spin
where T b c is defined using the M
M
c
=0
connection. Furthermore, Bianchi identities (T + T T ) = 0 and (T + T T )c

imply that
c
(s) = 0,
T b c = Tb
= (s)
=
(74)
bc
c
b c is defined using the
bc spin connection and Tb
where T
is defined using the M
M
spin connection.
At dimension one, the constraint Tc = Tc = 0 decomposes into

defg
Tc = Tc
Tc = T c
defg
(defg ) + Tcde (de ) + Tc = 0,
cde (de ) + T c = 0.
(defg ) + T
(75)
94
c = 0 and Tcde = T cde = 0 determine the vector components of the

The constraints Tc = T
(s) (s)
cdefg =
c de , whereas the constraint Tcdefg = T
spin connections c , c , c de and
bdefg
0 is implied by the Bianchi identities (DH + T H )bc (

) = 0 and (DH +
bdefg
T H )bc (
) = 0. The constraints Tc = i(c ) P
and Tc
= i(c ) P
for some P and P are implied by the Bianchi identities (T + T T ) = (T +
T T ) = 0. And P = P is implied by the Bianchi identity (T + T T )c = 0.
Similarly, all other constraints in (63) and (68) are either implied by Bianchi identities or

define C , C
and S in terms of the supervielbein.
The above constraints imply that all background superfields appearing in the action of
(56) can be expressed in terms of the spinor supervielbein EM and EM
. Furthermore, the
constraints

T c = i c ,

T c = i c ,
T c = 0
(76)
. So the constraints of (63) and (68)

imply the on-shell equations of motion for EM and EM
imply the Type II supergravity equations of motion. In the following section, the above
pure spinor description of Type IIB supergravity will be related to the HoweWest (HW)
description of [30]. It should similarly be possible to relate the pure spinor description of
Type IIA supergravity to the IIA superspace description of [33].
4. Relation with the SL(2, R)-covariant description of IIB supergravity

In this section we shall demonstrate that the constraints on the torsion for IIB derived
in the preceding section are indeed equivalent to the HW equations of motion of IIB
supergravity described in [30]. We shall do this by first showing that the latter are
generated by the standard dimension-zero torsion constraint and then exhibiting the explicit
transformation from the standard IIB superspace torsions to those derived above. In order
to carry through the first step we use the method of Weyl superspace and then we reduce
the structure group to the Lorentz group. In order to establish the result fully we also have
to examine the scalars in the theory.
The complete IIB supergravity theory was derived from a superspace perspective in
[30]. However, although complete results were given there for all of the superspace field
strength tensors, no attempt was made to identify a minimal generating set of constraints.
Moreover, the HW formalism is manifestly locally U (1)- and globally SL(2, R)-invariant
and this is not convenient for the applications we have in mind here. We shall work initially
in an SO(2) formalism (rather than U (1)) since this will be easier to adapt to our purposes.
For IIB superspace we use the same HW conventions as in [30], although we use to
denote the 16 16 spin matrices instead of . To convert SO(2) spinor indices i, j, . . . to
U (1) indices, we write

1
vi v = v1 v2
2
(77)
95
and

1
vi v = v 1 v 2 .
2
So the metric and J-tensor are
+ = 1,
J + = i,
The summation is therefore ui vi
(78)
J+ = i,
= u+ v+ +u v .4
J+ + = i.
(79)
To convert SO(2) vector indices r, s, . . .
to U (1) indices, we have

vr = (r )ij vij vij = r ij vr ,
(80)
where
1
r = (3 , 1 ).
2
We can then put

1
v = v 1 v 2
2
(81)
(82)
for vector indices and this is consistent since ( ++ )++ = 1.

In Subsection 4.1, we shall first show that the equations of motion of IIB supergravity
follow (up to topological niceties) from the usual dimension-zero constraint

Tij c = iij c .
(83)
We shall do this by working in Weyl superspace, i.e., we shall include a scale factor in
the structure group. Following through the consequences of this we find that the scale
curvature vanishes so that the scale connection is pure gauge. If we then take it to vanish
we recover the equations of [30]. This procedure is very similar to the approach used in
[31] to prove that the equations of motion of D = 11 supergravity follow from the standard
dimension-zero constraint.
Since the standard dimension-zero constraint of (83) is required by the nilpotency of
Q, it then follows that the equations of motion of IIB supergravity are indeed implied in
the pure spinor formalism. However, as we have seen, there are many other equations at
dimensions greater than zero that are required to hold either by the nilpotency or by the
holomorphicity of Q. In Subsection 4.2, we shall check these explicitly at dimension onehalf by comparing our results with those of Section 3.
In the HW superspace description of Type IIB supergravity, SL(2, R) global symmetry
is manifest since the two scalars are described by an SO(2)\SL(2, R) coset. However, this
SL(2, R) symmetry is not manifest in the pure spinor description since the dilaton and
4 This causes a slight problem in the superspace summation convention which should be taken to be
i
u vi = u+ v+ +u v , whereas in [30] one finds u v u v . So, in converting from HW conventions to
SO(2), one has to remember to insert an extra minus sign for downstairs indices. This means, for example, that
we must take Tij c = iij ( c ) since then one finds T+ c = i( c ) T c = i( c ) in agreement
with [30].
96
axion do not appear in an SL(2, R)-covariant manner. In Subsection 4.3, we will relate
these two descriptions of the Type IIB scalars and will show that the target-space metric
appearing in the pure spinor version of the Type IIB sigma model is in string gauge.
4.1. Weyl superspace
To get the superspace constraints under control it is useful to include a scale factor in
the connection. The structure group is then Spin(1, 9) Spin(2) R+ . The full connection
(denoted by a tilde) is
a b = a b + 2a b ,

i j = i j + i j + Ji j ,
(84)
(85)
where , , are, respectively, the connections for the Lorentz, U (1)(= Spin(2)) and
scale factors. We shall use the notation to denote the Spin(1, 9) U (1) connection, so
R + M R + M + N , where M
+ . Similarly, for the curvatures we have R
and N are, respectively, the U (1) and scale curvatures.
At dimension one-half we find, using the Bianchi identity,
|E| k) d = 0,
(i Tj k) d + T(ij E T
D
(86)
and the freedom to choose the dimension one-half components of the connection and the
even basis vectors Ea , that the dimension one-half component of the torsion tensor is

Tij k = i a a 2 () ij k ,
(87)
where ij k is totally symmetric and traceless on its Spin(2) indices, while
Tb c = 0.
This is exactly the same as in [30], and we identify the HW spinor field by
= 2 (222 + i111) = i
= 2 (222 i111) = i+++ .
(88)
(89)
(90)
At dimension 1 one has to solve two Bianchi identities

Ec d + Tci Jm T
Jmj d + Tcj Jm TJmi d ,
ij,c d = Tij E T
R
(i Tj k) l + T
(ij e Te k) l + T
(ij Jm T|Jm| k) l .
(ij, k) l = D
R
(91)
After a long and tedious calculation one can show that the only non-zero dimension
one components of the curvature and torsion tensors are those which correspond to the
dimension-one components of the IIB supergravity multiplet, that is Fabc , Pa , Gabcde
together with fermion bilinear terms. The tensors F, P and G are associated with the
antisymmetric tensor gauge fields of the theory and the scalar fields (Pa is essentially
the derivative of the scalar fields). One also determines the spinorial derivative of and
the dimension one component of the U (1) curvature M in terms of these physical fields.
Moreover, one finds that the dimension-one component of the scale curvature N vanishes,
Nij = 0. From this, one immediately concludes with the aid of the scale curvature
97
Bianchi identity, dN = 0, that the whole of N vanishes and so the scale connection is
pure gauge as anticipated. At this stage we can set the scale connection equal to zero and
recover the HW torsions and curvatures of [30]. From these results one can then construct
super extensions of F, G, P which satisfy corresponding Bianchi identities. In particular,
one can deduce the existence of the two scalar fields described by an SL(2, R)\U (1) coset
space.
4.2. Lorentz superspace
To recover the form of the torsion and curvature tensors derived from the pure spinor
formalism, we need firstly to restrict the structure group to be the ten-dimensional spin
group. This means that the components of and will appear in the redefined torsion.
Moreover, we shall choose a different scale gauge from = 0 which means that = dS
for some scalar field S and also that there is change of basis with respect to the HW basis,
a
, etc. Explicitly, we have
i.e., E a = e2s EHW
TAB C = TAB C + 2[A IB] C + 2[A JB] C ,
where
IA
b
b
= Ia j= 2a , j
Ii = i ,
(92)
b
JA = Ja j= 0, j
Ji = Ji ,
B
(93)
and where the mixed spinor-vector components of I and J are zero. In particular, at
dimension one-half, we have

Tij k = Tij k + i k j + Ji k j + j k i + Jj k i ,
(94)
c
c
c

Tib = Tib + 2b i .
(95)
We shall also have to shift the Lorentz connection as
(1,2)
bc bc
= bc + (bc ) Y(1,2).
(96)
The notation here is that the connection labelled i = 1, 2 will act on spinor indices with
the same internal index label. Since the two connections will be different, this procedure
manifestly breaks SO(2). For the moment we shall suppose that the vector indices are acted
upon by the original . Finally, in order to make a direct comparison to the earlier results
we shall have to shift the vectorial basis Ea by

Ea Ea + i a i Ei .
(97)
If we choose
(1)
Y1 = i111,
1 = i111,
(1)
Y2
(2)
Y1
(2)
Y2
2 = i222,
= i222,
= i111,
= i222,
(98)
98
and if, in addition,

i
1 = 111 ,
2
i
2 = 222 ,
2
1 = i222,
2 = i111,
(99)
then we find that all components of the redefined Tij k vanish except for
(s)
T11 1 = 2( )
,
(s),
T22 2 = 2(
)
(100)
where
(s) = i111,
(s) = i222.
(101)
(1,2)
If we also define new vectorial torsions with respect to the new connections bc , we find

(1) c
T1b
= 2 b c (s) ,
(1) c
T2b = 0,
(2)
T1b
= 0,

(2) c
(s) .
T2b = 2 b c
(102)
We shall verify that Eqs. (99) are indeed satisfied in the next section. For the moment,
assuming that they are, we are now in a position to compare directly with the dimension
one-half results coming from the pure spinor formalism. In order to do this, we remove the
fermionic scale connection from the Type II structure group. After identifying the indices
(1, 2) = (, ),
we find that the only non-vanishing components of the redefined torsion
with three spinorial indices are those of (100). The vectorial torsions, which do not need
to be redefined, are those of (102). We have therefore succeeded in demonstrating that the
torsions derived from the pure spinor formalism are indeed in agreement with those of [30]
after suitable field redefinitions. To complete the picture we must verify that = dS and
that the expressions given for i in (99) are correct. To do this, we need to examine the
scalar fields in the theory.
4.3. Scalar fields
The scalar fields take their values in the space SO(2)\SL(2, R). We describe then by a
real two by two matrix U acted on by U hUg 1 , for h SO(2), g SL(2, R). In index
notation we write Ur R . Note that r is vector SO(2) index while R is an SL(2, R) doublet
index. The MaurerCartan form M is given by
M = dU U 1 .
(103)
Since it is Lie-algebra valued it can be written as

Mr s = Pr s + 2Jr s ,
(104)
99
where is the U (1) connection of (85) and P rs is symmetric traceless, i.e., in complex
notation we have P ++++ := P , where P is the HW one-form defined in [30]. The Maurer
Cartan equation, dM + M2 = 0, implies that
DP = 0,
(105)
i
M = P P,
(106)
2
where M is the U (1) curvature tensor.
R , and we define
There is a HW SL(2R) doublet of three-form field strength tensors F
. Assuming that d F
= 0 we find that
F := U F
DFr = Pr s Fs .
(107)
As before, we can identify F = F ++ with the field F of [30]. In a complex basis (107)
reads

DF = P F
(108)
as in [30].
The field U is not quite the same as the HW field V. The two are related by
U = JV 1 J.
(109)
The MaurerCartan form is then

dU U 1 = JV 1 dV J

0 1
2i
P
0 1
=
1 0
P 2i
1 0

2i
P
=
.
P
2i
(110)
In the second line of (110), we have used the formula for the MaurerCartan form in [30]
(with instead of Q), and in the final line we have the correct expression in the new
conventions in a complex basis.
In the physical gauge we can write the components of U in terms of := 1 + i2 :=
C0 + ie , where is the dilaton and C0 is the axion. In the real basis we have

1
1 1
U=
(111)
2 0 2
and one can check that has the expected transformation under SL(2, R), i.e.,
a + b
,
c + d
where g SL(2, R) is

a b
g=
.
c d
(112)
(113)
100
In a complex basis (for both indices)

1
1 + i 1 i
.
U=
2 2 1 + i 1 i
(114)
If one computes the MaurerCartan form in this gauge one finds

id
1
d + ie dC0 =
,
2
22
1
= e dC0 .
4
P=
(115)
(116)
In the HW description, P ++++ is related to of (87) by

j kl
P ++++ = 2E + +++
+ E a Pa++++
(117)
= 0. This implies
and is chiral, i.e., D
D1 C0 = e D2 ,
D2 C0 = e
D1 .
(118)
(119)
Using this we can express the components of as

i
111 = D1 ,
4
i
222 = D2 .
4
We can also express the components of in terms of D as
(120)
(121)
1
1 = D2 ,
(122)
4
1
2 = D1 .
(123)
4
Now earlier we found what the U (1) and scale connections had to be chosen to be at
dimension one-half in order to achieve Q-integrability. We required

= i222,
1

2
= i111,
S=
.
8
(124)
(125)
i

,
1
(126)
=
2 111
i
=
,
2
(127)
2 222
where is the scale connection and where the prime indicates the basis which is related to
the unprimed HW one by E i = eS E i , E a = e2S E a . We also required = dS, since
it is pure gauge. So we can identify
(128)
101
In addition, if we compare the expressions for the components of and in terms of

D, we see that they agree, and so everything works as expected. If we define bosonic
metrics by
G = E b E a ab ,
b
a
G = E E ab ,
(129)
(130)
then G = e/2 G. This means we can identify G with the string metric and G with the
Einstein metric, so the conformal transformation we need to make is precisely the one
which goes between the two frames.
5. Higher-order corrections
In this paper we have verified to lowest order in that nilpotency and holomorphicity
of the pure spinor BRST operator implies the superspace equations of motion for the
background supergravity fields. The next question to investigate is how these superspace
equations of motion are modified by higher-order corrections to the nilpotency and
holomorphicity conditions. Since the sigma model is a free action in a flat background,
one can compute these corrections using standard sigma model methods by separating the
worldsheet variables into classical and quantum parts and expanding in normal coordinates
around a flat background.
When the background fields satisfy their string-corrected equations of motion, one
expects that the -functions of the sigma model should vanish, i.e., that the sigma model
remains conformally invariant at the quantum level. However, unlike the bosonic string
sigma model, quantum conformal invariance is not expected to imply the complete set
of equations of motion for the background fields. In addition, one needs to impose the
conditions that, at the conformal fixed point, d is holomorphic and nilpotent. It should
be possible to impose these nilpotency and holomorphicity conditions perturbatively in
by computing contributions of the quantum worldsheet variables and the FradkinTseytlin
term to the equations of motion and OPEs of and d .
The necessity of imposing BRST nilpotency and holomorphicity can be seen at the
linearized level by analyzing the superstring vertex operators of (17), (18) and (53). When
the superfields in these vertex operators are on-shell to linearized order, one can check that
the vertex operators have no poles with the stress tensor T and therefore preserve quantum
conformal invariance. However, the condition of having no poles with T is weaker than
BRST invariance (i.e., [Qflat , V ] = 0) and does not imply the complete set of linearized
on-shell conditions. Note that Q = Qflat + V to linearized level, so [Qflat , V ] = 0 implies
that Q is nilpotent to linearized order.
Besides the ChernSimons modifications to the three-form field strength, the first
superstring corrections to the supergravity equations of motion are expected to come at
order ( )3 , e.g., from the R 4 term. Since the supergravity equations of motion are implied
by classical nilpotency and holomorphicity of the BRST operator, one expects to see these
( )3 corrections to the equations at three loops in the nilpotency and holomorphicity
conditions. However, already at first order in , there are several non-trivial one-loop
102
contributions to the nilpotency and holomorphicity conditions which must be cancelled by

contributions from the FradkinTseytlin term and from the ChernSimons modification to
the three-form field strength.
w ) AP I JI )
For example, for the heterotic superstring, the term EP (P (
in Eq. (36) gets one-loop corrections from the chiral anomalies
appearing in d
1
N,
JI = [M AIN] Z M Z
2

1
N + 1 r ,
w = [M N] Z M Z
2
8
(131)
(132)
where r is the worldsheet curvature and the coefficient 18 in (132) can be obtained by
computing the coefficient of the triple pole of w with the pure spinor stress tensor T
gets a one-loop contribution
and dividing by four.5 So d

1
N.
2 rEP P(s) + EP P [M N] AIP [M AIN] Z M Z
(133)
2
After including other one-loop contributions coming from contractions of the quantum
worldsheet variables, one expects that the second term in (133) is completed to
1 P (CS)
M N
2 E wP MN Z Z , where

2
2
(CS)
wP MN = 3 Tr [P M N] + [P M N] A[P M AN] A[P AM AN] (134)
3
3
is the ChernSimons three-form constructed from the gauge, scale and Lorentz connections.
The first term in (133) is cancelled by the contribution from the FradkinTseytlin term

1
1
d 2 z r(Z),

2
2
. So the r contribution to d
is cancelled if the
which contributes 12 rD to d
(s)
heterotic dilaton superfield is related to the scale connection P by
D = 4EP P(s),
(135)
which can be checked to imply that the metric is in string gauge.

of (36) also contains the term 1 EP HP MN Z M Z
N , the second term in (133)
Since d
2
can be cancelled by redefining
HP MN HP MN wP(CS)
MN .
(136)
As in the RNS sigma model [35], the need for redefining HP MN can also be seen by
requiring gauge invariance of the sigma model action. Because of (131) and (132), the
action of (24) is invariant under local gauge, scale and Lorentz transformations only if
5 The triple pole of w with T can be computed using the formulas of [34] where w = 2 h is
1 N N ab 1 (h)2 2 2 h is the pure spinor stress tensor, and h(y)h(z)

the ghost-number current, T = 10
ab
2
log(y z).
BMN is defined to transform as

(s)
ab
ab ,
BMN = [M AIN] I [M N] (s) [M N]
103
(137)
where I , (s) and ab are the gauge parameters.

Similarly, for the Type II superstring, the anomalies

1
N + 1 r ,
w = [M N] Z M Z
2
8
1
1
N + r
Z M Z
w = [M
N]
2
8
(138)
imply from Eq. (65) that the Type II dilaton superfield is related to the scale connections
(s)
(s) by
P and
P
(s)
D = 4EP P ,
.
D = 4EP
P
(s)
(139)
One can check for the Type IIB superstring that (139) agrees with the relation found in
Eq. (121), which confirms that the metric is in string gauge. Furthermore, the terms in
(138) suggest that one should redefine the Type II three-form field strength as

HP MN HP MN wP(CS)
(140)
P(CS)
NM w
NM ,
where w(CS) is a ChernSimons three-form constructed from the unhatted spin connections
P(s) and Pab , and w (CS) is a ChernSimons three-form constructed from the hatted spin
(s) and
ab . However, since the differences of the vector components of the
connections
P
P
(s)
c(s) and cab
cab , are expected to vanish on-shell, the vector
spin connections, c
components of the three-form, Habc , are not expected to be affected by (140).
It would be interesting to verify that these and other one-loop corrections to the BRST
nilpotency and holomorphicity conditions are cancelled by the FradkinTseytlin term and
the ChernSimons modifications to the three-form. It would also be interesting to verify
that the sigma model actions of (24) and (56) are indeed conformally invariant at the
quantum level when the background fields are on-shell.
Acknowledgements
N.B. would like to thank CNPq grant 300256/94-9, Pronex grant 66.2002/1998-9,
FAPESP grant 99/12763-0, and the Clay Mathematics Institute for partial financial support.
The research of P.S.H. was supported in part by PPARC SPG grant 613.
References
[1] N. Berkovits, S. Gukov, B.C. Vallilo, Superstrings in 2D backgrounds with RR flux and new extremal black
holes, Nucl. Phys. B 614 (2001) 195, hep-th/0107140.
104
[2] N. Berkovits, Covariant quantization of the GreenSchwarz superstring in a CalabiYau background, Nucl.
Phys. B 431 (1994) 258, hep-th/9404162;
N. Berkovits, W. Siegel, Superspace effective actions for 4D compactifications of heterotic and Type II
superstrings, Nucl. Phys. B 462 (1996) 213, hep-th/9510106;
J. de Boer, K. Skenderis, Covariant computation of the low energy effective action of the heterotic
superstring, Nucl. Phys. B 481 (1996) 129, hep-th/9608078.
[3] N. Berkovits, C. Vafa, N = 4 topological strings, Nucl. Phys. B 433 (1995) 123, hep-th/9407190;
N. Berkovits, C. Vafa, E. Witten, Conformal field theory of AdS background with RamondRamond flux,
JHEP 03 (1999) 018, hep-th/9902098;
N. Berkovits, Quantization of the Type II superstring in a curved six-dimensional background, Nucl. Phys.
B 565 (2000) 333, hep-th/9908041.
[4] P.S. Howe, H. Nicolai, A. Van Proeyen, Auxiliary fields and a superspace Lagrangian for linearised tendimensional supergravity, Phys. Lett. B 112 (1982) 446.
[5] B.E.W. Nilsson, A. Tollsten, Supersymmetrization of the R 4 term in superstring theories, Phys. Lett. B 181
(1986) 63.
[6] C. Vafa, E. Witten, A one-loop test of string duality, Nucl. Phys. B 447 (1995) 261, hep-th/9505053.
[7] A. Strominger, Loop corrections to the universal hypermultiplet, Phys. Lett. B 421 (1998) 139, hepth/9706195.
[8] M.B. Green, M. Gutperle, Effects of D-instantons, Nucl. Phys. B 498 (1997) 195, hep-th/9701093;
M.B. Green, M. Gutperle, P. Vanhove, One loop in eleven dimensions, Phys. Lett. B 409 (1997) 177, hepth/9706175.
[9] A. Antoniadis, S. Ferrara, R. Minasian, K.S. Narain, R 4 couplings in M- and type II theories on CalabiYau
spaces, Nucl. Phys. B 507 (1997) 571, hep-th/9707013.
[10] J.G. Russo, A.A. Tseytlin, One loop four graviton amplitude in eleven-dimensional supergravity, Nucl. Phys.
B 508 (1997) 245, hep-th/9707134.
[11] N. Berkovits, C. Vafa, Type IIB R 4 H (4g4) conjectures, Nucl. Phys. B 533 (1998) 181, hep-th/9803145.
[12] M.B. Green, S. Sethi, Supersymmetry constraints on type IIB supergravity, Phys. Rev. D 59 (1999) 046006,
hep-th/9808061.
[13] M. Grisaru, A. van de Ven, D. Zanon, Four-loop beta functions for the N = 1 and N = 2 supersymmetric
non-linear sigma model in two dimensions, Phys. Lett. B 173 (1986) 423.
[14] E. Bergshoeff, M. de Roo, The quartic effective action of the heterotic string and supersymmtery, Nucl.
Phys. B 328 (1989) 439.
[15] M. de Roo, H. Suelmann, A. Wiedemann, The supersymmetric effective action of the heterotic string in ten
dimensions, Nucl. Phys. B 405 (1993) 326, hep-th/9210099.
[16] L. Bonora, M. Bregola, K. Lechner, P. Pasti, M. Tonin, Anomaly-free supergravity and super YangMills
theory in ten dimensions, Nucl. Phys. B 296 (1988) 877;
L. Bonora, M. Bregola, K. Lechner, P. Pasti, M. Tonin, A discussion of the constraints in N = 1 SUGRASYM in ten dimensions, Int. J. Mod. Phys. A 5 (1990) 461.
[17] S. Bellucci, S.J. Gates Jr., D = 10 N = 1 superspace supergravity and the Lorentz ChernSimons form,
Phys. Lett. B 208 (1988) 456.
[18] K. Peeters, P. Van Hove, A. Westerberg, Supersymmetric higher derivative actions in ten and eleven
dimensions, the associated superalgebras and their formulation in superspace, Class. Quantum Grav. 18
(2001) 843, hep-th/0010167.
[19] M. Cederwall, U. Gran, M. Nielsen, B.E.W. Nilsson, Manifestly supersymmetric M-theory, JHEP 10 (2000)
041, hep-th/0007035.
[20] S.J. Gates Jr., H. Nishino, Deliberations on 11D superspace for the M-theory effective action, Phys. Lett.
B 508 (2001) 155, hep-th/0101037.
[21] E. Witten, Twistor-like transform in ten dimensions, Nucl. Phys. B 266 (1986) 245.
[22] J.J. Shapiro, C. Taylor, Supergravity torsion constraints from the 10D superparticle, Phys. Lett. B 181 (1986)
67.
[23] I.L. Chau, B. Milewski, Light-like integrability and supergravity equations of motion in D = 10 N = 1
superspace, Phys. Lett. B 216 (1989) 330.
[24] M. Grisaru, P.S. Howe, L. Mezincescu, B.E.W. Nilsson, P.K. Townsend, N = 2 superstrings in a supergravity
background, Phys. Lett. B 162 (1985) 116.
105
[25] E. Bergshoeff, P.S. Howe, C.N. Pope, E. Sezgin, E. Sokatchev, Ten-dimensional supergravity from light-like
integrability in loop superspace, Nucl. Phys. B 354 (1991) 113.
[26] E. Bergshoeff, F. Delduc, E. Sokatchev, Lightlike integrability in loop superspace, KacMoody central
charges and ChernSimons terms, Phys. Lett. B 262 (1991) 444.
[27] P.S. Howe, Pure spinor lines in superspace and ten-dimensional supersymmetric theories, Phys. Lett. B 258
(1991) 141;
P.S. Howe, Pure spinors, function superspaces and supergravity theories in ten and eleven dimensions, Phys.
Lett. B 273 (1991) 90.
[28] N. Berkovits, Covariant quantization of the superparticle using pure spinors, JHEP 09 (2001) 016, hepth/0105050.
[29] N. Berkovits, Super-Poincar covariant quantization of the superstring, JHEP 04 (2000) 018, hepth/0001035;
N. Berkovits, Covariant quantization of the superstring, Int. J. Mod. Phys. A 16 (2001) 801, hep-th/0008145.
[30] P.S. Howe, P.C. West, The complete N = 2 D = 10 supergravity, Nucl. Phys. B 238 (1984) 181.
[31] P.S. Howe, Weyl superspace, Phys. Lett. B 415 (1997) 149, hep-th/9707184.
[32] W. Siegel, Classical superstring mechanics, Nucl. Phys. B263 (1986) 93.
[33] J.L. Carr, S.J. Gates Jr., R.N. Oerter, D = 10 N = 2A supergravity in superspace, Phys. Lett. B 189 (1987)
68.
[34] N. Berkovits, Relating the RNS and pure spinor formalisms for the superstring, JHEP 08 (2001) 026, hepth/0104247.
[35] C.M. Hull, E. Witten, Supersymmetric sigma models and the heterotic string, Phys. Lett. B 160 (1985) 398.

Penrose limit of N = 1 gauge theories

Jaume Gomis, Hirosi Ooguri
California Institute of Technology 452-48, Pasadena, CA 91125, USA
Received 19 March 2002; received in revised form 3 May 2002; accepted 7 May 2002
Abstract
We find a Penrose limit of AdS5 T 1,1 which gives the pp-wave geometry identical to the
one that appears in the Penrose limit of AdS5 S 5 . This leads us to conjecture that there is a
subsector of the corresponding N = 1 gauge theory which has enhanced N = 4 supersymmetry. We
identify operators in the N = 1 gauge theory with stringy excitations in the pp-wave geometry and
discuss how the gauge theory operators fall into N = 4 supersymmetry multiplets. We find similar
enhancement of symmetry in some other models, but there are also examples in which there is no
supersymmetry enhancement in the Penrose limit. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
The AdS/CFT correspondence [13] relates a conformal field theory in p +1dimensions
to string theory in AdSp+2 X, where X is a compact Einstein space. In the last few years,
we have learned much about nonperturbative aspects of string theory and conformal field
theories using this correspondence [4]. One of the major obstructions in making further
progress in this direction has been our lack of understanding of the worldsheet dynamics
describing string theory in AdS and related backgrounds. Understanding this problem is
essential, for example, to finding quantitative results from string theory for the large N
limit of gauge theories with finite t Hooft coupling. Although much progress has been
made for string theory in AdS3 with an NSNS background [5], worldsheet dynamics in
higher-dimensional AdS spaces and/or with RR field strengths remains a mystery. Thus
far, most of the results obtained from string theory in AdS have relied on the supergravity
approximation.
E-mail address: ooguri@theory.caltech.edu (H. Ooguri).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 9 6 - 6
J. Gomis, H. Ooguri / Nuclear Physics B 635 (2002) 106126
107
Recently, it was shown in [6] that the worldsheet theory of the Type IIB string on the
maximally supersymmetric pp-wave geometry [7]:
8

i i

ds = 4dx dx +
dr dr r i r i dx + dx + ,
2
(1.1)
i=1
with constant RR 5-form flux,

F+1234 = F+5678 = const,
(1.2)
is exactly soluble in the light-cone GreenSchwarz formalism [8,9]. Moreover, in a recent

interesting paper [10], it was pointed out that Type IIB string theory on this pp-wave
background is dual to the large N limit of a certain subsector of four-dimensional N = 4
SU(N) supersymmetric gauge theory. The subsector is characterized by choosing a U (1)R
subgroup of the SU(4)R R-symmetry of the gauge theory and by
considering states with
conformal weight and U (1)R charge R which scale as , R N and whose difference
( R) is finite in the large N limit. The claim is that, in the N limit with the string
2 finite, the subspace of the gauge theory Hilbert space and the operator
coupling gs = gYM
algebra preserving these conditions are described by string theory in the pp-wave geometry.
This duality was derived in [10] by starting with the familiar correspondence between
N = 4 SU(N) supersymmetric gauge theory and Type IIB string theory in AdS5 S 5 and
considering a scaling limit of the geometry near a null geodesic in AdS5 S 5 carrying
large angular momentum with respect to the U (1)R isometry of S 5 . This corresponds to
truncating to the appropriate subsector of the gauge theory in the scaling limit. The string
theory background that one obtains in the scaling limit is the pp-wave geometry with a
constant RR flux which can then be quantized in the light-cone gauge [6]. The scaling limit
is a special example of the Penrose limit which transforms any solution of supergravity to
a plane wave geometry [1114].
In [10], it was shown that operators in the appropriate subsector of N = 4 SU(N)
gauge theory can be identified with stringy oscillators in the pp-wave background. This
matching makes quantitative predictions about the spectrum of the gauge theory beyond
the supergravity approximation, and some of them were checked in [10] using gauge theory
computation of planar Feynman diagrams.
In this paper we consider a similar duality that exists between a certain four-dimensional
N = 1 gauge theory and Type IIB string theory in a pp-wave background. We derive this
duality by taking a scaling limit of the duality [15] between Type IIB string theory on
AdS5 T 1,1 and the four-dimensional superconformal field theory which consists of an
N = 1 SU(N) SU(N) super-YangMills multiplet with a pair of bifundamental chiral
) and (N
, N) representation of the gauge
multiplets Ai and Bi transforming in the (N, N
group. The gauge theory is flown to the IR fixed point and deformed by an SU(2)1 SU(2)2
invariant superpotential
W=
ij i
j
Tr Ai Bi
Aj Bj
,
2
(1.3)
108
which is exactly marginal at the fixed point. This gives the theory of [15] that lives1 on
N D3-branes sitting at the conifold singularity of a CalabiYau three-fold. The scaling
limit is obtained by considering the geometry near a null geodesic carrying large angular
momentum in the U (1)R isometry of the T 1,1 space which is dual to the U (1)R Rsymmetry in the N = 1 superconformal field theory.
The scaling limit around this null geodesic in AdS5 T 1,1 results in a pp-wave
geometry. We identify the light-cone Hamiltonian, longitudinal momentum and angular
momentum of string theory in the pp-wave geometry with linear combinations of the
conformal weight , U (1)R charge R and U (1)1 U (1)2 global charges Q1 and Q2 of
operators in the dual N = 1 gauge theory. The appropriate scaling limit requires truncating
the gauge theory Hilbert space to those operators whose conformal weight , R-charge R
and global charges Q1 and Q2 scaling like
, R, Q1 , Q2 N,
(1.4)
with
3
R,
2
1
Q1 R,
2
1
and Q2 R : finite,
2
(1.5)
in the large N limit.

Remarkably, the pp-wave geometry that one obtains in the scaling limit can be
transformed into the maximally supersymmetric background of (1.1) and (1.2) after a
suitable change of coordinates. Therefore, in this limit supersymmetry is enhanced. This
implies that the subsector of the Hilbert space of the N = 1 gauge theory of [15] obeying
the conditions (1.4) and (1.5) has a hidden N = 4 supersymmetry. We believe that this is a
very interesting prediction of our duality that deserves further study.
We find that the change of coordinates induces twisting of the light-cone Hamiltonian
of the string theory by
p = pS5 + J1 + J2 ,
(1.6)
where pS5 is the Hamiltonian of the maximally supersymmetric wave found in [6] and J1 ,
J2 correspond to rotation charges under an R2 R2 subspace of the transverse space of
the pp-wave geometry. From the gauge theory point of view, the light-cone Hamiltonian
p before the twisting is 32 R and the rotational charges are given by Ja = Qa 12 R
(a = 1, 2). Note that they remain finite in the limit (1.5). After the twisting, the light-cone
Hamiltonian is identified with N =4 RN =4 , in terms of the conformal weight and the
R-charge for the N = 4 supersymmetry algebra. Thus we find the following relation
3
N =4 RN =4 = R J1 J2
2
1
= R Q1 Q2 .
2
1 The details of the gauge theory will appear in Section 3.
(1.7)
109
The spectrum of stringy excitations in the N = 4 theory studied in [10] can then be turned
into that of the N = 1 theory by this twisting. The twisted string spectrum is highly
degenerate, and we show that it matches with gauge theory expectations.
The Penrose limit focuses on geometry near a null geodesic. When we have a gauge
theory whose supersymmetry is reduced by placing branes on a curved space, the Penrose
limit may flatten out the space and restore supersymmetry. We have found other examples
where similar enhancement of symmetry takes place. Those include the N = 1 pure superYangMills theory (with KaluzaKlein tower of fields) studied in [16] and gauge theories
realized on branes on a C 3 /Z3 orbifold singularity. The limit of the former is a variation
of the NappiWitten geometry [17] with 16 supercharges, and that of the latter is the
maximally supersymmetric pp-wave. On the other hand, there are cases in which such
enhancement does not happen, such as gauge theories on branes at a C 2 /Z2 orbifold
singularity.
The rest of the paper is organized as follows. In Section 2, we consider the scaling limit
around a null geodesic in AdS5 T 1,1 and show that one obtains a pp-wave background.
We identify the subsector of the Hilbert space of the dual superconformal field theory
that is dual to string theory in the pp-wave geometry. We show that there is a coordinate
transformation which brings the pp-wave background to the one which has maximal
supersymmetry (1.1). The Hamiltonian of the pp-wave is then obtained by twisting the
Hamiltonian of the Type IIB string in that background by angular momentum charges
J1 and J2 that the strings carry in the maximally supersymmetric pp-wave background.
In Section 3, we describe ingredients of the N = 1 theory of [15] that are required to
compare the gauge theory and the string theory. Precise matching is obtained by identifying
in a specific way string theory excitations with gauge theory operators. In Section 4, we
discuss other examples in which similar enhancement of symmetry takes place and show
an example where symmetry enhancement does not occur. We conclude with a discussion.
In the appendix we explicitly solve for the worldsheet theory of the pp-wave background
that we obtain in the limit. Explicit diagonalization of the Hamiltonian shows that it is
related to that of the maximally supersymmetric wave by twisting.
2. Penrose limit of AdS5 T 1,1

We start by considering the supergravity solution dual to the N = 1 superconformal
field theory of [15] that we describe in the next section. The background of interest is
AdS5 T 1,1 , where T 1,1 = (SU(2) SU(2))/U (1), with the U (1) diagonally embedded
in the two SU(2)s. The Einstein metric on AdS5 T 1,1 is given by

2
dsAds
= L2 dt 2 cosh2 + d 2 + sinh2 d3 ,

2
2 1
dsT 1,1 = L
(d + cos 1 d1 + cos 2 d2 )2
9
+

1 2
d1 + sin2 1 d12 + d22 + sin2 2 d22 ,
6
(2.1)
110
where d3 is the volume form of a unit S 3 and the curvature radius L of AdS5 is given
by L4 = 4gs N
2 27/16. Topologically, T 1,1 is a U (1) bundle over S2 S2 . The base
is parametrized by coordinates (1 , 1 ) and (2 , 2 ), respectively, and the Hopf fiber
coordinate has period 4 . The SU(2)1 SU(2)2 U (1)R isometry group of T 1,1 is
identified with the SU (2)1 SU (2)2 global symmetry and U (1)R symmetry of the dual
superconformal field theory of [15]. In addition, the solution has a constant dilaton and a
RR five-form flux
F = L4 (volAdS + volT 1,1 ),
(2.2)
T 1,1 .
where volAdS , volT 1,1 are the volume forms of AdS5 and
We now perform a scaling limit around a null geodesic in AdS5 T 1,1 which rotates
along the coordinate of T 1,1 , whose shift symmetry corresponds to the U (1)R symmetry
of the dual superconformal field theory.2 We introduce coordinates which label the
geodesic

1
1
1
L2
t + ( + 1 + 2 ) ,
t ( + 1 + 2 )
x+ =
(2.3)
x =
2
3
2
3
and consider a scaling limit around = 1 = 2 = 0 in the geometry (2.1). We take L
while rescaling the coordinates
6
6
r
1 =
1 ,
2 .
2 =
= ,
(2.4)
L
L
L
The metric one obtains in the limit is
+
ds = 4dx dx +
2
4

dr i dr i r i r i dx + dx +
i=1
da2
+ a2 da2 2a2 da dx +
a=1,2
= 4dx + dx +
+

4

dr i dr i r i r i dx + dx +
i=1

dza d z a + i(za dza za d z a ) dx + .
(2.5)
a=1,2
In the last line, we introduced complex Cartesian coordinates za in lieu of (a , a ). The

metric has a covariantly constant null Killing vector /x so that it is a pp-wave metric.
The pp-wave has a natural decomposition of the R8 transverse space into R4 R2 R2 ,
where R4 is parametrized by r i and R2 R2 by za . The geometry is supported by a null,
covariantly constant flux of the RR field,
F+1234 = F+5678 = const.
(2.6)
2 Shifts along the angles and generate an U (1) U (1) subgroup of the SU(2) SU(2) isometries
1
2
1
2
and correspond in the gauge theory side to the abelian charges Q1 and Q2 , which are the Cartan generators of
the SU(2)1 SU(2)2 global symmetry group of the gauge theory.
111
The obvious symmetries of this background are the SO(4) rotations in R4 and a U (1)
U (1) symmetry3 rotating R2 R2 . In the gauge theory side, the SO(4) symmetry
corresponds to the subgroup of the SO(2, 4) conformal symmetry (i.e., the rotations of
S 3 in the field theory space R S 3 ) and the U (1) U (1) rotation charges J1 and J2 with
the U (1) U (1) symmetry Q1 12 R and Q2 12 R, respectively, where R is the U (1)R
charge of the gauge theory and Q1 , Q2 are the Cartan generators of the SU(2)1 SU(2)2
global symmetry of the dual superconformal field theory.
In order to compare string theory in the pp-wave geometry (2.5) with the appropriate
subsector of the dual field theory determined by the limit, (2.3) and (2.4), we need to
establish the correspondence between conserved charges in string theory and in gauge
theory. In string theory, the light-cone momenta can be identified with combinations of the
conformal weight and the U (1)R charge R of the dual superconformal field theory by
noting that4
3
2p = ix + = i(t + 3 ) = R,
2

i
1
3
+
2p = ix = 2 (t 3 ) = 2 + R .
L
L
2
The J1 and J2 rotation charges of the string can be identified with

1
Ja = i
a = 1, 2,
= i
+i
= Qa R,
a x
a t,
t,i
2
(2.7)
(2.8)
such that the states of the dual gauge theory are also labeled by these global symmetry
charges.
Therefore, it follows from the identification (2.7) and the L limit, (2.3) and
(2.4), that string theory in the pp-wave background (2.5) with finite p , p+ and Ji
is dual to the N
= 1 gauge theory of [15] in a subsector of the Hilbert space where
, R, Qa L2 N with finite ( 32 R) and Qa 12 R in the large N limit. In particular
the duality with string theory predicts that there is a set of non-chiral primary operators,
which satisfy > 32 R, whose dimension and R-charge grow without bound but such that
the deviation from the BPS bound is finite in the large N limit. In the next section we
will make a proposal for which operators of the gauge theory obey this peculiar scaling
behavior.
Remarkably, the pp-wave geometry (2.5) that we have obtained in the scaling limit
reduces to the maximally supersymmetric pp-wave solution (1.1) after performing a
coordinate dependent U (1) U (1) rotation in the R2 R2 plane as
+
za = eix wa ,
z a = eix w a .
(2.9)
This means that the symmetry and supersymmetry of the original AdS5 T 1,1 background
is maximally enhanced in the Penrose limit, (2.3) and (2.4). We interpret this as saying
3 In fact, the metric and the flux is invariant under a larger symmetry as we shall see below.
4 The factor of two in the normalization of R is due to the 4 periodicity of the coordinate.
112
that the corresponding subsector of the dual N = 1 superconformal field theory has hidden
N = 4 supersymmetry.
The coordinate transformation (2.9) mapping the solution (2.5) to the maximally
supersymmetric solution (1.1) allows us to write down the string Hamiltonian p in (2.7)
in terms of the Hamiltonian pS5 of the maximally symmetric solution already computed in
[6]. Using the coordinate transformation (2.9) and the relation (2.7) between the isometry
and the gauge theory charges, we find that
3
R = 2p
2

=i +
x za

wa
=i + +
w a
x
wa
w a
wa
= 2pS5
+ J1 + J2 ,
(2.10)
where J1 and J2 are the U (1) rotation charges around an R2 R2 subspace of the
maximally supersymmetric pp-wave R8 transverse space,5 and
2pS5 = N =4 RN =4 .
(2.11)
We now briefly recall the string spectrum pS5 found in [6]. The spectrum consists of a
set of eight bosonic harmonic oscillators ani and eight fermionic harmonic oscillators Sn
with i, = 1, 2, . . . , 8. The increase in light-cone energy due to one oscillator is given by

2

n
2pS 5 = 1 +
(2.12)
,

p+
so that in particular the zero modes a0i and S0a increase pS5 by one. To obtain the spectrum
of p we use the twisting formula (2.10). In order to find the spectrum we need to know
the charges of ani and Sn under the U (1) U (1) subgroup of the SO(8) rotation group.
The charges of the bosonic oscillators follow from decomposing the bosonic oscillators,
which transform under the 8v representation of the SO(8) rotation group under SU(2)
SU(2) U (1) U (1), such that the R8 space on which SO(8) acts splits as R4 R2 R2 ,
and U (1) U (1) rotates R2 R2 . We organize the oscillators as
ani
J1 = J2 = 0,
wn1
w n1
wn2
w n2
i = 1, 2, 3, 4,
J1 = 1,
J2 = 0,
J1 = 1,
J2 = 0,
J1 = 0,
J2 = 1,
J1 = 0,
J2 = 1.
5 The U (1) U (1) rotation charges when changing from z to w coordinates remain identical.
(2.13)
Therefore, the contribution to p of each of the bosonic oscillators is

2
n
ani 2p = 1 +
,

p+

2

n
1
+ 1,
wn 2p = 1 +

p+

2

n
1
w n 2p = 1 +
1,

p+

2
n
wn2 2p = 1 +
+ 1,

p+

2

n
2
1.
w n 2p = 1 +

p+
113
(2.14)
The fermionic oscillator contribution to p follows by looking at the U (1) U (1) charges
carried by the SO(8) spinor 8s under SU(2) SU(2) U (1) U (1). The oscillators split
as
8s (2, 1)(1/2,1/2) (2, 1)(1/2,1/2) (1, 2)(1/2,1/2) (1, 2)(1/2,1/2),
(2.15)
where the charges in the subscript correspond to (J1 , J2 ) charges. Therefore, their
contribution to the light-cone Hamiltonian (2.10) is

2

n
++
2p = 1 +
+ 1,
Sn

p+

2
n
Sn
2p = 1 +
1,

p+

2

n
2p = 1 +
,
Sn+

p+

2
n
+
Sn
(2.16)
2p = 1 +
.

p+
Thus we see that the spectrum of bosonic oscillators and fermionic oscillators are identical.
We note that the light-cone energy is not increased by the action of the bosonic zero
mode oscillators w 01 and w 02 nor by the action of their supersymmetric fermionic zero mode
partners S0 . Therefore, the system has an infinitely degenerate spectrum labeled by the
number of times the vacuum state is acted on by the zero modes. In the original coordinates
of the pp-wave in (2.5), the degeneracy of the spectrum can be easily understood by
considering the zero-mode sector (i.e., the point particle limit) of string theory. In the
zero-mode sector, the Hamiltonian contains, on top of four free harmonic oscillators,
two decoupled Landau Hamiltonians describing an electron in a magnetic field in the
114
planes z1 and z2 . The degeneracy in these coordinates corresponds to the well-known

Landau level degeneracy of states of the electron where the degeneracy is labeled by the
angular momentum of the electron. In the appendix, we extend this to stringy excitations,
quantizing the bosonic string Hamiltonian in these coordinates. We show that indeed the
spectrum is twisted by (J1 + J2 ) as in (2.14) with respect to the maximally supersymmetric
case of [6].
In the next section we give a precise prescription of how to realize the infinite
degeneracy of states in string theory, which corresponds in the dual superconformal field
theory to having an infinite degeneracy of operators with a given conformal dimension, and
identify the operators dual to the insertion of the string theory oscillators.
3. Gauge theory spectrum

Type IIB superstring theory in AdS5 T 1,1 is dual to the N = 1 gauge theory with
gauge group SU(N) SU(N) with two chiral multiplets Ai (i = 1, 2) transforming in
) representation of the gauge group and two chiral multiplets Bi
(i
= 1, 2) in
the (N, N

the (N, N) representation. The theory is flown to the IR fixed point. We then turn on the
superpotential (1.3) involving the chiral superfields Ai and Bi
. At the fixed point, these
chiral superfields have conformal weight 3/4 and R-charge 1/2. They transform as (2, 1)
and (1, 2) under the SU (2)1 SU (2)2 global symmetry.
Now we are ready to compare states in the string theory with operators in the gauge
theory. We will mainly focus only on the bosonic excitations of the theory, denoted by
wna and w na in (2.14). Let us begin with the zero mode sector of string theory, which is
generated by w 01 and w 02 . Since we are dealing with the zero mode of the string, they are
supergravity modes in AdS5 T 1,1 . The correspondence between supergravity modes and
gauge theory operators has been discussed extensively in [21,22]. It is useful to rephrase
it in the stringy terminology of the last section. We will find some special feature of the
supergravity spectrum in the Penrose limit. Since the light-cone Hamiltonian 2p is equal
to ( 32 R), the lowest energy states of the string theory are chiral primary states. The
basic ones are of the form Tr(AB)R . Among them, we can identity the ground state |0 of
the w 0a oscillators with the gauge theory operator,

|0 Tr (A1 B1 )R .
(3.1)
Here and in the following, we ignore normalization factors in gauge theory operators which
may depend on N and R. We have chosen Ai and Bi
so that A1 and A2 carries Q1 charge
+ 12 and 12 and B1 and B2 carries Q2 charge + 12 and 12 , respectively. Their (J1 , J2 )
charges defined by Ja = Qa 12 R are, therefore, ( 14 , 14 ) and ( 34 , 14 ) for A1 and A2 ,
and ( 14 , 14 ) and ( 14 , 34 ) for B1 and B2 . Thus the operator Tr[(A1 B1 )R ] in (3.1) carries
J1 = J2 = 0, and 32 R = N =4 RN =4 = 0. Namely, it saturates the BPS bounds for
both N = 1 and N = 4 supersymmetry algebras.
The operators w 0a then act on it as
w 01 : A1 A2 ,
w 02 : B1 B2 .
(3.2)
115
In order for these to map the chiral primary state (3.1) into another chiral primary state,
their action has to be symmetrized along all A1 s and B1 s in the trace [15]. Since the
w 0a oscillators are absent in the worldsheet Hamiltonian, their action does not increase
2p = 32 R. This is consistent with the fact that the action of the w 0a s in the
gauge theory gives rise to chiral primary states saturating the BPS bound of the N = 1
supersymmetry. On the other hand, N =4 RN =4 = 32 R J1 J2 is increased by 1
every time we act with w 0 since J1 + J2 = 0 for A1 and B1 while it is 1 for A2 and B2 .
Note that (J1 +J2 ) is not positive for any of the operators. Therefore, these operators satisfy
the BPS bounds for both the N = 1 and the N = 4 supersymmetry algebras. This gives an
important consistency check of our conjecture about the supersymmetry enhancement.
From the way they act on Ai and Bi
, it is clear that w 01 , w 02 and their conjugates are
identified as the raising and lowering operators of the SU(2)1 SU(2)2 global symmetry
of the gauge theory. On the string worldsheet, they act as harmonic oscillators. On the other
hand, when acting on the gauge theory operators Tr[(A1 B1 )R ], they obey the constraints
(w 01 )R+1 = 0, (w 02 )R+1 = 0. This does not contradict with the correspondence between
string theory and gauge theory. Since Ja = Qa 12 R (a = 1, 2) has to remain finite in the
limit R , only a finite number of w 0 s can act on this operator and these constraints
become irrelevant.
The oscillators w0a s in (2.14) are more interesting. They change ( 32 R) by 2, thus
their action does not generate chiral primary states. Nevertheless the resulting states should
be in the supergravity sector. Candidates for such states can be found in the list of operators
given in [22] , where they are called semi-conserved superfields. Although they are not
chiral primaries, their conformal dimensions are protected. The ones we are interested in
here take the following form,

V n1 eV Be
V B n2 (AB)R ,
Tr AeV Ae
(3.3)
where V is the vector multiplet for the gauge group SU(N) SU(N). There is one
important subtlety in making the identification. It was pointed out in [21] that, in order for
the corresponding supergravity mode to have a rational conformal dimension, the integers
n1 and n2 must satisfy the Diophantine equation,
n21 + n22 4n1 n2 n1 n2 = 0.
(3.4)
This is true if
we are studying states with finite and R. Since we are studying the scaling
limit , R N , it is worth revisiting its origin. The constraint comes from the fact
that the eigenvalue E of the Laplacian on T 1,1 for the corresponding supergravity mode
takes the form

3
3
E = 6n21 + 6n22 + 8n1 n2 + (6R + 8)(n1 + n2 ) + R R + 4 .
(3.5)
2
2
One can then show that the conformal weight = 2 + 4 + E of the mode becomes
rational if (3.4) is satisfied. However, this condition is relaxed in the limit R . In this
limit, the meaningful quantity is ( 32 R), and it is given by

1
3
.
R = 2n1 + 2n2 + O
(3.6)
2
R
116
The right-hand side of this equation is clearly rational. In fact, they are even integers.
Therefore, the Diophantine constraint (3.4) is irrelevant in the subsector of the Hilbert
space we are looking. Moreover, the formula is exactly what we need to identify the action
of w0a . Since each of these operators is supposed to increase 2p = 32 R by 2 and they
carry the U (1) U (1) quantum numbers (1, 0) and (0, 1), the natural identification is
w01 : insertion of A1 eV A 2 eV ,
2 eV B1 .
w02 : insertion of eV B
(3.7)
Of course the insertion must be symmetrized along the trace so that ( 32 R) is minimized.
As we discussed in the last section, string theory in AdS5 T 1,1 acquires enhanced
N = 4 superconformal symmetry in the Penrose limit. This means that the spectrum of
the gauge theory operators in this subsector must fall into N = 4 multiplets. Since the
oscillators w0a and w 0a are part of the N = 4 superconformal generators (P I and J +I in
the notation of [6], see also [23]), we conjecture that the chiral primary fields of the form
Tr(AB)R and the semi-conserved multiplets of the form (3.3) combine to make N = 4
multiplets in the limit.6 Note that this can happen only in the limit. For finite R, the semiconserved multiplets have to obey the Diophantine constraint (3.4) in order for them to
have rational conformal weights.
Now let us turn to the stringy excitations. The string-bit interpretation suggests that a
worldsheet operator with non-zero Fourier mode n acts on a gauge theory operator just as
its n = 0 counter-part, but the action is summed over the trace with a position dependent
phase proportional to n. This point of view was adopted in [10] to identify operators in the
N = 4 gauge theory corresponding to stringy excitations. We extend their proposal to the
N = 1 theory we are studying.
Consider the oscillators w na . When n = 0, it is defined as the replacement A1 A2 for
a = 1 and B1 B2 for a = 2, averaged over the trace. It is then natural to identify
w n1 |0
R1
2 ink

Tr (A1 B1 )k A2 B1 (A1 B1 )R1k e R ,
k=0
w n2 |0
R1
2 ink

Tr (A1 B1 )k A1 B2 (A1 B1 )R1k e R .
(3.8)
k=0
Of course, the operators on the right-hand side vanish due to the cyclicity of the trace.
This corresponds to the string theory fact that the left-hand side does not satisfy the level
matching momentum constraint, as explained in
[10]. The
idea is to use an analogous

2 w
2 |0 such that
n
+
mi
= 0. For example,
definition for w n11 w n1s w m
i
mt
1
1
w n
w n2 |0
R1
2 ink

Tr A2 (B1 A1 )k B2 (A1 B1 )R1k e R .
(3.9)
k=0
6 The decomposition of N = 4 chiral multiplets into N = 1 multiplets has been discussed in [24], whose
results should be useful in proving this conjecture. There, the R-symmetry acting on N = 1 multiplets is chosen
to be the commutant of SU(3) in the SU(4) R-symmetry of the N = 4 theory. On the other hand, the R-symmetry
of the N = 1 theory of [15] is the commutant of SU(2) SU(2) in the SU(4) R-symmetry of the enhanced N = 4
supersymmetry.
117
Operators of this type are not chiral primaries. As pointed out in [15], these operators
vanish if we use the constraint due to the superpotential (1.3),
A1 Bi
A2 = A2 Bi
A1 ,
B1 Ai B2 = B2 Ai B1 .
(3.10)
Since they are defined against the potential wall, they gain additional conformal weights
beyond their naive values, and therefore 32 R > 0. The string theory computation in
Section 3 predicts that the action of w n changes ( 32 R) by

3
R =
2
n
=
n
1+
p+

1 + 3gs n2
2
1
N
1.
R2
(3.11)
We are taking the large N limit so that N/R 2 remains finite. When gs > 0, the righthand side is, indeed, strictly positive. One of the interesting features of this formula is that
( 32 R)n vanishes at gs = 0 even for n = 0, giving rise to further degeneracy of the
spectrum. In the large N limit, the only parameter of the string theory is gs while that of
the gauge theory is , the coefficient of the superpotential. Since the superpotential must
increase the conformal weights of these operators, one explanation of what happens at
gs = 0 is that vanishes in the gauge theory side. To our knowledge, a map between these
parameters has not been worked out, and this observation may give us some hint about the
correspondence between the gauge theory moduli and the string theory moduli. At gs = 0,
the string theory Hilbert space becomes the Fock space of first quantized strings, whose
spectrum is integral. It would be interesting to see whether such a structure emerges in this
subsector of the gauge theory at = 0.
The other set of operators wna are similarly interpreted in the gauge theory side. These
V and eV Be
V B in the trace and sum over insertion points along
operators insert AeV Ae
the trace with position dependent phases. The string theory computation predicts that the
amount of change of ( 32 R) is given by

2

3
n
R = 1+
+1
2

p+
n

N
= 1 + 3gs n2 2 + 1.
(3.12)
R
One can also consider operators corresponding to stringy excitations in the r i directions.
In [10], these are interpreted as taking derivative of operators with respect to spatial
coordinates in the gauge theory. In our case, one may be puzzled by the fact that there
seem to be two types of derivatives, those acting on As and those acting on Bs. Clearly
only particular combinations of them correspond to stringy excitations of the r i directions.
In general, there are many gauge invariant observables one can write down, and only some
of them correspond to string states. We expect that the others become infinitely heavy, i.e.,
( 32 R) becomes infinitely large, in the large N limit.
118
4. Other examples
The Penrose limit focuses on the geometry near a null geodesic. When we have a gauge
theory whose supersymmetry is reduced by placing branes on a curved space, the Penrose
limit may flatten out the space near the branes and restore supersymmetry. Thus we expect
that enhancement of symmetry takes place in a large class of theories which have gravity
duals.
Let us consider the supergravity solution found in [16] which is dual to pure N = 1
supersymmetric YangMills theory (with KaluzaKlein tower of fields). We will show that
it has a Penrose limit which is identical to that of a collection of five-branes in flat space.
Thus, in this case, symmetry is again enhanced in the corresponding subsector of N = 1
super-YangMills. To exhibit this we write the metric of NSNS five-branes in flat space

1
1
1
ds 2 = ds 2 R 1,5 + L2 d 2 + (d + cos d)2 + d 2 + sin2 d 2 , (4.1)
4
4
4
where L2 =
N . We introduce the following null coordinates

1
1
1 t
L2 t
+
x =
+ ( + ) ,
x =
( + )
2 L 2
2 L 2
(4.2)
and consider the limit around = = 0. We take the limit L while rescaling the
coordinates
2y
r
=
.
= ,
(4.3)
L
L
The metric that one obtains in this limit is

ds 2 = 4 dx + dx 2y 2 d dx + + ds 2 R8 ,
(4.4)
which by using the coordinate transformation in (2.9) reduces to the pp-wave metric

ds 2 = 4 dx + dx ww dx + dx + + dw d w + ds 2 R6 ,
(4.5)
where w is a complex coordinate on an R2 plane in R8 . Moreover, in the limit (4.4) the
dilaton becomes constant and the NSNS three-form flux H+12 = const becomes null.
The worldsheet theory describing the coordinates (x + , x , w, w)
can be identified with
the WZW model based on the non-semi-simple group which is the central extension of the
two-dimensional Poincare group, found by Nappi and Witten [17].
One can also show that the Penrose limit of the MaldacenaNuez solution [16] in the
region near = 0 gives rise to a generalization of the NappiWitten geometry with 16
supercharges and again the supersymmetry is enhanced.7 It would be interesting to explore
the consequences of this symmetry enhancement for the gauge theory.
Another interesting example is to consider the Penrose limit of the dual pair obtained
by placing N D3-branes at a C 3 /Z3 orbifold singularity. The gauge theory living on the
7 In the earlier version of this paper, we claimed that the Penrose limit of the MaldacenaNuez solution is
identical to the NappiWitten geometry (4.5). We thank Juan Maldacena and Horatiu Nastase for pointing out an
error in our argument. Our observation about the enhancement of supersymmetry in this case still remains true.
119
D3-branes is an N = 1 four-dimensional quiver gauge theory [25] and the gravity dual is
AdS5 S 5 /Z3 [26] , where the S 5 is described by the complex coordinates zi (i = 1, 2, 3)
constrained by
|z1 |2 + |z2 |2 + |z3 |2 = 1
(4.6)
and the Z3 generator g acts by

g zi = zi ,
i = 1, 2, 3,
3 = 1,
(4.7)
so that Z3 acts freely on the sphere. The U (1)R symmetry of the gauge theory can be
identified with shifts along the Hopf fiber coordinate when S 5 /Z3 is described as a
U (1) bundle over CP 2
2

ds 2 = d + sin2 3 + d2 + sin2 12 + 22 + cos2 32 ,
(4.8)
where i are a set of left-invariant SU(2) one forms satisfying di = ij k j k . The Z3
orbifold group acts by restricting the range of the coordinate to 1/3 of the usual value
for S 5 while leaving all other coordinates intact. We introduce null coordinates
1
x + = (t + ),
2
x =
L2
(t )
2
(4.9)
taking the L near = = 0 with8

r
y
,
= .
L
L
In the limit one gets
=
ds 2 = 4 dx + dx +
(4.10)
4

dr i dr i r i r i dx + dx + + dy i dy i + 2y 2 dx + 3 , (4.11)
i=1
where yi are Cartesian coordinates in R4 . By introducing a pair of complex coordinates za

for R4 one can show that (4.11) can be rewritten as
ds 2 = 4 dx + dx +
4

i i

dr dr r i r i dx + dx + +
dza d z a
i=1

+ i(za d z a z a dza ) dx ,
+
(4.12)
which we have already shown is the same as the maximally supersymmetric pp-wave
metric (1.1). So we have another example in which a subsector of a four-dimensional
N = 1 gauge theory is enhanced to N = 4. It would be interesting to match the string
oscillators in this background with operators in the quiver gauge theory.
We should note, however, that the Penrose limit does not always enhance supersymmetry. For example, we can consider the dual pair generated by placing a collection of
D3-branes at a C 2 /Z2 orbifold singularity. The gauge theory is a four-dimensional N = 2
8 The coordinate comes the AdS part of the metric.
5
120
gauge theory [25] and the gravity dual is AdS5 S 5 /Z2 [26], where the S 5 is described by
(4.6) and the Z2 generator g acts by
g z1 = z1 ,
g z2 = z2 .
(4.13)
described by |z3 = 1 at z1 = z2 = 0. The

Thus the Z2 action has as fixed locus an
coordinate parametrizing the fixed S 1 can be identified with the coordinate considered
in the Penrose limit of AdS5 S 5 [10]. Therefore, the corresponding pp-wave limit is
given by the Z2 orbifold of the maximally supersymmetric pp-wave (1.1), where Z2 acts
on an R4 subspace of the transverse R8 space. The main difference between this and the
previous orbifold is that is not acted by Z2 while the other angles on the S 5 have Z2
identifications. This is related to the fact that the S 5 has a fixed circle in this case.
To find the amount of supersymmetry left unbroken by the orbifold, one must find which
components of the Killing spinors of the pp-wave geometry (1.1) are left-invariant under
the Z2 action. The Killing spinors of (1.1) were found [7] and take the form

x, y, x + = f x, y, x + 0 ,
(4.14)
S1
|2
where x and y are the transverse R4 R4 coordinates, f is a function which can be found
in [7] and 0 is a constant SO(8) spinor. The amount of unbroken supersymmetry is the
number of Killing spinors (4.14) left-invariant under the Z2 action. Therefore, the unbroken
supersymmetries satisfy

x, y, x + = g x, y, x +

= 5678 x, y, x +

= f x, y, x + 56780 ,
(4.15)
namely,
56780 = 0 .
(4.16)
Therefore the orbifold preserves 1/2 of the supersymmetry which is generated by Killing
spinors satisfying this condition.
5. Discussion
In this paper we have given an explicit example of an N = 1 superconformal field
theory which, in the large N limit, has a subsector of the Hilbert space with enhanced
N = 4 superconformal symmetry. We have arrived at this perhaps unexpected conclusion
by taking the corresponding limit in the string theory side and by showing that it becomes
identical to the theory with higher supersymmetry. The subsector of the gauge theory that
should exhibit this symmetry enhancement is dictated by the Penrose limit which restricts
the space of states of the gauge theory to those whose conformal dimension and R-charge
diverge in the large N limit but which nevertheless have finite ( 32 R).
The light-cone Hamiltonian for the background that one obtains in the limit can be
found from the Hamiltonian of [6] for the maximally supersymmetric case by twisting it
121
with U (1) charges. In this way we get a prediction for the spectrum of ( 32 R) of the
N = 1 superconformal field theory. We proposed how stringy excitations are related to
gauge theory operators and made predictions about the gauge theory spectrum. Perhaps
the most striking one is that the various N = 1 multiplets should turn into multiplets
of N = 4 supersymmetry. In particular, the chiral multiplets and the semi-conserved
multiplets of N = 1 supersymmetry should combine into N = 4 chiral multiplets. It
would be interesting to explore this prediction further. The duality also predicts values
of ( 32 R) for certain operators in this strongly coupled gauge theory.
We have shown that the enhancement of symmetry in the Penrose limit is a fairly generic
phenomenon in theories which have gravity duals. One can intuitively understand this as
due to the fact that the limit flattens out parts of spacetime by focusing on a region near a
null geodesic. For example, we found that the Penrose limit of a collection of flat NSNS
five-branes is the NappiWitten geometry. The limit of the MaldacenaNuez geometry is
its variation and also has 16 supercharges. This is an interesting case since we can quantize
the worldsheet theory without taking the light-cone gauge, and we can compute correlation
functions and other observables using the standard techniques of conformal field theory
[27]. On the other hand, we also found cases in which the Penrose limit does not lead to
symmetry enhancement. It would be interesting to explore further the consequences of this
enhancement of symmetries for QCD-like theories and see which lessons this might teach
us for the familiar questions about strongly coupled dynamics of these gauge theories.
In this paper, we have focused on the particular class of geodesics along the circle fiber
of T 1,1 . It would be interesting to classify null geodesics in AdS5 T 1,1 and to see if
there is any other interesting limit in the gauge theory which has an exact string theory
description.
Note added
After posting this paper on the e-Print arXiv, we have received [18,19], where the
Penrose limit of AdS5 T 1,1 is studied and the supersymmetry enhancement is also noted.
We also received [20], where the Penrose limit of backgrounds with NSNS 3 form field
and its relation to a generalization of the NappiWitten model are also discussed.
Acknowledgements
We would like to thank Andreas Brandhuber, Sergio Ferrara, Juan Maldacena, Sunil
Mukhi, Horatiu Nastase, and John Schwarz for useful discussion. This research was
supported in part by DOE grant DE-FG03-92-ER40701 and by Caltech Discovery Fund.
Appendix A. String spectrum in the original coordinates

It may be instructive to show how we can quantize string theory using the metric (2.5)
before we make the coordinate transformation to its manifestly symmetric form (1.1). We
122
can read off the bosonic part of the light-cone action from the metric (2.5) as
1
S=
2
+
2
p

a=1,2

1 2
ri ri
2
2
4
i=1

1 2
x + ya2 xa
2 ya
2 xa ya + ya xa
2 a

,
where = and
= .
The spectrum of the ri part of the light-cone Hamiltonian is

2

n
(r)
Hr -part =
Nn
1+
,

p+
n=
(A.1)
(A.2)
as in the case of the N = 4 theory in [10], where n is the label of the Fourier mode around
the direction and Nn(r) is the number of excitations of that mode.
Let us examine the (xa , ya ) part of the Hamiltonian. In the following, we will ignore
the index a with the assumption that we are referring to either a = 1 or 2. Accordingly the
manifest global symmetry is U (1) SU(2). In the Fourier expansion,
1
i n
x=
xn e
p + ,
+
p n
1
i n
y=
yn e
p + ,
+
p n
(A.3)
the Hamiltonian becomes

Hxy -part =
Hn ,
(A.4)
n=0
where
H en=0 = (pxn yn )(pxn yn ) + (pyn + xn )(pyn + xn )
2

n
+
(xn xn + yn yn ),

p+
(A.5)
and
1
1
H0 = (px0 y0 )2 + (py0 + x0 )2 ,
2
2
(A.6)
where
pxn = i
,
xn
pyn = i
.
yn
(A.7)
To compare with the gauge theory spectrum, it is useful to use the complex coordinates,
zn = xn + iyn and z n = xn iyn , so that the U (1) part of the global SU(2) symmetry is
123
manifest. The Hamiltonians for the Fourier modes then become

i
i
i
i
Hn=0 = pz n + zn pzn z n + pz n zn pzn + z n
2
2
2
2
2

1
n
+
zn z n + zn z n ,
+
2 p
and
(A.8)

i
i
H0 = pz 0 + z0 pz0 z 0 .
2
2
(A.9)
Let us diagonalize them.

(a) Zero mode
The Hamiltonian H0 for the zero mode is nothing but the one for a charged particle
in two dimensions in a constant magnetic field. It has the Landau spectrum with infinite
degeneracy at each level. To compare with the gauge theory spectrum, it is useful to
introduce the following set of oscillators,
1
a0 = z 0 +
2
1
b0 = z0 +
2
,
z0
,
z 0
a0 = z0
,
2
z 0
1
b0 = z 0
.
2
z0
They obey the commutation relations

a0 , a0 = 1,
b0 , b0 = 1,
(others) = 0.
(A.10)
(A.11)
Under the U (1) subgroup of the SU(2) global symmetry, the operators a0 , b0 carry charge
1 and a0 , b0 carry charge +1
In terms of the oscillators, the Hamiltonian (A.9) is expressed as
H0 = 2a0a0 + 1.
(A.12)
In particular, it commutes with the oscillators b0 , b0 . The lowest energy states are
annihilated by a0 and thus have the form (b0 )k exp( 12 z0 z 0 ) where k is any non-negative
integer. The complete energy eigenstates are given by

m k
1
(0)
m,k
(A.13)
= a0 b0 exp z0 z 0 ,
2
with m, k 0. The energy of the state is 2m and the U (1) global charge is (k m).
(b) Non-zero modes
As in the case of the zero mode, we introduce the following set of oscillators,
1
,
an = z n +
2
zn
1
bn = zn +
,
2
z n
1
an = zn
2
1
bn = z n
2
,
z n
,
zn
(A.14)
124
obeying the commutation relations,

bn , bn = 1,
an , an = 1,
(others) = 0.
(A.15)
In the limit n2 /
p+ 0, the Hamiltonian Hn takes the same form as in the case of the
zero mode,
Hn = 2anan + 2an
an + 2.
(A.16)
In particular, each energy level has infinite degeneracy generated by the oscillators bn
.
For finite n2 /
p+ , the Hamiltonian Hn contains terms mixing the two oscillators,

1

+ (n n) + 2,
Hn = 2anan + 2 an + bn an + bn
2
(A.17)
where we introduced = n/
p+ to simply the following equations. The Hamiltonian can
be diagonalized by introducing a new set of oscillators n and n defined by
an = cosh n + sinh n
,
an = cosh n + sinh n ,
,
bn = cosh n + sinh n
bn = cosh n + sinh n .
They obey the commutation relations

n , n = 1,
n , n = 1,
(others) = 0.
(A.18)
(A.19)
The vacuum state for , is related to the one for a, b by the Bogolubov transformation.
Substituting these into (A.17) and requiring that the cross terms of and to vanish, we
find
e4 = 1 + 2 .
(A.20)
In terms of the new set of oscillators, the Hamiltonian Hn is then expressed as

Hn = 1 + 2 + 1 n n + 1 + 2 1 n n + (n n) + 2 1 + 2 .
(A.21)
Here we have chosen a solution so that, in the limit 0, the oscillators become n an
m
k k
and n bn . Energy eigenstates of Hn are then given by (n )m (n

) (n ) (n ) |0,
where |0 is the vacuum state for the , oscillators. The state carries the energy,

Em,m
,k,k
= (m + m
) 1 + 2 + 1 + (k + k
) 1 + 2 1 ,
and the global U (1) charge (k + k
m m
).
Combining this with the result for the zero mode, the complete spectrum for the (x, y)
part of the light-cone Hamiltonian is given by

2

n
()
Nn
Hxy -part =
1+
+1

p+
n=

2

n
()
+ Nn
(A.22)
1+
1 .

p+
125
This result agrees with the one we obtained in Section 2, with the identification:
n = w n ,
n = wn
(A.23)
in (2.14).
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys.
Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[5] J. Maldacena, H. Ooguri, Strings in AdS3 and SL(2, R) WZW model. Part 1: the spectrum, J. Math. Phys. 42
(2001) 2929, hep-th/0001053;
J. Maldacena, H. Ooguri, J. Son, Strings in AdS3 and the SL(2, R) WZW model. Part 2: euclidean black
hole, J. Math. Phys. 42 (2001) 2961, hep-th/0005183;
J. Maldacena, H. Ooguri, Strings in AdS3 and the SL(2, R) WZW model. Part 3: correlation functions,
hep-th/0111180, and references therein.
[6] R.R. Metsaev, Type IIB GreenSchwarz superstring in plane wave RamondRamond background, hepth/0112044.
[7] M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2002) 047, hep-th/0110242.
[8] M.B. Green, J.H. Schwarz, Covariant description of superstrings, Phys. Lett. B 136 (1984) 367.
[9] M.B. Green, J.H. Schwarz, Properties of the covariant formulation of superstring theories, Nucl. Phys. B 243
(1984) 285.
[10] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp-waves from N = 4 super-YangMills,
hep-th/0202021.
[11] R. Penrose, Any spacetime has a plane wave as a limit, in: Differential Geometry and Relativity, Reidel,
Dordrecht, 1976, pp. 271275.
[12] R. Gven, Plane wave limits and T -duality, Phys. Lett. B 482 (2000) 255, hep-th/0005061.
[13] M. Blau, J. Figueroa-OFarrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry, hepth/0201081.
[14] M. Blau, J. Figueroa-OFarrill, G. Papadopoulos, Penrose limits, supergravity and brane dynamics, hepth/0202111.
[15] I.R. Klebanov, E. Witten, Superconformal field theory on three-branes at a CalabiYau singularity, Nucl.
Phys. B 536 (1998) 199, hep-th/9807080.
[16] J.M. Maldacena, C. Nuez, Towards the large N limit of pure N = 1 super-YangMills, Phys. Rev. Lett. 86
(2001) 588, hep-th/0008001.
[17] C.R. Nappi, E. Witten, A WZW model based on a nonsemisimple group, Phys. Rev. Lett. 71 (1993) 3751,
hep-th/9310112.
[18] N. Itzhaki, I.R. Klebanov, S. Mukhi, PP-wave limit and enhanced supersymmetry in gauge theories, hepth/0202153.
[19] L.A. Zayas, J. Sonnenschein, On Penrose limits and gauge theories, hep-th/0202186.
[20] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NSNS and RR plane wave
backgrounds, hep-th/0202179.
[21] S.S. Gubser, Einstein manifolds and conformal field theories, Phys. Rev. D 59 (1999) 025006, hepth/9807164.
[22] A. Ceresole, G. DallAgata, R. DAuria, S. Ferrara, Spectrum of type IIB supergravity on AdS5 T 11 :
predictions on N = 1 SCFTs, Phys. Rev. D 61 (2000) 066001, hep-th/9905226.
126
[23] R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in RamondRamond plane wave
background, hep-th/0202109.
[24] S. Ferrara, M.A. Lledo, A. Zaffaroni, BornInfeld corrections to D3 brane action in AdS5 S 5 and N = 4,
d = 4 primary superfields, Phys. Rev. D 58 (1998) 105029, hep-th/9805082.
[25] M.R. Douglas, G.W. Moore, D-branes, quivers, and ALE instantons, hep-th/9603167.
[26] S. Kachru, E. Silverstein, 4d conformal theories and strings on orbifolds, Phys. Rev. Lett. 80 (1998) 4855,
hep-th/9802183.
[27] J. Gomis, H. Ooguri, E.Witten, unpublished.

Bulk and brane radiative effects in gauge theories

on orbifolds
G. von Gersdorff, N. Irges, M. Quirs
Instituto de Estructura de la Materia (CSIC), Serrano 123, E-28006 Madrid, Spain
Received 1 May 2002; accepted 8 May 2002
Abstract
We have computed one-loop bulk and brane mass renormalization effects in a five-dimensional
gauge theory compactified on the M4 S 1 /Z2 orbifold, where an arbitrary gauge group G is broken
by the orbifold action to its subgroup H. The spacetime components of the gauge boson zero modes
along the H generators span the gauge theory on the orbifold fixed point branes while the zero modes
of the higher-dimensional components of the gauge bosons along the G/H generators play the role of
Higgs fields with respect to the gauge group H. No quadratic divergences in the mass renormalization
of the gauge and Higgs fields are found either in the bulk or on the branes. All brane effects for the
Higgs field masses vanish (only wave function renormalization effects survive) while bulk effects
are finite and can trigger, depending on the fermionic content of the theory, spontaneous Hosotani
breaking of the brane gauge group H. For the gauge fields we do find logarithmic divergences
corresponding to mass renormalization of their heavy KaluzaKlein modes. Two-loop brane effects
for Higgs field masses are expected from wave function renormalization brane effects inserted into
finite bulk mass corrections. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.10Kk; 04.50+h; 11.25.Mj
1. Introduction
Extra dimensions (with respect to the four spacetime dimensions) are a common
ingredient in all fundamental theories aiming to unify gravity with the rest of the known
interactions. However, unlike gravitational interactions that propagate in the bulk of
the higher-dimensional space (with ten/eleven dimensions in string/M-theory), gauge
Work supported in part by CICYT, Spain, under contract AEN98-0816, and by EU under contracts HPRNCT-2000-00152 and HPRN-CT-2000-00148.
E-mail address: mariano@makoki.iem.csic.es (M. Quirs).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 9 5 - 4
128
G. von Gersdorff et al. / Nuclear Physics B 635 (2002) 127157
interactions can propagate on a (4 + n)-dimensional (n 1) I /I strings [1]. Moreover,

it has been shown that in these theories the radius R of the compact dimensions where
the gauge interactions propagate can be large enough [2] for the corresponding excitations
to be at the reach of future accelerators [311], while the string (or higher-dimensional
quantum gravity) scale Ms can be lowered to the TeV range [12] and show up in
colliders [13,14] and gravitational experiments [15]. This fact opened up for the first time
exciting possible experimental accessibility to fundamental theories and provided new
insight into long-standing problems of particle physics such as the hierarchy problem.
The hierarchy problem of the Standard Model, i.e., the appearance of quadratic
divergences in the quantum corrections to the Higgs mass is one of the most outstanding
problems in particle physics. It has motivated, as the prototype perturbative solution, the
introduction of supersymmetrya symmetry responsible for the absence or cancellation
of quadratic divergencesthat is being looked for extensively in experimental searches.
However, in view of the elusiveness of supersymmetry to show up in direct searches and the
robustness of the Standard Model predictions that is pushing up the scale of supersymmetry
breaking, it is interesting to explore new avenues and possible alternative solutions to the
hierarchy problem. The existence of TeV extra dimensions where the Standard Model fields
propagate provides new and useful tools for this search.
Large extra dimensions have shown to shed new light on the hierarchy problem. In
particular, in higher-dimensional supersymmetric theories it has been proved that one-loop
radiative corrections to the Higgs mass are finite (ultraviolet insensitive) and 1/R [16
25]. Of course, since the theory is non-renormalizable, higher-loop effects introduce
through wave function renormalization a certain ultraviolet sensitivity to the Higgs mass
that can be absorbed by the renormalization group running of the coupling constants and
hence it does not make this sensitivity explicit at low energy [26]. In this sense, the higher
dimensionality of the theory allows to improve the solution to the hierarchy problem with
respect to four-dimensional supersymmetry.
However, in the presence of large compact dimensions supersymmetry is not as
necessary an ingredient as it is in four dimensions. In fact, non-supersymmetric solutions
to the hierarchy problem based on toroidal compactifications were already explored in
the literature [27,28]. In those cases the Standard Model Higgs field should be identified
with an extra-dimensional component of a gauge field and electroweak symmetry breaking
proceeds by the Hosotani mechanism [29,30]. The higher-dimensional gauge invariance
protects the Higgs mass from quadratic divergences at the quantum level and radiative
corrections to the Higgs mass are finite and 1/R. In short, in these non-supersymmetric
models the role of supersymmetry preventing quadratic divergences is played by higherdimensional gauge invariance.
A word of caution should be said here about the proposed (perturbative) solutions
to the hierarchy problem. All of them are based upon introducing a symmetry at an
intermediate scale M0 between the electroweak scale Mweak and the Standard Model
cutoff (quantum gravity or string scale Ms ) such that quadratic divergences are canceled
at scales > M0 . In this way quadratic divergences survive only for scales smaller
than M0 and radiative corrections to the Higgs mass are M0 . For the case of fourdimensional supersymmetry, M0 = MSUSY , and at scales > MSUSY the Higgs mass is
protected by supersymmetry. For the case of a higher-dimensional non-supersymmetric
129
theory with electroweak Hosotani breaking, M0 = 1/R. For scales < 1/R the theory is
four-dimensional and it is not protected from quadratic divergences, while for > 1/R
the theory is higher-dimensional and the gauge invariance protects the Higgs mass from
quadratic divergences. In both cases, for the mechanism to be effective, the scale M0 has
to be stabilized and should be not much higher than the electroweak scale. This comment
applies to both the scale of supersymmetry breaking MSUSY and to the compactification
scale 1/R. In particular, fixing the radius R implies considering the gravitational sector of
the theory involving the radion field. This problem is outside the scope of the present paper
and we will assume that the radius has been fixed and stabilized by some mechanism [31].
In the construction of higher-dimensional theories the nature of the compact space plays
a prominent role in physics. In particular, theories with more than four flat dimensions or
higher-dimensional theories compactified on tori are non-chiral from the four-dimensional
point of view. The simplest solution to this problem1 is compactification on tori modded
out by a discrete symmetry group acting non-freely (with fixed points) on the compact
space, or orbifold compactifications [33,34]. Orbifolds are not smooth manifolds but have
singularities at the fixed points which are four-dimensional hypersurfaces or boundaries of
the higher-dimensional space. Those boundaries will be (and are) often named branes
by an abuse of language. At the field theory level brane contributions arise in the higherdimensional Lagrangian by means of Dirac delta functions. Non-supersymmetric models
on orbifold compactifications were already proposed in [35,36].
It has been proved that under radiative corrections a theory with no brane couplings
will generally flow to one with non-trivial physics on the brane [3740]. In particular,
wave function renormalization effects localized on the brane have been found. We could
understand the appearance of those renormalization effects since they are consistent with
the four-dimensional nature and symmetries of the branes. In a realistic model we should
perform electroweak breaking by the Hosotani mechanism in a higher-dimensional gauge
theory broken by the orbifold action to the Standard Model gauge theory on the brane.
The Higgs is a scalar field from the point of view of the brane and its mass is not
a priori protected by the higher-dimensional gauge invariance from acquiring quadratic
divergences localized on the brane. Possible mass terms on the branes that are not protected
by the residual gauge invariance are not obviously protected from the higher-dimensional
symmetries either and will be investigated in detail in this paper. Their absence should be
essential for any phenomenological applications aiming to solve the hierarchy problem in
the absence of supersymmetry.
The plan of this paper is as follows. In Section 2 we will consider a five-dimensional
(5D) theory compactified on the orbifold M4 S 1 /Z2 where an arbitrary gauge group G is
broken to the subgroup H by the orbifold action. We also consider fermions in an arbitrary
representation R of the gauge group and the associated, consistent with the orbifold, Z2
parity action. We fix the gauge consistently with the properties of the 5D theory and
introduce the corresponding FaddeevPopov ghosts with their transformation properties
under the orbifold action. The general Feynman rules for fields propagating in the bulk
1 There are other solutions that have been proposed involving smooth manifolds with non-trivial backgrounds [32].
130
of the orbifold are exhibited explicitly as well as some useful group theoretical formulas
that will be used in the rest of the paper. A general discussion of the allowed orbifold
gauge breaking patterns and associated consistent fermion representations is postponed
to Appendix A while the gauge fixing conditions and the unitary gauge are discussed
in Appendix B. In Section 3 we discuss the general structure of corrections generated
in the bulk and on the orbifold fixed planes by radiative corrections in the bulk. We
will restrict ourselves to radiative corrections to mass terms, i.e., corrections generated
by diagrams with vanishing external four-momentum: p = 0. In Section 4 the one-loop
radiative corrections to all field masses are computed in the theory where an arbitrary
gauge group G is broken to the subgroup H by the orbifold action with the fermions
in an arbitrary representations R of the gauge group. We have considered separately
gauge and fermion sectors and bulk and brane effects. We have found no quadratic
divergences either in the bulk or on the branes. While this effect in the bulk is justified
from the higher-dimensional gauge invariance, its interpretation on the brane for the extradimensional components of gauge fields is less clear, although it might be related to the
higher-dimensional Lorentz and gauge symmetries. On the other hand, squared mass terms
are generated for the extra-dimensional components of the gauge fields opening up the
possibility of spontaneous breaking of the residual H gauge symmetry on the branes. In
particular, the contribution from the gauge (fermion) sector to the squared mass terms
is positive (negative). In Section 5 the conditions for Hosotani breaking are discussed
and shown to depend on the group-theoretical invariants of the gauge group and fermion
representations. Also the possibility of reducing the rank of H by the Hosotani breaking is
briefly discussed. This possibility is essential in model building where one could identify H
with the Standard Model gauge group. Finally some comments about two-loop corrections
and our conclusions are drawn in Section 6.
2. Broken gauge symmetry on the Z2 orbifold

The model we will consider is a gauge theory coupled to matter in five flat space
time dimensions with coordinates x M = (x , x 5 ) and metric signature (+, , , , ).
The 5D geometry is M4 S 1 /Z2 , i.e., the fifth dimension is compactified on a Z2
orbifold, whereas the remaining part is four-dimensional Minkowski space with metric
. We have neglected gravity so the fixed planes having no tension are rigid geometrical
boundaries. Keeping this remark in mind we will refer to these planes as branes. We
denote the radius of the compact circle by R. The gauge group in the bulk is G (we denote
dim(G) dG ) and it is broken to H = H1 H2 (we denote dim(H) dH ) on the
fixed hyperplanes by our choice of the orbifold projection. For matter, we couple Dirac
fermions R that transform in the representation R (we denote dim(R) dR ) of G to
the gauge fields.2 We will use capital letters from the beginning of the Latin alphabet to
denote gauge indices (A, B, C, . . .), capital letters from the middle of the Latin alphabet
to denote five-dimensional Lorentz indices (M, N, R, . . .), small letters from the middle of
2 There should be no confusion between the representation R and the radius R.
131
the Greek alphabet to denote four-dimensional Lorentz indices (, , , . . .), and small
letters from the middle of the Latin alphabet to denote the discrete fifth-dimensional
momentum (k, l, m, . . .).
Our starting point is the action

1
5
MN
M
+ i R DM R ,
S5 = d x Tr FMN F
(2.1)
2
where
A
FMN = FMN
T A,
A
A
ABC B C
FMN
= M AA
AM AN
N N AM + gf
with the indices A, B, C running over the adjoint representation of the gauge group and
f ABC the corresponding structure constants. The gauge covariant derivative is DM =
A
A
3
M igAA
M TR , where TR are matrices corresponding to the representation R of the
gauge group satisfying
A B
TR , TR = if ABC TRC .
(2.2)
Our parity assignment is defined by

5
M AB B
AA
AM x , x 5
M x , x =
(no sum over M),

R x , x 5 = R i 5 R x , x 5 ,
(2.3)
(2.4)
where and R (i 5 ) represent the Z2 action on the gauge bosons and the fermions,
respectively (R acts on the representation indices), 5 = diag(i, i) and = +1,
5 = 1. In addition, R is a hermitian matrix that squares to one and therefore unitary.
Consistency of the 5D gauge symmetry with the orbifold action requires the condition [41]
f ABC = AA BB CC f A B C ,
(2.5)
where summation over repeated indices is understood. The above constraint comes from
A M AB F B (no sum over M), so that
the requirement that under the Z2 action FMN
MN
A
AMN
FMN F
is invariant and it is straightforward to check that it is an automorphism of the
Lie algebra of G.
On the other hand, the invariance of the fermion kinetic term requires the transformation
P M P = M M
(no sum)
(2.6)
while the invariance of the fermion-gauge boson coupling implies in addition that
P TRA P = AB TRB .
(2.7)
In five dimensions the only solution to these equations is P = R (i5 ) (with R

satisfying Eq. (2.7)), which corresponds to the lower sign in Eq. (2.6). With no loss of

generality we can diagonalize AA = A AA with A = 1, consequently Eq. (2.5) takes
3 When the subscript R is omitted it is implied that the matrices are in the fundamental representation.
132
the simpler form

f ABC = A B C f ABC
(no sum).
(2.8)
We will then often express (2.3) succintly as

5
5
M A A
(no sum).
AA
M x , x = AM x , x
(2.9)
A few remarks are in order. From (2.8) we can see that since the orbifold can break G
completely only if G = U (1), the unbroken subgroup is always (except in the U (1) case)
a non-trivial subgroup of G.4 Under the Z2 action we can naturally split the adjoint index
A into an unbroken part a and a broken part a so that the generators of the subgroup
H are TRa and the generators of the coset K = G/H (we denote dim(K) dK ) are TRa .
Notice that constraint (2.8) simply means that the matrix is a diagonal matrix with dH
elements equal to +1 and with the rest of the dK elements, corresponding to the broken
A
A
part of G, equal to 1. According then to Eq. (2.9), only AA
with = +1 and A5 with
A
= 1 acquire zero modes and they are non-vanishing at the fixed planes. The former
appear as the (massless) four-dimensional gauge fields that correspond to H and the latter
as massless scalar fields in four dimensions. Also (2.4) results in a non-trivial constraint
on the possible bulk fermion representation choices. Let us assume for now that we have
made a consistent choice for R . Then, the fermions that will appear massless in four
dimensions will be chiral because
of the i 5 appearing in (2.4) and they will transform

in some representation R = i ri of H = H1 H2 . Another important constraint
is that the resulting massless chiral spectrum on the fixed hyperplane should be anomaly
free. This puts further restrictions on realistic model building. There are several different
ways to arrive at an anomaly free model but this is not the subject of the present work. In
Appendix A we work out a few simple examples to illustrate issues related to the Z2 action
on the fermions.
We will now use the formalism of [37,38], i.e., we will work with exponential modes
for the fields. The modes for any field are related by
m = P m ,
(2.10)
where P is the parity operator of the field . For fermions it is P = R

as in
M AB with eigenvalues M A as in
(2.4), whereas for the gauge bosons it is PA = M M
(2.9). Eq. (2.10) is automatically satisfied if the fields are expressed by unconstrained ones:

1
m = m + P m .
(2.11)
2
The gauge propagator can therefore be written as
m m
1 (A)

A A = G (p , p5 ) + PA G(A) (p , p5 )PA mm
2

1
+ PA G(A) (p , p5 ) + G(A) (p , p5 )PA m+m .
(2.12)
2
(i 5 )
4 In particular, for models where the bulk gauge symmetry coincides with the gauged superalgebra of a locally
supersymmetric theory, this means that an extended bulk supersymmetry (N > 1) cannot be completely broken
on the fixed planes by the Z2 .
133
In the above, we have denoted by G(A) (p , p5 ) the 5D propagator that corresponds to the
compactification on S 1 , p is the momentum along M4 , p5 = m/R is the momentum in
the compact direction and we have used the notation k k,0 . Using the covariance of
G(A) (p , p5 ) under parity transformations (see (2.19))
PA G(A) (p , p5 )PA = G(A) (p , p5 )
(2.13)
Eq. (2.12) can be simplified to

1
Am Am = G(A) (p , p5 )(mm + PA m+m ).

2
(2.14)
The fermionic propagator can be computed similarly taking into account that P M P =
M M (no sum), which in turn implies the transformation (see (2.21))
P G( ) (p , p5 )P = G( ) (p , p5 ).
(2.15)
Then the simplified fermion propagator reads
m = 1 G( ) (p , p5 )(mm P m+m ).
m
(2.16)
2
The vertices then conserve 5D momentum and are the ones of the unorbifolded 5D theory.
All the information about the non-trivial Z2 action is encoded in the propagators in a
particularly simple way.
The next issue is gauge fixing and ghosts. We will work in the 5D covariant gauge
M AM = 0. The modified Lagrangian then including the gauge fixing term and the ghost
fields cA is in the standard way

L5 L5
1 M B N B
AM AN + Tr M cDM c.
2
(2.17)
By looking at the ghost-gauge field interaction term above, we can see that the ghost cB
has the same parity as the gauge field with the same index B, i.e.,

cA x , x 5 = AB cB x , x 5 .
(2.18)
The propagators and vertices can then be taken over from any standard textbook with the
indices properly generalized to five dimensions. For completeness we give their explicit
forms in 5D Minkowski spacetime [42]:
G(A) (p , p5 ) =
= i
G(c) (p , p5 ) =
G( ) (p , p5 ) =

BC
pM pN
(1
)
g
MN
p2 p5 2
p2 p5 2
=i
(2.19)
BC
,
p2 p5 2
(2.20)
a b
,
p + 5 p5
(2.21)
=i
134

= gf ABC (r q)M gNR + (q p)R gMN + (p r)N gRM
with p + q + r = 0,
(2.22)

= ig 2 f ABE f CDE (gMR gNS gNR gMS )
+ f CBE f ADE (gMR gNS gNM gRS )

+ f DBE f CAE (gSR gNM gNR gMS ) ,
(2.23)
c
= igM T A b ,
(2.24)
= gf ABC qM .
(2.25)
Finally, in our loop computations, we will make use of the following formulas:
f ABC f A BC = C2 (G) AA ,

1

ABC A BC C
f
f
= C2 (HA ) C2 (G) A + 1 AA ,
2
f ABC f A BC B C = C2 (G)A AA ,

tr TRA TRA = CR AA ,

tr TRA R TRA R = CR A AA ,
(2.26)
(2.27)
(2.28)
(2.29)
(2.30)
where C2 (G) is the quadratic Casimir of G and CR is the Dynkin index of the representation
R satisfying dR C2 (R) = CR dG . We normalize the fundamental representation to have
its Dynkin index equal to 1/2. In the second identity we have called HA the unbroken
subgroup which TRA belongs to.5 Note that this makes sense since the whole expression
vanishes if TRA is a broken generator TRA = TRa (i.e., when A = 1).
We will find it convenient to work in the Feynman gauge, = 1. For any other value of
B
, including the 5D Landau gauge = 0, there is a tree-level mixing AA
A5 as can be
M
seen from Eq. (2.19). Furthermore in the class of gauges defined by M A = 0 there is no
5 Since in the subspace of unbroken generators is just the identity one can find a basis in which the groups
Hi are generated by some {T Ai }.
135
B,n
value of for which AA,n
5 A5 = 0, n = 0, which would correspond to the unitary gauge
since the fields AA,n
(n = 0) are the Goldstone bosons eaten by the massive KaluzaKlein
5
gauge bosons. To reach the unitary gauge the gauge fixing condition has to be defined in
a 5D non-covariant fashion, just consistent with the compactified theory. This issue was
discussed in Ref. [43] for an Abelian gauge theory. It is discussed in Appendix B for a
general non-Abelian group broken to a subgroup by the orbifold compactification.
3. The structure of loop corrections

In this section we will analyze what kind of terms in the effective action could be
generated in the bulk and on the orbifold fixed planes by radiative corrections in the bulk.
We will focus on the bilinear terms in the effective action and also look only at terms not
involving derivatives with respect to infinite directions, . The most general terms are
then:

2 C
5
BC
[A] = d x AB
M MN 5 AN

B BC C
1

+ (x5 ) + (x5 R) AM MN i 5 , i 5 AN .
(3.1)
2
The arrows on the 5 indicate the field whose derivative is to be taken. The bulk and brane
terms in Eq. (3.1) can be read off from the generic two-point function

C,m
AB,m
=
M AN
(3.2)
in the following way: the contributions to Eq. (3.2) for which the five momentum in the
outgoing line is conserved, m = m, will give rise to bulk terms. We will denote them as
B,m C,m
+BC 2
BC 2
AM AN bulk = MN
(3.3)
m mm + MN
m m+m .
Contracting this expression with the external fields one finds
B,m

+BC 2
BC 2
m mm + MN
m m+m AC,m
AM MN
N .
(3.4)
We now take the Fourier transform of Eq. (3.1), sum over m :

2 C B,m BC 2

BC
dx5 AB
AM MN m mm AC,m
M MN 5 AN =
N
(3.5)
and read off the correspondence:

2
+BC 2
BC 2
BC
m = MN
m + N C MN
m .
MN
(3.6)
m,m
m,m
In the second term after summing over m there appears a factor of AC,m
which has to
N
C,+m
M
C
be transformed into AN
at the cost of an , according to (2.10). By doing a Taylor
136
expansion around m = 0 we obtain the possible bilinear operators in the bulk:

A m m2 A m
1
1
= Am (0) Am + Am (2) m2 Am + Am (4) m4 Am + .
2
24
(3.7)
The summation over m and the gauge and Lorentz indices are left implicit and (n) is the
nth derivative with respect to m evaluated at m = 0. The bulk mass term is just given by
(0) (= (0)).
The terms in Eq. (3.1) proportional to -functions are generated by contributions in
Eq. (3.2) which do not conserve five momentum. Instead we will find expressions of the
form
B,m C,m
AM AN brane

+BC (m, l)mm 2l +
BC (m, l)m+m 2l .
=
(3.8)
MN
MN
l
Note that the sum over l is now constrained through the Kronecker-, yielding an amplitude
BC (m, m ) we
which depends on the two independent variables m and m . To identify
MN
C,m
contract this matrix with AB,m
and perform the sum over l:
M and AN

+BC m, m m +
BC m, m + m
AB,m
AC,m
N .
MN
MN
M
2
2

(3.9)
m,m ; mm even
To compare this with Eq. (3.1) one takes the Fourier transform of the latter with respect to
the compact dimension:

C

1
BC
dx5 (x5 ) + (x5 R) AB
M MN i 5 , i 5 AN
2

C,m
1

BC
=
dx5 ei(m+m +k)x5 1 + eik AB,m
M MN m, m AN
2
m,m ,k

C,m

BC
=
AB,m
(3.10)
M MN m, m AN .
m,m ; mm even
We conclude that

m m
m + m
BC
BC m,
MN
+BC m,
+
m, m =
MN
MN
2
2
(3.11)
To understand the brane terms a bit better, one can make a Taylor expansion of (3.9) around
m, m = 0. One gets for even (E) and odd (O) fields, respectively:

m, m E m
Em
(2,0) m 1 m (0,2) 2 m
1
(0,0)E m + m2 E m

= Em
E + E
m E
+ ,
2
2

m, m O m
O m
(1,1) m 1
(1,3) 3 m

mO
+ mO m
m O
+ ,
= mO m
(3.12)
6
137
where we have suppressed the gauge and Lorentz indices for clarity and the (independent)
summations over m, m are implicit. In particular, it is now clear that the brane mass term
(0,0) (= (0,
0)), given by setting m = m = 0 in the momentumfor even fields is just
violating terms of Eq. (3.2). In contrast to Eq. (3.7) the expansion coefficients are now not
diagonal but democratic matrices in mode space.
Bulk terms are not expected to generate any divergences that were not present in the
original 5D theory. Brane terms, however, could generate new divergences. In particular,
we want to investigate the possible appearance of a scalar mass on the brane. Had we
introduced fundamental massless scalars in the 5D theory, their masses would pick up
corrections proportional to the cutoff of the five-dimensional theory and the same would
happen with their zero modes on the branes (provided they survive the orbifold projection).
In our model there are no scalars in the five-dimensional theory. The only bosonic fields
are the components of the bulk gauge fields whose masses are zero at tree level and remain
zero at all orders because they are protected by gauge invariance. On the branes, however,
there are scalars. Some of them are massless at tree level, namely the zero modes of
positive Z2 parity fields. The natural question that arises then is whether these masses
are protected against radiative corrections. We know that on general grounds this is not the
case and in order to avoid the Higgs mass to pick up corrections proportional to the cutoff,
it has to be protected by some symmetry, for example supersymmetry. Our model is not
supersymmetric and therefore such a mechanism is not possible. On the other hand since
the scalars are extra-dimensional components of the original gauge field, one would hope
that gauge invariance still protects them. The problem is that it is on the branes where the
zero modes of positive Z2 parity fields are seen as massless scalar fields and we have also
seen that it is on the branes that the only surviving symmetry is H; there is no apparent
symmetry that prohibits dangerous corrections to their masses. In the following we will
see in a one-loop calculation that even though there is no such apparent symmetry on the
branes, the larger, original bulk gauge symmetry arranges the couplings of the zero modes
and the KaluzaKlein (KK) towers in such a way that the masses of the zero modes remain
protected.
4. One-loop corrections
The one-loop corrections in the effective action coming from the exchange of 5D gauge
fields are given in Figs. 1 and 2. We first discuss the general structure of the brane and
bulk terms. As a first step, we will not compute the exact values of the diagrams. It is
enough to observe their general structure to deduce whether and when there appear bulk or
Fig. 1. The diagrams contributing from the gauge sector.
138
Fig. 2. The diagram contributing from the fermion sector.
brane terms. Then, once we have separated bulk from brane terms, we carry out the actual
calculation for each case separately and interpret the results.
Since we are mainly interested in mass corrections we will explicitly evaluate the
one-loop graphs only for vanishing external four momentum (p = 0). The amplitudes
corresponding to each graph will be denoted as in Eqs. (3.3) and (3.8). In addition, we add
a superscript (i), i.e.,

(i) AA 2
(i) AA
m ,
(m, l),
MM
(4.1)
MM
where i = 1 for the tadpole, i = 2 for the gauge loop, i = 3 for the ghost loop and i = 4
for the fermion loop. The other indices are as described in the beginning of Section 2. All
computations in this section will be done in Euclidean space [44].
4.1. The gauge sector
In the sector where 5D gauge bosons are exchanged as internal lines there are the three
different Feynman diagrams shown in Fig. 1. All our computations will be carried out in
the = 1 (Feynmant Hooft) gauge and dimensional regularization (with renormalization
scale ) will be used to handle divergent diagrams. Let us examine each case in turn.
4.1.1. The tadpole
The tadpole is the first graph appearing in Fig. 1. It is proportional to

ml+l m ll + M B l+l BB f ABC f A B C ,
which, using the identities (2.26)(2.30), can be written as

1

AA
M
C2 (HA ) C2 (G) A + 1 mm 2l AA .
C2 (G)mm
+
2
(4.2)
(4.3)
Notice that there is no term proportional to m+m and so no contributions to the two-point
(1) AA
function MM
will appear. The first term in the above is five momentum conserving
and thus gives only a bulk term. Its dependence on the gauge-index A is trivial and the
contribution is the same for all AA
M . The second term is momentum non-conserving and
gives rise to a brane term. Since it contains the factor (A + 1) it is zero if A = 1. Thus
there are only brane terms if the external lines correspond to a field AaM where a is an index
of the unbroken group H.
Applying the Feynman rules, in the bulk we obtain for either A = 1

+
1
d
dd q

(1) AA
MM+ = g 2 MM AA C2 (G)4d
(4.4)
.
d
2
2
(2)
q + l 2 /R 2
l=
139
Here we have defined d = . This expression is obviously the same along M4 or S 1 /Z2 .
The contribution of the tadpole to brane localized terms is non-zero only for A = +1
and is equal to
(1)+AA (m, l)= (d 1 ) g 2 MM AA
MM
4

A

4d
+ 1 2C2 (HA ) C2 (G)
M
1
dd q
. (4.5)
(2)d q 2 + l 2 /R 2
Observe that there is no sum over the loop fifth momentum.

4.1.2. The gauge and the GHOST loop
The second diagram in Fig. 1 is proportional to

mlk l +k m ll + N B l+l

NN BB kk + R C k+k RR CC f ABC f A B C ,
(4.6)
which gives rise to the four terms

C2 (G)mm AA NN RR ,
AA
N R A C2 (G)m+m NN RR ,

1

R A
+ 1 C2 (HA ) C2 (G) m+m 2l AA NN RR ,
2

1

N A
+ 1 C2 (HA ) C2 (G) mm 2l AA NN RR .
2
(4.7)
(4.8)
(4.9)
(4.10)
As in the case of the tadpole, there are bulk terms for both A = 1. However, once again,
one finds brane terms only for A = +1.
The structure for the ghost diagram is slightly simpler since the internal propagators
do not carry vector indices. The conclusion is unmodified, brane terms are proportional to
(A + 1).
The diagrams with momentum conserving external lines corresponding to (4.8) and
(4.9) and the analogous ghost diagrams give, for p = 0,
(i) AA 2
m
MM+

(i)
+
NMM
1
dd q

= g 2 AA C2 (G)4d
, (4.11)
d
2
2
2
2
8
(2)
(q + l /R )(q + (m l)2 /R 2 )
l=

(i) AA 2
(i) AA
MM
(4.12)
m = M A MM+ ,
where here of course i = 2, 3 only. For the gauge loop the numerators are given by

2

1 2
2l + 5m2 2ml
(2)

=
+
2
3
N
(4.13)
q ,
R2
d
(2)
N55 = 2q 2 +
d(m 2l)2
,
R2
(4.14)
140
while the ghost loop gives

(3)
NMM
= 2qM (p q)M .
(4.15)
Evaluated explicitly this becomes

2 2
(3)
N
= q ,
d
2l(m l)
(3)
N55 =
.
R2
(4.16)
(4.17)
(i)AA
are diagonal, in particular there is no mixing
Obviously all contributions to MM

between M = and M = 5. In deriving the above one should not forget that the fifth
component of the momenta entering in the Feynman rules for the vertices may be flipped
due to the l+l , etc., in the propagators.
Brane terms can be computed as well. We will again perform the calculation for p = 0.
We find
(i) AA
(m, l)
MM

1

= g 2 AA A + 1 2C2 (HA ) C2 (G) 4d
16

(i)
N
dd q
MM
,
(2)d (q 2 + l 2 /R 2 )(q 2 + (m l)/R 2 )
(4.18)
where
2
2
(2) = 6l + 3m 6ml 2 (5 3d)q 2 ,
N
R2
d
m(m
2l)
(2)

N
,
55 = d
R2
(3) = 2 q 2 ,
N
d
l(m l)
(3)

.
N
55 = 2
R2
(4.19)
(4.20)
(4.21)
(4.22)
4.1.3. Bulk effects from the gauge sector

In this section we will compute the bulk effects from the gauge sector obtained in
Sections 4.1.1 and 4.1.2. It is generally known from finite temperature field theory that
by compactifying on a circle no new divergences appear. We can give some reasoning on
why this should not happen by looking at a general one-loop amplitude

+
dd q 1
g(q , l/R).
(4.23)
(2)d R
l=
We can perform a Poisson re-summation

1
g(q , l/R) =
g(q
, 2kR),
R
l
(4.24)
141
where g is the Fourier transform of g with respect to q5 = l/R. This allows us to rewrite
the amplitude as

d d q dq5 i(2kR)q5
dq5
dd q
,
q
)
+
g(q , q5 ).
g(q
e
(4.25)
5
(2)d
2
(2)d
2
k=0
Here we have extracted the term in the sum corresponding to k = 0 which is just the fivedimensional amplitude. The remaining terms summed over give typically an exponentially
suppressed function of the four momentum squared.6 We will see explicit examples below.
To collect the total bulk contribution from the gauge sector, we have to add the different
terms according to Eq. (3.6):
3

2

(i)+AA 2
(i)AA 2
AA
MM
MM
m
=
m + M A MM
m .
(4.26)
i=1
Concentrating first on the scalar sector, we find from Eqs. (4.4), (4.11) and (4.12)

2 1 2 AA
dd q
AA
m = g C2 (G)4d
55
8
(2)d

+

4d
2q 2 + (d(m 2l)2 + 2l(m l))/R 2
+
2
.
q 2 + l 2 /R 2
(q 2 + l 2 /R 2 )(q 2 + (m l)2 /R 2 )
l=
(4.27)
Notice that the signs and in Eq. (4.12) exactly cancel the ones in Eq. (4.26) yielding a
global factor of 2 in the second term. Decomposing into partial fractions we obtain

2 1 2 AA
dd q
AA
4d
55 m = g C2 (G)
8
(2)d

+

1
1
+ 2d 2
2d 2
q + l 2 /R 2
q + (m l)2 /R 2
l=

q 2 + (l(m l))/R 2
4(d 1) 2
.
(4.28)
(q + l 2 /R 2 )(q 2 + (m l)2 /R 2 )
The first observation is that one can shift the summation index l in the second term
by m so that it cancels against the first term. To interpret the remaining term, let us
do the integral first. Naive power counting indicates that a quadratic divergence and a
logarithmic divergence will appear in the result. Recall that in dimensional regularization
the appearance of quadratic divergences in d = 4 are signaled by poles in d = 2 and notice
that the usual factor of d 2 multiplying the pole that appears in conventional gauge
theories is missing. Despite this fact we will now show that all divergences (quadratic and
logarithmic) are actually absent. According to Eq. (3.7), we can extract the bulk mass term
6 The separation of the amplitude into a five-dimensional part and a finite part is similar to the approach
of [45].
142
from Eq. (4.28) by evaluating it at m = 0:

1

AA
55
(0) = g 2 AA C2 (G)4d
8

+
q 2 l 2 /R 2
dd q
4(d 1) 2
.
(2)d
(q + l 2 /R 2 )2
l=
(4.29)
There seems to be a quadratically divergent piece left over. This is not unexpected at this
stage since the four-dimensional gauge invariance alone does not protect these scalars from
acquiring divergences. However, five-dimensional gauge-invariance does: according to our
discussion below Eq. (4.23) one can extract the five-dimensional part of the amplitude
by substituting the summation by an integration over q5 = l/R. In this case, however,
this integral turns out to be zero and therefore quadratic divergences are absent. The
remaining terms of the Poisson re-summation (sum over k = 0) give sinh2 (Rq) which
is exponentially suppressed for large q and renders the integration finite. Performing the
integration in (4.29) we find
9

g 2 AA C2 (G)(3).
(4.30)
32 4 R 2
It is a manifestly finite result, as expected, since it is the same result as the one we would
have obtained in an S 1 compactification.
The m-dependent terms and therefore all terms involving 5 derivatives vanish, which
can be seen by writing Eq. (4.28) for m = 0 as

l/R
q 2 + (l(m l))/R 2
(l m)/R
R
.
=
(q 2 + l 2 /R 2 )(q 2 + (m l)2 /R 2 ) m q 2 + l 2 /R 2 q 2 + (m l)2 /R 2
(4.31)
AA
(0) =
55
Shifting the summation index l l + m in the second term cancels the contribution from
the first one.
Let us now compute . As in the case of 55 the contributions of (2,3)+ and
(2,3)
effectively add, yielding a total of
2 1 2 AA
AA
= g C2 (G)4d
m
2

+
1
1 (5m2 2ml)
2
2
2
d d q (1 d)( d (d 2)q + l /R ) + 2
R2
.
d
2
2
2
2
2
2
(2)
(q + l /R )(q + (m l) /R )
l=
(4.32)
Performing the integral first we expect a quadratic divergence by power counting. However,
these divergences again cancel. This can be seen in dimensional regularization by noticing
that the coefficient of the pole is proportional to d 2. Here the factor of d 2 is
necessary because the integral that it multiplies is not zero. The logarithmically divergent
part however is now non-zero for m = 0. It fact it gives a contribution to the mass
renormalization of the heavy modes of the gauge bosons which is expected since the
KK modes being massive should not be protected by gauge invariance against such
divergences. On the other hand for m = 0 this would be a contribution (together with
143
(4.4)) to the mass of the massless H gauge bosons which should be forbidden by the gauge
invariance of the zero mode sector. Indeed, for m = 0 (4.32) reduces to
1 2 AA
AA
g C2 (G)4d
(0) =
2

+
dd q
q2
2
1
(1
d)
(1
d)
,
(2)d
q 2 + l 2 /R 2 d
(q 2 + l 2 /R 2 )2
l=
(4.33)
and one can check that the whole expression is zero in any d.
Alternatively, we could have re-summed first and separated the five from the fourdimensional part from the beginning. The five-dimensional part would then have a pole
at d = 1 (corresponding to a cubic divergence in d = 5). Then, it is the factor of 1 d in
(4.32) that would protect against these divergences, while the remaining integration would
be found to be logarithmically divergent for m = 0 and zero for m = 0.
4.1.4. Brane effects from the gauge sector
We will here gather brane effects from the brane sector results of Sections 4.1.1 and
55 according to Eq. (3.11). The
4.1.2. We have to add the different contributions to
contributions are taken from Eqs. (4.5) and (4.18) together with Eqs. (4.20) and (4.22):

d

AA
55
m, m = g 2 AA 2C2 (HA ) C2 (G) A + 1 4d

8

1
dd q
1
,
+
(4.34)
(2)d q 2 + m2 q 2 + m2+
1
1
where we have defined m = 2R
(m m ) and m+ = 2R
(m + m ). The last expression
seems to indicate a quadratic divergence. However, a closer look shows that this is not the
AA (m, m ) in (4.34) is an even function of m (and of m )
case. To see this, observe that
55
and therefore the sum
a,m

aa
55
m, m Aa,m
A5
= 0,
(4.35)
5
m,m
since only gauge components with A = +1 contribute to brane effects and therefore we
are considering only negative parity scalar fields Aa,m
5 . Thus, here we find a different reason
for the absence of divergences from the one we found in the bulk. There, the poles canceled
between the tadpole, the gauge and the ghost loop, as it happens in a d = 4 gauge theory;
here they are simply zero.
Next let us examine the brane terms for the gauge bosons M = . We find from
Eqs. (4.5) and (4.18) together with Eqs. (4.19) and (4.21)

1
AA m, m = g 2 AA 2C2 (HA ) C2 (G) A + 1 4d
,
16

4

2 + 3 m2 +m 2 + (d 2) (m+m )2
(d
2)(d
3)q
dd q
2
d
R
R2
.
d
2
2
2
2
(2)
(q + m )(q + m+ )
(4.36)
144
The first term is quadratically divergent but the pole cancels in d = 2 as expected. Notice
that here the divergence is quadratic (since there is no sum over l) so the poles should
necessarily cancel in d = 2. The brane mass term according to (3.12) is the zeroth order of
the above in an m, m expansion evaluated at m = m = 0. It is simply
AA (0, 0) = 0
(4.37)
as can be checked by doing the integral explicitly in dimensional regularization. The rest
of the terms in (4.36) correspond to a logarithmically divergent piece and a finite piece.
Recall that this last computation corresponds to the vacuum polarization of the unbroken
H gauge bosons, so naively one would have guessed that not only one should find no
quadratic divergences but moreover the logarithmically divergent and finite parts should
also be absent. What seems even more surprising is that the above amplitude is non-zero
for say m = 0 and m = 0 which implies a mixing between the zero mode and the heavy
modes thus apparently breaking gauge invariance. We believe that the resolution to this
puzzle is similar to the one in the Standard Model where the mixing between the photon
and the Z gauge boson computed in an R gauge is non-zero and gauge non-invariant until
vertex and box contributions are taken into account. We will not pursue this question any
further, since renormalization of these theories is not the main topic of this work.
4.2. The fermion sector
The diagram contributing from fermions living in the representation R is given in Fig. 2
and evaluates to

i
1

ll R i 5 l+l M TRA
(ig)2 ml+k l k m tr
l
q + R 5 2

5
1
i
A
k k R i k +k M TR .

q + kR 5 2
(4.38)
Expanding the latter and using the identities (2.26)(2.30), we obtain the momentum
conserving terms
(4) AA
MM+
(4) AA
MM
g2
1
1

CR AA 2
4
q + l 2 /R 2 q 2 + (l m)2 /R 2
tr {R M S M } qR rS ,
1
1

CR A AA 2
2
2
2
4
q + l /R q + (l m)2 /R 2

tr R (i5 )M S (i5 )M qR rS S ,
(4.39)
g2
(4.40)
where qR = (q , l/R), rS = (q , k/R). Using 5 S = S S 5 together with Eq. (3.6)

gives
(4)AA
MM
=
1
1
g2

CR AA 2
tr{R M S M }qR rS . (4.41)
2
2
2
2
q + l /R q + (l m)2 /R 2
145
The momentum violating terms are

(4)+AA =
MM
(4)AA =
MM

g2
1
1

tr R TRA TRA
4
q 2 + m2+ q 2 + m2

tr R (i5 )M S M qR qS+ S ,
(4.42)

1
1
g 2 A
tr TR R TRA
4
q 2 + m2+ q 2 + m2

tr R M S (i5 )M qR+ qS ,
(4.43)
where here qR+ = (q , m+ ), qS = (q , m ). These are potential brane terms for any value
of A . However, one can easily see that they are zero because of their gamma matrix
structure:

(4) m, m = 0,
(4) m, m = 0.
(4.44)
55
The bulk contribution (4.41) will be first evaluated for M = M = 5:

(4)AA 2
m
55

+
q 2 + (l(m l))/R 2
2[d/2] 2
dd q

g CR AA 4d
,
=
2
(2)d
(q 2 + l 2 /R 2 )(q 2 + (m l)2 /R 2 )
l=
(4.45)
where [x] is defined as usual as the integer part of x. In (4.29) we saw that this integral is
divergence free and that the only non-vanishing contribution comes from m = 0. Thus, the
result after summing and integrating, becomes

(4)AA 2
(4)AA
55
(0) =
m = 0 = 0 and 55
12

(4.46)
g 2 AA CR (3).
4
2
32 R
We stress that here we did not have factors of d 2 (d 1) to protect us against quadratic
(cubic) divergences so it is fortunate that this contribution is completely finite.
After a similar calculation, we obtain for the components along M4
(4)AA
2[d/2] 2

g AA CR 4d
m2 =
2

+
d d q d1 (d 2)q 2 + (l(m l))/R 2
.
(2)d
(q 2 + l 2 /R 2 )(q 2 + (m l)2 /R 2 )
(4.47)
l=
There is again no quadratic divergences since the corresponding pole vanishes in

dimensional regularization in d = 2. The logarithmic divergences for m = 0 correspond
to mass renormalization of the heavy KK gauge bosons just as in the contribution (4.32)
from the gauge sector. For m = 0 though these logarithmic divergences are absent, since
for d = 4 the integral is proportional to (4.33) which was found to be zero, i.e.,
(4)AA
(0) = 0.
(4.48)
146
As a final consistency check, let us see how does the cancellation of the pole work if
we perform the sum over l first. As we have said earlier, cubic divergences which should
somehow cancel will typically appear. Indeed, we can rewrite (4.47) as
(4)AA
2[d/2] 2

g AA CR 4d
m2 =
2

+
d d q d2 (1 d)q 2 + (q 2 + (l(m l))/R 2 )
.
(2)d
(q 2 + l 2 /R 2 )(q 2 + (m l)2 /R 2 )
(4.49)
l=
We have seen in Eq. (4.28) that the term in the bracket corresponds to just a single finite
contribution from m = 0. The first term is the one that has the cubic pole but it is multiplied
by the factor of 1 d so that the cubic divergence in d = 5 is actually absent.
5. The Hosotani mechanism
In this section we will discuss on the possibility that one of the scalars Aa,0
5 radiatively
acquires a vacuum expectation value (VEV) and breaks the gauge group H on the brane to
a subgroup. This would be relevant in model building, for example, when SU(2)W U (1)Y
is a subgroup of H, the scalars Aa,0

5 are interpreted as Higgs fields and the VEV
1 a,0
A R
(5.1)
2 5
breaks H down to U (1)Q . This mechanism is called the Hosotani mechanism [29,30]. It
is not easy to carry out a discussion as general as the one we had up to this point so we
will make a few simplifying assumptions. First, we will assume that there is only one type
of fermions, transforming either in the adjoint or in the fundamental representation of G.
We allow though for multiple flavors of fermions and we will call the number of different
flavors by Nf . Second, we assume that only one of the Higgs fields Aa,0
5 takes a VEV, we
will call this field h and write = vR/2, where v h.
The first step is to look at the squared mass at the origin of the Higgs field. Adding the
contribution (4.46) multiplied by the number of fermion flavors to the result (4.30) from
the gauge sector and noticing that in Euclidean space we are computing the negative mass
squared, we obtain the result

3
g 2 (3) 3C2 (G) 4CR Nf .
m2h =
(5.2)
4
2
32 R
We see that for models satisfying (we use C2 (G) = CG )
a
4
CG
< Nf
CR
3
(5.3)
this is negative and therefore could indeed break H. Of course, even when m2h > 0 the
true vacuum can be at some non-zero . To be more precise, we have to look at the full
effective potential which can be expressed as [18]

1
Tr V (rF ) V (rB ) ,
V=
128 6 R 4
147
with

V (r) = 3 Li5 (r) + Li5 r ,
rF,B = e2iqF,B ()
(5.4)
and where qF and qB are the shifts in the fermion and boson KK masses according to
n + qF ()
n + qB ()
,
mnB =
, n = 0, 1, 2, . . . .
(5.5)
R
R
The question then is whether this potential has a minimum for some non-zero value of ,
which would then trigger the breaking of H, and if this minimum is a global minimum.
It is hard to answer this question in full generality but one can make some progress in
special cases. Two simple examples have been provided in Ref. [46]. For Nf fermions in
the fundamental representation of SU(2) it was found that there is a global minimum at
= 1/2 7 which becomes degenerate with the one at = 0 when Nf 0. Thus, there is a
regime8 0 < Nf 3 where the minimum at = 0 does not correspond to the true vacuum.
However it was also pointed out that the U (1) symmetry generated by 3 which is left over
after the orbifolding remains unbroken. This is because the Wilson-line associated to the
vacuum = 1/2 becomes 1 which commutes with 3 . In the other example, G = SU (3)
with fermions in the fundamental, it was shown that there again exists a minimum at
= 1/2, which already becomes the true vacuum when Nf > 3/2. This is a less stringent
bound than Eq. (5.3), Nf > 9/2, the critical value at which the minimum at = 0 turns
into a maximum. By computing the Wilson line associated to = 1/2 it was finally shown
that this breaks only the SU(2) subgroup of H down to U (1). As in the case of SU(2) the
rank remains preserved.
If the fermions transform in the adjoint representation of G there is a slight simplification
because qF = qB . The dependence becomes the same for fermionic and bosonic
contributions and the symmetry breaking is determined by the global factor 3 4Nf in
front of the potential. For SU(2) one finds two degenerate minima which lie at = 0, 1/2
for Nf < 3/4 and at = 1/4, 3/4 for Nf > 3/4. For SU(3) one has only a minimum at
= 0 for Nf < 3/4 and two degenerate minima at 0.29, 0.71 for Nf > 3/4.
Our next observation is the following statement: the Hosotani mechanism does not
reduce the rank of H if the symmetry breaking global minimum is at = 1/2. To show
this statement, we can compute the Wilson line due to the VEV 1/R of a scalar along the
T A direction:
mnF =
W = eiT .
It is straightforward to show that
that

W , Hi = 0,
(5.6)
exp(iT A )
is a diagonal matrix. Thus, we always have

(5.7)
where Hi are the generators corresponding to the Cartan subalgebra of G, i.e., that the
Wilson loop commutes with at least those generators and therefore it leaves at least a
7 Note that the parameter in Ref. [46] is related to our as = 2.
8 According to Eq. (5.3), = 0 is a maximum for N 3.
f
148
U (1)1 U (1)rank(H) unbroken. On the other hand if = 1/2 then one can reduce
the rank of H by Hosotani breaking. An example of this possibility was provided above
where Nf > 3/4 fermions in the adjoint representation of SU(N) triggered a Hosotani
breaking with = 1/2.
One should however be careful with the interpretation of this effect. Let us look at
orb
the example of the orbifold breaking SU(2) U (1), where the subsequent Hosotani
mechanism according to the above argument apparently leaves the gauge group unbroken
Hos
when = 1/2: U (1) U (1). Strictly speaking, this is not correct. By looking at the
3,0 of H = U (1)
symmetry breaking pattern in detail it can be seen that the gauge boson A
becomes massive by the Hosotani vacuum expectation value and therefore H breaks to
nothing on the brane for generic values of . However, for special values of , such as the
value = 1/2 corresponding to fermions transforming in the fundamental representation
of SU(2), additional gauge bosons can become massless, as can be seen from the mass
matrix of the massive KK gauge bosons which has eigenvalues n2 , (n + 2)2 , (n
2)2 . Clearly, since n > 0, the gauge boson associated to the n = 1 level (some linear
1,1 and A3,1 in this case) becomes massless, so the correct statement is
combination of A
Hos
U (1) U (1) . In fact, in general, for special values of there will be an enhanced gauge
symmetry at the orbifold fixed points.9 The interesting feature here is that the special value
of that results in gauge symmetry enhancement is not arbitrary, instead it is obtained
by minimizing the fixed (once a bulk fermion representation is chosen) effective potential.
Recall that Rv and that having decoupled gravity, R is a parameter assumed to be
fixed to some reasonable value. Then, v is essentially the (only) classical modulus of the
theory. In supersymmetric theories v might remain a modulus even at the quantum level but
in non-supersymmetric theories such as the ones we analyze here, it can be fixed through
a non-trivial one loop potential as we have seen above.
It would be interesting to see if, by turning on gravity, R could be stabilized in the
same manner. Of course, in such a case, another important issue would be if R, being the
extra component of the metric (i.e., g55 ), is protected by gauge invariance against quadratic
divergences like Aa5 , gauge invariance now being general coordinate invariance.
6. Conclusions and outlook

In this paper we have analyzed the one-loop bulk and brane induced radiative effects in a
higher-dimensional gauge theory compactified on an orbifold that breaks an arbitrary gauge
group G into its subgroup H. We have restricted our explicit analysis to a five-dimensional
theory compactified on M4 S 1 /Z2 . The gauge group at the fixed point branes is H and
the higher-dimensional components of the gauge bosons along G/H are even scalar fields
whose (massless) zero modes can become tachyonic by radiative corrections in the bulk
and trigger spontaneous Hosotani breaking of H, i.e., they play the role of Higgs fields
with respect to the gauge group H.
9 Gauge symmetry enhancement at orbifold fixed points is a known effect in string theory.
149
Mass renormalization of Higgs fields in the bulk is expected to be protected from

quadratic divergences by the higher-dimensional gauge invariance G of the theory.
However, since the branes localized at the fixed points are four-dimensional spacetimes,
mass renormalization of the Higgs fields on the branes is not a priori protected from
quadratic divergences by the higher-dimensional theory.
We have computed the one-loop mass renormalization of the Higgs (and gauge) fields
in the bulk and on the branes and found no quadratic divergences at all for any of them.
While this effect in the bulk is justified from the higher-dimensional gauge invariance
its interpretation on the brane for the Higgs fields is less clear, although we believe it
might be related to the higher-dimensional Lorentz and gauge symmetries. In fact, we
have used a five-dimensional covariant gauge (Feynmant Hooft gauge) and dimensional
regularization that are both consistent with the higher-dimensional gauge invariance.
In particular, for the Higgs fields all brane effects vanish while bulk effects are finite
and can trigger, depending on the fermionic content of the theory, spontaneous Hosotani
breaking of the brane gauge group H. For the gauge fields we find only logarithmic
divergences consistent with the mass renormalization of heavy KK modes.
Our results are also consistent with the Higgs fields acquiring a VEV and thus
spontaneously break the gauge symmetry H on the branes with one-loop insensitivity to the
ultraviolet cutoff of the higher-dimensional theory. This insensitivity seems to be a remnant
of the properties of the higher-dimensional theory where the brane is embedded in. Our
results prove that the higher-dimensional gauge theory provides a one-loop solution to the
hierarchy problem, i.e., it replaces the cutoff by the scale 1/R above which the effective
theory becomes higher-dimensional, modulo the stabilization of the compactification
radius that should involve the gravitational sector of the theory. In any case, a quadratic
divergence on the brane would have recreated the hierarchy problem.
Of course our framework cannot be considered as a full solution to the absence of
quadratic divergences until they are proved to vanish at any order of perturbation theory,
or the symmetry protecting them is clearly identified. In fact, a naive analysis of twoloop diagrams prove that there should be non-vanishing Higgs mass renormalization
brane effects from two-loop diagrams. However, since there are one-loop wave function
renormalization effects localized on the branes, they should induce on finite one-loop mass
diagrams non-vanishing two-loop effects localized on the branes. This means that twoloop mass renormalization effects on the branes are mandatory. Only a genuine two-loop
calculation can disentangle a two-loop effect induced by the wave function renormalization
localized on the brane (and thus absorbable by renormalization group running as happens in
supersymmetric theories) from a genuine Higgs mass counter-term localized on it. A twoloop calculation is beyond the scope of the present paper but will be the subject of future
investigation by the present authors.
Note added in proof

After the appearance of the preprint version of this paper, another related paper appeared
[H.C. Cheng, K.T. Matchev and M. Schmaltz, hep-ph/0204342] in which slightly different
results for the masses of the non-zero modes of the gauge bosons A are obtained. This
150
is in no contradiction to our results, since these authors compute the propagator for nonvanishing four momentum p2 = p52 . This constitutes an approximation to the value of the
pole mass which includes wave function renormalization effects. On the other hand, our
masses are computed at p2 = p52 = 0 and therefore they are the running mass parameters
of the effective potential. Both methods yield zero masses for the zero modes, as can be
seen by setting p52 = 0 in their corresponding formulae. Analogously, both methods yield
the same finite result for the zero mode of A5 .
Acknowledgements
The work of G.G. was supported by the DAAD.
Appendix A. Orbifold gauge breaking patterns

We will not try to derive general rules for the allowed gauge breaking patterns and the
associated consistent fermion representations.10 Most of the general features of orbifold
actions on gauge fields and fermions have appeared in one way or another in the early string
theory literature. Instead, in this appendix, we work out a few examples demonstrating
the simplicity but also the restrictiveness of the Z2 orbifold action on the fermion
representations. We recall that the gauge group in the bulk is G, which breaks by the
orbifold action represented by

1dH
0
=
(A.1)
, such that f ABC = AA BB CC f A B C ,
0
1dK
as
G H,
(A.2)
where H = H1 H2 and the broken generators parametrize the coset K = G/H. The
fermion representation of the bulk then breaks up according to this, as
R = r1 r2 ,
where the ri have to be determined by the orbifold action on the fermions

R x , x5 = R i 5 R x , x 5
(A.3)
(A.4)
M A
and from the requirement that the coupling igAA
M R T R is Z2 invariant. The
resulting constraint from the latter is
R TRA R = A TRA ,
(A.5)
10 A treatment of parity assignments to pure gauge-theories and a list of all possible breaking patterns can be
found in Refs. [41,47].
which can be simplified to

R , TRa = 0,
R , TRa = 0.
151
(A.6)
To put it in simple words, for a given representation R, R has to be chosen in such a

way that it commutes with the unbroken generators TRa and anti-commutes with the broken
generators TRa .
The first class of models of interest is when is such that rank(G) = rank(H) (inner
automorphism). In particular, this means that none of the TRa is diagonal and therefore R
can always be chosen to be diagonal of the form

1d1
0
R =
(A.7)
,
0 1d2
where d1 and d2 are model dependent numbers. This is not the case for the second
interesting class of models, the one with chosen such that the rank of G is reduced
(outer automorphism). Reduced rank in particular implies that some of the TRa are diagonal
and therefore (for those diagonal TRa ) the second equality of (A.6) can never be satisfied if
R is diagonal. Thus, for the case of outer automorphism we have to find a non-diagonal
(and unitary) R that solves (A.6). The most interesting case will be
T
AB T B = T A ,
(A.8)
which, e.g., breaks SU(N) SO(N). From Eq. (A.5) it then follows that possible
representations must be real:
T
TRA = R TRA R
(A.9)
(recall that = = 1 ). For non-real representations R one can always choose R =
R R with generators

A
TR
0
.
TRA =
(A.10)
A )T
0 (TR
Comparing with Eq. (A.9) this fixes R to take the block form

0 1R
R =
.
1R 0
(A.11)
Thus R has dR eigenstates (ei , ei ) with positive parity and dR eigenstates (ei , ei ) with
negative parity, where with ei we denote the usual unit vectors. The zero mode spectrum
resulting from this action will always be vector-like and therefore anomaly free. Let us
now present a few examples for both the rank preserving and the rank breaking orbifold
actions.
The first example in the inner automorphism class is an SU(2) gauge group in the
bulk with a pair of Dirac fermions transforming as a doublet under the fundamental
representation of SU(2). First, we have to choose the action on the gauge fields. We
have essentially three choices. The first is to take = diag(+1, +1, +1). This choice
corresponds to an unbroken SU(2) on the fixed planes and therefore it is not an interesting
choice from our point of view. The second possibility is to take = diag(1, 1, 1)
152
which corresponds to a completely broken gauge group on the fixed planes. However, we
have seen in Section 2 that this choice is not compatible with the automorphism constraint
on the Lie algebra so this possibility cannot be realized. The third choice is to choose
= diag(1, 1, +1), which breaks SU(2) down to U (1) on the brane. The gauge boson
of positive parity corresponding to the unbroken U (1) is A3 , whereas the broken coset
1,2 . Similarly, the zero modes of the positive parity
is spanned by the negative parity A
A51,2 are seen as massless scalars in the four-dimensional theory but the negative parity A35
does not have a zero mode. This is an interesting possibility, so let us look at the action
on the fermions in the fundamental representation R = 2 of SU(2). One can easily check
that 2 = diag(+1, 1) (d1 = d2 = 1) commutes with TR3 = 12 3 and anti-commutes with
TR1 = 12 1 and TR2 = 12 2 . The surviving fermions on the brane are then (say) left-handed
Weyl fermions with U (1) charge +1 and right-handed fermions with U (1) charge 1.
Thus, the orbifold action resulted in a broken gauge group and chiral fermions on the
brane. The theory on the brane is anomaly free as can be readily checked.
As a second example, let us look at the breaking pattern SU(3) SU(2) U (1).
To achieve this breaking pattern, we take = diag(+1, +1, +1, 1, 1, 1, 1, +1),
so that the generators corresponding to H = SU(2) U (1) have positive parity and the
generators corresponding to K = SU(3)/(SU(2) U (1)) have negative parity. We also
choose the fermions to transform in the fundamental representation of SU(3), i.e., R = 3.
Then, 3 = diag(+1, +1, 1) (d1 = 2, d2 = 1) commutes with the generators of H and
anti-commutes with the generators of K, so on the brane we will have two massless Weyl
fermions transforming as a doublet under SU(2) and a singlet massless Weyl fermion of
the opposite chirality. Under SU(3) SU(2) U (1) we have
3 = 21 1+2 .
(A.12)
This model as it stands has a cubic U (1) and an SU(2) anomaly. One has, in principle, to
make modifications that render it anomaly free.11
The third and last example in the inner automorphism class is an SU(5) in the bulk
with fermions transforming under some representation of the gauge group. We will
not go through all the possibilities, instead we look at the breaking pattern SU(5)
SU(3) SU(2) U (1). Following the previous examples, we take to have +1s along
the diagonal corresponding to the 12 generators of SU(3) SU(2) U (1) and 1s
along the rest of the diagonal elements. Let us assume that the representation is R = 5.
Next we have to look for a 5 by 5 matrix R that commutes with all the generators
of H and anti-commutes with the generators of K. These constraints then fix 5 =
diag(+1, +1, +1, 1, 1) (d1 = 3, d2 = 2), since for SU(5) SU(3) SU(2) U (1)
5 = (3, 1)+1/3 (1, 2)1/2.
(A.13)
The fermionic zero mode sector consists then of an SU(3) triplet of Weyl fermions of say
right-handed chirality, and an SU(2) doublet of Weyl fermions of left-handed chirality. It
is also possible to carry out the same exercise for R = 10. Here it is more convenient to
11 An example can be found, e.g., in Ref. [36].
153
express 10 in a tensor rather than a matrix form since the ten is the anti-symmetrized
tensor product of the fundamental with itself. We find

1 i
5 m 5 l n 5 i n 5 l m ,
2
which implies that in
10 il mn =
(A.14)
10 = (3, 2)1/6 (3, 1)2/3 (1, 1)1
(A.15)
the zero mode spectrum consists of a (3, 2) of left-handed chirality and a (3, 1) (1, 1)
of right-handed chirality. For a model that has Dirac fermions in the 5 10 in the bulk,
the zero mode spectrum is clearly anomaly free, since the spectrum is that of the Standard
Model.
Before turning to examples of outer automorphisms, we would like to make the
connection between the general formalism reviewed in [41] and formulas (A.1) and (A.5)
again through simple examples. We recall that the action of the orbifold group on the fields,
when the action is an inner automorphism, is via group elements such that
g = e2iV H ,
(A.16)
where H = {Hi }, i = 1, . . . , rank(G), are the generators of the Cartan subalgebra of G and
V is the twist vector specifying the orbifold. For such group elements it is always true that
gHi g 1 = Hi
and gE g 1 = e2iV E ,
(A.17)
where = {i } are the roots of G and E the corresponding ladder generators in the
CartanWeyl basis, satisfying
[Hi , E ] = i E .
(A.18)
Let us first look at the SU(2) U (1) example. Taking the Cartan generator of SU(2)
in the adjoint representation, i.e., H1 = TA3 and requiring that g = = diag(+1, +1, 1)
in (A.16), fixes V = 1/2. Then, 2 is given again by (A.16) with H1 = T23 . A simple
calculation yields 2 = diag(+1, 1) (up to a sign) as we had found earlier. The exponent
in the second of Eq. (A.17) is non-zero for all non-zero roots of SU(2) ( = 1, so
exp (2i V ) = 1) and therefore the only unbroken generatoris H1 . A similar
calculation for the SU(3) SU(2) U (1) example gives V = (0, 3 ) and therefore
3 = diag(+1, +1, 1) (up to a sign). In this case, however, from (A.17) we can see
that in addition to H1 and H2 there are two more unbroken generators, namely, E1 ,
corresponding to 1 = (1, 0), since for those it is 1 V = 0. For the rest of the
generators E2 and E3 we find exp (2i V ) = 1 as expected.
Finally, we will present two examples with rank breaking actions. The first example is
the simplest possible one, namely, a U (1) in the bulk breaking down to nothing on the
branes. The choice that performs this breaking is = 1 and is a simple realization of
Eq. (A.8). Charged fermions are necessarily accompanied by oppositely charged partners.
Thus, the fermionic zero mode spectrum is a vector-like pair of Weyl fermions.
The second simplest example with rank breaking action is SU(3) SO(3). Solving
Eq. (A.8) to obtain , one has to give positive parities to anti-symmetric and negative ones
154
to symmetric generators. This results in the choice

= diag(1, +1, 1, 1, +1, 1, +1, 1).
(A.19)
The parities for fermions are given by Eq. (A.11). Since the positive generators T 2 , T 5 , T 7
form the fundamental representation of SO(3), we find that matter in 3 3 of SU(3) will
transform in 3 3 of SO(3). We checked also that after diagonalizing 33 the positive and
negative parity eigenstates transform in separate 3 representations of SO(3). The fermionic
zero mode spectrum is therefore an SO(3) triplet of Weyl fermions plus their vector-like
partners. Finally we mention that other choices of outer automorphisms not obtained from
Eq. (A.8) are related to this one by an inner automorphism.
Appendix B. The unitary gauge

In this appendix we will study the problem of gauge fixing and the physical (unitary)
gauge in the class of 5D models compactified on the orbifold S 1 /Z2 considered in this
paper, where the gauge group G is broken by the orbifold action into its subgroup H. As we
have seen in Section 5 in the presence of non-vanishing background values for the scalars
in the adjoint representation of G the subgroup H can be further broken and the
AA,0
5
mass pattern induced by the orbifold breaking will be modified. We will first consider, for
simplicity the case of zero VEV for AA,0
5 . The Hosotani breaking case will be subsequently
studied.
We have seen that the choice of the gauge
1
GA = M A A
M
(B.1)
does not lead, for any value of the parameter , to the unitary gauge. In the absence of
VEV for the fields AA,0
5 this can be achieved for the gauge fixing condition

1
5 A
GA = A A
+ A5 ,
(B.2)
consistent with the orbifold action. Using now the gauge-fixing condition (B.2) and the
infinitesimal transformation of the field AM under the gauge transformation (x , x 5 ),
1
AM AM + DM ,
g
standard techniques yield the FaddeevPopov Lagrangian

LFP = Tr c D + 5 D5 c.
The propagators can be worked out as in (2.19) and (2.20) and yield

(1 )p p
BC
(A )
(p, p5 ) = i
g
,
G
p2 p52
p2 p52
G(A5 ) (p, p5 ) = i
BC
,
p2 p52
(B.3)
(B.4)
(B.5)
G(c) (p, p5 ) = i
BC
.
p2 p52
155
(B.6)
After mode decomposition p5 = n/R and so for n = 0 one reaches the unitary gauge in the
A
limit . In this limit the massive modes of AA
5 and c decouple while the massive
A
modes of A only propagate their physical degrees of freedom. This corresponds to the
gauge fixing condition 5 AA
5 = 0, that does not fix the gauge in the zero-mode sector. For
the zero modes left out by the orbifold breaking the gauge symmetry is unbroken and, in
the absence of Hosotani breaking, one cannot define a unitary gauge, as it is obvious from
Eq. (B.4) to (B.6).
some of the massless gauge bosons in H acquire a
Turning now VEVs for fields AA,0
5
mass and the definition of the unitary gauge can be enlarged to also take into account this
effect. The analysis can be readily done in full generality as follows. Let us consider that
will acquire a VEV by quantum corrections. Their tree level potential
some fields AA,0
5
is flat, since they are part of the gauge bosons in 5D, and we can write the general
decomposition for them into a classical part and quantum fluctuations as

AA,0
(B.7)
x = vA + A x .
5
If we restrict ourselves to x -dependent gauge transformations = (x ) we can move
B,0 C
A
away from v A by means of gauge transformations AA,0
5 = f BC A5 which shows
A
C
that field fluctuations along f CB v correspond to Goldstone bosons for the zero-mode
sector. In fact, if we define the mass matrix
A
C
MA
B = gf CB v ,
the zero modes of the gauge bosons acquire the squared mass matrix
2 AB
AB
MA
= MMT
.
(B.8)
(B.9)
In order to incorporate the Hosotani breaking into the 5D formalism we can modify the
gauge fixing condition (B.2) and define the R gauge
5 A

1
A B
.
GA = A A
+ A 5 + MB
The Goldstone bosons (zero modes) acquire similarly a mass

2

MG AB = MT M AB .
(B.10)
(B.11)
Of course, not only the zero modes will acquire a symmetry breaking mass but also all
massive modes will get the mass
n A
M B = mn BA + MA
(B.12)
B,
where mn = n/R is the compactification mass and MA
B is the symmetry breaking mass
given in (B.8). Then propagators (B.4), (B.5) and (B.6) become

AB

i
(1 )p p
G(A ) (p, n) =
(B.13)
g
p2 Mn (Mn )T
p2 Mn (Mn )T
156

1
,
p2 (Mn )T Mn AB

AB
1
(c)
.
G (p, n) = i 2
p Mn (Mn )T
G(A5 ) (p, n) = i
(B.14)
(B.15)
The matrix character of the propagators implies that the matrix Mn should be invertible,
i.e., it satisfies the condition

det Mn = 0,
(B.16)
which is the necessary condition for gauging away the corresponding Goldstone boson.
References
[1] J. Polchinski, String theory. Vol. 2: Superstring Theory and Beyond, Cambridge Univ. Press, Cambridge,
UK, 1998.
[2] I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[3] I. Antoniadis, C. Munoz, M. Quirs, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309.
[4] I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 331 (1994) 313, hep-ph/9403290.
[5] P. Nath, M. Yamaguchi, Phys. Rev. D 60 (1999) 116004, hep-ph/9902323.
[6] I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 460 (1999) 176, hep-ph/9905311.
[7] T.G. Rizzo, J.D. Wells, Phys. Rev. D 61 (2000) 016007, hep-ph/9906234.
[8] R. Casalbuoni, S. De Curtis, D. Dominici, R. Gatto, Phys. Lett. B 462 (1999) 48, hep-ph/9907355.
[9] A. Delgado, A. Pomarol, M. Quirs, JHEP 01 (2000) 030, hep-ph/9911252.
[10] E. Accomando, I. Antoniadis, K. Benakli, Nucl. Phys. B 579 (2000) 3, hep-ph/9912287.
[11] I. Antoniadis, K. Benakli, Int. J. Mod. Phys. A 15 (2000) 4237, hep-ph/0007226.
[12] J.D. Lykken, Phys. Rev. D 54 (1996) 3693, hep-th/9603133.
[13] G.F. Giudice, R. Rattazzi, J.D. Wells, Nucl. Phys. B 544 (1999) 3, hep-ph/9811291.
[14] E.A. Mirabelli, M. Perelstein, M.E. Peskin, Phys. Rev. Lett. 82 (1999) 2236, hep-ph/9811337.
[15] EOT-WASH Group, E.G. Adelberger, hep-ex/0202008.
[16] A. Pomarol, M. Quirs, Phys. Lett. B 438 (1998) 255, hep-ph/9806263.
[17] I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quirs, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410.
[18] A. Delgado, A. Pomarol, M. Quirs, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[19] R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.
[20] N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hepph/0102090.
[21] A. Delgado, M. Quirs, Nucl. Phys. B 607 (2001) 99, hep-ph/0103058.
[22] A. Delgado, G. von Gersdorff, P. John, M. Quirs, Phys. Lett. B 517 (2001) 445, hep-ph/0104112.
[23] R. Contino, L. Pilo, Phys. Lett. B 523 (2001) 347, hep-ph/0104130.
[24] H.D. Kim, hep-th/0109101.
[25] V. Di Clemente, S.F. King, D.A.J. Rayner, Nucl. Phys. B 617 (2001) 71, hep-ph/0107290.
[26] A. Delgado, G. von Gersdorff, M. Quirs, Nucl. Phys. B 613 (2001) 49, hep-ph/0107233.
[27] H. Hatanaka, T. Inami, C.S. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067.
[28] H. Hatanaka, Prog. Theor. Phys. 102 (1999) 407, hep-th/9905100.
[29] Y. Hosotani, Phys. Lett. B 126 (1983) 309.
[30] Y. Hosotani, Ann. Phys. 190 (1989) 233.
[31] E. Ponton, E. Poppitz, JHEP 06 (2001) 019, hep-ph/0105021.
[32] G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307.
[33] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678.
[34] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[35] I. Antoniadis, K. Benakli, M. Quirs, Nucl. Phys. B 583 (2000) 35, hep-ph/0004091.
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
157
I. Antoniadis, K. Benakli, M. Quirs, hep-th/0108005.

H. Georgi, A.K. Grant, G. Hailu, Phys. Rev. D 63 (2001) 064027, hep-ph/0007350.
H. Georgi, A.K. Grant, G. Hailu, Phys. Lett. B 506 (2001) 207, hep-ph/0012379.
W.D. Goldberger, M.B. Wise, Phys. Rev. D 65 (2002) 025011, hep-th/0104170.
R. Contino, L. Pilo, R. Rattazzi, E. Trincherini, Nucl. Phys. B 622 (2002) 227, hep-ph/0108102.
A. Hebecker, J. March-Russell, Nucl. Phys. B 625 (2002) 128, hep-ph/0107039.
M.E. Peskin, D.V. Schroeder, An Introduction to Quantum Field Theory, AddisonWesley, Reading, USA,
1995, p. 842.
J. Papavassiliou, A. Santamaria, Phys. Rev. D 63 (2001) 125014, hep-ph/0102019;
A. Muck, A. Pilaftsis, R. Ruckl, hep-ph/0203032.
P. Ramond, Field Theory: A Modern Primer, Benjamin/Cummings, 1989.
S. Groot Nibbelink, Nucl. Phys. B 619 (2001) 373, hep-th/0108185.
M. Kubo, C.S. Lim, H. Yamashita, hep-ph/0111327.
R. Slansky, Phys. Rep. 79 (1981) 1.

Singular 7-manifolds with G2 holonomy and

intersecting 6-branes
Klaus Behrndt
Max-Plank-Institut fr Gravitationsphysik, Albert Einstein Institut, Am Mhlenberg 1, 14476 Golm, Germany
Abstract
A 7-manifold with G2 holonomy can be constructed as a R3 bundle over a quaternionic space.
We consider a quaternionic base space which is singular and its metric depends on three parameters,
where one of them corresponds to an interpolation between S4 and CP 2 or its non-compact analogs.
This 4d Einstein space has four isometries and the fixed point set of a generic Killing vector is
discussed. When embedded into M-theory the compactification over a given Killing vector gives
intersecting 6-branes as IIA configuration and we argue that membrane instantons may resolve the
curvature singularity. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Minimal supersymmetric field theories in four dimensions have many phenomenologically interesting features and can be obtained by CalabiYau compactification of heterotic
string theory. Over the years this has been the standard way for their construction, because
one obtains naturally chiral fermions and non-Abelian gauge groups. On the other hand,
it is much more difficult to obtain phenomenological interesting N = 1 models directly
from M-theory, see [1,2]. One reason is that the 7-manifold has to have G2 -holonomy and
these spaces are not yet well understood. Only few non-compact examples, that rely on the
construction done in [3,4], are explicitly known. Another reason are the difficulties to obtain a model with chiral fermions in four dimensions [5]. In fact, chiral fermions, but also
non-Abelian gauge groups, require that the 7-manifold is singular [68] and a supergravity
approximation may become questionable.
E-mail address: behrndt@aei.mpg.de (K. Behrndt).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 8 - 9
K. Behrndt / Nuclear Physics B 635 (2002) 158174
159
There are mainly two classes of known metrics: one is topologically a R3 bundle over
a quaternionic space and the other a R4 bundle over S3 . They have been introduced in
[3,4] and many generalizations, with more parameters or functions, have been discussed
in the past year. It is impossible to give a complete list of references, but relevant aspects
for our consideration can be found in [816]. In the first class one can further distinguish
between different quaternionic spaces as, e.g., the 4-sphere S4 = SO(5)
SO(4) but also the complex
projective space CP 2 = SU(3)
U (2) , which are the only compact homogeneous quaternionic
4-dimensional spaces [17]. Apart from their non-compact analogs, there are also nonhomogeneous quaternionic spaces, which we want to use for the construction of explicit
metrics.
Quaternionic spaces have been discussed in the physical literature mainly as moduli
spaces of N = 2 supergravity in four and five dimensions [18,19], for a more recent discussion see also [20,21]. The quantum moduli space is expected to be non-homogeneous,
but unfortunately not many examples are explicitly known. In a long quest to incorporate
brane world scenarios into gauged supergravity and to overcome numerous problems, see,
e.g., [22,23], a non-homogeneous quaternionic space has been explored recently in [24].
This space interpolates between the two homogeneous spaces and it is obvious to consider
this space also in the construction of metrics with G2 holonomy and therewith to unify the two spaces representing R3 bundles over S4 and CP 2 . More important is however,
that the Killing vectors of this non-homogeneous space have additional fixed points which
become additional D6-branes after compactification along this Killing vector [25]. To be
more clear, a co-dimension four fixed point set of a given isometry extends in 6 + 1 dimensions representing a NUT (= point) on the quaternionic space can be identified as a
6-brane. There are also co-dimension two fixed points, but the interpretation of these bolts
on the quaternionic space is unclear.
As it has been shown in [26] on any component of a non-compact homogeneous
quaternionic spaces can be at most one fixed point and for compact we expect at most two.
Hence, in order to find a configuration with three or more 6-branes one necessarily needs in
this setup non-homogeneous quaternionic spaces. As we will see, depending on the choice
of parameters, our space can have up to five NUT fixed points. The additional fixed points
reflect the topological non-trivial nature of this manifold, because, due to the Lefschetz
fixed point theorem, every NUT fixed point adds one unit to the Euler characteristic of the
manifold, see also [8,13].
The paper is organized as follows. In the next section we basically follow the literature
and solve the Killing spinor equations followed by a discussion of the closed and co-closed
3-form. This consideration is very general without using any specific quaternionic space. In
Section 3 we will investigate examples corresponding to different quaternionic spaces. In
Section 3.1 we start with the homogenous cases and in Section 3.2 we discuss the following
basic features of the non-homogenous space: it has four Killing vectors; it exhibits a
curvature singularity, which separates two asymptotic regions and finally it interpolates
between the two homogeneous spaces. In Section 4 we give a detailed analysis of the
isometries and the fixed point set; many details for the non-compact case were already
derived in [24]. Due to their interpretation as D6-branes, we especially identify NUT fixed
points. For compact quaternionic spaces there can be three or five and for non-compact
we found three or four non-degenerate fixed points, depending on the isometry groups
160
U (1) SU(2) or U (1) SL(2, R). Unfortunately, in the IIA description the string coupling
constant is not boundedevery model contains at least one non-compact direction in
which the dilaton diverges. We conclude with a discussion of two aspects that could be
interesting for future investigations: (i) resolution of the curvature singularity, e.g., by
of membrane instantons (or general G-fluxes) and (ii) the construction of new Spin(7)
manifold in complete analogy to the procedure of this paper.
2. Constructing metrics with G2 holonomy

Before we can discuss explicit examples let us summarize some aspects of the procedure
described in [3,27,28] which will also fix our notations. After discussing the ansatz for
the metric, we will derive the Killing spinor and the closed and co-closed 3-form. Both
conditions are sufficient to ensure supersymmetry for this background.
2.1. The metric ansatz
The construction of the 7-manifold relies on a 4d quaternionic base space and before we
discuss the metric ansatz we need some basic properties of these spaces; for a recent resume
about quaternionic geometry we refer to [20] and the appendix of [21]. Quaternionic
spaces are generalizations of complex spaces that allow for three complex structures J i
(i = 1, 2, 3) defined by the algebra
J i J j = I ij + ij k J k
(1)
and denoting the quaternionic vielbein by em , one obtains three 2-forms i by

i n
i = em Jmn
e .
(2)
The holonomy of a 4n-dimensional quaternionic spaces is contained in Sp(n) SU(2).

This statement is trivial for n = 1 and is replaced by the requirement that the Weyl-tensor
of 4-dimensional quaternionic space has to be anti-selfdual [18]
W + W = 0.
For a quaternionic space in any dimensions the triplet of 2-forms i is expressed in terms
of the SU(2)-part of the quaternionic connection Ai as
1
dAi + ij k Aj Ak = i
2
(3)
which ensures that the triplet of 2-forms is covariantly constant. Moreover, any quaternionic space is an Einstein space with the curvature implying that its metric gmn solves
the equation
Rmn = 3gmn .
(4)
161
The complex structures can be selfdual or anti-selfdual and in our notation we will take the
i = 1
i
latter one (Jmn
2 mnpq Jpq ) so that the triplet of 2-forms can be written as
1 = e4 e7 e5 e6 ,
2 = e4 e6 + e5 e7 ,
3 = e4 e5 + e6 e7 .
(5)
Moreover, the SU(2) connection is given as the anti-selfdual part of the spin connection
mn of the quaternionic space
1
i
Ai = mn Jmn
(6)
.
2
In the same way, the selfdual part gives the Sp(n) connection.
Having the basic relations for the quaternionic base space, the metric of the 7-manifold
is introduced by the ansatz [3,4]
2 = e2f i i + e2g em em ,
ds
where
(m = 4, . . . , 7) is the vielbein of the quaternionic space and
defined by
em
i ui = dui + ij k Aj uk
(7)
i
(i = 1, . . . , 3) is
(8)
with ui as local coordinates (in addition to the quaternionic once). Using the relation (3),
it is straightforward to verify that
i d i + ij k Aj k = ij k j uk .
(9)
The two unknown functions f and g in the metric ansatz are now fixed by the requirement
of G2 holonomy implying the existence of a Killing spinor or equivalently the existence of
a closed and co-closed 3-form. We will check both conditions.
2.2. Solving the Killing spinor equations
Consider M-theory on the manifold M4 X7 and assume that M4 is the flat 4d
Minkowski space. If we moreover assume the absence of G-fluxes, a Killing spinor is a
solution of the equation (a, b = 1, . . . , 7)

1 ab
d + ab = 0,
(10)
4
where ab is the spin connection 1-form and ab = [a b] with a as the 7d gamma
matrices. If this equation has exactly one solution, the resulting 4-dimensional field theory
has N = 1 supersymmetry and the manifold X7 has G2 holonomy. Recall, the unrestricted
holonomy of a 7-manifold is SO(7), but the existence of a (covariantly constant) Killing
spinor implies that the holonomy of the manifold is restricted. A generic spinor transforms
as representation 8 of SO(7), while a Killing spinor implies the decomposition 8 7 + 1,
which is exactly the decomposition under G2 SO(7). The exceptional group G2 appears
162
Fig. 1. The multiplication table of imaginary components of octonions: i1 i2 = i3 , i6 i2 = i4 , i4 i7 = i1 , . . . can be

obtained from this figure by following the lines in the direction of the arrows.
as automorphism group of octonions: o = x 0 I + x a ia , where ia satisfy the algebra

ia ib = ab + abc ic ,
for more details see, e.g., in [27,2931]. The G2 -invariant 3-index tensor abc can be
obtained from Fig. 1 and is given in the standard basis by
1
abc ea eb ec = e1 e2 e3 + e4 e3 e5 + e5 e1 e6 + e6 e2 e4
3!
+ e4 e7 e1 + e5 e7 e2 + e6 e7 e3
= e1 e2 e3 + ei i ,
(11)
where we used the complex structures as introduced in (2) and (5).

If the Killing spinor equation (10) has more than one solution, the holonomy is smaller
than G2 resulting in a 4d field theory with N > 1 supersymmetry. The holonomy is equal
to G2 if the Killing spinor equation has exactly one solution, which is equivalent to the
absence of covariantly constant 1-forms [3]. The existence of such a 1-form would imply
the existence of a covariantly constant Killing vector and hence to a factorization of the
space. As we will discuss in the next section this is not the case for the examples that we
consider and hence our models yield N = 1 supersymmetry in 4 dimensions.
Following the arguments from [27,28], one can define two orthogonal projectors P
ab +
ab with
that decompose the spin connection into two parts: ab = +

2 ab 1 abcd
ab
= P+abcd cd
cd ,
+
+
3
4

1 ab 1 abcd
ab

(12)
= Pabcd cd
cd ,
3
2
where abcd is the G2 invariant 4-index tensor which is dual to the 3-index tensor
(abcd = 3!1 abcdefg efg ). In order to solve (10) one imposes on the (constant) Killing
spinor and the spin connection cd the projector equations
P+abcd cd = 0,
Pabcd cd = 0
(13)
163
and finds as solution for the Killing spinor [27]

= c 8 ,
(14)
where = 1, . . . , 8 is the SO(7) spinor index and c is constant. The projector condition
on the spin connection can be simplified by contracting with the 3-index tensor abc and
using the relation: abc bcde = 4abc . One infers [28]
abc bc = 0
(15)
yielding first order differential equations for the unknown functions. We will assume in the
following that in the metric ansatz (7) f = f (|u|) and g = g(|u|) and obtain for the spin
connection
ij =
f [i j ]
u ij k Ak ,
|u|
mn = mn e2(f g) ij k uk Jmn i ,
g gf m i
j
e
e u ef g ij k uk Jmn en ,
mi =
|u|
j
(16)
where ( ) denotes the derivative with respect to |u|. In the contraction with the 3-index
tensor (11) we use the relation (6) and find

f
ab
jk
mn
2(f g) ij k j k
+e
0 = iab = ij k + imn =
u ,
|u|

g
i
i
0 = mab ab = mni ni = Jmn
(17)
ni =
e n ui .
e2(f g) egf Jmn
|u|
These equations have the symmetry: f f + , g g + , giving a constant conformal
rescaling of the metric (7) and a second symmetry: f f , u e u leaves the metric
invariant. Using these symmetries, only one (discrete) integration constant c = 1, 0
appears in the solution that can be written as [3]
e4f = 2|u|2 + c,
e4g = 2|u|2 + c
(18)
with |u|2 = (u1 )2 + (u2 )2 + (u3 )2 . Notice if < 0 (i.e., a non-compact quaternionic space),
the parameter range of u is bounded by 2|u|2 + c > 0, while for 0 u is unbounded.
2.3. The closed and co-closed 3-form
The holonomy reduction from SO(7) to G2 is equivalent to the existence a 3-form
which is closed and co-closed. Again following the procedure done in [3] we write this
3-form as
1
abc ea e b e c = e3f 1 2 3 + ef +2g i i
(19)
3!
and we will show now that the functions (18) ensure that: d = d = 0, i.e., it is closed
and co-closed. Since d(ui ui ) = d|u|2 = 2|u| d|u| = 2ui i and i = 0 as well as using
=
164
the relation (9) it follows

1

1
d ij k i j k = ij k i j k = |u| d|u| i i ,
3!
2

d i i = i i i i = 0,
3f

de
1 2 3 = e3f d|u| 1 2 3 = 0.
Therefore, from d = 0 we derive the equation

|u|e3f = ef +2g .
(20)
(21)
Next, for the co-closure we have to check whether the dual 4-form
1
= e2(f +g) ij k i j k + e4g e4 e5 e6 e7
2
(22)
is also closed. Employing the relation i j = 2 ij e4 e5 e6 e7 and again (9) one

obtains

1

= ij k i j k
i
j
k
2d ij k
= 4ui i e4 e5 e6 e7
(23)
and thus (recall ui i = |u| d|u|)

1
0 = d = e2(f +g) d|u| ij k i j k
2

+ 4|u|e2(f +g) + e4g d|u| e4 e5 e6 e7 .
This gives the two differential equations
2(f +g)
4g
e
= 0,
e
= 4|u|e2(f +g)
(24)
(25)
and it is straightforward to verify that these two equations together with (21) are equivalent
to the first order equations derived from (17) with the solution (18).
3. Explicit metric of the 7-manifold

In the previous section we have verified that the metric of the 7-manifold with G2
holonomy is given by [3,4]
i
2
1
ds 2 =
(26)
du + ij k Aj uk + 2|u|2 + c em em ,
2|u|2 + c
where em is the vielbein of the quaternionic space with the SU(2) connection Aj as
introduced in (6). By choosing different quaternionic spaces we can now discuss explicit
models.
165
3.1. Homogeneous quaternionic spaces

We will start with examples that played an important role in the recent literature. They
base on homogeneous quaternionic spaces, which appear in two classes [17,32], namely,
(i) the maximal symmetric spaces
(ii) the complex projective spaces
SO(5)
SO(4,1)
SO(4) and SO(4) with 10 isometries and
SU(3)
SU(2,1)
SU(2)U (1) and SU(2)U (1) each having 8 isometries.
The metric of the quaternionic space and the corresponding SU(2) connection for the
maximal symmetric case (i) can be written as

d 2
em em =
+ 1 2 d 2 + 2 d 2 + sin2 d 2 ,
2
1
A1 = 1 2 sin d,
A2 = 1 2 d,
A3 = d cos d.
For = +1 this metric describes the coset
(27)
SO(5)
SO(4)
= S4 ,
and for = 1 the corresponding
SO(4,1)
SO(4) .
non-compact analog,
These spaces with the maximal number of Killing vectors
become trivial for = 0 and have a vanishing Weyl tensor. The complex projective spaces
(ii) have less isometries and the Weyl tensor is non-trivial. The corresponding expressions
read (cp. [33])
2 d 2
2
+
(d cos d)2
(1 + 2 )2 2(1 + 2 )2
2

2
d + sin2 d 2 ,
+
2
2(1 + )
sin
1
d,
A2 =
d,
A1 =
1 + 2
1 + 2
em em =
A3 =
(2 + 2 ) cos
2
d
d.
2(1 + 2 )
2(1 + 2 )
(28)
As before, the space is compact for > 0 and otherwise non-compact.

Both spaces are spherical symmetric related to the S2 parameterized by (, ). But there
is also a generalization to any 2-space with constant curvature (as it will appear in the
next example). If we take into account this additional parameter, the resulting G2 metric
(26) depends in total on three parameters: , c and .
3.2. Quaternionic spaces with four isometries
More general, non-homogeneous quaternionic spaces may be classified by the number
of isometries, but not many concrete examples are known; see however [34,35]. One
example with four isometries has been discussed in [24] and when regarded as a hyper
multiplet moduli space yields upon gauging a supersymmetric RandallSundrum scenario.
166
The metric and connection for this space is given by

y dx x dy 2 2
d 2
dx 2 + dy 2
m m
2
+ V () d n
e e =
+
n
2 ,

2
V ()
1 + 4 (x + y 2 )
1 + 4 (x 2 + y 2 )

dy
dx
+n
+n
1
2
V
V
A =
,
A =
,
2
2
2
n 1 + 4 (x + y )
n 1 + 4 (x + y 2 )

y dx x dy
A3 = ( n) d + 2n(n )
(29)
,
1 + 4 (x 2 + y 2 )
with
( n)
V =
( + )( ),
+n

= n
4n2 +
(30)
As we mentioned in the last paragraph the parameter = 0, 1 determines the symmetry

group: for = +1 it is a spherical symmetric solution with the isometry group U (1)
SU(2); for = 1 the isometry group becomes U (1) SL(2, R) and for = 0 it is the
solvable subalgebra of SO(1, 4).
To understand this space better consider = 1, where it becomes the metric of the
Taub-NUT-(A)dS space [36], with n as NUT parameter and the mass parameter had been
fixed to ensure the quaternionic property (3) (or equivalently the anti-selfduality of the
Weyl tensor). For vanishing cosmological constant ( = 0) one obtains the well-known
Taub-NUT metric, and hence, this space represents topologically an orbifold. In fact, this
orbifold is related to the non-trivial periodicity of , which ensures the absence of conical
singularities:
+ 4n.
(31)
By complete analogy to the Ricci-flat Taub-NUT case we can make the orbifold action
explicit. For = 1 (the other case goes in complete analogy, see also [24]) the coset
S4 = SO(5)/SO(4) is defined by
(X0 )2 + (X1 )2 + (X2 )2 + (X3 )2 + (X4 )2 = 1
(32)
with the metric

ds 2 = (dX0 )2 + (dX1 )2 + (dX2 )2 + (dX3 )2 + (dX4 )2
(33)
subject to the constraint (32). Before imposing the constraint, the SO(5) symmetry group
is manifest, but afterwards only a subclass of these isometries are realized linearly and the
other symmetries are not manifest. Since we are interested here in the spherical symmetric
case ( = 1) we introduce polar coordinates in X1,2,3,4: (dX1 )2 + (dX2 )2 + (dX3 )2 +
(dX4 )2 = d 2 + 2 d3 and the constraint becomes X02 + 2 = 1. Since X0 dX0 = d
we can eliminate X0 and find for the metric
ds 2 =
d 2
d r 2 + r 2 d3 dz1 d z 1 + dz2 d z 2
2
+
d
=
=
3

2
2
2 2
2| 2
1 2
1 + r4
1 + |z1 | +|z
4
(34)
167
with = 1+rr2 /4 . As for the Taub-NUT space the Zn orbifold acts on the two complex
coordinates as
z1 e2i/n z1 ,
z2 e2i/n z2
(35)
which, after the change of coordinates

z1 = r cos(/2)ei(+)/2 ,
z2 = r sin(/2) ei()/2,
(36)
is equivalent to (31). Hence, this Zn orbifold acts on the

subspace as:
in
the usual way; for a discussion of orbifolds see also [9,14,15].
In addition to the reduced number of isometries, given by the subgroup that commutes
with the orbifold action, there are more significant differences to the homogeneous cases
that we discussed before. First, in the limit = 0 it becomes the Taub-NUT space, which is
hyper-Kaehler and since it is still non-trivial it may be of interest for G2 manifolds as well.
Second, the quaternionic space (29) has a curvature singularity at = n as indicated by
the curvature invariant

4n2 ( + 4n2 )2
.
Rmnst R mnst = 24 2 +
(37)
( + n)6
S3
S3
S3 /Zn
This singularity exist for any value of ui and hence also the G2 manifold (26) becomes
singular, but it is shielded by the Killing horizon where V = 0 representing a fixed point
set of k = , see below. Note, the zeros of V are regular points as it is obvious from the
curvature invariant (37). But at these zeros V changes its sign and therefore the metric is
well-defined only on the coordinate patch where
2 n2 0
V 0,
which allows for two physical regions that are disconnected by the curvature singularity,
see Fig. 2:
max{n, + }.
,
Since the metric is invariant under and n n we can assume that n > 0.
Before we discuss in the next section the Killing vectors and their fixed point set, let us
note that the solution (29) interpolates between the two homogeneous quaternionic spaces
that we introduced in Eqs. (27) and (28). To see this, let us set = +1. Obviously, if n = 0
we get V = 1 2 and obtain the maximal symmetric spaces S4 or EAdS4 in Eq. (27).
On the other hand, if we transform
+ n,
n
and take the limit
=
(38)
keeping = f ix
one finds the metric

ds 2 =
= 2n

1 y dx x dy 2
2d 2
dx 2 + dy 2
+ 2(1
4 )
d
+ 2
.
2
2
x 2 +y 2 2
(1
4 )
2 1 + x +y
1
+
4
4
(39)
168
Fig. 2. A positive definite metric requires: V > 0 as well as 2 > n2 . Hence, there are two allowed coordinate
regions, as we have indicated by the arrows.
Finally, the transformation

=
r2
,
4(1 + r 2 )
x + iy = 2 tanh ei
2
brings us to the complex projective space in Eq. (28).

Since negative n are equivalent to positive (after ), the two homogenous spaces
appear at the endpoints of the parameter space of n and in both limits the curvature
singularity disappears and the spaces become smooth. Let us also note, that in the limit of
vanishing cosmological constant the space (29) is the standard Taub-NUT space, whereas
the space (39) which appears in the large n limit becomes for = 0 the EguchiHanson
space.
4. Isometries and IIA description

Having the metric of the manifold X7 we can consider M-theory on M4 X7 and
if X7 has at least one isometry the dimensional reduction will give a IIA description.
Especially interesting are Killing vector fields with a fixed point set L of co-dimension
four implying that L extends in 6 + 1 dimensions, and hence, become D6-branes upon
compactification [25]. But before we come to the fixed point set let us discuss the Killing
vectors.
4.1. Isometries
An isometry is related to the existence of a Killing vector field k satisfying the equation
D(m kn) = 0. In a given coordinate patch one can introduce proper coordinates so that the
169
Killing vector field becomes k = and the ( -independent) metric reads

2
ds 2 = e2(x) d + m (x) dx m + gmn (x) dx m dx n .
(40)
In these (local) coordinates the Killing vector field becomes kM = e2 {1, m } and due
to the Killing property one find DM kN = D[M kN] = [M kN] . Obviously any covariantly
constant vector field is also a Killing vector with the consequence that and m are trivial
(i.e., d = 0). This in turn implies that the space factorizes and the holonomy is reduced.
Therefore, the holonomy in our case is equal to G2 iff there are no covariantly constant
(Killing) vector fields. The absence of such Killing vectors for 7-manifolds given as R3
bundles over S4 and CP 2 was shown in [3,4]. Since our case represents an interpolation
between these two spaces, where only a subclass of isometries survive, we are still dealing
with a manifold where the holonomy is equal to G2 .
To be more concrete, the isometries of X7 with the metric (26) are given by the
isometries of the quaternionic space (29) and in addition SU(2) transformations of the
coordinates ui . The quaternionic space has four Killing vectors [36] which can be written
as [24]

x2 y2
xy
k1 = ny 1 +
x y ,
4
2

2
xy
x y2
y ,
k2 = nx x 1
2
4
k4 =
k3 = 2n + yx xy ,
(41)
and fulfill the algebra
[ki , k4 ] = 0,
[ki , kj ] = fij l kl
(42)
with f123 = 1, f231 = f312 = . For the spherical symmetric case = 1 we get thus the
symmetry SU(2) U (1), where the U (1) corresponds to the orbifold action (35) and
the SU(2) is the subgroup of SU(3) or SU(2, 1) that commutes with the orbifold. For
the hyperbolic case ( = 1) one obtains the non-compact analog, namely the algebra
SL(2, R) U (1). For = 0 two Killing vectors become equivalent (k3 k4 ) and we should
take a different parameterization, namely,
k1 = ny x ,
k2 = nx y ,
k3 = yx xy ,
k4 = .
(43)
They satisfy the algebra

[ki , k4 ] = 0,
[k1, k2 ] = 2nk4 ,
[k2 , k3 ] = k1 ,
[k3 , k1 ] = k2
(44)
which is not anymore a direct product of the two Lie algebras.

From our discussion above it is clear that a covariantly constant Killing vector would
commute with the other Killing vectors. But the only commuting Killing vector k4 = is
not covariantly constant, due to the non-trivial U (1) fibration in (29). Therefore the space
cannot be factorized and the holonomy is equal to G2 , which we inferred already from the
relation to the other known G2 manifolds.
170
4.2. Fixed point set

For a given Killing vector k, the hypersurface L defined by |k|2 = 0 is the fixed point
set and L degenerates if the surface gravity vanishes, i.e., if |Dk|2 = 0 at |k|2 = 0;
see, e.g., [37,38]. For = 0 the fixed points of all Killing vectors are non-degenerate
implying a periodic identification along the Killing direction to ensure the absence of
conical singularities. This fact is well known from non-extreme black holes where the
event horizon is a non-degenerate fixed point set for a timelike Killing vector. In the case
here, the compactness of the Killing direction ensures that the KaluzaKlein gauge group
becomes U (1) and that the D6-branes, which are identified as the fixed point set L [25],
are at finite geodesic distance.
In order to identify the D6-branes we need a Killing vector that has a co-dimension
four fixed point set L, or in other words, L extends over three coordinates in X7 which,
in addition to M4 , become the world-volume of the D6-branes. We should therefore not
consider the Killing vectors related rotations of the ui coordinates. They have a fixed point
set at |u| = 0 which is a point in the 3d ui -space and hence a co-dimension three fixed
point set. On the other hand, the fixed points of the quaternionic Killing vectors can be
NUTs or bolts, depending on the rank of the 2-from dk; see [37]. The NUTs are points
on the quaternionic space and since the SU(2) connection Ai becomes trivial, these NUTs
correspond to isotropic D6-branes. On the other hand, if L is a bolt on the quaternionic
space it has co-dimension two and since at least one SU(2) connection remains nontrivial, there is no isotropic brane interpretation. After dimensional reduction we obtain
a supergravity solution which is singular at the fixed point set L and if L is a NUT the
singularity can be identified with the location of the D6-brane, but if L is a bolt there is no
clear interpretation. It may be related to some deformation of a given D6-brane, since, as
we will see, the bolts are connected with the NUTs for specific Killing vectors.
A generic Killing vector of the quaternionic space, can be introduced as a linear
combination of the four Killing vectors in (41)
k = 1 k1 + 2 k2 + 3 k3 + 4 k4

x2 y2
xy
= ny1 nx2 + 2n3 + 4 1 +
1 + 2 + y3 x
4
2

2
2
xy
x y
1 + 1
(45)
2 + x3 y .
2
4
We will now investigate the fixed point set of this Killing vector following the discussion in
[24]. Neglecting zeros of V (), zeros |k|2 are given by zeros of all components k m . Since
the Killing vector depends only on x and y the equations k m = 0 are not solvable for a
generic choice of m . We can solve, however, k x = k y = 0 and find as solution [24]
x = x =

22
3 || ,
2
+ 2
12
y = y =

21
3 || ,
2
+ 2
12
(46)
where ||2 = 12 + 22 + 32 . If one inserts these values into k = 0 one obtains a constraint
on 4 , but one would obtain bolts since and are arbitrary. In order to find a NUT we
171
keep 4 arbitrary and set instead

V () = 0
or = (n, )
(47)
which ensures that g k k = 0 and represent points in the (, ) space.1 The point = n
is special, because without fixing x, y we find |k|2 0.
Therefore, we have on the quaternionic space the following NUT fixed points of the
Killing vector k

> 0: (, x, y) = (n, x, y), ( , x , y )(+ , x , y )if >0 ,

< 0: (, x, y) = max[n, + ], x , y , ( , x , y )
(48)
with = n 4n2 + .
The dimensional reduction along the Killing vector (45) will yield a bound state of D6branes, located on the quaternionic space at these fixed points. The number depends on the
choice of parameters, but can be at most five (for , > 0) where two of them (at = )
are disconnected from the other three (at = n and = + ) by the curvature singularity
at = n. Recall, there are different (physical) regions for (see Fig. 2) and hence some
fixed points may not have a physical sensible interpretation. The correct identification of
the D6-branes is, however, a subtle point. E.g., because x+ x + y+ y = 4 the two points
(x+ , y+ ) and (x , y ) are antipodal points in the (x, y)-space (i.e., for = 1 on the S2 )
and one may be tempted to identify the two branes. But note, since the Killing vector is
zero at both points and non-singular in between, the eigenvalue of dk have to be different
at both fixed points, which implies that the 6-brane charge or tension will be different. It
would be interesting to see whether our model corresponds to one of the supersymmetric
intersecting brane worlds or the relation to the non-supersymmetric ones; see, e.g., [39,40]
and references therein.
A subtle question concerns also the dilaton. In adapted coordinates one can write the
7-metric as in Eq. (40), where x11 correspond to and is proportional to the dilaton.
Unfortunately, e2 |k|2 , which becomes the string coupling constant, is not bounded for
our solution. While for > 0 the range of is bounded, see Fig. 2, but |u| is unbounded,
see Eq. (18). On the other hand for a non-compact quaternionic space, with 0, is
situation is opposite. Therefore, in any case there are asymptotic regions where |k|2 e2
blows up and one should look for similar deformation as the ones discussed in [11] to
obtain a finite string coupling constant.
5. Outlook: membrane instantons and new Spin(7) metric

We mentioned already a few aspects that deserve further investigations, but let us add
another two.
1 Notice, V = 0 is a conical singularity which is resolved by the proper periodicity in .
172
5.1. Resolution of the curvature singularity by Membrane instantons

One genuine feature of our G2 manifold is the appearance of a curvature singularity;
otherwise the continues deformation of two topological different spaces (like S4 and CP 2 )
would be difficult to understand. Although for < 0 the singularity is inside an unphysical
coordinate range (with timelike coordinates), the validity of the supergravity solution is
a subtle question. One may therefore look for possibilities to resolve the singularity. An
interesting possibility could be to turn on appropriate fluxes along the lines discussed in
[41,42]. But non-trivial fluxes play an important role also in a somewhat different aspect.
Namely the curvature singularity in Eq. (37) disappears if + 4n2 = 0 (or = + ) and
since is the constant curvature of the 4d quaternionic space, this value can effectively
be changed by turning on 4-form fluxes. In fact, this possibility has been discussed
as neutralization of the cosmological constant and corresponds to take into account
membrane instanton effects [43,44]. What will happen with the quaternionic space in the
limit + = ? Setting = 1 and = 1 so that n = 1/2 we transform = 12 cosh R and
the metric becomes
ds 2 = dR 2 + sinh2 R d3
(49)
which is one parameterization of Euclidean AdS4 space. This metric covers only the region
> n, but for < n one obtains the same metric so that both physical regions are
decoupled and the region with the curvature singularity disappeared and moreover all
fixed points are joined at = n which corresponds to the point R = 0 where the metric
becomes flat. But note, the 4-form flux will not only change the effective value of the
curvature of the quaternionic space, it will also cause a back reaction on the 7-metric which
requires a detailed analysis. Needless to say, that it would be also interesting to calculate
the superpotential caused by these 4-form fluxes [13,45,46].
There is also another effect that may weaken or resolve the curvature singularity, namely
to change effectively n by a multi-center solution (as for Taub-NUT or the EguchiHanson
space). It is unclear whether these multi-center solution for quaternionic spaces exist, but
one could try to construct them similar to the multi-center supergravity solutions of [47,48].
5.2. New Spin(7) metrics
We have considered here only generalizations of known 7-manifolds with G2 holonomy,
but analogous calculations will also yield more general 8-manifolds with Spin(7)
holonomy. If one follows again the work by Bryant and Salamon, the corresponding metric
of the 8-manifold becomes
ds 2 = f 2 + g 2 em em ,
(50)
is again the metric of a quaternionic space and = du uA, where we used

where
the quaternionic notation u = u0 + iu1 + j u2 + ku3 and A = iA1 + j A2 + kA3 is the
SU(2) connection of the quaternionic space. The function f = f (|u|) and g = g(|u|) are
again the same as in [3] and new fixed points are again encoded by the quaternionic space.
Also, one may turn on fluxes and calculate the superpotential or replace the R4 , spanned
by the coordinates um , by a hyper-Kaehler space and finds a IIA description of 6-branes
em em
173
wrapping the quaternionic space. Some related work in these directions has been done,
e.g., in [4952].
Acknowledgements
We would like to thank R. Blumenhagen, G. DallAgata, S. Gukov and A. Klemm for
helpful discussions. This work is supported by a Heisenberg grant of the DFG.
References
[1] M.J. Duff, B.E.W. Nilsson, C.N. Pope, KaluzaKlein supergravity, Phys. Rep. 130 (1986) 1142.
[2] G. Papadopoulos, P.K. Townsend, Compactification of D = 11 supergravity on spaces of exceptional
holonomy, Phys. Lett. B 357 (1995) 300306, hep-th/9506150.
[3] R.L. Bryant, S.M. Salamon, On the construction of some complete metrics with exceptional holonomy,
Duke Math. J. 58 (1989) 829.
[4] G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S 3 , R 3 and R 4 bundles, Commun. Math.
Phys. 127 (1990) 529.
[5] E. Witten, Fermion quantum numbers in KaluzaKlein theory, PRINT-83-1056 (Princeton), October 1983,
in: Proceedings, Quantum Field Theory and the Fundamental Problems of Physics, Shelter Island, 1983,
pp. 227277.
[6] B.S. Acharya, On realising N = 1 super-YangMills in M-theory, hep-th/0011089.
[7] B. Acharya, E. Witten, Chiral fermions from manifolds of G(2) holonomy, hep-th/0109152.
[8] M. Atiyah, E. Witten, M-theory dynamics on a manifold of G(2) holonomy, hep-th/0107177.
[9] M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large-N duality, J. Math. Phys. 42 (2001) 3209
3220, hep-th/0011256.
[10] M. Cvetic, H. Lu, C.N. Pope, Massless 3-branes in M-theory, Nucl. Phys. B 613 (2001) 167188, hepth/0105096.
[11] A. Brandhuber, J. Gomis, S.S. Gubser, S. Gukov, Gauge theory at large-N and new G(2) holonomy metrics,
Nucl. Phys. B 611 (2001) 179204, hep-th/0106034.
[12] M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Cohomogeneity one manifolds of Spin(7) and G(2) holonomy,
hep-th/0108245.
[13] S. Gukov, D. Tong, D-brane probes of special holonomy manifolds, and dynamics of N = 1 threedimensional gauge theories, hep-th/0202126.
[14] T. Friedmann, On the quantum moduli space of M-theory compactifications, hep-th/0203256.
[15] H. Ita, Y. Oz, T. Sakai, Comments on M-theory dynamics on G(2) holonomy manifolds, hep-th/0203052.
[16] R. Blumenhagen, V. Braun, Superconformal field theories for compact G(2) manifolds, JHEP 12 (2001) 006,
hep-th/0110232.
[17] J. Wolf, Complex homogenous contact manifolds and quaternionic symmetric spaces, J. Math. Mech. 14
(1965) 1033.
[18] J. Bagger, E. Witten, Matter couplings in N = 2 supergravity, Nucl. Phys. B 222 (1983) 1.
[19] B. de Wit, P.G. Lauwers, A. Van Proeyen, Lagrangians of N = 2 supergravitymatter systems, Nucl. Phys.
B 255 (1985) 569.
[20] B. de Wit, M. Rocek, S. Vandoren, Hypermultiplets, hyper-Kaehler cones and quaternion-Kaehler geometry,
JHEP 02 (2001) 039, hep-th/0101161.
[21] R. DAuria, S. Ferrara, On fermion masses, gradient flows and potential in supersymmetric theories,
JHEP 05 (2001) 034, hep-th/0103153.
[22] R. Kallosh, A.D. Linde, Supersymmetry and the brane world, JHEP 02 (2000) 005, hep-th/0001071.
[23] K. Behrndt, M. Cvetic, Anti-deSitter vacua of gauged supergravities with 8 supercharges, Phys. Rev. D 61
(2000) 101901, hep-th/0001159.
174
[24] K. Behrndt, G. DallAgata, Vacua of N = 2 gauged supergravity derived from non-homogeneous

quaternionic spaces, hep-th/0112136.
[25] P.K. Townsend, The eleven-dimensional supermembrane revisited, Phys. Lett. B 350 (1995) 184187, hepth/9501068.
[26] D.V. Alekseevsky, V. Cortes, C. Devchand, A. Van Proeyen, Flows on quaternionic-Kaehler and very special
real manifolds, hep-th/0109094.
[27] B. de Wit, H. Nicolai, The parallelizing S(7) torsion in gauged N = 8 supergravity, Nucl. Phys. B 231 (1984)
506.
[28] A. Bilal, J.-P. Derendinger, K. Sfetsos, (Weak) G(2) holonomy from self-duality, flux and supersymmetry,
hep-th/0111274.
[29] M. Gunaydin, F. Gursey, Quark structure and octonions, J. Math. Phys. 14 (1973) 16511667.
[30] S. Fubini, H. Nicolai, The octonionic instanton, Phys. Lett. B 155 (1985) 369.
[31] B. de Wit, H. Nicolai, N.P. Warner, The embedding of gauged N = 8 supergravity into D = 11 supergravity,
Nucl. Phys. B 255 (1985) 29.
[32] D. Alekseevsky, Classification of quaternionic spaces with a transitive solvable group of motions, Math.
USSR Izv. 9 (1975) 297.
[33] K. Behrndt, M. Cvetic, Gauging of N = 2 supergravity hypermultiplet and novel renormalization group
flows, Nucl. Phys. B 609 (2001) 183192, hep-th/0101007.
[34] P.-Y. Casteill, E. Ivanov, G. Valent, Quaternionic extension of the double Taub-NUT metric, Phys. Lett.
B 508 (2001) 354364, hep-th/0104078.
[35] P.-Y. Casteill, E. Ivanov, G. Valent, U (1) U (1) quaternionic metrics from harmonic superspace, Nucl.
Phys. B 627 (2002) 403444, hep-th/0110280.
[36] D. Kramer, E. Herlt, M. MacCallum, H. Stephani, Exact solutions of Einsteins field equations, Cambridge
Univ. Press, 1979.
[37] G.W. Gibbons, S.W. Hawking, Classification of gravitational instanton symmetries, Commun. Math.
Phys. 66 (1979) 291.
[38] P.K. Townsend, Black holes, gr-qc/9707012.
[39] M. Cvetic, G. Shiu, A.M. Uranga, Chiral four-dimensional N = 1 supersymmetric type IIA orientifolds from
intersecting D6-branes, Nucl. Phys. B 615 (2001) 332, hep-th/0107166.
[40] R. Blumenhagen, B. Koers, D. Lst, T. Ott, The standard model from stable intersecting brane world
orbifolds, Nucl. Phys. B 616 (2001) 333, hep-th/0107138.
[41] M. Cvetic, G.W. Gibbons, H. Lu, C.N. Pope, Ricci-flat metrics, harmonic forms and brane resolutions, hepth/0012011.
[42] M. Cvetic, H. Lu, C.N. Pope, Brane resolution through transgression, Nucl. Phys. B 600 (2001) 103132,
hep-th/0011023.
[43] J.D. Brown, C. Teitelboim, Neutralization of the cosmological constant by membrane creation, Nucl. Phys.
B 297 (1988) 787836.
[44] R. Bousso, J. Polchinski, Quantization of four-form fluxes and dynamical neutralization of the cosmological
constant, JHEP 06 (2000) 006, hep-th/0004134.
[45] B. Acharya, X. de la Ossa, S. Gukov, G-flux, supersymmetry and Spin(7) manifolds, hep-th/0201227.
[46] C. Beasley, E. Witten, A note on fluxes and superpotentials in M-theory compactifications on manifolds of
G(2) holonomy, hep-th/0203061.
[47] L.A.J. London, Arbitrary dimensional cosmological multi-black holes, Nucl. Phys. B 434 (1995) 709735.
[48] J.T. Liu, W.A. Sabra, Multi-centered black holes in gauged D = 5 supergravity, Phys. Lett. B 498 (2001)
123130, hep-th/0010025.
[49] R. Hernandez, Branes wrapped on coassociative cycles, Phys. Lett. B 521 (2001) 371375, hep-th/0106055.
[50] G. Curio, B. Koers, D. Lst, Fluxes and branes in type II vacua and M-theory geometry with G(2) and
Spin(7) holonomy, hep-th/0111165.
[51] B. Acharya, X. de la Ossa, S. Gukov, G-flux, supersymmetry and Spin(7) manifolds, hep-th/0201227.
[52] R. Hernandez, K. Sfetsos, An eight-dimensional approach to G(2) manifolds, hep-th/0202135.

D0-branes in flux 5-brane backgrounds

Joan Simn
The Weizmann Institute of Science, Department of Particle Physics Herzl Street 2, 76100 Rehovot, Israel
Received 28 January 2002; received in revised form 6 May 2002; accepted 15 May 2002
Abstract
The existence of supersymmetric D0-branes sitting at an arbitrary distance from a flux 5-brane is
proven. The physical picture in type IIA is consistent with the KaluzaKlein reduction origin of the
flux 5-brane. An analysis of the fluctuations around these vacua is performed and the backreaction of
the D0-branes is computed. Non-threshold D0D2 bound states and similar stabilization mechanisms
for D2-branes in such backgrounds are also briefly discussed. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 11.15.kc; 11.30.Pb; 04.65.+e
Keywords: D-branes; Flux branes; Stabilization
1. Introduction
There has recently been a lot of interest in the construction and classification of
supersymmetric fluxbranes in string theory, either from a supergravity perspective [1,
2] or from a conformal field theory (CFT) one [3,4]. It is natural to ask about which
D-branes, NS-branes and other dynamical objects in string theory exist in this new sector
of the theory [5]. Some results in this direction were obtained using CFT and/or worldvolume techniques in [6,7]. In this note, we shall use low energy field theory descriptions
both in the open and closed string sectors to answer this question, and for simplicity, we
shall concentrate on D0-branes in flux 5-branes (F5-branes) [1].
A F5-brane is a type IIA configuration preserving one half of the spacetime supersymmetry (see Appendix A) with metric

g = 1/2 ds 2 E1,5 + dr 2 + r 2 d 2 + sin2 2 d 2
E-mail address: jsimon@wicc.weizmann.ac.il (J. Simn).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 0 2 - 9
176
J. Simn / Nuclear Physics B 635 (2002) 175191
+ r 2 1/2 {d + cos 2 d}2 ,
(1)
and non-trivial dilaton and RamondRamond (RR) 1-form

3
log ,
4
C(1) = (r)r1 (d + cos 2 d),
0 =
(2)
(3)
where the scalar function is defined by = 1 + (r)2 . The above configuration is

obtained by KaluzaKlein reduction of eleven-dimensional Minkowski spacetime along
the orbits of the Killing vector

F5 = x + x 6 7 x 7 6 + x 8 9 x 9 8 ,
where x = R stands for the eleventh compact dimension and for some canonically
normalised angular variable. Equivalently, the F5-brane is obtained in the limit1
R 0,
fixed.
(4)
2/3
gs
,
the type IIA description of the geometry is no longer

Notice that whenever r0
reliable, and one should replace it by its M-theory lift. As explained in [1], one can
associate some notion of charge to this object by computing the integral F(2) F(2) ,
on the full R4 transverse space to the F5-brane, F(2) being the corresponding RR 2-form
field strength. This transverse R4 space is parametrised by
x 6 + ix 7 = r cos ei(+) ,
x 8 + ix 9 = r sin ei() .
(5)
Thus, r stands for the radial distance to the F5-brane, measures the angle between the 67plane and the 89-plane, whereas are the polar angles (phases) in the corresponding
2-planes. The charge (proportional to ( 2 )1 ) is thus kept fixed in the KaluzaKlein
reduction (4).
Assuming the existence of D0-branes in this background at weak coupling, and taking
a strong coupling limit, the system should be described by an M-wave [8] propagating
in a locally flat spacetime satisfying topologically non-trivial identification conditions.2 It
is thus expected that the KaluzaKlein reduction of an M-wave configuration along the
orbits of the Killing vector F5 should give rise to a type IIA configuration describing a
composite system of D0-branes and F5-branes. One of the purposes of this note is to check
this interpretation.
In order to address these considerations, one can study the dynamics at low energy of
a single D0-brane in such an F5-brane background, by a probe analysis. If a D0-brane of
tension (mass) TD0 = l1s e0 sits in a F5-brane background at a distance r0 , the potential
felt by the probe

1/2
,
V = TD0 1 + 3(r)21
(6)
0
depends on the distance. In (6), 0 was defined as (r0 ). It is only at r0 = 0, the minimum
of the potential (6), that the probe feels no force. Equivalently, the classical equations
1 The author would like to thank H. Robins for discussions on this point.
2 Similar remarks were already mentioned and used in the context of M5-branes in the appendix of [4].
177
of motion for a massive particle propagating in a F5-brane background with no angular

velocity are only satisfied when motion is constrained to the hypersurface defining the
F5-brane.
It is a natural question whether there exists some physical mechanism by which the
D0-brane remains static3 at an arbitrary distance r0 from the F5-brane. The answer to this
question lies on the geometry of the background. Notice that even if the angular velocity
of the D0-brane vanishes, the RR 1-form induces some angular momentum through the
WessZumino coupling, yielding a non-static configuration. This is the reason why (6)
is not equal to minus the Lagrangian density, but includes the contribution from such
non-vanishing momenta. This argument raises the possibility of giving some angular
velocity to the D0-brane such that the total angular momentum vanishes, yielding an static
configuration of constant energy (independent of r0 ). This expectation will be confirmed
both for a single D0-brane in type IIA and a massless particle in eleven dimensions (which
is the lift to M-theory of the type IIA configuration).
The plan of the paper is as follows. In Section 2, we shall study the propagation of
a massless eleven-dimensional particle in a flat eleven-dimensional geometry satisfying
some non-trivial identification conditions, by using an adapted coordinate system to
the action of F5 . In Section 3, we shall move to the D0-brane setting and prove, by
construction, the existence of the aforementioned static configurations of finite energy.
Evidence is given in favour of their stability by an analysis of the fluctuations around these
vacua and due to the existence of a bound from below on their energy. In Section 4, the
supergravity type IIA configuration taking into the account the backreaction of a bunch
of D0-branes is found. It is then shown that an additional D0-brane probe having the
same angular velocity as in our previous vacua can be added at an arbitrary distance, thus
showing its BPS character, and confirming the physical interpretation of these supergravity
backgrounds. The same physical effect is used to construct D0D2 bound states in
Section 5 and to (meta)-stabilize curved D2-branes in F5-brane backgrounds in Section 6.
2. Eleven-dimensional massless particle

Let us study the propagation of an eleven-dimensional massless particle,

1
S = d L = d x M PM vhMN PM PN ,
2
(7)
in the eleven-dimensional geometry

h = ds 2 E(1,5) + dr 2 + r 2 d 2 + sin2 2 d 2 + R 2 d 2
+ 2r 2 R d(d + cos 2 d) + r 2 (d + cos 2 d)2 .
(8)
In (7), PM , M = 0, 1, . . . , , should be thought of some set of auxiliary fields, actually, the

conjugate momentum variables to the spacetime coordinates x M ; whereas v ensures the
mass-shell condition. The eleven-dimensional metric (8) was already written in adapted
3 In this note, by static, we shall refer to a configuration having vanishing angular momentum.
178
coordinates, where F5 = x . Thus, by KaluzaKlein reduction along the x = R

direction, the full F5-brane background is obtained.
Its classical equations of motion are given by,
PM = hMN x N v 1 ,
hMN PM PN = 0,
L
d PM
= M,
(9)
d
x
where v can always be set to one (using worldline reparametrisations) and the inverse of
the background metric hMN is
0 0
0
0
0 1 0
h1 =
, = 0, 1, . . . , 5,
0
0 0 r 2
1
0 0 0 h

1
R 2
0
R

1
2
1
2
2
h = R
1 + r sin 2 cos 2
0
cos 2
where the scalar function = (r sin 2 )2 was introduced and h 1 is written in the basis
{ , , }.
It will be assumed that all momenta is vanishing except for P , P and P , which
are conserved quantities, this being consistent with (9) and equivalent to setting r = =
x i = 0, i = 1, . . . , 5. The energy of the configuration is given by the mass-shell condition

2
2

P
1
2
+ P + 2 2
(P cos 20 P )2 + P r01
E =
(10)
R
r0 sin 20
and depends on the constant parameters r0 , 0 . Requiring the energy to be minimized
enforces an stationary requirement in the , angular directions
P = P = 0,
where the solution r0 was excluded, since we are interested in finite energy
configurations.
Using (9), we can solve the stationary conditions in terms of the velocities

r 2 + cos 2 + R cos 2 = 0,
r 2 + cos 2 + R = 0,

R 2 + Rr 2 + cos 2 = P ,
(11)
allowing three different solutions
r0 = 0, corresponding to a massless particle constrained to R5 S 1 .
= 0, = R 1 P = P , r0 .
0 = 0(/2), + () = P , r0 , corresponding to a massless particle moving in
the 67-plane (89-plane).
In all cases, the energy of the particle equals its velocity
E = x ,
179
along the compact direction, since P = R 2 = R x , thus suggesting the latter configurations do describe a massless particle propagating in the compact direction x . This can
explicitly be shown by mapping the trajectories in both the adapted coordinate system and
the one in which the metric is manifestly flat. The latter is governed by
= + x .
Thus, whenever = 0, there is no angular velocity in ( = 0); whenever 0 = 0(/2),
motion is constrained to a plane, thus the real physical phase variable is given by
+ cos 20 , which again has vanishing angular velocity; for r0 = 0, no motion is allowed
in the directions under consideration. Thus, indeed, all the found trajectories do correspond
to propagation in the compact direction, as the energy computation was suggesting to us.
All configurations described in this section preserve one quarter of the spacetime supersymmetry. This is obvious in the coordinate system in which the background is manifestly flat. Indeed, in that frame, the background preserves the supersymmetries satisfying
6789 = , whereas the propagating massless particle the ones obeying 0 = . Since
[6789 , 0 ] = tr 67890 = 0, the configuration preserves eight supercharges.
3. Static D0-brane in a F5-brane background
Once the motion of massless eleven-dimensional particle in the adapted coordinate
system has been worked out, the extension to the D0-brane is rather natural. One should
look for an angular velocity such that the full angular momentum of the massive particle
vanishes. We shall consider the most general ansatz consistent with this picture, in which
x 0 = (gauge choice), the coordinates parametrising the F5-branes worldspace are set to
a constant value (x0i ), the radial distance to the F5-brane (r0 ) and = 0 are constants and
the angular coordinates that do couple with the RR 1-form (3) are assumed to be
= 0 + 0 ,
= v0 + 0 .
(12)
The action describing the dynamics of this system at low energies and weak string coupling
constant is given by

0 r02 (0 + v0 cos 20)2 0 (r0 v0 sin 20 )2
SD0 = TD0 d 1
0

2
+ TD0 d 1
(13)
0 r0 (0 + v0 cos 20 ).
The non-trivial components of the angular momentum can be derived from (13) by standard
methods. In particular, computing L/0 one obtains

l0
2 1
P = TD0 r0 0 +
(14)
0 (r0 l0 )2 0 (r0 v0 sin 20 )2
and from L/v0

l0 cos 20 + 0 r02 sin2 20 v0

2 1
,
P = TD0 r0 0 cos 20 +
(15)
0 (r0 l0 )2 0 (r0 v0 sin 20 )2
180
where the constant parameter, l0 = 0 + v0 cos 20 , has been introduced.

When one requires both equation (14) and Eq. (15) to vanish, one finds three different
solutions to the corresponding system of equations:
r0 = 0, corresponding to a D0-brane restricted to move on the F5-brane;
0 = 0, /2, 0 v0 = r0 , corresponding to a D0-brane moving either in the 67plane (0 = 0) or in the 89-plane (0 = /2) with total angular velocity , in both
cases;
v0 = 0, 0 = 0 , r0 .
It is reaffirming to get exactly the same conditions we derived before. Since, by
construction, the angular momentum vanishes, the energy of the configuration equals
minus the Lagrangian density of the system (E = L), that is, it equals the tension (mass)
of the D0-brane. Thus, there exist configurations of static D0-branes sitting at any distance
r0 from the F5-branes, with non-vanishing angular velocity and finite energy equal to its
mass. This interpretation can also be derived from a pure Hamiltonian analysis of the
system. Using the general formalism developed in [9], the energy density can be computed
from the mass-shell condition
3/2
2
0
= 0,
P 2 + TD0
(16)
where P 2 = g MN (PM + TD0 CM )(PN + TD0 CN ), g MN being the inverse of the metric
(1) and CM the components of the RR 1-form (3). Notice that by requiring the spacelike
components of the momentum (Pm = 0, m = 1, . . . , 9) to vanish as corresponds to a static
configuration, the energy (P0 = E) equals the tension (mass) of the D0-brane.
Given a bosonic configuration on the world-volume of any brane, one way to check
whether it preserves some supersymmetry is by analysing whether it satisfies the kappa
symmetry preserving condition [10]
= ,
(17)
where is the Killing spinor of the corresponding background geometry and is the usual
field and background dependent matrix encoding the kappa symmetry transformations in
D-branes [11]. Eq. (17) reduces, in this particular case, to

1/2
1/2
0 + 0 r0 l0 + 0 r0 v0 sin 20 11 = .
(18)
Thus, for r0 = 0, Eq. (18) reduces to
0 11 =
(r0 = 0),
whereas in the remaining two static configurations, (17) reduces to

1/2
0 0 (r0 ) 11 = (l0 = ),
which corresponds to the supersymmetry projection condition of a boosted D0-brane.
As it is argued in Appendix A, the point dependence of the Killing spinors is of the
] is some matrix,
]0 , where 0 stands for a constant spinor and M[x;
form = M[x;
depending on the point, built from flat gamma matrices. Under such conditions, the analysis
181
of (18) is not straightforward, as has already been pointed out in the past in different
backgrounds [12].
An explicit computation involving the exact form for the Killing spinors should prove
that our static configurations preserve one fourth of the spacetime supersymmetry. To
give further evidence to the last statement, we shall first study the stability of these
configurations by an analysis of the fluctuations around them and the existence of a bound
from below for the energy, and shall postpone the discussion on the BPS character of these
configurations to the next section. Let us parametrise these arbitrary fluctuations by
x i = x0i + x i ( ),
r = r0 + r( ),
= 0 + ( ),
= 0 + 0 + ( ),
= v0 + 0 + ( ),
(19)
where will be the parameter counting the different orders in the expansion of the action
around the vacuum configuration,

SD0 = S(0) + S(1) + 2 S(2) + O 3 .
Let us study the different contributions to the expansion of the action SD0 by splitting the
latter into a DiracBornInfeld (DBI) part and a WessZumino (WZ), SD0 = SDBI + SWZ .
The contribution from the WZ term is given by
2
2

r0
r0
r02
l0 +
l0 r
SWZ = TD0 d
+ cos 20 + 2
0
0
0

r 2
+ 2 2 0 v0 cos 20 2 + sin 20
0

r0
+ 3 l0 1 3(r0 )2 r 2 + 2 2 r + cos 20
, (20)
0
0
where we already took into account that r0 v0 sin 20 = 0 for any of the three classical
solutions found previously.
In order to write the expansion of the DBI term, we shall distinguish between the
r0 = 0 classical solution and the ones involving an arbitrary r0 . Notice that for r0 = 0,
the contribution from the WZ term (20) vanishes and one is left with

1
1
1
SD0 = TD0 d + 2 TD0 d x i x j ij + r 2 + ()2 r 2
2
2
2
3
+O .
(21)
The zeroth order contribution in the parameter reproduces the energy of the vacuum
configuration (up to a sign), as it should, whereas the first order contribution vanishes, in
agreement with the fact that the classical configuration which we are expanding around
solves the classical equations of motion. The second order contribution describes the
dynamics of five massless scalar fields (the fluctuations along the five spacelike directions
along the F5-brane) plus a radion r, whose mass term has the wrong sign.
182
Let us now concentrate on the remaining two cases (r0 = 0). The contribution from the
DBI term can be summarized as

1 + 12 L1 x1
SDBI = TD0 d 1
0

+ 2 12 L2 14 L21 + x12 x2 12 x1 L1 ,
(22)
2 r0
2 2
r,
x2 =
r ,
0
0

L1 = 2(r0 )r0 + cos 20 ,

L2 = 4(r0 )r + cos 20 0 x i x j ij 0 r 2 0 r02 2

2
4(r0 )r0 v0 cos 20 2 + sin 20 r02 + cos 20

0 r02 sin2 20 2 0 v02 4r02 cos 40 2 + sin2 20r 2 .
x1 = 2
(23)
Joining the contributions from (20) and (22), we recover the zeroth order contribution
corresponding to the vacuum energy. At linear order in the expansion parameter , the
action

2 r0
2
r
x
S(1) = TD0 d 1
(24)
1
0
0
vanishes, as can be seen by the definition of x1 in (23). Finally, at second order, the action
reads as follows

2

S(2) = TD0 d 12 x i x j ij + 12 r 2 + 12 r02 2 + 12 r02 + cos 20

+ 12 r02 sin2 20 2 + 2r02 v02 cos 40 2 + 12 v02 sin2 20 r 2 . (25)
The above action describes a free scalar field theory in which only two mass like terms
appear with the wrong sign. Notice that v0 sin 20 is always vanishing for the vacua we are
discussing, thus radial fluctuations are both classically stable and massless. Furthermore,
when v0 = 0, the mass term for fluctuations vanish, thus we conclude that the static
classical configuration defined by v0 = 0 and 0 = is a stable configuration in which
all fluctuations {x i , r, , r0 ( + cos 20), r0 sin 20 } are massless, as it should be
for a D0-brane with no constraints on its motion.
Finally, for the classical configuration confined either to the 67-plane (0 = 0) or to
the 89-plane (0 = /2) the wrong sign mass like term for the fluctuation remains.
Notice that in this case, r0 sin 20 is also non-dynamical since the only physical degree
of freedom corresponds to fluctuations of the phase r0 ( ) in the plane where the
classical motion takes place.
Despite the apparent tachyonic behaviour of the radion r for the r0 = 0 vacuum and
for the 0 = 0(/2) vacua, one should check that the energy of these fluctuations is
negative, to consider them as real tachyonic modes. Actually, such a possibility is not
2 + TD0 3/2 = 0 (in general),
possible. Indeed, by solving the mass-shell condition P
one can express the energy of the system as

P 2
1
E 2 = (TD0 + P )2 +
+
(P cos 2 P )2
r
r 2 sin2 2
2
5
P
+ Pr2 +
+
(Pi )2 .
r
183
(26)
i=1
Thus, energy is bounded from below by TD0 , a bound which is saturated when the
configuration is static (P = P = Pr = P = Pi = 0 i). Due to the positivity of all
the terms in (26), any fluctuation around our vacua would give rise to a positive energy
contribution, thus showing the stability of all configurations discussed before.
4. Background reaction
If the above configuration is BPS, one could consider a set of N D0-branes sitting on the
same point in the presence of a F5-brane. For large enough N , such a configuration should
admit a reliable supergravity description. Once this background is known, one could probe
it with an additional D0-brane and the BPS feature should manifest itself with the existence
of classical configurations having the same features as the ones described in the previous
section, and allowing the additional D0-brane to sit at an arbitrary distance not only of
the F5-brane but also of the remaining D0-branes being described by the background. In
the following, we shall first construct the closed string description of N D0-branes in the
presence of a F5-brane, and afterwards, it will be checked that indeed an additional D0brane can sit at an arbitrary distance when having the correct angular velocity.
4.1. Supergravity background
The easiest way to look for the type IIA configuration we are interested in is to take the
strong coupling limit of the theory (R ), and look for the corresponding configuration
in M-theory.4 This eleven-dimensional configuration should be an M-wave propagating
in the compact direction, plus some topologically non-trivial conditions describing the
presence of the F5-brane. Thus, locally, the metric should look like the one of an
M-wave [8]
2
2

h = (U 2) dx 0 + U dx 2(U 1) dx 0 dx + ds 2 E9 ,
(27)
where x stands for the direction of propagation and U = U (r ) is an harmonic function
U = U (r ) = 1 +
k
,
r 7
on E9 . Notice that r 2 = r 2 + x i x j ij , where r is the radial coordinate transverse to the

F5-brane and x i (i = 1, . . . , 5) parametrise its worldspace. The configuration preserves
4 This remark was already mentioned in the appendix of [4].
184
one fourth of the supersymmetry. This is because the M-wave configuration is already a
one-half BPS object, whereas the non-trivial identifications that characterise the global
properties of this configuration break another half.
If one KaluzaKlein reduces the above configuration along the orbits of the Killing
vector F5 = x + (x 6 7 x 7 6 + x 8 9 x 9 8 ),5 one obtains a type IIA configuration
with metric

2
g = 1/2U 1/2 (U 1)2 dx 0 + 1/2 U 1/2 r 2 (d + cos 2 d)2
+ 2r 21/2 U 1/2 (U 1) dx 0(d + cos 2 d)

2

+ 1/2 U 1/2 (U 2) dx 0 + U 1/2 ds 2 E5

+ U 1/2 dr 2 + r 2 d 2 + sin2 2 d 2 .
RR 1-form potential C(1)

C(1) = 1 1 U 1 dx 0 + U 1 r 2 (d + cos 2 d) ,
(28)
(29)
and a non-vanishing dilaton

3
log(U ),
4
where the scalar function is defined as
=
(30)
= 1 + U 1 2 r 2 .
Notice that this is a type IIA background that correctly reproduces the F5-brane limit,
the one in which the charge of the D0-brane is sent to zero (k 0), and the D0-brane
limit, in which the charge of the F5-brane is sent to zero ( 0). In both limits, there
is an enhancement of supersymmetry. Furthermore, the ten-dimensional metric (28) takes
into account the backreaction of the boosted D0-branes by having non-vanishing g0 and
g0 components.
4.2. Probe computation
We shall now explicitly show that exactly the same static configurations that were found
in Section 3 are still static configurations in the above background, thus showing their BPS
character [13]. Notice that this computation also gives further evidence for the physical
interpretation given to the type IIA configuration obtained in the last subsection.
Using the same ansatz as in Section 3, the non-trivial angular momenta is given by
e
x m gm + TD0 C
P = TD0 m n
x x gmn
(31)
e
x m gm + TD0 C .
P = TD0 m n
x x gmn
(32)
and
5 This is just a particular case of a more general formulation discussed in [5].
185
Requiring the configuration to be static (P = P = 0), determines three inequivalent

solutions. It is life-reaffirming to check that these are precisely the ones found in
our previous analysis: r0 = 0, corresponding to a D0-brane moving on the F5-brane,
sin 20 = 0, corresponding to a D0-brane moving in the 67-plane or in the 89-plane and
= 0. As in that case, the angular velocity in the last two solutions is
0 + cos 20 v0 = .
Since, by construction, the configuration has vanishing momentum, its energy equals minus
the value of the Lagrangian density evaluated on it. Again, it is satisfactory to check that
E = TD0 .
This result can be confirmed by solving, after some algebra, the mass shell condition for E
g 00 (E + TD0 C0 )2 + 2g 0 (E + TD0 C0 )TD0 C + 2g 0 (E + TD0 C0 )TD0 C
2
2
2
2
+ TD0
U 3/2 3/2 = 0.
g C + g C2 + 2g C C + TD0
5. Static D0D2 bound state

In a flat and topologically trivial spacetime, either D0-branes or D2-branes can
be located anywhere, the resulting configurations being stable and supersymmetric.
Furthermore, non-threshold D0D2 bound states preserving one half of the spacetime
supersymmetry are known to exist. These can be realized on the D2-brane effective action
as constant magnetic fluxes. We have learnt in the previous section how to describe
static, supersymmetric D0-branes in F5-brane backgrounds. It is natural to ask whether
the forementioned bound states are also allowed in this new F5-brane backgrounds.
First of all, just as for the flat spacetime, one needs to identify the right vacuum state
describing a D2-brane. It is easy to show that a D2-brane parallel to the F5-brane sitting a
distance r0 from it, feels no force. In other words, the potential is constant and equals the
tension of the brane. It preserves one fourth of the spacetime supersymmetry.
It is natural to guess that the description of a D0D2 bound state in a F5-brane
background involves switching on some constant magnetic flux on the brane, F =
2 F12 , and giving some angular velocity to the system ( = + 0 ), such that the total
angular momentum P vanishes, to compensate for the effect of the non-trivial background
geometry.6 Using the static gauge, setting the rest of transverse coordinates to constant
values and the electric components of the gauge field to a pure gauge configuration, the
effective action description the D2-brane at low energies and weak string coupling constant
is

S = TD2 d 2+1 1
(33)
0 (r)2 0 + F 2 r 2 F .
0
6 We decided to concentrate on this description for the D0-brane, but the other two static configurations
described earlier in this note would lead to exactly the same picture found here.
186
The physical requirement of vanishing angular momentum fixes the angular velocity to be
P =
L
=0
F
7 =
,
1 + F2
for which the energy density of the configuration equals

E[7 ] = TD2 1 + F 2 ,
the energy density of a D0D2 non-threshold bound state [14]. We conclude that such
bound states exist in these new sectors of string theory described by F5-branes.
In the same philosophy followed in the previous section, we could compute the
backreaction of the D0D2 bound state, by taking the strong coupling limit and considering
the eleven-dimensional supergravity description of the above system. Locally, this should
be given by an M2-brane boosted in the eleventh compact dimension [15]. By Kaluza
Klein reduction along the orbits of the Killing vector F5 , one would obtain the desired
type IIA configuration.
6. Meta-stabilization of D2-branes
The existence of a linear coupling among the angular velocity and the RR 1-form was
crucial for the stabilization mechanism of D0-branes in F5-brane backgrounds to work.
D2-branes do also couple linearly to such RR 1-form through the electric components
of the gauge field strength F0a a = 1, 2. Whenever the relative orientation between the
probe and the F5-brane
is such that the above coupling is non-vanishing, the electric

L
field on the brane E a = (2 )1 F
will not vanish, even if F0a = 0. It will be
0a
examined below whether by requiring E a to vanish, the configuration is stabilized or
not. As when discussing D0-branes in F5-brane backgrounds, it is useful to interpret the
corresponding configurations in M-theory. In such a strong coupling limit, membranes
propagate in a locally flat spacetime satisfying topologically non-trivial conditions. Since
the relation among the world-volume membrane description and the world-volume D2brane description is a world-volume dualisation in 1 + 2 dimensions of the scalar field
x ( ) parametrising the eleventh dimension [16]

(1) = e det(G + F ) 7 F ,
dx + C
(2) , it is apparent that we do need no non-trivial embedding
where F = (2 )F + B

x = x ( ) in eleven dimensions to describe the corresponding D2-brane configurations.
Indeed, the source for the non-vanishing electric components of the field strength F is
(1) . We shall then be interested in those cases
entirely given in terms of the pull-back C
where such a pull-back is non-vanishing, the latter depending on the relative orientations
between the probe worldspace and the F5-brane background. In all of them, the elevendimensional configuration corresponds to a curved membrane in a locally flat spacetime.
187
6.1. =
This corresponds to a D2-brane in the -plane. When F0a = 0, the potential felt by
the probe

1/2
,
V = TD2 r02 sin 20 1 + 3(r0 )2 1
0
depends on the distance r0 to the F5-brane and the 0 angle among the 67-plane and 89plane. It will be useful to distinguish between the first factor r02 sin 20 and the second
1/2 . The first one accounts for the curved nature of the brane probe
{1 + 3(r0 )2 1
0 }
worldspace, whereas the second one encodes the non-trivial background geometry and
dilaton. Notice that the above configuration collapses to a pointlike configuration (r0 0).
Whenever F0a = 0, the effective action is given by

2 2
2
S = TD2 d 1+2 1
0 r0 r0 sin 20 0 (F cos 20 F )
0 sin2 20 F2
F = 2 F0
1/2

r0 (F cos 20 F ) ,
(34)
F = 2 F0 .
and
The physical requirement of vanishing electric
where
field on the brane determines the extremal values
F7 = 0,
F7 = (r0 )r0 sin 20 .
When one computes the energy density (E) for such configuration, it equals minus the
value of the Lagrangian density, since all momenta and electric field E a vanish. This energy
density equals

E F7 , F7 = TD2 r02 sin 20.
Thus, by switching on F7 on the brane, the physical effects due to the curved geometry,
non-vanishing dilaton and presence of RR 1-form are screened. The energy of the
configuration is the same as that for a curved D2-brane in a flat background, thus
confirming the eleven-dimensional interpretation of the configuration. As in that case, this
configuration will not be stable since there is no force that prevents the D2-brane from
contracting to a point (r0 0).
6.2. =
This corresponds to a D2-brane in the -plane. It will be checked that the same
screening effect described before happens here. We shall switch on F = 2 F0 on the
brane, such that the effective action is

2 F 2 r F .
r
r
S = TD2 d 1+2 1
(35)
0
0
0
0
0
Requiring the electric field to vanish, fixes
F7 = (r0 )r0 ,
188
for which the energy density of the configuration equals

E[F7 ] = TD2 r02 ,
the one of a curved D2-brane in a flat spacetime, as expected from the eleven-dimensional
discussion. As before, the configuration would collapse to zero size, but all effects due to
the original curved background were removed by F7 .
6.3. =
For completeness, we shall discuss a D2-brane on the -plane, even though the
conclusion is entirely analogous to the previous configurations. The effective action
describing the system when F = 2 F0 is different from zero, is given by

2 F 2 r F cos 2 ,
2 sin2 2
r
1
+
(r
)
r
S = TD2 1
(36)
0
0
0
0
0
0
0
0
and the requirement of vanishing electric field fixes
F7 = (r0 )r0 cos 20 ,
for which the energy density equals
E[F7 ] = TD2 r02 ,
which can be interpreted as in previous discussions.
Acknowledgements
The author would like to thank O. Aharony and M. Berkooz for pointing out
a mistake in the first version of this note and O. Aharony for proof reading this
manuscript. This research has been supported by a Marie Curie Fellowship of the
European Community programme Improving the Human Research Potential and the
Socio-Economic knowledge Base under the contract number HPMF-CT-2000-00480.
Appendix A. Supersymmetry
The type IIA F5-brane configuration is obtained from KaluzaKlein reduction of the
flat eleven-dimensional Minkowski space along the orbits generated by the Killing vector

F5 = x + x 6 7 x 7 6 + x 8 9 x 9 8 .
(A.1)
The Killing vector F5 acts on the constant Killing spinor via the spinorial Lie derivative
(see, e.g., [17] and also [18]), which in local coordinates is given by

LF5 = (F5 )m m + 14 [m (F5 )n] mn .
(A.2)
The condition that preserves some supersymmetry (L = 0) is equivalent to
6789 = ,
189
where it was assumed that = 0. This corresponds to having a F5-brane in the 12345plane. Thus, F5-branes preserve one half of the spacetime supersymmetry.
The above computation was done in eleven dimensions in a coordinate system in which
the eleven-dimensional geometry is manifestly flat. When dealing with kappa symmetry
and supersymmetry in type IIA/B string theories, one requires the explicit form for the corresponding Killing spinors associated with a given bosonic supersymmetric background.
In the case of F5-branes, the dilatino and gravitino supersymmetry transformations reduce
to
= m m + 38 e mn Fmn ,

1 pq
e Fpq m .
m = m + 14 m ab ab + 16
Thus, obtaining the Killing spinors is equivalent to solving = m = 0, when the
background is the F5-brane one.
It is useful to introduce an orthonormal basis
e = 1/4 dx ,
e =
r
1/4
dr,
= 0, . . . , 5,
e = 1/4 r d,
e = 1/4 r sin 2 d,
e = 1/4 r(d + cos 2 d).
In such a basis, the RR 2-form field strength F2 = dA1 is given by

F2 = 2 2 er e 3/2 e e ,
whereas the non-trivial spin connection reads as

r = r 1 5/4 + 12 (r)2 e ,
r = 12 (r)r5/4 e ,

r = r 1 5/4 + 12 (r)2 e ,
r = r 1 5/4 12 (r)2 e ,
cos 2
e ,
sin 2
= r 1 3/4 e .
= r 1 3/4 e + 2r 1 1/4
= r 1 3/4 e ,
The dilatino equation gives rise to an algebraic equation

rr 1/2 + r = 0,
this being equivalent to

r 1/2 r = .
(A.3)
On the other hand, if we consider the linear combinations ea m m = 0, the gravitino

equation can be rewritten as
m

ea m + 14 a bc bc 14 5/4 2 ra r 14 3/4 a [b c] Fbc
= 0,
(A.4)
190
where we have already used (A.3). Notice that (A.4) is trivially satisfied for a = , due to
(A.3). The remaining equations

r = 12 1 12 r + ,

= 12 r + 1/2 ( + r ) ,

( cos 2 ) = 12 sin 2 r + 1/2 ( + r ) cos 2 ,

= 12 1 1/2 r + + 1/2 rr
(A.5)
suggest the form of the Killing spinors should be
(l)
= M[r, . . . , ]l efl (r,...,) 0 ,

where 0 is some constrained constant spinor and (l) stands for appropriate antisymmetrised products of gamma matrices, whereas fl (r, . . . , ) are generically non-constant
functions just as M[r, . . . , ], with the difference that the latter might also be a matrix.
References
[1] M. Gutperle, A. Strominger, Fluxbranes in string theory, J. High Energy Phys. 06 (2001) 035, hepth/0104136.
[2] F. Dowker, J.P. Gauntlett, S.B. Giddings, G.T. Horowitz, Nucleation of p-branes and fundamental strings,
Phys. Rev. D 53 (1996) 71157128, hep-th/9512154;
C.M. Chen, D.V. Galtsov, S.A. Sharakin, Intersecting M-fluxbranes, Grav. Cosmol. 5 (1999) 4548, hepth/9908132;
M.S. Costa, M. Gutperle, The KaluzaKlein Melvin solution in M-theory, J. High Energy Phys. 03 (2002)
027, hep-th/0012072;
P.M. Saffin, Gravitating fluxbranes, Phys. Rev. D 64 (2001) 024014, gr-qc/0104014;
M.S. Costa, C.A.R. Herdeiro, L. Cornalba, Flux-branes and the dielectric effect in string theory, Nucl. Phys.
B 619 (2001) 155190, hep-th/0105023;
A.M. Uranga, Wrapped fluxbranes, hep-th/0108196;
C. Chen, D.V. Galtsov, P.M. Saffin, Supergravity fluxbranes in various dimensions, hep-th/0110164;
J.M. Figueroa-OFarrill, J. Simn, Generalised supersymmetric fluxbranes, J. High Energy Phys. 12 (2001)
011, hep-th/0110170.
[3] J.G. Russo, A.A. Tseytlin, Magnetic backgrounds and tachyonic instabilities in closed superstring theory
and M-theory, Nucl. Phys. B 611 (2001) 93124, hep-th/0104238;
A.A. Tseytlin, Magnetic backgrounds and tachyonic instabilities in closed string theory, hep-th/0108196;
T. Suyama, Properties of string theory and KaluzaKlein Melvin backgrounds, hep-th/0110077;
T. Takayanagi, T. Uesugi, Orbifolds as Melvin geometry, hep-th/0110099.
[4] J.G. Russo, A.A. Tseytlin, Supersymmetric fluxbrane intersections and closed string tachyons, hepth/0110107.
[5] J.M. Figueroa-OFarrill, J. Simn, Supersymmetric KaluzaKlein reductions of M-branes, to appear.
[6] E. Dudas, J. Mourad, D-branes in string theory Melvin backgrounds, hep-th/0110186;
T. Takayanagi, T. Uesugi, D-branes in Melvin backgrounds, hep-th/0110200.
[7] T. Takayanagi, T. Uesugi, Flux stabilization of D-branes in NSNS Melvin background, hep-th/0112199.
[8] C.M. Hull, Exact pp wave solutions of eleven-dimensional supergravity, Phys. Lett. B 139 (1984) 3941.
[9] E. Bergshoeff, P.K. Townsend, Super D-branes revisited, Nucl. Phys. B 531 (1998) 226238, hepth/9804011.
[10] K. Becker, M. Becker, A. Strominger, Fivebranes, membranes and non-perturbative string thoery, Nucl.
Phys. B 456 (1995) 130152, hep-th/9507158;
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
191
E. Bergshoeff, R. Kallosh, T. Ortn, G. Papadopoulos, Kappa-symmetry, supersymmetry and intersecting

branes, Nucl. Phys. B 502 (1997) 149169, hep-th/9705040.
M. Cederwall, A. von Gussich, B.E.W. Nilsson, A. Westerberg, The Dirichlet super-three-brane in tendimensional type IIB supergravity, Nucl. Phys. B 490 (1997) 163178, hep-th/9610148;
M. Cederwall, A. von Gussich, B.E.W. Nilsson, P. Sundell, A. Westerberg, The Dirichlet super-p-branes in
ten-dimensional type IIA and IIB supergravity, Nucl. Phys. B 490 (1997) 179201, hep-th/9611159;
E. Bergshoeff, P.K. Townsend, Super D-branes, Nucl. Phys. B 490 (1997) 145162, hep-th/9611173.
J. Gomis, A. Ramallo, J. Simn, P.K. Townsend, Supersymmetric baryonic branes, J. High Energy Phys. 11
(1999) 019, hep-th/9907022;
M.T. Grisaru, R.C. Myers, O. Tafjord, Susy and goliath, J. High Energy Phys. 08 (2000) 040, hepth/0008015;
A. Hashimoto, S. Hirano, N. Itzhaki, Large branes in AdS and their field theory dual, J. High Energy Phys. 08
(2000) 051, hep-th/0008016;
J.M. Camino, A.V. Ramallo, Giant gravitons with NSNS B field, J. High Energy Phys. 09 (2001) 012,
hep-th/0107142;
J.M. Camino, A.V. Ramallo, M-theory Giant gravitons with C field, hep-th/0110096.
A.A. Tseytlin, No-force condition and BPS combinations of p-branes in 11 and 10 dimensions, Nucl. Phys.
B 487 (1997) 141154, hep-th/9609212.
J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
J.G. Russo, A.A. Tseytlin, Waves, boosted branes and BPS states in M-theory, Nucl. Phys. B 490 (1997)
121144, hep-th/9611047.
P.K. Townsend, Four lectures in M-theory, Proceedings of the ITP summer school on High Energy Physics
and Cosmology, Trieste, June 1996, hep-th/9612121.
Y. Kosmann, Drives de Lie des spineurs, Annali di Mat. Pura Appl. (IV) 91 (1972) 317395.
J.M. Figueroa-OFarrill, G. Papadopoulos, Homogeneous fluxes, branes and a maximally supersymmetric
solution of M-theory, J. High Energy Phys. 06 (2001) 036, hep-th/0105308.

D2-branes with magnetic flux in the presence

of RR fields
D.K. Park a,b , S. Tamaryan c,d , H.J.W. Mller-Kirsten c
a Department of Physics, Kyungnam University, Masan, 631-701, South Korea
b Michigan Center for Theoretical Physics, Randall Laboratory, Department of Physics, University of Michigan,
Ann Arbor, MI 48109-1120, USA

c Department of Physics, University of Kaiserslautern, 67653 Kaiserslautern, Germany
d Theory Department, Yerevan Physics Institute, Yerevan-36, 375036, Armenia
Abstract
D2-branes are studied in the context of BornInfeld theory as a source of the 3-form RR gauge
potential. Considering the static case with only a radial magnetic field it is shown that a locally
stable hemispherical deformation of the brane exists which minimises the energy locally. Since the
D2-brane carries also the charge of D0-branes, and the RR spacetime potential is unbounded from
below, these can tunnel to condense on the D2-brane. The corresponding instanton-like configuration
and the tunneling rate are derived and discussed. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Lm; 11.27.+d; 05.70.Fh
1. Introduction
The low energy dynamics of D-branes described by BornInfeld theory on the
worldvolume of the brane is a topic of intense investigation and has led to useful insights
into how string theory is interwoven with electromagnetic phenomena. Such investigations
are useful in domains where gravitational (closed string) effects can be ignored in the
leading approximation. The dynamics of these branes changes drastically in the presence
of external (spacetime) forces, such as those of RR fields contained in type II superstring
theories, since the D-brane carries the appropriate RR charges and so couples to the
E-mail addresses: dkpark@hep.kyungnam.ac.kr (D.K. Park), sayat@moon.yerphi.am (S. Tamaryan),
mueller1@physik.uni-kl.de (H.J.W. Mller-Kirsten).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 1 - 6
D.K. Park et al. / Nuclear Physics B 635 (2002) 192214
193
potentials. In the following we will be concerned with these interactions in the particular
case of D2-branes.
BornInfeld theory by itself (i.e., without RR fields) has been shown [1] to imply stringlike brane excitations which owing to their charge and tension can be identified in the
appropriate limit as fundamental strings. Such a string may be looked at as a collapsed
brane or the original brane as one with a dissolved fundamental string [2]. The conserved
(axion) charge of these strings along (say) x 1 results from the electric component B01
of the NS B-field contained in the BornInfeld action, and their tension is of order 1 (as
distinct from the 1/(string coupling g) behaviour of the tension of D-branes). Considering
BornInfeld theory of the D2-brane in the static limit and with only the electric component
of the U (1) gauge field in its worldvolume, these strings are globally stable. However, on
application of the RR field, i.e., with minimal coupling of the brane to the RR potential, the
BornInfeld string has been shown to be only locally stable and can tunnel to or expand into
a D2-brane [2,3], which is also described as the polarisation of a system of fundamental
strings into a higher-dimensional brane. This tunneling has been considered in detail in
Ref. [3].
It is natural to extend such investigations to the magnetic counterpart or rather to the
fully electromagnetic formulation which, of course, introduces complications as soon as
Lorentz boosts or deviations from a static case are required. Such investigations have been
carried out recently in various directions [411], and our objective here is to extend some
considerations of Refs. [2] and [4] with particular reference to braneantibrane pairs as in
Ref. [1] and the case of dielectric branes [10,11].
Since the D2-brane couples not only to the three-form RR potential but in the presence
of magnetic flux also to the one-form potential of D0-branes, the magnetic case is very
different from that of the purely electric case. Static torus-like brane configurations in the
presence of only a pure magnetic field have been derived in Ref. [4] as well as locally
stable spherical configurations which can be related to the dielectric D-branes of Ref. [10].
The presence of the magnetic flux provides these theories locally with energy minima at
nontrivial expanded configurations analogous to the separation of charges in a dielectric
medium. The related stabilisation of branes (i.e., prevention of their collapse to trivial or
pointlike pure tension configurations) by the presence of magnetic flux was pointed out
earlier in Ref. [12] in the context of WZW models.
In the following we remain in the context of the model of Refs. [2] and [4] and
show that a locally stable hemispherical brane configuration can be shown to exist for
a pure magnetic field in the worldvolume. We demonstrate this explicitly by considering
fluctuations around the brane. The considerations are analogous to those of branes and their
antibranes of Ref. [1] whose stability was studied in detail in Ref. [13]. We then consider
the Euclidean time pseudoparticle brane configuration and its relation to the classical brane
solution. Having found these, we calculate the transition rate of the 2-brane through the RR
potential barrier and interpret the result as a set of D0-branes condensing on the D2-brane
as suggested some time ago [12]. Irrespective of the physical significance of the result,
we consider the explicit calculations which the model permits, to be very instructive for
comparison with other cases such as those of Refs. [1] and [13], and in providing hints on
what one may expect in higher-dimensional cases. Some specific calculations, such as the
194
comparison with the spherical D2-brane of Ref. [4,10], the evaluation of the fluctuation
determinant and the Lorentz transformation of the action are shifted into appendices.
2. D2-branes in the presence of both electric and magnetic fields

The BornInfeld action describing a D2-brane as the source of the 3-form RR gauge
potential A and its coupling to this RR 3-form gauge potential in type IIA superstring
theory is

I = T2 d 3
det( X X + 2 F )

1
+ A X X X ,
(1)
3!
where T2 = 1/4 2 g is the p = 2 volume tension of the D2-brane obtained from
Tp =
2
,
(2ls )p+1 g
(2)
where we set ls = 1 and = 2ls2 = 2 = 2 . Here p = Tp is the RR charge of the

brane under the (p + 1)-form RR potential. We follow Ref. [2] but include also a magnetic
field. With = 1,
X0 = t,
X1 = z,
X2 = R(t, z) cos ,
X3 = R(t, z) sin ,
others = const
(3)
Fz = Az z A
(4)
H = dA,
(5)
and
and
H0123 = h
and the target space metric with signature 1, +1, +1, +1, . . ., and , = t, z, , Ez =
2Ft z , E = 2Ft , B = 2Fz :

+ Ez E
1 + R 2
RR

X X + 2 F = Ez + R R
(6)
B .
1 + R 2
E
B
R2
Then

I=
dt dz d L(R, A0 , Az , A ),
where for E = 0

1
2 1R
2 + R 2 Ez2 + B 2 1 R 2 + h R 2 .
R
L=
2
4 2 g
(7)
(8)
195
The four resulting EulerLagrange equations reduce to two constraints and one equation
of motion but imply also certain conditions and therefore have to be considered carefully.
The equations are, respectively:
1. For R we obtain the equation of motion

2
R 2 R + RB
t
R 2 (1 R 2 + R 2 Ez2 ) + B 2 (1 R 2 )

R R2

+
z
R 2 (1 R 2 + R 2 E 2 ) + B 2 (1 R 2 )
z
R(1 R 2
+ R 2
Ez2 )
R 2 (1 R 2 + R 2 Ez2 ) + B 2 (1 R 2 )
+ hR = 0.
(9)
2. For A0 we obtain the equation equivalent to the Gauss law in Maxwell theory, i.e.,
DE
= 0,
z
DE
R 2 Ez
R 2 (1 R 2 + R 2 Ez2 ) + B 2 (1 R 2 )
(10)
3. For A we obtain the equation

(2D)
= 0,
z
2D
2B(1 R 2 )
R 2 (1 R 2 + R 2 Ez2 ) + B 2 (1 R 2 )
(11)
4. Finally for Az we obtain the equation
(2DE ) +
(2D) = 0.
t
(12)
We observe that for the purely electric case considered in Ref. [2] the second and the fourth
equations imply that the electric quantity DE is independent of both z and t and is therefore
a constant. In the purely magnetic case, which we concentrate on here, we conclude from
the third and fourth equations that the magnetic quantity D does not possess an explicit
z-dependence, nor an explicit -dependence, but can depend explicitly on t. In addition
and this is a vital pointD is a functional of R which is a function of z and and possibly
of t.
In the static and purely magnetic case the two remaining equations are

R(1 + R 2 )
R R2

(13)

+ hR = 0
z
R 2 (1 + R 2 ) + B 2
R 2 (1 + R 2 + B 2 )
and
D=
2B
R 2 (1 + R 2
+ B2)
(14)
196
Setting

P 2 = 1/ 1 D 2
and solving the latter equation for B we obtain

B = P DR 1 + R 2
and
(15)

R2 1 + R 2 + B 2 = R2 1 + R 2 P 2 .
(16)
This equation allows us to express the quantity P 2 and so D in terms of R and B 2 . Using
this, Eq. (13) leads to

RR
1 + R 2 + hR = 0.
(17)
z P 1 + R 2
P
With further manipulations this equation can be converted into

hR 2
hR 2
d
R
R
= C,
= 0,
dz P 1 + R 2
2
2
P 1 + R 2
(18)
where C is a constant. Eq. (18) is also contained in the work of Ref. [4] and can also be
obtained from a Legendre transformed Lagrangian density
LB := L + B
L
,
B
D = 4 2 g
L
,
B
(19)
where

1
R 1 + R 2 h 2
LB (R) =
R
.
4 2 g
P
2
(20)
We comment on other derivations later.

We now define the Hamiltonian density H by
H = PR R + PAz A z + PA A L,
(21)
etc. Then
where PR , PAz , PA are the conjugate momenta of R, Az , A , i.e., P = L/ R,

1
1
R2 + B 2 + R2 R 2
h
H=
(22)

R 2 + DE A0 .
2
g
4 g
(1 R 2 )(R 2 + B 2 ) + R 2 (R 2 E 2 ) 2
z
This expression is to be supplemented by the two constraint equations (10) and (11) for Ez
and B.
The magnetic flux in the worldvolume is for R = 0 given by

1
1
R2 D 1 + R 2
.
N=
(23)
d dz Fz =
dz
2
2
2 + R 2 (1 D 2 )
DE
This integral of the worldvolume 2-form field strength is in general not finite. Thus to
obtain a finite expression for this charge, one has to close the membrane configuration
197
as already discussed in Ref. [14]. This is essentially the charge associated with the
worldvolume vector potential. The latter is a 1-form which couples to a D0-brane.
The Hamiltonian H now becomes in going to the static case (R = 0)

2)
2

h
(1
+
R
1
2
R2 .
dz d R + DE
H=
(24)
2 + R2 R2 D2 )
2
4 2 g
(DE
In performing the variation

H
d H
= 0,
dz R
R
one has to remember that the expression D is a functional of R as expressed by Eq. (16).
Considering the static magnetic case this means that before the variation of the Hamiltonian

1
h 2

2
H=
(25)
dz d P R 1 + R R
2
4 2 g
is performed, one has to replace P with the help of expression (16) in terms of B. Finally
again replacing the expression with B by that with P one arrives at the same equation
which was obtained earlier from the Lagrangian. The Hamiltonian will be needed later
for the calculation of the energy of our minimum energy configuration. In the above we
considered the case of a cylindrical geometry. For comparison we provide in Appendix A
the main formulae (for flux and energy) of the spherical case considered in Refs. [4,10].
3. Pure magnetic case

We consider the static and purely magnetic case. The total magnetic flux is nonzero
which means that we have objects dissolved in the D2-brane carrying magnetic charge.
Being the singularities of the magnetic field, these objects must be pointlike, hence they
are D0-branes. In Ref. [4] it was shown that the number of D0-branes can in this case be
simply related to the magnetic flux. The Hamiltonian H is given by Eq. (24). We know that
if the D2-brane is coupled to magnetic flux it also carries the charge of the D0-brane [15].
Thus it is sensible to consider the limit of the spherical configuration shrinking to a point.
The worldvolume magnetic flux is given by

PD
P DS

2
N=
(26)
,
S
=
dz
d
R
1 + R 2 ,
dz R 1 + R =
2
(2)2
where S is the area of the worldvolume which follows simply from the fieldless static metric contained in Eq. (6). One can see from this relation that if D were kept constant in a
variation, a nonvanishing variation of S would have to be compensated by a nonvanishing
variation of N , and thus the number of D0-branes would not be fixed unless compensated
appropriately from a reservoir of D0-charge around the D2-brane. From the relation (26)
we obtain
D2 =
(NT0 )2
,
(ST2 )2 + (NT0 )2
NT0 = ST2 DP ,
(27)
198
so that for S shrinking to zero D 1 and the energy can be presented as the mass of N
D0-branes, i.e.,
ER0 = T0 N.
(28)
One can also express the energy like Eq. (27), as observed earlier in Ref. [4]:

h
2
2
H = (ST2 ) + (NT0 )
dz R 2 .
4g
(29)
Varying this expression (and so the surface area S with respect to R) one again obtains
the same equation of motion as with the other methods for D = D(R). The equation of
motion, Eq. (18), can be rewritten to give

h
2 R2 R2 R2
R+
R = 2
(30)
hR + 2C
with

4C 2
4
2
2
(31)
,
R+
+ R
= 2 2 1 CP 2 h .
2
h
P h
We are looking for nonperiodic, finite energy solutions. Consequently we set R = 0
yielding for the integration constant C = 0. Then Eq. (30) simplifies to

2 R2
R+
,
R =
(32)
R
2 2
R+
R =
where R+ = P2h and the configuration R is geometrically a radius. The solutions of this
equation are the configurations R with, respectively, positive (z0 > z) and negative (z0 < z)
derivatives as in Eq. (32), i.e.,
z
R(z) =
z0 R+
z0+R+
R(z) = +

z
(z z0 ) dz
2
R+
(z z0
)2
(z z0 ) dz
2 (z z )2
R+
0
2 (z z )2 ,
R+
0
2 (z z )2
R+
0
(33)
with the enveloping sphere

2
.
R 2 + (z z0 )2 = R+
(34)
This sphere is thus the envelope of the pair of spherical shells with z0 R+ z z0 , z0
z z0 + R+ , or the pair of circles of radius R at positions
z = z0
R
2
R+
R2
=
R+
R dR
(35)
2 R2
R+
Thus z(R) is double valued. The two possible signs define the two hemispherical
configurations on the enveloping sphere given by Eq. (34) and as indicated in Fig. 1.
199
Fig. 1. The two circular shells and the enveloping sphere.
These two hemispherical configurations defined by, respectively, positive or negative R

(observe that R originally defined as a radius is always positive, the variable angle being
understood) can be looked at as a brane and its antibrane analogous to the appearance of
branes and their antibranes in Ref. [1]. In fact, that the sphere (34) represents an unstable
braneantibrane pair can be seen by differentiating (34) with respect to z and reinserting
the equation, which gives
z z0
dR
=
.
dz
2 (z z )2
R+
0
(36)
We see that this is an odd function which reverses its sign on passing through z0 . This
is one of the characteristic properties of the configuration called a bounce [16]. (Like
a periodic instanton it can be loosely looked at as an instantonantiinstanton pair, the
instanton (or antiinstanton) being a monotonically increasing (or decreasing) function of
its argument contrary to the behaviour of the combination). The sphere is the limiting
form of a spheroidal bulge like that discussed in the electric case in Ref. [2]; hence the
behaviour of R here is also that in this limit, i.e., that of a function with the shape typical
of an odd first excited state wave function. Thus as expected in Ref. [1] for the D3-brane
model considered there and demonstrated for this in Ref. [13], the braneantibrane pair
is unstable, which means that small fluctuations in its neighbourhood possess a negative
mode, the tachyon (how the braneantibrane system with magnetic flux N = 2 is related to
tachyon condensation is described, for instance, in Ref. [15]). The appearance of these
braneantibrane pairs is a consequence of the two possible signs of the derivative in
Eq. (32), and so of two possible solutions, and these in turn are a consequence of the
square-root form of the BornInfeld Lagrangian density. Thus BornInfeld theory leads
very naturally to braneantibrane configurations through a linkage of the solutions or
continuation of the one solution to the other with the opposite sign of its derivative as
in the cases considered in Refs. [1] and [2].
200
For the energy E of the solution with R 1 + R 2 = R+ and z0 R+ z z0 as

integration domain one finds from H
3
4R+
hT2
NT0
.
E=
(37)
2D
3
2
With Eq. (27) the first part can be rewritten so that
3
4R+
hT2
1
2
2
.
E=
(NT0 ) + (ST2 )
2
3
2
(38)
(With the factor 1/2 the volume part is the same as that in the electric case of Ref. [2].)
This result is very physical with the first part representing the square root of the sum of
the squares of the masses of the D2-brane and the N D0-branes, and the second the energy
stored in the volume. The negative sign of the latter indicates that the (hemi)spherical
configuration of radius R+ minimises the energy locally as will be shown below. We
observe from Eqs. (24) and (25) that the potential is unbounded below for large values
of R. Thus tunneling of the locally stable hemispherical configuration is possible, and one
can calculate the appropriate transition rate. This will be examined in the following.
4. Small fluctuations and the pre-exponential factor

We now consider the second variation of H around the positive derivative solution (33)
and demonstrate that it is positive definite under small fluctuations. This means that any
small deviation from this solution of (33) increases the energy. Consequently we have a
minimum of the energy functional. This minimum is not a global minimum because the
potential is unbounded below as observed earlier; thus it is only a local minimum. Next
we determine the fluctuation operator describing the behaviour of the second variation
of the energy functional in the vicinity of this classical solution. BornInfeld theory is
a covariant theory; therefore, the existence of the static finite energy solution implies
the existence of the instanton-type pseudoparticle solution which is really a bounce. In
calculating the tunneling transition rate in the next section, we use the semiclassical
approximation around this instanton-type solution. The argument of the exponential of
the semiclassical amplitude is the pseudoparticle action, and the pre-exponential factor
is, as usual, the determinant of the fluctuation operator of the pseudoparticle solution.
From Lorentz invariance it follows that the fluctuation operators of both the static and
the pseudoparticle solutions must coincide up to a possible sign factor, so that it suffices to
calculate the determinant in the static case.
The second variation of H is given by

2

1
2H
2 H
H
2

2
.
(R
)
+
2
RR
+
(R)
dz
2 H =
(39)
2
RR
R 2
R 2
This is evaluated in Appendix B and the result is

1

= 1 dR R MR,
2 H =
dz R MR
2
2
R
(40)
201
is
where the fluctuation operator M
=
M
1 1 d RR 2 (1 + D 2 R 2 ) d 1
.
2gP R dz (1 + R 2 )3/2 dz R
(41)
The derivative R is positive for one hemisphere as explained earlier, and negative for the
as
other. Thus one of these minimises the energy locally, the other does not. We present M
a product because this has the advantage that its determinant can be easily calculated. This
is our next step.
must be normalized. As a normalization
The determinant of the fluctuation operator M
point we choose D = 0 (P = 1) which is the case of a vanishing magnetic field. We denote
the corresponding classical solution by R0 and the corresponding fluctuation operator by
0 . The normalized determinant is then
M
=
detn M

det M
.
0
det M
(42)
We use the following two properties of determinants:

det(AB) = det(BA),
det(T A) det A
=
.
det(T B) det B
(43)
The expression for the normalized determinant then simplifies and we have
=
detn M
det[R(1 + D 2 R 2 )(1 + R 2 )3/2 /P ]

.
det[R0 (1 + R0 2 )3/2 ]
(44)
For our solution with integration constant C = 0 we have

R(1 + R 2 )3/2 h3 R 4 2
=
P ,
P
8
resulting in
=
detn M
2)
det(R 4 + R 2 P 2 D 2 R+
det R04
2

D 2 R+
1
1 + D2 R 2 = 2 +
P
R2
(45)
(46)
2 is a c-number and its determinant yields an integral:

The expression R 4 + R 2 P 2 D 2 R+

2
2
det R 4 + R 2 P 2 D 2 R+
= exp Tr ln R 4 + R 2 P 2 D 2 R+
0

dz 4
2 2 2 2
= exp
(47)
ln R + R P D R+ .
R+
R+
The evaluation of the integral is straightforward and yields

0
R+

2

dz 4
P +1
2
= 4 + 2 ln 2R+
+ ln P 2 1 + P ln
ln R + R 2 P 2 D 2 R+
R+
P 1
(48)
202
which gives for the determinant

P 2 1 P + 1 P D2 P + 1 P

= 2
.
detn M =
P4
P 1
P
P 1
(49)
One may note that the expressions (48) and (49) are finite for P 1.
5. The tunneling amplitude

To describe the tunneling by the instanton-type pseudoparticle solution in Euclidean
time, we make a Wick rotation and set t i . The way to find this pseudoparticle
solution in a case like the one here was already proposed in [2]. One performs an exchange
z, so that having the static solution one obtains the pseudoparticle one. However,
the Lagrangian of Eq. (8) does not exhibit an explicit symmetry under this exchange.
This might seem strange but has a simple explanation if we include in our considerations
the polar component E of the electric field which has therefore been given explicitly in
Eq. (6). We now perform a right angle rotation in the (z, ) plane, i.e.,
z z = ,
= z,
(50)
under which electromagnetic components transform as follows:

Ez Ez
(51)
and
E B,
B E ,
(52)
where B is as before the magnetic field. The above transformations show that in
the purely magnetic case the action does not have the explicit symmetry under the
exchange z. On the other hand the transformations (52) are electromagnetic duality
transformations [17] leaving the equations of motion invariant [17,18]. Therefore, the
Lorentz transformations yield for the pseudoparticle the same equation (26). We verify
this explicitly in Appendix C. Also one can make the backward Lorentz rotation and obtain
the pseudoparticle solution in the original frame. The shape of the pseudoparticle is that
of a sphere which, however, is not invariant under this rotation and instead becomes an
ellipsoid. It is instructive to obtain the same ellipsoid directly from the EulerLagrange
Eq. (9). We set Ez = 0, R = 0, so that in this nonstatic case
B2 =
D2 R2
1 R 2 D 2
and obtain for the time dependent solution the equation

R R
d

1 R 2 D 2 + hR = 0.
dt
1 R 2 D 2
(53)
203
Going to Euclidean time with t = i and using the fact that the equation does not contain
explicitly, allows us to convert it into the following first order differential equation
h
R

R 2 = C,
2
2
P 2 1 D 2 + ( dR
d )
(54)
where C is a constant. For the discussion below it is more transparent to consider first the
general case with C = 0. Then

h
dR
2 R2 R2 R2 ,
=
(55)
R+
h
2
d
2P (C + 2 R )
where
2
=
R
2
P 2 h2
1 ChP 2

1 2ChP 2 .
(56)
For large R the energy given by the Hamiltonian H of Eq. (25) decreases without limit.
Thus the motion of the pseudoparticle in Euclidean time starts with zero velocity dR/d
at R = R+ and bounces back from the potential wall at point R at the time-symmetric
point = 0 until it again reaches R = R+ . At the time-symmetric point the reversal of
the velocity implies a switch from one sign of the square root to the other. Now we again
choose the integration constant C to be zero as in Eq. (32). We thus have R = 0 and
obtain
2
R2
2 dR
+ 1 = +2 .
P
(57)
d
R
From this we obtain the solution

2
2
2
2 R2
2
R R+ R
+ ( 0 )2 = 0,
P
+
(58)
where R is the bouncing or turning point at Euclidean time 0 . However, in evaluating the
action of the bounce and hence the tunneling rate it is more convenient to integrate with
respect to R by setting d = dR/R and using the derivative relation (57), i.e., also

dR 2
2 + D2 P 2 R2 .
= R+
PR 1 +
(59)
d
The pseudoparticle action IE is now obtained from Eq. (7), i.e.,

dR 2 h 2
IE = T2 dz d P R 1 +
+ R .
d
2
(60)
The orbit of the bounce is that from R+ to R and back, so that

IE = 2P T2

R+
2 + D2 P 2 R2
R+
R2
h
2
.

dz dR
2 R2
2 R2 R2
R+
2
(61)
204
Evaluating this by integrating over the compact length L of a torus we obtain [19]
3

4R+
2
2 2
IE = 4P T2 LR+
1,
1/2,
3/2;
D
+
F
P
LP hT2 .
2 1
3
(62)
Using Eq. (27) the action IE can be rewritten as
3

4R+
NT0
2 2
IE = L
+
LP hT2
2 F1 1, 1/2, 3/2; D P
D
3

= L (ST2 )2 + (NT0 )2 2 F1 1, 1/2, 3/2; D 2P 2
3
4R+
P hT2 .
+L
3
(63)
Since action corresponds to energy length, this result is, as expected, similar to the
energy (38) of the sphere but takes into account the ellipsoidal deformation. In computing
the decay or tunneling rate we have to subtract from the action of the pseudoparticle the
action of the initial state which means here the square-root part of the action. For the
argument of the hypergeometric function we also have

N 2
.
D2 P 2 =
(64)
2
R+
We observe that for no magnetic field and so charge, or no D0-branes, the radius R+ is
zero, that is, there is no dielectric effect.
With this the semiclassical approximation of the tunneling amplitude is given by
1/2 exp(IE ).
= (det M)
(65)
Inserting the appropriate quantities, the final expression for the tunneling rate is seen to be
3
4R+
P P 1 P /2
P hT2
=
exp L
D P +1
3
P /2
P P 1
8L
=
(66)
exp
.
D P +1
3gP 2 h2
In the limit of D 0, P 1, there is no electromagnetic field (cf. (11) and (16)).
But we still have a static spherical solution which is the excitation of the original brane
under the influence of the RR field. In the opposite limit of D 1, P of a strong
magnetic field, it is better to express the tunneling amplitude via the D0-brane number.
From Eqs. (26) and (27) it follows that
2
1
4
1
+
O
, h2 N 1.
P2 =
(67)
2
h N
P2
Thus the tunneling rate becomes

Lh2 N 2
4
exp

.
h2 N
6g
(68)
205
6. The purely temporal case

We consider here the cylindrical D2-brane with magnetic field in a purely timedependent case and demonstrate an analogy with the static, purely electric case [2,3]. This
permits the consideration of quantum classical transitions in the magnetic case to be taken
over from the electric case.
We start from our original D2-brane action

h 2
2

2
2
2
2
SD2 [R, A , Az ] = T2 dt d dz R 1 + R R + B 1 R R
2
(69)
with and L/2 z L/2. Varying the action (69) yields the following
equations of motion:
D=
D = 0,
z
(R 2 + B 2 )R
R2 R

t R 2 (1 + R 2 R 2 ) + B 2 (1 R 2 ) z R 2 (1 + R 2 R 2 ) + B 2 (1 R 2 )
R(1 + R 2 R 2 )
+
(70)
hR = 0,
R 2 (1 + R 2 R 2 ) + B 2 (1 R 2 )
where now with B = 2Fz
D=
B(1 R 2 )
.
R 2 (1 + R 2 R 2 ) + B 2 (1 R 2 )
(71)
Since the equations z D = D = 0 have already been solved in the static case (R = R(z)),
we now consider the solution of z D = D = 0 in the purely temporal case, i.e., for
R = R(t).
In this case Fz = Fz (t) could be an arbitrary function of time. Since, however, we
are considering the purely magnetic case, it is appropriate to assume again Fz = const,
because if not, the time-dependent magnetic field usually generates an electric field. Thus,
we choose
Fz =
N
,
L
where
N=
1
2
(72)

d dz Fz
(73)
is the total number of D0-brane particles or quantised flux through the cylindrical surface
of length L.
Using (72) and R = 0 the last of Eqs. (70) reduces to

R2 + 2
1 R 2
hR = 0,
+
R
(74)
R
t
R2 + 2
1 R 2
206
where
N
, = 2.
L
One can also show that Eq. (74) is obtained directly by varying the action

h

SD2 [R] = 2LT2 dt
R 2 + 2 1 R 2 R 2 .
2
=
(75)
(76)
In general, we cannot insert a classical solution into the action before varying it, however
here this is permissible for the constant solution (72), which can also be shown by varying
action (76). Before solving (74) it is helpful to consider the potential VD2 (R) which can be
read off from (76),

h
VD2 (R) = 2LT2
(77)
R2 + 2 R2 .
2
It is interesting to reexpress the potential as

VD2 (R) = (ST2 )2 + (NT0 )2 hVT2 ,
(78)
where S = 2RL and V = R 2 L. In fact, S and V are, respectively, surface area and
volume of the cylindrical D2-brane in flat spacetime. Hence, the potential consists of two
terms, i.e., the surface energy of the D2-brane with N D0-branes dissolved in it and the
volume energy.
The shape of the potential VD2 (R) is as follows. If h > 1, VD2 (R) is a monotonically
decreasing function and R = 0 becomes a point of instability.
If h < 1, VD2 (R) has a
local minimum at R = 0 and a global maximum at R R = 1/ h2 2 . Thus we have
quantum tunneling in this case. Here we confine ourselves to the latter case ( h < 1). We
summarize several particular values computed from the potential:
VD2 (R = 0) = 2LT2 NT0 ,

LT2
1 + h2 2 ,
VD2 (R = R ) =
h
1
VD2 (R = 0) = h > 0,

VD2 (R = R ) = h 1 h2 2 < 0.
(79)
We solve Eq. (74) for h < 1. Since we have tunneling in this case, it is more convenient
to go to Euclidean time by introducing = it. Then Eq. (74) becomes

1 + R 2
d R2 + 2
+R
hR = 0,
(80)
R
d
R2 + 2
1 + R 2
where a dot denotes differentiation with respect to . In fact, Eq. (80) can be derived by
varying the Euclidean version of action (76)

h
2 + 2 1 + R 2 R 2 .
=
2LT
IDEuc
(81)
d
R
2
2
2
One can show that Eq. (80) can be converted into the following first order form

R2 + 2 h 2
R = C,
2
1 + R 2
207
(82)
where C is an integration constant. After some manipulations one can reexpress Eq. (82)
in the following way:

h
2 R2 R2 R2 ,
R+
R =
(83)
2C + hR 2
where
4(1 Ch)
4(C 2 2 )
2 2
(84)
,
R+
R =
.
2
h
h2
Comparing Eq. (84) with corresponding equations in Refs. [2,3], one can see that Eq. (83)
is exactly the same as that of the purely electric case there if the electric displacement
D there is identified with . Thus the general periodic instanton solution of (84) and its
classical Euclidean action can be read off directly from Ref. [3].
Here we consider only the vacuum solution (R = 0, C = ), which is determined by

2 R2

R+ + R+
2
2 R2 +
= h(0 ),
R+
ln
(85)
hR+
R
where R+ = 2 1 h/ h and the corresponding Euclidean action is

4 3 hT2
R
.
Icl = NT0 d + L
(86)
3 + 2
2
2
R+
+ R
=
Here the first term is the contribution of N D0-branes and the second term is the
contribution of the D2-brane.
Finally we show that our formulation allows D to be time-dependent but N to be fixed.
Using (71) and (82) D is expressed in Euclidean space as

1 + R 2
D=
(87)
=
.
R2 + 2
C + h2 R 2
Since R is dependent on time, D should also depend on time. Using (71) and (73) N can
be generally expressed in Euclidean space as

D
R 1 + R 2 + R 2
.
N=
(88)
d dz
2
(1 + R 2 )(1 + R 2 D 2 )
Thus if we consider R = R( ), N reduces to
N=
L
DR

.
1 + R 2 D 2
Inserting (87) into (89) we obtain N = L/, which is our original definition of N .
(89)
208
One can also calculate the quantum-classical transition in this case using the periodic
instanton or sphaleron solutions. The criterion for a first-order transition can be read off
directly from Ref. [3] as
N 1
< .
(90)
L
2
Thus the number of D0-branes as well as the RR-potential are involved in the criterion.
h
7. Concluding remarks
In the above we have considered D2-branes in the presence of spacetime RR fields
in the context of a model with worldvolume cylindrical symmetry, and we have found
locally stable hemispherical deformations of the brane, the complementary hemispherical
configurations being unstable. We have also demonstrated that these two configurations
together comprise an enveloping sphere representing a braneantibrane pair which in view
of its associated Euclidean time bounce configuration is unstable. This configuration is
analogous to the braneantibrane configuration constructed in Ref. [1] where the presence
of the bounce was anticipated. The stability/instability of the associated field configurations
was investigated in detail in Ref. [13], where it was pointed out in particular that the
braneantibrane configurations (constructed from the combination of a stable brane and
an unstable antibrane) are again unstable. We have derived explicitly the operator of small
fluctuations about such configurations from which the local stability or instability of such
configurations in the sense of minimising the energy locally may be deduced. We then
calculated the corresponding transition rate for the decay of such a locally stable brane
configuration through the hump of the RR potential. It might be somewhat easier to repeat
the same steps in the case of a pure electric field aligned with the RR field in view of the
explicit symmetry of the action under the exchange of time and space coordinates. More
interesting is the consideration of the polar component of the electric field together with the
magnetic field. In this case the action has a nice symmetry as discussed in Section 5. Also,
having both electric and magnetic fields in this case, one might expect the existence of
both strings as well as D0-branes. Another interesting direction of extension is to consider
the same phenomena with D3-branes. The D3-brane is selfdual and physical quantities
in different regions of the energy and/or coupling constant can have the same analytic
expression [10]. But in all these cases the states are unstable because the strong RR field
makes the potential unbounded from below. One might try to consider D0 and spherical
D2-branes applied to a radially decreasing RR field at infinity to avoid this instability. The
proper principle, however, is to somehow take into account gravity, which might be a good
candidate to stabilize the system as has been discussed, for instance, in Ref. [8].
Acknowledgement
S.T. acknowledges support by the Deutsche Forschungsgemeinschaft (DFG). D.K.P.
was supported by grant No. R05-2001-000-00106-0 from the Basic Research Program of
the Korea Science and Engineering Foundation.
209
Appendix A
Here we summarise a few points in relation to Refs. [4,10] in which a spherical D2brane was considered instead of the cylindrical one considered here. In a simplified way
Ref. [10] has
X0 = t,
X1 = r(t) sin cos ,
X3 = r(t) cos ,
so that with
X0
X2 = r(t) sin sin ,
others = const (Dirichlet)
(A.1)
=t
9

i 2
ds 2 = dt 2 + dr 2 + r 2 d 2 + sin2 d 2 +
dX .
i=4
Now one takes as worldvolume coordinates of the D2-brane (denoted by indices

, , . . .) the variables t, , . Thus r, or the function r(t), originally the third of the three
polar coordinates, acts as a scalar excitation of the brane. Then F = A A and
H = dA, H0123 = h, and as in [4] we take the background RR four-form field strength to
be H0123 = h = const h123 . This is a field strength aligned with the 123 subspace of
spacetime, i.e., orthogonal to the 4 9 part. The target space metric is thus changed from
Minkowsky to S2 (r) R7 . Then the BornInfeld action integral becomes

I = d 3 L(r, A , A ),
(A.2)
where
L = T2

2
1 + r 2 r 4 sin2 + 4 2 F

4 r (0 A ) + sin (0 A )
2 2
1/2

hr 3 sin
.
3
(A.3)
The expression F = N2 sin (apart from its normalisation) is not a choice; rather it is
dictated by the EulerLagrange equations derived from L for F in the present case.
Consider for simplicity the static case and ignore the WessZumino contribution. Set
F
r 4 sin2 + 4 2 ( A A )2
Ds .
Then we obtain the equations

Ds
= 0,
Ds
=0
so that Ds is independent of and , but, of course, depends on r. Now we can solve the
former equation for F and obtain
Ds r 2 sin
.
F =
1 Ds2 .4 2
(A.4)
210
This relation corresponds

exactly to the relation one obtains in the cylindrical case, there
with sin replaced by R 1 + R 2 . Since

2N = d d F
the flux, obtained by integrating over the closed surface of the 2-sphere is quantised.
The number N is identified with the number of D0-branes. From here on many of the
considerations leading to the minimised energy parallel those in our considerations above
and therefore will not be given here. We cite only the expression for the potential

hr 3
hr 3
4
2
2
= (ST2 )2 + (NT0 )2 4T2
.
VD2 = 4T2 r + N
(A.5)
3
3
Appendix B
We present here the main steps involved in the determination of the operator of small
fluctuations since the nontrivial procedure can be useful in other analogous considerations.
Our starting point is the Hamiltonian

1
h 2

2
H=
(B.1)
dz P R 1 + R R ,
2g
2
where
P = 1/ 1 D 2 , D = D(R), and for convenience we set p(D) P R 1 + R 2 =

R 2 (1 + R 2 ) + B 2 p(B). Thus, since D = D(R), we first replace in the Hamiltonian
p(D) by p(B) and then perform the variation. Then after each variation we can return to
expressions in terms of D. Proceeding in this way the first variation yields
R(1 + R 2 )
hR
H
=
,
R
2gp(B)
2g
H
R2 R
.
=
R 2gp(B)
(B.2)
From
H
d H
(B.3)
=0
R
dz R
we obtain the equation of motion which when integrated and with integration constant
chosen equal to zero implies the relation
h
R
R 2 = 0.
(B.4)
P 1 + R 2 2
It is at the configuration given by this equation that the second variation of H is to be
evaluated. First we obtain

1 + R 2 R 2 (1 + R 2 )2
2 H
1
1 D2 1 + R 2
h
(B.5)
=
h =
2g p(B)
2g P
R
R 2
p(B)3
and similarly (omitting now the intermediate step)
2 H
1 R(1 + D 2 R 2 )
=
,
R 2 2g P (1 + R 2 )3/2
2H
1 R (D 2 + 1)
2H
.
=
=
RR 2g P 1 + R 2 R R
(B.6)
These expressions are inserted into the second variation and give

2

1 1
2 H
2 H
H
2

2
2 H =
(R
)
+
2
RR
+
(R)
dz
2g 2
RR
R 2
R 2

1
D2
=
dz[A] +
dz[B],
4gP
4gP
211
(B.7)
where

R
R
dR
dR 2
hP (R)2 ,
A=
+
2
R

2
3/2

2
(1 + R )
dz
dz
1+R
RR 2
dR 2
1 + R 2
R
dR
B=
+
(R)2 .
+
2
R

2
dz
R
(1 + R 2 )3/2 dz
1+R
(B.8)
By writing

d
d 1
d R
R = R
R
dz
dz R
dz R
(B.9)
one can arrive after some manipulations at the expression
2

2
dR 2
2 d R
d (R)
=R
+
R
.
dz
dz R
dz R
(B.10)
We insert this expression into the expressions for A and B. Considering first the quantity
A and ignoring total derivatives, we can rewrite this as

d R 2
RR 2
+ V (R)2 ,
A=
(B.11)
(1 + R 2 )3/2 dz R
where
V =
RR
R
d
1 d
hP .
R dz (1 + R 2 )3/2 dz 1 + R 2
(B.12)
One can show that for the solutions of Eq. (B.4) V = 0. Considering now the quantity B
and ignoring total derivatives, we can rewrite this as

RR 4
d R 2
B=
(B.13)
+ U (R)2 ,
(1 + R 2 )3/2 dz R
where
d
1 d RR 2 R
R
+
U =

2
3/2
R dz (1 + R )
dz 1 + R 2
1 + R 2
.
R
(B.14)
One can show that for the solutions of Eq. (B.4) U = 0. Thus finally we are left with
(replacing dz by dR/R )

1
dR RR 2 (1 + D 2 R 2 ) d R 2
.
2 H =
(B.15)
4gP
R
dz R
(1 + R 2 )3/2
212
Thus this remaining term is positive definite for positive derivative R , and negative for
negative R . Thus for solutions R with R positive the energy is minimised and for those
with R negative it is maximised. We can now write the second variation

1

2 H =
(B.16)
dz R MR,
2
is
where the fluctuation operator M
=
M
1 1 d RR 2 (1 + D 2 R 2 ) d 1
.
2gP R dz (1 + R 2 )3/2 dz R
(B.17)
Appendix C
The action I is given by Eq. (7) with the Lagrangian (8) and in this the magnetic field
B is given by Eq. (11). Then we make a Wick rotation by setting t = i . The resulting
Euclidean action is

IE = d dz d LE ,
(C.1)
where (dots now refer to Euclidean time)

h 2
1
2
2

2
2
2
LE =
R
1
+
R
+
B
1
+
R
+
+R
,
R
4 2 g
2
(C.2)
with magnetic field B given by

D 2 R 2 1 + R 2 + R 2
.
1 + R 2 1 + R 2 D 2
The induced metric is

0
RR
1 + R 2
g =
1 + R 2 0
R R
0
0
R2
and the field tensor (cf. Eq. (6))

0
0
0
2F = 0
0
B .
0 B 0
Now we make the right angle rotation to
B2 =
z z = ,
= z
(C.3)
(C.4)
(C.5)
(C.6)
so that (omitting tildes on R for simplicity)

R R,
R R,
R R .
(C.7)
becomes
The magnetic field B
2 2
2
2
2 = D R 1 + R + R
B
1 + R 2 1 + R 2 D2
(C.8)
and the induced metric g

1 + R 2 RR
g = R R 1 + R 2
0
0
and the field tensor

0
0
=
F
0
0
0
B
0
0
R2

B
0 .
0
These expressions together yield the Lagrangian

1
2 1+R
2 + R 2 + B
2 1 + R 2 + h R 2
LE =
R
4 2 g
2
and the action
IE = d d z d L E .
213
(C.9)
(C.10)
(C.11)
(C.12)
Since the Lagrangian is a Lorentz scalar the result could also have been written down
directly from (A.5). For the instanton solution R = 0, and the Lagrangian assumes the
following form which demonstrates its equivalence with that of the static solution:

1
2 + h R2 .
1
+
R
P
R
LE =
(C.13)
2
4 2 g
References
[1] C.G. Callan, J.M. Maldacena, Brane dynamics from the BornInfeld action, Nucl. Phys. B 513 (1998) 198,
hep-th/9708147.
[2] R. Emparan, BornInfeld strings tunneling to D2-branes, Phys. Lett. B 423 (1998) 71, hep-th/9711106.
[3] D.K. Park, S. Tamaryan, Y.-G. Miao, H.J.W. Mller-Kirsten, Tunneling of BornInfeld strings to D2-branes,
Nucl. Phys. B 606 (2001) 84, hep-th/0011116.
[4] Y. Hyakutake, Torus-like dielectric D2-branes, JHEP 0105 (2001) 013, hep-th/0103146.
[5] D. Mateos, P.K. Townsend, Supertubes, Phys. Rev. Lett. 87 (2001) 011602, hep-th/0103030.
[6] D. Bak, K. Lee, Noncommutative supersymmetric tubes, Phys. Lett. B 509 (2001) 168, hep-th/0103148.
[7] R. Emparan, D. Mateos, P.K. Townsend, Supergravity supertubes, JHEP 0107 (2001) 011, hep-th/0106012.
[8] I. Bena, The polarization of F1 strings into D2 branes, hep-th/0111156.
[9] Y. Hyakutake, Expanded strings in the background of NS5-branes via a M2-brane, a D2-brane and D0branes, hep-th/0112073.
[10] R.C. Myers, Dielectric branes, JHEP 9912 (1999) 022, hep-th/9910053.
[11] N.R. Constable, R.C. Myers, O. Tafjord, The noncommutative BIon core, Phys. Rev. D 61 (2000) 106009,
hep-th/9911136.
[12] C. Bachas, M. Douglas, C. Schweigert, Flux stabilisation of D-branes, JHEP 0005 (2000) 048, hepth/0003037.
[13] D.K. Park, S. Tamaryan, H.J.W. Mller-Kirsten, J.-Z. Zhang, D-branes and their absorptivity in BornInfeld
theory, Nucl. Phys. B 584 (2001) 243, hep-th/0005165;
See also: R. Manvelyan, H.J.W. Mller-Kirsten, J.-Q. Liang, Y. Zhang, Absorption cross section of scalar
field in supergravity background, Nucl. Phys. B 579 (2000) 177, hep-th/0001179.
214
[14]
[15]
[16]
[17]
P.K. Townsend, D-branes from M-branes, Phys. Lett. B 373 (1996) 68, hep-th/9512062.
J. Schwarz, TASI lectures on non-BPS D-brane systems, hep-th/9908144.
S. Coleman, Fate of the false vacuum: semiclassical theory, Phys. Rev. D 15 (1977) 2929.
G.W. Gibbons, D.A. Rasheed, Electric-magnetic duality rotations in non-linear electrodynamics, Nucl. Phys.
B 454 (1995) 185, hep-th/9506035.
[18] G.W. Gibbons, K. Hashimoto, Non-linear electrodynamics in curved backgrounds, JHEP 0009 (2000) 013,
hep-th/0007019.
[19] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, 1965, p. 285, formula
3.196(1).

Super-Liouville theory with boundary

Takeshi Fukuda, Kazuo Hosomichi
Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
Abstract
We study N = 1 super-Liouville theory on worldsheets with and without boundary. Some
basic correlation functions on a sphere or a disc are obtained using the properties of degenerate
representations of superconformal algebra. Boundary states are classified by using the modular
transformation property of annulus partition functions, but there are some of those whose wave
functions cannot be obtained from the analysis of modular property. There are two ways of putting
boundary condition on supercurrent, and it turns out that the two choices lead to different boundary
states in quality. Some properties of boundary vertex operators are also presented. The boundary
degenerate operators are shown to connect two boundary states in a way slightly complicated than
the bosonic case. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Conformal field theories on worldsheets with boundary play an important role in
understanding some aspects of string theory, since they give worldsheet descriptions
of D-branes. Conformal field theories in general have large symmetry which includes
Virasoro symmetry, and they are generated by holomorphic currents. On worldsheets
without boundary there are two copies of the same symmetry algebra corresponding to the
left- and the right-moving sectors, and on the boundary of worldsheets the two are related
to each other by certain boundary condition which preserves one copy of the symmetry
algebra. From the representation theoretical point of view the classification of boundary
states reduces to that of possible boundary conditions and their solutions.
On the other hand, some conformal field theories are endowed with Lagrangians, and
they are also available even for worldsheets with boundary if suitable boundary terms are
E-mail addresses: tfukuda@yukawa.kyoto-u.ac.jp (T. Fukuda), hosomiti@yukawa.kyoto-u.ac.jp
(K. Hosomichi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 7 - 7
216
T. Fukuda, K. Hosomichi / Nuclear Physics B 635 (2002) 215254
incorporated. From this viewpoint, the classification of boundary states corresponds to that
of possible boundary terms along with the boundary conditions on the fields.
These two viewpoints have been shown to be consistent in [1,2] for Liouville theory. The
boundary states are classified in a complete way through the analysis of modular property
of annulus partition functions, and some boundary states correspond to the addition of a
boundary interaction term with certain values of coupling constant. It was also the first case
where the classification of Cardy states was done for non-compact CFTs having continuous
spectrum of representations. Based on this idea the Liouville theory with boundary has
been analyzed in [3,4], and the analysis of boundary states has also been made recently for
more involved CFTs such as CFT on Euclidean AdS3 [58].
In this paper we consider the N = 1 supersymmetric extension of Liouville theory
in the presence of boundary, using the techniques developed in [1,2]. This theory has
been analyzed for decades and some old references include [916]. One will face some
complexity due to the presence of NS and R sectors, and a careful analysis reveals what
kind of new features arises as a result of supersymmetrization. For the case without
boundary, the exact results for basic correlators have been obtained in [17,18]. There have
also been some recent works on the case with boundary [19]. Since super-Liouville theory
is one of the simplest CFTs with N = 1 worldsheet supersymmetry, our result should
contain many of the properties which all the N = 1 supersymmetric CFTs have in common.
This paper is organized as follows. In Section 2 we analyze the N = 1 super-Liouville
theory without boundary, especially on a sphere. The calculation of basic correlation
functions which has been done in [17,18] is reviewed. We first summarize the spectrum
of degenerate representations in N = 1 superconformal algebra. Then we calculate various
structure constants using the most fundamental degenerate operators which will be denoted

as b/2
, under the reasonable assumption that the product of them with any operators are
expanded into two discrete terms. The consistency of this assumption is investigated by
. In Section 3
solving the differential equation for four-point functions containing b/2
we analyze the theory on worldsheets with boundary, especially on a disc. The modular
property of annulus partition functions is investigated, from which we obtain the wave
functions for some Cardy states. There are some others whose wave functions cannot be
determined from the analysis of modular property, and we determine them through the
analysis of disc one-point functions. The two-point functions of boundary operators are
also obtained. The results for reflection coefficients are consistent with the argument of
density of open string states, but it turns out that the reflection coefficients differ for each
operator in a single supermultiplet. The last section gives a brief summary of our results
and some discussions.
2. N = 1 super-Liouville theory
The supersymmetric extension of Liouville theory was found in [20]. It is described by
a boson and its superpartner , and the action on flat worldsheet reads

1
+ 2i d 2 z d d eb ,
I=
(2.1)
d 2 z d d D D
2
217
where we employed the superfield formalism
= + i + i + i F,
D = + z ,
= + z .
D
(2.2)
The reader should note that there is a linear dilaton coupling hidden in the action. In [21] it
was analyzed as a two-dimensional theory of supergravity with superconformal symmetry.
Superfield expression for linear dilaton coupling can be found in [22].
We regard this theory as a free CFT of and with a linear dilaton coupling,

1
+ QR +
+
F 2 ,
Q = b + b1
d 2 z
I=
(2.3)
4
2
perturbed by the following interaction

2i d 2 z ibF eb + b2 eb .
(2.4)
In what follows we shall neglect the auxiliary field F which yields a contact interaction,
assuming the analyticity of correlators or OPEs that allow us to calculate any quantity by
the continuation from the region where contact interactions can be neglected. See [14,23,
24] for more detailed argument on this point. Thus we shall treat the super-Liouville theory
as the free CFT of and perturbed by

2
d 2 z eb .
Sint 2ib
(2.5)
The stress tensor T and the supercurrent TF of the free theory are given by the Feigin
Fuchs representation

1
Q 2 + ,
2
TF = i( Q).
T =
They satisfy the super-Virasoro algebra with c =
(2.6)
3c
2
= 32 (1 + 2Q2 )
3c
2T (0) T (0)
,
+
+
4
4z
z2
z
3TF (0) TF (0)
,
T (z)TF (0)
+
2z2
z
c
2T (0)
.
TF (z)TF (0) 3 +
(2.7)
z
z
We shall concentrate on the left-moving sector for the time being, in order to clarify
the symmetry structure of the theory. We work with the primaries V = e and their
superpartners = ie , which satisfy
T (z)T (0)
h V (0) V (0)
,
+
z2
z

h + 12 (0) (0)
,
T (z) (0)
+
z2
z
(0)
,
TF (z)V (0)
z
T (z)V (0)
218
2h V (0) V (0)
+
(2.8)
,
z
z2
with h = (Q )/2. They are NS vertices and correspond to spacetime bosons. We
also consider the R vertices corresponding to spacetime fermions, which are given by
spin fields = e . They obey the following transformation property

1
(0) (0)
h + 16
,
+
T (z) (0)
z
z2
p (0)
i(Q 2)
TF (z) (0)
(2.9)
.
+ ,
p =
3/2
2
2z
TF (z) (0)
The most important property of spin fields is that the supercurrent TF becomes doublevalued around them. The spin fields are defined to satisfy
(0)
(z) (0)
(2.10)
.
2 z1/2
We can analyze the theory perturbatively, by expanding any quantity as a power series
in the cosmological constant . However, due to the momentum conservation in Linear
dilaton theory, any correlators of operators of definite Liouville momentum have only one
contribution from a specific order of . This can easily be seen by employing the path
integration approach and perform the integration over the zero-mode of as discussed in
[25]. Then we find, for example,

1
N
Vi (zi ) = b #(N)
Vi (zi )Sint
,
(2.11)
i
Wick
where the suffix Wick represents the ordinary Wick contraction with respect to free fields
and N is defined by

bN = Q(1 g)
(2.12)
i
i
for worldsheets with g handles. Although the expression (2.11) can be used to evaluate
correlators by first assuming N to be a non-negative integer and then extending the result
to generic N , we do not use it this way. We would rather read off from it one important
property that the correlator diverges when a non-negative integer insertions of Sint can
screen the non-conserving Liouville momentum, and the residue of the divergence is given
by the free field correlator with an appropriate number of Sint inserted.
2.1. Two-point functions on a sphere
Here we rederive the basic correlation functions on a sphere which were obtained
by [17,18] as an introduction to our method to analyze the theory on a disc. The most
important among them are the two-point functions. We first consider the following one:

V1 (z1 )V2 (z2 ) = |z12 |4h1 2 (p1 + p2 ) + (p1 p2 )D(1 )

Q
i = + ipi .
(2.13)
2
219
The global superconformal symmetry yields some relations between two-point correlators:
V1 (z1 )V2 (z2 ) ,

1 2 V1 (z1 )V2 (z2 ) = 2h1 z12

1 (z1 )V
2 (z2 ) = 2h1 z 1 V1 (z1 )V2 (z2 ) ,
1 2 V
12

12 22 V1 (z1 ) V2 (z2 ) = 4h21 |z12 |2 V1 (z1 )V2 (z2 ) ,
(2.14)
so that all the two-point functions of NSNS sector vertices are described by a single
structure constant, D(). To study the two-point functions of RR-sector vertices, we adopt
the following convention. We first define the spin fields in the right-moving sector by
the equations
i (0)
.
(0)(z)
2 z 1/2
Then the spin fields,
(z, z ) e (z, z ) e (z, z )
(2.15)
(, = )
(2.16)
are shown to satisfy the OPE relations

p , (0)
,
2 z3/2
p , (0)
i (0)TF (z)
.
2 z 3/2
TF (z), (0)
(2.17)
We assume that + , + commute and , anti-commute with fermions. Thus

commute with fermions while anti-commute. Using this, the superconformal Ward
identity becomes

dw

(w z)1/2 (w z )1/2 TF (w) (z) (z ){ }
2i

,
1

(z ){ }
= (z z )1/2 p , (z) (z ){ } + i p (z)

2

dw

(w z)1/2 (w z )1/2 TF (w){ },
+ (z) (z )
2i

d w

(w z )1/2 (w z )1/2 { } (z) (z )TF (w)
2i

1
= (z z )1/2 { }p (z) , (z ) + { }i p , (z) , (z )
2

dw

(w z)1/2(w z )1/2 { }TF (w)
+
(2.18)
(z) (z ).
2i
The normalization of spin fields is given by the following correlators:

(z) (0) free = |z|1/4,
(z) (0) free = i|z|1/4,
(2.19)
where one should be careful for that only Grassmann-even combinations can have
non-vanishing correlators. Note also that all these are related via superconformal
220
transformations. From this, we put the following ansatz for the two-point functions of RR
vertices:

1
i
(z1 )
(z2 ) =
(z1 )
(z2 ) = |z12 |4h1 4 2(p1 + p2 ),
1
2
1
2

(z2 ) =
(z1 )
(z2 )
i 1 (z1 )
2
1
2
1 ).
= |z12 |4h1 4 2(p1 p2 )D(
1
(2.20)
These two-point functions give relations between operators carrying Liouville momentum
and Q :
D() =
V
VQ

D()
= ,
.
Q

V
V
2 V
=
=
,
(Q )VQ
(Q ) VQ
(Q )2 VQ
(2.21)

They are referred to as the reflection relation in what follows. D(), D()
are called the
reflection coefficients.
The ansatz (2.20) for the two-point functions of spin fields might seem peculiar at first
sight, because it leads to the reflection relation which flips the indices , as well as the
momentum. In the following we obtain the reflection coefficients using some properties of
degenerate primary fields, and there we will convince ourselves that the above ansatz is the
only one which is consistent with the OPEs involving degenerate operators.
Degenerate fields and their OPEs
Let us summarize here some basic properties of operators belonging to degenerate
representations of superconformal algebra. As was found in [2628], they are given by
the following Liouville momentum r,s

1
Q rb sb1 ,
(2.22)
2
in close analogy with the case of bosonic Liouville theory. The difference is that the
degenerate representations with odd r + s sit in the R sector, while those with even r + s
are in the NS sector. They are known to have null states at level rs/2. The corresponding
vertex operators are Vr,s (r + s even) or r,s (r + s odd).
(and 1/2b
)
We will frequently use the most fundamental degenerate operators b/2
in the following analysis. The property of the corresponding degenerate primary states
|2, 1 are summarized as follows:
i(2b 2 + 1)
i 2
G1 |2, 1 +
L1 |2, 1 = 0.
G0 |2, 1 =
(2.23)
|2, 1 ,
b
2 2b
r,s
We first discuss the OPEs involving these degenerate operators to find out the expressions
for reflection coefficients. Then in later sections we will give a detailed analysis of the
four-point functions involving them by solving the associated differential equations.

with arbitrary primary fields or spin
Here we just assume that the OPEs of b/2
fields yield only two discrete terms. This assumption will be justified in later sections by
221
analyzing the four-point functions. We begin with one particular example:

++
++
b/2
(z1 )V (z2 ) |z12 |b C+ ()b/2
(z2 ) + |z12 |b(Q)C ()+b/2
(z2 ),
(2.24)
where the coefficients C () are calculable using the techniques of free CFT as proposed
in [29] and utilized in the analysis of bosonic Liouville theory in [1]:
C+ () = lim
z1 z2
C () = lim
++
++
Q+b/2
(w)b/2
(z1 )V (z2 )Wick
++
++
|z12 |b Q+b/2
(w)b/2
(z2 )Wick
= 1,
++
Qb/2
(w)b/2
(z1 )V (z2 )(Sint )Wick
|z12 |b Qb/2
(w)+b/2
(z2 )Wick

bQ
= b2 bQ
2 (1 b) b 2 .
z1 z2
(2.25)
Here we introduced the notation (x) #(x)/ #(1 x). The free field correlator in the
numerator can be evaluated by using

(z1 ) (z2 ) (z3 ) = i (z1 ) (z2 ) (z3 )
i
= |z12 |3/4|z13 z23 |1 ,
(2.26)
2
which follows from (2.10), (2.15) and (2.19). Summarizing similar OPE relations we
obtain
,

b/2
(z1 )V (z2 ) |z12 |b b/2
(z2 ) + |z12 |b(Q)C ()+b/2
(z2 ).
(2.27)
Taking the superconformal transformation of both sides we obtain

1/2
2 z21 b/2 (z1 )V (z2 )
,
,
(z2 ) |z12 |b(Q)(Q )C ()+b/2
(z2 ),
|z12 |b b/2
1/2
i 2 z 21 b/2 (z1 ) V (z2 )
,
,
|z12 |b b/2
(z2 ) |z12 |b(Q)(Q )C ()+b/2
(z2 ),

(z1 ) 2 V (z2 )
2i|z21|b/2
,

|z12 |b 2 b/2
(z2 ) + |z12 |b(Q)(Q )2 C ()+b/2
(z2 ).
(2.28)
Combining them with the reflection relations (2.21), we obtain the following recursion
relations between structure constants:
+ b/2),
D() = C ()D(
b/2) = C (Q )D().
D(
(2.29)

To find the OPE of b/2
with generic spin fields, let us start with
+
+ ()|z12 |b+ 4 Vb/2 (z2 )
2b/2
(z1 )+ (z2 ) C
3
()|z12 |b(Q) 4 V+b/2 (z2 ).

+C
1
(2.30)
222
The free field integrals with screenings give

+ () = 1,
C

() = 2ib2 bQ 1 b b
C
2
2
b2
2
= 2iC (Q b/2).
(2.31)
Collecting similar OPE relations we have
2b/2
(z1 ) (z2 )
(z1 ) (z2 )
2ib/2
()|z12 |b(Q) 4 V+b/2(z2 ).

|z12 |b+ 4 Vb/2(z2 ) + C
3
(2.32)
Taking its superconformal transformations we obtain
2 b/2
(z1 ) (z2 )

(z1 ) (z2 )
i 2 b/2
1
() 1
1
1
(2 + b)C
2
2
b/2 (z2 ) +
z12
|z12 |b 4 V
|z12 |b(Q) 4 V+b/2 (z2 ),
z 12
2(2Q 2 b)
2 b/2 (z1 ) (z2 )

i 2 b/2
(z1 ) (z2 )
1
2
z12
|z12 |b 4 Vb/2 (z2 )
() 1
1
(2 + b)C
+b/2(z2 ),
z 2 |z12 |b(Q) 4 V
+
2(2Q 2 b) 12
(z1 ) (z2 )
b/2
(z1 ) (z2 )
ib/2
()
3
(2 + b)2 C
|z12 |b(Q)+ 4 V+b/2(z2 ).
4(2Q 2 b)2
(2.33)
Combining them with the reflection relations (2.21), we obtain the same set of recursion
relations as (2.29). Here one can also see that the reflection of the Liouville momentum of
spin fields should be accompanied by the flip of signs , ,
because in the right-hand side
of the above OPEs there are always two NS operators of opposite fermion number (when
counted for either one of the left/right sectors separately).
A simple solution of (2.29) satisfying the unitarity
1
|z12|b 4 Vb/2(z2 ) +
D(Q
) = 1
D()D(Q ) = D()
and exhibiting the b 1/b duality can be found rather easily:
1

Q

Q2 # b Q
2 # b 2
b
,
D() = (bQ/2)
1

Q
# b Q
2 # b 2
1 1

Q

Q2 # 12 + b Q
#
+
2
2
b
2

D()
= (bQ/2) b
.

1

Q 1
Q
1
# 2 b 2 # 2 b 2
(2.34)
(2.35)
(2.36)
223
Before moving on to the analysis of three-point functions, we note that the correlators
involving NSR vertices are difficult to analyze since the screening operator becomes
double-valued around them. Therefore, they are not well-defined vertex operators from
the viewpoint of a perturbed free CFT.
We also note that the following relations for products of two spin fields hold:
(z1 ),
(z2 ) = i ,
(z1 ),
(z2 ), (2.37)
1 (z1 )2 (z2 ) = i ,
1
2
1
2
in all the expressions for two-point functions and OPEs obtained so far. We assume this
to hold in arbitrary correlation functions containing two spin fields. As an evidence, the
analysis of four-point functions becomes much simpler without any contradiction if we
employ this relation. However, one should not expect this relation to hold in correlators
containing more than two spin fields.
2.2. Three-point functions on a sphere
We first put the following ansatz for them:

V1 V2 V3 = C1 (1 , 2 , 3 ),

i1 V1 V2 V3 = C2 (1 , 2 , 3 ),

2 V1 V2 V3 = C3 (1 , 2 , 3 ),

1 (1 ; 2 , 3 ),
V1
=C
2
3

2 (1 ; 2 , 3 ),
=C
V1
2
3

3 (1 ; 2 , 3 ),
=C
V1
1
2
3
(2.38)
where we omit the coordinate dependence which can easily be restored knowing the
conformal weights of operators:

O1 (z1 )O2 (z2 )O3 (z3 )
h h1 h2 h1 h2 h3 h2 h3 h1 h 3 h 1 h 2 h 1 h 2 h 3 h 2 h 3 h 1
z23
z13
z 12
z 23
z 13
.
z123
(2.39)
Other three-point functions are related to the above expressions via leftright symmetry,
or are obtained by taking the superconformal transformations of them. Using the
superconformal symmetry we can also find that C1,2,3 (i ) are symmetric in the three
arguments.

To obtain them we again use the degenerate field b/2
. We focus on the left-moving
sector of the theory and consider the four-point functions in chiral CFT:

(z0 )1 (z1 ) = f00
(zi ),
V3 (z3 )V2 (z2 )b/2

(zi ),
i2 V3 (z3 )V2 (z2 )b/2 (z0 )1 (z1 ) = f01

i3 V3 (z3 )V2 (z2 )b/2 (z0 )1 (z1 ) = f10 (zi ),

(z0 )1 (z1 ) = f11
(zi ).
2 3 V3 (z3 )V2 (z2 )b/2
(2.40)
As was shown in [17] and reviewed in the following, a set of differential equations
among them can be derived from the superconformal symmetry and the degeneracy of
b/2 which has two independent solutions. This justifies the assumption in the previous

paragraph that the OPEs involving b/2
yield only two discrete terms. However, in order
to obtain the three-point structure constants it is sufficient to know that the first four of the
224
, satisfy an ordinary hypergeometric differential equation. If we simply

above sixteen, f00
assume this we can skip the lengthy analysis of the differential equation and easily write
down the solution, since the behavior when z0 approach either of z1,2,3 is known from the
OPE formula.
Differential equation for correlators V V

Firstly, the above sixteen functions are related via global superconformal transformadw
tion. By multiplying 2i
(w z0 )1/2 (w z1 )1/2 TF (w) onto the product of operators in
the bracket we find
1/2
z01
1/2 1/2
1/2 1/2
,
(pb/2 p1 )f00
z30 z31 f10
+ z20 z21 f01
+
= 0,
2

1/2
z01
h2
h2
1/2 1/2
1/2 1/2
,

z30 z31 f11 + z20 z21 2 +
+

= 0,
f00
(pb/2 p1 )f01
z20 z21
2

1/2
z01
h3
h3
1/2 1/2
1/2 1/2
,

z30 z31 3 +
+
z20 z21 f11

= 0,
f00
(pb/2 p1 )f10
z30 z31
2

h3
h3
h2
h2
1/2 1/2
1/2 1/2

f01
f
z30 z31 3 +
+
z20 z21 2 +
+
z30 z31
z20 z21 10
1/2
z01
,
(pb/2 p1 )f11
+
= 0.
2
(2.41)
Here the relation (2.37) was assumed. By introducing a new set of functions Fij () of the
z23
cross ratio = zz01
:
03 z21
=
f00

0j = 2h0 1/8,
,
zijij F00
i>j

f01
=

=
f10

=
f11
1/2
2 z
z30
21

zijij F01
,
2 z z
z32
01 20
i>j
1/2
2
z30 z21

zijij F10
,
2
z01 z31 z32
i>j
z30 z21
2
z20 z31 z32
1/2
i>j
,
zijij F11
j =0

1j = 2h1 1/8,
j =1

2j = 2h2 ,
j =2

3j = 2h3 .
j =3
The above relations can be rewritten in a simpler way:

(1 ) D
1
,

F10
+ F01
+ (pb/2 p1 )F00
= 0,
2

1
,

D + 23 + (h2 + 21 ) F00
(pb/2 p1 )F01
= 0,
F11
2
(2.42)
225

1
,

D + 23 + (h3 + 03 ) F00
F11
(pb/2 p1 )F10
= 0,
2

D + 23 1 + (h3 + 03 + 1) F01
+ D + 23 1 + (h2 + 12 + 1) F10
1
,
+ (pb/2 p1 )F11
= 0.
(2.43)
2
Secondly, we derive another set of differential equations by translating the equation for
null states

bpb/2
2b2 + 1
G1 |2, 1 = L1 G0 +
G1 |2, 1 = 0
L1 G0 +
(2.44)
4
2i
into a differential equation for correlators involving the corresponding degenerate operator.
Here we deal with correlators involving only two spin fields one of which is degenerate,
and use the following translation law:

dz2 1 n+1 n+1

z z z
Ln b/2 (z0 )1 (z1 ) =
T (z2 )b/2
(z0 )1 (z1 ),
2i 01 20 21
z0

dz2 1/2 n+1/2 n+1/2

Gn b/2
z
(z0 )1 (z1 ) =
z
z21
TF (z2 )b/2
(z0 )1 (z1 ),
2i 01 20
z0
(2.45)
and write down the differential equation in the following way:

(z0 )1 (z1 ) = 0.
+ G1 b/2
2 ib1 L1 b/2
(2.46)
This yields a set of differential equations for the functions Fij :

1 1
F01 + F10 ,
+
0 = i 2 b1 D 02 (1 )03 F00

1
1

D 02 + (1 )(03 + 1) F01
0 = i 2 b
2

(1 )F11
+ D + 23 + (12 + 3h2 ) F00
,

1
1

0 = i 2 b
D 02 (1 ) 03 +
F10
2

(1 ) D + 23 + (03 h3 ) F00
+ F11
,

1
1

F11
0 = i 2 b1 D 02 + (1 ) 03 +
2
2

1
D + 23 1 + (03 + 1 h3 ) F01
+

1
+ D + 23 1 + (12 + 1 + 3h2 ) F10
.
(2.47)
Obviously, (2.43) and (2.47) are redundant. Indeed, (2.43) yields only two relations

between {F00
, F01
, F10
, F11
}, so that we are left with two independent functions
out of four after imposing the superconformal symmetry. (2.47) also contains only
226
two independent equations, giving a second-ordered differential equation for one of the

remaining two independent functions. Thus the solution for f00
is written in terms of the
hypergeometric function,
h h1 h2 h0 1/8 h1 h2 h3 +h0 +1/8 h2 h3 h1 +h0 2h0 1/8
z23
z13
z03
01
02

f00
(z0,1,2,3) = z123
(1 ) F (A, B; C; ),
b3
1
+ 2h0 + ,
A = 01 + 02 +
2
8
b(Q 3 )
1
B = 01 + 02 +
+ 2h0 + ,
2
8
1
C = bQ + 201 + 4h0 + ,
4
bQ b(21 Q) 3
bQ + b(21 Q) 1
01 =
or
+ ,
4
8
4
8
b(Q 2 )
b2
02 =
(2.48)
or
.
2
2
The two choices for 01 give two independent solutions, while two choices of 02 lead to
the same solution owing to the formula of hypergeometric functions.
By taking its square in a crossing symmetric way we can construct the four-point
function on a sphere involving a degenerate field,

++
(z0 )
(z1 )
V3 (z3 )V2 (z2 )b/2
1
= |z12 |2(h3 h1 h2 h0 1/8) |z23 |2(h1 h2 h3 +h0 +1/8)
|z13 |2(h2 h3 h1 +h0 ) |z03 |4h0 1/4

P1 G1 (1 , 2 , 3 ; )G1 (1 , 2 , 3 ; )

+ P2 G2 (Q 1 , 2 , 3 ; )G2 (Q 1 , 2 , 3 ; )
,
(2.49)
where G1,2 are given by (we use the notation like p12+3 p1 p2 + p3 )

+ 34 ; ibp1 + 32 ; ,
b1 1
b2

1 ib
1
1
G2 (i ; ) = 2 8 (1 ) 2 F ib
2 p+1+2+3 + 4 , 2 p+1+23 + 4 ; ibp1 + 2 ; ,
G1 (i ; ) =
b1 3
2 +8
(1 )
b2
2
ib
2
p+1+2+3 + 34 ,
ib
2 p+1+23
(2.50)
and the coefficients P1,2 can be fixed from the crossing symmetry up to an overall
normalization:

2
2

P1
= 12 + ibp1 12 ibp1 34 + ib
p+1+2+3
2
P2

3 ib
3 ib

34 + ib
2 p+1+23 4 + 2 p+12+3 4 + 2 p+123 .
(2.51)
Considering in the same way, we find that there is no crossing symmetric solution for
++
+
, so that they must vanish. The correlator
four point functions like V3 V2 b/2
1
++
++
V3 V2 b/2 1 is obtained simply by replacing 1 with Q 1 in the above.
227
From the above solutions we can derive a recursion relation for three-point structure
i (1,2,3 ). From the fact that some four-point functions vanish it
constants Ci (1,2,3 ) and C
follows that
2 (1 , 2 , 3 ) = 0.
C2 (1 , 2 , 3 ) = C
(2.52)
By comparing the limit z0 z1 of (2.49) with the OPE formula (2.32) we find

+ (1 )C3 1 b , 2 , 3
C
P1
1
2
(p1 , p2 , p3 ) =

,
2
b
P2
(1 )C1 1 + 2 , 2 , 3
1 b2 C

(1 )C3 1 + b , 2 , 3
C
P1
1
2
(p1 , p2 , p3 ) =

.
+ (1 )C1 1 b , 2 , 3
P2
(2Q 21 b)2 C
(2.53)
These give a set of recursion relations for C1 (1 , 2 , 3 ) and C3 (1 , 2 , 3 ), whose

solution can be expressed in terms of function introduced in [30,31]. It is defined by

ln (x) =
2

sinh2 [(Q/2 x)t]
dt 2t Q
x
e
,
t
2
sinh[bt] sinh[t/b]
(2.54)
and satisfies the following relations

(x + b) = (x)b12bx (bx),

x + b1 = (x)b2x/b1 (x/b),
(x) = (Q x).
(2.55)
It has zeroes at
(x) = 0 at x = mb nb1 ,
If we define
NS (x) =
x x+Q
,
2
2
x = Q + mb + nb1
R (x) =
x+b x+b1
,
2
2
(m, n Z0 ).
(2.56)
(2.57)
the solution for the recursion relation can be expressed as

(0)
NS
NS (1+2+3 Q)
NS (21 )NS (22 )NS (23 )
,
NS (1+23 )NS (2+31 )NS (3+12 )

(0)

1b2 (Q:i )/b
2NS
C3 (i ) = i bQ
b
2
R (1+2+3 Q)
NS (21 )NS (22 )NS (23 )
(2.58)
,
R (1+23 )R (2+31 )R (3+12 )
where we used the notations like 1+23 1 + 2 3 . The functions NS , R are also
useful to construct reflection-symmetric quantities because of the relations

1b2 Q2
NS (2)
b
,
D() = bQ
2 b
NS (2Q 2)

1b2 Q2
R (2)
b

D()
= bQ
(2.59)
.
2 b
R (2Q 2)

1b2 (Q:i )/b
C1 (i ) = bQ
2 b
228
The three-point structure constants containing spin fields can be obtained in a similar
way. This time we take the limit z0 z2 of the solution (2.49). Using the formulae for
hypergeometric functions we find

++
(z0 )
(z1 )
V3 (z3 )V2 (z2 )b/2
1
= |z12 |2(h3 h1 h2 h0 1/8) |z23 |2(h1 h2 h3 +h0 +1/8)
|z13 |2(h2 h3 h1 +h0 ) |z03 |4h0 1/4

Q1 H (1 , 2 , 3 ; 1 )H (1 , 2 , 3 ; 1 )

,
+ Q2 H (1 , Q 2 , 3 ; 1 )H (1 , Q 2 , 3 ; 1 )
(2.60)
where H is given by
H (i ; 1 )
=
b1 3
2 +8
(1 )
b2
2
ib
2
p+1+2+3 + 34 ,
ib
2 p+1+23

+ 34 ; ibp2 + 1; 1 ,
(2.61)
and the ratio Q1 /Q2 reads

Q1
= b2 p22 (ibp2 )2 34 + ib
p1+2+3 34 + ib
p1+23
2
2
Q2
1 ib

14 + ib
2 p1+2+3 4 + 2 p1+23 .
(2.62)
By making a comparison with the OPE formula (2.27) we find recursion relations

1 3 ; 2 b , 1
C+ (2 )C
Q1
2
(p1 , p2 , p3 ) =

,
3 3 ; 2 + b , 1
Q2
C (2 )C
2

3 3 ; 2 b , 1
C+ (2 )C
Q1
2
(p1 , p2 , p3 ) =
(2.63)

.
1 3 ; 2 + b , 1
Q2
C (2 )C
2
This can be straightforwardly solved and we obtain

1 (3 ; 2 , 1 ) = bQ b1b2 (Q:i )/b
C
2
(0) (2 ) (2 ) (2 )
NS
NS
3 R
2 R
1
,
R (1+2+3 Q)R (1+23 )NS (2+31 )NS (3+12 )

3 (3 ; 2 , 1 ) = bQ b1b2 (Q:i )/b
C
2
(0) (2 ) (2 ) (2 )
NS
NS
3 R
2 R
1
.
NS (1+2+3 Q)NS (1+23 )R (2+31 )R (3+12 )
(2.64)
We can also check that these three-point structure constants are consistent with the
reflection symmetry.
Although these three-point structure constants (2.58) and (2.64) have complicated form,
each factor has a clear physical meaning. Note first that the zeroes of NS , R are at
NS (x) = 0 at x = mb nb1 ,
R (x) = 0
at x = mb nb
x = Q + mb + nb1
x = Q + mb + nb
(m + n even),
(m + n odd).
(2.65)
229
Each three-point structure constant therefore has eight sequences of poles. This agrees with
our naive expectation, since by combining the perturbative consideration, b b1 and the
reflection symmetry we can easily guess that, for example, C1 (i ) diverges at
(1 or Q 1 ) + (2 or Q 2 ) + (3 or Q 3 ) = Q mb nb1
(m + n even).
(2.66)
The last condition is because of the fermionic nature of the screening operator. The pole
structure of other structure constants can also be understood in the same way if we take
into account that the reflection of spin fields also flips their Grassmann parity.
Differential equation for correlators
Let us analyze here the correlation functions of four spin fields, one of which is
degenerate. The reason for this is that we will use the solution to obtain the one-point
structure constant on a disc in the next section. The solution can also be used to crosscheck the three-point structure constants which were obtained previously.
We first consider the holomorphic sector as in the previous case, and begin by
introducing some notations:

0
b/2
(z0 )11 (z1 )22 (z2 )33 (z3 ) f 0 1 2 3 (zi ).
(2.67)
Then we translate the expression for the null vector (2.44) into differential equation for
correlators. In doing this, note first that the Ramond algebra is generated by L1 and G1 .
So it suffices to give the rule of translation for these two:

dz4 2
0
0
z40 T (z4 )b/2
L1 b/2
(z0 )
ii (zi ) =
(z0 )
ii (zi ) ,
2i
i

0
(z0 )
G1 b/2

=
z0

ii (zi )
z0
dz4
z41 z42 z43
2i z40 z01 z02 z03
1/2
0
(z0 )
TF (z4 )b/2

ii (zi )
Using them, the rule for other generators can be obtained easily:

0
G0 b/2
(z0 )
ii (zi )
1
=
3

z0
dz4 z40 z41 z42 z43

2i
z01 z02 z03
1/2
z01 z02 z03

+
+
z41 z42 z43

0
TF (z4 )b/2
(z0 )
ii (zi ) ,
i
(2.68)
230

0
(z0 )
L1 b/2
1
3

z0

ii (zi )

dz4 z41 z42 z43 z01 z02 z03
0
T (z4 )b/2
+
+
(z0 )
ii (zi ) . (2.69)
2i z01 z02 z03 z41 z42 z43
i
Then the equation for the null vector (where we use the notation ),

0
0
2 ib1 L1 b/2
(z0 )
ii (zi ) + G1 b/2
(z0 )
ii (zi ) = 0,
i
(2.70)
can be recast into the form of a differential equation,

z12 z13 1/2
z21 z23 1/2
0
p1 f 0 1 2 3 + 0 1
p2 f 0 1 2 3
z10
z20

z31 z32 1/2
+ 0 1 2
p3 f 0 1 2 3
z30

2h0 + 18 1
2i
1
1
1/2
z01 + z02 + z03 f 0 1 2 3 .
= (z01 z02 z03 )
0 +
b
3
(2.71)
As in the previous case we rescale the functions f 0 1 2 3 in the following way:

ij
z01 z23
zij F 0 1 2 3 (),
=
,
f 0 1 2 3 =
z03 z21
i,j

i(=j )
1
ij = 2hj .
8
(2.72)
Then the above equation can be rewritten into the form

2ib10 DF 0 1 2 3 = p1 ( 1)1/2F 0 1 2 3 + 1 p2 ()1/2 F 0 1 2 3
+ 1 2 p3 1/2 (1 )1/2 F 0 1 2 3 ,

1 2bQ
1 2bQ
D = (1 ) + (1 ) 01 +
02 +
.
8
8
(2.73)
According to the signs of i , there are sixteen components of F 0 1 2 3 . The above

equations separate them into two groups, each containing eight components with even
(odd) number of minus signs in i . We try to reduce the number of independent components
further by putting the assumption
F 0 1 2 3 = c(0 , 1 , 2 , 3 )F 0 1 2 3 .
(2.74)
The consistency with (2.73) yields

c(0 , 1 , 2 , 3 ) = c( 0 , 1 , 2 , 3 ) = 1,
c(0 , 1 , 2 , 3 ) = c( 0 , 1 , 2 , 3 ) = c( 0 , 1 , 2 , 3 ) = c( 0 , 1 , 2 , 3 ).
(2.75)
231
Denoting c(+, +, +, +) = and F0,1,2,3 = (F ++++ , F ++ , F ++ , F ++ ) we have

2ib1DF0 = p1 ( 1)1/2F1 p2 ()1/2F2 p3 1/2(1 )1/2F3 ,
2ib1DF1 = +p1 ( 1)1/2F0 + p2 ()1/2 F3 p3 1/2(1 )1/2 F2 ,
2ib1DF2 = p1 ( 1)1/2F3 + p2 ()1/2 F0 + p3 1/2(1 )1/2 F1 ,
2ib1DF3 = +p1 ( 1)1/2F2 p2 ()1/2 F1 + p3 1/2(1 )1/2 F0 .
(2.76)
One can see that the above system of differential equations exhibits a symmetry in three
generic spin fields. It is also consistent with the reflection symmetry: for example, F1 obeys
the same equation as that for F0 with the signs of p2 and p3 flipped. Here we will not go
into any further detail to determine or choose explicitly one appropriate branch for each
square root, since different choices lead to different correlators and we would like to study
their mutual relation later in detail.
Let us step aside for a while and try another way to construct the correlators (2.67).
Recall that the correlators are characterized by the analyticity and the asymptotic behavior
around z0 z1,2,3 dictated by the OPE formula:
z0 zi
bi
f 0 1 2 3 (zi ) z0i2
18
bi
z0i2
+ 38
b(Qi ) 1
8
2
z0i
b(Qi ) 3
+8
2
z0i
(2.77)

In the previous analysis of reflection coefficients it was assumed that the OPEs of b/2
with generic spin fields (of definite Grassmann parity) yield only two discrete terms.
However, since the differential equation is of the fourth order, there should be four
independent solutions. Therefore, in solving the differential equation we should not adhere
to the idea that each spin field in any correlator has a definite Grassmann parity. Thus we
assume that the leading order behavior of the four independent solutions should be given
by (2.77).
Let us then put an assumption that the correlator can be expressed as a double contour
integral of the following form:
ij

zij
dw dw (zi w)i (zi w )i (w w )
f 0 1 2 3 (zi ) =
(2.78)
i<j
as one can guess by analogy with simpler cases where we encounter with hypergeometric
functions. The global conformal invariance yields the conditions on the exponents i , i ,
and ij :

1
i =
i = 2,
ij + i + i = 2hi .
(2.79)
8
i
j (=i)
Analyzing the behavior at, say, z0 z1 we find that the double integral approximately
breaks into several terms with different asymptotic behavior. In doing this, note first that
the limit z0 z1 can also be viewed as the limit z2 z3 . Then there are two possibilities
for w to be either near z0,1 or near z2,3 . Similarly, there are also two possibilities for w ,
so that they altogether give four terms in the limit z0 z1 :
f 0 1 2 3 (zi ) z0101 ,
01 +1+0 +1
z01
+1+0 +1
z0101
01 ++2+0 +1 +0 +1
z01
.
(2.80)
232
Comparing this with the OPE formula, we can easily find that must be 0 or 1. Among
1
3
these three, = 1 is the only solution consistent with h0 + 16
= 16
(1 2bQ). Going back
to the analysis of Eq. (2.76), we are thus lead to conjecture that after setting 01 = 02 = 0
in (2.73), the function F0 () should be given by
F0 (p1 , p2 , p3 ; )
=
b1 3
2 +8
(1 )
b2
2
+ 38

3/4
dw dw w (w 1)(w )
w 4 + 2 p1+2+3 (w 1) 4 + 2 p12+3 (w ) 4 2 p1+2+3 (w w )

b1 3
b2 3
+8
+8
2
2
(1 )
dw dw f (w, w ; )(w w ),
ib
ib
ib
(2.81)
with certain integration contours for w and w . Other three functions should be obtained
from the symmetry of Eq. (2.76): by flipping the signs of p1,2,3 and make a suitable change
of coordinates we find a set of contour integrals

b1 3
b2 3
+8
+8
2
2
F0 () =
(1 )
dw dw f (w, w ; )(w w ),

b1 3
b2 1
F1 () = 2 + 8 (1 ) 2 8 dw dw f (w, w ; )(ww w w + ),

b1 1
b2 3
F2 () = 2 8 (1 ) 2 + 8 dw dw f (w, w ; )(ww ),

b1 1
b2 1
8
8
2
2
F3 () =
(1 )
dw dw f (w, w ; )(ww w w + ),
(2.82)
satisfying the following differential equations
2ib1DF 0 () = p1 (1 )1/2F1 () + p2 1/2 F2 () p3 1/2 (1 )1/2 F3 (),
2ib1DF 1 () = p1 (1 )1/2F0 () + p2 1/2 F3 () p3 1/2 (1 )1/2 F2 (),
2ib1DF 2 () = p1 (1 )1/2F3 () + p2 1/2 F0 () + p3 1/2 (1 )1/2 F1 (),
2ib1DF 3 () = p1 (1 )1/2F2 () + p2 1/2 F1 () + p3 1/2 (1 )1/2 F0 (),

1 2bQ 1 2bQ
+
D (1 ) +
(2.83)
.
8
8( 1)
Hence they can be used to express the solutions of (2.76). They are not yet functions
because the contours are not specified. However, a notable property is that, as far as the
form of the integrand is concerned, the four transform into one another under the change
of integration variables. Some typical ones are given below:
w 1 w, w 1 w :
(F0 , F1 , F2 , F3 )()
(F0 , F2 , F1 , F3 )(1 )(p1 ,p2 )(p2 ,p1 ) ,
w w

w w
(F0 , F1 , F2 , F3 )() (F2 , F3 , F0 , F1 )()(p1 ,p3 )(p1 ,p3 ) ,
1
(F0 , F1 , F2 , F3 )() (F2 , F3 , F0 , F1 )().
(2.84)
233
Now that we have found a way to express the solutions of the differential equation in
an integral form, we have to investigate the property of them under the monodromy and
find some formulae for the change of basis like those of hypergeometric functions. This
can be done straightforwardly because the functions Fi () in the above are all expressible
as sums of products of two hypergeometric functions.
By fixing the integration contours and making a suitable rescaling, we define the
function F0 in the following way:

8# 12 + ibp1 # 32
F0 (p1 , p2 , p3 ; )
1 ib
1 2
# 14 + ib
2 p1+23 # 4 + 2 p12+3 # 4
b1
2
+ 38
(1 )
b2 3
2 +8
1

3/4
dw dw w (1 w )(1 w )
34 + ib
2 p1+23
3
ib
(1 w) 4 + 2 p12+3
ib
(1 w) 4 2 p1+2+3 (w w).
(2.85)
Using the same normalization and contour to define all the functions Fi , we find F0 = F1
and F2 = F3 . Thus we introduce the following functions
4G0 (p1 , p2 , p3 ; ) F0 (p1 , p2 , p3 ; ) = F1 (p1 , p2 , p3 ; )

b2 3

1 + 2ibp12+3 b1 + 3
=
2 8 (1 ) 2 + 8 F 34 , 54 , 32 ;
1 + 2ibp1

1
ib
3
F 34 + ib
2 p1+2+3 , 4 + 2 p1+23 , 2 + ibp1 ;
(p2,3 p2,3 ),
2G1 (p1 , p2 , p3 ; ) F2 (p1 , p2 , p3 ; ) = F3 (p1 , p2 , p3 ; )

b2 1

1 + 2ibp12+3 b1 1
=
2 8 (1 ) 2 8 F 14 , 14 , 12 ;
1 + 2ibp1

1
ib
3
F 34 + ib
2 p1+2+3 , 4 + 2 p1+23 , 2 + ibp1 ;
+ (p2,3 p2,3 ).
(2.86)
We furthermore introduce the notations

G2 (p1 , p2 , p3 ; ) = G0 (p1 , p2 , p3 ; ),
G3 (p1 , p2 , p3 ; ) = G1 (p1 , p2 , p3 ; ),
(2.87)
so that the solution of the differential equations for f 0 1 2 3 of (2.67) be linear

combinations of the functions Gi (p1 , p2 , p3 ; ) or of Gi (p1 , p2 , p3 ; ). Note that the
hypergeometric functions in the above which do not depend on pi can also be written as
1/2

3 5 3
2
1/2
F 4 , 4 , 2 ; = (1 )
,
1+ 1

1 + 1 1/2
F 14 , 14 , 12 ; =
(2.88)
,
2
234
so that they are related to the four-point functions of spin operators in Ising model [32].
The above functions Gi are not single valued on the entire CP1 , and the basis of solutions
are chosen so as to diagonalize the monodromy around = 0. Our next task is to find
the transition coefficients giving a relation between different bases. Using the formula for
hypergeometric functions they are given by
G0 (p1 , p2 , p3 ; )
x+ x+ x++ x++
1 x+ x+ x++ x++
G1 (p1 , p2 , p3 ; )
=
x x x+ x+
G2 (p1 , p2 , p3 ; )
2
G3 (p1 , p2 , p3 ; )
x
x
x+ x+
G0 (p2 , p1 , p3 ; 1 )
G (p , p , p ; 1 )
1 2 1 3
(2.89)
,
G2 (p2 , p1 , p3 ; 1 )
G3 (p2 , p1 , p3 ; 1 )
x++ =
x+ =
x+ =
x =
3
3
3
3
1

1
2 + ibp1 # 2 + ibp2

,
1
ib
+ ib
2 p1+2+3 # 4 + 2 p1+23

# 12 + ibp1 # 12 ibp2
1 ib
,
+ ib
2 p123 # 4 + 2 p12+3

# 12 ibp1 # 12 + ibp2
1 ib
,
+ ib
2 p1+23 # 4 + 2 p1+2+3

# 12 ibp1 # 12 ibp2
1 ib
.
+ ib
2 p12+3 # 4 + 2 p123
(2.90)
Using the above formulae we shall then try to find monodromy invariant combinations of
the left and right sectors. First, there is a diagonal product of the following form:

0 + G1 G
1 ) + x+ x++ (G2 G
2 + G3 G
3 ) (p1 , p2 , p3 ; )
x x+ (G0 G

0 + G1 G
1 ) + x+ x++ (G2 G
2 + G3 G
3 ) (p2 , p1 , p3 ; 1 ),
= x x+ (G0 G
(2.91)
i (pi ; ) Gi (pi ; ).
where we denoted G
Up to an overall constant, it has the following
asymptotic behavior at 0:

2

1
ibp23 2 b1 + 3
4 + ||b(Q1 ) 4 1 + ibp1 1 + ib p1+2+3
||
2
4
2
1 + 2ibp1
3 ib
3 ib

14 + ib
2 p123 4 + 2 p1+23 4 + 2 p12+3

2

ibp2+3 2 1
b1 14
b(Q1 )+ 34
+ ||
+ ||
2 + ibp1 14 + ib
p1+2+3
2
1 2ibp1
1 ib
3 ib

4 + 2 p123 4 + 2 p1+23 34 + ib
(2.92)
2 p12+3
Let us compare the asymptotic behavior of the above solution with the previous OPE
is expanded into two discrete terms. Apart
analysis based on the assumption that b/2
235
from the coordinate dependences which are irrelevant, the four-point functions should obey
the following asymptotic behavior:

(z0 )++
(z1 )++
(z2 )++
(z3 )
b/2
1
2
3

3
1
ibp23 2
|z01 |b1 + 4
C1 1 b2 ; 2 , 3 + |z01 |b(Q1 ) 4
1 + 2ibp1

C (1 )C3 1 + b2 ; 2 , 3
,
2i
++

b/2 (z0 )++
(z1 )++
(z2 )++
(z3 )
1
2
3

ibp23 2
b1 14
b
b(Q1 )+ 43
|z01 |
C3 1 2 ; 2 , 3 + |z01 |
1 2ibp1

b

C (1 )C1 1 + 2 ; 2 , 3
.
2i
(2.93)
Comparing them with (2.92) we find that the crossing symmetric solution (2.91) of the
differential equation corresponds to the sum of four-point functions
0 + G1 G
1 ) + x+ x++ (G2 G
2 + G3 G
3 )
x x+ (G0 G
++ ++ ++
++ ++ ++
b/2 1 2 3 + b/2 1 2 3
++ ++ ++
++ ++ ++
+ b/2
1 2 3 + b/2 1 2 3 .
(2.94)
By flipping the momentum of one of the three generic spin fields we obtain another
diagonal type solution of the differential equation, and these two are the only diagonal
solutions we could find. One might think that a more careful analysis would lead to another
solution and thereby enable us to write down the expression for each of the summand in
the above. However, there should not be any more solutions once we admit that all the
solutions are expressed in terms of Gi (pi ; ). It is also difficult to argue that the solutions
be duplicated due to the ambiguity in choosing the square-root branches. Thus we conclude
that there are only two independent solutions of diagonal type.
The above result seems to indicate that in correlation functions involving spin fields,
all one can fix by hand is the total fermion number of the product of spin fields and not
the Grassmann parity of each spin field. Nevertheless, our previous analysis of two- and
three-point structure constants still remains valid since they involve no more than two spin
fields.
Finally, there are also solutions of the off-diagonal type:

1 G1 G
0 ) x+ x++ (G2 G
3 G3 G
2 ) (p1 , p2 , p3 ; )
x x+ (G0 G

1 G1 G
0 ) x+ x++ (G2 G
3 G3 G
2 ) (p2 , p1 , p3 ; 1 )
= x x+ (G0 G
+ + ++
++ + + ++
1 2 3 + b/2 1 2 3
b/2
++ + + ++
++ + +
+ b/2 1 2 3 + b/2 1 2 3 .
(2.95)
236
3. Super-Liouville theory with boundary

If the worldsheet has a boundary, the action has boundary terms:

1
QR
2
2
+ + + 2ib
I=
d z +
d 2 z eb
2
4

QK
+ aB beb/2 ,
+ dx
4
(3.1)
where K is the curvature of the boundary which is defined by the Euler number formula

1
1
gR +
g 1/4 K = = 2 2g h
(3.2)
4
2
for worldsheets with g handles and h holes. The coupling B will be referred to as
the boundary cosmological constant as in the bosonic case. It can take different values for
different connected components of the boundary, and it may also jump at points where
the boundary primary operators are inserted. The boundary interaction term contains a
Grassmann odd constant a satisfying a 2 = 1, in order to avoid the Lagrangian becoming
Grassmann odd [19,33,34].
On worldsheets with boundary we can insert operators on the boundary. One of the most
fundamental boundary operator is
B (x) = e/2 (x) = eL (x),
(3.3)
and of course there are some other operators like

B = eL ,
= eL .
(3.4)
In free-field scheme the correlators of bulk and boundary fields on the upper half-plane can
be calculated as usual using mirror image techniques. One can also perform the integration
over the zero-mode of and derive that any correlator diverges when

Q(2 2g h) 2
(3.5)
i
j = b(2n + nB )
i (bulk)
j (boundary)
for non-negative integers n and nB , and the residue is given by a sum of free field
correlators with n bulk and nB boundary screening operators.
3.1. Classification of boundary states
The classification of boundary states can be done by studying the modular property of
annulus partition functions. We follow the same path as the reference [2] which analyzed
the bosonic Liouville theory. We first introduce the Ishibashi states, and then classify all the
possible Cardy states by expressing them as superpositions of Ishibashi states. Note that in
super-Liouville theory there is a freedom in choosing the spin structure on the worldsheet,
so that we have to consider the characters with ()F inserted as well as the ordinary ones.
237
Boundary states in generic superconformal field theory are defined as the solutions of
the boundary condition on currents on the real axis:
T (z) = T(z),
TF (z) = TF (z).
(3.6)
Mapping the upper half-plane onto a unit disc the above condition can be rewritten into the
following form
n ,
Ln = L
r .
Gr = i G
(3.7)
The boundary states |B; and B; | are therefore solutions of the equations
n ) = B; |(Gr + i G
r ) = 0,
B; |(Ln L

(Ln Ln )|B; = (Gr i Gr )|B; = 0.
(3.8)
The NS (R) boundary states satisfy the above condition with r Z + (r Z).
Given a highest weight state |h; NS of the superconformal algebra, one can construct
the Ishibashi state |h; +, I in the following way:
1
2
|h; +, I = |h; NSL |h; NSR + (descendants).
(3.9)
In the same way one can construct the Ishibashi state |h; , I from a highest weight state
|h; R in the R sector:

|h; , I = h; R+ L h; R+ R i h; R L h; R R + (descendants),

+
(3.10)
h; R+ + i h; R h; R + (descendants),
I h; , | = h; R
R
where we assumed h; R+ |h; R+ = h; R |h; R to be non-zero. Note that we will only
consider the Grassmann-even combinations in what follows. From these definitions it
follows that the annulus partition function bounded by two Ishibashi states is given by
the character:

c
0 c )
ic (L0 +L
12 |k; +, I = h,k Trh(NS) e 2ic (L0 24 ) ( )F ,
I h; +, |e
2ic (L c )

0 c )
ic (L0 +L
0 24 ( )F .
12 |k; , I = h,k Tr
(3.11)
I h; , |e
h(R) e
Here the expression is symbolic in the sense that the delta symbol h,k merely represents
the Ishibashi states diagonalize the annulus partition function as seen from the closed string
channel. R Ishibashi states and NS Ishibashi states are orthogonal to each other.
Cardy states are defined by the property that the multiplicity of open string modes
between two of them is given by the fusion coefficient Nh,kl . In superconformal field
theories they are given by the sum of NS and R pieces:
C h; | = C h; +, | + C h; , |,
|h; C = |h; +, C + |h; , C . (3.12)
NS (R) piece of any Cardy state is itself a solution of boundary condition on currents, and
should be expressed as a superposition of NS (R) Ishibashi states. The partition function
on an annulus bounded by two of them as seen from the open string channel is expressed
as the sum of characters with coefficients Nh,kl

c
Nh,kl Trl(NS) e2io (L0 24 ) (1)F
( = ),

Z(h,, ),(k,, ) (o ) =
(3.13)
c
2io (L0 24
)
l
F
( = ).
(1)
Nh,k Trl(R) e
238
Let us then consider the boundary states in super-Liouville theory. First of all, as the
labels h of representations of the superconformal algebra, we use the Liouville momentum
p for non-degenerate representations or the index (r, s) for degenerate ones. The characters
for non-degenerate representations are given by (q e2i )
+
Trp(NS) q L0 24 = p(NS)
( ) = q
c
p2
1
2 16

1 qn
n=1
Trp(NS) ()F q
c
L0 24
= p(NS)
( ) = q

p2
1
2 16
1
1
1 + q n 2 ,
1 qn
1
1
1 q n 2 ,
n=1
Trp(R) q
c
L0 24
+
= p(R)
( ) = 2q
p2
2

1

1 qn
1 + qn .
(3.14)
n=1
c
Note that TrR ()F q L0 24 vanishes for any representations except for the R vacuum. They
obey the following modular transformation property
+
u(NS)
( ) =
+
dp e2ipup(NS)
(1/ ),
21/2 u(NS)
( ) =
+
21/2 u(R)
( ) =
+
dp e2ipu p(R)
(1/ ),
dp e2ipup(NS)
(1/ ).
(3.15)
For degenerate representations, the characters are given by those of corresponding Verma
modules subtracted by those of null submodules:
+
r,s(NS)
= +
i
1 )(NS)
r,s(NS)
=
i
1 )(NS)
2 (rb+sb
2 (rb+sb
+
=
r,s(R)
i
2 (rb+sb
1 )(R)
+
i
2 (rbsb
1 )(NS)
()rs
i
2 (rbsb
2 (rbsb
1 )(R)
1 )(NS)
,
(3.16)
We would like to find the expression for the wave functions (p; h ) which express
the NS and R Cardy states as superpositions of Ishibashi states belonging to normalizable
representations:

C h; , | = 2
dp
(p; h ) I p; , |,
2

|h; , C = 2
0
dp
|p; , I (p; h ).
2
(3.17)
239
Here we have taken care for the equivalence of representations with momentum p and p.
The Ishibashi states are normalized to satisfy
0 c )
ic (L0 +L
12
|p , +, I = 2(p p )p(NS) (c ),
0 c )

ic (L0 +L
12 |p , , I =
2 (p p )p(R) (c ).
I p, , |e
I p, +, |e
(3.18)
We also assume that (p; h ) = (p; h ).

There is one important notice regarding the equivalence of R Ishibashi states under the
reflection p p. Recall that the highest weight states in the R sector can be created by
multiplying a spin operator onto the vacuum. Therefore, if we write the R Ishibashi states
as in (3.10), we obtain
++

I p; , | = Q/2+ip Q/2+ip + (descendants),

++
|p; , I = Q/2ip Q/2ip + (descendants).
(3.19)
As a consequence, if = +1 there arises a minus sign in flipping the sign of p. Hence
the wave functions for R Cardy states should depend also on how the supercurrents in the
left and the right sectors are glued: they must be odd functions of p for = 1 and even
functions for = 1. On the other hand, there is no such subtleties for NS Ishibashi states
so that one may well expect that the NS wave functions do not depend on .
The open/closed duality for annulus partition functions together with the obvious fusion
k
relation N(1,1),h
= hk yields

+ (p; u )+ p; (1, 1) = cos(2pu),

+ p; (r, s) + p; (1, 1) = 2 sinh(prb) sinh(ps/b),

(p; u ) p; (1, 1) = cos(2pu),

irs
sinh ps
p; (r, s) p; (1, 1) = 2 sinh prb + irs
(3.20)
2
b 2 .
Another equation comes from the fact that the wave functions (p; h ) are proportional
to the disc one-point functions of V Q +ip or Q +ip . Therefore they must be consistent with
2
2
the reflection relations:

+ (p; h ) = D Q
2 + ip + (p; h ),

Q + ip (p; h ).
(p; h ) = D
(3.21)
2
These equations determine the form of almost all the wave functions:

ip/b

1
1
3
+ p; (1, 1) = 2 2 2 bQ
ip#(ipb)# ip
,
2
b

bQ ip/b
ip
1
1
ip#(ipb)# b
+ p; (r, s) = 2 2 2 2
sinh(prb) sinh(ps/b),
ip/b

1
1
+ (p; u ) = 2 2 2 bQ
ip#(ipb)# ip
2
b cos(2pu),

bQ ip/b 1
1 ip 1
1
3
# 2 ipb # 2 b
,
p; (1, 1) = 2 2 2 2
240

ip/b 1

1
1
p; (r, s) = 2 2 2 bQ
# 2 + ipb # 12 + ip
2
b

irs
sinh ps
sinh prb + irs
2
b 2 ,

ip/b 1

1
1
(p; u ) = 2 2 2 bQ
# 2 + ipb # 12 + ip
2
b cos(2pu).
(3.22)
All we are left with is the R wave functions (p; h+ ). However, due to the requirement
that they must be a product of Gamma functions multiplied by an odd function of p, one
can actually find no analytic expressions for them. Consequently, one should conclude that
the R wave functions cannot be found by analyzing the modular property. They will be
proposed later by an analysis of one-point function on a disc and, moreover, we will find
that the degenerate Cardy states (r, s) must satisfy r + s = even (odd) for = 1 (+1).
If we accept this, the absence of (1, 1)+ state explains why we could not find the wave
functions the Cardy states with = 1 from the modular property.
It might seem strange that one can find the R Cardy states with = 1 only,
because naively one tends to think that the two choices for the boundary conditions on
supercurrent should be equivalent. However, it turned out that the two choices are actually
inequivalent and affects the parity of the wave functions under the sign-change of the
momentum p.
3.2. One-point functions of bulk operators
Let us try to reproduce these wave functions from a different approach, by calculating
the one-point functions on a disc. We define various one-point structure constants by the
equations

V (z) u = U+ (; u )|z z |2h ,
(z) u
= U (, ; u )|z z |2h 8 .
(3.23)
The one-point functions of spin fields , always vanish because we restricted the R
boundary states to be Grassmann-even in (3.10). The periodicity of supercurrents when
we go around the boundary of the disc is unambiguously determined by how many spin
fields are inserted on the disc. All the other one-point functions are zero or obtained by
superconformal transformations from the above ones. For example, the one-point function
of descendants in the NS sector is given by

V (z) u = i (Q ) 1 U+ (; u )|z z |2h 1 ,
(3.24)
and the one-point functions of spin fields depend on the index in the following way:

(z) u = , (z) u ,
(3.25)
in consistency with the previous analysis of modular property.

To obtain the one-point structure constants, we derive a set of recursion relations for
them from the solution of differential equation for two-point functions with one degenerate
241
operator. Let us first consider

1
1

V (z)b/2
(w) u = |w z |4hb/2 4 |z z |2hb/2 2h 8

b

b2 3
C+ ()U b2 , ; u 2 (1 ) 4 + 8

2
1b2
F 1b
2 + ibp , 2 , 1 + ibp ;

b(Q)
b2 3
+ C ()U + b2 , ; u 2 (1 ) 4 + 8

2
1b2
F 1b
(3.26)
2 ibp , 2 , 1 ibp ; ,
and C () are the OPE coefficients defined in (2.25). The coefficients

where = |zw|
|zw|
2
are chosen so that the behavior when the two bulk operators approach each other agrees

approaches the boundary, it should
with the OPE analysis. On the other hand, when b/2
be expanded as a discrete sum of boundary degenerate operators:
2

(w) |w w|
2hb/2 +hb + 8 r+ Bb (w) + |w w|
2hb/2 8 r B0 (w)
b/2
3
(3.27)
with certain coefficients r . Comparing this with the behavior of the solution around 1
we obtain a recursion relation:

2

# b + 1b
#(b 2 )
2
b
C+ ()U 2 , ; u
1b2
#(b b2)# 2

2
2

# 3+b
2 b #(b )
+ C ()U + b2 , ; u
1b2 = r (, )U+ (; u ).
#(1 b)# 2
(3.28)
The coefficient r can be calculated using free fields

2

3b4 38
(w)Bb (x)BQ (y)

dx b/2
r = abB |w w|

2 2
,
= 2 r bB # b 2 # 1b
(3.29)
2
where r is related to the free field correlator on a disc:

(x) (w) = 21/2a r (, )|x w|1 |w w|
3/8.
(3.30)
For this correlator to be non-vanishing, we have to identify with on the real axis
up to some constants.
Another recursion relation can be obtained from the analysis of the correlation functions
of two spin fields on a disc:

,
(z) (w) u
2i b/2
4hb/2 14
= |z w|
|w w|
2h +2hb/2

+ ()U+ b ; u G0 (p , p b , p ; )
2i C
2
2

b

+ C ()U+ + 2 ; u G3 (p , p b , p ; ) ,
2
(3.31)
242
2

and the coefficients are determined from the consistency with the OPE
where = zw
zw
of two spin fields, as before. Note that there seems to be another possibility of writing
down the solution using Gi (p , pb/2 , p ) instead of Gi (p , pb/2 , p ). However, it
will not lead to a recursion relation consistent with the analysis of modular property. As
was discussed in the previous section, it seems that we must use appropriate solutions of
differential equation according to the total fermion number.
Around 1 the above solution can be expressed as a certain linear combination of
Gi (pb/2 , p , p ; 1 ). Note that all the four of them show up, as opposed to what

one naively expects as a limit of b/2
approaching the boundary. According to our
observation in the previous section, this is due to the fact that we can only fix the total
is actually mixed with , ,
fermion number, so that the correlator b/2
b/2
with equal weights.
The terms proportional to G3 (pb/2 , p , p ; 1 ) can be identified with the

contribution from the case where b/2
approaches the boundary and turns into the
boundary identity operator. Thus we obtain a recursion relation
2
2 ir (, )U (, ; u )

2

# b b2 #(b 2 )
b

= 2i C+ ()U+ 2 ; u

2
2 # b b2 12 # 1b
2

2

# 1 + b2 b #(b 2 )
b

+ C ()U+ + 2 ; u
(3.32)
.

2
2 # 12 b # 1b
2
Here a factor 2 was inserted, because the solution of the differential equation is actually
a mixture of correlators as mentioned above and it is not known how they are mixed in
generic solutions.
Let us solve the system of two recursion relations. We first put the following ansatz:
2Q

1
1
Q 1
Q
2b
Q
U+ = 2 2 2 bQ
2
2 # b 2 # b 2 U+ ,
2Q 1

1
1
Q 1
Q
1
2b #
U = 2 2 2 bQ
2
2 + b 2 # 2 + b 2 U ,
(3.33)
to simplify the recursion relation into the following form

b , ; u + U
+ b , ; u
U
2
2

+ (; u ) 1 cos b2 1/2 ,
= 21 r (, )B U
2
2

+ + b ; u
+ b ; u + U
U
2
2

(, ; u ) 1 cos b2 1/2 .
= 21 r (, )B U
2
2
(3.34)
The solution for = 1 is given by

+ (; u ) = cosh (2 Q)u ,
U

(, ; u ) = cosh (2 Q)u ,
U
2 1/2
cosh(ub)
B = cos(b
2 /2)
(3.35)
243
together with r (, ) = 1. The solution for = +1 becomes

+ (; u+ ) = cosh (2 Q)u ,
U

(, ; u+ ) = sinh (2 Q)u ,
U
2 1/2
sinh(ub)
B = cos(b
2 /2)
(3.36)
with the condition that r (, +) = . The above conditions on r are met when we assume
= on the boundary (real axis). The results for = 1 agree with those obtained
in [19].
Our result gives the relation between the label u of Cardy states and the boundary
cosmological constant B . Remarkably, the relation is different according to the choice of
boundary condition on supercurrent. The one-point structure constants for spin fields also
differ according to . Although the corresponding wave function for R Cardy states with
= 1 could not be obtained from the modular property, it should be possible to account
for this quantity using the modular property.
Due to the subtlety in the correspondence between the solutions of differential equation
and the correlators, we are left with an undetermined constant . In later subsection we
will see that should be unity. However, we simply set = 1 for the time being until we
check it using the (3, 1) degenerate boundary operator.
One-point functions for degenerate boundary states
In the same way we can analyze the one-point structure constants for boundary states
belonging to degenerate representations. The main difference as compared to the previous
analysis is that they have no interpretation in terms of boundary interaction term. For
bosonic Liouville theory, it was found that the geometry of the open worldsheet becomes
a pseudosphere [2], so that the boundary is infinitely far from generic points in the bulk.
In this case, disc two-point functions are expected to factorize to products of one-point
functions in the limit where the two operators approach the boundary.
The recursion relations for degenerate boundary states are obtained by a simple
modification of (3.28) and (3.32):

2

# b + 1b
#(b 2 )
b
2
U+ () = C+ ()U 2 ,
1b2
#(b b2 )# 2

2
2

# 3+b
b
2 b #(b )
+ C ()U + 2 ,
1b2 ,
#(1 b)# 2

2
b

# b b2 #(b 2 )

b

U 2 , U (, ) = C+ ()U+ 2

2
# b b2 12 # 1b
2

2

# 1 + b2 b #(b 2 )
b
()U+ +
+C
.

2
2
2i# 12 b # 1b
2
U b2 ,
(3.37)
244
Assuming
1

Q
# b Q
# b Q
2
2
2
+ ()
U+ () = U
,
Q bQ Q
2
2 # 2 # 2b
1 1

Q

bQ /b # 12 + b Q
2 # 2 + b 2

U (, +) = U () 2
Q bQ Q
2 # 2 # 2b

bQ /b
together with (3.25), they are simplified to the following form:

b U
+ () = U
+ b ,
b U
U
2
2
2

b U
() = U
+ + b .
+ b U
U
2
2
2
(3.38)
(3.39)
Solving them together with those obtained by b 1b , we find the following solutions:

Q
sin s
Q
2
b
2
+ () =
,
U
sQ
sin brQ
2 sin 2b

1

Q
sin s 12
r+s sin r 2 + b 2

U () = i
sQ
sin brQ
2 sin 2b

sin br
1
b
Q
2
(3.40)
where r, s are integers whose sum must be even for = 1 and odd for = +1. There
as is obvious from the structure of the
exists an ambiguity of sign in front of U
recursion relation. The minus sign was chosen from the consistency with the analysis of
modular property. The results for = 1 again agree with [19]. By making a comparison
with the wave functions obtained in the previous subsection, we find that the solutions of
(3.37) can all be expressed in terms of them:

+ i Q
2 ; (r, s)
U+ ; (r, s) =
,

+ iQ
2 ; (r, s)

i Q
2 ; (r, s)
,
U , +; (r, s) =
(3.41)

+ iQ
2 ; (r, s)
if we define the R wave functions for degenerate representations as follows:

ip/b 1

1
1
p; (r, s) = i r+s 2 2 2 bQ
# 2 + ipb # 12 +
2

sin r 12 + ipb sin s 12 + ip
b .
ip
b
(3.42)
The appearance of + ( iQ
2 ; (r, s) ) expresses that the one-point functions are normalized
by the zero-point function with the same boundary condition.
Our result for one-point structure constants for degenerate representations shows that we
should associate the representations of NS (R) superalgebras to the boundary states with
= 1 (+1). This property can also be observed by a detailed analysis of the boundary
two-point functions.
245
3.3. Two-point functions of boundary operators

We would like then to find the expression for two-point functions of boundary operators
on a disc. Equivalently, we shall find the boundary reflection coefficients d(|u, u ) and
d(|u,
u ) defined by
(B )u ,u = d(|u , u )(BQ )u ,u ,

|u , u )
u ,u = a d(,
Q u
,u
(3.43)
Here the boundary states on both sides of the boundary operators are expressed by two
pairs of (u, ). To obtain these coefficients, all we have to do is to study the OPE of

. Note that, as is the
generic boundary operator with boundary degenerate operator b/2
case with bosonic Liouville theory with boundary, the consistency with fusion algebra
requires that the two boundary states appearing in the two sides of boundary degenerate
operators should be related to each other. It is expected that the two boundary states (u, )

are related via
and (u , ) connected by b/2
ib
,
= .
(3.44)
2
The condition is actually more stringent as we will see below by a detailed analysis.
Let us consider the OPE formula
u u =

(x)u ,u B (y)u ,u
b/2
b

|x y| 2 c+ b/2
(x) + |x y|
b(Q)
2

ac +b/2
(x)
!
u ,u
(3.45)
and calculate the coefficients c (, |u, u , u ; ) using free fields. One finds c+ = 1 as
usual, and that c is given in terms of free field correlators with one boundary screening
operator inserted. The latter consists of three terms corresponding to three boundary
segments, and we have to take the sum of these three carefully. The result reads
c (, |u, u , u , )

#(1 b)# b bQ
2
r (, )B
=

2 r (, )
# 1 bQ
2
bQ

bQ bQ
#(1 b)# 2
# b 2 # 2
+ r (, )B
+ r (, )B
#(b)
# 1 + bQ
2 b

br (, ) bQ
# 2 #(1 b)# b bQ
=
2
2 r (, )

2
2
iB cos b2 + iB cos b b2 + B sin b ,
(3.46)
b
where the coefficients such as r represent the prefactors arising in front of free field
correlators:

(x) (y) ,, = r (, )|x y|1/8 ,
246

1
(x) (y)(z) ,, = r (, )|x y|3/8|y z|1/2 |x z|1/2 ,
2

1
(x)(y) (z) ,, = r (, )|x z|3/8|z y|1/2|y z|1/2,
2

1
(x) (y) (z) ,, = r (, )|y z|3/8|x y|1/2|x z|1/2 . (3.47)
2
There are some relations among them due to the requirement of the consistency with cyclic
permutations and the analyticity in the upper half-plane with respect to the coordinate of .
Fixing them in the following way:
r (, 1)
= 1,
r (, 1)
r (, 1)
= i,
r (, 1)
(3.48)
and calculating further we obtain

1/2

2ib
bQ
#(1 b)# b bQ
2
2

b
sin b
2 (iu + iu + ) sin 2 (iu iu + ),
1/2

2ib
bQ
#(1 b)# b bQ
= +1 c (, |u, u , ) =
2
2

b
cos b
2 (iu + iu + ) sin 2 (iu iu + )
= 1 c (, |u, u , ) =
(3.49)
assumption u
= u ib
2 .
u
= u + ib
2
under the
On the other hand, when
the coefficient
c does not reduce to such a simple form. This reflects the fact that the boundary states
(u, ) and (u, ) are not strictly equivalent when = 1. Hence it is expected that
+

b/2 (u ib ) ,u ,
(3.50)
b/2 (u+ ib ) ,u
2
are the only boundary (2, 1) degenerate operators that are indeed degenerate. Combining
the reflection equivalence with the OPE formula as in the spherical case we obtain the
following recursion relations

d(|u , u )c (Q , |u, u , ) = d b2 , u , u ib
2 ,

d + b2 , u , u ib
(3.51)
2 c (, |u, u , ) = d(|u , u ).
In the same way, let us consider the OPE

(x)u ,u B (y)u ,u
b/2
b

|x y| 2 c+
b/2
(x) + |x y|
b(Q)
2

ac
+b/2
(x)
and calculate the coefficients using free fields. One finds

r (, )
i
1

= ( = 1); ( = 1) ,
c+ =
2 r (, )
2
2
!
u ,u
(3.52)

Q
b
# b bQ
#(1 b)# bQ
2
2
2r (, )

2
r (, )B cos b2 + r (, )B cos b

r (, )B sin b

b(Q )
#(1 b)# bQ
=
# b bQ
2
2
2

2
2
B cos b2 + B cos b b2 iB sin b .
247

c
=
b2
2
(3.53)
Here we used the free field correlators

r (, )
(x1 ) (x2 )(y1 )(y2 ) ,, =
2
(x1 y1 )(x2 y2 ) + (x1 y2 )(x2 y1 )
,
|x1 y1 |1/2|x1 y2 |1/2|x2 y1 |1/2 |x2 y2 |1/2 |x1 x2 |1/8 |y1 y2 |

r (, )
(x1 )(y1 ) (x2 )(y2 ) ,, =
2
(x1 y1 )(x2 y2 ) (x1 y2 )(y1 x2 )
. (3.54)
|x1 y1 |1/2|x1 y2 |1/2|x2 y1 |1/2 |x2 y2 |1/2 |x1 x2 |1/8 |y1 y2 |

Assuming again u = u
ib
2 ,
the coefficients can be rewritten further:

1

(, |u, u , ) = i2 2 ,
= 1 c+
1

c
(, |u, u , ) =
1/2

22 b Q
bQ
#(1 b)# b
2
bQ
2

b
cos b
2 (iu + iu + ) cos 2 (iu iu + ),
1

= +1 c+
(, |u, u , ) = 2 2 ,
1

c
(, |u, u , ) =
1/2

2 2 ib Q
bQ
#(1 b)# b
2
bQ
2

b
sin b
2 (iu + iu + ) cos 2 (iu iu + ).
(3.55)
If we require that the reflection relation holds also for descendants B , possibly with the
coefficient different from that of the primaries
(B )u ,u = d (|u , u )(Q )(BQ )u ,u ,
we obtain another set of recursion relations
Q

d (|u , u )c
(Q , |u, u , )

= c (, |u, u , )d b , u , u
+
Q

d (|u , u )c+
(Q , |u, u , )

= c (, |u, u , )d + b , u , u
ib
2
ib
2

.
(3.56)

,
(3.57)
248
It is straightforward to write down the solutions of the recursion relations (3.51) and
(3.57) in terms of the functions G and S introduced in [1]. They are defined in the following
way:

log G(x) =
0

log S(x) =
2

et Q
dt
1 Q
eQt /2 ext
+
x
, (3.58)
+
t (1 ebt )(1 et /b )
2
2
t 2

sinh[(x Q/2)t]
dt 2x Q
.
t
t
2 sinh[bt/2] sinh[t/2b]
(3.59)
The function G(x) has zeroes at x = mb nb1 (m, n Z) and no poles. The functions
S and are expressed in terms of G:
S(x) = G(Q x)/G(x),
(x) = G(Q x)G(x).
(3.60)
The shift relations for G and S

1
G(x + b) = G(x)(2) 2 b 2 bx #(bx),

S(x + b) = S(x)2 sin(bx),

1 x 1
G x + b1 = G(x)(2) 2 b b 2 #(x/b),

S x + 1b = S(x)2 sin(x/b)
(3.61)
can be used to write down the solutions of the recursion relations. If we define the functions
GNS , GR and SNS , SR by

x+b1

,
GR (x) = G x+b
,
GNS (x) = G x2 G x+Q
2
2 G
2
x x+Q
x+b x+b1
SNS (x) = S 2 S 2 ,
SR (x) = S 2 S 2 ,
(3.62)
the solution becomes
d(|u , u )

1b2 Q2
1
2b G
bQ
NS (Q 2)GNS (2 Q)
2 b
,
=
SNS ( + iu + iu )SNS ( iu + iu )SNS ( + iu iu )SNS ( iu iu )
d (|u , u )

1b2 Q2
1
2b G
bQ
NS (Q 2)GNS (2 Q)
2 b
,
=
SR ( + iu + iu )SR ( iu + iu )SR ( + iu iu )SR ( iu iu )
|u , u+ )
d(,

1b2 Q2
2b G (Q 2)G (2 Q)1
i bQ
R
R
2 b
,
=
SNS ( + iu + iu )SNS ( iu + iu )SR ( + iu iu )SR ( iu iu )
d(|u+ , u+ )

1b2 Q2
1
2b G
bQ
NS (Q 2)GNS (2 Q)
2 b
,
=
SR ( + iu + iu )SNS ( iu + iu )SNS ( + iu iu )SR ( iu iu )
249
d (|u+ , u+ )

1b2 Q2
1
2b G
bQ
NS (Q 2)GNS (2 Q)
2 b
,
=
SNS ( + iu + iu )SR ( iu + iu )SR ( + iu iu )SNS ( iu iu )
|u+ , u )
d(,

1b2 Q2
2b G (Q 2)G (2 Q)1
i bQ
R
R
2 b
.
=
SNS ( + iu iu)SNS ( iu iu)SR ( + iu + iu)SR ( iu + iu)
(3.63)
They satisfy the unitarity condition and are consistent with the equivalence of boundary
states u and u when = 1. Note also that the structure of poles of these quantities
implies that we should identify = 1 (+1) boundary states with the representations for
NS (R) superalgebras.
From the above result one can easily read off that d and d differ, and d does depend
on . This means that the reflection coefficients of boundary operators differ for each
operator in a single supermultiplet at least if the transformation law is defined in a
naive way. As a consequence, it follows that the supersymmetry transformation and the
reflection do not commute. However, from the representation theory involving degenerate
representations it is natural since, if we assume (3.50), it follows that different operators in
a single (degenerate) supermultiplet connect two boundary states in a different way. As a
simple example, let us consider how the two boundary states on the two sides of Bb or
Bb are related to each other. If these operators are regarded as created by multiplying

, the only possibilities are
two degenerate operators b/2
[Bb ](uib) ,u ,
[Bb ]u ,u .
(3.64)
This is reasonable if we make comparison with the OPE relations

[Vb ] [V ] [Vb ] + [ V ] + [V+b ],
+b ].
[ Vb ] [V ] [ Vb ] + [V ] + [ V
(3.65)
The generalization of this argument to higher degenerate representation is straightforward.

Let us denote boundary degenerate operators as Bkbhb1 , Bkbhb1 or kbhb
1
using two non-negative half-integers k, h. Then the two boundary states with labels u and
u are related via
u = u + i(r k)b + i(s h)b 1
(0 r 2k, 0 s 2h),
(3.66)
where r +s must be even for Bkbhb1 , kbhb

1 and odd for Bkbhb1 , kbhb1 .
Of course this is merely a conjecture, and the consistency should be proven by a more detailed analysis of this model.
Density of open string states

The reflection coefficients can be regarded as phase shifts, so are related to the density
of certain open string states through the formulae of the following form:

1 d

log d Q
2 + is u , u = (s|u , u ),
2i ds
250

1 d
log d Q
+ is u , u = (s|u , u ),
2
2i ds

1 d
log d Q
(3.67)
+ is, u , u = (s,
|u , u ).
2
2i ds
Using the integral expression for S and discarding terms independent of the labels u and
u , some of them are given by
(s|u , u ) 2

dp
(s|u , u ) 2
e2ips cosh(Qp) cos(2up) cos(2u p)

sinh(2bp) sinh(2p/b)

dp 2ips
e
+ (p; u )+ (p; u ) + (p; u ) (p; u ) ,

dp
e2ips cosh (b b1 )p cos(2up) cos(2u p)

sinh(2bp) sinh(2p/b)

dp 2ips
e
+ (p; u )+ (p; u ) (p; u ) (p; u ) .
(3.68)
This agrees with the analysis of modular property since the annulus partition function
bounded by two Cardy states u and u is given by
=

Zu ,u =
1
=
2
dp
+
2+ (p; u )+ (p; u )p(NS)
(c )
2

+
(c )
+ 2 (p; u ) (p; u )p(R)

+
(o )
ds (s|u , u ) s(NS)
+ s(NS)
+

s(NS)
+ (s|u , u ) s(NS)
(o ) .
(3.69)
The other quantities can be written as

(s|u+ , u+ ) =
(s|u+ , u+ ) =
(s,
|u , u+ ) =

dp 2ips
e
+ (p; u+ )+ (p; u+ ) + (p; u+ ) (p; u+ ) ,

dp 2ips
e
+ (p; u+ )+ (p; u+ ) (p; u+ ) (p; u+ ) ,

dp 2ips
e
+ (p; u )+ (p; u+ ) + (p; u ) (p; u+ ) ,
251

dp 2ips
e
+ (p; u+ )+ (p; u ) + (p; u+ ) (p; u ) ,
(3.70)
if we assume that the R wave function for = 1 is given by
ip/b 1

1
1
(p; u+ ) = i2 2 2 Q
(3.71)
# 2 + ipb # 12 + ip
2
b sin(2pu).
(s,
|u+ , u ) =
Hence, in all the cases the phase shifts give the density of open string states with definite
worldsheet fermion number. This also suggests that the wave function for R Cardy states
with u+ is given by (3.71), in consistency with the analysis of one-point structure constants.
Bulk and boundary cosmological constants
As a further check, let us consider the OPE of boundary operators involving (3, 1)
degenerate representation. Above all, the OPE coefficients involving them depend on both
of and B . The consistency with the previous arguments fixes the constant which was
left undetermined.
Let us consider the OPE relation
[Bb ]u ,u [B ]u ,u
c+ [Bb ]u ,u + c0 [B ]u ,u + c [B+b ]u ,u
(3.72)
with u = u + ib, and use c to derive another recursion relation for boundary reflection
coefficient. Here again c+ = 1, so we concentrate on the calculation of c . There are
largely two contributions to c , which are proportional to and 2B , respectively.
Calculating first the contribution proportional to using free fields, one finds

2
2
2b2 d 2 z |z|2b |1 z|2b |z z |b 1 = 2b2 I0 sin b2 sin2 (b),
(3.73)
where I0 is given by

bQ
2
#(1
b)#
1
I0 =
2 sin b2
bQ
2

b #(b)# b
bQ
2 .
(3.74)
There are six contributions proportional to 2B , since there are six ways of inserting two
boundary screening operators onto the boundary divided into three segments. Restoring ,
they are summarized into the following form

2
2
2
b2 I0 2B sin b2 cos b2 B sin(b) cos b b2

2
2
+ B sin(b) cos b + b2

2
2
2
2
2
+ 2b2 I0 sin b2 B B cos b2 cos b b2 B B cos b2 cos b + b2

2
2
+ B B cos b b2 cos b + b2 ,
( = 1)

b
= 22 b2 I0 sin b2 sin2 (b) + 4 sin b
2 (iu + iu + ) sin 2 (iu iu + )

b
cos b
2 (iu + iu + + b) cos 2 (iu iu + + b) ,
252
( = 1)

b
= +22 b2 I0 sin b2 sin2 (b) 4 cos b
2 (iu + iu + ) sin 2 (iu iu + )

b
sin b
2 (iu + iu + + b) cos 2 (iu iu + + b) .
(3.75)
One can see that when = 1 there are some cancellation between the contributions from
the bulk and the boundary, and the result for c yields a recursion relation for d(|u , u )

.
consistent with the analysis using b/2
In this paper the N = 1 super-Liouville theory was analyzed on a sphere and on
a disc. The analysis was based on the approach developed in [1,2,29,31] for bosonic
Liouville theory. Various quantities were obtained in a form very similar to the bosonic
case. However, there are some new features in N = 1 theory largely due to the fermionic
nature of screening operators. As one of the consequences, the reflection of spin fields
is always accompanied by the flip of the chirality. We also presented all the solutions
of differential equations for four-point functions containing one degenerate spin operator

b/2
. For four-point functions of four spin fields, the differential equation becomes of the
fourth order and there are therefore four independent solutions in apparent contradiction

with any operators is decomposed into two
with the assumption that the product of b/2
discrete terms. However, our solutions obey a special transformation property under the
change of basis so that the crossing symmetric combination of the left and the right sectors
can be identified with a particular sum of four-point functions of spin fields.
Contrarily to the bosonic case, the analysis of modular property of annulus partition
functions was not sufficient to obtain all the wave functions that define Cardy states. It
was also found that the two ways of putting boundary condition on supercurrent leads to
two boundary states which differ in quite a non-trivial way. Indeed, we were unable to find
the R Cardy states with = 1 from the modular analysis. We found the wave functions
for them through the analysis of disc one-point functions and found the correspondence
between = 1 (+1) boundary states and the representations of NS (R) superalgebras.
The two-point functions of boundary operators were also obtained and the density of
open string states which can be read off from them were shown to be consistent with
the analysis of modular property. Remarkably, the reflection coefficients for boundary
operators depend on the label of boundary states in such a way that they are different
for different components in the same supermultiplet. To understand this we need a more
detailed analysis of the property of boundary operators.
Our analysis have shown that the non-compact superconformal theory with boundary
can be analyzed using the techniques developed in the analysis of bosonic theory, if an
appropriate care is taken. It would then be interesting to analyze similar superconformal
theories or those with higher worldsheet supersymmetry in the same way.
253
Note added
For disc one-point functions, after the submission of this paper we were informed of the
preceding analysis [19] which covers the = 1 case of our result.
Acknowledgements
The authors thank C. Ahn, T. Eguchi, K. Ito, I. Kishimoto, C. Rim, M. Stanishkov
and Al.B. Zamolodchikov for discussions and comments. The works of the authors were
supported in part by JSPS research fellowships for young scientists.
References
[1] V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, Boundary Liouville field theory I. Boundary state and
boundary two-point function, hep-th/0001012.
[2] A. Zamolodchikov, Al. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[3] K. Hosomichi, Bulk-boundary propagator in Liouville theory on a disc, JHEP 0111 (2001) 044, hepth/0108093.
[4] B. Ponsot, J. Teschner, Boundary Liouville field theory: boundary three point function, Nucl. Phys. B 622
(2002) 309, hep-th/0110244.
[5] A. Giveon, D. Kutasov, A. Schwimmer, Comments on D-branes in AdS3 , Nucl. Phys. B 615 (2001) 133,
hep-th/0106005.
[6] A. Parnachev, D.A. Sahakyan, Some remarks on D-branes in AdS3 , JHEP 0110 (2001) 022, hep-th/0109150.
[7] P. Lee, H. Ooguri, J.W. Park, Boundary states for AdS2 branes in AdS3 , hep-th/0112188.
[8] B. Ponsot, V. Schomerus, J. Teschner, Branes in the Euclidean AdS3 , hep-th/0112198.
[9] J.F. Arvis, Classical dynamics of the supersymmetric Liouville theory, Nucl. Phys. B 212 (1983) 151;
J.F. Arvis, Spectrum of the supersymmetric Liouville theory, Nucl. Phys. B 218 (1983) 309.
[10] E. DHoker, Classical and quantal supersymmetric Liouville theory, Phys. Rev. D 28 (1983) 1346.
[11] O. Babelon, Monodromy matrix and its Poisson brackets in supersymmetric Liouville string theory, Phys.
Lett. B 141 (1984) 353;
O. Babelon, Construction of the quantum supersymmetric Liouville theory for string models, Nucl. Phys.
B 258 (1985) 680.
[12] A.B. Zamolodchikov, R.G. Pogosian, Operator algebra in two-dimensional superconformal field theory,
Sov. J. Nucl. Phys. 47 (1988) 929.
[13] E. Abdalla, M.C. Abdalla, D. Dalmazi, K. Harada, Correlation functions in super-Liouville theory, Phys.
Rev. Lett. 68 (1992) 1641, hep-th/9108025.
[14] 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.
[15] K.I. Aoki, E. DHoker, On the Liouville approach to correlation functions for 2-D quantum gravity, Mod.
Phys. Lett. A 7 (1992) 235, hep-th/9109024;
K.I. Aoki, E. DHoker, Correlation functions of minimal models coupled to two-dimensional quantum
supergravity, Mod. Phys. Lett. A 7 (1992) 333, hep-th/9109025.
[16] D. Dalmazi, E. Abdalla, Correlators in noncritical superstrings including the spinor emission vertex, Phys.
Lett. B 312 (1993) 398, hep-th/9302032.
[17] R.H. Poghosian, Structure constants in the N = 1 super-Liouville field theory, Nucl. Phys. B 496 (1997)
451, hep-th/9607120.
[18] R.C. Rashkov, M. Stanishkov, Three-point correlation functions in N = 1 super-Liouville theory, Phys. Lett.
B 380 (1996) 49, hep-th/9602148.
254
[19] C. Ahn, C. Rim, M. Stanishkov, Exact one-point function of N = 1 super-Liouville theory with boundary,
hep-th/0202043.
[20] A.M. Polyakov, Quantum geometry of fermionic strings, Phys. Lett. B 103 (1981) 211.
[21] J. Distler, Z. Hlousek, H. Kawai, Super-Liouville theory as a two-dimensional, superconformal supergravity
theory, Int. J. Mod. Phys. A 5 (1990) 391.
[22] S. Chaudhuri, H. Kawai, S.H. Tye, Path integral formulation of closed strings, Phys. Rev. D 36 (1987) 1148.
[23] M.B. Green, N. Seiberg, Contact interactions in superstring theory, Nucl. Phys. B 299 (1988) 559.
[24] M. Dine, N. Seiberg, Microscopic knowledge from macroscopic physics in string theory, Nucl. Phys. B 301
(1988) 357.
[25] M. Goulian, M. Li, Correlation functions in Liouville theory, Phys. Rev. Lett. 66 (1991) 2051.
[26] M.A. Bershadsky, V.G. Knizhnik, M.G. Teitelman, Superconformal symmetry in two dimensions, Phys.
Lett. B 151 (1985) 31.
[27] D. Friedan, Z.A. Qiu, S.H. Shenker, Superconformal invariance in two-dimensions and the tricritical Ising
model, Phys. Lett. B 151 (1985) 37.
[28] S.K. Nam, The Kac formula for the N = 1 and the N = 2 superconformal algebras, Phys. Lett. B 172 (1986)
323.
[29] J. Teschner, On the Liouville three point function, Phys. Lett. B 363 (1995) 65, hep-th/9507109.
[30] H. Dorn, H.J. Otto, Two and three point functions in Liouville theory, Nucl. Phys. B 429 (1994) 375, hepth/9403141.
[31] A. Zamolodchikov, Al. Zamolodchikov, Structure constants and conformal bootstrap in Liouville field
theory, Nucl. Phys. B 477 (1996) 577, hep-th/9506136.
[32] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Infinite conformal symmetry in two-dimensional
quantum field theory, Nucl. Phys. B 241 (1984) 333.
[33] S. Ghoshal, A.B. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional integrable
quantum field theory, Int. J. Mod. Phys. A 9 (1994) 3841;
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353, hep-th/9306002, Erratum.
[34] R.I. Nepomechie, The boundary supersymmetric sine-Gordon model revisited, Phys. Lett. B 509 (2001)
183, hep-th/0103029.

Asymmetric ChernSimons number diffusion from

CP-violation
Bert-Jan Nauta, Alejandro Arrizabalaga
Institute for Theoretical Physics, Valckenierstr. 65, 1018 XE Amsterdam, The Netherlands
Received 19 February 2002; accepted 3 May 2002
Abstract
We study ChernSimons number diffusion in a SU(2)-Higgs model with CP-odd dimension-eight
operators. We find that the thermal average of the magnitude of the velocity of the ChernSimons
number depends on the direction of the velocity. This implies that the distribution function of the
ChernSimons number will develop an asymmetry. It is argued that this asymmetry manifests itself
through a linear growth of the expectation value of the third power of the ChernSimons number. This
linear behavior of the third power of a coordinate of a periodic direction is verified by a numerical
solution of a one-dimensional Langevin equation. Further, we make some general remarks on thermal
averages and on the possibility of the generation of the baryon asymmetry in a non-equilibrium
situation due to asymmetric diffusion of the ChernSimons number. 2002 Published by Elsevier
Science B.V.
1. Introduction
An important cosmological observation is the matterantimatter asymmetry in the
present universe. This asymmetry may be quantified by the baryon-to-photon ratio, whose
observational value is [1]
nB
= (1.554.45) 1010,
n
(1)
with nB the baryon number density and n the photon density. An interesting question is
how this asymmetry was generated in the early universe. In his 1967 paper Sakharov [2]
notes that baryon number can only be generated when
E-mail addresses: nauta@science.uva.nl (B.-J. Nauta), arrizaba@science.uva.nl (A. Arrizabalaga).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 5 - 3
256
B.-J. Nauta, A. Arrizabalaga / Nuclear Physics B 635 (2002) 255285
(1) baryon number is not conserved,

(2) the transformations C and CP are not symmetries,
(3) there is a departure from equilibrium.
In 1976 t Hooft [3] showed that in the electroweak theory the first requirement is
satisfied. The non-trivial vacuum structure of the SU(2)-gauge theory in combination
with the anomaly equation implies that a change in ChernSimons number, NCS , is
accompanied by a change in baryon number, B,
B(t) B(0) = 3[NCS (t) NCS (0)] =
3g 2
32 2
t
dt
a a
,
d 3 x F
F
(2)
0
a the field strength, and F
a its dual. At zero
with g the SU(2)-gauge-coupling, F
temperature changes in the ChernSimons number involve instanton processes. The rate
of these processes is very small [3]. However, at high temperature the rate is much larger.
As has long been recognized this opens up the possibility that the baryon asymmetry was
generated at temperatures of about 100 GeV [4], when the (elementary) particles that made
up the plasma may be described by the electroweak theory.
However, within the standard model the amount of CP-violation is probably too small to
account for the observed asymmetry (1) [5]. One way to deal with this is to include effective
non-renormalizable operators that parameterize the CP-violation of a more fundamental
a F
a , with
theory. The lowest-dimensional operator in a SU(2)-Higgs theory is F
the Higgs field. It may be included in the effective action as
0
3g 2 CP
a a
S [6] =
(3)
,
d 4 x F
F
32 2 M 2
0 a coefficient that ideally may be derived from the fundamental
with M a mass and CP
theory. In the current investigation dimension-eight operators will play a more important
role. They may be included in the action as

1

2 a a
3g 2
2
b
3
,
S [8] =
F
F b + CP
d 4 x CP
D (D ) CP
F F
2
4
32 M
(4)
with D the covariant derivative.

In this paper we study the combination of the first two of Sakharovs requirements,
although we were unable to refrain from making some remarks on the inclusion of the
third one in our study. That is, we study the motion of the ChernSimons number in the
SU(2)-Higgs action extended with (3) and (4). It is well known that in a pure SU(2)-Higgs
theory (without extra CP-odd operators and without fermions) the long-time behavior of
the ChernSimons number may be viewed as a diffusion process in one dimension, that
can be specified by the expectation values
NCS (t) NCS (0) = 0,

[NCS (t) NCS (0)]2 = 2V spht,
(5)
(6)
257
with V the volume and sph the sphaleron rate. The sphaleron rate plays an important
role in scenarios for electroweak baryogenesis, since it determines the rate of the baryonnumber violating processes. Much effort has gone into the determination of the sphaleron
rate, both in the symmetric phase [68] as well as in the broken phase [911]. Here, only
the broken-phase sphaleron-rate will be needed
sph = T 4 eEsph ,
(7)
with the sphaleron energy Esph , and a dimensionless coefficient that depends on the
temperature, the Higgs mass and expectation value, and the gauge coupling. The physical
picture that underlies (7) is that of classical transitions from one (classical) vacuum to the
next over a potential barrier of height Esph . On the basis of this picture, the gauge-Higgs
dynamics will be treated classically in this paper.
The idea that the inclusion of CP-violating operators may have an interesting effect
on the diffusion of the ChernSimons number stems from the fact that the ChernSimons
number itself is CP-odd. Therefore the motion over the barrier towards positive Chern
Simons numbers may differ from the motion in the direction of negative ChernSimons
numbers. It is clear then, that the distribution function of the ChernSimons number does
not need to develop symmetrically. Or put differently: there is no symmetry argument
why expectation values of odd powers of NCS (t) NCS (0) should vanish. In fact, we will
show that an asymmetry in the distribution function will develop indeed (as may have been
inferred from the title). That is, an initially symmetric distribution function will develop an
asymmetry. In the presence of the CP-odd operators in (4) an equation for the expectation
value of the third power of NCS (t) NCS (0) needs to be added to Eqs. (5) and (6) to
characterize the diffusion. We will argue that this expectation value increases linearly in
time in the broken phase:

[NCS (t) NCS (0)]3 t.
(8)
A more detailed conjecture is given in Eq. (48).

We end the introduction with a short outline of the rest of the paper. In the next section
the dynamics of the gauge-Higgs system will be projected on a one-dimensional path
through the configuration space. This will simplify the analysis of the rest of the paper.
In Section 3 we will calculate the thermal average of the velocity when the system moves
in the direction of positive or negative ChernSimons numbers. In Section 4 a conjecture
for the time dependence of [NCS (t) NCS (0)]3 is presented, where it is argued to grow
linearly in time. This linear growth is also present in a simple random walk model as is
shown in Appendix B. Section 5 contains some general remarks on statistical averages.
In the following section, we derive a simple stochastic equation that is useful to verify
some of the analytic results. Also some numerical solutions are presented. In Section 7
we discuss how asymmetric diffusion of the ChernSimons number may yield a non-zero
baryon number in a out-of-equilibrium situation. Also some comments are made on the
constraints on such a model. We end with a summary in Section 8.
258
2. Projection of the dynamics of the gauge and Higgs fields onto one special
direction in configuration space
To simplify the later study of the effect of the CP-violating operators in (3) and (4) on
the diffusion of the ChernSimons number, the gauge-Higgs-field dynamics is projected
on a single path through the configuration space. In order to include sphaleron transitions,
we use a path that runs from a classical vacuum to the sphaleron configuration and further
to the next vacuum. Manton [12] defined such a path in his proof that a non-contractible
loop exists in configuration space of the SU(2)-Higgs model. This is the path that will
be used in the following. The path of Manton is not the minimal-energy path, which
was constructed in [13]. But the precise path will not be important for the following
rather general arguments and we expect that the results will suffice as order of magnitude
estimates.
We parameterize the path of Manton by the time-dependent coordinate . The fields
are given in the radial gauge (i Aia = 0) by
2i
f ()[ U ()]U 1 (),
g

1
0
0
Higgs =
2 v h()U ()
i[1 h()] cos
,
1
1
2
gauge Aa a =
(9)
(10)
with v the Higgs expectation value and = gvr, where r = |

x |. The -dependent SU(2)matrix is given by

1
z
x + iy
i 0
U () =
(11)
sin +
cos .
z
0 i
r x + iy
The functions f and h satisfy the boundary conditions
f 0,
h 0,
r 0,
f 1,
h 1,
r .
(12)
The form of the fields (9), (10) is a non-static generalization of the fields considered
in [12]. When is chosen to be time-independent the fields given by (9), (10) can be
identified (up to a global U(1) gauge transformation) with the fields (3.17) in [12] (the
coordinate corresponds to in [12]).
For the functions f and h the Ansatz b of Klinkhamer and Manton [14] will be used

2
,
A,
A(A+4)
f () =
(13)
4
1
1 A+4 exp[ 2 (A )], A,

B+1
,
B,
h() = B( B+2)
(14)
B 1
1
1 B+2 exp[ 2 (B )], B,
1
with = (/2g 2 ) 2 , where is the Higgs self-coupling constant. The parameters A, B

were determined by minimizing the energy for the static field configuration at = /2
in [14], in order to obtain an estimate for the sphaleron energy. In this way the parameters
259
depend only on . We take = 1, for which the parameters have the values A = 1.15 and
B = 1.25 [14].
Now that the dynamics has been restricted to the path described by (9) and (10) we may
rewrite the SU(2)-Higgs action, S, in terms of the coordinate

2

4v
2
2
4
a3 sin + a4 sin .
dt a1 + a2 sin
S[, ] =
(15)
g
(gv)2
The CP-odd action (4) in terms of reads

2 3 2
1
1
4v 2
[8]
2
2
sin sin ,
+ b2 CP
+ b3 CP
+ b4 CP
S [, ] =
dt b1 CP
M4
(16)
where total time-derivatives have been neglected.
, gives only a total
The action S [6] (3), that includes the dimension-six operator F F
2
a
a

in (4). Therefore, the inclusion of
time-derivative, just as the operator ( ) F F
S [6] will not affect the motion along the path parameterized above. When the full fielddynamics is considered, the dimension-six operator may have an effect on the motion
along the ChernSimons number direction in a non-equilibrium situation, see, e.g., [15].
In equilibrium however, the system oscillates around one vacuum or, during a transition,
around the minimal-energy path that connects two classical vacua (the path that we have
approximated by (9) and (10)). Since, the oscillations average out in equilibrium, we do
not expect the inclusion of S [6] to affect the diffusion of the ChernSimons number.
The coefficients a1 , a2 , a3 , a4 , b1 , b2 , b3 b4 are given by the integrals

a1 =

1
d 2 ( f )2 + (h f )2 = 2.03,
2
(17)

1
2
2
2
2 2
d 8f (1 f ) + 1 h 1 f = 2.53,
2
(18)

a2 =
0

1
a3 = d 4( f )2 + 2 ( h)2 + h2 (1 f )2
2
0

2
2
2f h(1 f )(1 h) + f (1 h) = 1.41,

a4 =
(19)

d 8f 2 (1 f )2 / 2 + 2f h(1 f )(1 h)

1 2
2 2
f (1 h) + 1 h
= 0.70,
4

9
b1 =
d (h f )2 ( f )f (1 f ) = 0.09,
2
2
(20)
(21)
260

b2 = 9
d ( f )3 f (1 f ) = 0.10,
(22)
9
b3 =
2

d 1 h2 ( f )f (1 f ) 1 f 2 = 0.19,
(23)

b4 = 72
1
( f )f 3 (1 f )3 = 0.16.
2
(24)
Let us shortly discuss the actions (15) and (16). In the action (15) we recognize the
potential barrier [12],
V () =

4v
a3 sin2 + a4 sin4 ,
g
between different vacua. The sphaleron energy reads

4v
1
4v
= 2.11
.
Esph = V = = (a3 + a4 )
2
g
g
(25)
(26)
The sphaleron energy determines the exponential suppression of the sphaleron rate (7).
Another quantity that enters the sphaleron rate is (the real part of) the frequency of the
negative mode, , at the sphaleron configuration [9]. From the action (15) it may be
determined as
2
a3 + 2a4
(gv)2 0.61(gv)2 .
a1 + a2
(27)
In agreement with the order of magnitude estimate gv in [9].

The action (16) is odd in . This can be understood by realizing that a CPtransformation in terms of is
,
(28)
since the action (4) is CP-odd, so should (16).

Now that the SU(2)-Higgs action including the effective CP-odd operators has been
immensely simplified, namely to the one-dimensional action (15) + (16), the question:
does the ChernSimons number diffuse asymmetrically under the influence of the
dimension-eight operators in the action (4)? may be cast in the form: does the factor
3 in (16) lead to an asymmetric diffusion of ?. What is meant by diffusion of
may require some explanation. From Eqs. (9)(11), one sees that = 0 and = 2
correspond to the same (static) gauge and Higgs field configuration. Even = 0 and
= correspond to the same physical state, since they are related by a simple gauge
transformation. Hence, one may view configuration space as a circle with circumference .
This is the same for the full gauge-Higgs fields. When a transition over the sphaleron
barrier is considered, the system starts and ends in the same state (since the different vacua
are related by a large gauge transformation). Nevertheless the winding number is changed:
/NCS = 1. The (change in the) winding number is the physical quantity of interest, since it
261
Fig. 1. The distribution function of in a periodic potential. On the left it is viewed on the circle [0, [,
then the distribution function is static. On the right it is unwound to the infinite line ], [ accounting
thus for the winding number naturally. On the infinite line the system is dynamical and the distribution function
diffuses.
is related to the (change in the) baryon number through (2). The relation between a change
in the ChernSimons number and is (see Appendix A)
2
NCS (t) NCS (0) =
t
dt sin2 .
(29)
Instead of considering the system on a circle and keeping track of the winding number,
it is convenient to unwind the system and consider the system on an infinite line
], [ and follow the dynamics of (see Fig. 1). Thus, we consider the diffusion
of ], [, where we keep in mind that actually we consider the winding number
on a circle.
To discuss the diffusion of in the next section it is convenient to derive the
Hamiltonian. Firstly, we write the Langrangian corresponding to (15) and (16) as
L = z1 () 2 + z2 () 3 V (),
(30)
with

4
(31)
a1 + a2 sin2 ,
g3 v
2 2
1
4v 2 1
2
2
z2 () =
(32)
b1 CP + b2 CP
sin sin .
+ b3 CP
+ b4 CP
4
M
Clearly the kinetic energy in (30) is unbounded. And, for instance, a (CP-even) 4 is
required to put a lower bound on the kinetic energy. Such a 4 term should come from
a higher (non-renormalizable) operator, so it is expected to be suppressed compared to
the other CP-even 2 -term. Therefore the details of the 4 -term are unimportant for the
1 and 2 (which are assumed
following calculations. In fact strictly to first order in CP
CP
to be small), one can work with the unbounded Lagrangian (30) (and the unbounded
Hamiltonian (34)), although it may be kept in mind that a 4 -term (or higher even powers
are required for a bounded energy. We assume that these higher-order CP-even terms
of )
will make the Lagrangian convex, so that we can go to a Hamiltonian description. The
conjugate momentum then reads
z1 () =
p = 2z1 () + 3z2 () 2 ,
(33)
and the Hamiltonian

H (p, ) =
1
z2 ()
p2
p3 + V ().
4z1 ()
8[z1 ()]3
(34)
262
This is the Hamiltonian that will be used for the calculation of statistical averages in the
next section.
The point of this section was to show that the CP-odd operators in S [8] (4) projected
onto the one-dimensional path in phase space gives a p3 -term in the Hamiltonian and a
3 -term in the Lagrangian, and to give reasonable estimates for the coefficients in the
projected action and Hamiltonian. The main subject of this paper, the effect of the CP-odd
terms on the dynamics, will be discussed in the next and following sections.
3. Velocity expectation values

We study the effect of the action (4) on the diffusion of the ChernSimons number. For
1
this the Hamiltonian (34) will be used. We will work to first order in the coefficients CP
2
and CP (and, hence, to first order in z2 ). So the question is what does the diffusion of
1 and 2 . In particular, will the distribution function of
look like to first order in CP
CP
develop an asymmetry?
The first thing one may notice is that (the magnitude of) the velocity differs for positive
More precisely, at a given momentum (p) or at a given energy (H ) the
or negative .
magnitude of the velocity depends on its direction. This implies that the motion over the
barrier towards negative ChernSimons number differs from the motion towards positive
ChernSimons number. Hence, an asymmetry is expected to develop.
It is useful to define a quantity that measures the asymmetry. A simple and useful
quantity is the mean velocity, at a certain position, when moves either to the right or to
the left. We define

= ||H(
,
v ()
(35)
)(
)
H()(
)

,
v () = ||H(
(36)
)(
)
H()(
)
where the brackets denote a thermal average, and H the Heaviside function. When the
difference
= v ()
v ()

/v()
(37)
is non-zero, an asymmetry will develop. Consider, for instance, the evolution of the
distribution function of . When the particles1 that move to the left do it faster (slower)
than the particles that move to the right, the tail of the distribution function to the
left extends more (less) than to the right. Hence, the distribution function develops an
asymmetry. In particular, we expect expectation values of odd powers of to become
negative (positive) when the distribution function is initially symmetric and thermal.
In the calculation of the velocities (35) and (36), firstly one may remark that their
numerators are equal. This follows from the fact that the flux through a point

= ||H(
||H(

(
)
)(
)
)(
)
(38)
1 It is convenient to think of the distribution function as a normalized sum of a lot of single particle positions.
263
vanishes. This may be seen from

= p H ( )

(
)
= Z1
/2
H
d p H ( )e
dp
/2
= Z1 T
/2
dp
p eH = 0.
d ( )
(39)
/2
Here we used that exp H vanishes at p = , for which we had to keep in mind that
a p4 -term in the Hamiltonian is required to make it bounded and convex. (Alternatively,
1 and 2 , the momentum integrations in (39) also
strictly working to first order in CP
CP
vanish.) The normalization factor Z is defined as usual, see (51). It will drop out of the
calculation of the velocities v and v . The fact that the flux vanishes, implies that the
expectation value of remains constant:
(t) (0) = 0,
(40)
as will be discussed more fully in Section 5. Hence, asymmetric diffusion will affect only
expectation values of higher odd powers of .
The result for the numerator on the r.h.s. of (35) (equal to the numerator on the r.h.s. of
(36)) is

||H(
)(
)
= Z1
/2
dp
H
d p H H(p)( )e
/2

= Z1 T eV () .
(41)
The denominator of (35) is given by

= Z1 eV ()
H()(
)
= Z1

dp e
+T
z1 ()T

z2 ()
1
2
p3
p + 8[z ()]
3
4z1 ()
1
V ()
z2 ()

,
e

z1 ()
to first order in z2 . The denominator of (36) yields

V ()

z2 ()

1

z1 ()T T
.
e
H()( ) = Z

z1 ()
1 and 2 we have
Hence, up to first order in CP
CP

T
T
=

v () ()
(+)
z (),

2 2
z1 ()
[z1 ()]
(42)
(43)
(44)
264
and
=
/v()
2T

z (),
2 2
[z1 ()]
(45)
The relative velocity difference,

=
/vrel ()

T z2 ()
/v()
,
=
+ v ()

3/2
[z1 ()]
v ()
(46)
gives a (dimensionless) measure for the effect of CP-violation on diffusion.

In conclusion, we have established that, when = 0, the magnitude of the thermal
expectation value of the velocity depends on the direction in which moves.
Let us consider what the difference in thermal expectations values of the velocity
implies for the evolution of the distribution function of and p, P (p, , t). When
the momentum, p, is thermally distributed independent of the position , the velocity
difference is present in the entire (-)space. Now it is not hard to imagine that, when the
particles move faster to the left than to the right, the tail of an initially symmetric thermal
distribution function will (start to) extend more to the left than to the right. Therefore, a
symmetric thermal distribution function will not remain symmetric. The diffusion develops
asymmetrically when CP-odd terms like 3 are included in the Lagrangian.2 When we
translate this result back to the gauge-Higgs system it implies that the inclusion of the
CP-odd operators in (4) leads to an asymmetric diffusion of the ChernSimons number.
4. Asymmetric diffusion: a conjecture for large times

In the previous section, it was shown that there is an asymmetry in the average velocity
(46), which implies that the distribution function will develop an asymmetry. Hence,
for instance, the expectation value of the third power of the ChernSimons number,
[NCS (t) NCS (0)]3 , will be non-zero. In this section, we will argue that this expectation
value will grow linearly in time in the broken phase of the SU(2)-Higgs model.
We assume that after a transition over the sphaleron barrier the system thermalizes
before a following transition. This implies that the transitions are uncorrelated. This
assumption is reasonable when the height of the barrier, Esph , is large compared to the
temperature, T , and the motion over the barrier is damped sufficiently. (In our description
in terms of , this would require to be not-too-weakly coupled to the other modes in
the plasma.)
On the basis of this assumption we can establish the time-dependence of the third
power of the ChernSimons number. We have established in the previous section that an
asymmetry will develop in the distribution function starting from an symmetric thermal
initial distribution function. Hence, after a time /t the expectation value [NCS (/t)
2 Strictly, the argument given only implies that P (, p, t) = P (, p, t) cannot remain to hold as time

evolves. In principle, it might still be possible that dp P (, p, t) = dp P (, p, t) remains true. Since, in
the broken phase at least, we expect the momenta to be thermally distributed independent of the position, , we
do not consider this possibility further.
265
NCS (0)]3 has a non-zero value. During the time /t the distribution function has
spread over different vacua. Since we assume that the distribution is thermal in each
vacuum, during the time from /t to 2/t, from each vacuum the distribution will diffuse
asymmetrically in the same way as it did in the time interval from time 0 to time /t. The
thermalization in each vacuum implies that in each time interval /t, from each vacuum the
same diffusion process takes place (relative to that vacuum). From this argument follows
the result that the expectation value of the third power of the ChernSimons number in
the presence of CP-odd dimension-eight operators in the broken phase of the SU(2)-Higgs
model grows linearly in time. That the repetition of the same asymmetry in the diffusion
process in each time interval leads to a linear growth, may be seen, for instance, from a
random walk model, as is shown in Appendix B.
In Section 5.1, we show that in general (without the above assumption) the expectation
value of a third power of a coordinate x, x 3 , either stays constant or grows linearly in
time in thermal equilibrium. A stochastic model that will be introduced later, shows that,
with potential barriers, x 3 grows linearly in time, and that without barriers x 3 goes to a
constant. This supports the view that barriers are required to ensure that the asymmetry in
the diffusion process is independent of the position (which vacuum the system is in) and
time, from which the linear behavior follows. In the symmetric phase, different vacua are
not well separated by a barrier, therefore it is for us not possible to determine the timedependence of asymmetric ChernSimons number diffusion.
To obtain a more detailed conjecture for diffusion in the broken phase, we extend our
one-dimensional model by coupling to the other modes (that form the heat bath at
temperature T ) as in the full SU(2)-Higgs model. In this way the rate for transitions over
the barrier is given by (gv)3 sph , with (gv)3 the volume of the sphaleron. Then, we
expect

3 (t) (gv)3 sph t/vrel,sph ,
(47)
with the relative velocity difference (44) at the sphaleron configuration /vrel,sph =
/vrel (/2).
By viewing the entire space as made up of blocks of volume (gv)3 in each of which
a -coordinate is diffusing, one can translate (47) into an expectation value for the third
power of the ChernSimons number

[NCS (t) NCS (0)]3 = cV spht/vrel,sph ,
(48)
with V the volume and c a constant of order one. In going from (47) to (48)
we have in a very non-sophisticated manner included the zero mode for translation
invariance. (It is conventional to include rotational zero modes in sph and exclude the
translational zero modes.) A better way would have been to include the zero modes in
the parameterization (9) and (10). However, our aim is to investigate the occurrence of
asymmetries for which the zero modes play no essential role.
266
5. Remarks on statistical averages

The ChernSimons number diffusion is a non-equilibrium process. Still we, as many
others, see, e.g., [9], use a thermal (equilibrium) average to calculate the mean or average
velocity of the ChernSimons number (of actually). In this subsection, we will briefly
consider the use of statistical averages. In the next subsection, we return to dynamical
issues and consider the possible time-dependence that an expectation value can have.
Consider a particle in one dimension coupled to a heat bath. In general one expects that
a probability distribution function, P (p, x, t), of the position, x, and momentum, p, in the
long time limit, goes to a thermal distribution function:
t
P (p, x, t) Z 1 eH (p,x),
(49)
Z 1
with
the normalization factor. Then, the long-time limit of expectation values can
be calculated using the thermal distribution function. However, in the case that the
Hamiltonian is periodic, the thermal distribution function is not normalizable. A wellknown consequence is that x 2 (t) in a flat or periodic potential does not have a thermal
limit-value at large times (since the thermal average is not defined). Instead it grows
linearly in time. Also thermal averages of other positive powers of x cannot be calculated.
In particular, since x n is not well-defined in equilibrium, one cannot conclude from a
symmetric potential, V (x) = V (x), that for odd n it should vanish.
Nevertheless there is still a lot that may be calculated using a thermal distribution. This
is based on the notion that the long-time limit of the distribution function in a periodic
potential with period satisfies

t
P (p, x + n, t) Z1 eH (p,x),
(50)
n
where Z1 is the normalization factor, with

/2
Z =

dx
/2
dp eH .
(51)
Eq. (50) expresses the fact that when the points x + n are identified, that is when we
view the dynamics of the system on a circle with circumference , the system thermalizes
as usual. There are two different ways to look at a periodic system. When the system is
considered in terms of the coordinate x ], [, the system is dynamical. There is
diffusion, transitions to other classical vacua, etc. Whereas, when we view the system as
living on a circle, with coordinate = xmod , the system is static after thermalization (of
course, the dynamics returns when the winding number is considered).
Thermal averages can be calculated of functions, f (p, x), that are either independent
of x or -periodic in x. In the long-time limit these functions have a thermal limit value:

dx
dp f (p, x)P (p, x, t)
Z1
2
12

dx
dp f (p, x)eH (p,x).
(52)
267
In particular the average velocity can be calculated with a thermal distribution (as we did
in Section 3). Since the average velocity vanishes
x
= 0,
(53)
one expects that x(t) is constant (in the long-time limit). There exists no such argument
for expectation values of other odd powers of x. For instance, the expectation value of the
time-derivative of x 3 , 3xx
2 , is itself not well defined. Hence, x 3 does not need to vanish
and has no thermal limit-value. Indeed, as we have argued earlier, it may grow in time.
5.1. Possible power laws
Next, we will determine on general grounds the possible time-dependence of x 3 (t) in
a periodic potential in the long-time limit. We assume that it behaves as a power law
3
x (t) = C[Pin ]t p ,
(54)
with C and p constants. The brackets denote a classical average, that is, an average over
initial conditions weighted by an initial distribution function Pin . The constant C may
depend on the initial distribution.
We will use that
x(t) x(0) = 0,
(55)
see (40). It is convenient to choose the origin such that x vanishes at the initial time.
Since then x(t) = 0 for all times, and the disconnected contributions to x 3 (t) vanish.
To determine the possible values for p it is useful to consider

[x(t2 ) x(t1 )]3 = x 3 (t2 ) 3 x 2 (t2 )x(t1 ) + 3 x(t2 )x 2 (t1 ) x 3 (t1 ) .
(56)
When t2 t1 , we can apply (54) to the left-hand side

[x(t2 ) x(t1 )]3 = C[P (t1 )](t2 t1 )p .
(57)
The constant C[P (t1 )] depends on the distribution function at time t1 (instead of the initial
distribution function). A small subtlety is that P (t1 ) should be considered as a distribution
function of the shifted coordinate y(t) = x(t) x(t1 ) instead of x(t).
Consider the correlation functions x(t2 )x 2 (t1 ) and x 2 (t2 )x(t1 ) on the right-hand side
of (56) in the same limit t2 t1 . When the system thermalizes it follows from
the fact that the expectation of x goes to a constant (55) that the correlation functions
[x(t2 ) x(t1 )]x 2 (t1 ) and [x(t2 ) x(t1 )]2 x(t1 ) go to a (small) constant in the limit
t2 t1 . Therefore, when p > 0, we have in this limit

x(t2 )x 2 (t1 ) x 3 (t1 ) ,
(58)

3
2
x (t2 )x(t1 ) x (t1 ) .
(59)
When both t2 and t1 are send to infinity also, (56) gives
p
C[P (t1 )](t2 t1 )p = C[Pin ]t2 C[Pin ]t1 .
(60)
268
This equation has two possible solutions

p = 1,
with C[P (t1 )] = C[Pin ] = C,
p = 0.
(61)
(62)
Hence, the expectation value x 3 (t) either grows linearly in time or stays constant. When
the expectation value x 3 (t) goes to a constant in the long-time limit, it is not so that
this constant should be zero. The constant can be non-zero when it depends on the initial
distribution. (In Appendix C we consider a stochastic equation where x 3 indeed goes to
a non-zero constant in the long-time limit.)
It is reasonable to expect that a similar reasoning can be applied to connected correlation
function of higher powers of x, with the result that these too are either constant or linearly
growing in the long-time limit. Then, in the case that x 3 (t) grows linearly in time, the
expectation values of higher powers of x are dominated by their disconnected parts:

x 2n+1 (t)
(2n + 1)!
2n1 3!(n 1)!

n1
.
x 3 (t) x 2 (t)
(63)
For a linearly growing expectation value of x 3 , we get x 2n+1 t n .

To summarize, on general grounds it has been shown that the expectation value of
x 3 (t) goes to a constant or grows linearly in the long-time limit. As we have argued
in Section 4 we expect that, in the broken phase, the expectation value of the third power
of the ChernSimons number grows linearly in time. It is encouraging that the result of the
argumentation in Section 4 is consistent with the more general analysis presented in this
section. Nevertheless, it is desirable to have a more realistic model that would enable one
to test the conjecture that x 3 in a periodic potential with high barriers grows linearly in
time. In the next section we present such a model, namely a simple stochastic equation.
6. A stochastic equation
In Section 2 a one-dimensional Lagrangian was derived for the motion of the Chern
Simons number. In this section we want to (re)introduce the effect of the (infinite number
of) other modes [10]. They provide a heat bath at temperature T . The simplest way to
mimic their effect is by the introduction of a damping term and a stochastic noise in the
equations of motion. In fact, the motion of the ChernSimons number in the symmetric
phase is to leading order indeed determined by a stochastic (field) equation [7]. In the
broken phase, the case that is of interest to us, a local damping term and stochastic force
is probably only suitable for illustrative purposes, and not of direct quantitative interest for
ChernSimons number diffusion. Nevertheless the stochastic equation derived below will
provide a simple and realistic model for asymmetric diffusion.
Consider the Lagrangian
L=

1 2
x + x 3 + d 2 x 4 V (x),
2
(64)
with a small parameter and d a dimensionless coefficient. The x 3 is the time-reversal

non-invariant term similar (but simpler) as the 3 term in (30). The x 4 has been added
269
so that the kinetic energy is bounded from below. We will demand that the Lagrangian is
convex, which requires d > 9/24; we choose for the following d = 1.
The equation of motion is

x 1 + 3 x + 6 2 x 2 = x V .
(65)
We now introduce damping and a stochastic force in the following way

x 1 + 3 x + 6 2 x 2 = x V x + ,
(66)
with the damping coefficient and a Gaussian white noise

(t) = 0,

(67)

(t)(t ) = 2 T (t t ).
(68)
The introduction of the damping term ( x)

looks standard. However, there is one subtlety.
This may be made clear by going to the Hamiltonian formulation. The Hamiltonian
corresponding to (64) (with d = 1) is
H=

1 2
p p3 + 2 p4 + V (x) + O 3 .
2
(69)
The Hamilton equations up to order 2 including noise and damping in the same way as in
(66) read
3
x = vp = p p2 + 2 2 p3 ,
2
p = x V vp + .
(70)
(71)
However, the introduction the damping term as vp instead of p is arbitrary at this point,
and requires an explanation.
One argument for the introduction of the damping term as vp goes as follows. In
a microscopic derivation (for example, using influence-functional techniques [19]), one
would find that integrating out the modes of the heatbath yields a memory kernel of the
t
form 0 dt 9(t t )x(t ). In a short-time expansion, this memory kernel will reduce to x.
This reduction is independent of the (possible complicated) structure of the kinetic term
in the Lagrangian or Hamiltonian. Therefore, we expect that in time-reversal non-invariant
theories with complicated kinetic terms, a derivation from first principles would yield a
damping term of the form x as introduced in (66) and (70), (71).
Perhaps a more convincing argument is given by considering the equilibrium distribution function. The equation for the time evolution of distribution functions, the Fokker
Planck equation, may be derived in a standard manner, as for instance in [16]. It reads

t P = p T p + (x V ) + vp P vp x P ,
(72)
with P = P (p, x, t) the time-dependent distribution function. The static solution to this
FokkerPlanck equation is found to be exp H . This shows that damping is correctly
introduced in (66) and (71).
The remarks of Section 5 apply to the stochastic system (70) and (71). Namely, for
periodic functions or functions that are only dependent on the momentum p, the thermal
270
average determines the long-time limit (52).3 For instance odd powers of the velocity have
the following limit
n(2n + 2)! n+1
T
,
(73)
2n (n + 1)!
with n = 0, 1, 2, . . . . Similarly, it may be concluded that, for example, the long-time limit
of odd powers of sin(2x) vanishes when the potential is symmetric and periodic with
period .
The stochastic Eq. (66) introduced here will be used in Appendix C for some sample
calculations with potentials V = 0 and V = 12 2 x 2 .
vp2n+1
6.1. Numerical simulations

Perhaps the most relevant case for the stochastic model described above is when
the potential V is a periodic function similar to (25) describing a potential barrier. For
simplicity we take V to be V (x) = Vb sin2 (x) where Vb is then the potential barrier height.
The solution to this case is provided by numerical analysis of Eqs. (70) and (71) for a
considerable amount of realizations. Average quantities are then obtained by averaging
over all those realizations. The results from a numerical simulation are displayed in Fig. 2
for the evolution of the system up to large times (t 2500). One immediately sees from
Fig. 2 that x 3 (t) grows linearly in time. In Section 5 we obtained two possible long-time
behaviours for x 3 , namely, x 3 = constant or x 3 t. Therefore, the numerical results
support our arguments in favour of the latter for a periodic potential. The fact that it grows
to a positive value is somewhat surprising since our equilibrium considerations showed
is larger than the velocity in the positive
that the velocity in the negative direction v ()

direction v () for CP > 0, so /v = v v < 0. This behaviour for the asymmetry
/v in the velocities is also found in our numerical simulations. It would seem to follow
that x 3 (t) evolves in time to a negative value. This is also supported by the random
walk model considerations of Appendix B.4 The numerical results however indicate the
opposite, which means that probably non-equilibrium effects determine the sign of growth
of x 3 (t) . It might also be that the sign depends on the details of the interactions, so that a
different model could lead to a different sign.
Expectation values of higher powers of x are dominated by their disconnected parts in
the long time limit as argued in Section 5. Therefore, according to Eq. (63) one has
fn (t)
x 2n+1 (t)
(2n+1)!
x 3 (t) [x 2 (t) ]n1
2n1 3!(n1)!
1.
(74)
We have explicitly checked this for x 5 (t) and x 7 (t) (Fig. 3).
In the case of x 3 (t) , its disconnected part x 2 (t) x(t) is negligible. This is because
x(t) is constant in time and thus equal to x(0) which was chosen to be x(0) = 0 in
3 Also for periodic functions or functions that only depend on the momentum p the ergodic theorem implies
that the statistical average equals the time average. Even when the time-reversal symmetry is broken [17].
4 Nevertheless it is important to say that one cannot make a direct correspondence between the / in the
random walk model of Appendix B and the considered here in this stochastic model, not even for the sign.
271
Fig. 2. Evolution of x 3 (t) (left) and x 2 (t) (right) for = 0.1, = 0.3 and Vb = 4 (in units where T = 1)
with 8 105 realizations. Statistical errors of one standard deviation are also included, together with their time
n
evolution (small pictures). These grow as t 2 for x n (t) .
Fig. 3. Evolution of f5 (t) =
x 5 (t)
x 7 (t)
(left) and f7 (t) =
(right).
10x 3 (t) x 2 (t)
105x 3 (t) x 2 (t) 2
our simulations. Actually in the numerical results |x | is of the order of 0.02 0.1. The
large-time limit of x 3 (t) in Fig. 2 corresponds therefore to its connected part.
We have also checked in the simulations that velocity expectation values vp2n+1 have
the limit given by (73). Thus we have verified that our stochastic model thermalizes in the
way as expected in Section 5.
7. On baryon-number generation from asymmetric ChernSimons number

diffusion
A natural question to ask is whether the asymmetric diffusion of the ChernSimons
number may yield a non-zero baryon number in some out-of-equilibrium situation. This
section contains some remarks on this question. In particular, we will present a simple way
in which asymmetric diffusion of a particle, with coordinate x, will lead to a (temporary)
non-zero expectation value of x itself. It is then argued that such a situation occurred
during the electroweak phase-transition yielding a non-zero ChernSimons number (and
272
baryon number). In any electroweak baryogenesis scenario the two important questions
are: Is the resulting baryon asymmetry sufficient to account for the observed asymmetry
in the universe (1), and what are the requirements to prevent a wash out of the generated
asymmetry? We will present some estimates to answer these questions.
Let us reconsider the diffusion of a particle in one dimension with coordinate x, that
starts at x = 0. As time evolves the distribution function will spread. In the long-time limit
the distribution will become Gaussian with on top a small asymmetry (assuming that the
symmetry-breaking terms are small). We have argued that x 3 (t) grows linearly in time.
To be definite, let us say it grows in the negative direction. Then the tail of the distribution
in the negative-x direction (tail) will be larger than the tail of the distribution function in
the positive-x direction (+tail). This difference in the tails accounts for the negative values
of x 3 (and of expectation values of higher powers of x). That the expectation value of x
remains zero is due to an asymmetry in the distribution function closer to x = 0. (A nice
way to imagine this that the peak of the distribution function (the most probable position)
moves in the positive-x direction. The motion of the peak would give a growing x except
that the contribution of the tails is precisely opposite. For higher powers of x the tails
dominate and a negative and growing value is the result.) With this picture of the evolution
of the distribution function it is not hard to construct a system for which x itself becomes
(temporarily) non-zero. All that one has to do is to prevent the -tail to compensate for the
positive contribution of the peak moving in the positive-x direction (if the peak does not
move in the positive x-direction there is still a positive contribution from the region around
x = 0 and the argument goes through unchanged). Consider the diffusion of this particle in
a box with symmetrically placed walls at x = a. The tail will hit the wall earlier than
the +tail. Then the -tail can no longer compensate for the asymmetry of the distribution
function close to x = 0 (for instance, the motion of the peak of the distribution function
in the positive-x direction). Hence, x will grow and become non-zero. Eventually, when
the system goes to equilibrium, x relaxes back to zero.
Instead of putting the system in a box we could also have let the system evolve in a
harmonic potential V = 12 2 x 2 (to be superimposed on the periodic potential in which it
diffuses). Basically by the same arguments as above it follows that the expectation value of
x becomes temporarily non-zero. A slightly more general case that may be considered, is
a system that is initially in thermal equilibrium in the presence of a harmonic potential
V1 = 12 12 x 2 which changes at the initial time to V2 = 12 22 x 2 , with 12 > 22 . In the
evolution towards the new equilibrium state, the expectation value x will temporarily
become non-zero. In the next section we will show in a short-time expansion that x will
become non-zero, indeed, after such a change in a harmonic potential.
7.1. Short-time expansion
We show in the following that an instantaneous change from a potential V1 =
Vper + 12 12 x 2 to V2 = Vper + 12 22 x 2 , with Vper a periodic potential, will lead to a nonzero expectation value x . We will use the FokkerPlanck equation (72), with V2 as
the potential, for the time evolution of the distribution function P (p, x, t). The initial
273
distribution function at time t = tin is given by

Pin (p, x) = Z 1 e[ 2 (p
1
2 p 3 + 2 p 4 )+V
1 (x)]
and the expectation value x by

x(t) = dx dp xP (p, x, t).
(75)
(76)
We calculate this expectation value in a short-time expansion

x(t) =
1
xn t n ,
n!
n
with coefficients

dn
xn = dx dp x n P (p, x, t)
.
dt
t =tin
Define the operator

O = p T p + vp + (x V ) vp x ,
(77)
(78)
(79)
such that the FokkerPlanck equation (72) may be written as t P = OP , with potential
V = V2 . The coefficients xn may be calculated by

xn = dx dp xO n Pin (p, x).
(80)
The first three terms in the expansion (77) vanish
x0 = x in = 0,
(81)
x1 = vp in = 0,

x2 = (1 3p)x V2 vp 3 T p2 in = 0,
(82)
where
in =
(83)

dx
dp Pin (p, x).
For x3 we find

x3 = 3 T x2 V2 (x V2 )2 in .
(84)
(85)
When V2 = V1 this gives x3 = 0. To first order in 12 22 the result for x3 is independent

of the periodic potential, Vper ,

x3 = 3 12 22 T .
(86)
This shows that a change in the potential will lead to a (temporary) non-zero value for x ,
even if the initial and final potential is symmetric.
274
Fig. 4. Evolution of x(t) (left) and x 3 (t) (right) after the instantaneous potential change Vbefore Vafter (at
t 1000) with statistical errors included.
7.2. Numerical results

To obtain the behaviour of the evolution of x for longer times we numerically solve
the stochastic model discussed in Section 6, with the instantaneous potential change
Vbefore = Vb sin2 (x) + 12 12 x 2 Vafter = Vb sin2 (x) + 12 22 x 2 with 12 and 22 chosen to
be equal to 0.1 and 0.01, respectively. The system is first let to be equilibrated in Vbefore
and then the instantaneous potential quench is performed. The numerical results show that
indeed a temporary non-zero value for x is obtained (see Fig. 4).
7.3. Baryon-number generation
We now turn to the problem of baryon-number generation. The back reaction of
the generated baryons on the ChernSimons number may be described by the effective
potential given by the free energy at a given baryon-number B [9,10]
1
F (B) = B2 B 2 /V ,
2
(87)
with B2 = 13/6T 2 and V the volume. Due to the relation between a change in the Chern
Simons number to a change in the baryon number, the effective potential generates a force
on the ChernSimons number. As a result a non-zero baryon number will be pushed back
to B = 0 by the potential. This is the wash out of baryon number.
In terms of the effective potential is given by
Veff () =
9
(gv)3 B2 2 ,
2 2
(88)
where the factor (gv)3 is the inverse volume of a sphaleron.

Now consider the effect of a first-order electroweak phase-transition. Besides the
complicated dynamics of bubble nucleation and moving bubble walls, the quarks will
acquire a mass through the Higgs mechanism. This produces a small change in the backreaction of the quarks on sphaleron transitions. In the description of this back-reaction
in terms of an effective potential, this means that the potential (87) is changed by mass
275

corrections: B2 B2 (1 i ci m2i /T 2 ), with mi the masses of the different quark species
and ci dimensionless coefficients. As argued before, a change in a (harmonic) potential will
lead to a temporary non-zero value of the one-point function. Hence, the expectation value
of the baryon and ChernSimons number will acquire temporarily a non-zero value.
Before we discuss how temporarily may become forever, we will turn to the question
of how large we may expect the expectation value of the baryon number to grow. From the
sample calculations in Appendix C, we learn that the asymmetry has typically a maximum
value of (1 22 /12 ), with a dimensionless measure of the amount of CP-violation
and 1 and 2 the two frequencies of the potential before and after the phase transition.
The factor (1 22 /12 ), that is a measure of the amount of the departure from equilibrium,
is of the order of the mass correction to the potential (87). Therefore, it is dominated by
the top quark mass: (1 22 /12 ) m2t /6T 2 , with mt the top quark mass. In the case of
ChernSimons number diffusion we identify with the relative asymmetry in the velocity
at the sphaleron configuration /vrel,sph . Hence, when we consider the motion of in this
potential, we estimate for the maximum after the electroweak phase-transition
max
m2t
|/vrel,sph|.
6T 2
(89)
Imagine the universe made up of blocks of size (gv)3 (the size of the sphaleron), with
in each block a -variable. At the first-order electroweak phase-transition bubbles fill out
the universe and in each block the harmonic potential for changes. The time that the
bubbles need to fill out the universe is much shorter than the time for a single sphaleron
1
. Therefore, as far as the motion of is concerned, the change in the
transition (gv)3 sph
potential occurs effectively at the same time in the whole universe. This implies that the
expectation values of s in different regions in the universe move simultaneously towards
their maximum value. Then we get for the maximum baryon-number density, nB = B /V ,

nB
(gv)3
(90)
4 3 max ,

n max
T
with the photon density n = 0.24 T 3 . In this estimate we used that s in different blocks
of space move independently, similar as in the case for pure diffusion in going from (47)
to (48). This is only valid when the back reaction of the baryons on sphaleron transitions
may be considered as arising from the potential (87) and non-localities may be neglected
on the length scale (gv)1 . This is what is usually done in the literature, see, e.g., [18]. If
this assumption is incorrect extra suppression factors may be expected.
When the generated baryon-number is not washed out, the asymmetry (90) results, at
the present time, in an asymmetry

4 100 GeV 4
1
nB
2
7CP + 6CP 10
,
(91)
n max,now
M
where we used g = 0.6, mt = 170 GeV, and v T 100 GeV. This rough estimate for
the asymmetry may indicate that in the process considered here, sufficient baryons may
have been generated, provided that the CP-violation is strong enough and occurs at a nottoo-high energy-scale. It may be noted that the mass M, that gives the energy scale of new
276
physics, may not be as large as in scenarios for baryon generation based on CP-violation
through CP-odd dimension-six operators, since (91) is suppressed by four powers of M
(instead of two). Other constraints for the scenario based on asymmetric diffusion may be
more important and are discussed in the next section.
7.4. Constraints
The immediate question is whether a once created baryon asymmetry survive to present
times. The decay of the baryon-number density, nB , for small densities is given by [9]
dnB
= 9B2 sphnB .
dt
The formal solution is
nB (t) = nB (tpt )eR(t ),
(92)
(93)
with
t
R(t) = 9
dt B2 sph (t ),
(94)
tpt
with tpt the time of the electroweak phase-transition and sph (t) the time-dependent
sphaleron rate (its time-dependence enters through the changing temperature and Higgs
expectation value as the universe cools down). The requirement that a generated baryon
asymmetry survives to present times is
R = R() 1.
(95)
This may be translated into a bound on the phase transition strength

vpt > vcr 100 GeV,
(96)
with vpt the Higgs expectation value just after the first-order electroweak phase transition.
In [20] one of the authors believed that the bound to prevent the wash out of baryon
number could be much weaker. This was based on the realization that the linear response
result is not directly applicable to the situation where a non-zero expectation value of a
coordinate is generated by asymmetric diffusion in a symmetric potential. (Since according
to linear response this expectation value can never become non-zero, which as was seen
in Section 7.1, is not the case.) However, the time scale for equilibration given by linear
response (93) is still correct. Therefore, the bound (95) still applies.
In the scenario discussed there is another bound, since the asymmetry itself has to be
generated after the phase transition. Therefore, to generate a large asymmetry sufficient
sphaleron transitions must occur after the phase transition. This requires
R 1.
(97)
Hence, to generate sufficient baryons without a large wash out in the scenario that we
discuss here, the number of sphaleron transitions after the phase transition has to lie in a
277
small range around 1. We expect typically something like

e1 < R < e.
(98)
This requirement for the number of sphaleron transitions after the electroweak phasetransition may be translated into a requirement on the strength of the phase transition,
that may be indicated by the Higgs expectation value just after the transition, vpt . In
Appendix D we show that from (98) it follows that this Higgs expectation value lies in
the range

1
1
< vpt < vcr 1 +
,
vcr 1
(99)
pt Esph
pt Esph
with the inverse temperature at the phase transition pt (100 GeV)1 . Since, pt Esph
45 the range of allowed phase transition strengths is quite small. Hence, for the above
scenario for the generation of matter to work, the strength of the electroweak phase
transition has to satisfy strict bounds (99). For a given particle model these bounds may
be translated into bounds for the Higgs mass, as has already been done for the bound (95)
or (96) for the minimal supersymmetric standard model [2123].
7.5. Discussion
Perhaps it is useful to consider the difference between the effect of the dimension (the lowest-dimensional CP-odd operator) and the dimension-eight
six operator F F
and (F F )(F F
) in the context of electroweak baryogenesis. As
operators (D) (D)F F
we have argued the dimension-six operator does not affect the diffusion of the Chern
Simons number in equilibrium. However, in a non-equilibrium situation it may have an
effect. In particular, when the Higgs expectation value changes in time the dimensionsix operator introduces an effective force Feff t on the motion of the Chern
Simons number. This implies that in the case of a time-dependent Higgs expectation value
the diffusion of the ChernSimons number yields a non-zero expectation value of NCS
itself: [NCS (t) NCS (0)] sph Feff t. This in contrast to the asymmetry generated by
the dimension-eight operators in equilibrium that affect only expectation values of third
and higher powers of NCS (t) NCS (0).
It is well-known that the dimension-six operator may play a role in the generation
of the baryon asymmetry. For instance, during the electroweak phase-transition when
the Higgs expectation value grows the effective force can be used to push the Chern
Simons number and baryon number to some positive value. The problem is that during
the growth of the Higgs expectation value the sphaleron rate decreases. Therefore, it
is difficult to generate a sufficient baryon asymmetry in the short time that the Higgs
expectation grows. A way to circumvent this problem has been given in [15]. In that paper
the situation was considered that the reheating temperature after inflation is below the
temperature at which the electroweak phase-transition takes place. This means that there
never was an electroweak phase-transition. The Higgs expectation value grows during
the reheating of the universe. At this time the CP-odd dimension-six operator generates
the earlier mentioned effective force. Further it has been argued in [15] that during the
period of reheating the exponential suppression of the sphaleron rate is absent. Therefore,
278
sufficient sphaleron transitions can take place and a baryon asymmetry may be generated.
Since the sphaleron rate becomes exponentially suppressed when the plasma thermalizes,
a generated baryon asymmetry may be preserved.
When the asymmetric diffusion of the ChernSimons number from CP-odd dimensioneight operators plays a role in the generation of a baryon asymmetry, as we have argued is
possible, then the time-scales for the generation and the wash out of a baryon asymmetry
are similar. Therefore, there is no need to try to avoid the exponential suppression of
the sphaleron rate. The price to pay is for these similar time scales is that the strength
of the phase transition needs to be finely tuned in order to ensure sufficient sphaleron
transitions after the phase-transition and avoid a subsequent wash out of baryon number
(see Section 7.4).
Of course, in Section 7 we have neglected various important dynamical aspects of the
problem. Such as the motion of the bubble walls of a first-order phase transition, the motion
of the baryons and dynamical aspects of their back-reaction on the motion of the Chern
Simons number, the growth in energy of sphaleron configuration as a bubble wall passes a
certain region in space (that is as the Higgs expectation value increases), the non-Brownian
beginning of the motion of the ChernSimons number [24], etc. These neglected aspects
may well modify the here presented estimate for the final baryon asymmetry. It is even
possible that, being non-equilibrium processes, one of these will provide, in combination
the included CP-violation, a different mechanism for baryon number generation. If so, the
dimension-eight operators in (4) may prove more effective than the dim-six operator in (3)
in providing the necessary CP-violation when the Higgs expectation value is not rapidly
changing. Although all these non-equilibrium phenomena could yield a non-zero baryon
number in combination with CP-violation, we believe that the mechanism presented in
Section 7 is the most natural way to generate an asymmetry on the basis of the CP-odd
dimension-eight operators.
8. Summary
We have argued that the inclusion of CP-odd operators, especially the dimension-eight
operators in (4), in an effective action will result in an asymmetric diffusion of the Chern
Simons number (in the absence of fermions). That is, in thermal equilibrium the correlation
function [NCS (t) NCS (0)]3 becomes non-zero. In Section 5, we noted that, on general
grounds, the expectation value of the third power of a coordinate either grows linearly in
time or goes to a constant in the long time limit. We have argued that in the broken phase
[NCS (t) NCS (0)]3 grows linearly in time. A more detailed conjecture was given in
(48). Unfortunately, the study presented here did not allow us to determine the sign of the
expectation value (the direction of the growth).
Further we noted that asymmetric diffusion may lead to a non-zero expectation value
of the coordinate itself in a non-equilibrium situation. For instance, when the potential
is changed (even if before, during, and after the change the potential is symmetric). This
may have implications for the generation of the baryon asymmetry in the early universe.
Although there are severe constraints on the strength of the electroweak phase-transition,
279
we found that the observed baryon asymmetry could be generated by asymmetric diffusion
in combination with the electroweak phase-transition.
Acknowledgements
We would like to thank Jan Smit and Jeroen Vink for useful comments and discussions.
A.A. was supported by Foundation FOM of the Netherlands.
Appendix A. Relation between the ChernSimons number and

The change in the ChernSimons number is given by
g2
NCS (t) NCS (0) =
32 2
t
dt
a a
.
d 3 x F
F
(A.1)
Using the parametrization (9) of the non-contractible loop in configuration space found by
Manton [12], Eq. (A.1) may be rewritten in terms of :
12
NCS (t) NCS (0) =
t
dt sin2

dr(r f )f (1 f ).
(A.2)
The time and spatial integrations may be factorized. The spatial integration is completely
determined by the boundary values of the function f (it does therefore not depend on the
Ansatz that is made for f , given the boundary conditions)
1
dr (r f )f (1 f ) = .
6
(A.3)
Hence, the relation between the change in ChernSimons number and is given by
2
NCS (t) NCS (0) =
t
dt sin2 .
(A.4)
For instance, when we change from (0) = 0 to (t) = n , crossing n barriers, the
ChernSimons changes as
NCS (t) NCS (0) = n,
as expected.
(A.5)
280
Appendix B. Random walk model

A random walk model is a convenient toy model to study diffusion. For asymmetric
diffusion we consider a one-dimensional random walk model with probabilities
PR = (1 + )/2,
PL = (1 )/2,
(B.1)
of moving right (R) or left (L). The distance the particle moves in one time step differs
also between the two different directions. When the particle moves to the right it travels a
distance /xR and when it moves to the left a distance /xL :
/xR = (1 + )1 ,
xL = (1 /)1 .
(B.2)
The relation
PR /xR PL /xL = 0,
(B.3)
implies that the flux vanishes.

Thus, the random walk model defined above contains the two important features of the
motion of the ChernSimons number established in Section 3. The flux vanishes (at every
point), see (39), and the left and right velocities differ, see (45), which translates in the
different distances (B.2) travelled in one time step in the random walk model.
The distribution function of the number of steps to the right, R and the number of steps
to the left, L, after a total of N = L + R steps, reads
N! 1
(1 )L (1 + )R .
L!R! 2N
When the particle starts at the initial time, N = 0, at position x = 0, x is given by
P (L, R, N) = L+R,N
(B.4)
x = R/xR L/xL .
(B.5)
The relation L + R = N and (B.5) can be used to convert (B.4) in a distribution function
of x. Instead, we will calculate expectation values. Let us start with x . As mentioned
before, the vanishing of the flux should imply that x is constant and, since the particle
starts at x = 0, is zero. This may be checked for the random walk model. The expectation
value of x is given by
x(N) =
N
N

N! 1
(1 )L (1 + )R
L!R! 2N
L=0 R=0

R(1 + )1 L(1 )1 .
L+R,N
(B.6)
We define R = R 1 and L = L 1 and write (B.6) as

x(N) =
N N1

L+R ,N1
L=0 R =0
N1
N

L =0 R=0
N! 1

(1 )L (1 + )R

N
L!R ! 2
L +R,N1
N! 1

(1 )L (1 + )R .

N
L !R! 2
(B.7)
281
Since in the first line on the right-hand side of (B.7) the term L = N in the sum over L does
not contribute due to the delta function, and similarly the term R = N does not contribute
in the second line, the sums in the first and second line cancel. Hence,
x(N) = 0,
(B.8)
in accord with (39).

Next, consider the expectation value

N
N

N! 1
L+R,N
(1 )L (1 + )R
x 3 (N) =
L!R! 2N
L=0 R=0
3

R(1 + )1 L(1 )1 .
(B.9)
This expression may be evaluated in a similar manner as (B.6). The result is

1
3
x (N) = N (1 + )2 (1 )2 .
(B.10)
2
Hence, for positive , the expectation value x 3 is negative and grows linearly in time.
This calculation supports our argument that when the asymmetry is independent of
space and constant in time the expectation value x 3 grows linearly in time.
Appendix C. Some solutions to the stochastic equation
In this appendix we present some calculations using the Langevin equation (66) with a
constant potential or a harmonic potential.
We start with a calculation of x in the case V = 0. The Langevin equation (66) with
V = 0 reads

x 1 + 3 x + 6 2 x 2 = x + ,
(C.1)
with
(t) = 0,
(C.2)
(t)(t ) = 2 T (t t ),
(C.3)
and initial conditions x(0) = xin , x = vin .

We solve the equations of motion in an expansion in :
x(t) = x0 (t) + x1 (t) + ,
(C.4)
where x0 and x1 satisfy the equations

x0 = x 0 + ,
(C.5)
x1 = x 1 3x0x0 ,
(C.6)
with initial conditions

x0 (0) = xin ,
x0 (0) = vin ,
(C.7)
x1 (0) = 0,
x1 (0) = 0.
(C.8)
282
The solution for x0 reads

vin
(1 e t ) +
x0 (t) = xin +
t
dt G(t t )(t ),
(C.9)
0

with the Green function G(t t ) = [1 e (t t ) ]/ . The solution for x1 is

t
x1 (t) = 3
dt G(t t )x0 (t )x0 (t ).
(C.10)
When the average over the stochastic force is performed, we obtain

vin
1 e t ,
x0 (t) = xin +

1
3 2
t
2 t
e
.
x1 (t) =
+e
v T
in
2
(C.11)
(C.12)
We take thermal initial conditions: xin = 0 and vin is thermally distributed. The average
over initial momenta gives for the initial velocity and velocity squared
vin in = 0,
2
vin
= T + O().
(C.13)
Hence the average over initial conditions of (C.11) and (C.12) gives

x0 (t) in = 0,

x1 (t) in = O().
(C.15)
Therefore the combined thermal and stochastic average of x vanishes,

x(t) in = 0,
(C.16)
(C.14)
up to order .
This is an explicit check that when the velocity is thermally distributed, and hence
x
= 0, that then x(t) = constant.
The second quantity that we have calculated is x 3 (t) . We find that it goes to a constant
(that depends on the initial distribution) in the long time limit. For the initial conditions
xin = 0 and vin thermally distributed, we find

3

21 T 2
x (t) in
(C.17)
.
2
Hence, differently than we have argued for the diffusion of the ChernSimons number,
in this case the expectation value of the third power does not grow linearly in time.
Since there are no high barriers, we do not expect the velocity to thermally distributed,
independent of the position in space. Therefore, the asymmetry in the velocities does
not need to be independent of the position in space. In this way the argument for the
linear growth of the expectation of x 3 (t) does not hold in this case. Hence, the fact that
x 3 (t) goes to a constant when V = 0, does not contradict the expected linear growth of
283
[NCS (t) NCS (0)]3 in the long time limit. It does raise the following question. When we
add a periodic potential Vb sin2 (x), how does the long time behavior of x 3 (t) depend on
Vb ? We know that for Vb = 0 it goes to a constant, and we have argued that for Vb /T 1
it grows linearly in time. So, at what value of Vb does the behavior change?
Next, we consider the time-evolution in a harmonic potential V = 12 22 x 2 . As initial
2 and v is thermally distributed. This
distribution we use for the position exp 12 12 xin
in
situation may be viewed as follows. The system is in thermal equilibrium before the initial
time, t = 0, with a potential V = 12 12 x 2 . At the initial time the harmonic potential changes
1 2 2
1 2 2
2 1 x 2 2 x . We calculate the following time evolution of x . In the case without
damping, = 0, We get to first order in

22
T
1 2 sin(2 t) 1 cos(2 t) .
x(t) =
(C.18)
2
1
For the case , we present only the contribution of the slowest decaying mode

3T 22
22 (2 / )t
2
x(t) =
(C.19)
+ O e2(2 / )t .
1
e 2
3
2
2
1
We learn from (C.18) and (C.19) that the expectation value x becomes non-zero after the
potential is changed, as we have argued in Section 7. And further that the magnitude of x
is proportional to (1 22 /12 ) and .
Appendix D. Bound on the strength of the phase transition
In this appendix we show how the bound (98)
e1 < R < e,
(D.1)
with

R=9
dt B2 (t ),
(D.2)
tpt
may be translated into a bound on the Higgs expectation immediately after the electroweak
phase transition, vpt .
In order to simplify the following calculations we take in (7) to be constant, then
39
T eEsph .
(D.3)
2
As a further simplifying approximation, we will assume that the Higgs expectation value
stays constant after the phase transition and that the only time dependence enters through
the temperature as [25]
9B2 sph =
T 2 = 0.03
Mpl
,
t
where Mpl = 1.2 1019 GeV is the Planck mass.
(D.4)
284
When the Higgs expectation value, v, is time-independent so is the sphaleron energy

Esph , and the integration in (D.2) can easily be performed. The result is
R=K
Mpl pt Esph
e
,
Esph
(D.5)
with K = 1.08 and pt the inverse temperature at the electroweak phase-transition.

It is useful to determine the value vpt = vcr for which
R = 1.
(D.6)
Inserting the result (D.5) for R and taking the natural logarithm gives
ln R = ln K
Mpl
pt Esph = 0.
Esph
(D.7)
From [9] we obtain the value for = 1.3 104 . Inserting this value for in K and then in
(D.7) yields
pt Esph = 45.
(D.8)
This gives for the phase transition strength

vcr Tpt 100 GeV.
(D.9)
In [11] the time- or temperature-dependence of the Higgs expectation value is taken into
account in the derivation of vcr . We are more interested in the window of phase-transition
strengths determined by (D.1).
A small shift of vpt by an amount of v, vpt = vcr + v, saturates the lower bound
in (D.1)
R(vcr + v) = e1 .
(D.10)
This equation gives for v

v
(1 pt Esph ) = 1.
vcr
(D.11)
Thus the bound (D.1) translates into the following bound for the phase-transition strength
vpt

1
1
< vpt < vcr 1 +
,
vcr 1
(D.12)
pt Esph
pt Esph
for pt Esph 1.
References
[1]
[2]
[3]
[4]
K. Olive, Nucl. Phys. Proc. Suppl. 80 (2000) 79.

A.D. Sakharov, JETP Lett. 5 (1967) 24.
G. t Hooft, Phys. Rev. Lett. 37 (1976) 8.
V. Kuzmin, V. Rubakov, M. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
V. Rubakov, M. Shaposhnikov, Phys. Usp. 39 (1996) 461.

P. Arnold, D. Son, L. Yaffe, Phys. Rev. D 55 (1997) 6264.
D. Bdeker, Phys. Lett. B 426 (1998) 169.
D. Bdeker, G. Moore, K. Rummukainen, Phys. Rev. D 61 (2000) 056003.
P. Arnold, L. McLerran, Phys. Rev. D 36 (1987) 581.
S. Khlebnikov, M. Shaposhnikov, Nucl. Phys. B 308 (1988) 885.
G.D. Moore, Phys. Rev. D 59 (1999) 014503.
N.S. Manton, Phys. Rev. D 28 (1983) 2019.
T. Akiba, H. Kikuchi, T. Yanagida, Phys. Rev. D 38 (1988) 1937.
F. Klinkhamer, N.S. Manton, Phys. Rev. D 30 (1984) 2212.
J. Garcia-Bellido, D.Yu. Grigoriev, A. Kusenko, M. Shaposhnikov, Phys. Rev. D 60 (1999) 123504.
J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon, Oxford, 1989.
Ya. Sinai, Topics in Ergodic Theory, Princeton University Press, 1994.
M. Joyce, T. Prokopec, N. Turok, Phys. Rev. D 53 (1996) 2930;
M. Joyce, T. Prokopec, N. Turok, Phys. Rev. D 53 (1996) 2957.
R.P. Feynman, A. Hibbs, Quantum Mechanics and Path Integrals, McGrawHill, 1965.
B.J. Nauta, Phys. Lett. B 478 (2000) 275.
J.M. Moreno, M. Quiros, M. Seco, Nucl. Phys. B 526 (1998) 489.
J.M. Cline, G.D. Moore, Phys. Rev. Lett. 81 (1998) 3315.
M. Laine, K. Rummukainen, Nucl. Phys. B 597 (2001) 23.
A. Rajantie, P.M. Saffin, E.J. Copeland, Phys. Rev. D 63 (2001) 123512.
E. Kolb, M. Turner, The Early Universe, AddisonWesley, 1989.
285

The exclusive B (K, K )+ decays in a CP

softly broken two Higgs doublet model
Gray Erkol, Grsevil Turan
Middle East Technical University, Physics Department Inonu Bul. 06531, Ankara, Turkey
Abstract
We study the differential branching ratio, forward-backward asymmetry, CP-violating asymmetry,
CP-violating asymmetry in the forward-backward asymmetry and polarization asymmetries in the
B K+ and B K + decays in the context of a CP softly broken two Higgs doublet
model. We analyze the dependencies of these observables on the model parameters by paying a
special attention to the effects of neutral Higgs boson (NHB) exchanges and possible CP-violating
effects. We find that NHB effects are quite significant for both decays. A combined analysis of abovementioned observables seems to be very promising as a testing ground for new physics beyond the
SM, especially for the existence of the CP-violating phase in the theory. 2002 Published by Elsevier
Science B.V.
1. Introduction
At the quark level, B K + and B K+ decays ( = e, , ) are induced by
the b s+ transition, which has received considerable attention [112], as a potential
testing ground for the effective Hamiltonian describing the flavor changing neutral current
processes in B decays. They are also expected to open a window to investigate the new
physics prior to any possible experimental clue about it.
It is well known that the inclusive rare decays, although theoretically cleaner than the
exclusive ones, are more difficult to measure. This fact stimulates the study of the exclusive
decays, but the situation is contrary then: their experimental study is easy but the theoretical
investigation is hard. For inclusive semileptonic B-meson decays, the physical observables
can be calculated in heavy quark effective theory (HQET) [13]; however, the description
E-mail addresses: gurerk@newton.physics.metu.edu.tr (G. Erkol), gsevgur@metu.edu.tr (G. Turan).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 4 9 - 8
G. Erkol, G. Turan / Nuclear Physics B 635 (2002) 286308
287
of the exclusive decays requires the additional knowledge of decay form factors, i.e., the
matrix elements of the effective Hamiltonian between the initial B and final meson states.
Finding these hadronic transition matrix elements is related to the nonperturbative sector
of the QCD and should be calculated by means of a nonperturbative approach. The form
factors for B decays into K and K have been calculated in the framework of different
methods, such as chiral theory [14], three point QCD sum rules method [15], relativistic
quark model [16], effective heavy quark theory [17], and light cone sum rules [18,19].
From the experimental side, there exist upper limits on the branching ratios of B 0
0
K + and B + K + + , given by CDF collaboration [20]

BR B 0 K 0 + < 4.0 106 ,

BR B + K + + < 5.2 106 .
With these measured upper limits and also the recent measurement of the branching ratio
of B K+ with = e, ,

6
BR B K+ = 0.75+0.25
0.21 0.09 10 ,
at KEK [21], the processes B (K, K )+ have received great interest so that
their theoretical calculation has been the subject of many investigations in the SM
and beyond, such as the SM with fourth generation, multi-Higgs doublet models,
minimal supersymmetric extension of the SM (MSSM) and in a model independent
method [2238].
In this paper we will investigate the exclusive B (K, K )+ decays in a CP softly
broken two Higgs doublet model, which is called model IV in the literature [39].
CP violating asymmetry ACP is an important observable that may provide valuable
information about the models used. In the SM the source of CP violation is the complex
CabibboKobayashiMaskawa (CKM) matrix elements and due to unitarity of this matrix
,A
+
together with the smallness of the term Vub Vus
CP for B (K, K ) decays almost
vanishes in the SM. However, like many extensions of the SM, model IV predicts a new
source of CP violation so that we have an opportunity to investigate the physics beyond
the SM by analysing the CP violating effects.
In model IV, up-type quarks get masses from Yukawa couplings to the one Higgs
doublet H2 , and down-type quarks and leptons get masses from another Higgs doublet H1 .
In such a 2HDM, all the parameters in the Higgs potential are real so that it is CPconserving, but one allows the real and imaginary parts of 1+ 2 to have different selfcouplings so that the phase , which comes from the expectation value of Higgs field,
cannot be rotated away, which breaks the CP symmetry (for details, see Ref. [39]). In
model IV, interaction vertices of the Higgs bosons and the down-type quarks and leptons
depend on the CP violating phase and the ratio tan = v2 /v1 , where v1 and v2 are the
vacuum expectation values of the first and the second Higgs doublet respectively, and they

are free parameters in the model. The constraints on tan are usually obtained from BB,

KK mixing, b s decay width, semileptonic decay b c and is given by [40]

mH
,
0.7 tan 0.52
(1)
1 GeV
and the lower bound mH 200 GeV has also been given in [40].
288
In addition to the CP asymmetry ACP , differential or total branching ratios and the
forward-backward asymmetries, polarization asymmetries are also thought to play an
important role in further investigations of the structure of the SM and for establising
new physics beyond it. It has been pointed out in Refs. [10,28] that the longitudinal
polarization PL of the final lepton may be accessible in the B (K, K ) + mode in
the near future. It has been shown [12] that together with PL , the other two orthogonal
components of polarization, PT and PN , are crucial for the + mode since these
three components contain the independent, but complementary information because they
involve different combinations of Wilson coefficients in addition to the fact that they are
proportional to m /mb . Lepton polarizations in B K(K )+ decays are analyzed in
the model II version of the 2HDM and in a general model independent way in Refs. [30]
([35]) and [41] ([42]), respectively. Ref. [43] gives an analysis of the lepton polarization
asymmetries in the processes B (K, K )+ in a supersymmetric context.
As pointed out before, (see, for example, [4448]), in models with two Higgs doublets,
like MSSM, 2HDM, etc., neutral Higgs boson (NHB) effects could contribute largely to the
semileptonic rare B meson decays, especially for heavy lepton modes and for large tan .
However, in the literature there was a disagreement about the results of NHB exchange
diagrams contributing to b s+ transition in the context of the 2HDM [46,47]. This
situation seems to be resolved now [47,49], and in view of new forms of the Wilson
coefficients CQ1 and CQ1 due to NHB effects, it is quite worthwhile to return to the
exclusive processes B (K, K ) + in order to investigate the NHB effects together
with the CP violating effects in model IV.
The paper is organized as follows. In Section 2, after we give the effective Hamiltonian
and the definitions of the form factors, we introduce basic formulas of observables.
Section 3 is devoted to the numerical analysis and discussion of our results.
2. Effective Hamiltonian and form factors

At the quark level, the effective Hamiltonian describing the rare semileptonic b
s+ transition can be obtained by integrating out the top quark, Higgs bosons and W ,
Z bosons:

10
10

4GF
Ci ()Oi () +
CQi ()Qi () ,
Heff = Vt b Vt s
(2)
2
i=1
i=1
where Oi are currentcurrent (i = 1, 2), penguin (i = 1, . . . , 6), magnetic penguin (i =
7, 8) and semileptonic (i = 9, 10) operators and Ci () are the corresponding Wilson
coefficients renormalized at the scale [50,51]. The additional operators Qi , (i =
1, . . . , 10) and their Wilson coefficients are due to the NHB exchange diagrams, which
can be found in [45,47,49].
Neglecting the mass of the s quark, the above Hamiltonian leads to the following matrix
element:

GF
+ C10 s (1 5 )b
5
M = Vt b Vts C9eff s (1 5 )b
2 2
2C7eff

mb
i
q
(1
+
)b

,
5
q2
289
(3)
where q is the momentum transfer. Here, Wilson coefficient C9eff () contains a perturbative
part and a part coming from long-distance effects due to conversion of the real cc
into
lepton pair + :
pert
C9eff () = C9 () + Yreson (s),
(4)
where
pert
C9 () = C92HDM ()

+ h(z, s) 3C1 () + C2 () + 3C3 () + C4 () + 3C5 () + C6 ()

1
h(1, s) 4C3 () + 4C4 () + 3C5 () + C6 ()
2

1
h(0, s) C3 () + 3C4 ()
2

2
+ 3C3 () + C4 () + 3C5 () + C6 () ,
(5)
9
and z = mc /mb . The functions h(z, s) arises from the one loop contributions of the four
quark operators O1 , . . . , O6 and their explicit forms can be found in [51]. It is possible to
parametrize the resonance cc
contribution Yreson (s) in Eq. (4) using a BreitWigner shape
with normalizations fixed by data which is given by [52]
(Vi + )mV
3
i
2
2 m + im
em
sm
Vi
Vi Vi
B
Vi =i

(3C1 () + C2 () + 3C3 () + C4 () + 3C5 () + C6 ()) .
Yreson (s) =
(6)
The phenomenological parameter in Eq. (6) is taken as 2.3 so as to reproduce the correct
= BR(B J /X)BR(J /
value of the branching ratio BR(B J /X X)
X).
Next we proceed to calculate the differential branching ratio dBR/ds, forwardbackward asymmetry AFB , CP violating asymmetry ACP , CP asymmetry in the forwardbackward asymmetry ACP (AFB ) and finally the lepton polarization asymmetries of the
B K+ and B K + decays. In order to find these physically measurable quantities at hadronic level, the necessary matrix elements are M(pM )|s (1 5 )b|B(pB ) ,
M(pM )|s i q (1 + 5 )b|B(pB ) and M(pM )|s (1 + 5 )b|B(pB ) for M = K, K ,
which can be parametrized in terms of form factors. Using the parametrization of the form
factors as in [30] and [32], we find the amplitudes governing the B K+ and the
B K + decays as follows:

GF

MBK = Vt b Vts 2A1 pK + B1 q
2 2

5 + E1
+ F1
5 ,
+ 2G1 p + D1 q
(7)
290
and

GF
2A9 pK
MBK = Vt b Vts
pB
2 2

+ iB iC(pB + pK ) q iD( q)q
5 2E9 pK
+
pB

q
+ iF iG q (pB + pK ) iH q q + i Q

5 N q ,
+ i
(8)
where
fT
,
mB + mK

mB (m2 m2 q 2 )
,
B1 = C9eff f + + f + 2C7eff 2 fT B
q
mB + mK
G1 = C10 f + ,

D1 = C10 f + + f ,

1
2
mB m2K f + + f q 2 ,
E1 = CQ1
mb

1
2
F1 = CQ2
mB m2K f + + f q 2 ,
mb
V
mb
+ 4 2 C7eff T1 ,
A = C9eff
mB + mK
q

4mb
eff
eff
B = (mB + mK ) C9 A1 + 2 (mB mK )C7 T2 ,
q

A
m
q2
2
b
C = C9eff
+ 4 2 C7eff T2 + 2
T
3 ,
mB + mK
q
mB m2K
mK
mb
D = 2C9eff 2 (A3 A0 ) 4C7eff 2 T3 ,
q
q
V
,
E = C10
mB + mK
F = C10 (mB + mK )A1 ,
A2
,
G = C10
mB + mK
mK
H = 2C10 2 (A3 A0 ),
q
mK
A0 ,
Q = 2CQ1
mb
mK
N = 2CQ2
A0 .
mb
A1 = C9eff f + 2mB C7eff
(9)
Here f + , f and fT and A0 , A1 , A2 , A3 , V , T1 , T2 and T3 are the relevant form factors in

B K and B K transitions, respectively. For B K, we use the results calculated
291
Table 1
B K transition form factors in light cone QCD sum
rules.
ABK
1
ABK
2
V BK
T1BK
T2BK
BK
T3
F (0)
aF
bF
0.34 0.05
0.60
0.023
0.28 0.04
1.18
0.281
0.46 0.07
1.55
0.575
0.19 0.03
1.59
0.615
0.19 0.03
0.49
0.241
0.13 0.02
1.20
0.098
in the light cone QCD sum rules framework, which can be written in the following pole
forms [30]

0.29
,
f + q2 =
q2
1 23.7

0.21
,
f q2 =
q2
1 24.3

0.31
fT q 2 =
(10)
.
2
1 q23
As for the B K transition, we use the result of [19], where q 2 dependence of the form
factors can be represented in terms of three parameters as given by

F (0)
F q2 =
2 2 ,
q2
1 aF 2 + bF q 2
mB
mB
where the values of parameters F (0), aF and bF for the B K decay are listed
in Table 1. The form factors A0 and A3 in Eq. (9) can be found from the following
parametrization:
T3 q 2
,
mK mb
mB + mK
mB mK
A3 =
(11)
A1
A2 .
2mK
2mK
Using Eqs. (7) and (8) and performing summation over final lepton polarization, we get
for the double differential decay rates:
2

G2 2
d 2 BK
= 11F 5 Vt b Vts m3B v m2B 1 z2 v 2 |A1 |2 + s v 2 |E1 |2 + |F1 |2
ds dz
2

+ m2B 1 z2 v 2 + 16rm2 |G1 |2 + 4sm2 |D1 |2
+ 4m2 (1 r s) Re G1 D1 + 2vm z Re A1 E1

+ 2m (1 r s) Re G1 F1 + s Re D1 F1 ,
(12)
A0 = A3
292
and

2 G2F
d 2 BK
2
Vt b Vt s
= 15
v 4s 2 + v 2 z2 1 |A|2
5
ds dz
2 mB

+ 4v 2 sm4B 1 + z2 |E|2 + 16m2B svz Re BE + Re AF

1

1 z2 v 2 + 2r s 5 2v 2 |B|2 + m4B 2 1 z2 v 2 |C|2
+
r

+ 1 z2 v 2 2r s 1 4v 2 |F |2

2

+ m4B 1 + r 1 v 2 z2

+ 1 + z2 st 2 8(1 + r )t 2 |G|2

+ 2m2B W 1 z2 v 2 Re BC

2m2B W 1 z2 v 2 4t 2 Re F G

+ m2B 4sm m |H |2 + Re H N + s |N|2 + v 2 |Q|2

4t Re F 2tH + N /mB + 4(1 r )m Re G 2m H + N

+ 4tmB vz2 Re W B + m2B W2 4r s C Q
(13)
.

2 + (s
Here s = q 2 /m2B , r() = m2K(K ) /m2B , v = 1 4t 2 /s, t = ml /mB , () = r()
1)2 2r() (s + 1), W() = 1 + r() + s and z = cos , where is the angle between the
three-momentum of the lepton and that of the B-meson in the center of mass frame of
the dileptons + .
Having established the double differential decay rates, let us now consider the forwardbackward asymmetry AFB of the lepton pair, which is defined as
0
1
d 2
d 2
0 dz ds dz 1 dz ds dz
.
AFB (s) = 1
(14)
0
d 2
d 2
0 dz ds dz + 1 dz ds dz
The AFB s for the B K+ and B K + decays are calculated to be

1

2
dsv
=
ds
tv
Re
A
E
,
ABK
1 1
FB
=
ABK
FB
(15)

ds 2m3B v 2 4mB s Re BE + Re AF
1

W Re BQ + mB Re CQ
ds v
.
+
r
(16)
We note that in the SM, AFB in B K+ decay is zero because of the fact that
hadronic current for B K transition does not have any axial counterpart. As seen from
Eq. (15), it is also zero in model IV unless we do not take into effect the NHB exchanges.
293
Therefore, B K+ decay may be a good candidate for testing the existence and the
importance of NHB effects.
In this work, we also analyse the CP violating asymmetry ACP , which is defined as
ACP =
M
+ )
d /ds(B M+ ) d /ds(B
,
M
+ )
d /ds(B M+ ) + d /ds(B
(17)
where M = K, K and d /ds are the corresponding differential decay rates, which are
obtained by integrating the expressions in Eqs. (12) and (13) over the angle variable
G2 2
d BK
= 10F 5 |Vt b Vts |2 m3B v,
ds
2
(18)
where

4m2
1
(12rs + )|G1 |2
= m2B 3 v 2 |A1 |2 + |G1 |2 +
3
3s

+ 4m2 s|D1 |2 + s v 2 |E1 |2 + |F1 |2 + 4m2 (1 r s) Re G1 D1

+ 2m (1 r s) Re G1 F1 + s Re D1 F1 ,
(19)
and

2 G2F mB
d BK
2
V
=
V
v
t
b
t
s
ds
212 5
(20)
where

8
= m6B s 3 v 2 |A|2 + 2v 2 |E|2
3

4
m2B m Re F m2B (1 r )G m2B sH N
r

1
1
4
+ mB sv 2 |Q|2 + m2B 3 v 2 |C|2 + s|N|2 + m2B s 2 1 v 2 |H |2
r
3

2

3 v 2 W 3s 1 v 2 Re F G 2s 1 v 2 Re F H
+
3

2

2
2
2
+ 2mB s(1 r ) 1 v Re GH + 3 v W Re BC
3

1 2
+
mB ( + 12r s) 3 v 2 |B|2 + m4B 3 v 2
3r

3s(s 2r 2) 1 v 2 |G|2 + 3 v 2 + 24r sv 2 |F |2 .
(21)
We would also like to present the CP asymmetry in the forward-backward asymmetry
ACP (AFB ), which is another observable that can give information about the physics beyond
the SM. It is defined as
AFB A FB
,
ACP (AFB ) =
(22)
AFB + A FB
where A FB is the CP conjugate of AFB .
294
Finally, we would like to discuss the lepton polarization effects for the B K+
and B K + decays. The polarization asymmetries of the final lepton is defined as
Pn (s) =
(d (Sn )/ds) (d (Sn )/ds)

(d (Sn )/ds) + (d (Sn )/ds)
(23)
for n = L, N, T . Here, PL , PT and PN are the longitudinal, transversal and normal

polarizations, respectively. The unit vectors Sn are defined as follows:

p+
,
SL = (0, eL ) = 0,
|p+ |

p p+
SN = (0, eN ) = 0,
,
|p p+ |

ST = (0, eT ) = 0, eN eL ,
(24)
where p = pK , pK and p+ are the three-momenta of K, K and + , respectively. The
longitudinal unit vector SL is boosted to the CM frame of + by Lorentz transformation:

|p+ | E p+
.
,
SL,CM =
(25)
m m |p+ |
It follows from the definition of unit vectors Sn that PT lies in the decay plane while PN is
perpendicular to it, and they are not changed by the boost.
After some algebra, we obtain the following expressions for the polarization components of the + lepton in B K+ and B K + decays:

4m3B
2mB Re A1 G1 6t (1 r s) Re G1 E1 6st Re D1 E1
3

+ 3s Re E1 F1 ,
m3B
=
2mB (1 r s)t Re A1 G1 2mB st Re A1 D1
s

+ s 4t 2 Re G1 E1 + s Re A1 F1 ,

m3B s
(26)
=
2mB t Im G1 D1 Im A1 E1 Im G1 F1 ,
PLBK =
PTBK
PNBK
and
PLBK =

4m2B
mB t Re F + m2B (1 r )G + m2B sH Q
r
1

8
+ m4B r s Re AE Re B m2B (1 r s)G ( + 12r s)F
3
3

1

+ C m2B (1 r s)F 2 m4B G + m2B s Re QN ,
2
PTBK
PNBK
295
m2B
=
8m2B r st Re AB m2B t (1 r ) (1 r s) Re BG
r s

m2B Re CG m2B t Re F C

1
mB s Re B(1 r s) m2B C N
2
1

+ m4B st Re CH + smB 2 Re F (1 r s) m2B G Q
2
2

t (1 r s) Re mB sH F B
,

1
m3B s
=
4mB tr Im BE + AF + m2B Im 2mB tH G
r
2

GN + CQ (1 + 3r s)mB t Im GF

1
1
(1 r s) Im mB tH F NF + QB .
2
2
3. Numerical results and discussion

In this section we present the numerical analysis of the exclusive decays B K+
and B K + in model IV. We will give the results for only = channel, which
demonstrates the NHB effects more manifestly. The input parameters we used in this
analysis are as follows:
mK = 0.493 GeV,
mB = 5.28 GeV,
mb = 4.8 GeV,
mc = 1.4 GeV,
m = 1.77 GeV,
mK = 0.893 GeV,
mH 0 = 125 GeV,
mh0 = 100 GeV,
mH = 250 GeV,

Vt b V = 0.04,
ts
= 129,
GF = 1.17 105 GeV2 ,
B = 1.64 1012 s.
(27)
The masses of the charged and neutral Higgs bosons, mH , mH 0 , mA0 and mh0 , and
the ratio of the vacuum expectation values of the two Higgs doublets, tan , remain as free
parameters of the model. The restrictions on mH , and tan have been already discussed
in Section 1. For the masses of the neutral Higgs bosons, the lower limits are given as
mH 0 115 GeV, mh0 89.9 GeV and mA0 90.1 GeV in [53].
Before we present our results, a small note about the calculations of the long-distance
effects is in order. There are five possible resonances in the cc system that can contribute
to the decays under consideration and to calculate them, we need to divide the integration
region for s into two parts so that we have
4m2
m2B
s
(m2 0.02)2
m2B
(m2 + 0.02)2
m2B
s
(mB mM )2
,
m2B
(28)
296
where m2 = 3.686 GeV is the mass of the second resonance, and M = K, K .

In the following, we give results of our calculations about the dependencies of the
differential branching ratio dBR/ds, forward-backward asymmetry AFB (s), CP violating
asymmetry ACP (s), CP asymmetry in the forward-backward asymmetry ACP (AFB )(s) and
finally the components of the lepton polarization asymmetries, PL (s), PT (s) and PN (s),
of the B K + and B K + decays on the invariant dilepton mass s. In order
to investigate the dependencies of the above physical quantities on the model parameters,
namely CP violating phase and tan , we eliminate the other parameter s by performing
the s integrations over the allowed kinematical region (Eq. (28)) so as to obtain their
averaged values, AFB , ACP , ACP (AFB ) , PL , PT and PN .
Numerical results are shown in Figs. 126 and we have the following line conventions:
dot lines, dashed-dot lines and solid lines represent the model IV contributions with
tan = 10, 30, 50, respectively, and the dashed lines are for the SM predictions. The cases
of switching off NHB contributions i.e., setting CQi = 0, almost coincide with the cases of
2HDM contributions with tan = 10, therefore we did not plot them separately.
In Fig. 1, we give the dependence of the dBR/ds on s for B K + . From this
figure NHB effects are very obviously seen, especially in the high-s region.
In Figs. 2 and 3, AFB (s) and AFB of B K + as a function of s and CP violating
phase are presented. Since AFB arises in the 2HDM only when NHB effects are taken into
account, it provides a good probe to test these effects. We see that AFB is quite sensitive
to tan and it is negative for all values of and s except in the region. AFB in
B K + is between (0.04, 0.01), which is non-zero but hard to observe.
Figs. 4 and 5 show the dependence of ACP (s) on s and ACP on for B K +
decay. We see that ACP (s) is quite sensitive to tan and its sign does not change in the
allowed values of s except in the resonance mass region of when tan = 50. It follows
from Fig. 5 that ACP is also sensitive to , and it varies in the range (0.8, 0.8) 102 ,
which may provide an indication for the existence of new physics since ACP is zero in
the SM.
ACP (AFB )(s) and ACP (AFB ) of B K+ as a function of s and CP violating
phase are presented in Fig. 6 and 7, respectively. We note that in both of these figures,
predictions for the different values of tan completely coincide which indicates that
ACP (AFB ) is not sensitive to this parameter in B K + decay. As seen from Fig. 7,
ACP (AFB ) strongly depends on CP violating phase and it can reach about 6% for some
values of .
In Figs. 810, we present the s dependence of the longitudinal PL , transverse PT and
normal PN polarizations of the final lepton for B K + decay. We see that except the
region, PN is negative for all values of s, but PL and PT change sign with the different
choices of the values of tan . The effects of NHB exchanges are also very obvious. In
Figs. 1113, dependence of the averaged values of the longitudinal PL , transverse PT
and normal PN polarizations of the final lepton for B K+ decay on are shown.
It is obvious from these figures that PL (PN ) is weakly (strongly) sensitive to while
PT is totaly insensitive to . We also note that PN is zero in the SM and it is at the
order of 1% in model IV for tan = 30. Thus, measurement of this component in future
experiments may provide information about the model IV parameters.
297
Fig. 1. The dependence of the dBR/ds on s for B K + decay. Here dot lines, dashed-dot lines and solid
lines represent the model IV contributions with tan = 10, 30, 50, respectively, and the dashed lines are for the
SM predictions.
Fig. 2. The dependence of AFB (s)(B K + ) on s. Here dot lines, dashed-dot lines and solid lines represent
the model IV contributions with tan = 10, 30, 50, respectively, and the dashed lines are for the SM predictions.
Fig. 3. The dependence of AFB (B K + ) on . Here dot lines, dashed-dot lines and solid lines represent
298
Fig. 4. The same as Fig. 2, but for ACP (s)(B K + ).
Fig. 5. The same as Fig. 3, but for ACP (B K + ).
Fig. 6. The same as Fig. 2, but for ACP (AFB )(s)(B K + ).
299
Fig. 7. The same as Fig. 3, but for ACP (AFB ) (B K + ).
Fig. 8. The dependence of PL (s)(B K + ) on s. Here dot lines, dashed-dot lines and solid lines represent
Fig. 9. The same as Fig. 8, but for PT (s)(B K + ).
300
Fig. 10. The same as Fig. 8, but for PN (s)(B K + ).
Fig. 11. The dependence of PL (B K + ) on . Here dot lines, dashed-dot lines and solid lines represent
Fig. 12. The same as Fig. 11, but for PT (B K + ).
301
Fig. 13. The same as Fig. 11, but for PN (B K + ).
Fig. 14. The dependence of the dBR/ds on s for B K + decay. Here dot lines, dashed-dot lines and solid
lines represent the model IV contributions with tan = 10, 30, 50, respectively, and the dashed lines are for the
SM predictions.
Figs. 1426 are devoted to the B K + decay. In Fig. 14, dependence of the
dBR/ds on s is given. We see that dBR/ds of this process is not as sensitive to the effects
of NHB exchanges as B K + decay and these effects begin to be significant when
tan 40.
In Fig. 15 and 16, AFB (s) and AFB of B K + as a function of s and CP
violating phase are presented. As in B K + decay, AFB here is also quite sensitive
to tan , and its magnitude gets smaller than the SM prediction with the increasing values
of tan . As seen from Fig. 16, AFB in B K + is of the order of 10% and strongly
dependent on , especially when tan = 50.
Figs. 17 and 18 show the dependence of ACP (s) on s and ACP on for B K +
decay. We see that ACP (s) is quite sensitive to tan and and it does not change sign in
the allowed values of s. It follows from Fig. 18 that ACP is of the order of 0.1% and hard
to observe.
302
Fig. 15. The dependence of AFB (s)(B K + ) on s. Here dot lines, dashed-dot lines and solid lines
represent the model IV contributions with tan = 10, 30, 50, respectively, and the dashed lines are for the SM
predictions.
Fig. 16. The dependence of AFB (B K + ) on . Here dot lines, dashed-dot lines and solid lines represent
Fig. 17. The same as Fig. 15, but for ACP (s)(B K + ).
303
Fig. 18. The same as Fig. 16, but for ACP (B K + ).
Fig. 19. The same as Fig. 15, but for ACP (AFB )(s)(B K + ).
ACP (AFB )(s) and ACP (AFB ) of B K + as a function of s and CP violating

phase are presented in Figs. 19 and 20, respectively. We see that ACP (AFB ) comes mainly
from exchanging NHBs and its magnitude can reach 0.3 exhibiting a strong dependence
on the CP-violating phase .
In Figs. 2123, we present the s dependence of the longitudinal PL , transverse PT
and normal PN polarizations of the final lepton for B K + decay. We see that
NHB exchanges modify the spectrums of PT and PN greatly while its effect is relatively
weak for PL . We also observe that except the region, PL is negative for all values of s,
but PT and PN change sign with the different choices of the values of tan . In Figs. 2426,
dependence of the averaged values of the longitudinal PL , transverse PT and normal
PN polarizations of the final lepton for B K + decay on are depicted. It is
obvious from these figures that PL , PT and PN in model IV are larger as absolute
values than the corresponding SM predictions. Sensitivity of these observables to the
parameter is significant when tan is not smaller than 30.
304
Fig. 20. The same as Fig. 16, but for ACP (AFB ) (B K + ).
Fig. 21. The dependence of PL (s)(B K + ) on s. Here dot lines, dashed-dot lines and solid lines represent
Fig. 22. The same as Fig. 21, but for PT (s)(B K + ).
305
Fig. 23. The same as Fig. 21, but for PN (s)(B K + ).
Fig. 24. The dependence of PL (B K + ) on . Here dot lines, dashed-dot lines and solid lines represent
Fig. 25. The same as Fig. 24, but for PT (B K + ).
306
Fig. 26. The same as Fig. 24, but for PN (B K + ).
We now summarize our results:

We observe an enhancement in the differential branching ratio for both B K +
and B K + processes in model IV compared to the SM when the NHB effects
are taken into account. The NHB effects are more manifest in B K + decay
with respect to B K + decay.
AFB comes only from NHB contributions in B K + , and its average is between
(0.04, 0.01), which is non-zero but hard to observe. However for B K +
decay, it is of the order of 10%, which should be within the luminosity reach of coming
B factories.
ACP is between (0.8, 0.8) 102 and (0.3, 0.3) 102 in B K + and
B K + decays, respectively. Since ACP for these decays is practically zero in
the SM, a non-zero value measured in future experiments for ACP will be a definite
indication of the existence of new physics.
ACP (AFB ) is at the order of 1% for B K + decay and it is very sensitive to the
CP violating phase , but not to tan . As for B K + decay, it comes mainly
from exchanging NHBs, and can be as large as 30% for some values of .
Model IV contributions modify the spectrums of PL , PT and PN greatly compared to
the SM case for both decays. These quantities are sensitive to the NHB effect and also
the CP violating phase , except the PT component for B K + decay.
Therefore, the experimental investigation of AFB , ACP , ACP (AFB ) and the polarization
components in B K+ and B K + decays may be quite suitable for testing
the new physics effects beyond the SM.
Acknowledgement
We would like to thank Shou Hua Zhu for his comments about the previous version of
this work.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
B. Grinstein, M.J. Savage, M.B. Wise, Nucl. Phys. B 319 (1989) 271.
N.G. Deshpande, X.-G. He, J. Trampetic, Phys. Lett. B 367 (1996) 362.
Y.B. Dai, C.S. Huang, H.W. Huang, Phys. Lett. B 390 (1997) 257.
C.S. Huang, Q.S. Yan, Phys. Lett. B 442 (1998) 209;
C.S. Huang, W. Liao, Q.S. Yan, Phys. Rev. D 59 (1999) 011701.
S. Fukae, C.S. Kim, T. Yoshikawa, hep-ph/9908229.
S. Fukae, C.S. Kim, T. Morozumi, T. Yoshikawa, Phys. Rev. D 59 (1999) 074013.
Y.G. Kim, P. Ko, J.S. Lee, Nucl. Phys. B 544 (1999) 64.
T. Goto et al., Phys. Rev. D 55 (1997) 4273;
T. Goto, Y. Okada, Y. Shimizu, Phys. Rev. D 58 (1998) 094006.
E. Lunghi et al., Nucl. Phys. B 568 (2000) 120, and references therein.
J.L. Hewett, Phys. Rev. D 53 (1996) 4964.
Y. Grossman, Z. Ligeti, E. Nardi, Phys. Rev. D 55 (1997) 2768.
K. Krger, L.M. Sehgal, Phys. Lett. B 380 (1996) 199.
I.I. Bigi, M. Shifman, N.G. Vraltsev, A.I. Vainstein, Phys. Rev. Lett. 71 (1993) 496;
B. Blok, L. Kozrakh, M. Shifman, A.I. Vainstein, Phys. Rev. D 49 (1994) 3356;
A.V. Manohar, M.B. Wise, Phys. Rev. D 49 (1994) 1310;
S. Balk, T.G. Krner, D. Pirjol, K. Schilcher, Z. Phys. C 64 (1994) 37;
A.F. Falk, Z. Ligeti, M. Neubert, Y. Nir, Phys. Lett. B 326 (1994) 145.
R. Casalbuoni, A. Deandra, N. Di Bartolemo, R. Gatto, G. Nardulli, Phys. Lett. B 312 (1993) 315.
P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 53 (1996) 3672;
P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 57 (1998) 3186, Erratum.
W. Jaus, D. Wyler, Phys. Rev. D 41 (1990) 3405.
W. Roberts, Phys. Rev. D 54 (1996) 863.
T.M. Aliev, A. zpineci, M. Savci, Phys. Rev. D 56 (1997) 4260.
P. Ball, V.M. Braun, Phys. Rev. D 58 (1998) 094016.
CDF Collaboration, T. Affolder et al., Phys. Rev. Lett. 83 (1999) 3378.
Belle Collaboration, K. Abe et al., Phys. Rev. Lett. 88 (2002) 021801.
N.G. Deshpande, J. Trampetic, Phys. Rev. Lett. 60 (1988) 2583.
P.J. ODonnell, H.K. Tung, Phys. Rev. D 43 (1991) 2067.
D.S. Du, C. Liu, Phys. Lett. B 317 (1993) 179.
G. Burdman, Phys. Rev. D 52 (1995) 6400;
W. Roberts, Phys. Rev. D 54 (1996) 863.
D.S. Liu, Phys. Rev. D 52 (1995) 5056.
C. Greub, A. Ioannissian, D. Wyler, Phys. Lett. B 346 (1995) 149.
C.Q. Geng, C.P. Kao, Phys. Rev. D 54 (1996) 5636.
D. Melikhov, N. Nikitin, S. Simula, Phys. Lett. B 442 (1998) 381.
T.M. Aliev, M. Savci, A. zpineci, H. Koru, J. Phys. G 24 (1998) 49.
T.M. Aliev, E.O. Iltan, Phys. Lett. B 451 (1998) 175.
A. Ali, P. Ball, L.T. Handoko, G. Hiller, Phys. Rev. D 61 (2000) 074024.
F. Krger, J.C. Romo, Phys. Rev. D 62 (2000) 034020.
T.M. Aliev, C.S. Kim, Y.G. Kim, Phys. Rev. D 62 (2000) 014026.
T.M. Aliev, M. Savci, Phys. Lett. B 481 (2000) 275.
T.M. Aliev, A. zpineci, M. Savci, Phys. Lett. B 511 (2001) 49.
E.O. Iltan, hep-ph/0102061.
D. Guetta, E. Nardi, Phys. Rev. D 58 (1998) 012001.
C.-S. Huang, S.H. Zhu, Phys. Rev. D 61 (1999) 015011;
C.-S. Huang, S.H. Zhu, Phys. Rev. D 61 (2000) 119903, Erratum.
D. Buskulic et al., ALEPH Collaboration, Phys. Lett. B 343 (1995) 444;
J. Kalinowski, Phys. Lett. B 245 (1990) 201;
A.K. Grant, Phys. Rev. D 51 (1995) 207.
T.M. Aliev, M.K. Cakmak, M. Savci, Phys. Rev. D 64 (2001) 055007.
307
308
[42] T.M. Aliev, M.K. Cakmak, M. Savci, Nucl. Phys. B 607 (2001) 305.
[43] Q.-S. Yan, C.-S. Huang, W. Liao, S.H. Zhu, Phys. Rev. D 62 (2000) 094023.
[44] Y.B. Dai, C.S. Huang, H.W. Huang, Phys. Lett. B 390 (1997) 257;
Y.B. Dai, C.S. Huang, H.W. Huang, Phys. Lett. B 513 (2001) 429, Erratum;
C.S. Huang, L. Wei, Q.S. Yan, Phys. Rev. D 59 (1999) 011701.
[45] C.S. Huang, L. Wei, Q.S. Yan, S.H. Zhu, Phys. Rev. D 63 (2001) 114021.
[46] H.E. Logan, U. Nierste, Nucl. Phys. B 586 (2000) 39.
[47] C. Bobeth, T. Ewerth, F. Krger, J. Urban, Phys. Rev. D 64 (2001) 074014.
[48] Z. Xiong, J.M. Yang, hep-ph/0105260.
[49] C.S. Huang, L. Wei, Q.S. Yan, S.H. Zhu, Phys. Rev. D 64 (2001) 059902.
[50] B. Grinstein, R. Springer, M.B. Wise, Nucl. Phys. B 339 (1990) 269;
R. Grigjanis, P.J. ODonnell, M. Sutherland, H. Navelet, Phys. Lett. B 213 (1988) 355;
R. Grigjanis, P.J. ODonnell, M. Sutherland, H. Navelet, Phys. Lett. B 286 (1992) 413;
G. Cella, G. Curci, G. Ricciardi, A. Vicer, Phys. Lett. B 325 (1994) 227;
G. Cella, G. Curci, G. Ricciardi, A. Vicer, Nucl. Phys. B 431 (1994) 417.
[51] A.J. Buras, M. Mnz, Phys. Rev. D 52 (1995) 186;
M. Misiak, Nucl. Phys. B 393 (1993) 23;
M. Misiak, Nucl. Phys. B 439 (1995) 461, Erratum.
[52] A. Ali, T. Mannel, T. Morozumi, Phys. Lett. B 273 (1991) 505.
[53] A. Heister et al., ALEPH Collaboration, CERN-EP/2001-095;
L3 Collaboration, Phys. Lett. B 503 (2001) 21.

(1 + 1)-dimensional baryons from the SU(N)

color-flavor transformation
J. Budczies a , S. Nonnenmacher b , Ya. Shnir a , M.R. Zirnbauer a
a Institut fr Theoretische Physik, Universitt zu Kln, 50937 Kln, Germany
b Service de Physique Thorique, CEA-Saclay, 91191 Gif-sur-Yvette Cedex, France
Received 20 December 2001; accepted 23 April 2002
Abstract
The color-flavor transformation, an identity that connects two integrals, each of which is over one
of a dual pair of Lie groups acting in the fermionic Fock space, is extended to the case of the special
unitary group. Using this extension, a toy model of lattice QCD is studied: Nf species of spinless
fermions interacting with strongly coupled SU(Nc ) lattice gauge fields in 1 + 1 dimensions. The
color-flavor transformed theory is expressed in terms of gauge singlets, the meson fields, organized
into sectors distinguished by the distribution of baryonic flux. A comprehensive analytical and
numerical search is made for saddle-point configurations of the meson fields, with various topological
charges, in the vacuum and single-baryon sectors. Two definitions of the static baryon on the square
lattice, straight and zigzag, are investigated. The masses of the baryonic states are estimated using
the saddle-point approximation for large Nc . 2002 Elsevier Science B.V. All rights reserved.
PACS: 02.20.-a; 11.15.Ha; 11.15.Tk; 12.38.Lg; 11.30.Rd
Keywords: Lattice gauge theory; Low-energy effective theory; Chiral symmetry; Non-perturbative calculations;
Group theory
1. Introduction
In quantum chromodynamics (QCD), a hierarchy of scales is provided by 1 GeV,
the scale of chiral symmetry breaking, and QCD 0.18 GeV, defined as the location of
the Landau pole of the one-loop beta function. The running coupling constant increases
from weak to strong coupling as the momentum scale is lowered from the perturbative
regime above down to QCD .
E-mail address: shnir@thp.uni-koeln.de (Ya. Shnir).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 1 - 0
310
J. Budczies et al. / Nuclear Physics B 635 (2002) 309356
In the past two decades a great deal was learned about the non-perturbative structure
of QCD at scales between and QCD . The guiding idea was to construct low-energy
effective theories which encode the symmetries of the fundamental QCD Lagrangian. To
obtain these effective theories, one may start from full QCD, and integrate out the highenergy degrees of freedom (quarks and gluons) in order to produce a low-energy effective
action in terms of mesons and baryons. In this way it was possible to recover the chiral
Lagrangian [14] that had been introduced phenomenologically by Weinberg [5].
In a more recent development, it was shown [6] how to extract the effective longdistance degrees of freedom by starting from the lattice [7,8] formulation of QCD. In
that approach it is assumed that the long-distance physics of lattice QCD (LQCD) can
be described by a strongly coupled lattice theory. From the latter, one gets the continuum
chiral Lagrangian by expanding the effective action in powers of the lattice spacing and
external momenta. All the terms of the GasserLeutwyler continuum effective Lagrangian
[9] can be recovered in this way [10]. The lattice formulation is, however, deficient in
one respect: by the technical difficulties with chiral symmetry for lattice fermions, the
chiral anomaly is lost, i.e., for Nf massless quark flavors the chiral symmetry of the lattice
effective theory is U(Nf ) rather than SU(Nf ).
This type of approach was initiated in [11,12]; it relied on a bosonization of the
strong-coupling LQCD action, and a large-Nc or large-dimension expansion. Technically,
the heart of the method is the computation of integrals over the group SU(Nc ) with Haar
measure, weighted by eS(U ) . Some general results for such integrals have recently been
reviewed in [13].
A few years ago, an alternative kind of bosonization scheme was introduced [14],
relying on a mathematical formalism later called the color-flavor transformation [15].
This transformation relates two different formulations of a certain class of theories. In
condensed matter theory, the transformation has found a number of applications, among
others to the random flux model [16].
The color-flavor transformation in its original version applies to the gauge group U(Nc ).
For this group, all gauge singlets are of mesonic (or quarkantiquark) type. In order
for baryons to appear, one needs to replace U(Nc ) by the special unitary group SU(Nc ).
In Section 2 of the present paper we extend the color-flavor transformation to SU(Nc ),
by decomposing the (colorless) flavor sector of Fock space into disconnected subsectors
labeled by the baryonic charge.
In Sections 47 we apply the formalism to a toy model of LQCD: Nf species of
spinless fermions interacting with strongly coupled SU(Nc ) lattice gauge fields in 1 + 1
dimensions. The color-flavor transformation yields a dual representation of this nonAbelian model. Combining numerical computations with analytical considerations, we
conduct a comprehensive search for saddle-point configurations in various baryonic sectors
with different topological properties. We use these configurations (without fluctuation
corrections) to estimate the mass of a single baryon in our model. In doing so we ignore
the MerminWagnerColeman theorem (asserting that spontaneous breaking of continuous
global symmetries does not occur in 1 + 1 dimensions), by assuming the pattern of chiral
symmetry breaking that is known to occur in the physical case of 3 + 1 dimensions.
After the present work had been completed, we learned that the SU(N) generalization
of the color-flavor transformation has also been worked out by Schlittgen and Wettig [17].
311
2. Color-flavor transformation for SU(Nc )

2.1. Group action on fermionic Fock space
In this section we set up some algebraic structures, which are needed to establish the
color-flavor transformation for the special unitary group. Our discussion follows the line
of reasoning of Ref. [14] but is somewhat simpler, as we do not need the superalgebraic
framework employed there.
We start by considering a set of fermionic creation and annihilation operators fAi and
fAi , which obey the canonical anticommutation relations
i j
i j
i j
fA , fB = 0,
fA , fB = 0,
fA , fB = AB ij .
The lower index takes the values +a or a, with range a = 1, . . . , Nf , and the upper
index takes the values i = 1, . . . , Nc . Having QCD in mind, we interpret the operators
i and fi as creation operators for quarks and antiquarks, respectively; the index i
f+a
a
corresponds to the gauge (or color) degrees of freedom and the index a labels the different
quark flavors. (The quarks are regarded here as being spinless.) The operators fAi and fAi
act on a Fock space with vacuum |0 and its conjugate 0|, by fAi |0 = 0 and 0|fAi = 0 for
all A and i.
ij
We next consider the set of quadratic operators EAB defined by
i f ,
E+a,+b = f+a
+b
ij
i f ,
E+a,b = f+a
b
ij
ij
i j
fb .
Ea,b = fa
i
Ea,+b = fa
f+b ,
ij
The C-linear span of these operators has the structure of a complex Lie algebra, G.
ij
More precisely, the operators EAB obey the commutation relations of a set of canonical
generators of the Lie algebra gl(2Nf Nc ):
ij

kj
kl
il
EAB , ECD
= j k BC EAD
li DA ECB .
Thus we have a Lie algebra isomorphism from gl(2Nf Nc ) (i.e., the space of complex
matrices of size 2Nf Nc 2Nf Nc , with the Lie bracket given by the commutator) to G:
ij ij
m tm :=
mAB EAB .
t : gl(2Nf Nc ) G,
ij,AB
This isomorphism lifts to an isomorphism of the corresponding complex groups:

T : GL(2Nf Nc ) G,
M = exp(m) TM = exp(tm ),
(1)
which forms a (reducible) representation of GL(2Nf Nc ) on Fock space. The representation

is single-valued (which means there are no U(1) obstructions from the multi-valuedness of
ii is the set {0, 1}.
the logarithm) as the spectrum of each operator EAA
The Lie algebra gl(2Nf Nc ) has two subalgebras gl(Nc ) and gl(2Nf ) which are
embedded in a natural way: a matrix X gl(Nc ) is identified with I2Nf X, and a
matrix Y gl(2Nf ) with Y INc . Through these embeddings, gl(Nc ) and gl(2Nf ) form a
pair of maximal commuting subalgebras of gl(2Nf Nc ), also known as a dual pair [18].
312
The subgroups GL(Nc ) and GL(2Nf ) are embedded into GL(2Nf Nc ) in the same way.
Their adjoint action on the fermionic creation and annihilation operators is described in
Appendix A.
We define the color group to be the subgroup SU(Nc ) of GL(Nc ), and the flavor group
to be the subgroup U(2Nf ) of GL(2Nf ). GL(Nc ) contains an extra U(1) subgroup which
lies outside the color group and, being generated by the unit matrix, commutes with the
+ Nf where
whole group GL(2Nf Nc ). This U(1) is generated by Q

1 i i
ii
i
i
= 1
Q
(2)
EAA
Nf =
fa
f+a f+a fa
Nc
Nc
A,i
a,i
counts the difference between the number of particles and the number of antiparticles:
= 1 (N+ N ). In contrast, the operator giving the total number of particles,
Q
Nc

i f i + fi f i ), does not commute with the generators of gl(N ). We will
= a,i (f+a
N
f
+a
a a

call Q the baryon charge operator.
2.2. From color group integrals to flavor group integrals
Let Ai and Ai be two independent sets of Grassmann variables, referred to as quark
fields, and consider the color group integral

i
j
i ij j
=
U b .
dU exp +a
U ij +a + b
Z(, )
(3)
SU(Nc )

The Haar measure dU of SU(Nc ) is understood to be normalized by SU(Nc ) dU = 1. We
also adopt the convention that repeated occurrence of an index implies summation.
The color-flavor transformation will replace the integral (3) by an integral over the flavor
as the matrix
group U(2Nf ). A key step in doing the transformation is to interpret Z(, )
element of an operator P that projects on the colorless sector (or flavor sector) of Fock
space. This sector is the subspace of all states |flavor which are invariant under the color
group: TU |flavor = |flavor for all U SU(Nc ).
The first step towards the color-flavor transformation is to express the projector P as
dU TU .
P=
(4)
SU(Nc )
Let us now introduce the fermion coherent states

i i

i i

i
i
i
i
| = 0| exp a

,
| = exp fa
|0.
fa + +a
f+a
a + f+a
+a
(5)
By making use of the first set of relations in Appendix A, it is straightforward to show that

i
j
i ij j
|TU | = exp +a

U ij +a + b
U b
for U SU(Nc ). This yields the simple formula
= |P| .
Z(, )
(6)
as an integral over the flavor group, we will derive an alternative
To express Z(, )
representation of the projector P, as an integral over coherent states of the flavor sector.
313
2.3. The flavor sector

The subspace of states in Fock space which are invariant under U(Nc ) was described in
[14]. It consists of the vacuum and of mesonic excitations on top of it. The prototype of
such an excitation is the one-meson state
|mab =
ii
E+a,b
|0 =
i i
f+a
fb |0.
ii
(where we have gone back to
By the multiple action of the gl(2Nf ) generators E+a,b
using the summation convention), one can build states containing up to Nc Nf mesons, with
different flavors. These states are automatically U(Nc )-invariant; conversely, all U(Nc )invariant states are linear combinations of such multi-meson states. The group U(2Nf )
acts irreducibly on this invariant subspace.
The set of SU(Nc )-invariant states is larger. To obtain it, one relaxes the constraint

Q|
= 0. Thus there exist colorless sectors of Fock space on which the central
takes a non-zero value. These sectors contain the baryons, which are totally
generator Q
antisymmetric combinations of Nc quarks. A baryon with flavors a1 , . . . , aNc is defined as
|bA1 ...ANc =
1
i
i ...i fi1 fANNc |0,
c
Nc ! 1 Nc A1
(7)
where the Ak = ak are taken either all positive (baryon), or all negative (antibaryon).
A matrix g GL(Nc ) acts on this state simply by multiplication with Det(g) (respectively, Det1 (g)). Therefore, the state is invariant under the color group SU(Nc ).
The above baryon (respectively, antibaryon) is an eigenstate of the baryon charge
with eigenvalue +1 (respectively, 1). Acting on it with the generators E ii
operator Q
AB
of the flavor algebra gl(2Nf ), one builds other colorless states with the same baryon
number, which form an irreducible subspace for U(2Nf ): the one-baryon (respectively,
one-antibaryon) sector.
The one-baryon sector can be generated from the state (7) with all Aj = 1. One can
similarly build Q-baryon (respectively, Q-antibaryon) states from
|BQ =
Q

a=1
Nc
1
f+a
|0,
f+a
|B0 = |0,
|BQ =
Nc
1
fa
|0.
fa
(8)
a=1
The values of the baryon charge range from Nf to Nf , according to Paulis exclusion
principle. As with Q = 1, acting on |BQ with the algebra gl(2Nf ) builds the full Qbaryon part of the flavor sector, so the group U(2Nf ) acts irreducibly on this part. This
can be proved by using the dual-pair property of the subalgebras gl(2Nf ) and gl(Nc ), as
exposed in [18].
To summarize, the flavor sector of Fock space decomposes into 2Nf + 1 subsectors,
characterized by their baryon charges Q. Each sector carries an irreducible unitary
representation of the flavor group U(2Nf ).
314
2.4. Coherent states

Having decomposed the flavor sector as described above, we can now express the
projector P in a different way. For this purpose we will use coherent states, in the spirit
of Perelomov [19]. On each subsector with a fixed baryon charge Q, we consider the
generalized coherent states built by the action of G U(2Nf ) on the reference state |BQ ,
i.e., the states
g G,
def
Q = Nf , . . . , Nf :
|gQ = Tg |BQ ,
def
gQ | = BQ |Tg .
(9)
The crucial property of coherent states we will now use, is that they supply a resolution of
unity. Because of the irreducibility of the U(2Nf ) action on each Q-subsector, the operator
def
PQ = Q dg |gQ gQ |
(10)
G
coincides with the orthogonal projector on that subsector, the only provision being that
the normalization constant Q be chosen appropriately. Indeed, the operator PQ trivially
commutes with every element of the flavor group; Schurs lemma then ensures that it
is proportional to the identity on each irreducible space of this group, therefore, on
each subsector with fixed baryonic charge. Owing to orthogonality, PQ vanishes on all
subsectors with Q = Q, whereas it is the identity on the Q-subsector if we take
1

Q =
dg |BQ |Tg |BQ |2
(11)
Some particular values of the constant (namely 0 , 1 ) are computed in Appendix B.

For the matrix element (6) of the projector P on the full flavor sector,
P=
Nf

PQ ,
Q=Nf
we now have a new representation:

=
Z(, )
Nf

Q=Nf
|gQ gQ | .
dg
(12)
|gQ and gQ | , it is convenient to use a Gauss decomposiTo compute the overlaps
B
G can be factored as
tion of G = U(2Nf ): almost any matrix g = A
CD

A B
1 Z
A 0
1 0
=
(13)
1 ,
C D
0 1
0 D
Z
= D 1 C, and A = A BD 1 C hold. The decomposiwhere the relations Z = BD 1 , Z
tion becomes singular if D does, but this happens only on a submanifold of codimension
= D Z A 1 and
one (and, hence, measure zero) of G. The unitarity of g implies Z
315
allows to write the central matrix in the form

(1 + ZZ )1/2
U 0
A 0
0
.
=
(14)
0 V
0 D
0
(1 + Z Z)1/2

The matrices U and V are unitary, so U0 V0 is an element of the diagonal U(Nf ) U(Nf )
subgroup of G, which we call H . It can thus be shown that the elements g of an open dense
subset of G are in one-to-one correspondence with the triplets (Z, U, V), where the pair
diag(U, V) is an element of H , while Z represents a point in the coset space G/H and can
be any complex Nf Nf matrix. Moreover, the Haar measure dg of G factorizes as

(15)
dg =
d(gH ) dh =
d Z, Z
dU dV.
G
G/H
Nf Nf
Both dU and dV are normalized Haar measures on U(Nf ), and

2Nf

ij
d Z, Z = CNf Det 1 + ZZ
dZij d Z
i,j
is the normalized invariant measure on G/H . The normalization factor CNf is computed
in Appendix B; see Eq. (B.6).
We now explain how to use this decomposition to compute the overlaps. The Gauss
decomposition (13) carries over to any representation of G, so for every g G we can
write the operator Tg as
Tg = T Tdiag(A,D)
(16)
T ,

1 0
where = 10 Z1 and = Z
. According to the relations given in Appendix A, the factors
1
T and T act trivially on the reference states:
Q = Nf , . . . , Nf :
T |BQ = |BQ ,
BQ |T = BQ |.
(17)
The action of the block-diagonal operator is slightly more subtle. Using the third set of
relations given in Appendix A, we get
|0 = (Det D)Nc |0,
Tdiag(A,D)
Tdiag(A,D)
|B1 = (Det D)Nc
Nc

i
|0,
A a1 f+a
i=1
Tdiag(A,D)
|B1 = (Det D)Nc
Nc

1
i
|0.
D 1a fa
i=1
(To make sense of these formulas one must remember that we are using the summation
convention: the flavor index a under the product is understood to be summed over.) These
|:
formulas directly yield the desired overlaps with
|g0 = (Det D)Nc

Nc

i=1

i
i
,
exp +a
Zab b
316
|g1 = (Det D)Nc

Nc

i
i
i
,
Ac1 exp +a
Zab b
+c
i=1
Nc

|g1 = (Det D)Nc

i

1 i
i
c exp +a
D1c
Zab b
,
i=1
as well as the overlaps with | :

Nc
N

i
i
,
g0 | = Det D c
exp a
Z ab +b
i=1
Nc

i
N

i
i
g1 | = Det D c
exp a
Z ab +b
,
A 1c +c
i=1
Nc
N
1

i
i
i
D c1 exp a
.
g1 | = Det D c
c
Z ab +b
i=1
The overlaps with the coherent states |gQ containing more than one baryon (|Q| > 1) can
be computed in the same way; in front of the exponential factors, there will be |Q| similar
products, with flavor indices 1, . . . , |Q|.
We now insert the above expressions for the overlaps into (12), and use the factorization
(15) to arrive at an integral over triples (Z, U, V). Leaving the Z-integral for later, we next
carry out the integrations over the unitary matrices U and V. They enter in the overlaps via
the matrix elements of A and D; see Eq. (14). To simplify the notation, we first perform a
flavor rotation on the Grassmann fields:
i
i
i
i
+b
= ( 1 + ZZ )ba +a
,
b
= a
( 1 + Z Z )ab ,

i =
i ( 1 + ZZ ) ,
i =
i .
+a
b
1 + Z Z ba a
+b
ab
The integrals we need to compute then read as follows (assuming Q > 0):
Q ( + , + ) = Q
def
dU
U(Nf )
def
Q ( , ) = Q
Q
Nc

(18)
c=1 i=1
U(Nf )
i 1
i
+a
Uac +b
Ucb ,
dV
Q
Nc

i

i
1
a
b Vbc .
Vca
(19)
c=1 i=1
; Z) Q ( + , + ), and Q (,
; Z) Q ( ,
We also set 0 0 , and Q (,
). The function 1 ( + , + ) will play a distinguished role in the lattice gauge theory
application in Section 3, and we therefore evaluate it explicitly in the next subsection.
The integrations over H having been done, we are left with an integral over G/H , i.e.,
over a Z-dependent integrand, in each Q-subsector. Putting everything together, we finally
arrive at the following identity:
317
i

j
i ij j
dU exp +a
U ij +a + b
U b
SU(Nc )
Nf

Q=Nf
; Z)
d Z, Z Q (,
j
j
i Z
i
exp( +a
ab b + b Z ba +a )
Det(1 + ZZ )Nc
N N
C f f
(20)
which is called the color-flavor transformation for SU(Nc ), and is the central result of the
present section. Note that the right-hand side of the transformation has the attractive feature
of organizing the contributions according to the different baryonic sectors.
An effective action in the bosonic variable Z can be obtained by doing the (Gaussian)
integral over the Grassmann fields. This will be done in a lattice gauge context in Section 3.
2.5. Evaluation of 1
In this subsection we evaluate the coefficient
1 ( + , + ) = 1
dU
U(Nf )
Nc

i
i

b1 .
U
+a Ua1 +b
i=1
Only the first column of the unitary matrix U occurs in the integrand, so the integral
is effectively over a unit sphere in Nf -dimensional complex space, S2Nf 1 = CNf /R+ .
Parametrizing the latter by a complex vector z = (z1 , . . . , zNf ) with unit norm |z| = 1, we
have

c i
i
) N
i=1 (+a za )(+b z b )
|z|=1 d(z, z

,
1 (+ , + ) = 1
)
|z|=1 d(z, z
where d(z, z ) is a U(Nf )-invariant measure on the unit sphere |z| = 1. By homogeneity
in z and z , we may use the trick of replacing the numerator and denominator by integrals
2
over CNf , with a Gaussian weight function e|z| included in the integrands. The answer
then easily follows from Wicks theorem:
1 ( + , + ) = 1
Nc

(Nf 1)!
(i)
i
sgn
+a
+a
(Nc + Nf 1)!
SNc
= 1
i=1
Nc

(Nf 1)!
(i)
i
1 + ZZ ab +b
sgn
,
+a
(Nc + Nf 1)!
SNc
(21)
i=1
where SNc denotes the group of permutations of the numbers 1, . . . , Nc .

This result was already obtained in [22]. More general considerations based on the
group theoretical approach were presented in [17]. Our calculation of the pre-exponential
factors 0 , 1 are in agreement with the latter work where a general formula for Q was
derived.
318
3. Color-flavor transformation on the lattice

We consider a Euclidean SU(Nc ) gauge theory in 1 + d dimensions placed on a
hypercubic lattice with lattice constant a. The fermions bi (n), with colors i = 1, . . . , Nc
and flavors b = 1, . . . , Nf , are put on lattice sites labeled by n = (n0 , . . . , nd ), while the
gauge matrix variables U (n + 2 ) = exp[iagA (na + a2 )] SU(Nc ) are placed on lattice
links n + /2
(we label links by their middle points), starting from sites n in any of the
directions = 0, . .. ,
d. In the limit of a strong gauge coupling g, the gauge theory has the
partition sum Z =
n d(n) d (n) Z(, ) with [11]

(n)
m,,
exp SU,, n +
.
Z(, ) =
(22)
e
dU n +
2
2
n
SU(Nc )
The fermions on two neighboring sites n and n + are coupled through the gauge fields
on the connecting link n + /2
in a gauge-invariant way:

ad i
j
SU,, n +
=
b (n)U ij n +
b (n + )
2
2
2

j
j i
i
b (n) ,
n+
b (n + )U
(23)
2
while the (bare) quark mass m couples the fermions diagonally
Sm,, (n) = a d+1 m(n)(n).
(24)
We are not going to worry here about the fermion doubling problem and will restrict our
considerations to this naive discretization of the fermionic action. Also, for simplicity we
do not take into account the spin degrees of freedom, leaving their inclusion for a future
publication.
We rescale the fermionic fields so as to absorb the prefactor a d /2. This just adds a global
prefactor to Z, and has no effect on the physical quantities. The SU(Nc )-integral over U
on each link is then identical to (3) after the following substitutions:
+ = (n),
+ = (n + ),
= (n),
+ ).
= (n
On each link, we perform the color-flavor transformation (20), thereby introducing a

complex flavor matrix field Z(n + 2 ), Z (n + 2 ). The outcome of the transformation
reads

=
Z(, )
e2am(n)(n)
{Q} n
Nf Nf

d Z, Z n +
Z,, n +
2
2

exp SZ,, (n + 2 )
Det(1 + Z Z(n + 2 ))Nc
(25)
319
where the sum on the right-hand side extends over all possible distributions {Q} of baryonic
charge (actually, baryonic flux) over the links of the lattice. The color-flavor transformed
action on a link n + /2
is

= ai (n)Zab n +
bi (n)
SZ,, n +
2
2

j
j
ba n +
,
+ b (n + )Z
(26)
a (n + )
2
and the -coefficients are 0 (n + 2 ) = 0 and (with Q = Q(n + 2 ) > 0)

Q
Z,,
n
+
2

= Q(n+ ) (n)
1 + ZZ n +
, 1 + ZZ n +
(n + )
, (27)
2
2
2

Q
Z,, n +
2

= Q(n+ )
1+Z Z n+
(n), (n + )
1+Z Z n+
. (28)
2
2
2
The fermions are now coupled through their flavor indices, whereas in the original action
the coupling had been mediated by the color degrees of freedom. Moreover, the coupling
has become ultralocal: the fermions at a site n couple only to one another, via Z(n + 2 ),
and so do the fermions at site n + ,
via Z (n + 2 ). Correlations between neighbors
are solely due to the relation between Z and Z by Hermitian conjugation. A graphical
description of the change of coupling scheme is given in Fig. 1.
The partition function (25) is a sum over all configurations of baryonic fluxes
{Q(n + 2 )}. For most of these configurations, the Grassmann integral vanishes identically.
To see that, we expand the integrand for a given configuration into a polynomial in the
Grassmann fields, and count (for each site n) the number of fermions (n), (n)
in the
various monomials:
Fig. 1. Coupling of the fermion fields before and after the color-flavor transformation.
320
For every direction , the coefficient Q (n + 2 ) contains Nc |Q| Grassmann variables

ai (n) if Q > 0, and the same number of Grassmann variables ai (n) if Q < 0.
For the coefficients Q (n 2 ) the situation is the same, except that (n) and (n)
switch roles.
Each term of the expansion of eZ+Z +2am involves as many (n)

as (n).
The Grassmann integral
d (n)
d(n)
d(n) d (n)
extracts the coefficient of the top-monomial,
Nf
Nc

ai (n) ai (n) = 1,
i=1 a=1
setting all others to zero. This monomial contains as many (n)

as (n). Hence, in view
of the counting above, the contribution from a configuration {Q(n + 2 )} vanishes unless
the following condition is met:
d

=0

d
Q n+
Q n
=
.
2
2
(29)
=0
The physical meaning of this equation is conservation of the baryon current: the (algebraic)
number of baryons arriving at the site n (from the links n /2)
must equal the number

of baryons leaving the site (via the links n + /2).
The general structure of the partition function (25) corresponds to the hadronic
correlation function written in terms of colorless Nc -quark currents [20,21].
3.1. Integration over the fermions
Based on the general considerations above, we perform the integration over the fermions
sector by sector, and present below two particular cases: the vacuum, i.e., the sector
where the baryonic flux Q(n + 2 ) vanishes on every link n + /2,
and a toy model of a
static baryon on a mesonic background (with Q(n + 2 ) = 1 along a time axis).

In each case, integration over the Grassmann variables yields a purely bosonic effective
action, which depends on the configuration of the fields Z and Z . After computing
these effective actions, we will look for their saddle-point configurations to estimate the
respective partition functions.
Before computing the effective actions in particular cases, we emphasize the consequences of chiral symmetry, which emerges in the limit of zero quark mass. The hypercubic lattice is bipartite, so it can be split into two sublattices according to the parity of
def
|n| = n . Given this splitting, the effective actions S(Z, Z ) of all sectors are invariant under the following global transformation:

U1 Z n +
U2 ,
|n| even: Z n +
2
2

U2 Z n +
U1 ,
Z n +
2
2
|n| odd:

Z n+
U2 Z n +
U1 ,
2
2

Z n +
U1 Z n +
U2 ,
2
2
321
(30)
for any pair (U1 , U2 ) U(Nf ) U(Nf ). Therefore, in each sector, the saddle-point
configurations in the chiral limit form a continuous set (namely an orbit) generated by
acting with the chiral symmetry group U(Nf ) U(Nf ). As soon as the quark masses
are turned on, this degeneracy disappears, and the saddle points become isolated. Eq. (30)
show that the fields Z, Z (n + 2 ) transform differently according to the parity of |n|.
To stress this difference, we give different names to the fields on different sublattices: the
fields living on the even lattice links will be called V , V (n + 2 ), while the fields on the
odd links will be denoted by W, W (n + 2 ).
3.1.1. Vacuum action
For the vacuum sector we have zero baryonic flux (Q = 0) everywhere on the lattice;
the integral over the fermions, being Gaussian, is then easily done and yields
d(n)Zvacuum (, )
Zvacuum = d (n)

exp(Nc Svacuum [Z]),

0 d Z, Z n +
=
(31)
2
n,
where the result of the integration has been sent back to the exponent. The integration
measure in curly brackets will be denoted by D(Z, Z ) in the following. The factor Nc
)
in the exponent comes from the color content of the fermions: since the action SZ (,
does not couple fermions with different colors, the Grassmann integral is a product of Nc
identical integrals. The effective action is
Svacuum =
Tr ln M(n) +
d

n =0

,
Tr ln N n +
2
(32)
where

d

+ Z n +
,
Z n+
2
2
=0

def
N n+
= 1+Z n+
Z n+
.
2
2
2
def
M(n) = 2am +
(33)
(34)
3.1.2. Static baryon action

By the static baryon we mean the following distribution of baryonic fluxes over the
lattice: Q(n + 2 ) = 1 along the links of the world line (or string) n = (t, 0, . . . , 0)
with t = 0, . . . , T 1; on all other links Q = 0 (see Fig. 2). This distribution
Z1+d , = 0,
satisfies the current conservation law (29) at all sites but the ends t = 0 and t = T
322
Fig. 2. Baryon string placed on a two-dimensional lattice.
of the world line. There it does, too, if we impose periodic (or antiperiodic) boundary
conditions on the Grassmann fields: for a lattice of size T in the time direction, we set
(T + 1, r ) = (1, r ), (T
+ 1, r ) = (1,
r ). We again write the partition function in
the form (31),

= D Z, Z exp(Nc Sbaryon [Z]).
Zbaryon = D D Zbaryon (, )
(35)
The effective action Sbaryon contains the sea term Svacuum [Z], plus an extra part coming
from the factors 1 along the world line of the baryon. These factors depend on the values
of the Z field along this line and on the adjacent links, through the following matrix:

3
1
3
def
1
0 M(1 0) N
0 N T
G=N
0
2
2
2

1 .
1 N T 1 0 M(T 0)
M((T 1)0)
(36)
2
(We use the abbreviation t 0 0 + t 0 to denote the sites or links on the world line of
the baryon.) This product of matrices runs over all sites n on the baryon world line (it
is expressed as a quark propagator along that line). In Appendix C we show that the
effective action takes the form

T
1 0 Nc + Nf 1
Nc Sbaryon [Z]
=
e
Nc ! 1
Nc

N
S
c
N ( )
Nc

Tr Gl
cl ( )
eNc Svacuum [Z] .
(37)
l=1
In the non-vacuum factor of (37), runs over all conjugacy classes of the group SNc of
permutations of the set {1, . . . , Nc }. Every representative of the class can be decomposed
323
as a product of cycles of various lengths l, such that cl ( ) cycles of length l occur; thus,
each class is uniquely specified by the sequence {cl }, or equivalently by a Young diagram.
The weight factor N ( ) is simply the cardinality of the class , and is given by
N ( ) = N
c
l=1 l
Nc !
cl ( ) c
)!
l (
(38)
For the lowest numbers of colors the explicit expressions are

Nc = 1:
Sbaryon = Svacuum ln Tr G + const,

1
Nc = 2: Sbaryon = Svacuum ln (Tr G)2 + Tr G2 + const,
2

1
Nc = 3: Sbaryon = Svacuum ln (Tr G)3 + 3 Tr G2 Tr G + 2 Tr G3 + const. (39)
3
The constants, which are not given above, make contributions to the baryon mass, so they
need to be taken into account in the final answer.
In the following section, we look for the saddle-point configurations of the effective
actions Svacuum and Sbaryon .
3.2. Saddle-point equations
In the two sectors that we are interested inthe vacuum and the static baryonwe
wish to compute, or at least estimate, the partition functions (31), respectively, (35). Since
we are unable to provide an exact answer, we will treat both integrals in a saddle-point
approximation, valid in the limit of a large number of colors Nc . For both the vacuum
and the static baryon, we will restrict ourselves to a purely classical approximation, which
is to say we will identify the saddle points, evaluate the action functional on them, and
approximate the partition function as Z eS(Zs.p. ) . Thus we neglect all loop corrections,
which are of higher order in 1/Nc .
In the vacuum sector, where Nc appears explicitly as a factor of Svacuum [Z], the saddlepoint approximation is fully justified in the large-Nc limit. The situation is less transparent
in the static-baryon sector (37). However, for the ansatz made below, the matrix G is
proportional to unity: G = gINf . (Note that G transforms under the chiral transformation
(30) as G U GU 1 , so the multiples of unity are fixed points of this group action.) If
one decides to consider only those configurations of Z and Z for which G is scalar, the
static-baryon action (37) simplifies to
Sbaryon [Z] = Svacuum [Z] log g + const,
so the saddle-point expansion is rigorously justified (for large Nc ) if the integral is
restricted to these configurations. We will use it to approximate the full integral.
4. Vacuum saddle-point configurations

The saddle-point analysis for the action functional Svacuum [Z] has already been carried
out in [16,22,23], so we are going to be brief here. In varying the action (32), the complex
324
matrices Z and Z are to be considered as independent, which leads to two sets of

equations. Variations of Z(n + 2 ) affect only the blocks M(n) and N(n + 2 ), with the
linear response being

1
N n+
.
Svacuum = Tr M(n) + Z n +
Z n +
2
2
2
The resulting saddle-point equation reads M(n)1 = Z (n + 2 )N(n + 2 )1 or, by taking
the inverse on both sides,

1
+ Z n +
M(n) = Z n +
(40)
.
2
2
Similarly, the variation Z (n 2 ) influences M(n) and N(n 2 ), and yields the saddlepoint equation M(n)1 = N(n 2 )1 Z(n 2 ), which is equivalent to

1
.
M(n) = Z n
(41)
+Z n
2
2
As an immediate corollary, we have

1
1
Z n+
+ Z n +
+Z n
= Z n
2
2
2
2
(42)
at every site n and for any pair , .

4.1. Homogeneous vacuum
The simplest possibility for the field Z, Z is the scalar ansatz Z = Z = zI, with
z a spacetime-independent real number. The vacuum saddle-point Eqs. (40) and (41) are
solved by this ansatz if we put

am
1
(am)2
.
1+
z = z =
(43)
2d + 1 2d + 1
2d + 1
If m > 0, the action Svacuum takes different values on these solutions. Expanding it in
powers of (am), we get

2d + 1
(2d + 2)d
Svacuum [z ] = Ld+1 Nf ln
(44)
am
,
(2d + 1)d+1/2
d +1
which shows that for a positive quark mass, the configuration z z+ minimizes the action.
In the chiral limit (m = 0), a continuous set of solutions is obtained by applying the
transformations (30) to the homogeneous configuration Z = Z = zvac I for
zvac = (2d + 1)1/2 .
This vacuum configuration is invariant under the transformations of the diagonal
subgroup U1 = U2 U(Nf ) of the chiral symmetry group, but it maps to a new vacuum
325
by taking U1 = U U(Nf ), U2 = INf . By the Goldstone mechanism, the breaking of the

continuous U(Nf ) symmetry leads to the existence of massless modes, namely the mesons,
an effective Lagrangian for which was obtained by expanding Svacuum near this vacuum in
[16,22].
If I = U U(Nf ), the vacua obtained by translating the homogeneous one by U are
staggered, in the sense that the value of Z depends on the parity of its position. However,
on adopting the notations V (n + 2 ), W (n + 2 ) for fields on the even and odd sublattices,
the staggered vacua become homogeneous for each sublattice:

= (2d + 1)1/2U,
= (2d + 1)1/2U .
V n+
(45)
W n +
2
2
In [23], it was proved that, modulo the U(Nf ) degeneracy, the configuration Z zvac
is the unique solution of the vacuum saddle-point equations in the chiral limit (except in
dimension d = 0, where the symmetry of the action is larger). The proof proceeds by a local
argument, showing that for each site n the 2(d + 1) saddle-point equations involving M(n)
imply the equality of the matrices Z(n + 2 ), Z (n + 2 ) for all = 0, . . . , d; iteration of
this result then trivially leads to the set of staggered configurations (45). The proof strongly
relies on Z being the Hermitian conjugate of Z, a constraint which is not mandatory. By
relaxing it, we are now going to find a plethora of additional solutions of the vacuum
saddle-point equations.
4.2. Nonhomogeneous vacuum configurations
By local considerations, as stated above, the only solutions of the vacuum saddlepoint equations (and the Hermiticity constraint relating Z to Z ) in the chiral case are
homogeneous in both sublattice fields V and W . However, on a finite lattice, say with
the topology of a (d + 1)-dimensional torus Ld T , there is also a global aspect to
consider: one has to make a choice of boundary conditions for the various fields. The
simplest choice are periodic boundary conditions in all directions, but one can also impose
-twisted boundary conditions, say along the first spatial direction 1:

V n + + L1 = ei V n +
,
2
2

i

,
W n + + L1 = e W n +
(46)
2
2
for all n and .
An opposite twist for the fields V , W is natural in view of their opposite
behavior under the chiral transformations (30).
Now, accepting these twisted boundary conditions, let us investigate which configuration will minimize the action (32). A homogeneous configuration suffers from a phase
jump along a d-dimensional boundary, which is energetically very costly. A more reasonable ansatz for a minimum of the action is the following: the fields V , W smoothly
rotate their phase, starting from V , W zvac for n1 = 1, to V = ei zvac , W = ei zvac at
n1 = L, with a linear phase evolution in between. In this way, everywhere in spacetime the
configuration locally looks like one of the degenerate homogeneous vacua.
326
4.2.1. Contour deformation

The above ansatz for V , W is only qualitative. In order to actually obtain field
configurations that satisfy both the twisted boundary conditions (46) and the saddle-point
equations (40), (41), we need to relax the Hermiticity relation between the fields Z and Z .
By its construction via the color-flavor transformation, the integrand eSvacuum is to be
viewed primarily as a function of the real variables {(Zab +Zba

)(n+ 2 ), i(Zab Zba
)(n+
2
d
2 )}, the total number of which is D = 2Nf (d + 1) (T L ). From [23], this function for
/ 2Z has no saddle points on RD , but it can be analytically continued into CD , where

complex saddle points may exist. On such a saddle point there must exist at least one link
(n + 2 ) where the matrix Z differs from the Hermitian conjugate of Z.
If a complex saddle point is not too far from the original contour of integration,
it contributes to the vacuum-sector partition function, upon deforming the contour of
integration so as to reach that point.
4.2.2. Vacuum saddle-point equations for twisted fields
We will demonstrate below the existence of complex saddle points for the vacuum
sector with any -twist. To make things simpler, we restrict ourselves to a 2-dimensional
spacetime, with a twist in the spatial boundary conditions (we call the time index t, the
spatial index x). The above qualitative ansatz for the fields V , W suggests the following
symmetries:
All fields are scalar, i.e., at each point Z and Z are multiples of the identity matrix.
The fields V , W are time-independent. For each position x, there are 4 field variables
associated with the time-like link which we denote by v0 (x), w0 (x), v0 (x), w0 (x),
and 4 field variables associated with the space direction, which we denote by
v1 (x + 1/2), w1 (x + 1/2), v1 (x + 1/2), w1 (x + 1/2).
Thus, at each position x = 0, . . . , L 1 we have 8 independent complex variables. In the
chiral limit (m = 0), the saddle-point equation (40) pertaining to M(x, t) on an even site
(x, t) read
v0 (x) + w0 (x) + v1 (x + 1/2) + w1 (x 1/2)
= v0 (x) + 1/v0 (x)
= w0 (x) + 1/w0 (x)
= v1 (x + 1/2) + 1/v1 (x + 1/2)
= w1 (x 1/2) + 1/w1 (x 1/2).
(47)
Eq. (41) pertaining to M(x, t + 1) are obtained by interchanging v w, v w . For a

finite quark mass, 2am is to be added to the left-hand side.
The two first equations, together with their v w exchange analogs, allow us one
more simplification. Indeed, they imply the identities v0 (x) = w0 (x), v0 (x) = w0 (x)
(the alternative possibility, v0 (x) = 1/w0 (x) and v0 (x) = 1/w0 (x), is incompatible
327
with other relations that need to be satisfied). So there remain only 6 complex variables
for each x.
In the next section, we will provide approximate solutions for Eq. (47) together with their
v w partners and assuming the above symmetries.
4.2.3. Linearized problem
To solve (at least approximately) the above equations, we will use the fact that we
expect the fields to be locally close to one of the configurations (45). We can then expand
the saddle-point equations to first order in the perturbations from that configuration, and
solve the linear
problem. We start by expanding the fields around the real positive vacuum
V , W = I/ 3:

1
v0 (x) = 1 + v0 (x) ,
3

1
v0 (x) = 1 + v0 (x) ,
3

1
v1 (x + 1/2) = 1 + v1 (x + 1/2) ,
3

1
v1 (x + 1/2) = 1 + v1 (x + 1/2) ,
3

1
w1 (x + 1/2) = 1 + w1 (x + 1/2) ,
3

1
w1 (x + 1/2) = 1 + w1 (x + 1/2) .
3
After inserting these expressions into (47) and expanding to linear order, we obtain a
transfer matrix representation of these equations, i.e., a linear equation relating the
vector of deviations of the spatial components of the fields {v1 , v1 , w1 , w1 } at position
x + 1/2, to the same vector at position x 1/2. The structure of the 4 4 transfer matrix
allows to decompose it into two 2 2 matrices, upon considering at each point the vectors

v1 + w1
v1 w1
,
.
R=
I
=
(48)
v1 + w1
v1 w1
In terms of these two vectors, the linearized equations read

6 1
def
R(x + 1/2) =
R(x 1/2) = Tr R(x 1/2),
1 0

3/2 1/2
def
I(x + 1/2) =
I(x 1/2) = Ti I(x 1/2).
1/2 1/2
(49)
(50)
Similarly, the deviations of the temporal components v0 (x), v0 (x) are determined by the
variations at x 1/2:

v0 (x)
3/8 1/8
3/4 1/4
(51)
R(x 1/2) +
I(x 1/2).
=
v0 (x)
3/8 1/8
3/4 1/4
328
Thus, the transfer matrix allows to express the linear variation of all fields by the deviations
at position 1/2.
To make the x-dependence more explicit, we seek to diagonalize the transfer matrices
Tr and Ti . The first transfer matrix Tr has the eigenvalues

1
def
def
e = (3 2 2 ), associated to the vectors R = .
(52)
e
The second transfer matrix Ti cannot be diagonalized but only put in Jordan normal form.
Indeed, it acts on the vectors

1
def 1
def
I+ =
,
I =
(53)
1
1
as
T i I+ = I+ + I ,
T i I = I .
Therefore, an initial deviation

R(1/2) = c+r R+ + cr R ,
I(1/2) = c+i I+ + ci I
propagates through the transfer matrix as follows:

x

x
R(x + 1/2) = c+r e R+ + cr e R ,
I(x + 1/2) = c+i I+ + (ci + xc+i ) I .
(54)
This linear evolution is only valid as long as the deviations from 1/ 3 are small compared
to unity. This cannot be the caseuniformly for our twisted ansatz, where the fields near
x = L take values close to ei / 3. Still, the fact that I(x + 1/2) depends linearly on the
position is encouraging: this is exactly the behavior we expect for the phases of the fields
in the ansatz.
The linearization of the saddle-point equations can actually be performed near any of
the degenerate
family of vacua (45). Linearizing the equations in the vicinity of a vacuum
v = ei / 3, we obtain for the deviations the same transfer matrix as before. We can
therefore construct local solutions near various -vacua using (54), and glue them together
to obtain a global, rotating solution. An equivalent procedure is to exponentiate the
deviations,
1
v0 (x) = exp{v0 (x)},
(55)
3
etc., and extend Eq. (54) for I(x + 1/2) to a larger domain of validity. This we do as follows.
First of all, the R-part of the deviations grows exponentially, and is staggered with
respect to x. Our ansatz excludes both features, so we simply set c+r = cr = 0. Next,
we note that the I-deviations depend on two coefficients, c+i and ci . According to (55),
their real parts describe the moduli of the fields, and the imaginary parts
the phases. In our
ansatz, we expect the moduli of the fields to be constant and close to 1/ 3 (a linear growth
would be incompatible with the boundary conditions). Therefore, we set Re c+i = 0.
The other coefficient ci causes a global shift of the fields, which can be interpreted
as a generalized chiral rotation: the generalization consists in taking in Eqs. (30) for U1
329
and U2 any invertible complex matrix, and replacing U1 and U2 by U11 and U21 . One
easily checks that the effective action is invariant under this GL(Nf ) GL(Nf ) extension
of the chiral symmetry group. The complex extension appears since we have relaxed the
Hermiticity condition. The parameter ci is then seen to parametrize the C -manifold of
scalar complex homogeneous vacua.
def
The remaining coefficient, which we abbreviate to = Im c+i , is responsible for a nonhomogeneous solution. As we explained above, deviations I(x) ix of order O(1)
can
be rescaled to deviations of order O() by changing the reference vacuum ei / 3. As
a result, the fields built from (55), (54) with cr = 0, Re c+i = 0 satisfy the saddle-point
equations up to order O( 2 ) uniformly in x.
One can add terms of higher order in to the exponent, so as to kill the higher-order
terms in the expansion of (47). By iterating the procedure, one obtains for the fields a series
expansion in powers of , such that the saddle-point equations are satisfied to any order.
We conjecture that this expansion can be (re)summed, at least in a certain domain in ,
thereby yielding an exact solution of (47). The solution up to order 2 is
eix
2
v0 (x) = eci i/2+5 /8 ,
3
eix
2
v1 (x + 1/2) = eci +i /8 ,
3
eix
2
v1 (x + 1/2) = eci +i /8 .
3
(56)
The expressions for the fields w, w are obtained by replacing , ci ci .

The coefficient parametrizes the slope of the phase with respect to x, and it must be
tuned according to the boundary conditions:
L = + 2Qw ,
(57)
where Qw is some integer. For a finite twist , can be chosen small only in the large
volume limit L % 1, in which case can take several values labeled by the integers
Qw & L.
4.2.4. Topologically non-trivial configurations
We now return to the original problem with periodic boundary conditions ( = 0).
We have shown that there exist non-trivial solutions, for which the fields are positiondependent, with their phases rotating Qw times when the position x goes from 0 to L.
The integer Qw can be called the winding number of the configuration. We can associate
a winding number to a (discrete) configuration because the phases of the fields v, w are
varying smoothly with position. More generally, when the lattice has the topology of a
(1 + d)-dimensional torus, one can associate to any smooth scalar configuration a set
of winding numbers {Qw, }, each number specifying the number of times arg(v) rotates
between the positions (n + 2 ) and (n + L + 2 ).
Using Newtons algorithm, we have searched for numerical solutions of the vacuum
saddle-point equations (47), starting from trial configurations with Qw = 1, Qw = 2. We
330
Fig. 3. Vacuum configuration computed numerically, with winding number Qw = 1 in the chiral limit. Fields
which are numerically indistinguishable (e.g., |v1 | and |w1 |) are represented by the same symbol. There is a
perfect fit with formulas (56), including the 2 correction. The difference between |v0 | and |v0 | comes from
Re ci = 0.
plot some results in Fig. 3. These plots are very well described by our approximation (56),
including the O( 2 ) corrections.
4.2.5. Nontrivial vacua for finite quark mass
So far we have constructed non-trivial vacua only in the chiral limit, where a continuum
of homogeneous vacuum configurations exists. What happens to these non-trivial vacua
when the chiral symmetry is broken explicitly by switching on the quark mass?
Recall that for m = 0 there remain only two homogeneous scalar vacua (43) out of
the former U(1) continuum, with one of them (zvac = z+ ) being an absolute minimum of
331
the action. One can again study the linearized saddle-point equations near this solution.
Unlike before, the results will apply only locally, since we cannot use the trick of rescaling
the homogeneous reference vacuum any more.
We have performed numerical searches for (1 + 1)-dimensional topologically nontrivial vacua, assuming the same symmetries as before (we took scalar, time-independent,
smoothly varying fields). The saddle-point equations are modified by the addition of the
quark mass term 2am to the left-hand side of Eq. (47). The outcome of these calculations
(see below) can be understood in large part by analytical reasoning, as follows.
To linearize the saddle-point equations around the point z+ , we set
v0 (x) = z+ ev0 (x),
etc.
(58)
As in the chiral case, the 4 4 transfer matrix splits into two 2 2 matrices that apply to
the vectors R, I defined in Eq. (48). These matrices can be written for an arbitrary value of
am, using the exact expression (43) for z+ (am) (we only consider the case d = 1). Using
the same notations as above, they are
2

2
2 + 2z4
z+ 2 z+
2z+
1
+
,
Tr (am) =
2 2z4
2 + 2z4 + 3z6
4
2z+
z+
1 z+
+
+
+
2

2
2 2z4
z+ 2 + z+
2z+
1
+
(59)
Ti (am) =
,
2 + 2z4
2 + 2z4 3z6
4
2z+
z+
1 z+
+
+
+
and the deviations on links pointing in the time direction propagate as

2
1 z+
1
v0 (x)
=
R(x 1/2)
2 ) 1 z2
v0 (x)
2(1 z+
+

2
1 z+
1
+
I(x 1/2).
2
2 ) 1
z+
2(1 + z+
(60)
One easily checks that both transfer matrices have the property Det Tr/ i (am) = 1.
We expand both matrices and their spectra in powers of am, since we are interested in
the case of a small quark mass. In the massless limit the matrix Tr (0) is hyperbolic, with
negative real eigenvalues that are well separated from each other (one expanding, the other
contracting). Thus, a perturbation of order O(am) is still diagonalizable, with eigenvalues
and eigenspaces shifted by that same order O(am). We will keep calling the eigenvalues
e , with the expansion

(am) = ln(3 + 2 2 ) + am/ 6 + O (am)2 .
(61)
On the other hand, Ti (0) was non-diagonalizable with eigenvalue +1, so a perturbation
can change its qualitative features. For any positive am, Ti (am) becomes
diagonalizable,

with real positive eigenvalues e (am) associated to eigenvectors e11(am) . To express the
deviations of the field v0 (x), we use the coefficients
def
e0 =
2 e1
1 + z+
2
1 + z+
332
For small am, these data have the following expansions:

5
2
(am) = 1/4 (am)1/2 + 7/4 (am)3/2 + O (am)5/2 ,
3
3

4
4
1/2
7/4 (am)3/2 + O (am)5/2 ,
1 (am) = 1/4 (am)
3
3

1
3
1
am
0 (am) = 1/4 (am)1/2 +
(62)
(am)3/2 + O (am)2 .
7/4
2
3
23
For a finite mass am, one expects the linear approximation to be valid only for small
deviations. However, the error introduced in the saddle-point equations by the linear
approximation is at most of order O(am|z|), so it remains small if the mass is small.
The numerical solution we obtained for a mass am = 0.01 and winding number Qw = 1
(see Figs. 4, 5) suggest that the vectors R are negligible in a large domain of x around the
point x = 0 where fields are close to z+ . This indicates that the coefficients cr vanish,
like in the chiral case. The fields will, therefore, depend on two complex parameters C :
v1 (x + 1/2) = C+ e x + C e x ,
v1 (x + 1/2) = C+ e1 + x C e1 x ,
v0 (x + 1) = C+ e+0 + x + C e0 x .
(63)
As opposed to the chiral case, both coefficients C are complex, so that both the phases and
the moduli of the fields vary with x. This ansatz fits the numerical solution even when the
deviations from z+ become of order O(1), which is quite surprising. However, it is unable
to reproduce the zone where the fields cross the negative real axis (near x = L/2). For a
quark mass am = 0.01 the values of the various exponents are
= 0.1527,
1 = 0.3045,
+0 = 0.0673,
0 = 0.0845.
(64)
These values are used in the fits to the numerical solution shown in Fig. 5.
5. Static baryon saddle-point equations

The effective action Sbaryon for the static-baryon sector, Eq. (37), contains a string
term in addition to the sea term S|0 . While the sea term depends on every one of the
matrices Z(n + 2 ), the string term involves only those matrices Z and Z that are situated
in the near vicinity of the string. More precisely, what enters into the baryon world line
and M(t 0).
Of these, the former depend only
propagator, G, are the matrices N((t + 12 )0)
on Z and Z along the string, whereas the latter also involve the matrices Z(t 0 + /2)
and
The Z field on the remaining links (away from the string) appears only in
Z (t 0 /2).
S|0 , so the variation with respect to these matrices yields the same Eqs. (40) and (41) as
in the vacuum sector.
For simplicity let us consider the two particular cases Nc = 2 and Nc = 3, using the
expressions (39) for the effective action. The most general variation yields
Nc = 2 : Sbaryon = Svacuum
Tr(G G) + Tr(G) Tr( G)

,
Tr(G2 ) + (Tr G)2
333
Fig. 4. Numerical vacuum configuration with winding number Qw = 1 and chiral symmetry explicitly broken by
a finite quark mass. The fields are normalized with respect to the corresponding value z+ . The phases arg(v0 ) and
arg(v0 ) are indistinguishable, so we plot them together (idem for arg(v1 ) and arg(w1 ), respectively, arg(w1 )
and arg(v1 )).
Nc = 3 : Sbaryon
= Svacuum
(Tr G)2 Tr G + 2 Tr G Tr(G G) + Tr(G2 ) Tr( G) + 2 Tr(G2 G)

.
(Tr G)3 + 3 Tr G Tr(G2 ) + 2 Tr G3
We then work out how the various traces of powers of G respond to variations of each
matrix Z and Z entering in the definition of G. For instance, variations of Z(t 0 + /2)
whereas varying Z((t + 1/2)0)

affects both
with = 0 affect only the matrix M(t 0),
and N((t + 1/2)0).
These computations are simplified by the use of cyclicity
M(t 0)
334
Fig. 5. Same vacuum configuration as in the previous figure, plotted on a logarithmic scale. We fit the fields in the
range 0 < |x| < 30 (where they are close to z+ ) with the ansatz (63), using the theoretical values for , 1 , 0
from Eq. (64). The best fit is obtained with C+ = (2.80 + i7.62) 105 and C = (1.82 i5.48) 105 .
properties: given any decomposition G = G1 G2 , we may replace G in the static-baryon

action by the matrix
G = G2 G1 , as G always appears under a trace.
We provide detailed calculations for the variation with respect to Z((t + 1/2)0).
1
The modified factors of G in this case are M(t 0) and N((t + 1/2)0), and the modified
matrix G reads

1
1 + N (t + 1/2)0
1 1 + M(t 0)
G + G = M(t 0)

N (t + 1/2)0 .
335
It is now natural to conjugate G + G into

1 + N (t + 1/2)0 1
T (G + G)T 1 =
G + M(t 0)
G,
def
1 N((t 1/2)0)M(t
1 . The saddle-point
where
G = N((t + 1/2)0)M((t
+ 1)0)
0)
equation that follows from varying Z((t + 1/2)0) then takes the succinct form

1 + Z (t + 1/2)0 N (t + 1/2)0 1
Z (t + 1/2)0 : 0 = M(t 0)

INf FNc (
G) ,
with the case-by-case definition of the matrix-valued function FNc (G) being
Nc = 2 : F2 (G) =
G2 + G Tr G
,
Tr(G2 ) + (Tr G)2
Nc = 3 : F3 (G) =
G(Tr G)2 + 2G2 Tr G + G Tr(G2 ) + 2G3

.
(Tr G)3 + 3 Tr G Tr(G2 ) + 2 Tr(G3 )
is similar. In terms of the

The saddle-point equation obtained by varying Z ((t 1/2)0)
= M(t 0)
1 N((t + 1/2)0)
M((t 1)0)
1 N((t 1/2)0)
it is best expressed:
matrix G

) M(t 0)
1 + N (t 1/2)0 1 Z (t 1/2)0 ,
0 = INf FNc (G
with the same definitions for FNc as above. The term multiplying the unit matrix INf stems
from Svacuum . Note that this term factors out in both of the above variations.
To get an idea of the matrix INf FNc (G), we compute it in the vacuum configuration
Z zvac INf . In this case we have G =
G=
G INf . We then notice that FNc (INf ) =
1
Nf INf for any number of colors and any = 0. Therefore, on the vacuum configuration,
we get INf FNc (G) = (1 Nf1 )INf , which is invertible as soon as Nf > 1. More
generally, this equation holds as long as G is a multiple of the unit matrix, which is a
property of the inhomogeneous scalar ansatz we will make in the next section.
Clearly, as long as the matrix INf FNc (G) remains non-singular, the saddle-point
and Z ((t + 1/2)0)
are identical to those in the
equations due to varying Z((t + 1/2)0)
vacuum sector, Eqs. (40) and (41). As was said earlier, this is also the case for the equations
due to varying all matrices Z and Z not involved in the matrix G, i.e., those away from the
string. The only difference to the vacuum equations comes from the matrices on the links
adjacent to the string, namely Z(t 0 + /2),
Z (t 0 /2)
for the directions = 1, . . . , d.

These matrices are contained only in some M 1 factor of G, and their variations give the
following saddle-point equations:
1
1 + Z (t 0 + /2)N(t
Z(t 0 + /2):
M(t 0)
0 + /2)
1 FNc (
= M(t 0)
G),
(65)
1 + N 1 (t 0 /2)
M(t 0)
Z(t 0 /2)
Z (t 0 /2):
1

= FNc (G ) M(t 0) ,
(66)
336
are the same as before. These equations represent the only

where the matrices
G, G
obstruction that prevents the vacuum configuration Z = Z zvac I from being also a
saddle point of the static-baryon sector.
5.1. Configurations for the static baryon in 1 + 1 dimensions
In this section we present approximate solutions of the saddle-point equations in
the static baryon sector for the simplest non-trivial case, which is the two-dimensional
Euclidean square lattice (d = 1). We use the same notations and assume the same
symmetries as in Section 4.2.2, so at each position x there are 8 independent complex
scalar variables.
The baryonic string is placed on the Euclidean time axis at position x = 0. The equations
to solve are the vacuum saddle-point equations (47) off the string (x = 0), and the modified
equations
v0 (0) + w0 (0) + v1 (1/2) + w1 (1/2) + 2am
= v0 (0) + 1/v0 (0)
= w0 (0) + 1/w0 (0)

= 1 Nf1 v1 (1/2) + 1/v1 (1/2)

= 1 Nf1 w1 (1/2) + 1/w1 (1/2)
(67)
on the string, together with the equations obtained by exchanging v w, v w .

As in the vacuum sector, these equations imply the identifications v0 (x) = w0 (x) and
w0 (x) = v0 (x) for all x.
5.1.1. Physical requirements
Recall from Section 4.1 that demanding Z to be the Hermitian conjugate of Z (and
assuming the vacuum saddle-point equations) leads to a homogeneous configuration,
where the fields are constant on each sublattice. Such a homogeneous configuration cannot
satisfy the last two of Eq. (67). To get a solution, we must relax the Hermiticity condition
(cf. Section 4.2.1), and consider the fields Z and Z as independent variables.
We want the baryon to be a localized object, in the sense that a baryonic saddle-point
configuration should differ from a vacuum configuration only in some neighborhood of the
baryon world line. The baryon can then be interpreted as a spatially localized excitation of
this vacuum. A priori, baryon excitations may exist on top of each of the vacua described
in Section 4.
In Eq. (67), the number of flavors Nf enters just as a parameter, so one can extend the
equations to any real value of Nf . In the limit Nf = , we recover the vacuum saddlepoint equations. We can therefore obtain a solution of the baryon saddle-point equations by
starting from a given vacuum configuration (at Nf = ), and deforming the configuration
by continuous variation of Nf down to its physical value (say Nf = 2). Any baryon
configuration obtained in this way carries the same topological charge as the vacuum it
is associated to.
337
Fig. 6. Numerical baryon configuration in the chiral limit. All fields are real. Top: the fields converge exponentially
fast to zvac . Bottom: we compare the numerical data (circles, triangles, squares) with the theory of Section 5.1.2
(3 lines, cross for z0 (0)).
5.1.2. Topologically trivial sector Qw = 0, chiral limit

In the sector with zero winding number, we numerically found a unique solution
(see Fig. 6) asymptotic to the homogeneous vacuum zvac , i.e., satisfying the asymptotic
condition
|x|
v , v , w , w zvac = 1/ 3.
All the fields of this configuration are real and time-independent. Various components
coincide pairwise or in quadruples:
v0 = w0 = v0 = w0 z0 ,
v1 = w1 z1 ,
v1 = w1 z1 .
338
One easily checks that the saddle-point equations are invariant under the following
transformation:
z1 (x + 1/2) z1 (x 1/2),
z0 (x) z0 (x),
(68)
which represents a reflection at the baryon world line. The solution found numerically is
invariant under this transformation, and we believe the same to be true for the exact solution
(or else we would get a second solution by reflection). We can therefore restrict our study
to the domain 0 x.
Our numerics show an exponential convergence of all the fields towards zvac as we
depart from the string (see Fig. 6), and the signs of the deviations alternate with x.
This phenomenon can be explained by the linearized saddle-point equations studied in
Section 4.2.3. The linear theory indeed applies if the fields are close to zvac , which is
the case for large enough x. Eqs. (51), (54), together with the physical condition that the
deviations decay as x , require c+r = 0. We thus get the ansatz

x
1
z0 (x) = 1 + cr 2 e e ,
3

x
1
z1 (x + 1/2) = 1 + cr e ,
3

x
1
z1 (x + 1/2) = 1 + cr e e ,
(69)
3
with e = 3 + 2 2 as before. According to this linear approximation, the fields oscillate

around the asymptotic value zvac , and the amplitude of the oscillations is controlled by a
def
unique coefficient, which we denote by C1 = cr .

The results (69) fit the numerical configuration not only far from the string (where this
is expected), but even down to the baryon string, where the fields deviate significantly from
zvac . More precisely, the ansatz fits z1 , z1 for all x, whereas z0 departs from it only at x = 0.
def
0 ) together with the parameter C1 can be computed using the
The value z0 (0) = 1 (1 + C
3
(nonlinear) saddle-point equations on the string (67) and the reflection symmetry (68). We
obtain two equations:
02 + 2C1 C
0 + 2C1 + 4C
0 = 0,
C

0 + C1 3 + 7e + 4 = 0.
0 C1 + 4C
3e C12 + 4e C
The equations have four pairs of solutions, two real ones and two complex ones, conjugate
0 = 0 as we vary Nf from
to one another. The physical solution (which deforms to C1 = C
0 = 0.0504, giving
2 to ) is C1 = 0.0971 and C

(70)
2 e C1 = 0.8002.
The relative smallness of these constants (except for the last one, which governs the
amplitude z0 (0)) may explain why the ansatz works well down to the string.
The above saddle-point configuration is indeed situated outside of the original contour
of integration: it is a complex saddle point (although all fields have real values).
339
A contour deformation has to be performed for the variables i(z1 z1 )(x), which move
away from the real axis as |x| decreases.
This ansatz is tailored to the limit L , but owing to the fast decrease towards zvac ,
it is already quite good for short lattices. In Fig. 6 (bottom), we show a logarithmic plot
of the deviations of the fields from the homogeneous vacuum, for a lattice of total length
L = 20, as well as the values predicted by the above ansatz.
This configuration might be called a non-topological soliton, cf. [21]. Its characteristic
length (in units of the lattice spacing a) is 1 = 0.5673, and its mass will be computed in
the next section.
This real scalar configuration is just one point on a U(Nf )-manifold of solutions,
obtained by the action (30) of the chiral symmetry group. For a generic point on
this manifold, the configuration is staggered in time. These solutions are saddle-point
configurations for Nf = 2, and do not depend on the number of colors Nc (which does
not appear in the saddle-point equations). However, the value of the action for these
configurations does depend on Nc ; see the next section.
5.1.3. Topological baryon, chiral limit
We also obtained topologically non-trivial configurations, characterized by a nonvanishing winding number Qw . In Fig. 7 we plot a solution in the baryon sector with
Qw = 1. All fields are scalar, and have the symmetries described in Section 4.2.2.
Away from the string (x % 1), the moduli of the fields are close to zvac , and the phase
varies linearly. In this region, we can apply the linear theory described in Section 4.2.3,
in particular the ansatz (55). Now only the coefficient c+r has to vanish, to prevent the
deviations from exploding as x . As in the vacuum case, the coefficient ci can
take any value, yielding only a chiral shift of the fields. We find that c+i = i is purely
imaginary, as in the vacuum. The fields are well fitted by

x
1
v1 (x + 1/2) = ei(x+1) 1 + cr e ,
3
etc.
(71)
for positive x. Near the string, the moduli of the fields behave in a similar way as in the
non-topological sector, whereas their phases make a small jump at x = 0. Both this jump
and the value of cr can be computed from the full saddle-point equations near the string
(see below). The value of the parameter depends on the height of this jump: will not
be exactly equal to its value in the vacuum, which is 2/L for this topological sector,
but rather to (2 )/L. As a consequence, the convergence of the fields towards the
vacuum configuration away from the string will not be exponential, but only linear (the
fields coincide at the antipode of the baryon, x = L/2).
From Fig. 7, we can assume that the above ansatz is still a good approximation
for v1 (1/2), and for v1 (1/2) up to a phase jump of /2. Given this assumption and
setting Nf = 2, the saddle-point equation on the baryon worldline, v0 (0) + 1/v0 (0) =
1/2{v1 (1/2) + 1/v1 (1/2)}, yields two real equations, one of which reads
3(1 + C1 )
sin(3/2 + /2)
=
.
sin(/2)
2(1 + C1 e )
340
Fig. 7. Numerical baryon in the chiral limit with winding number Qw = 1 (we only plot the vicinity of the
baryon worldline). The absolute values are very similar to the case Qw = 0, but now the phases vary linearly
away from the worldline. The slope and the phase jump at the baryon are in good agreement with the theory
of Section 5.1.3: we find 2/ = 123.42 and arg v0 (1) arg v0 (1) = 0.046 5.24.
For small angles and , this equation gives a linear relation between them: using the
value for C1 obtained in the non-topological sector, we get 3.24 . The value of the
slope then is = 2Qw /(3.24 + L).
5.1.4. Topological baryon, m = 0
As in the vacuum sector, our results for a finite quark mass are mostly numerical (see
Figs. 8, 9). We obtained solutions of the saddle-point equations with various winding
numbers, which are close to the corresponding vacuum configurations except in the vicinity
of the baryon string. Near the antipode of the string (x = L/2), the field approaches the
341
Fig. 8. Numerical baryon with winding number Qw = 1 and broken chiral symmetry. Away from the worldline,
the fields converge exponentially fast to z+ . The logarithms of the phases are linear in the range 3 < x < 55, with
slopes 0.1523 0.1533 in agreement with the theoretical value from Eqs. (64).
homogeneous value z+ , and the fields can be fitted by the linear theory developed in
Section 4.2.5.
6. Mass of the static baryon (Nf = 2)

The mass Mbaryon of the baryon is defined by comparing the static baryon partition
function to the vacuum one. Since the saddle-point configurations are classified according
342
Fig. 9. Numerical baryon with winding number Qw = 2 and broken chiral symmetry. The symmetry with respect
to x = 0 is only approximate, due to a numerical loss of accuracy.
their winding number Qw , the comparison is performed within a given topological class,
i.e., between a baryon configuration and the corresponding vacuum sector.
In the limit of large lattices, the ratio of partition functions is expected to behave as
Zbaryon,Qw (L T )
eMbaryon,Qw T
Zvacuum,Qw (L T )
as T .
(72)
As explained in Section 3.2, we estimate both partition functions through their respective
lowest-order saddle-point approximations:

Zbaryon
exp Nc Sbaryon [zbaryon] Svacuum [zvac ] .
Zvacuum
(73)
343
As before, we will only treat the (1 + 1)-dimensional case.

6.1. Mass of the static non-topological baryon
We first study the baryonic excitation on top of a homogeneous vacuum, which has
vanishing winding number. The value of the action for the homogeneous vacuum was
given in Eq. (44). It is proportional to the volume of the lattice, and is called a sea term
in the literature.
The action Svacuum is part of the full static baryon action (37), which leads to a sea
contribution to the baryon mass:
Svacuum [zbaryon] Svacuum [zvac ]
.
T
We now assume the limit m = 0 (Section 5.1.2). By the time-independence of both
configurations, the sea contribution reads
L/2

N1 (x + 1/2)N0 (x)
Msea = Nf Nc
ln

M(x)
Z=zbaryon
x=L/2

N1 (x + 1/2)N0 (x)
.
ln

M(x)
Z=zvac
def
Msea = Nc
Since the deviations of the fields from zvac are small and decrease exponentiallysee
Eq. (69)it is reasonable to keep only the linear order, and extend the sums to L = .
Quite remarkably, this linear approximation gives a vanishing sea term (for m = 0).
Alternatively, we can compute the sea term from the Nf = 2 configuration obtained
numerically. In this way we obtain

Msea 0.02324 Nc in units of a 1 .
This answer is small (5%) compared to the second term we compute below, so the linear
approximation (giving Msea = 0) is rather good in this respect.
The remaining contribution to the baryon mass comes from the sum over traces of G,
which involve only the fields on or adjacent to the string. In the QCD context this is
generally referred to as the valence quark contribution [21]. For the time-independent
configuration described in Section 5.1.2, the valence term for Nf = 2 becomes

0 Nc + Nf 1
2z0 (0) + 2z1 (1/2)
Mvalence = ln
(74)
+ Nc ln
.
Nc
1
1 + z0 (0)2
zbar
The term proportional to Nc evaluates to

0 /2
1 + C1 /2 + C
ln 3 + C1 /2 = 0.5493 0.0485.
ln 3 + ln
2
0 /2 + C
/4
1+C
0
We notice that the second term due to the deviations |zbaryon zvac | is small compared to
the first term ln 3, obtained by inserting the vacuum configuration zvac into (74).
344
Table 1
Qw
m=0
am = 0.002
am = 0.01
0
1
2.839
2.840
2.840
3.097
2.852
3.692
The result of this linear approximation is very close to the exact (numerical) value,
0.5004. On including the combinatorial and normalization terms (see Eq. (B.7)), we finally
get, for the non-topological static baryon in the chiral limit:
Mbaryon 0.5236Nc + ln(1 + Nc /2) + ln(1 + Nc ).
(75)
This yields for example

Nc = 2 : Mbaryon = 2.839 a 1,
Nc = 3 : Mbaryon = 3.873 a 1,
where we have reinstated the mass scale given by the lattice constant.
6.2. Masses of topological baryons
To compute the baryon masses in the topologically non-trivial sectors, we first need
to evaluate Svacuum on the corresponding vacua with winding number Qw . This is
straightforward in the chiral limit, where we have the accurate approximation (56) at our
disposal. The result up to second order in the small parameter = 2Qw /L is

3 2 3 2 Q2w
=
Nf .
T 1 Svacuum [zvac, ] Svacuum [zvac ] = Nf L
16
4L
The vacuum energies are obtained from this by multiplication with the number of colors,
Nc . They agree well with the values computed numerically for L = 120 (and Nc = 2):
(0)
(0)
E1 = 0.25, E2 = 0.98.
The baryon mass in each topological sector is defined relative to the corresponding
(0)
vacuum energy EQw . For a non-vanishing quark mass, both the vacuum energy and the
baryon energy are computed numerically. In Table 1, we summarize our results for a lattice
L = 120, with Nc = Nf = 2, for various quark masses: In Fig. 10, we plot the baryon
masses in the topological sectors Qw = 0 and Qw = 1 as a function of the quark mass.
7. Zigzag baryon
In the standard formulation of the theory on the Euclidean 2-dimensional square lattice,
the temporal and spatial directions are given by the lattice generators, t = (1, 0), x = (0, 1).
However, it is also possible to use different spacetime axes. For instance, on the same
square lattice, we define the zigzag spacetime axes as t = (1/2, 1/2),
x = (1/2, 1/2)
(see Fig. 11). The unit spacetime separation now has the length a/ 2. For convenience,
we choose the spacetime origin on the middle of a link (say, a link in the direction t + x),
345
Fig. 10. Masses of static baryons as functions of the quark mass am, for the topological sectors Qw = 0 and
Qw = 1.
Fig. 11. Zigzag baryon string on a square lattice.
so that the coordinates of lattice sites will be half-integers, while links will be indexed by
integers.
The division of the square lattice into two sublattices is now expressed only in terms of
the spatial coordinate: the links at even positions x = 2n carry the fields V , V , while the
links at odd x = 2n + 1 carry the fields W, W .
The vacuum effective action and its corresponding saddle-point equations are still given
by the formulas (40), (41), after a suitable change of labels for links and sites.
346
7.1. Time-independent vacua

Once again, we make a scalar time-independent ansatz for the fields:
x even, t :
V (x, t) = v(x)I,
x odd, t :
W (x, t) = w(x)I,
V (x, t) = v (x)I,
W (x, t) = w (x)I.
(76)
(77)
In particular, this implies the equality of fields situated on links (x, t) and (x, t + 1). There
are 2 scalar variables at each position.
With these symmetries, the vacuum saddle-point equations read:
x even :
2v(x) + 2w (x 1) + 2am = v(x) + 1/v (x)

= w (x 1) + 1/w(x 1),
x odd :
2w(x) + 2v (x 1) + 2am = w(x) + 1/w (x)

= v (x 1) + 1/v(x 1).
(78)
In the chiral limit, the homogeneous vacuum configurations are still given by v = w =
ei zvac and v = w = ei zvac . For fields close to this vacuum (v(x) = ei zvac exp{v(x)},
etc.) the linearized saddle-point equations yield the following transfer matrix equation:

z(x 1)
z(x 1)
z(x)
3/2 1/2
=
=
.
T
(79)
zig
1/2 1/2
z (x 1)
z (x 1)
z (x)
The symbol z stands for either v or w, depending of the parity of x. In contrast with
Section 4.2.3, we now have just one transfer matrix, which relates deviations of v, v to
deviations of w, w and vice versa. This transfer matrix Tzig is related to the matrix Ti
described in Section 4.2.3. Indeed, it acts on the vectors I+ , I as follows:
Tzig I+ = I+ I ,
Tzig I = I .
The deviations v(0), v (0) are parametrized by two complex parameters c as

v(0)
= c+ I + + c I .
v (0)
The deviations will then depend on position as follows (x is even):

v(x)
c + (x + 1)c+
c (x + 2)c+
w(x + 1)
=
=
,
.
v (x)
c (x 1)c+
c + xc+
w (x + 1)
(80)
The rest of the discussion is identical to the one following Eq. (55). The coefficient c
plays the role of a global shift, or generalized chiral rotation. If Re c+ = 0, the absolute
values of the fields vary linearly with x, which is incompatible with their periodicity. On
the other hand, taking c+ = i will linearly rotate the phases of the fields, keeping them
close to some vacuum configuration, as in Eq. (56). Taking for a multiple of 2/L, we
obtain a topologically non-trivial configuration.
The case of broken chiral symmetry (m = 0) can be treated along the same lines as in
Section 4.2.5; the above linear evolution in position is then replaced by an exponential
347
one, at least for fields in the vicinity of the value z+ . The transfer matrix takes the
z+2 /2 1/2
form
2 /2 , and has the eigenvalues exp(zig ) associated to the eigenvectors
1/2 3z+

1
exp(zig ) , with the expansions

2
zig = 1/4 (am)1/2 + O (am)3/2 ,
(81)
3

2 2
zig = 1/4 (am)1/2 + O (am)3/2 .
(82)
3
Up to this order, the exponent zig differs from the corresponding exponent of Eq. (62)
compensates for the ratio of unit lengths between

by a factor of 2. This factor actually
the two frameworks (a versus a/ 2). For the quark mass am = 0.005 used in compiling
the figures, we have zig = 0.0762 and ezig = 0.85905.
7.2. Zigzag baryon configurations
In the new labeling conventions, the worldline of a static baryon situated at position
x = 0 forms a zigzag curve (see Fig. 11). Assuming that the fields are scalar and timeindependent, the G matrix appearing in the baryonic part of the action is

1

1 + v(0)v (0)
G = 2v(0) + 2w (1) + 2am

T /2

I.
2w(1) + 2v (0) + 2am 1 + v(0)v (0)
The resulting saddle-point equations on the string read
2v(0) + 2w (1) + 2am = v(0) + 1/v (0)

= 1 Nf1 w (1) + 1/w(1) ,
2w(1) + 2v (0) + 2am = v (0) + 1/v(0)

= 1 Nf1 w(1) + 1/w (1) .
(83)
The saddle-point equations are invariant under the transformation

w(x) w (x),
v(x) v (x),
(84)
which is also a symmetry of our numerical solutions.

In the chiral limit, the linear dependence in (80) makes it impossible for the absolute
values of the fields to approach zvac at infinity, unless the deviations are purely imaginary;
this latter possibility (|z(x)| zvac ) is incompatible with the saddle-point equations on
the baryon string (83). However, for a finite lattice, infinity is the antipodal point
x = L/2. In numerical searches (see Fig. 12, top), we found a solution which comes
close to zvac near the antipode (but does not converge exponentially to it). To describe these
fields, it is convenient to use the position variable centered at the antipode, x = x + L/2
(we assume that L/2 is even, so that x and x have the same parity). The symmetry (84) also
holds if one replaces x by x . For x even, L/2 < x < L/2, the fields are well described
by
348
Fig. 12. Numerical zigzag baryon for the case with chiral symmetry (top) and without (bottom). We plot the
(real) fields on a logarithmic scale. Top: a linear fit yields the slope c+ = 0.04 and the field on the baryon
v(0) = 0.1735, in excellent agreement with formulas (86). Bottom: we use the exponential ansatz of Section 7.2.1,
with coefficients C fitted over the domain |x | < 30. The values C+ = 0.062, C = 0.053 are in good agreement
with the analytical theory.

v(x ) = zvac ec+ (x +1) ,
v (x ) = zvac ec+ (x 1) ,

w(x + 1) = zvac ec+ (x +2) ,
w (x + 1) = zvac ec+ x ,
(85)
where the value of the real coefficient c+ is small (for a lattice of length L = 80, we
found c+ 0.04). At all points x = 0, the fields are close to a generalized homogeneous
vacuum.
The coefficient c+ can be analytically estimated by using the above ansatz for w, w at
the position x = 1 + L/2, and then enforcing Eq. (83). One obtains (to lowest order in
349
c+ ) the transcendental equation

c+ ec+ L = 1,
(86)
with approximate solution c+ L1 ln L. The field on the baryon then takes the value
3zvac
c+ .
2
This configuration is very distorted compared to the original contour of integration.
Indeed, the ratios v/v and w/w are of order L in the vicinity of the string. In the case
L = 80, the above equations yield c+ = 0.04018, v(0) = 0.1736, which are in excellent
agreement with our numerical data.
v(0) = v (0) =
7.2.1. Zigzag baryonbroken chiral symmetry

If chiral symmetry is broken by a non-vanishing quark mass, the deviations from z+
evolve exponentially with x away from the antipodal point x = 0. More precisely, in
some domain |x | & L/2 the fields should follow the ansatz

z(x )
C+ (ezig )x + C (ezig )x
.

C+ ezig (ezig )x C ezig (ezig )x
z (x )
From the mirror symmetry (84), the coefficients C are related by
C = ezig C+ .
(87)
We numerically computed a solution with L = 80, am = 0.005 (see Fig. 12, bottom), for
which this ansatz works well up to the string. Using this fact, it is possible to estimate the
value of C+ (the last remaining parameter), as in the chiral case. The crudest approximation
yields the equation

def
C+ zig C e2CC+ = 2, with C = exp zig (L/2 1) .
The solution of this equation in the case L = 80, am = 0.005 is C+ = 0.0614. This
configuration has a mass M = 2.885 a 1 with respect to the homogeneous vacuum.
There also exist topologically non-trivial vacuum and baryon saddle-point configurations with the symmetry (84). For example, Fig. 13 shows the solution in the sector Qw = 1
with quark mass am = 0.005 on a lattice of length L = 160. With respect to the corresponding vacuum configuration, this configuration has a mass M = 3.18 a 1 .
The color-flavor transformation introduced in [14,15] replaces an integral over the
gauge group U(Nc ) by an integral over the flavor degrees of freedom. In the present
paper we extended this transformation to the gauge group SU(Nc ).
The color-flavor transformation can be interpreted as a kind of duality, linking two
different formulations of the theory. We believe that this duality transformation may be
useful for treating realistic non-perturbative QCD. Here we have applied it to a simple
model of two-dimensional lattice fermions. The non-Abelian theory we have treated is
350
Fig. 13. Numerical zigzag baryon with winding number Qw = 1 and broken chiral symmetry. We plot the real
(top) and imaginary (bottom) parts of log(z/z+ ) on a logarithmic scale, together with a fit by the ansatz (7.2.1) in
the region |x | < 60. The coefficients take the values C+ (3.48 + i1.07) 103 , C (3.00 + i0.92) 103 ,
and satisfy quite well the relation (87).
of course too far from realistic four-dimensional QCD for our results to be of direct
phenomenological relevance.
The main approximation we made was to assume the strong-coupling limit for the
lattice gauge fields. In this approximation gluons do not propagate, and the connection
to asymptotic freedom at short distances is lost. An unphysical consequence is the absence
of the U(1) chiral anomaly. Thus, the chiral symmetry group of our low-energy effective
action is not SU(Nf ) but the larger group U(Nf ).
351
Among the gauge groups SU(Nc ) the case Nc = 2 is special, as the vector and covector
representations of SU(2) happen to be equivalent. Since these representations correspond
to quarks and antiquarks, respectively, there is no physical distinction between baryons
and mesons in that case. This symmetry between baryons and mesons is obscured in the
present treatment which, by the use of a saddle-point approximation valid only for Nc % 1,
is geared to the large-Nc limit. It can, however, be made manifest by identifying SU(2) with
the compact symplectic gauge group Sp(2) and using the color-flavor transformation for
the latter [15].
To extend the formalism to lattice QCD in four dimensions, we need to take into account
the spin degrees of freedom and put the chiral fermions properly on the lattice. We hope to
address these issues in a separate publication. Here we only note that a first step towards
a more realistic color-flavor transformed theory of the strong interaction was described in
[22,24], where we discuss the effect of spontaneous chiral symmetry breaking and estimate
the numerical values of the chiral condensate, the pion decay constant and the mass of the
pion.
Acknowledgements
We are grateful to A. Altland, A. Andrianov, D. Diakonov, A. Morel, V. Petrov,
P. Pobylitsa, M. Polyakov, R. Seiler, A. Smilga and A. Wipf for useful discussions and
comments.
Appendix A. Action of the color and flavor groups

Transformations U GL(Nc ) of color space act on the one-fermion operators by
kj
j
k
TU f+a
TU1 = f+a U 1 ,
k
TU fa
TU1 = fa U j k ,

k T 1 = fj U 1 kj .
TU fa
a
U
j
k T 1 = f U j k ,
TU f+a
+a
U
A B
Transformations C D GL(2Nf ) of flavor space can be decomposed in the way shown
in Eq.
of the various factors may be described separately. An element
(13),
and the action

= 10 Z1 = exp 00 Z0 acts by
j
k T 1 = f k fk Z , T f k T 1 = f k + fk Z ,
T f+b
b
a ba
+a ab
+b
b
1
1
k
k
k
k
T f+b T = f+b ,
T fb T = fb ,
an element diag(A, D) =
A 0
0 D
GL(2Nf ) by
1
k T 1
k
k T 1
k
, Tdiag(A,D)fb
Tdiag(A,D) f+b
diag(A,D) = f+a A
diag(A,D) = fa Dab ,
ba
1
k T 1
k T 1
k
k
Tdiag(A,D)fb
,
Tdiag(A,D) f+b
diag(A,D) = f+a Aab ,
diag(A,D) = fa D
ba
352
and an element =
1 0
1
Z
GL(2Nf ) by
k
k
k
k
T 1 = f+b
,
T fb
T 1 = fb
,
T f+b
1
1
k
k
k
k
k
k
T f+b T = f+b + fa Zab , T fb T = fb f+a

Zba .
All these formulas are particular cases of the fermionic Fock-space representation of the
Lie group GL(2Nf ) expounded in Chapter 9 of [19].
Appendix B. Normalization constants

1
We are going to calculate the normalization constants Q
= G dg |BQ |Tg |BQ |2
introduced in Eq. (11)for the values Q = 1, 0, 1. To that end, we employ the
decomposition of the group G = U(2Nf ) given by Eqs. (13) and (14). This yields
2Nc

B1 |Tg |B1 2 = |( 1 + ZZ )1a Ua1 |
Det(1 + ZZ )Nc
for Q = 1, and similar expressions for the other two cases. The first step now is to do the
integral over U U(Nf ), which for Q = 1 is effectively an integral over a (2Nf 1)dimensional sphere. Carrying it out by the method of Section 2.5, we get the preliminary
expressions
dZ dZ
01 = CNf
(B.1)
,
Det(1 + ZZ )2Nf +Nc
C
11
1
= 1
Nf Nf
= CNf
(Nf 1)!Nc !
(Nc + Nf 1)!
N N
C f f
[(1 + Z Z)11 ]Nc dZ dZ

,
Det(1 + ZZ )2Nf +Nc
where CNf is defined by
dZ dZ
1
CN
=
.
f
Det(1 + ZZ )2Nf
C
(B.2)
(B.3)
Nf Nf
For later convenience, we have made a change of integration variables Z Z in the

numerator of the integral in (B.2).
In the second step we perform the integration over the Nf Nf matrix Z using a
recursion procedure similar to that in [25]. From here on we use the simplified notation
n = Nf . The recursion consists in slicing the matrix Z into vertical vectors, step by step.
We now detail the first step of the recursion. We decompose Z as Z = (Zn,n1 , z1 ), where
z1 is a (column) n-vector, and Zn,n1 is a n (n 1) matrix. We then have the expressions

+ z1 z1 ,
Z Z 11 = Zn,n1
Zn,n1 11 .
ZZ = Zn,n1 Zn,n1
Using the (positive definite) n n matrix 1 which is defined as the square root of
12 = 1 + Zn,n1 Zn,n1
,
353
we make a change of variables, from z1 to w1 = 11 z1 . From 1+ZZ = 1 (1+w1 w1 )1 ,

we get the relation

.
Det 1 + ZZ = 1 + w1 w1 Det 1 + Zn,n1 Zn,n1
The change of variables from Z to {Zn,n1 , w1 } has the Jacobian Det(1 + Zn,n1 Zn,n1
).
Each of the integrals (B.1), (B.2), and (B.3) can now be written as the product of a Zn,n1
integral times a w1 -integral.
The former can in turn be expressed as the product of a Zn,n2 -integral times a w2 integral (with w2 a n-vector), which can be decomposed in turn, and so on, until we
reach, at the nth step, a Zn,1 -integral, i.e., an integral over the first column of the original
matrix Z. We call this column vector wn for reasons of homogeneity.
The successive Jacobians multiply to give the following integration measure:

n1
dwn dwn .
dZ dZ = dw1 dw1 1 + w2 w2 dw2 dw2 1 + wn wn
The integrands in (B.1)(B.3) also have simple expressions in the new variables, due to the
identities (Z Z)11 = wn wn and

Det 1 + ZZ = 1 + w1 w1 1 + w2 w2 1 + wn wn .
The wi -integrals to be performed are all of the type (N n)
dw dw
(N n)!
.
= n
(1 + w w)N+1
N!
Cn
The resulting expressions for the normalization constants are

0 =
1
CNf
Nf2
(2Nf + Nc 1)! (Nf + Nc )!

,
(Nc + Nf 1)! Nc !
1 = 1 =
(B.4)
Nf (2Nf + Nc 1)! (Nf + Nc + 1)!

,
(Nc + Nf 2)! Nc !
(B.5)
CNf
1 (2Nf 1)! Nf !
CNf =
(B.6)
,
2
Nf (Nf 1)! 0!
where we have reinstated n = Nf . The quantity entering into the baryon mass is the ratio
Nf2
Nf
1
=
.
0 Nf + Nc
(B.7)
Appendix C. Static baryon

In this appendix we prove the formula (37) for the action functional of the static baryon
sector. We need to integrate polynomials in the quark fields along the worldline of the
baryon (the baryon string), weighted by the same Gaussian as in the vacuum sector.
We start out using the short-hand notation t 0 0 + t 0 of Section 3.1.2 for sites and links
on the string. The part of the integrand containing the quark fields situated on the string,
354
(t
for t = 0, . . . , T 1, then reads
0)
namely (t 0),

1
3
1 (0), N
0 (10) 1 (10), N
0 (20)
2
2
!

T
1

1
i (t 0)
.
exp
1 (T 1)0 , N T
0)
i (t 0)M(t
0 (T 0)
2
t =0
we are faced with the integral

Isolating the terms with fermions at the site n = t 0,

N(n 0/2)(n)
0),
d (n)
d(n) 1 (n

e(n)M(n)(n) 1 (n),
N(n + 0/2)(n
+ 0)

2

k
(Nf 1)!
k
d ai (n) dai (n)ec (n)Mcc (n)c (n)
= 1
(Nc + Nf 1)!
i,a

(i)
ab (n 0/2)
ai (n 0)N
sgn sgn
b (n)
, SNc
j
(j )
a (n)Na b (n + 0/2)
b (n + 0)

= 1
2
1
(Nf 1)!
ai bi (n 0/2)
sgn( )
ai (i) (n 0)N
(Nc + Nf 1)!
i
, SNc

i
i
d i (n) d i (n) bi i (n) ai (n)e (n)M(n) (n)

i

(i)
,
Nai bi (n + 0/2)
(n + 0)
b
i
where the first equality sign uses the expression (21) for the function 1 . Note that the
integral between curly brackets involves only fermions of color i. The fermionic version
of Wicks theorem yields for it the value Mb1a (n) Det M(n), so after combining the
i i
permutations and , the above expression becomes
12
(Nf 1)!2
Nc ! Det M(n)Nc
(Nc + Nf 1)!2

a b (n 0 n + 0)
(i)
ai i (n 0)G
sgn
(n + 0)
i i
b
SNc

Nc + Nf 1 1
G(n 0 n + 0)(n
,
1 (n 0),
= 1 Det M(n)
+ 0)
Nf 1
Nc
1 N(n + 0/2).
def
= N(n 0/2)M(n)
with the propagator G(n 0 n + 0)
Repeating the procedure, we successively integrate over the quark fields along the string,
by which process the matrices N and M get organized into a single propagator. In the final
integration step, we need to take into account the periodic boundary conditions for the
355
= (0). The final integral over (0) then reads

quark fields: (T 0)
(0)M(0)(0)

G(0 T 0)(0)
e
.
d (0)
d(0) 1 (0),
We now use the following expression for the function 1 :
) = 1
1 (,
Nc
(Nf 1)!
ai i ai (i) ,
(Nc + Nf 1)!
{ai } SNc i=1
which is easily obtained from Eq. (21) by interchanging the product over colors with the
sum over flavors. Wicks theorem then yields for the (0)-integral the result

(Nf 1)!
1 (0).
Det M(0)Nc (1)Nc
Ga (i) bi (0 T 0)M
1
bi ai
(Nc + Nf 1)!
{ai ,bi } SNc i
The last matrix product may also be expressed in terms of the propagator G defined in
1 .
Eq. (36), G = G(0 T 0)M(0)

What is the interpretation of the sign factor (1)Nc ? To answer that question,
recall that we evaluated the Grassmann field integral using time-periodic boundary
conditions (instead of the conventional time-antiperiodic ones). In a d-dimensional
quantum mechanical frame work with Hamiltonian H and inverse temperature , this
would mean that we are computing not the usual partition function but rather the
supertrace Tr(1)NF eH with NF the total fermion number. The overall sign factor
(1)Nc originates from that very fermion number, and is simply telling us that the baryon
is a fermion (boson) if Nc is odd (respectively, even).
Let us take a closer look at the contributions from the sum over permutations SNc .
Each permutation can be uniquely decomposed into a product of independent cycles.
Denoting by cl ( ) the number of cycles of length l in this decomposition, the contribution
from to the partition function can be written as
Nc

{ai } i=1
Gai a (i) =
Nc

Tr Gl
cl ( )
l=1
The permutation group SNc may be partitioned into disjoint classes with respect to
conjugation (, are said to be conjugate to each other iff there exists a permutation
such that = 1 ). Two permutations and are in the same conjugacy class iff
they have the same cycle structure, i.e., l : cl ( ) = cl ( ). This allows to rewrite the sum
Nc , taking into account the cardinality of
over as a sum over the conjugacy classes S
each class, N ( ), given in Eq. (38). We then obtain the result (37).
References
[1] U.G. Meiner, Rep. Prog. Phys. 56 (1993) 903.
[2] T. Schafer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323.
[3] D. Diakonov, Chiral quarksoliton model, Lectures given at advanced summer school on nonperturbative
quantum field physics, Peniscola, Spain, June 1997, hep-ph/9802298.
356
[4] D.I. Diakonov, V.Yu. Petrov, Nucl. Phys. B 245 (1984) 259.
[5] S. Weinberg, Phys. Rev. Lett. 18 (1967) 188;
S. Weinberg, Phys. Rev. 166 (1968) 1568.
[6] S. Myint, C. Rebbi, Nucl. Phys. B 421 (1994) 241.
[7] K. Wilson, Phys. Rev. D 10 (1974) 2445.
[8] M. Creutz, Quarks, Gluons and Lattices, Cambridge Univ. Press, Cambridge, 1983.
[9] J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142;
J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
[10] A.R. Levi, V. Lubicz, C. Rebbi, Phys. Rev. D 56 (1997) 1101.
[11] H. Kluberg-Stern et al., Nucl. Phys. B 190 (1981) 504.
[12] N. Kawamoto, J. Smit, Nucl. Phys. B 192 (1981) 100.
[13] A.B. Balantekin, Character expansions, ItzyksonZuber integrals, and the QCD partition function, hepth/0007161.
[14] M.R. Zirnbauer, J. Phys. A 29 (1996) 7113.
[15] M.R. Zirnbauer, The color-flavor transformation and a new approach to quantum chaotic maps, Proceedings
of the XIIth international congress of mathematical physics, Brisbane, July 1997, chao-dyn/9810016.
[16] A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
[17] B. Schlittgen, T. Wettig, Nucl. Phys. B 632 (2002) 155.
[18] R. Howe, Remarks on classical invariant theory, Trans. Amer. Math. Soc. 313 (1989) 539.
[19] A. Perelomov, Generalized Coherent States and Their Applications, Springer-Verlag, Berlin, 1985.
[20] B.L. Ioffe, Nucl. Phys. B 188 (1981) 317.
[21] Chr.V. Christov et al., Prog. Part. Nucl. Phys. 37 (1996) 91.
[22] J. Budczies, Y. Shnir, Color-flavor transformation for the special unitary group and application to low energy
QCD, in: M.N. Dubinin, V.I. Savrin (Eds.), Proceedings of 15th International Workshop on High Energy
Physics and Quantum Field Theory, MSU, Moscow, 2001, p. 317, hep-lat/0101016.
[23] T. Nagao, S.M. Nishigaki, Phys. Rev. D 64 (2001) 014507, hep-lat/0012029.
[24] J. Budczies, Y. Shnir, AIP Conf. Proc. 508 (2000) 172.
[25] L.K. Hua, Harmonic Analysis of Functions of Several Variables in the Classical Domains, Science Press,
1959.

Potential NRQCD and heavy-quarkonium spectrum

at next-to-next-to-next-to-leading order
Bernd A. Kniehl a , Alexander A. Penin a,b , Vladimir A. Smirnov c ,
Matthias Steinhauser a
a II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
b Institute for Nuclear Research, Russian Academy of Sciences, 60th October Anniversary Prospect 7a,
117312 Moscow, Russia

c Institute for Nuclear Physics, Moscow State University, 119899 Moscow, Russia
Received 11 February 2002; accepted 15 May 2002
Abstract
The next-to-next-to-next-to-leading order (N3 LO) Hamiltonian of potential nonrelativistic QCD is
derived. The complete matching of the Hamiltonian and the contribution from the ultrasoft dynamical
gluons relevant for perturbative bound-state calculations is performed including one-, two-, and
three-loop contributions. The threshold expansion is used to disentangle and match contributions of
different scales in the effective-theory calculations. As a physical application, the heavy-quarkonium
spectrum is obtained at N3 LO for the case of vanishing QCD beta function. Our results set the stage
for a full N3 LO analysis of the heavy-quarkonium system. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 12.38.Aw; 12.38.Bx
1. Introduction
The theoretical study of nonrelativistic heavy-quarkantiquark systems [1] and its
applications to bottomonium [2] and topantitop [3] physics rely entirely on first
principles of QCD. These systems allow for a model-independent perturbative treatment.
Nonperturbative effects [4,5] are well under control for the topantitop system and, at
least within the sum-rule approach, also for bottomonium. This makes heavy-quark
antiquark systems an ideal laboratory to determine fundamental parameters of QCD,
E-mail address: msteinh@mail.desy.de (M. Steinhauser).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 0 3 - 0
358
B.A. Kniehl et al. / Nuclear Physics B 635 (2002) 357383
such as the strong-coupling constant s and the heavy-quark masses mq . The study of
t t threshold production should even allow for a precision study of Higgs-boson-induced
effects. Recently, essential progress has been made in the theoretical investigation of the
nonrelativistic heavy-quark threshold dynamics based on the effective-theory approach
[6,7]. In such a framework, one has two expansion parameters, s and the relative velocity
v of the heavy quarks. The corrections are classified by the total power of s and v, i.e.,
Nk LO corrections contain terms of O(sl v m ), with l + m = k. This has the consequence
that, in general, different loop orders, which are counted in powers of s , contribute to the
Nk LO result.
Analytical results for the main parameters of the nonrelativistic heavy-quarkantiquark
system are now available through the next-to-leading order (NLO) and the next-to-nextto-leading order (NNLO) [815]. They have been applied to bottomonium [8,11,13,15
17] and topantitop [12,14,1823] phenomenology. Some specific classes of the nextto-next-to-next-to-leading order (N3 LO) corrections have also been investigated [2427]
(see Ref. [28] for a brief review). These corrections have turned out to be so sizeable
that it appears to be indispensable to gain full control over this order, both with respect to
phenomenological applications and in order to understand the structure and peculiarities of
the nonrelativistic effective theory. Besides its phenomenological importance, the heavyquarkonium system is very interesting from the theoretical point of view because it
possesses a highly sophisticated multiscale dynamics and its study demands the full power
of the effective-theory approach. No qualitatively new theoretical effects are expected
beyond N3 LO, so that the N3 LO analysis would bring us much closer to the full
understanding of the perturbative nonrelativistic dynamics.
In the present paper, we set the stage for the complete N3 LO analysis of perturbative
heavy quarkonium. In particular, we elaborate in detail the nonrelativistic effective
Hamiltonian, which is the key object of the heavy-quarkonium theory, and its matching
to the contribution associated with the emission and absorption of dynamical ultrasoft
gluons, which are relevant for perturbative bound-state calculations [24,29]. To this end,
we employ the technique of Ref. [30] based on the effective-theory approach [6,7,31]
implemented with the threshold expansion [32].
This paper is organized as follows. In Section 2, we introduce the basic ingredients
of the nonrelativistic effective-theory formalism and describe the main features of our
approach. In Section 3, we recall lower-order results and analyze the one- and two-loop
contributions to the N3 LO Hamiltonian. In Section 4, we discuss the matching of the
Hamiltonian to the ultrasoft contribution, which necessitates the inclusion of one-, twoand three-loop contributions. In Section 5, we convert our results into corrections to the
heavy-quarkonium spectrum and illustrate their phenomenological relevance. Section 6
contains our conclusions. In Appendix A, we present some details of our calculation of the
1/mq corrections to the heavy-quark potential.
2. Effective theory of nonrelativistic heavy quarks

The nonrelativistic behaviour of the heavy-quarkantiquark pair is governed by a
complicated multiscale dynamics. In the nonrelativistic regime, where the heavy-quark
359
velocity v is of the order of the strong-coupling constant s , the Coulomb effects are
crucial and have to be taken into account to all orders in s . This makes the use of the
effective theory mandatory. The effective-theory approach allows us to separate the scales
and to implement the expansion in v at the level of the Lagrangian. Let us recall that
the dynamics of a nonrelativistic quarkantiquark pair is characterized by four different
regions and the corresponding modes [32]:
(i) the hard region (the energy and three-momentum scale like mq );
(ii) the soft region (the energy and three-momentum scale like mq v);
(iii) the potential region (the energy scales like mq v 2 , while the three-momentum scales
like mq v); and
(iv) the ultrasoft region (the energy and three-momentum scale like mq v 2 ).
The ultrasoft region is only relevant for gluons, ghosts, and light quarks. Nonrelativistic
QCD (NRQCD) [6,7] is obtained by integrating out the hard modes. Subsequently
integrating out the soft modes and the potential gluons results in the effective theory of
potential NRQCD (pNRQCD) [31], which contains potential heavy quarks and ultrasoft
gluons, ghosts, and light quarks as active particles. The effect of the modes that have
been integrated out is two-fold: higher-dimensional operators appear in the effective
Hamiltonian, corresponding to an expansion in v, and the Wilson coefficients of the
operators in the effective Hamiltonian acquire corrections, which are series in s .
The theory of pNRQCD is relevant for the description of the heavy-quarkonium system.
Let us recall its basic ingredients. In pNRQCD, the (self)interactions between ultrasoft
particles are described by the standard QCD Lagrangian. The interactions of the ultrasoft
gluons with the heavy-quarkantiquark pair are ordered in v by the multipole expansion.
For the N3 LO analysis, only the leading-order (LO) emission and absorption of ultrasoft
gluons have to be considered. They are described by the chromoelectric dipole interaction,
which is of the form gs r E, where gs is the QCD gauge coupling, r is the difference
of coordinates of the quark and antiquark, and E = E a t a is the chromoelectric field,
with t a being the generators of the colour gauge group SU(3). The propagation of the
quarkantiquark pair in the colour-singlet (s) and colour-octet (o) states is described by
the nonrelativistic Green function Gs,o of the Schrdinger equation,
s,o

H E Gs,o r, r , E = r r ,
(1)
where E is the energy of the quarkantiquark pair counted from the threshold 2mq and
Hs,o is the effective nonrelativistic Hamiltonian,
s,o
+ ,
Hs,o = HC
s,o
HC
=
r
+ VCs,o (r),
mq
(2)
with r = r2 and r = |r|. The ellipses stand for higher-order terms in s and v. The
Coulomb (C) potentials for the singlet and octet states are attractive and repulsive,
respectively, and are given by

s
s
CA
VCs (r) = CF ,
(3)
VCo (r) =
CF
,
r
2
r
360
where CA = 3 and CF = 4/3 are the eigenvalues of the quadratic Casimir operators
of the adjoint and fundamental representations of the colour gauge group, respectively.
Throughout this paper, we assume that s = s () if no argument is specified.
The LO approximation for the Green function is given by the Coulomb solution,
which sums up terms singular at threshold and describes the leading binding effects.
The corrections to the Coulomb Green function due to higher-order terms in the effective
Hamiltonian can be found in RayleighSchrdinger time-independent perturbation theory
as in standard quantum mechanics. The Green functions have the following spectral
representations:

d3 k ks (r)ks (r )
ns (r)ns (r )
+
,
Gs r, r , E =
(4)
En E
(2)3 k 2 /mq E
n=1

d3 k ko (r)ko (r )
Go r, r , E =
(5)
,
(2)3 k 2 /mq E
where ns and ks,o are the wave functions of the quarkantiquark bound and continuum
states, with principal quantum number n and relative three-momentum k, respectively, and
the E + i rule is implied. In Eqs. (4) and (5), the orbital and spin quantum numbers,
l and m, respectively, are suppressed. Note that a discrete part of the spectrum (bound
states) only exists for the singlet Green function. Through the emission or absorption of
an ultrasoft gluon, the quarkantiquark pair changes its colour state, so that one switches
from Eq. (4) to Eq. (5) and vice versa.
Let us now turn to the problem of perturbative calculations in the effective theory.
Both NRQCD and pNRQCD have specific Feynman rules, which can be used for a
systematic perturbative expansion. However, this is complicated because the expansion
of the Lagrangian corresponds to a particular subspace of the total phase space. Thus, in
a perturbative calculation within the effective theory, one has to formally impose some
restrictions on the allowed values of the virtual momenta (see, e.g., Refs. [33,34], for
examples of highly sophisticated calculations performed in this scheme).
Explicitly separating the phase space introduces additional scales to the problem, such
as momentum cutoffs, and makes the approach considerably less transparent. A much more
efficient and elegant method is based on the expansion by regions [32,35], which is a
systematic method to expand Feynman diagrams in any limit of momenta and masses. It
consists of the following steps:
(i) consider various regions of a loop four-momentum k and expand, in every region,
the integrand in Taylor series with respect to the parameters that are considered to be
small there;
(ii) integrate the expanded integrand over the whole integration domain of the loop
momenta; and
(iii) put to zero any scaleless integral.
In step (ii), dimensional regularization, with d = 4 2! spacetime dimensions, is used to
handle the divergences. In the case of the threshold expansion in v, one has to deal with
the four regions and their scaling rules listed above.
361
In principle, the threshold expansion has to be applied to the Feynman diagrams of full
QCD. However, after integrating out the hard modes, which corresponds to calculating the
hard-region contributions in the threshold expansion, it is possible to apply step (i) to the
diagrams constructed from the NRQCD and pNRQCD Feynman rules [30]. Equivalently,
the Lagrangian of the effective theory can be employed for a perturbative calculation
without explicit restrictions on the virtual momenta if dimensional regularization is used
and the formal expressions derived from the Feynman rules of the effective theory are
understood in the sense of the threshold expansion. In this way, one arrives at a formulation
of effective theory with two crucial virtues: the absence of additional regulator scales and
the automatic matching of the contributions from different scales. The second property
implies that the contributions of different modes, as computed in the effective theory, can be
simply added up to get the full result. This automatic-matching property of effective-theory
calculations in dimensional regularization was observed in Ref. [36] and used for highorder calculations in the theory of QED bound states in Ref. [37]. We should emphasize,
however, the crucial rle of the threshold expansion in effective-theory perturbative
calculations because, in general, the nave use of the effective-theory Feynman rules and
dimensional regularization leads to incorrect results. Another advantage of the effectivetheory realization of the threshold expansion is that the individual contributions from
the hard, soft/potential, and ultrasoft regions are manifestly gauge invariant. Indeed, the
Lagrangians of NRQCD and pNRQCD are gauge invariant, and dimensional regularization
preserves the gauge symmetry as well. Thus, the QCD calculation of the hard corrections
and the NRQCD calculation of the soft and potential corrections to the on-shell amplitudes
can be performed in the covariant gauge suitable for relativistic problems, while the
pNRQCD calculations can be done in the Coulomb gauge appropriate for nonrelativistic
problems. In the next sections, we illustrate the power of the approach outlined above by
an explicit analysis in pNRQCD at N3 LO.
3. Nonrelativistic effective Hamiltonian

Let us start this section with a general remark on the structure of higher-order
corrections in pNRQCD. As already mentioned in the Introduction, the corrections are
classified by the total power of s , counting the number of loops, and v or, equivalently,
1/mq . In particular, the N3 LO Hamiltonian includes one-loop corrections of O(s v 2 ),
two-loop corrections of O(s2 v), and three-loop corrections of O(s3 ). Note that purely
relativistic corrections of O(v 3 ) are absent [36]. In NLO, the only source of corrections
is the renormalization and running of the Coulomb potential. In NNLO, relativistic
corrections in v due to higher-dimensional operators start to contribute. In N3 LO,
retardation effects, which cannot be described by operators of instantaneous interaction,
enter the game. They will be discussed in the next section. At this order, the nonrelativistic
effective Hamiltonian becomes infrared (IR) sensitive. No qualitatively new theoretical
effects are expected beyond N3 LO.
362
The general form of the Hamiltonian valid up to N3 LO reads

2

p
p4
+ Cc (s )VC |q| + C1/m (s )V1/m |q|
H = (2)3 (q)
3
mq
4mq

p 2 + p 2
CF s
C
(
)
+
C
(
)
+ Cs (s )S 2
+
s
p
s
m2q
2q 2

+ C (s )(p, q) + Ct (s )T (q) ,
(6)
where the following operators are involved

4CF s
VC |q| =
,
q2
S (p q)
(p, q) = i
,
q2
2 2 CF s
1 + 2
,
S=
,
V1/m |q| =
mq |q|
2
(q 1 )(q 2 )
T (q) = 1 2 3
.
q2
(7)
Here, p and p are the three-momenta of the incoming and outgoing quarks, respectively,
q = p p is the three-momentum transfer, and 1,2 are the quark and antiquark spin
operators. Note that the effective Hamiltonian is defined for on-shell quarks, with p 2 =
p 2 = mq E. The Wilson coefficients are power series in s ,

s n i
cn (mq , |q|, ),
Ci (s ) =
(8)
n=0
where the modified minimal-subtraction (MS) scheme for the renormalization of s is

implied.
In the following, we discuss the nontrivial terms in Eq. (6), which are sorted in terms of
the inverse heavy-quark mass. The contribution from the hard-virtual-momentum region
is analytic in v 2 and starts at O(v 2 ). Thus, it does not affect VC (|q|) and V1/m (|q|).
Corrections to the static Coulomb potential only arise from the soft contribution. Using
renormalization group (RG) arguments, they can be rewritten as

Cc (s )VC |q|

s (|q|)
s (|q|) 2
4CF s (|q|)
1+
a1 +
a2
=
q2
4
4

s (|q|) 3
2
2 3
(9)
a3 + 8 CA ln 2 + .
+
4
q
The RG logarithms can be recovered from Eq. (9) by recalling that (see, e.g., Ref. [38])

s (|q|) s ()
s ()
s () 2 2
=
1+
0 L +
L 0 L + 1
3

5
s ()
3 2
L 0 L + 0 1 L + 2 + ,
+
(10)
where L = ln(2 /q 2 ) and [39]

1 11
4
CA TF nl ,
0 =
4 3
3

1 34 2 20
C CA TF nl 4CF TF nl ,
1 =
16 3 A
3

1 2857 3 1415 2
205
CA
CA TF nl
CA CF TF nl + 2CF2 TF nl
2 =
64 54
27
9

158
44
2 2
2 2
CA TF nl + CF TF nl
+
27
9
363
(11)
are the first three coefficients of the QCD beta function. Here, TF = 1/2 is the index of
the fundamental representation, and nl is the number of light-quark flavours. The nonRG logarithmic term of O(s3 ) in Eq. (9) reflects the IR divergence of the static potential
[40]. The corresponding pole is canceled against the ultraviolet (UV) one of the ultrasoft
contribution [24,41]. For convenience, this pole is subtracted in Eq. (9) according to the
MS prescription, so that the coefficient a3 is defined in the MS subtraction scheme both
for UV and IR divergences. For consistency, the UV pole of the ultrasoft contribution has
to be subtracted in the same way. Throughout the calculation, we use the same procedure
to render the contributions from the various regions finite. The actual cancellation of the
spurious divergences appearing in the process of expanding by regions is reflected in the
independence of the final result.
In the literature [24,40,41], the coefficient in front of the IR logarithm in the O(s3 )
static potential is given as CA3 /(24), which differs from CA3 /(8) in Eq. (9). This is
a consequence of the consistent use of dimensional regularization in our analysis based
on the threshold expansion. The difference is due to the fact that we perform all three
loop integrals in d dimensions, not just the one that is IR divergent. The logarithmic
terms not associated with IR-divergent integrals are unphysical and are exactly canceled
by similar terms from the three-loop ultrasoftpotentialpotential contribution, in which
only the ultrasoft integral is UV divergent (see Section 4), while the physical logarithmic
integral between soft and ultrasoft scales results in ln s corrections to the spectrum. The
calculation of the coefficients ai can be performed in the static limit mq of NRQCD.
Due to the exponentiation of the static potential [42], these coefficients only receive
contributions form the maximum non-Abelian parts. In the language of the threshold
expansion, the selection of these parts effectively separates the contribution of the soft
region. The Abelian colour factor CF indicates the presence of the Coulomb singularity
and implies that at least one loop momentum is potential. All such contributions are just
iterations of the lower-order potential and are taken into account in the perturbative solution
of the Schrdinger equation (1) around the Coulomb approximation.
The one-loop coefficient,
31
20
CA TF nl ,
(12)
9
9
has been known for a long time [42,43], while the two-loop coefficient has only recently
been found [44,45]. In our previous communication [30], we confirmed the result of
a1 =
364
Ref. [45],

4343
4 22
1798 56
2
2
+ 4
+ (3) CA
+ (3) CA TF nl
a2 =
162
4
3
81
3

2
55
20
16 (3) CF TF nl +
TF nl ,
3
9

(13)
where is Riemanns zeta function, with value (3) = 1.202057 . . . . At present, only Pad
estimates of the tree-loop MS coefficient are available, namely, [46]

142 if nl = 3,
a3
(14)
=
98
if nl = 4,
43
60 if n = 5.
l
Although the Pad estimates are in reasonable agreement with the exact results where the
latter are available, the reliability of Eq. (14) is not guaranteed, and it is very desirable
to exactly evaluate the coefficient a3 . However, in Section 5, we show that even a 100%
uncertainty in a3 would not result in a significant error in the N3 LO corrections to the
spectrum for the states with small principal quantum number.
The 1/mq -suppressed terms of Eq. (6) receive contributions from the soft and potential
regions and, applying RG techniques, can be written in the following form:

C1/m (s )V1/m |q|

2
2 CF s2 (|q|)
s (|q|)
4
=
(15)
b1 +
b2 CA2 + 2CA CF ln 2 + .
mq |q|
3
q
The one-loop coefficient b1 reads [47,48]
CF
.
(16)
2
Our result [30] for the two-loop coefficient b2 is given by Eq. (31) in Section 3.2, where
also a detailed description of its derivation is presented. A more detailed description of
the calculation is presented in Section 3.2. The coefficient of the two-loop IR logarithm
in Eq. (15) can be extracted from the UV divergence of the ultrasoft contribution [24].
However, similarly to the case of the IR divergence of the static potential, one has to take
into account additional logarithmic terms resulting from the consistent use of dimensional
regularization.
The RG analysis results in the following representation of the 1/m2q part of the effective
Hamiltonian:

s () 4
2
d1 + (CA 2CF ) ln 2 + ,
s ()C (s ) = s |q| d0 +
3
q

p s () p 8
2
s ()Cp (s ) = s |q| d0 +
d1 CA ln 2 + ,
3
q

i s () i
d1 + , i = s, , t,
s ()Ci (s ) = s |q| d0 +

s (mq ) i,a
s ()Cia (s ) = s (mq ) d0i,a +
(17)
d1 + , i = , s,
b1 = CA +
365
where the contributions from the annihilation channel are marked by the superscript a. The
normalization scale of s in the one-loop scattering terms is not fixed because they receive
contributions from both the soft and hard regions. According to the RG, the scale of s
should be chosen to be h mq for the hard contribution and s s mq for the soft one.
The IR one-loop logarithms which match the UV behaviour of the ultrasoft contribution
[24] are written out explicitly in Eq. (17). The calculation of the one-loop coefficients
d1i within the threshold expansion is discussed in Section 3.1. Some of these coefficients
also contain logarithms of the form ln(m2q /q 2 ) originating from the logarithmic integration
between the soft and hard scales.
The purely relativistic tree-level O(v 2 ) corrections are given, up to a colour factor, by
the standard Breit Hamiltonian and read
d0 = 0,
d0,a = 0,
d0 = 4,
d0s,a = 0.
4
d0s = ,
3
d0 = 6,
1
d0t = ,
3
(18)
The QED effective Hamiltonian for nl light fermions is obtained from the above
expressions by setting CA = 0 and CF = TF = 1. Note that the QED Breit Hamiltonian
has a nonvanishing one-photon annihilation coefficient,
s,a
= 1,
d0,QED
(19)
which is absent in the case of colour-singlet quarkonium due to colour conservation.

In Refs. [36,49], dealing with QED bound-state calculations in the Coulomb gauge, one
finds Wilson coefficients different from the Abelian parts of those listed in Eqs. (16) and
(18). Using them would lead to
b1 = CA ,
d0 = 1.
(20)
These two sets of coefficients are equivalent, and the difference is related to the use of offshell operators in the Hamiltonian. This problem is discussed in more detail in Section 3.2.
3.1. One-loop operators
The O(s v 2 ) operators, contributing to N3 LO at one loop, have attracted some attention,
and an essential part of the results can be found in the literature [4754]. Here, we present
a consistent derivation of these corrections within the threshold-expansion framework.
The O(s v 2 ) operators receive contributions from the hard and soft/potential regions.
The contribution from the hard region requires a fully relativistic treatment. A part of
it is directly related to the Wilson coefficients cF , cD , cS , and d2 parameterizing the
Fermi, Darwin, spinorbital, and heavy-quark vacuum-polarization terms in the NRQCD
Lagrangian, respectively [52]. The residual part arises from the on-shell scattering and
annihilation box diagrams at threshold [53].
The calculation of the soft contribution can be performed in NRQCD. In the effectivetheory language, we study the reduction from NRQCD to pNRQCD and compute the effect
of the soft modes being integrated out. Apart from the standard LO terms of the NRQCD
Lagrangian, we need the 1/mq -suppressed terms originating from the covariant-derivative
366
operator D 2 /(2mq ) acting on the quark and antiquark fields, the Fermi, Darwin, and spin
orbital terms. Note that the covariant-derivative operator includes the quark kinetic-energy
term k 2 /(2mq ). According to the threshold expansion, it should be treated as a perturbation
if k is soft or kept in the nonrelativistic quark propagator,
S(k) =
1
k0 k 2 /(2mq ) + i
(21)
if k is potential. The potential region is connected with the contribution of the pole of
the propagator of Eq. (21) to the integral over k0 . In this connection, we can make an
interesting observation. Let us consider two-particle-irreducible diagrams that are free of
Coulomb singularities, so that the k0 contour is not pinched between the poles of the quark
and antiquark propagators. In this case, one can close the contour of the k0 integral keeping
the poles either inside or outside. The contribution from the potential region is obviously
different in these cases, although the result for the integral is the same. This means that
separating the contributions from the soft and potential regions for diagrams without
Coulomb singularity is useless, and the propagator of Eq. (21) can thus be safely expanded
in 1/mq in both regions. In fact, the expansion of the pole contribution yields familiar
generalized functions of k0 , namely (n) (k0 ). This observation dramatically simplifies the
calculation, which can be performed in a covariant form after substituting k0 = v0 k, where
v0 = (1, 0).
By contrast, the two-particle-reducible diagrams including the product of the quark and
antiquark propagators,
1
k0 k 2 /(2mq ) + i k0 + k 2 /(2mq ) i
(22)
where k is the two-particle-reducible loop four-momentum, suffer from a Coulomb

singularity. In this case, after expanding these propagators, one obtains ill-defined pinched
products like
1
1
.
(k0 + i)m (k0 i)n
(23)
Thus, separating the soft and potential regions is unavoidable. In the soft region, the pole
contributions of the quark and antiquark propagators have to be excluded, and the above
product should be defined to be its principal value,

1
1
1
(24)
+
,
2 (k0 + i)m+n (k0 i)m+n
which again allows for a covariant treatment. In the potential region, the quark and
antiquark propagator poles produce contributions of the form

mq
k2
k2
+ k0 +
,
k0
i 2
(25)
2mq
2mq
k i
where the 1/v Coulomb singularity shows up explicitly. After integration over k0 , Eq. (25)
yields the nonrelativistic Green function of the free Schrdinger equation. A contribution
of this type can always be related to iterations of the operators of the effective Hamiltonian
367
in time-independent perturbation theory. Therefore, it should not be considered as a

correction to the Hamiltonian. However, there is one subtle question here, namely as to
whether the operators that vanish for on-shell quarks should be included in the effective
Hamiltonian or not. This does not affect the 1/m2q potential, but it matters for the 1/mq
operator discussed in Section 3.2.
One has to be careful with the definition of commutators of the Dirac/Pauli matrices
within dimensional regularization. Since poles in ! are only present in the individual
contributions from the hard and soft regions and drop out in the sum, one can explicitly
retain the commutators during the analysis of the these regions and replace them by the
four/three-dimensional expressions in the final result [37]. Otherwise one has to use the
same prescription for the evaluation of the commutators of the Dirac/Pauli matrices in
the hard and soft regions. Throughout the calculation, we use the four/three-dimensional
antisymmetric ! tensor for the definition of the commutators as was done in Ref. [54].
(This differs from the prescription of Ref. [53].)
The one-loop calculation poses no technical problems, and our results for the Wilson
coefficients read
4
2
d1 + (CA 2CF ) ln 2
3
q

1 17 2
5 1 2
4
ln 2 CA +
ln
=
CF TF
4
6
mq
3 3 m2q
15
h

2
2
9 25
1 7
+
ln 2 CA +
ln
+
CF ,
4
6
q
3 3 q2
s

2
2
8
31 8
20
p
d1 CA ln 2 = ln 2 CA + TF nl ,
3
q
9
3 q
9
s

2
4 7
2
14 7 2
20
+ ln 2 CA CF + ln 2 CA TF nl ,
d1s =
3 6 mq
3
27 6 q
27
h
s

2
2
7
10
2 ln 2 CA TF nl ,
d1 = 4 + 2 ln 2 CA + 4CF +
6
3
mq
q
h
s

2
2
1
1 1
1
13
5
+ ln 2 CA + CF +
ln 2 CA TF nl ,
d1t =
3 6 mq
3
108 6 q
27
h
s
d1,a = (4 + 4 ln 2 i2)TF ,
d1s,a = (2 2 ln 2 + i)TF ,
(26)
where the contributions from the hard (h) and soft (s) regions are explicitly separated.
The first two equations of Eq. (26) are written in a way appropriate for Eq. (17). As in
NNLO, the one-photon-annihilation channel provides an additional contribution to the
QED Wilson coefficient, namely

nl
5
s,a
2 ln 2 + i
.
d1,QED
(27)
= 2 2 ln 2 + i
3
3
368
The QED part of Eq. (26) is in agreement with Ref. [31]. The non-Abelian part of Eq. (26)
agrees with Ref. [54], including the nonlogarithmic CA term in the coefficient d1 , which
differs from the one of Ref. [48].
3.2. Two-loop operators
The O(s2 v) part of the effective Hamiltonian is given by the two-loop corrections to the
1/mq potential. These corrections are solely generated by the covariant-derivative operator
in the NRQCD Hamiltonian. The calculation of the two-loop 1/mq potential is simplified
by the absence of the hard contribution, but, in turn, it is complicated by the presence of
off-shell operators.
To introduce the problem, let us start with the one-loop 1/mq potential. For illustrative
purposes, we use the Feynman gauge, where the Coulomb and transverse gluons do not
mix. Nonzero contributions come from the planar and crossed box diagrams and a diagram
with one three-gluon vertex. The last two diagrams do not contain Coulomb singularities.
According to the procedure described in the previous section, the quark and antiquark
propagators should be expanded in 1/mq . The result reads

2 s2
.
CA CF CF2
mq |q|
(28)
The soft contribution to the planar box diagram vanishes because the 1/mq terms from
the expansion of the quark and antiquark propagators cancel. The potential contribution
corresponds to the second iteration of the operators generated by the exchange of one
potential gluon. The operators that are defined for on-shell quarks enter the effective
Hamiltonian, and their iteration is taken into account when Eq. (1) is solved. However,
the potential-gluon exchange also generates operators proportional to the energy transfer
q0 = (p 2 p2 )/(2mq ), which vanish for on-shell quarks. Such operators do not enter the
effective Hamiltonian, which is defined on-shell, and their iterations should be considered
as corrections to the effective Hamiltonian. The factor q0 cancels the denominator of
the free nonrelativistic Green function and effectively makes the diagram two-particle
irreducible. The Coulomb singularity brings a factor of mq , so that for the calculation of
the 1/mq corrections we need the iteration of the 1/m2q off-shell operator and the leading
Coulomb potential. In the case under consideration, the relevant off-shell operator is

CF s p 2 p2 2
(29)
.
m2q
q2
Explicit evaluation of the corresponding potential contribution yields
2 CF2 s2
,
2mq |q|
(30)
and the coefficients of Eqs. (28) and (30) sum up to Eq. (16). In fact, by using the Coulomb
equation of motion, it is straightforward to check that the matrix elements of Eqs. (29) and
(30) between Coulomb states are the same.
Note that, in QED calculations performed in the Coulomb gauge [36,49], the off-shell
tree-level operator analogous to Eq. (29) naturally appears. If one includes this operator in
369
the effective Hamiltonian, the Wilson coefficient d0 is given by Eq. (20) and the coefficient
b1 is purely non-Abelian.
The equivalence of the two formulations is obvious from the above analysis. The use of
off-shell operators is advantageous in QED because it allows one to reduce the number of
loops by means of the Coulomb equation of motion, as may be seen by comparing Eqs. (29)
and (30). However, here we use the on-shell formulation and the general covariant gauge,
which is more suitable for multiloop QCD calculations.
The structure of the expansion remains intact at two loops, and our final result reads [30]

101 4
65 8
49
+ ln 2 CA2 +
ln 2 CA CF + CA TF nl
b2 =
36
3
18 3
36
2
CF TF nl .
(31)
9
There is no fully Abelian CF3 contribution in b2 , as is known from the QED analysis [36,
49]. In the calculation of the two-loop two-particle-irreducible diagrams, which completely
determine the maximum non-Abelian CA2 CF and CA CF TF nl structures of the result, we
used an expanded form of Eq. (21). Typical diagrams contributing to the CA2 CF part are
depicted in Fig. 1. The analysis of the relevant two-particle-reducible diagrams, shown in
Fig. 2, is conceptually similar to the one-loop case described above. Let us discuss it in
more detail. The reducible loop momentum can be either soft or potential. If it is soft, the
quark and antiquark propagators can be expanded, and the only nonzero contribution corresponds to the situation where the single gluon is the Coulomb one and the 1/mq term is
kept in the expansion of the one-loop block, B. If the reducible loop momentum is potential, then one only has to take into account contributions from the off-shell operator. There
are two possibilities: (i) the off-shell operator is generated by the single-gluon exchange,
and the block B stands for the one-loop corrections to the Coulomb potential; or (ii) the
Fig. 1. Examples of two-particle-irreducible two-loop diagrams. The standard quarkgluon vertex represents the
leading Coulomb interaction. The black circles correspond to the three types of 1/mq terms generated by the
quark covariant-derivative term.
Fig. 2. Example of a two-particle-reducible two-loop diagram. B stands a general one-loop two-particleirreducible subgraph. The threshold expansion of this diagram is discussed in the text.
370
off-shell operator comes from the 1/m2q part of the one-loop block B, and the single gluon
is the Coulomb one. The analysis of the potential contribution is rather straightforward
from the technical point of view. The calculation of the two-particle-irreducible diagrams
and the soft parts of the two-particle-reducible diagrams is more involved. Some details
are presented in Appendix A.
We performed a number of nontrivial checks for our analysis. (i) We worked in the
general covariant gauge and verified that the gauge parameter cancels in our final result.
(ii) The two-loop expression from which Eq. (31) is obtained contains both UV and IR
divergences. The UV ones were removed in Eq. (31) by the renormalization of s in
the one-loop result of Eq. (15). On the other hand, the IR divergences were canceled by
the UV ones of the ultrasoft contribution (see Section 4) leaving a finite result for the
spectrum. The RG logarithms proportional to 0 and the IR logarithms are in agreement
with Eq. (15). (iii) To test our program, we also recalculated the two-loop corrections to the
static heavy-quarkantiquark potential and found agreement with Ref. [45]. Note that, with
our prescription for the calculation of the soft and potential contributions, we explicitly
obtained zero for the partially Abelian corrections to the static potential.
4. Ultrasoft contribution
For the N3 LO Hamiltonian, only the leading retardation effects are needed. They arise
from the chromoelectric dipole interaction of the heavy quarkonium with a virtual ultrasoft
gluon, as depicted in Fig. 3. In the analysis of the ultrasoft contribution, we proceed along
the lines of the original analysis [24]. As has been mentioned in Section 2, there is freedom
in the choice of gauge. We work in the Coulomb gauge, which is especially appropriate for
N3 LO calculations in pNRQCD because the Coulomb gluon does not propagate and the
dynamical gluon is transverse.
The analytical expression for the corrections to the singlet Coulomb wave function can
be obtained by using the pNRQCD Feynman rules of Section 2. It reads

(x) (y)

m
n
s
Jmn (E),
G (x, y, E) =
(32)
(Em E)(En E)
m n
where
Jmn (E) = CF gs2 (us )

k2
d3 k
ij
E
,
r

r

I
j mk i kn
mq
(2)3
(33)
Fig. 3. Feynman diagram giving rise to the ultrasoft contribution at N3 LO. The shaded and light double lines stand
for the singlet and octet Green functions, respectively. The loopy line represents the ultrasoft-gluon propagator in
the Coulomb gauge, and the black circles correspond to the chromoelectric dipole interaction.
with

I (t) = i
ij
dd l l02 ( ij l i l j /l 2 )
.
(2)d
l 2 (t l 0 )
371
(34)
The sum/integral in Eq. (32) goes over the whole spectrum, and m and n stand for
the complete set of quantum numbers characterizing the discrete/continuum part of the
spectrum. The scale us s2 mq of gs in Eq. (33) reflects the ultrasoft-momentum flow
through the gluon propagator. The matrix element rkn is taken between the singlet
Coulomb wave function of quantum number n and the octet Coulomb wave function
of three-momentum k. Performing in Eq. (34) the integration over l0 , we recover the
expression of time-independent nonrelativistic perturbation theory. The remaining integral
over l is UV divergent. Subtracting the UV pole according to the MS prescription, we
obtain

|E1C |
5
k2
ij
k2 3
, (35)
ln
2
+
+
ln
I ij E
=
E
ln
mq
6 2
mq
|E1C | 6
k 2 /mq E
where, for convenience, the Coulomb energy,
CF2 s2 mq
,
(36)
4n2
with n = 1, has been introduced into the arguments of the logarithms. The k-dependent
logarithmic term in Eq. (35) represents a pure retardation effect and cannot be interpreted
in terms of some instantaneous interaction. It receives contributions from Coulomb-gluonexchange diagrams of all orders and leads to a QCD analogue of the familiar Bethe
logarithms in the corrections to the spectrum, to be discussed in the next section. On the
other hand, making use of the completeness relation,

k2 l
d3 k
o l
r

r

= r E HC
r mn ,
E
(37)
j mk i kn
3
(2)
mq
EnC =
the remaining part of Eq. (35), excluding the k-dependent logarithmic term, is reduced to
an instantaneous interaction of the form

o 3
r mn
r E HC

3
s2
C 3
s
= A s CA2 + 2CA CF
+ 4(CA 2CF ) 2 (r)
2
8 r
mq r
mq

s
1
+ reducible part,
+ CA 2 r ,
(38)
mq
r mn
s
acting directly on a wave
where the reducible part includes terms with the operator E HC
function. By using Eqs. (32), (35), (37) and (38), one arrives at the following representation
of the corrections to the Green function:

Gs r, r , E = d3 r GsC r, r , E Hus r GsC r , r , E
+ contact terms,
(39)
372
where, in momentum space,

2
CF s 1
5
ln C + ln 2
Hus =
3
2 (E1 )2 6

s2

s
3
CA3 s2 + 4 CA2 + 2CA CF
8(CA 2CF ) 2
q
mq |q|
mq

2

2
s p + p
+ 16CA 2
.
mq
2q 2
(40)
The first term in Eq. (39) imitates the corrections to the Green function due to the N3 LO
term of Eq. (40) in the effective Hamiltonian. The contact terms correspond to the reducible
part of Eq. (38). By the equation of motion (1), one or both Coulomb Green functions in
Eq. (39) are converted into functions. The corresponding contribution cannot be imitated
by a term of the effective Hamiltonian. It does not affect the energy levels, but it leads
to corrections to the wave functions. A part of such corrections, namely the on-shell
renormalization of the heavy-quarkonium wave function at the origin, was computed in
Ref. [24]. In the present paper, we refrain from discussing this type of corrections.
Although dimensional regularization was used in deriving Eq. (40), the latter is not
consistent with the threshold expansion because the three-dimensional expression for the
Coulomb Green function was used instead of the (d 1)-dimensional one, which is not
available. Thus, Eq. (40) derived in Ref. [24] requires some additional matching. For
this purpose, we separate the divergent contributions from the diagram of Fig. 3 and
compute them according to the threshold expansion. Eq. (40) implies that only the one-,
two-, and three-loop contributions encoded in the diagram of Fig. 3 are divergent. They
include one divergent ultrasoft loop integration and zero, one, or two convergent potential
loop integrations. The three-dimensional form of the Coulomb Green function used in the
evaluation above implies that the integrations over the potential momenta are performed in
three dimensions, while, for the corresponding contributions obtained within the threshold
expansion, they are done in d 1 dimensions. Thus, the matching term is given by the
difference of the diagrams computed in d 1 dimensions and the same diagrams with
three-dimensional integrals over the potential momenta in the limit ! 0. Only two- and
three-loop diagrams have to be considered. Examples of such diagrams are presented in
Fig. 4. The calculation is simplified by the fact that, for the matching, we only need the
pole part of the ultrasoft integral, which factorizes. For the two-particle-reducible diagrams
Fig. 4. Examples of two- and three-loop diagrams encoded in the diagram of Fig. 3, which require additional
matching to bring Eq. (40) in agreement with the threshold expansion. The dashed and loopy lines represent the
potential (Coulomb) and ultrasoft (transverse) gluon propagators in the Coulomb gauge, respectively. The black
circles correspond to the interaction generated by the quark covariant-derivative term.
373
of the type shown in Fig. 2, one has to take into account the off-shell operators generated
by the two-particle-irreducible block B with one ultrasoft loop in a way similar to the case
of the 1/mq potential in Section 3.2. In addition to the O(s v 2 ) operator proportional to
Eq. (29) with an extra factor of s , the one-loop ultrasoft exchange generates an off-shell
operator proportional to
CF s p2 + p 2 2mq E
,
m2q
q2
(41)
which, in three dimensions after adding one extra Coulomb gluon, results in the same 1/mq
term as the operator of Eq. (29). The matching terms are found to be

CF s
2
2 3 s3
us
H =
ln 2 CA 2 + 1 + 4 ln 2 + 2 ln 2 CA2
3
q
q
q

2
s2
4 2 2 ln 2 ln 2 CA CF
(42)
.
q
mq |q|
Incidentally, the three-loop coefficient does not have a constant term. The dependence
of the ultrasoft contribution Hus + Hus exactly cancels the dependence of the N3 LO
Hamiltonian given in Eqs. (9), (15) and (17) ensuring the cancellation of IR and UV poles.
5. Heavy-quarkonium spectrum
Let us now apply the results of the previous sections to the analysis of the heavyquarkonium spectrum. We restrict the analysis to the perturbative corrections, neglecting
issues like nonperturbative contributions in the case of bottomonium and finite-width
effects in the case of the topantitop system. This is justified because the problem of large
perturbative corrections seems to be crucial for the heavy-quarkonium theory. Furthermore,
we only consider the zero-orbital-momentum states, with l = 0, which are of primary
phenomenological interest.
5.1. Perturbative s5 mq heavy-quarkonium spectrum
The O(s3 ) corrections to the energy levels arise from several sources:
(i) matrix elements of the N3 LO operators of the effective Hamiltonian between Coulomb wave functions;
(ii) higher iterations of the NLO and NNLO operators of the effective Hamiltonian in
time-independent perturbation theory;
(iii) matrix elements of the N3 LO instantaneous operators generated by the emission and
absorption of ultrasoft gluons; and
(iv) retarded ultrasoft contribution.
Parts (i) and (ii) include corrections due to the running of s in the lower-order operators
of the effective Hamiltonian proportional to i , with i = 0, 1, 2. The logarithmic part of
374
these corrections can be taken into account by choosing the relevant soft normalization
scale of s in the NNLO result for the spectrum to be s s mq . However, there are
also nonlogarithmic corrections proportional to the QCD beta function. We postpone the
calculation of these corrections to a future publication. Here, we focus our attention on
the conceptually interesting non-RG corrections. In the absence of the running of s ,
the calculation of part (ii) is reduced to a redefinition of s in the leading Coulomb
approximation. The matrix elements relevant for parts (i) and (iii) are conveniently
evaluated in coordinate space. All necessary formulae, including Fourier transforms, can be
found in Ref. [48]. Part (iv) corresponds to the k-dependent term of Eq. (35). The ultrasoft
corrections to the nth energy level are given by Jnn (En ), and its retarded part can be written
as
2CF3 s3 C E
En Ln ,
(43)
3
where we have introduced the QCD Bethe logarithms [24]

2 C
E1C
1
k2 3
d3 k

LE
(44)
r
E
=
ln
.
kn
n
n
mq
(2)3
CF2 s2 EnC
EnC k 2 /mq
The latter can be reduced to one-parameter integrals of elementary functions [24].1 For the
readers convenience, we list the relevant formulae here. They read

LE
n
d YnE ()Xn2 (),
(45)
where
n2 2
26 n5 ( 2 + 1) exp[4 arccot(/n )]
ln
,
n2 ( 2 + n2 )3 [exp(2) 1]
2 + n2
X1 () = 1 + 2,
2 (222 + 92 + 8) 22 (2 + 4)
,
X2 () =
( 2 + 22 )
YnE () =
X3 () =
4 (833 + 6032 + 1233 + 66) 2 2 32 (632 + 413 + 54) + 334 (3 + 6)

3( 2 + 32 )2
(46)
with

CA
n
1 = .
n = n
(47)
2CF
8
The expressions for Xn with n > 3 are usually irrelevant for practical applications. For
n = 1, 2, 3, we obtain the following numerical values:
LE
1 = 81.5379,
LE
2 = 37.6710,
LE
3 = 22.4818.
(48)
1 There are two misprints in Ref. [24]: in Eq. (A.3), n should be in the numerator; in Eq. (A.5), arctan should
be replaced by arccot.
375
Putting everything together and writing

En = EnC + En(1) + En(2) + En(3) + ,
(49)
N3 LO
corrections to the heavy-quarkonium energy levels

we obtain our final result for the
in the approximation of putting (s ) = 0:

En(3) (s )=0

C s3
7
9
S(S + 1)
CA CF
a1
a1 a2 + a3

+ +
+
= En
+
CF2
2
32
2
4 16n
2
n

1
5
+ ln 2 E ln n 1 (n + 1) + Ls CA3
+
36 6

CA2 CF
97 4
+ + ln 2 + E ln n + 1 (n + 1) + Ls
36 3
n

41
7
139
+ 4 ln 2 + E ln n + 1 (n + 1) + Ls
+
36
6
6

1
47 2
+ ln 2 + E + ln n + 1 (n + 1) Ls
+
24 3
n

CA CF2
7
7
107
+ E ln n + 1 (n + 1) Ls S(S + 1)
+
108 12n 6
n

7
8
7
79
S(S + 1) CF3
+ ln 2 + E ln n + 1 (n + 1) + 3Ls
+
18 6n 3
3
3
n
2

C TF
32
+ + 2 ln 2 + (1 ln 2)S(S + 1) F
15
n

2
CF TF nl 2 3 E
8
5
10
49CA CF TF nl
+
S(S + 1)
+ CF Ln ,
+
(50)
36n
9 18n 27
n
3
where S is the spin quantum number,
Ls = ln(CF s ),
1 (z) = d;(z)/dz,
;(z) is Eulers gamma function, and E = 0.577216 . . . is Eulers constant. The terms
proportional to a1 correspond to iterations of lower-order operators. We have not included
in Eq. (50) the imaginary part corresponding to the partial width of the decay of the S = 0
state to two gluons,
CF4 TF s5 mq
.
2n3
The logarithmic part of Eq. (50),

3 1 3
7
2
4
41
CA + CF CA2 +
S(S + 1) 2 CF2 CA
En(3) = EnC s
6
3n
6n 6n
3n

1
3 3
+ CF ln ,
n
s
;gg =
(51)
(52)
376
is known from previous analyses [25,26]. For the corrections to the n = 1, 2, 3 energy
levels, we find numerically

3 a3
(3)
+ 177.716 11.611nl + 0.274n2l 0.004n3l
E1 s E1C
32 2

1
1
60.500 ln + 18.853 + 1.312 nl + 6.222 ln
s
s

S(S + 1) ,
E2(3)
(3)

s3 C a3
E
+ 102.917 8.034nl + 0.274n2l 0.004n3l

2 32 2

1
1
33.389 ln + 9.603 + 0.658nl + 3.111 ln
s
s

S(S + 1) ,
E3 =

s3 C a3
E3
+ 75.919 6.690 nl + 0.274 n2l 0.004 n3l
32 2

1
1
23.957 ln + 6.425 + 0.439nl + 2.074 ln
s
s

S(S + 1) .
(53)
5.2. Numerical estimates and phenomenological examples

To illustrate the phenomenological relevance of our results, let us consider two
important physical examples: the resonance in topantitop threshold production by e+ e
annihilation via a virtual photon and the lowest resonance. We neglect nonperturbative
contributions and finite-width effects, so that the masses of the resonances are determined
by the perturbative expressions with principal quantum number n = 1 and spin quantum
number S = 1. The complete NLO and NNLO corrections may be found in Refs. [8,
13,18], and the N3 LO ones for (s ) = 0 are given in Eq. (53). To take into account
the N3 LO RG logarithms, we normalize s in NLO and NNLO at the soft scale s =
CF s (s )mq . The setting of the normalization scale in the O(s3 ) corrections is a more
subtle problem. In this order, the hard and ultrasoft regions start to contribute. This
results in RG logarithms with corresponding scales. Furthermore, the contributions from
different regions are not separately finite, and the operators of the effective Hamiltonian
acquire anomalous dimensions, which result in non-RG logarithms (see Eqs. (9), (15)
and (17)). Thus, starting with the next order, the RG and non-RG logarithms mix. The
correct treatment of the logarithmic corrections is possible within the effective-theory RG
approach [20,5561]. For simplicity, we ignore these sophistications for the time being and
employ the soft normalization point for the whole of the O(s3 ) corrections. As an estimate
of the nonlogarithmic corrections proportional to i , we use the 03 term, which is currently
377
only known for n = 1 [27],

4s 0 3 1 2
4
2
3
E1(3) 3 = E1C
+
+ (3)
(3) + (5) ,
+
0
8 16 1440
8
2
(54)
with (5) = 1.036928 . . . , and is expected to dominate the corrections of this type.
For the topantitop system, we use mt = 176 GeV and s (s ) = 0.14 to find the
perturbative expansion of the binding energy to be

E1 = 1.53 GeV 1 + 0.448 + 0.322 + (0.006 + 0.011|a3 + 0.073| 3 ) + ,
0
(55)
N3 LO
03
contributions due to the a3 and

terms are given separately. Thus,
where the
the resonancepeak position is decreased by approximately 140 MeV in comparison to
the NNLO result. Although the O(s3 ) corrections are still important, the series shows
reasonable convergence, which makes us optimistic about an accurate determination of mt
and s from this observable.
As for the (1S) resonance, we use mb = 4.8 GeV and s (s ) = 0.31 to find the
perturbative expansion of the binding energy to be

E1 = 205 MeV 1 + 1.11 + 1.88 + (0.49 + 0.19|a3 + 1.02| 3 ) + .
(56)
0
This implies that the value of mb extracted from the (1S) resonance is increased by
approximately 170 MeV in comparison to the NNLO result. Although there is no further
growth of the perturbative corrections, the O(s3 ) corrections seem to be too large to expect
a reliable prediction from the N3 LO result, and some optimization, e.g., by mass and/or
coupling-constant redefinition, is necessary to improve the convergence of the series.
In the above estimates we used the Pad results of Eq. (14) for the coefficient a3 . The
accuracy of the Pad approximation is difficult to estimate, and a significant deviation
from the exact result does not seem impossible. However, from Eqs. (55) and (56) we
observe that the corresponding contribution only provides about 10% of the total O(s3 )
corrections, and even a 100% variation of a3 merely results in a 10% variation of the O(s3 )
corrections. This is not crucial for the topantitop system, but it could be essential in the
bottomonium case, where the magnitude of the O(s3 ) contribution is very sizeable. The
analytical evaluation of the coefficient a3 is thus quite important.
The origin of the large NLO and NNLO corrections in Eqs. (55) and (56) is usually
attributed to the IR-renormalon contribution. This contribution is absent in the IR-safe
short-distance masses, and the perturbative series for such masses are expected to exhibit
faster convergence (see, e.g., Refs. [1517]). We observe, however, that, in the topantitop
case, the perturbative series numerically converges even in the pole-mass scheme. On
the other hand, in the bottomonium case, the O(s3 ) corrections remain sizeable even
for (s ) = 0, i.e., the nave subtraction of the renormalon contribution through a mass
redefinition does not completely solve the problem of large perturbative corrections.
Our next comment concerns the corrections logarithmic in s . Using the effectivetheory RG equations, it is possible to sum up the logarithmic corrections to the energy
levels to all orders. The presence of several correlated scales renders the problem
378
very interesting and nontrivial from the conceptual point of view. Now there are two
contradictory results [59,60] on the resummation of the sn+4 lnn s terms in the series
for the energy levels, the first of which is given by Eq. (52). The result of Ref. [60] is
obtained within pNRQCD, while an alternative formulation of effective theory, velocity
NRQCD (vNRQCD), was used in Ref. [59]. Since there can be only one correct result,
this issue has to be clarified. In any case, it is interesting to check the accuracy of the
logarithmic approximation. Numerically, the logarithmic series is dominated by its first
term, which provides about 80% in both the bottomonium and topantitop cases [59,60,62].
From Eq. (53), we observe that the full result has approximately the same magnitude, but
the opposite sign compared to the logarithmic contribution. We thus conclude that, while
the logarithmic contributions in some order can give us a hint at the order of magnitude
of the full contribution of that order, the practical relevance of the high-order resummation
is questionable. This is not unexpected because the resummation parameter sn lnm s is
neither large for s 0 nor for s 1.
6. Conclusion
In this paper, we took a crucial step towards the N3 LO analysis of the heavy-quark
threshold dynamics. We used the effective theory of pNRQCD and the threshold expansion
for a detailed analysis of the nonrelativistic Hamiltonian in this order. Explicit expressions
for the N3 LO Hamiltonian in one and two loops were given. We also presented the full
matching of the Hamiltonian to the contribution from the ultrasoft gluons, which enter the
bound-state dynamics in this order. The matching calculation includes one-, two- and threeloop operators. To complete our analysis, the three-loop MS coefficient a3 of the corrections to the static potential, for which only Pad estimates are available, has to be computed.
With the full expression for the Hamiltonian and the ultrasoft contribution at hand,
it is straightforward to complete the N3 LO analysis of the heavy-quarkonium spectrum.
In this paper, we derived the heavy-quarkonium spectrum in this order, neglecting the
nonlogarithmic terms proportional to the QCD beta function. For the latter, only the 03
term for the ground-state energy is known so far.
Collecting all available contributions, we found the N3 LO corrections to be sizable
for the topantitop system, where, however, the perturbative series for the resonancepeak energy exhibits a tendency to converge even in the pole-mass scheme. In the case
of the (1S) resonance, the corrections are so sizeable, that some kind of optimization
of the perturbative expansion is needed, e.g., by removing the pole mass in favour
of a renormalon-free short-distance mass. However, we should emphasize that, in the
bottomonium case, the N3 LO corrections remain sizeable even for (s ) = 0, and the
bad behaviour of the perturbative series cannot be solely explained by the renormalon
contribution.
In order to render the analysis more accurate, the remaining nonlogarithmic terms
proportional to the QCD beta function have to be evaluated along with the a3 coefficient.
However, we do not expect a qualitative change of our result.
The result of this paper also provides a starting point for the calculation of the
N3 LO single-logarithmic s3 ln s corrections to the heavy-quarkonium production and
379
annihilation rates, similarly to a number of QED bound-state problems [6366]. While

the analysis of the nonlogarithmic terms in this order requires the calculation of the
three-loop hard renormalization of the relevant production and annihilation amplitudes,
which still is a challenging theoretical problem, the logarithmic terms are universal
and essentially determined by the effective Hamiltonian and the ultrasoft contribution
presented in this paper. Along with the known double-logarithmic s3 ln2 s terms [26],
the single-logarithmic contribution would constitute an essential part of the N3 LO
corrections to the heavy-quarkonium production and annihilation rates. The calculation of
the single-logarithmic terms would also lead to further progress in the resummation of the
logarithms in v via the nonrelativistic effective-theory RG [58,61], since it determines an
anomalous dimension necessary for the NNLO logarithmic analysis of heavy-quarkonium
production and annihilation. Although it is, in general, dangerous to rely on the logarithmic
approximation, as we observed in the case of the spectrum, the situation could be different
for the cross section normalization, where the N3 LO double-logarithmic contribution is
known to be sizeable and the resummation of the logarithmic terms could stabilize the
behavior of perturbation theory [58].
Acknowledgements
A.A.P. is grateful to A.H. Hoang and A. Pineda for useful communications. We
thank J. Soto for helpful comments on this manuscript. This work was supported in
part by the Deutsche Forschungsgemeinschaft through Grant No. KN 365/1-1 and by
the Bundesministerium fr Bildung und Forschung through Grant No. 05 HT1GUA/4.
The work of V.A.S. was supported in part by the Russian Foundation for Basic Research
through Project No. 01-02-16171 and by INTAS through Grant No. 00-00313.
Appendix A
To evaluate the two-loop contribution to the heavy-quark potential, in particular, the
1/mq corrections, one needs analytical results, at least as Laurent expansions in ! up to
some order (typically, ! 0 and ! 1 ), for the following family of two-loop Feynman integrals:

J a1 , . . . , a8 ; q 2 ; !

dd k dd l
=
(k 2 )a1 (l 2 )a2 [(k q)2 ]a3 [(l q)2 ]a4 [(k l)2 ]a5
1
,
(A.1)
a
6
(v0 k) (v0 l)a7 [v0 (k l)]a8
where the four-vector v0 is defined below Eq. (21), k and l are loop four-momenta, and
+i is omitted in all the denominators.
As in Refs. [44,45,67], we use a reduction procedure that expresses any integral of the
form of Eq. (A.1) in terms of some master integrals. To this end, we employ the following
380
identities, which can be obtained by means of the method of integration by parts [68],

(d 2a1 a3 a5 a6 ) + a3 3+ q 2 1 a5 5+ (1 2 ) a8 6 8+ = 0,

(d 2a2 a3 a5 a7 ) + a4 4+ q 2 2 a5 5+ (2 1 ) + a8 7 8+ = 0,
(d a1 a3 2a5 a6 a8 ) + a1 1+ (2 5 ) + a3 3+ (4 5 ) + a6 6+ 7 = 0,
(d a2 a4 2a5 a7 a8 ) + a2 2+ (1 5 ) + a4 4+ (3 5 ) + a7 6 7+ = 0,

(d a1 2a3 a5 a6 ) + a1 1+ q 2 3 a5 5+ (3 4 ) a8 6 8+ = 0,

(d a2 2a4 a5 a7 ) + a2 2+ q 2 4 a5 5+ (4 3 ) + a8 7 8+ = 0,
2a1 1+ 6 + 2a33+ 6 + a5 5+ 8 + v 2 a6 6+ + v 2 a8 8+ = 0,
2a2 2+ 7 + 2a44+ 7 a5 5+ 8 + v 2 a7 7+ v 2 a8 8+ = 0,
(A.2)
as well as the trivial identity 6 7 = 8 . Here, the standard notation for raising and
lowering operators has been used, e.g.,
1 3+ J (a1 , . . . , a8 ) = J (a1 1, a2 , a3 + 1, . . . , a8 ).
(A.3)
We developed a reduction procedure very similar to the one of Ref. [67]. In our problem,
however, we need a larger class of integrals that arise in the calculation of the 1/mq
corrections in the general covariant gauge. The main difference between our reduction
procedure and the one of Ref. [67] is that we stop the reduction if we arrive at integrals
expressed in terms of gamma functions for finite !. There are two subclasses of the integrals
of Eq. (A.1) that are only evaluated as expansions in ! up to some order. The first of them
was described in Ref. [44], namely,

I a1 , . . . , a5 ; q 2 ; !

dd k dd l
=
(A.4)
.
(k 2 )a1 (l 2 )a2 [(k l q)2 ]a3 (v0 k)a4 (v0 l)a5
In particular, we have
J (0, a2, a3 , 0, a5, a6 , 0, a8) = I (a5 , a3 , a2 , a8 , a6 ),
J (a1 , 0, 0, a4, a5 , a6 , a7 , 0) = I (a1 , a4 , a5 , a6 , a7 ).
The master integrals for this subclass are

I 1, 1, 1, 1, 1; q 2; !

2
(i d/2eE ! )2
2 2
7 4
2
2
4
!
+
O
!
=
24
,
3!
9
(q 2 )2!

I 1, 1, 2, 1, 1; q 2; !

5 2
10 2 64(3)
4
(i d/2eE ! )2 2
+
8
16
=
!
(q 2)1+2!
!2 !
3
3
3

+ O !2 .
(A.5)
(A.6)
381
We need also integrals of the type of Eq. (A.4) with a numerator that can be chosen to be
q k or q l. The reduction of these integrals is quite similar and also results in two master
integrals.
The second subclass of the integrals of Eq. (A.1) that are not expressed in terms of
gamma functions for general ! consists of integrals with a5 = a6 = a7 = 0. Using Feynman
parameters, the general integral of this kind can be represented in terms of the following
MellinBarnes representation:
J (a1 , . . . , a4 , 0, 0, 0, a5)
= 5
(1)a 2a5 1 (i d/2 )2
i=1 ;(ai )(q

+i

1
2i
dz
i
2 )aa5 /24+2!
;(a1 + a2 + a3 + a4 + a5 /2 4 + 2! + z);(a1 + z);(a3 + z)

;(a1 + a3 + 2z);(z)
;(a1 a3 a4 a5 /2 + 4 2! z);(a1 a2 a3 a5 /2 + 4 2! z)
;(2a1 a2 2a3 a4 a5 + 8 4! 2z)

;(a1 + a3 + a5 /2 2 + ! + z);(a1 a3 + 2 ! z),
(A.7)
5
where a = i=1 ai . We performed a reduction to integrals with a2 = a3 = a4 = a5 = 1
and evaluated them by means of Eq. (A.7) for the required values of a1 . In particular, we
found

J 1, 1, 1, 1, 0, 0, 0, 1; q 2; !

2

5 4
(i d/2eE ! )2
2
=
(A.8)
!+O !
4 ln 2 1 + !(4 + ln 2) +
.
(q 2 )1/2+2!
3
References
[1] T. Appelquist, H.D. Politzer, Phys. Rev. Lett. 34 (1975) 43.
[2] V.A. Novikov, L.B. Okun, M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, V.I. Zakharov, Phys. Rev. Lett. 38
(1977) 626;
V.A. Novikov, L.B. Okun, M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, V.I. Zakharov, Phys. Rev. Lett. 38
(1977) 791, Erratum;
V.A. Novikov, L.B. Okun, M.A. Shifman, A.I. Vainshtein, M.B. Voloshin, V.I. Zakharov, Phys. Rep. C 41
(1978) 1.
[3] V.S. Fadin, V.A. Khoze, Pisma Zh. Eksp. Teor. Fiz. 46 (1987) 417, [JETP Lett. 46 (1987) 525].
[4] M.B. Voloshin, Nucl. Phys. B 154 (1979) 365;
M.B. Voloshin, Yad. Fiz. 36 (1982) 247, [Sov. J. Nucl. Phys. 36 (1982) 143].
[5] H. Leutwyler, Phys. Lett. B 98 (1981) 447.
[6] W.E. Caswell, G.P. Lepage, Phys. Lett. B 167 (1986) 437.
[7] G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 51 (1995) 1125;
G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 55 (1997) 5853, Erratum.
[8] A. Pineda, F.J. Yndurain, Phys. Rev. D 58 (1998) 094022;
A. Pineda, F.J. Yndurain, Phys. Rev. D 61 (2000) 077505.
382
[9] A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 80 (1998) 2531;

A. Czarnecki, K. Melnikov, Phys. Rev. D 65 (2002) 051501.
[10] M. Beneke, A. Signer, V.A. Smirnov, Phys. Rev. Lett. 80 (1998) 2535.
[11] J.H. Khn, A.A. Penin, A.A. Pivovarov, Nucl. Phys. B 534 (1998) 356.
[12] A.H. Hoang, T. Teubner, Phys. Rev. D 58 (1998) 114023.
[13] A.A. Penin, A.A. Pivovarov, Phys. Lett. B 435 (1998) 413;
A.A. Penin, A.A. Pivovarov, Nucl. Phys. B 549 (1999) 217.
[14] A.A. Penin, A.A. Pivovarov, Nucl. Phys. B 550 (1999) 375;
A.A. Penin, A.A. Pivovarov, Yad. Fiz. 64 (2001) 323, [Sov. J. Nucl. Phys. 64 (2001) 275].
[15] K. Melnikov, A. Yelkhovsky, Phys. Rev. D 59 (1999) 114009.
[16] A.H. Hoang, Phys. Rev. D 59 (1999) 014039;
A.H. Hoang, Phys. Rev. D 61 (2000) 034005.
[17] M. Beneke, A. Signer, Phys. Lett. B 471 (1999) 233.
[18] K. Melnikov, A. Yelkhovsky, Nucl. Phys. B 528 (1998) 59.
[19] O.I. Yakovlev, Phys. Lett. B 457 (1999) 170.
[20] M. Beneke, A. Signer, V.A. Smirnov, Phys. Lett. B 454 (1999) 137.
[21] T. Nagano, A. Ota, Y. Sumino, Phys. Rev. D 60 (1999) 114014.
[22] A.H. Hoang, T. Teubner, Phys. Rev. D 60 (1999) 114027.
[23] A.H. Hoang, M. Beneke, K. Melnikov, T. Nagano, A. Ota, A.A. Penin, A.A. Pivovarov, A. Signer, V.A.
Smirnov, Y. Sumino, T. Teubner, O. Yakovlev, A. Yelkhovsky, Eur. Phys. J. C 3 (2000) 1.
[24] B.A. Kniehl, A.A. Penin, Nucl. Phys. B 563 (1999) 200.
[25] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Phys. Lett. B 470 (1999) 215.
[26] B.A. Kniehl, A.A. Penin, Nucl. Phys. B 577 (2000) 197.
[27] Y. Kiyo, Y. Sumino, Phys. Lett. B 496 (2000) 83.
[28] A.A. Penin, Nucl. Phys. B (Proc. Suppl.) 96 (2001) 418.
[29] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Nucl. Phys. B 566 (2000) 275.
[30] B.A. Kniehl, A.A. Penin, M. Steinhauser, V.A. Smirnov, Report No. DESY 01-075, hep-ph/0106135,
accepted for publication in Phys. Rev. D, Rapid Communication.
[31] A. Pineda, J. Soto, Nucl. Phys. B (Proc. Suppl.) 64 (1998) 428.
[32] M. Beneke, V.A. Smirnov, Nucl. Phys. B 522 (1998) 321.
[33] M. Nio, T. Kinoshita, Phys. Rev. D 55 (1997) 7267.
[34] G.S. Adkins, R.N. Fell, J.R. Sapirstein, Phys. Rev. Lett. 84 (2000) 5086.
[35] V.A. Smirnov, Phys. Lett. B 465 (1999) 226;
V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses, Springer-Verlag, Heidelberg, 2001.
[36] A. Pineda, J. Soto, Phys. Lett. B 420 (1998) 391;
A. Pineda, J. Soto, Phys. Rev. D 59 (1999) 016005.
[37] A. Czarnecki, K. Melnikov, A. Yelkhovsky, Phys. Rev. A 59 (1999) 4316;
A. Czarnecki, K. Melnikov, A. Yelkhovsky, Phys. Rev. Lett. 83 (1999) 1135.
[38] B.A. Kniehl, Z. Phys. C 72 (1996) 437.
[39] O.V. Tarasov, A.A. Vladimirov, A.Yu. Zharkov, Phys. Lett. B 93 (1980) 429.
[40] T. Appelquist, M. Dine, I.J. Muzinich, Phys. Rev. D 17 (1978) 2074.
[41] N. Brambilla, A. Pineda, J. Soto, A. Vairo, Phys. Rev. D 60 (1999) 091502.
[42] W. Fischler, Nucl. Phys. B 129 (1977) 157.
[43] A. Billoire, Phys. Lett. B 92 (1980) 343.
[44] M. Peter, Phys. Rev. Lett. 78 (1997) 602;
M. Peter, Nucl. Phys. B 501 (1997) 471.
[45] Y. Schrder, Phys. Lett. B 447 (1999) 321.
[46] F.A. Chishtie, V. Elias, Phys. Lett. B 521 (2001) 434.
[47] S.N. Gupta, S.F. Radford, Phys. Rev. D 24 (1981) 2309;
S.N. Gupta, S.F. Radford, Phys. Rev. D 25 (1982) 3430.
[48] S. Titard, F.J. Yndurain, Phys. Rev. D 49 (1994) 6007.
[49] S.N. Gupta, W.W. Repko, C.J. Suchyta III, Phys. Rev. D 40 (1989) 4100.
[50] W. Buchmller, Y.J. Ng, S.H.H. Tye, Phys. Rev. D 24 (1981) 3003.
[51] J. Pantaleone, S.H.H. Tye, Y.J. Ng, Phys. Rev. D 33 (1986) 777.
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
A.V. Manohar, Phys. Rev. D 56 (1997) 230.

A. Pineda, J. Soto, Phys. Rev. D 58 (1998) 114011.
A.V. Manohar, I.W. Stewart, Phys. Rev. D 62 (2000) 074015.
M.E. Luke, A.V. Manhoar, I.Z. Rothstein, Phys. Rev. D 61 (2000) 074025.
A.V. Manohar, I.W. Stewart, Phys. Rev. D 62 (2000) 014033;
A.V. Manohar, I.W. Stewart, Phys. Rev. D 63 (2001) 054004.
A. Pineda, J. Soto, Phys. Lett. B 495 (2000) 323.
A.H. Hoang, A.V. Manohar, I.W. Stewart, T. Teubner, Phys. Rev. Lett. 86 (2001) 1951;
A.H. Hoang, A.V. Manohar, I.W. Stewart, T. Teubner, Phys. Rev. D 65 (2002) 014014.
A.H. Hoang, A.V. Manohar, I.W. Stewart, Phys. Rev. D 64 (2001) 014033.
A. Pineda, Report No. TTP01-22, hep-ph/0109117.
A. Pineda, Report No. TTP01-27, hep-ph/0110216.
A. Pineda, private communication.
B.A. Kniehl, A.A. Penin, Phys. Rev. Lett. 85 (2000) 1210;
B.A. Kniehl, A.A. Penin, Phys. Rev. Lett. 85 (2000) 3065, Erratum;
B.A. Kniehl, A.A. Penin, Phys. Rev. Lett. 85 (2000) 5094.
R. Hill, G.P. Lepage, Phys. Rev. D 62 (2000) 111301.
K. Melnikov, A. Yelkhovsky, Phys. Rev. D 62 (2000) 116003;
K. Melnikov, A. Yelkhovsky, Phys. Rev. Lett. 86 (2001) 1498.
R.J. Hill, Phys. Rev. Lett. 86 (2001) 3280.
Y. Schrder, Ph.D. Thesis, Hamburg University, 1999, Report No. DESY-THESIS-1999-021.
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
383

On the quantum moduli space of M-theory

compactifications
Tamar Friedmann
Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
Abstract
We study the moduli space of M-theories compactified on G2 manifolds which are asymptotic to a
cone over quotients of S3 S3 . We show that the moduli space is composed of several components,
each of which interpolates smoothly among various classical limits corresponding to low energy
gauge theories with a given number of massless U (1) factors. Each component smoothly interpolates
among supersymmetric gauge theories with different gauge groups. 2002 Published by Elsevier
Science B.V.
1. Introduction
The study of M-theory compactifications on seven-dimensional manifolds X of G2
holonomy has been motivated by the fact that such compactifications result in unbroken
supersymmetry in four dimensions. The properties of the compactification manifold X
determine the particle spectrum of the corresponding four-dimensional theory. It has been
shown in recent years that compactifications on singular manifolds can result in low energy
physics containing interesting massless spectra. Specifically, certain singular G2 manifolds
give rise to N = 1 supersymmetric gauge theories at low energies, as shown, for example,
in [13]. There, X was taken to be asymptotic to a quotient of a cone on S3 S3 , and
the singularities of X took the form of families of ADE singularities giving ADE gauge
theories at low energies.
Subsequently, the quantum moduli space of M-theories on G2 manifolds X which are
asymptotic to a cone on S3 S3 or quotients thereof has been studied in [4]. It was shown
E-mail address: tamarf@feynman.princeton.edu (T. Friedmann).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 0 8 - X
T. Friedmann / Nuclear Physics B 635 (2002) 384394
385
that the moduli space is a Riemann surface of genus zero, which interpolates smoothly
between different semiclassical spacetimes.
The purpose of this paper is to generalize the construction of [4] to other quotients
of S3 S3 and obtain the moduli spaces for those as well. Our quotients contain those
in [4] as special cases. We propose that the moduli space for our quotients consists of
several branches classified according to the number of massless U (1) factors that appear
in the low energy gauge theories corresponding to semiclassical points. Each branch of the
moduli space interpolates smoothly between the different semiclassical points appearing
on it; hence, we get smooth interpolation between supersymmetric gauge theories with
different gauge groups.
This paper is organized as follows: in Section 2 we review the M-theory dynamics on
the cone on Y = S3 S3 given in [4]. In Section 3, we describe quotients of this cone by
discrete groups of the form = 1 2 3 where the i are ADE subgroups of SU(2);
these ADE groups must be chosen carefully in order to obtain known low energy gauge
theories from the compactification. In Section 4, we turn to the description of the moduli
space N of M-theories on these quotients, beginning with the classical moduli space and
concluding with the quantum moduli space.
While this paper was being completed, we received [5] which has overlap with the case
where 3 is trivial (or r = 1 in our notation of Section 3).
2. Dynamics of M-theory on the cone over S3 S3
In this section, we review the M-theory dynamics on a manifold X of G2 holonomy
which is asymptotic at infinity to a cone over Y = S3 S3 [4]. The manifold Y can be
described as a homogeneous space Y = SU(2)3 /SU(2), where the equivalence relation
is (g1 , g2 , g3 ) (g1 h, g2 h, g3 h), gi , h SU(2). Viewed this way, this manifold has
SU(2)3 symmetry via left action on each of the three factors in Y , as well as a triality
symmetry S3 permuting the three factors. Up to scaling, there is a unique metric with such
symmetries given by
d 2 = da 2 + db2 + dc2 ,
(1)
Tr(a 1 da)2 ,
=
the trace is taken in the fundamental
where a, b, c SU(2),
representation of SU(2), and a, b, c are related to g1 , g2 , g3 by a = g2 g31 and cyclic
permutations thereof.
The metric for a cone on Y is
da 2
ds 2 = dr 2 + r 2 d 2 ,
(2)
where d 2 is the metric on Y . Such a cone can be constructed by filling in one of the three
SU(2) S3 factors of Y to a ball. We denote the manifold obtained by filling in a given gi
by Xi . The metric on a manifold X, asymptotic to Xi at infinity, can be written with a new
radial variable y, which is related to r by
y =r

r03
+ O 1/r 5 ,
2
4r
(3)
386
as
6

r
r03
y2
2
2
2
2
2
2
ds = dy +
,
da + db + dc 3 f1 da + f2 db + f3 dc + O 06
36
2y
y
(4)
where r0 is a parameter denoting the length scale of Xi , and (fi1 , fi , fi+1 ) = (1, 2, 1)
(indices are understood mod 3). When y or r , this becomes precisely the
cone (2).
We will need to study the 3-cycles of Y in order to understand the relations between the
periods of the M-theory C-field and the membrane instanton amplitudes, which we shall
need in order to describe the moduli space.
The 3-cycles Dj of Y are given by projections of the j th factor of SU(2)3 to Y . Hence,
Dj
= S3 . The third Betti number of Y is two, so the three Dj satisfy the relation
2
D1 + D2 + D3 = 0.
(5)
The intersection numbers of the Di are given by

Di Dj = j,i+1 j,i1 .
(6)
At Xi , where the ith factor is filled in, Di shrinks to zero and the relation (5) reduces to
Di1 + Di+1 = 0 (where again the indices are understood mod 3).
At each Xi , there is a supersymmetric 3-cycle Qi given by gi = 0. It can be shown that
Qi is homologous to Di1 and Di+1 , where the sign depends on orientation.
A manifold still has G2 holonomy up to third order in r0 /y if we take the fj of (4) to
be any linear combination of (1, 2, 1) and its permutationsso we have G2 holonomy
as long as
f1 + f2 + f3 = 0.
(7)
These fj can be interpreted as volume defects of the cycle Dj at infinity: the volume
of Dj depends linearly on a positive multiple of fj . Furthermore, since at the classical
manifold Xi , only one of the Dj vanishes, only one of the fj (namely fi ) can be negative.
So the classical moduli space may contain manifolds with the relation (7) as long as only
one of the fj is negative [4,6].

The periods of the C-field along the cycles Dj are j = Dj C. We combine them with
the fj into holomorphic observables j where now the C-field period is a phase:

2k
k
fj 1 + fj + ij ,
j = exp
(8)
3
3
where k is a parameter. The relation (7) means that the j are not independent, but instead
they obey

1 2 3 = exp i
(9)
j .
(It can be shown that due to a global anomaly in the membrane effective action, the righthand side above is 1.)
387
The moduli space at the classical approximation is given by three branches Ni , each
of which contains one of the points Xi with r0 . On Xi , i vanishes and the
parameters fj are such that i = 1. So on Ni the functions j obey
i = 1,
i1 i+1 = 1.
(10)
At the quantum level, there are corrections to this statement. It has been suggested in [3]
that the different classical points Xi are continuously connected to one another. Hence
they should appear on the same branch of the moduli space N . We proceed now with the
assumption that the only classical points are the Xi , which are the points where some of
the j have a zero or pole. As explained in [4], since a component of N which contains a
zero of a holomorphic function j must also contain its pole, and since the only points at
which the j are singular are associated with one of the Xi , it follows indeed that all Xi
are contained on a single component of N . Furthermore, each j has a simple zero and
simple pole in N . The existence of such functions on N means that the branch containing
the zero and pole has genus zero. In addition, any of the j can be identified as a global
coordinate of N . Choosing any j gives a complete description for this branch of N .
3. Quotients and low energy gauge groups

Here, we begin our study of manifolds which are asymptotic to a cone over quotients
of Y . We shall consider a discrete group action of = 1 2 3 on Y where the i
will be chosen from ADE subgroups of SU(2) in such a way that the low energy physics is
known.
We begin with the simplest case where = Zp Zq Zr . Each Zn is embedded in
SU(2) via

2ik/n
0
e
,
k =
(11)
0
e2ik/n
where is the generator of Zn and k = 0, 1, . . . , n 1. The action of on Y =
SU(2)3 /SU(2) is given by
( , , () Zp Zq Zr : (g1 , g2 , g3 ) ( g1 , g2 , (g3 ),
(12)
and we denote the resulting quotient space by Y .

The spaces Xi, , obtained by filling in the ith SU(2) factor of Y , are quotients of R4
3
S where the R4 corresponds to the filled-in factor. Choosing i = 1 and gauging g2 away
using the right diagonal SU(2) action, the identification corresponding to ( k , l , ( m )
is

(g1 , 1, g3 ) k g1 l , 1, ( m g3 l ,
(13)
where g1 R4 and g3 SU(2) S3 . The set (0, 1, g3 ) with g3 varying in SU(2) is a fixed
point of the action of the Zp subgroup of , and this singularity is identical to the standard
Ap1 singularity of codimension four of the form R4 /Zp or C2 /Zp , which gives an SU(p)
gauge theory at low energies.
388
Depending on the values of the integers q and r, there may be additional, unfamiliar
singularities for which we do not know the low energy physics. Namely, there may be
values of g3 which are fixed under a nontrivial subgroup of Zq Zr , i.e., where the
following holds
( m g3 l = g3 .
(14)
This is the same as looking for elements g3 of SU(2) which diagonalize l :

( m = g3 l g31 .
(15)
Choosing the orders q and r of and ( to be relatively prime, (q, r) = 1, ensures that there
are no solutions of this equation (since then the orders of the left- and right-hand sides
of (15) are relatively prime). Similarly, we choose (p, q) = (p, r) = 1, and so there are no
singularities at Xi, other than the ADE singularities whose low energy physics is known:
an Ap1 singularity on S3 /(Zq Zr ) at X1, , an Aq1 singularity on S3 /(Zr Zp )
at X2, , and an Ar1 singularity on S3 /(Zp Zq ) at X3, , with the discrete group action
on S3 given by the appropriate cyclic permutation of the action on g3 in (13).
Now consider also the non-abelian ADE groups. Again, we would like to choose such
that we will only get singularities whose physics at low energies we understandnamely,
ADE singularities. For this purpose we review the relevant properties of the DE groups.
For information about these groups, see [7].
As in the abelian case, we let = 1 2 3 act on Y by
( , , () 1 2 3 : (g1 , g2 , g3 ) ( g1 , g2 , (g3 ),
(16)
from which Eqs. (13) and (14) follow in the same way as before.
The binary dihedral groups Dq have order 4q 8 and are generated in SU(2) by two
elements:
i

0 1
e q2
0
Dq =
(17)
,
.
i
1 0
0
e q2
Since all Dq groups share an element of order 4, we cannot choose more than one of the
i to be a dihedral group, since otherwise we would get solutions to (14). Hence we let
= Zp Dq Zr with (p, r) = (p, 2(q 2)) = (r, 2(q 2)) = 1.
We turn to the E series. A singularity R4 /G which gives at low energies E6 , E7 ,
or E8 gauge groups corresponds to G being the tetrahedral group T24 , the octahedral
group O48 , or the icosahedral group I120 . The orders of these groups are 24, 48, and 120,
respectively, and each of them has elements of orders 3 and 4, so we cannot have more
than one E group appearing in . The group I120 also has elements of order 5. Hence,
in addition to (p, r) = 1, for = Zp E6 Zr or = Zp E7 Zr , we need also
(p, 2 3) = (r, 2 3) = 1, and for = Zp E8 Zr , we need (p, 2 3 5) = (r, 2 3 5) = 1.
Therefore, our group is always chosen to be of the form = Zp 2 Zr where
2 is an A, D, or E group, and p, r, and 2 satisfy the conditions noted above, which can
be summarized by
(p, N) = (r, N) = (p, r) = 1,
389
Table 1
Low energy gauge theories at Xi,
2
Zq
Dq
T24
O48
I120
SU(p)
SU(p)
SU(p)
SU(p)
SU(p)
X1,
S3 /(Zq Zr )
S3 /(Dq Zr )
S3 /(T24 Zr )
S3 /(O48 Zr )
S3 /(I120 Zr )
SU(q)
SO(2q)
E6
E7
E8
X2,
S3 /(Zr
S3 /(Zr
S3 /(Zr
S3 /(Zr
S3 /(Zr
Zp )
Zp )
Zp )
Zp )
Zp )
SU(r)
SU(r)
SU(r)
SU(r)
SU(r)
X3,
S3 /(Zp Zq )
S3 /(Zp Dq )
S3 /(Zp T24 )
S3 /(Zp O48 )
S3 /(Zp I120 )
where N is the order of the group 2 . At X1, we have an Ap1 singularity on

S3 /(2 Zr ), at X3, we have an Ar1 singularity on S3 /(Zp 2 ), and at X2, we
have an A, D, or E singularity on S3 /(Zr Zp ), where here the discrete group action is
given by the appropriate cyclic permutation of the action on g3 in (13).
The low energy gauge theories obtained from compactifying M-theory on R4 Xi, are
listed in Table 1. Each entry contains the gauge group and the compact 3-manifold which
is the locus of the ADE singularity.
As we shall see below, for the cases where 2 is a D or E group, there are additional
semiclassical points where the low energy gauge group is different from those listed above.
We note that for the case with r = 1, X3, is smooth and its low energy theory has no
gauge symmetry. If also p = 1, X1, is smooth as well (this is the case studied in [4]).
4. The curve of M theories on the quotient
4.1. Classical geometry
The 3-cycles Di of Y are the projections of the ith factor of SU(2)3 to Y . Hence, for
= Zp 2 Zr we have
D1 = S3 /Zp ,
(18)
D2
D3
(19)
= S /2 ,
3
= S /Zr .
3
(20)
Using the relation (5) in Y and the fact that D1 Y projects to a p-fold cover of
as well as cyclic permutations of this fact, we find
pD1 + ND2 + rD3 = 0,
D1
Y ,
(21)
where N is the order of the group 2 . To study the intersection numbers of the Di we note
that D1 Y lifts to NrD1 Y , and similar statements are true for the other Di . Counting
the intersection numbers in Y and then dividing by pNr (since there are pNr points in Y
which project to one point in Y ), we get
D1 D2 = r,
D2 D3 = p,
D3 D1 = N.
(22)
Here we see that the Di generate the third homology group of Y : since (r, N) = 1, we
can find integers m, n such that
D1 (mD2 + nD3 ) = mr nN = 1,
(23)
390
and similarly for the other cycles.

We define the periods of the M-theory C-field at infinity by

j = C mod 2.
(24)
Dj
Note that these are related to the j of Y by

1 = p1 ,
2 = N2 ,
3 = r3 .
(25)
4.2. Classical moduli space

We define our holomorphic observables to be the following functions of the periods j
and of the volumes fj :

2k
k
1 = exp
f3 +
f1 + i1 ,
3p
3p

2k
k
f1 +
f2 + i2 ,
2 = exp
3N
3N

2k
k
f2 + f3 + i3 .
3 = exp
3r
3r
These functions are adopted from (8), where we substitute the expressions in (25) for
the periods and then take the largest possible root that still leaves the i invariant under
j j + 2 .
The periods of the C-field are interpreted as the phases of the holomorphic observables.
Due to (7), we have

p N r
j .
1 2 3 = exp i
(26)
j
The j have zeros or poles at the semiclassical points Xi, with large r0 in which the fj
diverge. As in Section 2, classically at the point X1, , 1 = 1 and 1 = 0. Hence, at this
point

2N 3r = exp i(2 + 3 ) ,
(27)
so when 2 has a pole, 3 has a zero and vice versa. In fact, the order of the zeros or
poles of 2 must be a multiple of r, and similarly the order of the zeros or poles of 3
must be a multiple of N for this equation to hold. In the classical approximation, there are
three branches Ni of the moduli space, on which we have i = 1 and i1 obeying the
relation (27) for i = 1 or cyclic permutations of it for i = 2, 3.
4.3. Quantum curve via membrane instantons
To study the quantum curve, we study the singularities, i.e., the zeros and poles of the
holomorphic observables j , which correspond to the classical points Xi, with r0 .
391
We shall use a relation between the j and the amplitude for membrane instantons which
wrap on supersymmetric cycles Q in X. Using chiral symmetry breaking of the low energy
gauge theories, we find a clear relation between the local parameter on the moduli space
and our observables, and hence can describe the moduli space.
A supersymmetric cycle in Xi, is given by the 3-manifolds Qi given by gi = 0:
Q1 = S3 /(2 Zr ),
(28)
Q2 = S /(Zr Zp ),
(29)
Q3 = S /(Zp 2 ).
(30)
3
3
At X1, , Q1 is homologous (up to orientation) to the Dj as follows:

rQ1 D2 ,
(31)
NQ1 D3 ,
(32)
and cyclic permutations of that give the relations at X2, to be pQ2 D3 and rQ2 D1 ,
and at X3, we have NQ3 D1 and pQ3 D2 .
We now study the zeros and poles of the j . To understand the orders of the zeros and
poles, we must compare the j to the true local parameter on N around each Xi, with
large r0 .
One would expect at first that the membrane instanton amplitude u itself, given
near Xi, by

u = exp T V (Qi ) + i
C ,
(33)
Qi
where T is the membrane tension and V (Qi ) is the volume of Qi , would be a good local
parameter near Xi, . However, at low energies we have a supersymmetric A, D, or E
gauge theory in four dimensions, and due to chiral symmetry breaking, we expect the good
local parameterthe gluino condensateto be u1/ h where h is the dual Coxeter number
of the gauge group.
We now compare phases of the j to the phase of u. Let Pi, correspond to the
manifolds Xi, with large r0 . For the case where
2 = Zq , atP1, Eq. (31) implies that
the phase D C of 2 is related to the phase Q1 C of u by D C r Q1 C. Since the
2
good local parameter is actually u1/p due to chiral symmetry breaking of the SU(p) gauge
theory at P1, , the true order of the zero of 2 at P1, is pr. The same calculation for the
other j and Pi, gives the orders of zeros and poles shown in Table 2.
Table 2
Behavior of i for 2 = Zq
2 = Zq
P1,
P2,
1
1
qr
2
0pr
1
3
pq
0pq
P3,
0qr
pr
1
392
Table 3
Behavior of i for 2 = Dq
2 = Dq
P1,
P2,
1
1
rh
2
0rp
1
3
Np
0hp
2rh
1

02h p
0rN
rp
1
Table 4
Behavior of i for 2 = Ea
2 = Ea
P1,
P2,
Pt,
P3,
rh
rtht
e2 i/t
0ptht
0rN
rp
1
1
2
3
1
0rp
Np
0hp
P2 ,
P3,
The cases where 2 is a D or E group give similar tables, except that in these cases
we get extra semiclassical points in the same way as in [4]: for the case 2 = Dq , we have
Table 3, where h = 2q 2, h = q 3, and h + 2h = N . The low energy gauge theory at
P2 , has gauge group Sp(q 4).
For 2 in the E series, we have Table 4, where t, ht , are given for each Ea as follows:
let ki be the Dynkin indices of Ea , and let t be the positive integers which divide some of
the ki ; runs over positive integers less than t that are prime to t, unless t = 1 in which
case = 0; ht is the dual Coxeter number of the associated group Kt whose Dynkin
indices are ki /t where
here the ki run through the indices of Ea that divide t. The t and ht
obey the relation
tht = N . The low energy gauge group is given by the ADE group
corresponding to Kt .
From the relation tht = N and Tables 24, we see that for each j , the total number
of zeros and poles is equal. Since the total number of zeros is the same as the total number
of poles for each of the j , it seems reasonable to assume that we have found all the
zeros and poles, and hence all the semiclassical limits in our moduli space. It would seem,
therefore, that we can now proceed to describe the moduli space completely, by writing
our functions i explicitly and identifying the points Pi, with values of a good coordinate
on the moduli space. However, as we shall see, we run into a few puzzles.
The first question we ask is: what can be said about the genus of N ? For the cases
p = r = 1, which are the cases considered in [4], the function 2 has a simple zero and
a simple pole, and hence can be identified with a global coordinate on the moduli space,
which can then be claimed to have genus zero. If p, r > 1, this is not so: none of our j
have just a simple zero and pole, so we cannot identify the moduli space with any of the j ,
and we do not know the genus.
However, the simplest result would be that the curve has genus zero, and we proceed
with this assumption. Hence, we assign the curve a global coordinate z, write the j as
holomorphic functions of z, and see how well we can describe the curve.
For the case 2 = Zq , this turns out to be straightforward; we may fix P1, at z = 0,
P2, at z = 1, and P3, at z = , and then write our functions:
393
1
,
(34)
(1 z)qr
2 = zpr ,
(35)
(1 z)pq
3 =
(36)
.
zpq
This description is unique up to possible overall factors which are related to an anomaly in
the membrane effective action, analogous to the one described in Section 5 of [4].
For 2 = Dq , we run into a puzzle. Once we fix the first three points, we have to find at
what value z4 the fourth point P2 , sits: our functions in this case are
1 =
z42rh
,
rh
(1 z) (z4 z)2rh
2 = zrp ,
1 =
3 =
(1 z)ph (z4
zNp

z)2h p
(37)
(38)
(39)
again up to overall factors. The forms of 1 and 3 do not constrain z4 , but to satisfy
pr
2 (z4 ) = 1, we need z4 = 1 for which there are pr solutions. A similar situation
arises for 2 in the E series, where there are pr choices for each point beyond the first
three.
The pr solutions, however, should correspond to the same point in the moduli space
of M-theories, since they correspond to the same theory. Hence, it seems that we have a
redundancy in our description of the moduli space; we should impose a symmetry on N
which identifies the different values of z4 .
There is another, more serious puzzle which shows up, also involving possible extra
classical points on N : from Table 1, we see that our low energy gauge theory is
compactified on a manifold which is not simply connected, but rather is of the form S3 /H
for some discrete group H . Hence its fundamental group is equal to H . Therefore, it is
possible to construct theories which have gauge fields with nontrivial Wilson loops which
break the gauge symmetry. Where in N do these theories lie?
For the case 2 = Zq , the point P1, can have Wilson loops which are conjugacy classes
of elements of SU(p) of order qr. One can show that, when p, q, r are relatively prime,
the number of inequivalent such elements is

(p + qr 1)!
1 p + qr 1
=
(40)
qr 1
p
p!(qr)!
with cyclic permutations for P2, and P3, . Furthermore, for Wilson loops that break
SU(p) in a way that leaves s 1 factors of U (1), i.e.,
SU(p)
where
s

SU(n1 ) U (1)s1,
i=1
ni = p, the number of inequivalent Wilson loops is

s
p
qr
.
s
pqr s
(41)
394
Each set of theories with a given number s 1 of U (1) factors should lie on a separate
component Ns, of the moduli space, since smooth interpolation means that the number of
massless modeswhich corresponds to U (1) fieldsis constant on each component. For
s > 1, we know that the theories on Ns, do not have a mass gap due to the massless U (1)
field. On the other hand, the theories corresponding to the points Pi, with no nontrivial
Wilson loops are believed to have a mass gap. Hence we claim that N1, contains theories
with a mass gap.
For the case r = 1, we obtain no singularity at X3, . Hence, for that case the mass gap
of the theory at X3, means that all of N1, has a mass gap.
Continuing with the case where r = 1, we note a manifest symmetry between p and q in
the expression (41) for the number of possible Wilson loops at each level s. At first sight,
this could support the assertion that these points lie on their own branch of the moduli
space, which will interpolate smoothly among them and contain no other singular points.
However, chiral
symmetry breaking means that the number of vacua at each classical point
is given by ni which is clearly not symmetric between p and q, and spoils the counting
of the orders of zeros and poles.
Going back to general r and looking at N1, only, we see that we have smooth
interpolation among theories with different gauge groups: SU(p), SU(q), and SU(r) when
2 = Zq ; SU(p), SO(2q), Sp(q 4), and SU(r) when 2 = Dq ; and analogously for 2 in
the E series, where we interpolate between SU(p), Kt , and SU(r), with Kt as described
before. Similarly, the other branches Ns, smoothly interpolate among theories with these
gauge groups broken by Wilson lines.
Acknowledgements
The author is very grateful to Edward Witten for useful and inspiring discussions and
for guidance at every stage of this project. This research has been supported in part by a
Paul and Daisy Soros Fellowship for New Americans and in part by a National Science
Foundation Graduate Research Fellowship.
References
[1] B. Acharya, M theory, Joyce orbifolds and super-YangMills, Adv. Theor. Math. Phys. 3 (1999) 227, hepth/9812205.
[2] B. Acharya, On realizing N = 1 super-YangMills in M theory, hep-th/0011089.
[3] M.F. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large N duality, J. Math. Phys. 42 (2001) 3209,
hep-th/0011256.
[4] M.F. Atiyah, E. Witten, M-theory dynamics on a manifold of G2 holonomy, hep-th/0107177.
[5] H. Ita, Y. Oz, T. Sakai, Comments on M theory dynamics on G2 holonomy manifolds, hep-th/0203052.
[6] M. Cvetic, G.W. Gibbons, H. Lu, C. Pope, Supersymmetric M3 branes and G2 manifolds, Nucl. Phys. B 620
(2002) 3, hep-th/0106026.
[7] F. Klein, Lectures on the ikosahedron and the solution of equations of the fifth degree. Translated by
G.G. Morrice. London: Trubner & Co., 1888.

Type II theories compactified on CalabiYau

threefolds in the presence of background fluxes
Jan Louis, Andrei Micu
Fachbereich Physik, Martin-Luther-Universitt Halle-Wittenberg,
Friedemann-Bach-Platz 6, D-06099 Halle, Germany
Received 18 March 2002; accepted 15 April 2002
Abstract
Compactifications of type II theories on CalabiYau threefolds including electric and magnetic
background fluxes are discussed. We derive the bosonic part of the four-dimensional low energy
effective action and show that it is a non-canonical N = 2 supergravity which includes a massive twoform. The symplectic invariance of the theory is maintained as long as the flux parameters transform
as a symplectic vector and a massive two-form which couples to both electric and magnetic field
strengths is present. The mirror symmetry between type IIA and type IIB compactified on mirror
manifolds is shown to hold for RR fluxes at the level of the effective action. We also compactify
type IIA in the presence of NS three-form flux but the mirror symmetry in this case remains unclear.
2002 Elsevier Science B.V. All rights reserved.
1. Introduction
CalabiYau compactifications of heterotic and type II theories have been studied
intensively in the past since they lead to consistent string theories below the critical
dimension d = 10. In particular compactifications on CalabiYau threefolds Y3 result in
four flat Minkowskian spacetime dimensions (d = 4) and a low number of unbroken
supersymmetries. Their effective theories are supergravities coupled to a set of vectorand matter multiplets with N = 1 supersymmetry in the case of the heterotic string and
N = 2 supersymmetry for type II strings. The low energy effective theories all share the
Work supported by: DFGThe German Science Foundation, GIFthe GermanIsraeli Foundation for
Scientific Research, European RTN Program HPRN-CT-2000-00148 and the DAADthe German Academic
Exchange Service.
E-mail addresses: j.louis@physik.uni-halle.de (J. Louis), micu@physik.uni-halle.de (A. Micu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 8 - 3
396
J. Louis, A. Micu / Nuclear Physics B 635 (2002) 395431
feature that they contain a (large) number of gauge neutral moduli multiplets which are flat
directions of the effective potential and thus parameterize the vacuum degeneracy of the
theory.
Generalization of CalabiYau compactifications are possible if one allows a p-form
field strength Fp to take a non-trivial background
value eI along appropriate cycles

I in the compact CalabiYau manifold, i.e., I Fp = eI . Depending on the choice
of these background fluxes the metric is deformed and the direct product of a fourdimensional Minkowskian spacetimes a CalabiYau threefold is replaced by a warped
product [1,2]. This generically introduces a potential for the moduli and turns an ordinary
supergravity into a gauged or massive supergravity. The consistency of such generalized
compactifications was discussed in [36] while various other aspects have been studied
previously in Refs. [1,2,719].
The background fluxes eI are quantized in units of the string scale [7,9] and thus do
not represent a continuous deformations of the CalabiYau compactification. However,
in the low energy supergravity they do appear as continuous parameters and hence can
be discussed as continuous deformations of the well-known low energy effective theories
derived for vanishing flux background [9]. In the gauged supergravity the flux parameters
play the role of masses and gauge charges.
The purpose of this paper is to perform a KaluzaKlein reduction of the ten-dimensional
type II supergravities on compact CalabiYau threefolds with all possible background
fluxes turned on. We derive the bosonic part of the resulting low energy effective action
and showwhenever possibleits consistency with gauged supergravity.1 We find that
if magnetic charges are turned on a two-form becomes massive and the resulting N = 2
supergravity is not easily related to the known N = 2 supergravities [2022]. We show
that the symplectic invariance of the ungauged N = 2 supergravity continues to hold as
long as the background flux parameters transform as a symplectic vector. This leads to
a symplectically invariant potential which for compactifications of type IIB theories was
previously derived in Refs. [1113]. However, the issue of the symplectic invariance was
not resolved and here we show that the presence of a massive two-form is crucial for the
symplectic invariance of the theory. We believe that this is a more general feature of gauged
supergravity and that the existing N = 2 supergravities have to be amended by including
the possibility of massive two-forms. Only in this more general framework a symplectically
invariant theory can arise.
A second aspect of our paper concerns the mirror symmetry between type IIA
3 . This duality holds
compactified on Y3 and type IIB compactified on the mirror manifold Y
for vanishing background fluxes but its validity in the presence of fluxes is unclear. We
show that for the background fluxes of RamondRamond (RR) p-forms mirror symmetry
holds at the level of the effective theories while for background fluxes in the Neveu
Schwarz (NS) sector it is not easily established [12,16,19,23]. A similar problem arises
for the non-perturbative dualities relating type IIA compactified on K3 to heterotic on T 4
[24] and type IIA compactified on Y3 to heterotic on K3 T 2 [16,18].2
1 For the heterotic string we performed a similar analysis in Ref. [18].
2 An interesting suggestion for its cure has been put forward in Refs. [25,26].
397
This paper is organized as follows. In Section 2 we focus on type IIA supergravity

compactified on CalabiYau threefolds Y3 in the presence of background fluxes. It turns out
that in order to establish the mirror symmetry to type IIB compactifications it is necessary
to start from the N = 2 massive version of the ten-dimensional type IIA supergravity
where the NS two-form B2 is massive [27]. We first briefly recall this theory including its
symmetry properties. In Section 2.1 we then perform the compactification on Y3 turning
2 and the RR
on the 2h1,1 possible background values of the RR two-form field strength F

four-form field strength F4 . (h1,1 is the Hodge number of the cohomology group H 1,1 (Y3 ).)
Furthermore, the four-dimensional low energy effective theory includes a three-form (with
a four-form field strength) which in d = 4 is Poincar dual to a constant. This constant is an
additional parameter of the theory so that together with the ten-dimensional mass parameter
of B2 the theory depends on 2h1,1 + 2 fluxes. The resulting low energy effective action
is not a standard N = 2 gauged supergravity in that also in d = 4 the two-form B2 is
generically massive. The mass terms depend on the magnetic flux parameters and vanish
for purely electric fluxes. As far as we know this situation has not been discussed previously
in the supergravity literature.
In Section 2.2 we discuss the gauge invariance of this massive theory and furthermore
show that as long as the flux parameters are appropriately transformed, the equations of
motions are invariant under a generalized electricmagnetic duality which is part of an
Sp(2h1,1 + 2) transformation. To establish this symmetry it is crucial to start from the
massive type IIA theory in d = 10 and to also include the constant dual of the threeform. A similar observation was made in [24] where K3 compactifications of type IIA in
d = 6 were studied. The perturbative SO(4, 20) T-duality which is present in the absence of
background fluxes can only be maintained if one starts from the massive ten-dimensional
IIA theory and furthermore transforms the flux parameter in the vector representation of
SO(4, 20). Nevertheless the appearance of the symplectic invariance in the d = 4 theory is
somewhat unexpected since it is usually lost in gauged supergravities.
In Section 2.3 we discuss the relation with the standard gauged N = 2 supergravities.
By fixing the symplectic invariance one can go to a particular gauge where all magnetic
charges vanish and B2 remains massless. We show that in this gauge the theory is a
special case of the known N = 2 gauged supergravities. However, in an arbitrary gauge
B2 is massive and the corresponding supergravity couplings are not independently known.
Nevertheless, in d = 4 a massive two-form is dual to a massive one-form [2830] and
therefore one might suspect that in the dual formulation consistency with the standard
gauged supergravity is achieved. In Section 2.3 (and Appendix E.3) we explicitly perform
the duality transformation and show that also in the dual basis the gauged supergravity is
non-canonical.
3 which
In Section 2.4 we turn our attention to non-trivial fluxes of the NS three-form H
have not been discussed previously in the literature. We derive the low energy effective
theory and establish the consistency with N = 2 gauged supergravity. In this case the
potential depends non-trivially on the scalar in the hypermultiplets whereas the dependence
on the scalars in the vector multiplets is only via an overall volume factor. This is exactly
opposite to the case of RR fluxes which induce a potential for the vector-scalars leaving
the hyper-scalars (except the dilaton) undetermined. Furthermore, only the graviphoton
398
participates in the gauging but no charged states with respect to the other gauge fields
appear.
In Section 3 we briefly discuss type IIB compactifications on Y3 with non-trivial RR
3 . This case has been considered previously in Refs. [1114,17] and
three-form flux F
therefore we keep our presentation short. However, we do establish the fact that also in
this case B2 becomes massive for non-vanishing magnetic charges. With this result we are
able to establish the mirror symmetry to type IIA compactified on the mirror threefold in
the presence of RR background fluxes at the level of the low energy effective theories.
Section 4 contains our conclusions and some of the more technical aspects of this
paper are relegated to five appendices. Appendix A summarizes our notation. Appendix B
briefly recalls N = 2 supergravity in d = 4. Appendix C assembles the necessary facts
about the moduli space of CalabiYau threefolds. Appendix D redoes the compactification
of massless type IIA on Y3 with special emphasis on the role of the RR three-form C3
and its Poincar dual constant. Already in this simpler case the constant is an additional
parameter of the effective theory and turns the ordinary type IIA supergravity into a gauged
supergravity. Finally Appendix E discusses the Poincar dualities of massless and massive
two-forms and of three-forms in d = 4.
2. Compactification of massive type IIA supergravity on CalabiYau threefolds with

background fluxes
Let us start by compactifying the ten-dimensional massive type IIA theory [27] on
a CalabiYau threefold in the presence of background fluxes. The compactification of
ordinary massless type IIA supergravity on CalabiYau threefolds was performed in
Ref. [31] which we briefly recall in Appendix D. The massless modes of the ten2 and
dimensional IIA theory comprise in the NSNS sector the metric, a two form B
3 . In the
the dilaton while the RR-sector contains a vector field A 1 and a three-form C
2 is massive and a cosmological constant is present. The action reads
massive version B
[27]3

1
1
H
S=
e2 R
1 + 2 d d H
3
3
2
4
1
1 2
m
F
(2.1)
+
F
1
,
+
L
F
F
2
2
4
4
top
2
2
where the field strengths are defined as
2 )2 ,
4 = d C
3 B
2 d A 1 m (B
3 = d B
2 ,
F
H
2
and the topological terms read

1
2 )3 d A 1 d A 1
3 d C
3 (B
2 )2 d C
3 d A 1 + 1 (B
Ltop = B2 d C
2
3
2 = d A 1 + mB
2 ,
F
3 This differs from [27] by a redefinition of the mass parameter m m.
(2.2)
2
m 3
2 )4 d A 1 + m (B
2 )5 .
3 + m (B
(B2 ) d C
3
4
20
399
(2.3)
Throughout the paper we use differential form notation (summarized in Appendix A) and
2 B
2 = (B
2 )2
denote differential forms in d = 10 by a hat . Furthermore we abbreviate B
etc. The massive IIA supergravity has an unbroken ten-dimensional N = 2 supersymmetry
and in the limit m 0 the action for ordinary type IIA theory (recorded in (D.1))
is recovered. The action (2.1) is invariant under the following three Abelian gauge

2 ,
1 )
transformations (with parameters ,

A 1 = d ,
2 = d
1 ,
B
3 = d
2 ,
C
3 =
1 d A 1 ,
C
1 .
A 1 = m
(2.4)
As already mentioned, in the limit m 0 we recover the standard type IIA supergravity
and hence the degrees of freedom described by the two theories are the same. However,

in the massive theory these degrees
of freedom are redistributed in that B2 describes a
massive two-form (which carries 92 = 36 degrees of freedom) while A 1 carries no degree
of freedom since it can be gauged away using (2.4). This is the analog of the unitary gauge
in the standard Higgs mechanism where A 1 plays the role of the Goldstone boson which is
2 .
eaten by B
Compactification of massive IIA supergravity on a CalabiYau threefold Y3 results in
an effective theory in d = 4 with N = 2 supersymmetry and proceeds as in the massless
type IIA case (cf. Appendix D). The ten-dimensional metric gives rise to h1,1 + 2h1,2
scalar fields related to h1,1 Khler deformations v i , i = 1, . . . , h1,1 and the h1,2 (complex)
2 decomposes
deformations za , a = 1, . . . , h1,2 of the complex structure.4 The two-form B
into a two form B2 in d = 4 and h1,1 scalar fields bi according to
2 = B2 + bi i ,
B
i = 1, . . . , h1,1 ,
(2.5)
where i are harmonic (1, 1)-forms which form a basis of H 1,1(Y3 ). The b i combine with
3 decomposes into a fourthe v i to form complex fields t i = bi + iv i . The three-form C
i
dimensional three form C3 , h1,1 vector fields A and 2h1,2 + 2 (real) scalar fields A , A
3 = C3 + Ai i + A A + A A ,
C
A = 0, . . . , h1,2 ,
(2.6)
where (A , A ) are harmonic three-forms which form a real basis of H 3 (Y3 ) (cf.
Appendix C.2). Finally the 1-form in d = 10 only results in a four-dimensional 1-form,
i.e., A 1 = A0 . Together these (bosonic) fields assemble into the N = 2 gravity multiplet
(g , A0 ), h1,1 vector multiplets (Ai , t i ), h1,2 hypermultiplets (za , a , a ) and one tensor
multiplet (B2 , , 0 , 0 ). In d = 4 a two-form is dual to a scalar and hence the tensor
multiplet can be dualized to an additional (universal) hypermultiplet. A four-dimensional
three form C3 is dual to a constant and thus the compactified massive type IIA theory
already contains two parameters, the mass m of the ten-dimensional action and the constant
e0 which is dual to C3 . Both of these constants are related to the cosmological constant
4 A summary of the CalabiYau geometry is assembled in Appendix C.
400
and as we will see shortly they turn the ordinary N = 2 supergravity into a gauged
supergravity.5
2.1. Turning on RR fluxes
Let us now turn to the compactification of massive type IIA supergravity in the presence
2 and F
4 . This will add 2h1,1 parameters
of background fluxes for the RR field strengths F
i
(ei , m ) into the action. We assume that the fluxes are turned on perturbatively so that the
light d = 4 spectrum is not modified. More specifically we start from the standard reduction
ansatz (cf. Appendix D)
2 = B2 + bi i ,
B
A 1 = A0 ,
3 = C3 + Ai i + A A + A A ,
C
(2.7)
and modify the RR-field strength according to6

3 d C
3 + ei i ,
dC
d A 1 d A 1 mi i .
(2.8)
The flux parameters (ei , mi ) are constants and i are harmonic (2, 2)-forms which
form a basis of H 2,2 (Y3 ) and which are dual to the (1, 1)-forms i , i.e., they obey the
normalization of Eq. (C.8). Notice that we can consistently do this modification at the
3 appear only through their Abelian
level of the action since in (2.1) the fields A 1 and C

field strengths d A1 and d C3 . In terms of the field strengths defined in (2.2) turning on
fluxes according to (2.8) amounts to
3 = dB2 + dbi i ,
H

2 = dA0 + mB2 mi mbi i ,
F

4 = dC3 B2 dA0 m (B2 )2 + dAi dA0 bi + mi B2 mB2 bi i
F
2

A
1 i j
A
i j
+ d A + d A + b m mb b Kij k k + ei i ,
(2.9)
2

where Kij k = Y3 i j k and the last equation used (C.9).
To derive the four-dimensional effective action we insert (2.9) into (2.1). Before giving
the final result let us discuss the new terms which arise due to the presence of the
parameters (e0 , ei , m, mi ) and are absent in the standard IIA theory. The kinetic term of
A 1 gives a contribution to the potential

V1 = 2K mi mbi mj mbj gij ,
(2.10)

where gij (v) = 41K Y3 i j is the metric on the space of Khler deformations and K
defined in (C.3) denotes the volume of Y3 . In addition the following interaction and mass
5 In Ref. [31] the case e = 0 was considered. In Appendix D we derive the action of massless type IIA
0
compactified on Y3 for arbitrary e0 . The quantization condition on e0 has been discussed in Refs. [32].
6 The minus sign in the last relation was chosen to make the symplectic invariance explicit later.
401
terms for B2 arise

m2 K
B2 B2 .
(2.11)
2
3 also contributes to the potential
The kinetic term of C

1
1 k l
1 m n
k l
m n
V3 =
ei + b m Kikl mb b Kikl ej + b m Kj mn mb b Kj mn g ij ,
8K
2
2
(2.12)

k
ij
i
j
j
k
where g = 4K Y3 is the inverse metric obeying gij g = i . In addition the
following interaction terms arise

Lint = 4K mi mbi B2 dAj dA0 bj gij

2K mi mbi mj mbj gij B2 B2

m
m
K
dC3 B2 dA0 (B2 )2 dC3 B2 dA0 (B2 )2 .

2
2
2
(2.13)
Finally the topological terms (2.3) are modified according to

Ltop = B2 dAi ei + bi dAj mk Kij k bi ei dA0 bi bj mk Kij k dA0

1
m
2b i ei + bi bj mk Kij k b i bj bk Kij k dC3

2
3

i

m
1
mi ei mbi ei + bi mj mk Kij k
+ B2 dA dA0 bi bj bk Kij k
2
2

3m i j k
m2 i j k
b b m Kij k +
b b b Kij k (B2 )2 .
(2.14)
2
2
To arrange the above expressions in the form of the standard gauged N = 2 supergravity
we are going to proceed as in Appendix D. First we dualize the three-form C3 to a constant.
Collecting all couplings of C3 we obtain

K
m
m
0
2
0
2
LC3 =
dC3 B2 dA (B2 ) dC3 B2 dA (B2 )
2
2
2

1
m
bi ei + bi bj mk Kij k b i bj bk Kij k dC3 .
(2.15)
2
6
The dualization of a three-form in d = 4 is summarized in Appendix E.2. Applying
the formulae of this appendix yields the dual action where the three-form C3 is traded
for a constant e0

2
1
1
m
e0 + ei bi + bi bj mk Kij k b i bj bk Kij k 1
LC3 Le0 =
2K
2
6

1
m
e0 + ei bi + bi bj mk Kij k b i bj bk Kij k
2
6

m
B2 dA0 + (B2 )2 .
(2.16)
2
Lint = mKB2 dA0
402
Let us stress that the appearance of the parameter e0 obtained by dualizing C3 does not
depend on the fact that we have turned on other fluxes. Le0 does not vanish in the limit
m = mi = ej = 0 and thus is also present in the compactification of massless type IIA
supergravity without any fluxes turned on. In Appendix D we show that e0 becomes the
charge of the scalar a which is dual to B2 and a potential consistent with the standard
N = 2 gauged supergravity is induced.
Defining the four-dimensional dilaton via e2 = e2 K, using the formulae (D.5),

(D.6) together with (2.10)(2.14) and (2.16) the low energy effective action takes the form7

1
1
S = e2 R 1 + 2 d d H3 H3
2
4

gab dza d z b gij dt i d tj
AB

1
BD d D
Im M1
d A + MAC d C d B + M
2
1

1
1
+ H3 A d A A d A + Im NI J F I F J + Re NI J F I F J
2
2
2
1 2
1 2
B2 J2 M B2 B2 MT B2 B2 V ,
(2.17)
2
2
where I = 0, . . . , h1,1 and NI J , MAB are standard supergravity couplings defined in (C.7)
and (C.23). The new couplings J2 , M 2 , MT2 which depend on the fluxes only appear as
couplings to B2 and are found to be8

J2 = eI F I mI GI ,
+
M 2 = mI Im NI J mJ ,
MT2 = mI Re NI J mJ + mI eI ,
(2.18)
where we denoted m by m0 and introduced the vectors mI = (m0 , mi ), eI = (e0 , ei ).

Furthermore, we introduced the magnetic dual of F I dAI by
GI Im NI J F J + Re NI J F J .
(2.19)
Finally the string frame potential in (2.17) is found to be

V =

1
I K mK (Im N )1I J eJ NJ L mL ,
eI N
2
(2.20)
where (Im N )1 is given in (C.11).

The action (2.17) together with the definitions (2.18) is our first non-trivial result. It
gives the low energy effective action for massive type IIA supergravity compactified on
a CalabiYau threefold in the presence of RR-background flux. As we see the 2h1,1 flux
parameters (ei , mi ) naturally combine with the mass parameter m0 of the ten-dimensional
7 Strictly speaking also the Khler moduli t i have to be redefined by a dilaton dependent factor [31]. In order
not to overload the notation we use the same symbol t i also for the redefined moduli.
8 Note that M 2 is positive since in our conventions Im N
I J is negative definite.
403
massive IIA theory and the dual e0 of the four-dimensional three-form C3 to form
the vectors (eI , mI ). As we are going to show next these vectors enjoy an action of
a symplectic group Sp(2h1,1 + 2). Furthermore the flux parameters introduce a potential V ,
GreenSchwarz type couplings B2 J2 , a regular and a topological mass term M, MT
for B2 .
2.2. Gauge and symplectic invariance
Let us first focus on the gauge invariance of the action (2.17). First of all (2.17)
is manifestly invariant under the standard one-form gauge transformation AI = d I .
However, the gauge invariance related to the two-form B2 is less obvious. After
compactification the ten-dimensional gauge transformations (2.4) of the two-form B2
become
B2 = d1 ,
C3 = 1 dA0 ,
AI = mI 1 .
(2.21)
As in d = 10, the three-form C3 transforms under this gauge transformations. However,

in the dualization of C3 the gauge invariant combination dC3 dA0 B2 m2 B2 B2
appeared and is dual to the (gauge invariant) constant e0 . The gauge invariance in the dual
action is most easily seen by rewriting the action (2.17) as

1
2 1
R 1 + 2 d d H3 H3
S= e
2
4

a
b
i j
gab dz d z gij dt d t
AB

1
BD d D
Im M1
2
1

1
1
+ H3 A d A A d A + Im NI J F I F J + Re NI J F I F J
2
2
2

I
1
I
B2 F + dA eI V ,
(2.22)
2
where we defined
+
F I dAI + mI B2 .
(2.23)
Under the transformations

B2 = d1 ,
F I
AI = mI 1 ,
(2.24)
is invariant and it can be easily checked that the action (2.22) is also not modified.
Let us discuss the symplectic invariance of the theory described by (2.22). The
ungauged N = 2 supergravity is invariant under generalized electricmagnetic duality
transformations which are part of a symplectic Sp(2h1,1 + 2) invariance [20,22]. However,
this is not a symmetry of the action, but it leaves the equations of motion and Bianchi
identities invariant. In gauged supergravity this invariance is generically broken as charged
states appear and the action is no longer expressed in terms of only the field strength F I .
However, for the case at hand a symplectic invariance can be maintained as long as the
404
fluxes (mI , eI ) are also transformed. To see this consider the Bianchi identities and the
equations of motions derived from the action (2.22)
d dAI = d F I mI dB2 = 0,
L
I eI dB2 = 0,
= dG
AI

1
L
I eI F I = 0,
= d e2 dB2 + mI G
B2 2
where
I Re NI J F J + Im NI J F J .
G
(2.25)
(2.26)
These equations are invariant under the symplectic transformations given in (B.10) with
I ) transforming as symplectic vectors
(mI , eI ) and (F I , G

I
I

I
I
U Z
m
U Z
m
F
F
,
(2.27)
I W V
I ,
W V
G
G
eI
eI
where U , V , W , Z obey (B.11). Similarly one checks the invariance of V given in (2.20).
The symplectic invariance of the Eq. (2.25) is our second non-trivial result. In particular
it shows that the symplectic invariance of the potential as observed in [12,13] has its deeper
origin in the symplectic invariance of (2.25). However, the supergravity which displays this
invariance is not the canonical one but instead features a massive two-form B2 with very
specific couplings to the gauge fields. It would be interesting to investigate this situation
in more detail from a purely supergravity point of view without any reference to a flux
background of string theory.
2.3. Relation with gauged supergravity
Let us now investigate the relation between the action derived in (2.17) or (2.22) and the
standard gauged N = 2 supergravity as summarized in Appendix B. The new ingredients
in the action (2.17) are the mass terms for B2 . Let us first observe that they all vanish for
mI = 0. Since we have established the symplectic invariance of the theory we can always
do a symplectic transformation on the vector (mI , eI ) and go to a basis where all mI
vanish9
I

m
0
(2.28)
.
eI
eI
In this basis the new couplings considerably simplify and from (2.18) and (2.20) one
immediately obtains
M = MT = 0,
J2 = eI F I ,
1
V = eI (Im N )1I J eJ ,
2
(2.29)
9 We should stress again that from a pure supergravity point of view the fluxes mI , e are just continuous
I
parameters and so there always exist an Sp(2h1,1 + 2, R) transformation such that the rotated magnetic fluxes
vanish. In a quantum theory however, the fluxes become quantized and the Sp(2h1,1 + 2, R) invariance is
generically broken to Sp(2h1,1 + 2, Z). In this case, it is impossible to set the magnetic charges to zero by
an Sp(2h1,1 + 2, Z) rotation.
405
where the prime indicates the rotated basis. The drawback of this basis is that also the
gauge couplings N of the action (2.17) change according to (B.13) and the relation to
the prepotential as given in (B.7) is more complicated. So we have the choice to work
either with the standard gauge couplings and a set of complicated interactions of B2 or to
transform to a new basis where the gauge couplings are more complicated but B2 remains
massless. In this latter basis the consistency with gauged supergravity is easily established
so let us first discuss this case.
For mI = 0 B2 is massless and thus can be dualized to a scalar a as in Appendix E.1.
After a Weyl rescaling g e2 g the dual action reads

1
1
1
R 1 + Im NI J F I F J + Re NI J F I F J
S=
2
2
2
i j
u
v
gij dt d t huv Dq Dq VE ,
(2.30)
where
huv Dq u Dq v
= d d + gab dza d z b

e4
Da + A d A A d A Da + A d A A d A
+
4
AB

e2
BD d D ,
Im M1
2
(2.31)
and
Da = da + 2eI A I .
(2.32)
The covariant derivative of a arises from the GreenSchwarz type interaction B2 J2

in (2.17) and as a consequence a couples like a Goldstone boson and is charged under an
Abelian gauge symmetry with gauge charges eI . VE represents the potential in the Einstein
frame and is given by
e4
e (Im N )1I J eJ .
(2.33)
2 I
In Ref. [33] it was shown that huv of (2.31) is a quaternionic metric in accord with the
constraints of N = 2 supergravity that the scalars in the hypermultiplets span a quaternionic
manifold. In order to establish the further consistency with gauged N = 2 supergravity we
need to show that the potential (2.33) is consistent with the general form of the potential
(B.18) of gauged supergravity. Let us first note that only one scalar a in the hypermultiplets
carries gauge charge while the scalars t i in the vector multiplets remain neutral. In terms
of the Killing vectors defined in Eqs. (B.14) and (B.17) (2.32) implies
VE =
kIu = 2eI ua ,
kIi = 0.
(2.34)
Inserted into (B.18) using (2.31) one arrives at

VE =
I J

1
J e4 eI eJ P x P x .
(Im N )1 PIx PJx + 4eK XI X
I J
2
(2.35)
406
We are left with the computation of the Killing prepotentials PIx defined in (B.16).
Following Ref. [11] one first observes that for the constant (field independent) Killing
vectors as in Eqs. (2.34), (B.16) are solved by
PIx = kIu ux ,
x = 1, 2, 3,
(2.36)
where ux is the SU(2) connection on the quaternionic manifold. For the case at hand ux
has been computed in [33] and here we only need their result ax = 12 e2 3x . Inserted into
(2.36) using (2.34) we obtain
PI1 = PI2 = 0,
PI3 = e2 eI ,
(2.37)
which implies PIx PJx = huv kIu kJv . Thus the last term in (2.35) vanishes while the first
one reproduces the potential (2.33). This establishes the consistency with N = 2 gauged
supergravity.
Let us return to the discussion of the action in the unrotated basis where both eI and
mI are non-zero. In this case B2 is massive and the relation with the standard gauged
supergravity is not obvious and, as far as we know, has not been discussed in the literature.
However, one can use the fact that a massive two-form in d = 4 is Poincar dual to a
massive vector [2830]. This generic duality is briefly summarized in Appendix E.3. In the
following we perform the duality transformation and display the dual action in terms of
only vector fields.
Starting from the action (2.17) it is straightforward to apply the results in Appendix E.3.
Denoting by AH the dual of the massive B2 the resulting action reads

1
S = e2 R 1 + 2 d d gab dza d z b gij dt i d tj
2
AB

1
BD d D V
+ Im M1
2
1
1
+ Im NI J F I F J + Re NI J F I F J e2 AH AH
2
2
H

M2
1
F J2 F H J2
2 M 4 + MT4
+
H

MT2
1
F J2 F H J2 ,
2 M 4 + MT4
(2.38)
where
F H = dAH ,

1
J2 = J2 + d A d A A d A ,
2
(2.39)
and the quantities M, MT and J2 are defined in (2.18). The above action contains an
explicit mass term for the vector field AH which can equivalently be written as the
covariant derivative of a Goldstone boson
1
e2 AH AH = e2 Da Da,
4
(2.40)
407
where
Da = da + 2A H .
(2.41)
(A H denotes the gauge transformed vector potential.) Inserting (2.40) into (2.38) and
absorbing 12 (A d A A d A ) into a further redefinition of AH results in

1
R 1 gij dt i d tj huv Dq u Dq v
S=
2
1
F I F J + 1 Re N
F I F J VE ,
+ Im N
(2.42)
IJ
IJ
2
2
where also a Weyl rescaling g e2 g has been performed and we introduced the
index I = (I, H ). VE is the Weyl rescaled potential related to V of (2.20) by VE = e4 V .
huv Dq u Dq v is again the standard quaternionic metric given in (2.31) with the only
difference that (2.32) is replaced by (2.41). Moreover, the new (h1,1 + 2) (h1,1 + 2) is given by
dimensional gauge coupling matrix N
IJ

K
I J = NI J i eI NI K m
H H = i,
N
eJ NJ L mL ,
N

I H = i eI NI K mk ,
N
M 2 + iMT2
M 4 + MT4
(2.43)
I J mJ = eI and hence Im N
has a null vector while Re N
has
One easily shows that N
IJ
IJ
one constant eigenvalue. This implies that one (linear combination) of the vector fields
only has a topological coupling.10
The dualization of B2 resulted in an additional massive vector AH and we chose to write
the mass term as the coupling of a Goldstone boson a. The number of physical degrees of
freedom is of course unchanged since the action (2.38)/(2.42) is still invariant under the
gauge transformations (2.24) which after dualization become
AI = mI 1 ,
AH = 0.
(2.44)
AH being the Poincar dual of H3 is invariant under (2.24) but one of the other h1,1 + 1
vector fields in (2.42) can be gauged away by (2.44). In this unitary gauge the symplectic
invariance is lost. Thus the theory can be formulated in terms of only vector fields but
symplectic invariance demands the presence of an additional auxiliary vector field with
only topological couplings. In any physical gauge the symplectic invariance is broken.
To conclude this section let us discuss another aspect of the dualization of the massive
I in terms of electric and magnetic potentials
B-field. Eq. (2.25) can be solved for F I and G
AI and A I
F I = mI B2 + dAI ,
I = eI B2 + d A I .
G
(2.45)
Now the equation of motion for B2 becomes

1 2
d e
dB2 + mI d A I eI dAI = 0.
2
10 We thank B. de Wit and S. Vandoren for discussions on this point.
(2.46)
408
This suggests that we can introduce a scalar field a (the dual of B2 ) which obeys
e2 dB2 = Da da 2mI d A I + 2eI dAI .
(2.47)
This definition has the feature that it maintains explicitly the symplectic invariance closely
related to the proposal of [11,13]. However, in (2.45) B2 and not dB2 appears and thus it
is not possible to give an action in terms of the dual scalar a with electric and magnetic
couplings. Nevertheless, one can compute the electric and magnetic Killing prepotentials
corresponding to the gauging (2.47) as suggested in [11,13]. They are very similar to the
ones found only for the electrically charged particles (2.37)
PI3 = e2 eI ,
I 3 = e2 mI ,
P
I 1 = PI 2 = 0.
PI1 = PI2 = P
(2.48)
Using the formula for the potential suggested in [11]

KI
J huv k u k uK NKI kIv k vK N
VE = 4eK XI X
I

1
KJ , (2.49)
J PIx PKx NKI PJx PKx N
(Im N )1I J + 4eK XI X
2
which is the symplectic invariant extension of (B.18), one immediately recovers the
potential obtained in (2.20).
2.4. NS fluxes
So far we concentrated on non-trivial flux related to the modification of the RR field
3 and d A 1 . In this section we discuss the modifications which appear in the
strength d C
type IIA compactification due to the presence of NS fluxes. For the consistency of the
procedure, we require that the fields which acquire a background value appear in the action
only via the corresponding field strengths. This is easily achieved for the RR fields A 1 and
3 as can be seen in the form of the actions (2.1). Clearly, for massive type IIA theory,
C
2 cannot appear only via its field strength so we only have a chance to
the NSNS field B
turn on NS fluxes if we start from the massless type IIA theory given in (D.1). However
3 + A 1 B
2 in order to have the action
3 C
we need to perform the field redefinition C
2 . This turns (D.1) into
3 but not B
only depend on the field strength H

1
1
S = e2 R
1 + 2 d d H
3 H3
2
4
1 1
3 ,
F2 F2 + F4 F4 + H3 C3 d C
(2.50)
2
2
3 A 1 H
3 and H
3 = d B
2 . Note that this field redefinition changes the
4 = d C
where F
form of the gauge transformations (D.3) which now become
2 = d
1 ,
B

A 1 = d ,
3 = d
2 ,
C
3 =
2 .
dB
C
(2.51)
The compactification of this action proceeds as in Appendix D with the following

modification of the reduction ansatz (D.4)
3 = H3 + dbi i + pA A + qA A ,
H
(2.52)
409
3 . Using
where H3 = dB2 and (pA , qA ) are the 2h1,2 + 2 possible flux parameters of H

(2.52) F4 is reduced according to

4 = dC3 A0 H3 + dAi A0 dbi i + D A A + D A A ,
F
(2.53)
where
D A = d A pA A0 ,
D A = d A qA A0 .
(2.54)
Inserting (2.52) and (2.53) into the action (2.50) results in terms very similar to (D.5). The
3 now contains the covariant derivatives D A ,
differences are that the kinetic term for C
A
2 (D.5)
D A instead of the ordinary ones d , d A . Furthermore, the kinetic term for B

now induces a potential
V =

1 e2
BD pD ,
qA + MAC pC (Im M)1AB qB + M
4 K
(2.55)
where M is defined in (C.23). The last modification due to (2.52) appears from the
topological term in (2.50) which reads

Ltop = pA A qA A dC3 .
(2.56)
As before we have to dualize the three-form C3 to a constant. Collecting the terms

including C3 we find
LC3 =

K
dC3 A0 H3 dC3 A0 H3 + pA A qA A dC3 .
2
(2.57)
By using the formulas of Appendix E.2 we obtain the dual Lagrangian

LC3 Le =
2

1 A
p A qA A + e 1 + pA A qA A + e A0 H3 , (2.58)
2K
where e is an arbitrary constant, the dual of C3 . The first term in the above expression
contributes together with (2.55) to the scalar potential which reads
V =

e2
+ 1 pA A qA A + e 2 .
(q + pM)(Im M)1 (q + pM)
4K
2K
(2.59)
The final thing to do in order to have the compactified action in the standard form of
a gauged supergravity is to dualize B2 to a scalar. Collecting the terms including B2 we
find
e2
H3 H3 +
4
e2
H3 H3 +
=
4
LB2 =

1
H3 A d A A d A 2 pA A qA A + e A0
2

1
H3 A D A A D A pA A qA A + 2e A0 .
2
(2.60)
410
Using Appendix E.1 we obtain the following action for the dual scalar a
LB2 La =

e2
Da + A D A A D A Da + A D A A D A ,
4
(2.61)
where

Da = da pA A qA A A0 2eA0.
(2.62)
The final form of the action is obtained after going to the Einstein frame and is similar to
the one found in the massless case (D.13)

1
S=
2
1
1
+ Im NI J F I F J + Re NI J F I F J VE ,
(2.63)
2
2
where now
huv Dq u Dq v

e4
Da + A D A A D A Da + A D A A D A
+
4
AB

e2
BD D D ,
(2.64)
Im M1
D A + MAC D C D B + M
2
with the covariant derivatives D A , D A and Da given in (2.54) and (2.62). The Einstein
frame potential will only differ from (2.59) by a factor of e4
4

e2
+ e pA A qA A + e 2 .
(2.65)
(q + pM)(Im M)1 (q + pM)
4K
2K
The crucial difference to the previous case of RR fluxes is that now the potential depends
on the scalars in the hypermultiplets but not on the scalars in the vector multiplets.
As before the fluxes turn ordinary supergravity into a gauged supergravity where
a certain isometry of the scalar manifold has been gauged. The corresponding gauge
invariance is just the compactified version of the 10-dimensional gauge invariance (2.51).
Inserted into the compactification ansatz (D.4) modified according to (2.52), the fourdimensional gauge invariance reads
VE =
A0 = d,
= p ,
A
dAi = dbi ,
A = qA .
C3 = H3 ,
(2.66)
The covariant derivatives defined in (2.54) and the action (2.61) is invariant under this
transformations provided a transforms according to

a a + 2e + pA A qA A .
(2.67)
As before let us establish the consistency of the theory with the standard gauged
supergravity. Compared to the massless type IIA theory the effect of the non-trivial
411
(pA , qA ) NS-fluxes is the replacement of ordinary derivatives by the covariant derivatives

(2.54), (2.62) and the appearance of the potential (2.59). Otherwise the structure of the
theory is unchanged. To show agreement with gauged supergravity we need to demonstrate
that with the gaugings (2.54) and (2.62) the potential (B.18) reduces to (2.65). The key
point here is to notice that for the case at hand (B.18) considerably simplifies in the sense
that the term which contains the Killing prepotentials vanishes. To see this we first note
that since only one vector field A0 participates in the gaugings (2.54) and (2.62), the only
non-trivial components of PIx will be the ones for which I = 0. Using (C.5), (C.11) and
the fact that X0 = 1 one immediately sees that
1
0 = 0.
(Im N )100 + 4eK X0 X
2
Inserting (2.68) into (B.18) we arrive at
VE = 4eK huv k0u k0v ,
(2.68)
(2.69)
where we already used ki = 0. The Killing vectors k0u can be read off from the covariant
derivatives (2.54) and (2.62) to be

B
k = pB ,
(2.70)
k B = qB ,
k0a = 2e + pA A qA A .
0
Using the metric components of the charged scalars from (2.64), the evaluation of (2.69)
precisely results in the potential (2.59) and thus establishes the consistency with gauged
supergravity.
As we said before the difference for NS-fluxes is that the potential depends on the
scalars in hypermultiplets (, za , A , A ) while the vector multiplet scalars t i remain
undetermined. Furthermore, the only gauge field which appears in the covariant derivatives
is the graviphoton A0 while all other Ai do not participate in the gauging.
3. Type IIB compactified on CalabiYau threefolds with background fluxes and

mirror symmetry
So far we concentrated on type IIA theories compactified on CalabiYau threefolds
Y3 and derived the four-dimensional effective theory when non-trivial background fluxes
are turned on. Without fluxes these theories are equivalent to type IIB compactified on the
3 . At the level of the low energy effective action the precise map between
mirror threefold Y
these compactifications was derived in Refs. [3335]. However, when background flux is
turned on, the validity of perturbative and non-perturbative dualities become obscure and
is not fully understood at present [16,19,2326,36]. Before we discuss this issue in more
detail let us focus on type IIB compactification with background fluxes. Without fluxes
the effective theory is given in [34,35] while turning on three-form flux was considered
in Refs. [1114,17]. Here we do not redo the computation in detail but focus on the
situation where both electric and magnetic charges (eA , mA ) are present. For this case the
potential has been computed in [11,12] but the complete couplings of the two forms were
not derived. (Ref. [17] only considered the case mA = 0.) The purpose of this section is to
give the complete bosonic Lagrangian including the couplings and mass terms of the NS
412
two-form B2 and to discuss the mirror symmetry to type IIA compactification with fluxes
as derived in the previous section. For simplicity we only turn on the background fluxes of
the RR two-form C2 .
Let us start by recalling the structure of the ten-dimensional type IIB supergravity. The
2 and the dilaton ,
while the RR sector
NSNS sector features the graviton, a two-form B
contains a second scalar l, a second two form C2 and a four form A4 with a self-dual field
strength. The bosonic part of the low energy effective action in ten dimensions reads [37]11

1
1
SIIB = e2
H
R 1 + 2 d d H
3
3
2
4

1
5 F
3 + 1 F
5 1 A 4 H
3 F
3 d C
2 ,
(3.1)
dl dl + F
2
2
2
where the field strengths are defined as12
3 = d B
2 ,
H
3 = d C
2 ld B
2 ,
F
5 = d A 4 + B
2 d C
2 .
F
(3.2)
When compactified on a CalabiYau threefold the resulting low energy spectrum

contains as in type IIA h1,1 Khler deformations v i , i = 1, . . . , h1,1 of the CalabiYau
metric and h1,2 complex deformations za , a = 1, . . . , h1,2 of the complex structure. The
2 , C
2 and the four-form A 4 decompose according to
doublet of two-forms B
2 = C2 + ci i ,
2 = B2 + bi i ,
C
B
d A 4 = dD i i + F A A GB B + di i ,
2
(3.3)
where as before (A , B ) span a real (2h2,1 + 2)-dimensional basis of H 3 (Y3 ), i is a h1,1 dimensional basis of H 1,1(Y3 ) and i is the dual basis on H 2,2 (Y3 ). The normalizations
of these basis are given in Appendix C. The self-duality condition F5 = F5 implies that
the GB are the dual magnetic field strengths of F A while i are the duals of the tensors
D i . Together these fields combine into a gravitational multiplet with bosonic components
(g , A0 ), a double-tensor multiplet (B2 , C2 , , l), h1,1 tensor multiplets (D2i , v i , bi , ci )
and h1,2 vector multiplets (Aa , za ).13 Dualizing the two-forms to scalars turns the tensor
and double tensor multiplets into hypermultiplets each containing four real scalars. In
this dual basis the low energy spectrum features h1,1 + 1 hypermultiplets and h1,2 vector
multiplets apart from the gravitational multiplet.
2
Next we turn on background fluxes for the RR three-form by modifying d C
2 d C
2 + mA A eA A ,
dC
(3.4)
11 In this action the self-duality condition on the 5-form field strength has not been imposed. A covariant action
including the self-dual 4-form has been constructed in [38]. The field equations for type IIB supergravity were
originally derived in Refs. [39].
1
12 Commonly one uses the convention F = d A 1 C

2 2 d B2 + 2 B2 d C2 which is obtained from F5 by
5
4
the redefinition A4 = A4 12 B2 C2 . Note that this redefinition modifies the topological term only by a total
derivative while the self-duality constraint F5 = F5 remains unchanged.
13 We consider AA to be electric potentials, i.e., F A = dAA .
3 turns into
where eA , mA are the constant background fluxes. The field strength F

3 = F3 lH3 + dci ldbi i + mA A eA A ,
F
413
(3.5)
where F3 = dC2 and H3 = dB2 and the only modification due to (3.4) coming from the
term F3 F3 is a potential

1
(3.6)
eA mC MCA (Im M)1AB eB MBD mD .
2
The matrix M is defined in (C.23) and to obtain the above expression we used (C.20) and
5 (3.2) which is modified according
2 appears in the definition of F
(C.22). Furthermore, d C
to

5 = dD i + bi dC2 + B2 dci i + F A A
F
2

A A + di + bj dck Kij k i ,
G
(3.7)
V =
where we defined
F A F A + mA B2 ,
A GA + eA B2 ,
G
(3.8)
and used (C.9) to express the product of harmonic two forms as a harmonic four form.
Finally, the topological term produces a GreenSchwarz type interaction

1 A
(3.9)
F eA GA mA B2 .
2
In the standard compactification of the type IIB theory the four-dimensional action is
5 . This however, cannot be done
obtained after imposing the self-duality condition for F
5 = F
5 in (3.1) the kinetic term for A 4 vanishes identically and
directly since imposing F
the same happens with the kinetic terms of the fields which come from the reduction of A 4 .
The correct 4-dimensional action is obtained by adding appropriate Lagrange multipliers in
order to impose the self-duality condition. At the level of the reduced fields this condition
can be obtained from (3.7) using the expressions for the Hodge duals of the harmonic forms
on a CalabiYau threefold (C.9), (C.20), (C.22) and reads
Ltop =
A = Im MAB F B + Re MAB F B .
G
(3.10)
Similarly one obtains

di + Kij k bj dck = 4Kgij dD j + bj dC2 + B2 dcj .
(3.11)
Strictly speaking this relation should involve field strengths as in (3.10) rather than only
exterior derivatives of some potentials. However, since the fluxes play no role in this
relation we do not need to be more precise here and instead rely to the standard Calabi
Yau compactification of type IIB theory. On the other hand, (3.10) does depend on the flux
defined in (3.8) and reduces to the usual formula (B.8) which
parameters through F and G
relates the electric and magnetic field strengths only for vanishing fluxes. To obtain the
Lagrangian for the vector fields we proceed as in the massless case [17]. First, treating F A
and GA as independent fields, inserting (3.7) into (3.1) and taking into account (3.9) we
414
obtain

1
BD
A F C MAC G
B F D M
L(F, G) = (Im M)1AB G
4

1
F A eA GA mA B2 .
(3.12)
2
Furthermore, to impose (3.10) as the equation of motion for G we have to add to (3.12) the
term 12 F A GA . Eliminating GA using (3.10) one is left with the following Lagrangian
for F A

1
1
L(F ) = F A GA F A eA GA mA B2
2
2

1
1
1
A
= Im MAB F F B + Re MAB F A F B F A eA + F A eA B2 .
2
2
2
(3.13)
The scalar sector is not modified by the introduction of fluxes in (3.4) and so we can use
the results in the literature [3335] to obtain the following 4-dimensional action

1
1
S = e2 R 1 + 2 d d H3 H3
2
4

gab dza d z b gij dt i d tj
I J

1
J L d L
Im N 1
d I + NI K d K d J + N
2
1

1
1
+ H3 A d A A d A + Im MAB F A F B + Re MAB F A F B
2
2
2
A

1
A
B2 F + dA eA V ,
(3.14)
2
where V is given in (3.6) and the scalars I and I are functions of l, ci , i and the dual
of C2 specified by the mirror map [35].
Mirror symmetry at the level of the low energy effective actions is now obvious since
the (2.22) and (3.14) have the same form. The only difference is the range of indices
and the fact that the matrices N and M are interchanged which just expresses the mirror
symmetry h1,1 h1,2 . Moreover the potentials (3.6) and (2.20) are the same and the flux
parameters (mA , eA ) in type IIB case are the mirror partners of (mI , eI ) in type IIA. Due
to the appearance of F in (3.14) B2 is also massive in IIBa fact which has not been
stressed previously. Finally, the two extra parameters in type IIA, the one coming from the
mass m of the massive type IIA supergravity and the other coming from the dualization of
the 4-dimensional C3 to a constant e0 , are crucial in order to make mirror symmetry work
when RR fluxes are turned on.
A similar analysis can be performed for the NS three-form. Apart from the factor of the
dilaton one obtains the same result as computed in Refs. [11,12,17]. However, in this case
there is no obvious mirror dual set of fluxes in type IIA. This has led to the proposition [23]
that in type IIA the holomorphic three-form ceases to be holomorphic and effectively
an NS two- and four-form is induced. It would be interesting to make this proposal more
+
415
explicit at the level of the effective action. Similarly there is no obvious mirror dual in type
IIB of the NS-fluxes (pA , qB ) discussed in Section 2.4.
4. Conclusions
In this paper we studied the compactification of type II theories on CalabiYau
threefolds in the presence of background fluxes. First we concentrated on the massive
4 .
2 , F
type IIA theory and considered non-trivial background values for the RR fields F
These fluxes are in one to one correspondence with harmonic (1, 1)- and (2, 2)-forms
on the CalabiYau manifold and therefore induce 2h1,1 flux parameters into the effective
action. Furthermore, dualizing the three-form C3 in the d = 4 effective action to a constant
provides an additional parameter. Together with the (ten-dimensional) mass of massive
type IIA theory one finds a total of 2h1,1 + 2 flux parameters which naturally combine
into symplectic vector of Sp(2h1,1 + 2). Furthermore, we showed that in fact the entire
theory is invariant under Sp(2h1,1 + 2) rotations due to the presence of a massive twoform B2 which couples to both electric and magnetic field strengths. In our analysis we
just followed the compactification of type IIA supergravity but there is obviously a more
general story to be discovered here. It would be interesting to construct the most general
gauged supergravity including massive two-forms and investigate the conditions for the
symplectic Sp(2h1,1 + 2) invariance of the theory entirely within supergravity and without
any particular reference to flux backgrounds of string theory.
For vanishing magnetic fluxes mI = 0 the two-form B2 is massless and can be dualized
to a scalar a. This dual basis is a standard gauged N = 2 supergravity with a potential
(2.33) depending on the scalars in the vector multiplets. Only a is charged but with respect
to linear combination of all h1,1 + 1 gauge fields. The form of the potential coincides with
the potential found in [18] for the case of the heterotic string compactified on T 2 K3
when a set of very specific fluxes along K3 are turned on. In particular, in the heterotic
case more than one scalar is generically charged and thus the fluxes have to be chosen to
point in the direction of a in order to find agreement with the type IIA case [16].
The perturbative duality of type IIA compactified on Y3 to type IIB compactified on the
3 was established at the level of the effective Lagrangian for RR fluxes
mirror threefold Y
turned on in both theories. For non-vanishing mI also the type IIB effective theory features
a massive B2 . Furthermore, the validity of the duality crucially depends on including the
ten-dimensional mass parameter m0 and the dual constant e0 of three-form C3 .
The situation for NS-fluxes remains murky. We derived the effective theory of type
IIA in the presence of NS three-form flux H3 . In this case a standard gauged N = 2
supergravity is found but contrary to RR flux backgrounds the potential depends nontrivially on the scalars in the hypermultiplets. Furthermore, the hyper-scalars are charged
only with respect to the graviphoton but not with respect to any of the other h1,1 gauge
fields.
In the type IIB compactification we found no obvious set of NS mirror fluxes and leave
this puzzle for further studies. As suggested in [23] it might be related to the fact that the
CalabiYau geometry is deformed and the precise nature of the deformation has to be taken
into account in more detail.
416
We also did not address the issue of holomorphic superpotentials in relation with
spontaneous N = 2 N = 1 supersymmetry breaking [12] which we leave to a separate
publication [40].
Acknowledgements
This work is supported by DFGThe German Science Foundation, GIFthe German
Israeli Foundation for Scientific Research, the European RTN Program HPRN-CT-200000148 and the DAADthe German Academic Exchange Service.
We have greatly benefited from conversations with L. Andrianopoli, J.-P. Derendinger,
B. de Wit, R. Grimm, B. Gunara, S. Gurrieri, D. Lst, P. Mayr, T. Mohaupt, F. Quevedo,
H. Singh, A. Strominger, C. Vafa, S. Vandoren, A. Van Proeyen, and M. Zagermann.
Appendix A. Notations and conventions

In this appendix we assemble the conventions used throughout the paper.
The spacetime metric has signature (, +, +, ).
The components of a differential p-form are defined as follows
Ap =
1
A dx 1 dx p .
p! 1 p
(A.1)
A hat on a p-form, e.g., A p denotes differential forms in d = 10. p-forms without the
hat are four-dimensional quantities.
The Hodge operation is defined in such a way that
dAp dAp =
(A.2)
(dA)1 p+1 (dA)1 p+1 d d x
p!
reproduces the correct kinetic term for a p-form in d spacetime dimensions. In
particular we denote 1 = g d d x.
After compactification the Hodge operation splits into a Hodge-star on the fourdimensional space and another one acting on the internal CalabiYau space. For
example, in the expansion of a p form one encounters terms like A p = + Apk k +
, where k is some harmonic k form on the internal space. The Hodge dual is given
by
A p = + (1)k(pk)Apk k + ,
(A.3)
where the first on the RHS acts only in spacetime while the second acts only in the
internal space. The (1)k(pk) assures that the kinetic term of A p produces

(A.4)
Ap Ap = + Apk Apk k k + .
Y3
Y3
417
The indices i, j, k, . . . label harmonic (1, 1) and (2, 2) forms on the CalabiYau
threefold and run from 1 to h1,1 ; the indices I, J, . . . label the vector fields in type
IIA compactifications and include the zero I = 0, 1, . . . , h1,1 . The indices a, b . . . run
from 1 to h1,2 and label (1, 2)-forms on Y3 . The indices A, B . . . include the zero and
label all three-forms including the (3, 0)-form, i.e., A = 0, 1, . . . , h1,2 . A, B, . . . also
label vector fields in type IIB compactifications.
Appendix B. N = 2 supergravity in d = 4
The purpose of this appendix is to give a short review about N = 2 supergravity in
d = 4 [2022,41]. A generic spectrum contains the gravitational multiplet, nV vector
multiplets and nH hypermultiplets. The vector multiplets contain nV complex scalars t i ,
i = 1, . . . , nV while the hypermultiplets contain 4nH real scalars q u , u = 1, . . . , 4nH .14
Due to supersymmetry the scalar manifold factorizes
M = MV MH ,
(B.1)
where the component MV is a special Khler manifold spanned by the scalars while
MH is a quaternionic manifold spanned by the scalars q u .
A special Khler manifold is a Khler manifold whose geometry obeys an additional
constraint [20]. This constraint states that the Khler potential K is not an arbitrary real
function but determined in terms of a holomorphic prepotential F according to

I
I (X)
,
(t )FI (X) iXI (t)F
K = ln i X
(B.2)
ti
where XI , I = 0, . . . , nV are (nV + 1) holomorphic functions of the t i . FI abbreviates the

(X)
derivative, i.e., FI F
and F (X) is a homogeneous function of XI of degree 2, i.e.,
X I
I
X FI = 2F .
The 4nH scalars q u , u = 1, . . . , 4nH in the hypermultiplets are coordinates on a
quaternionic manifold [21]. This implies the existence of three almost complex structures
(J x )w
v , x = 1, 2, 3 which satisfy the quaternionic algebra
J x J y = xy + iB xyz J z .
Associated with the complex structures is a triplet of Khler forms
w
x
Kuv
= huw J x v ,
(B.3)
(B.4)
where huw is the quaternionic metric. The holonomy group of a quaternionic manifold
is Sp(2) Sp(2nh ) and K x is identified with the field strength of the Sp(2) SU(2)
connection x , i.e.,
1
K x = dx + B xyz y z .
2
(B.5)
14 There also exist N = 2 tensor multiplets which contain a two-form B and three real scalars as bosonic
2
components. Upon dualizing B2 to a scalar the tensor multiplet can be treated as an additional hypermultiplet.
With this in mind we do not discuss tensor multiplets in this appendix.
418
The bosonic part of the (ungauged) N = 2 action is given by

1
R 1 gi dt i d tj huv dq u dq v
S=
2
1
1
I J
I
J
+ Im NI J F F + Re NI J F F ,
2
2
(B.6)
where gi = i K, F I = dAI (F 0 denotes the field strength of the graviphoton) and the
gauge coupling functions are given by
I J + 2i Im FI K Im FJ L X X .
NI J = F
Im FLK XK XL
K
(B.7)
The equations of motion of the action (B.6) are invariant under generalized electric
magnetic duality transformations. From (B.6) one derives the equations of motions
L
= dGI = 0,
GI = Re NI J F J + Im NIJ F J ,
AI
while the Bianchi identities read
(B.8)
dF I = 0.
(B.9)
These equations are invariant under the generalized duality

F I U I J F J + Z I J GJ ,
rotations15
GI VI J GJ + WI J F J ,
(B.10)
where U , V , W and Z are constant, real, (nV + 1) (nV + 1) matrices which obey
U T V W T Z = V T U Z T W = 1,
U T W = W T U,
Z T V = V T Z.
Together they form the (2nV + 2) (2nV + 2) symplectic matrix

U Z
O=
,
O Sp(2nV + 2).
W V
(B.11)
(B.12)
(F I , GI ) form a (2nV + 2) symplectic vector; (XI , FI ) enjoy the same transformations

properties and also transforms as a symplectic vector under (B.10). Clearly, the Khler
potential (B.2) is invariant under this symplectic transformation. Finally, the matrix N
transforms according to
N (V N + W )(U + ZN )1 .
(B.13)
Let us now turn to gauged N = 2 supergravity [22]. One can gauge the isometries on
the scalar manifold M. Such isometries are generated by the Killing vectors kIu (q), kIi (t)
q u = I kIu (q),
t i = I kIi (t).
(B.14)
15 This is often stated in terms of the self-dual and anti-self-dual part of the field strength F J and the dual
+J , G N
I J F J .
quantities G+
I NI J F
I
419
kIu (q), kIi (t) satisfy the Killing equations which in N = 2 supergravity can be solved in
terms of four Killing prepotentials (PI , PIx ). The Killing vectors on MV are holomorphic
and obey
kIi (t) = g i j j PI ,
(B.15)
while the Killing vectors on MH are determined by a triplet of Killing prepotentials PIx (q)
via

x
kIu Kuv
(B.16)
= Dv PIx v PIx + B xyz vy PIz .
Gauging the isometries (B.14) requires the replacement of ordinary derivatives by covariant
derivatives in the action
q u D q u = q u kIu AI ,
t i D t i = t i kIi AI .
(B.17)
Furthermore the potential

J gj kI k j + 4huv kIu kJv + g i Di XI D X
J PIx PJx 3XI X
J PIx PJx
VE = e K X I X
j
J

J gj k k j + 4huv k u kJv 1 (Im N )1I J + 4eK XI X
J P x P x ,
= eK XI X
I J
I
I J
2
(B.18)
has to be added to the action in order to preserve supersymmetry. The bosonic part of the
action of gauged N = 2 supergravity is then given by

1
S=
R 1 gi Dt i D tj huv Dq u Dq v
2
1
1
+ Im NI J F I F J + Re NI J F I F J VE .
(B.19)
2
2
The symplectic invariance of the ungauged theory is generically broken since the action
now explicitly depends on the gauge potentials AI through the covariant derivatives
Dt i , Dq u .
Appendix C. The moduli space of CalabiYau threefolds

The moduli space of CalabiYau threefolds Y3 splits into the space of Khler
deformations of the CalabiYau metric and deformations of the complex structure. The
Khler deformations are harmonic (1, 1)-forms and thus are elements of H 1,1(Y3 ). The
complex structure deformations are harmonic (1, 2)-forms and thus are elements of
H 1,2(Y3 ). In string theory a two-form B2 always appears together with the (spacetime)
metric in the NSNS sector. Its compactification on a CalabiYau threefold produce
h1,1 additional scalars which together with the Khler class deformations form the
complexified Khler cone M1,1 . The moduli space M of CalabiYau manifolds is be a
direct product of M1,1 and the space M1,2 spanned by the complex structure deformations
M = M1,1 M1,2 .
(C.1)
420
M1,1 and M1,2 are both special Khler manifolds and M1,1 describes the vector multiplet
sector in type IIA theories while M1,2 characterizes the vector multiplet sector in type IIB
case. In this appendix we briefly summarize these geometries following Refs. [4144].
C.1. The complexified Khler cone
The Khler class deformations together with the zero modes of B2 are harmonic (1, 1)forms on Y3 . Hence both the Khler form J and B2 can be expanded in a basis i of
H 1,1(Y3 )

B2 + iJ = bj + iv j j t j j , j = 1, . . . , h1,1 .
(C.2)
It is useful to define the following quantities:

1
J J J,
Ki = i J J,
K=
6
Y3
Y3

Kij =
i j J,
Y3
Kij k =
i j k .
(C.3)
Y3
Note that we have introduced the factor 1/6 in the definition of K so that it is precisely
the volume of Y3 . The metric on the complexified Khler cone M1,1 is Khler, i.e.,
gij = i j K and given by [31,42]

1
1 Ki 1 Ki K
= i j ( ln 8K)
i j =
gij =
4K
4 K
4 K2
Y3
= i j ( ln 8K).
(C.4)
Furthermore, the Khler potential K is determined in terms of a holomorphic prepotential

F via

I
I , FI I F , I = 0, . . . , h1,1 ,
FI X I F
eK = 8K = i X
(C.5)
where
F =
1 Kij k Xi Xj Xk
.
3!
X0
(C.6)
The (complex) Khler class deformations t i are the so-called special coordinates related to
the XI via XI = (1, t i ). (XI , FI ) transforms as a symplectic vector under (B.12) and K is
a symplectic invariant.
The scalar manifold M1,1 describes the moduli space of the vector multiplets in the low
energy effective action of type IIA supergravity compactified on a CalabiYau threefold.
Therefore, the matrix N defined in (B.7) plays the role of generalized gauge couplings.
Inserting (C.6) into (B.7) it is straightforward to derive

1
1 Ki Kj i j
bb ,
Im N00 = K + Kij
Re N00 = Kij k bi bj bk ,
3
4 K

1 Ki Kj j
Im Ni0 = Kij
b ,
4 K

1 Ki Kj
Im Nij = Kij
.
4 K
1
Re Ni0 = Kij k bj bk ,
2
Re Nij = Kij k bk ,
421
(C.7)
In the main text we encounter the inverse matrices g ij and (Im N )1I J . These can be
expressed in terms of harmonic (2, 2) forms. On a CalabiYau threefold H 2,2(Y3 ) is dual
to H 1,1 (Y3 ) and it is useful to introduce the dual basis i normalized by

j
i j = i .
(C.8)
Y3
With this normalization the following relations hold

g ij = 4K i j ,
i = 4Kgij j ,
Y3
1 ij
i
g j ,
(C.9)
=
i j Kij k k ,
4K
where the symbol denotes the fact that the quantities are in the same cohomology class.
Introducing Kij via
Kij Kj k = ki ,
one derives
(Im N )1 =
(C.10)
1
K
bK
.
i j
i vj
Kij bKb v2K
Using (C.10) one can also write an explicit formula for the inverse metric g ij

vi vj
ij
ij
.
g = 4K K
2K
i
bK
(C.11)
(C.12)
C.2. The special geometry of H 3

The complex structure deformations of a threefold parameterize H 1,2(Y3 ). However,
it turns out be convenient to discuss the entire H 3 (Y3 ) = H 3,0 H 2,1 H 1,2 H 0,3
including H 3,0 and H 0,3 which have only the holomorphic (3, 0)-form and the complex
as elements. One commonly chooses a real basis (A , B ) on H 3
conjugate (0, 3)-form
which obeys

B
A B = A
= B A ,
A, B = 0, . . . , h1,2 ,
Y3
A B =
Y3
Y3
A B = 0.
Y3
(C.13)
422
Note that these relations are invariant under symplectic rotations

U Z
W V
(C.14)
where the matrices U , V , W , Z satisfy (B.11). The unique holomorphic (3, 0) can be
expanded in terms of this basis according to
= Z A A GA A ,
where
Z A , GA
ZA =
(C.15)
are the periods of defined as

A,
GA = A .
Y3
(C.16)
Y3
is invariant under (C.14) and hence (Z A , GA ) transforms as a symplectic vector. The

GA are functions of Z A and determined in terms of a homogeneous function of degree two
G(Z) as
GA =
G
A G.
Z A
(C.17)
Furthermore, is homogeneous of degree one in Z, i.e., = Z A A with

A = A GAB B .
(C.18)
The deformations of the complex structure za , a = 1, . . . , h1,2 which reside in H 1,2 (Y3 )
are related to the coordinates Z A via za = Z a /Z 0 or in other words one can choose
Z A = (1, za ). The metric ga b on the space of complex structure deformations M1,2 is
Khler ga b = a b K with the Khler potential K given by

A
A .
GA Z A G
= ln i Z
K = ln i
(C.19)
Y3
As we see K is determined in terms of the holomorphic prepotential G(Z) and hence M1,2
is a special Khler manifold. Note that K is symplectically invariant.
Finally, let us discuss the action of the Hodge on the basis (C.13). A and B are
both three-forms again so that they can be expanded in terms of and according to
A = AA B B + BAB B ,
= C AB B + D A B B .
(C.20)
Using (C.13) one derives

BAB = A B = B A = BBA ,
Y3
C AB =
Y3
A B =
Y3
AA B =
Y3
B A = C BA ,
Y3
B A =
Y3
A B = D B A .
(C.21)
423
Furthermore, the matrices A, B, C can be determined in terms of a matrix M [41,43]

A = (Re M)(Im M)1 ,
B = (Im M) (Re M)(Im M)1 (Re M),
C = (Im M)1 ,
(C.22)
where
AB + 2i
MAB = G
(Im G)AC Z C (Im G)BD Z D

.
Z C (Im G)CD Z D
(C.23)
MAB determines the gauge couplings in type IIB compactifications on Y3 .
Appendix D. Massless type IIA supergravity compactified on CalabiYau threefolds

without fluxes
In this section we recall the compactification of massless type IIA supergravity on a
CalabiYau threefold first performed in Ref. [31]. The purpose of this appendix is twofold.
One the one hand we need to redo the computation in order to fix the notation and
convention for the case studied in Section 2 where fluxes are turned on. On the other hand
the detailed dualization of the three-form C3 to our knowledge has not been presented
previously. This is of importance for our analysis in Section 2 as the three-form C3 turns
out to be dual to a constant e0 which nicely combines with other flux parameters to build
symplectic invariant combinations.16 Furthermore, as we will show e0 is the charge of the
scalar a which is dual to B2 and a potential consistent with the standard N = 2 gauged
supergravity is induced.
Let us start from the ten-dimensional action of massless type IIA supergravity. It
2 and the dilaton in the NSNS sector, a vector field
features the graviton, a two form B

A1 and a three-form C3 in the RR-sector and reads

1
2 1

S= e
R 1 + 2 d d H3 H3
2
4
1
F2 F2 + F4 F4 + Ltop ,
(D.1)
2
where
4 = d C
3 d A 1 B
2 ,
2 = d A 1 ,
3 = d B
2 ,
F
F
H

1
2 )3 d A 1 d A 1 .
3 d C
3 (B
2 )2 d C
3 d A 1 + 1 (B
Ltop = B
2 dC
2
3
(D.2)
16 In Ref. [31] the case e = 0 was considered.
0
424
We have chosen the conventions such that (D.1) is the m 0 limit of the action
(2.1).17 In these conventions (D.1) is invariant under the following three Abelian gauge
transformations

A 1 = d ,
2 = d
1 ,
B
3 = d
2 ,
C
3 = A 1 d
1 .
C
(D.3)
In order to compactify the action (D.1) on a CalabiYau threefold Y3 we expand the tendimensional fields in harmonic forms on Y3 . As already reviewed in Section 2 the Khler
class deformations of the metric are in one to one correspondence with the harmonic (1, 1)forms while the complex structure deformations are in one to one correspondence with the
2 , C
3 are expanded according
harmonic (1, 2)-forms on Y3 . Furthermore, the forms A 1 , B
to
A 1 = A0 ,
2 = B2 + bi i ,
B
3 = C3 + Ai i + A A + B B ,
C
(D.4)
where as reviewed in Appendix C, i , i = 1, . . . , h1,1 are harmonic (1, 1) forms on Y3 and

(A , A ), A = 0, . . . , h1,2 is a real basis on H 3 (Y3 ). A0 is the graviphoton and together
with the graviton g describe the bosonic components of the gravitational multiplet. The
other h1,1 vector fields Ai combine with the t j = bj + iv j into h1,1 vector-multiplets. The
h1,2 complex structure deformations za together with a , a form h1,2 hypermultiplets and
B2 , , 0 , 0 form a tensor multiplet. In d = 4 the two-form B2 can be dualized to a scalar
and, hence, the tensor multiplet can be turned into an additional (universal) hypermultiplet.
C3 carries no degrees of freedom in d = 4 but is dual to a constant.
Using (C.3), (C.22) and (D.4) the various terms in the Lagrangian integrated over the
CalabiYau space become

1
3 H
3 = K dB2 dB2 Kgij dbi dbj ,
H
4
4
Y3
2 F
2 = K dA0 dA0 ,
F
2
Y3
4 F
4
F
Y3

K
dC3 dA0 B2 dC3 dA0 B2
2

2Kgij dAi dA0 bi dAj dA0 bj
17 There is an ambiguity in the definition of C

3 in that C
3 C
3 + A 1 B
2 changes the form of the action and
the form of the gauge transformations. It is this second formulation that we use in Section 2.4 where we turn on
NS three-form flux.
425
AB

1
BD d D ,
Im M1
(D.5)
2
where gij and K were defined in (C.3), (C.4). Finally, for the topological terms we find

1
Ltop = B2 d A d A A d A + b i dAj dAk Kij k
2
+
Y3
1
bi bj dAk dA0 Kij k + bi bj bk dA0 dA0 Kij k .
3
(D.6)
Defining the four-dimensional dilaton via e2 = e2 K the action in d = 4 becomes18

1
2 1
R 1 + 2 d d H3 H3
S= e
2
4

i j
a
b
gij dt d t gab dz d z

1
KF2 F2 + 4Kgij dAi dA0 bi dAj dA0 bj

2
AB

1
BD d D
+ Im M1
2

1
+
H3 A d A A d A
2

1
1
b i dAj dAk bi bj dAk dA0 + b i bj bk dA0 dA0 Kij k

2
3

1
(D.7)
K dC3 dA0 B2 dC3 dA0 B2 .
2
The next step is the dualization of C3 following Appendix E.2. Using the results in this
appendix we find that the dual of the Lagrangian
LC3 =

K
dC3 dA0 B2 dC3 dA0 B2
2
(D.8)
is given by
1 2
(D.9)
e 1 e0 dA0 B2 ,
2K 0
with e0 being an arbitrary constant parameter. Replacing (D.8) by (D.9) in the action (D.7)
and collecting the terms involving H3 we obtain (after partial integration)

1
1
LH3 = e2 H3 H3 + H3 A d A A d A + e0 H3 A0 .
(D.10)
4
2
Le0 =
We see that e0 induces a four-dimensional GreenSchwarz term H3 A0 into action. The

standard form of the type IIA action is obtained by also dualizing B2 to an axionic scalar
18 Strictly speaking also the Khler moduli t i have to be redefined by a dilaton dependent factor [31]. In order
not to overload the notation we use the same t i also for the redefined moduli.
426
a. For details we refer the reader to Appendix E.1 while here we only record the final result
LH3 La =

e2
Da + A d A A d A Da + A d A A d A ,
4
(D.11)
where
Da = da + 2e0 A0 .
(D.12)
As anticipated the axionic scalar a is charged under a local PecceiQuinn gauge symmetry
with e0 being the gauge charge. Thus even ordinary CalabiYau compactifications of
type IIA give a one-parameter family of four-dimensional effective theories which are
generically gauged rather then ordinary supergravities. For e0 = 0 one recovers the
standard type IIA supergravity of Ref. [31].
The final task of this appendix is to rewrite the action in the form of standard gauged
supergravity [22]. To do this we use the dual action (D.11) instead of (D.10) and perform
the Weyl rescaling g e2 g in order to go to the Einstein frame. Furthermore
the scalars (, a, za , A , A ) which together form h1,2 + 1 hypermultiplets are denoted
collectively by q u . In these variables the action (D.7) reads

1
S=
2
1
1
+ Im NI J F I F J + Re NI J F I F J VE .
(D.13)
2
2
where
huv Dq u Dq v

e4
Da + A d A A d A Da + A d A A d A
4
AB

e2
BD d D ,
Im M1
2
(D.14)
and MAB was defined in (C.23). In Ref. [33] it was shown in that huv is a quaternionic
metric in accord with the constraints of N = 2 supergravity that the scalars in the
hypermultiplets span a quaternionic manifold.
Using (C.4) it is straightforward to show that the gauge couplings in (D.13) are given
by (C.7). Finally the potential VE in (D.13) is given by
VE =
e4 2
e 1.
2K 0
(D.15)
This potential coincides with (2.33) for ei = 0 and thus the consistency with gauged
supergravity can be shown analogously to Section 2.3.
427
Appendix E. Poincar dualities

For an arbitrary p-form in d dimensions one always has the choice to describe the action
in terms of a Poincar dual form. The nature of the dual form differs
in the massless and
physical degrees of
massive case.19 A massless p-form in d dimensions describes d2
d1 p
freedom while a massive p-form in d dimensions contains p degrees of freedom. The
difference can be easily understood from a generalized Higgs mechanism where a p-form
d2
.
eats a massless (p 1)-form and thus the number of degrees of freedom change by p1
Therefore a massless p-form in d dimensions is dual to a (d p 2)-form while a massive
p-form is dual to a (d p 1)-form. A massless (d 1)-form is special in that it is dual to
a constant. In d = 4 this implies that a massless three-form is dual to a constant, a massless
two-form is dual to a scalar while a massive 2-from is dual to a vector (a 1-form). Let us
discuss these cases in turn.
E.1. Dualization of a massless B2
Let us first consider the dualization of a massless two-form B2 with field strength
H3 = dB2 to a scalar a. We start from the generic action

g
1
H3 H3 H3 J1 ,
SH3 =
(E.1)
4
2
where g is an arbitrary function of the scalars while J1 is a generic 1-form depending
on the scalars and possibly some gauge field A1 . The dualization can be carried out by
introducing a scalar field a as a Lagrange multiplier and adding the term H3 da to SH3 .
Treating H3 as an independent three-form (not being dB2 ) the equation of motion for a
implies H3 = dB2 while the equation of motion for H3 reads H3 = g1 (da + J1 ). Inserted
back into the action (E.1) one obtains the dual action

1
(da + J1 ) (da + J1 ).
Sa =
(E.2)
4g
There is an another way of treating the dualizations which turns out to be useful in
understanding the dualization of a three-form in four dimensions. Consider the equation of
motion for B2
d(g H3 J1 ) = 0,
(E.3)
which can be derived from (E.1). It is solved by

g H3 J1 = da,
(E.4)
with a being some arbitrary scalar field. The equation of motion for this field is dictated by
the Bianchi identity of H3

1
0 = dH3 = d
(da + J1 ) ,
(E.5)
g
19 We thank F. Quevedo for educating us on this subject.
428
which in turn can be obtained from the action (E.2). This implies that the two ways
described for the dualization of B2 are equivalent.
E.2. Dualization of the three-form
Next we consider the dualization of a three-form in 4 dimensions. We start from a
generic action for a three-form C3 possibly coupled to two-forms, 1-forms and scalars

g
h
SC3 =
(E.6)
(dC3 J4 ) (dC3 J4 ) + dC3 ,
4
2
where g, h denote two arbitrary scalar functions and J4 is a 4-form which can depend on
the two-forms, 1-forms and scalars present in the spectrum.
For the field strength of a three-form in 4 dimensions there is no proper Bianchi identity
since no 5-forms exist. That is why the second way of dualizing forms presented in the
previous section, by exchanging the equation of motion with the Bianchi identity, cannot
work in this case. The only consistent way to proceed is to add a Lagrange multiplier to
the action (E.6) [45]

g
h
e0
(dC3 J4 ) (dC3 J4 ) + dC3 + dC3 ,
SC3 =
(E.7)
4
2
2
where e0 is a constant. The equation of motion for dC3 imply
h + e0
g
.
(dC3 J4 ) =
2
2
Inserted back into the action (E.6) and using dC3 = dC3 one obtains

1
1
Se0 =
(h + e0 )2 1 + (h + e0 )J4 .
4g
2
(E.8)
(E.9)
As we see a potential for the scalar fields is induced and e0 play the role of a cosmological
constant.
E.3. Dualization of a massive two-form
Finally, let us discuss the dualization of a massive two-form B2 [2830]. We start from
a generic action

gH3 H3 + M 2 B2 B2 + MT2 B2 B2 + B2 J2 ,
SB2 =
(E.10)
where g, M, MT can be field dependent couplings and J2 is a two-form which can depend
on the gauge potential A1 and/or some scalar fields. (J2 does not depend on B2 .) We can
treat B2 and H3 as independent fields and ensure H3 = dB2 by the equations of motion.
This is achieved in the action

gH3 H3 + 2gH3 dB2
SB2 =

+ M 2 B2 B2 + MT2 B2 B2 + B2 J2 ,
(E.11)
429
which indeed has H3 = dB2 as the equation of motion for H3 . So by inserting H3 = dB2
into (E.11) we obtain (E.10). On the other hand one can eliminate B2 through its equation
of motion and obtain an action expressed only in terms of H3 . The equation of motion for
B2 from (E.11) is
2M 2 B2 + 2MT2 B2 + J2 2 d (gH3 ) = 0,
(E.12)
which is solved by
B2 =
or
B2 =
MT2
1
M2
2
2
J
M
d
(gH
)
+
M
d
(gH
)
J
3
3
2
2
T
2
2
M 4 + MT4
MT2
1
M2
2
2
J
d
(gH
)
M
d
(gH
)
+
J
M
3
3
2
2 .
T
2
2
M 4 + MT4
(E.13)
Inserted back into the action (E.11) results in

M2
1
1

J
J
d
d
gH3 H3
(gH
)
(gH
)
SB2 =
3
2
3
2
2
2
M 4 + MT4

2
MT
1
1
+
(E.14)
d (gH3 ) J2 d (gH3 ) J2 .
4
4
2
2
M + MT
We can now replace H3 by its Poincar dual one-form AH = g H3 and the dual action for
the massive field AH is

1 H H
M2
1
1
H
H
A A +
SAH =
dA J2 dA J2
g
2
2
M 4 + MT4

2
MT
1
1
(E.15)
dAH J2 dAH J2 .
4
4
2
2
M + MT
As promised this is the action for a massive one-form AH .
References
[1] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[2] B. de Wit, D.J. Smit, N.D. Hari Dass, Residual supersymmetry of compactified d = 10 supergravity, Nucl.
Phys. B 283 (1987) 165.
[3] E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Duality of type II 7-branes and
8-branes, Nucl. Phys. B 470 (1996) 113, hep-th/9601150.
[4] P.M. Cowdall, H. Lu, C.N. Pope, K.S. Stelle, P.K. Townsend, Domain walls in massive supergravities, Nucl.
Phys. B 486 (1997) 49, hep-th/9608173.
[5] I.V. Lavrinenko, H. Lu, C.N. Pope, From topology to generalised-dimensional reduction, Nucl. Phys. B 492
(1997) 278, hep-th/9611134.
[6] J. Scherk, J.H. Schwarz, How to get masses from extra dimensions, Nucl. Phys. B 153 (1979) 61.
[7] R. Rohm, E. Witten, The antisymmetric tensor field in superstring theory, Ann. Phys. 170 (1986) 454.
[8] C. Bachas, A way to break supersymmetry, hep-th/9503030.
430
[9] J. Polchinski, A. Strominger, New vacua for type II string theory, Phys. Lett. B 388 (1996) 736, hepth/9510227.
[10] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Duality in superstring compactifications with magnetic
field backgrounds, Nucl. Phys. B 511 (1998) 611, hep-th/9708075.
[11] J. Michelson, Compactifications of type IIB strings to four dimensions with non-trivial classical potential,
Nucl. Phys. B 495 (1997) 127, hep-th/9610151.
[12] T. Taylor, C. Vafa, RR flux on CalabiYau and partial supersymmetry breaking, Phys. Lett. 474 (2000) 130,
hep-th/9912152.
[13] P. Mayr, On supersymmetry breaking in string theory and its realization in brane worlds, Nucl. Phys. B 593
(2001) 99, hep-th/0003198.
[14] G. Curio, A. Klemm, D. Lst, S. Theisen, On the vacuum structure of type II string compactifications on
CalabiYau spaces with H-fluxes, Nucl. Phys. B 609 (2001) 3, hep-th/0012213.
[15] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, hep-th/0105097.
[16] G. Curio, A. Klemm, B. Krs, D. Lst, Fluxes in heterotic and type II string compactifications, Nucl. Phys.
B 620 (2002) 237, hep-th/0106155.
[17] G. DallAgata, Type IIB supergravity compactified on a CalabiYau manifold with H-fluxes, JHEP 11 (2001)
005, hep-th/0107264.
[18] J. Louis, A. Micu, Heterotic string theory with background fluxes, hep-th/0110187.
[19] G. Curio, B. Krs, D. Lst, Fluxes and branes in type II vacua and M-theory geometry with G(2) and Spin(7)
holonomy, hep-th/0111165.
[20] B. de Wit, A. Van Proeyen, Potentials and symmetries of general gauged N = 2 supergravityYangMills
models, Nucl. Phys. B 245 (1984) 89.
[21] J. Bagger, E. Witten, Matter couplings in N = 2 supergravity, Nucl. Phys. B 222 (1983) 1;
B. de Wit, P.G. Lauwers, A. Van Proeyen, Lagrangians of N = 2 supergravitymatter systems, Nucl. Phys.
B 255 (1985) 569;
R. DAuria, S. Ferrara, P. Fre, Special and quaternionic isometries: general couplings in N = 2 supergravity
and the scalar potential, Nucl. Phys. B 359 (1991) 705.
[22] For a review and a more complete list of references see, for example, L. Andrianopoli, M. Bertolini,
A. Ceresole, R. DAuria, S. Ferrara, P. Fre, T. Magri, N = 2 supergravity and N = 2 super YangMills
theory on general scalar manifolds: symplectic covariance, gaugings and the momentum map, J. Geom.
Phys. 23 (1997) 111, hep-th/9605032.
[23] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[24] M. Haack, J. Louis, H. Singh, Massive type IIA theory on K3, JHEP 04 (2001), hep-th/0102110.
[25] K. Behrndt, E. Bergshoeff, D. Roest, P. Sundell, Massive dualities in six dimensions, hep-th/0112071.
[26] H. Singh, Romans type IIA theory and the heterotic strings, hep-th/0201206.
[27] L.J. Romans, Massive N = 2a supergravity in ten-dimensions, Phys. Lett. B 169 (1986) 374.
[28] Y. Takahashi, R. Palmer, Gauge-independent formulation of a massive field with spin one, Phys. Rev. D 1
(1970) 2974.
[29] F. Quevedo, C.A. Trugenberger, Phases of antisymmetric tensor field theories, Nucl. Phys. B 501 (1997)
143, hep-th/9604196.
[30] A. Smailagic, E. Spallucci, The dual phases of massless/massive KalbRamond fields, J. Phys. A 34 (2001)
L435, hep-th/0106173.
[31] M. Bodner, A.C. Cadavid, S. Ferrara, (2, 2) vacuum configurations for type IIA superstrings: N = 2
supergravity Lagrangians and algebraic geometry, Class. Quantum Grav. 8 (1991) 789.
[32] R. Bousso, J. Polchinski, Quantization of four-form fluxes and dynamical neutralization of the cosmological
constant, JHEP 0006 (2000) 006, hep-th/0004134;
J.L. Feng, J. March-Russell, S. Sethi, F. Wilczek, Saltatory relaxation of the cosmological constant, Nucl.
Phys. B 602 (2001) 307, hep-th/0005276.
[33] S. Ferrara, S. Sabharwal, Quaternionic manifolds for type II superstring vacua of CalabiYau spaces, Nucl.
Phys. B 332 (1990) 317.
[34] M. Bodner, A.C. Cadavid, Dimensional reduction of type IIB supergravity and exceptional quaternionic
manifolds, Class. Quantum Grav. 7 (1990) 829.
[35] R. Bhm, H. Gnther, C. Herrmann, J. Louis, Compactification of type IIB string theory on CalabiYau
threefolds, Nucl. Phys. B 569 (2000) 229246, hep-th/9908007.
431
[36] B. Janssen, Massive T-duality in six dimensions, Nucl. Phys. B 610 (2001) 280, hep-th/0105016.
[37] J. Polchinski, String theory, Cambridge Univ. Press, 1998.
[38] G. DallAgata, K. Lechner, D. Sorokin, Covariant actions for the bosonic sector of D = 10 IIB supergravity,
Class. Quantum Grav. 14 (1997) L195, hep-th/9707044;
G. DallAgata, K. Lechner, D. Sorokin, D = 10, N = IIB supergravity: Lorentz-invariant actions and
duality, JHEP 07 (1998) 017, hep-th/9806140;
G. DallAgata, K. Lechner, D. Sorokin, Action for IIB supergravity in 10 dimensions, hep-th/9812170.
[39] N. Marcus, J.H. Schwarz, Field theories that have no manifestly Lorentz invariant formulation, Phys. Lett.
B 115 (1982) 111;
J.H. Schwarz, Covariant field equations of chiral N = 2 D = 10 supergravity, Nucl. Phys. B 226 (1983) 269;
P.S. Howe, P.C. West, The complete N = 2, d = 10 supergravity, Nucl. Phys. B 238 (1984) 181.
[40] B. Gunara, J. Louis, in preparation.
[41] A. Ceresole, R. DAuria, S. Ferrara, The symplectic structure of N = 2 supergravity and its central extension,
Nucl. Phys. Proc. Suppl. 46 (1996) 67, hep-th/9509160;
L. Castellani, R. DAuria, S. Ferrara, Special Kahler geometry: an intrinsic formulation from N = 2
spacetime supersymmetry, Phys. Lett. B 241 (1990) 57;
L. Castellani, R. DAuria, S. Ferrara, Special geometry without special coordinates, Class. Quantum Grav. 7
(1990) 1767.
[42] P. Candelas, X. de la Ossa, Moduli space of CalabiYau manifolds, Nucl. Phys. B 355 (1991) 455.
[43] H. Suzuki, CalabiYau compactification of type IIB string and a mass formula of the extreme black holes,
Mod. Phys. Lett. A 11 (1996) 623, hep-th/9508001.
[44] A. Strominger, Special geometry, Commun. Math. Phys. 133 (1990) 163.
[45] P. Binetruy, F. Pillon, G. Girardi, R. Grimm, The 3-form multiplet in supergravity, Nucl. Phys. B 477 (1996)
175, hep-th/9603181;
For a review see, P. Binetruy, G. Girardi, R. Grimm, Supergravity couplings: a geometric formulation, Phys.
Rep. 343 (2001) 255, hep-th/0005225.
Nuclear Physics B 635 [PM] (2002) 435472

Rational Generalised Moonshine from abelian

orbifoldings of the Moonshine Module
Rossen Ivanov a,b , Michael Tuite a,c
a Department of Mathematical Physics, National University of Ireland, Galway, Ireland
b Institute for Nuclear Research and Nuclear Energy, 72 Tzarigradsko shosse, 1784 Sofia, Bulgaria
c Dublin Institute for Advanced Studies, 10 Burlington Road, Dublin 4, Ireland
Received 28 February 2002; accepted 22 April 2002
Abstract
We consider orbifoldings of the Moonshine Module with respect to the abelian group generated
by a pair of commuting Monster group elements with one of prime order p = 2, 3, 5, 7 and the
other of order pk for k = 1 or k prime. We show that constraints arising from meromorphic orbifold
conformal field theory allow us to demonstrate that each orbifold partition function with rational
coefficients is either constant or is a hauptmodul for an explicitly found modular fixing group of
genus zero. We thus confirm in the cases considered the Generalised Moonshine conjectures for all
rational modular functions for the Monster centralisers related to the Baby Monster, Fischer, HaradaNorton and Held sporadic simple groups. We also derive non-trivial constraints on the possible
Monster conjugacy classes to which the elements of the orbifolding abelian group may belong.
PACS: 11.25.Hf; 02.10.De; 02.20.Bb
Keywords: Conformal fields; Modular groups; Moonshine; Orbifolds
1. Introduction
Orbifold constructions [1] in Meromorphic Conformal Field Theory [2,3] (MCFT)
and Vertex Operator Algebras [46] provide the most natural setting for understanding
Moonshine phenomena [79]. The Moonshine Module [4,10] whose automorphism group
is the Monster finite sporadic group M, is an orbifold MCFT constructed by orbifolding
the Leech lattice MCFT with respect to the group generated by a reflection involution.
E-mail addresses: rossen.ivanov@nuigalway.ie (R. Ivanov), michael.tuite@nuigalway.ie (M. Tuite).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 8 - 8
436
R. Ivanov, M. Tuite / Nuclear Physics B 635 [PM] (2002) 435472
The Moonshine Module partition function is the classical elliptic J function which is a
hauptmodul for the genus zero modular group SL(2, Z) and is believed to be the unique
MCFT with this partition function [4]. Orbifolding the Moonshine Module with respect
to the group generated by g M leads naturally to the notion of the orbifold partition
function, i.e., the Thompson series Tg . Monstrous Moonshine is mainly concerned with the
property, conjectured by Conway and Norton [11] and subsequently proved by Borcherds
[12], that each Thompson series Tg is a hauptmodul for some genus zero fixing modular
group. Assuming the uniqueness of the Moonshine Module, this genus zero property is
believed to be equivalent to the following statement [8]: the only orbifold MCFT that
can arise by orbifolding with respect to g is either the Moonshine Module itself (for g
belonging to a so-called Fricke Monster conjugacy class) or the Leech lattice MCFT (for
g belonging to a non-Fricke Monster conjugacy class).
The Generalised Moonshine conjecture of Norton [13] is concerned with modular
functions associated with a commuting pair g, h M and asserts that each such modular
function is either constant or is a hauptmodul for some genus zero fixing group. No
extension of the Borcherds approach to Monstrous Moonshine has yet been shown to
be possible for Generalised Moonshine. We argue that the most natural setting for these
conjectures is to consider orbifoldings of the Moonshine Module with respect to the abelian
group g, h generated by g, h [14]. In the cases where g, h can be generated by a
single Monster element, for example when g and h have coprime orders, then Generalised
Moonshine follows directly from Monstrous Moonshine. In this paper we consider the
case where g is of prime order p = 2, 3, 5 and 7 and is of Fricke type and h is of order
pk for k = 1 or k prime. We confirm Nortons conjecture for modular functions with
rational coefficients in these cases by considering orbifold modular properties and some
consistency conditions arising from the orbifolding procedure. We also demonstrate a
number of other non-trivial aspects of Generalised Moonshine such as properties of the
character expansion of Generalised Moonshine functions in terms of irreducible characters
for the centraliser of g in M and constraints on the possible Monster conjugacy classes to
which the elements of g, h may belong.
We begin in Section 2 with a general review of abelian orbifold constructions in
Meromorphic Conformal Field Theory (MCFT). We also briefly review the construction of
the Moonshine Module and the relationship between the genus zero property of Thompson
series Tg in Monstrous Moonshine for g M and evidence to support the claim that the
only possible MCFTs obtainable by orbifolding the Moonshine Module with respect to
the group generated by g are the Moonshine Module, for g Fricke, and the Leech lattice
MCFT, for g non-Fricke.
In Section 3 we begin with a discussion of general properties for Generalised Moonshine
Functions (GMF) following from the orbifold considerations of Section 2. We then prove
two theorems concerning constraints that arise from the consistency of orbifolding the
Moonshine Module with respect to g, h under various choices of generators. These
constraints are exploited in Section 4 in order to determine the residues of singular cusps
of GMFs. In particular if all elements of g, h are non-Fricke the GMF is constant. We
then prove some modular properties of GMFs with rational coefficients in the cases where
g is Fricke of prime order p = 2, 3, 5 and 7 and h is of order pk for k = 1 or k prime.
This analysis in part relies on properties of the characters of the centralisers of the Monster
437
related to the Baby Monster, Fischer, Harada-Norton and Held sporadic simple groups. We
also highlight the importance of the Monster conjugacy classes to which the elements of
g, h belong in determining the possible singularities of a GMF.
In Section 4 we give a comprehensive analysis of the possible singularity structure of
GMFs for the cases under consideration. In each case we demonstrate that either the given
singularity structure is inconsistent or else all singularities of the GMF can be identified
under some genus zero fixing group for which the GMF is a hauptmodul. Thus we obtain
non-trivial constraints on the Monster classes to which the elements of g, h may belong
and verify the Generalised Moonshine conjecture in these cases.
In Appendix A we review the definitions of standard modular groups. In Appendix B we
consider the first ten coefficients of a general GMF as character expansions for the Monster
centralisers related to the Baby Monster and Fischer centraliser subgroups.
2. Abelian orbifolds and Monstrous Moonshine

2.1. Self-dual C = 24 meromorphic CFTs
A meromorphic CFT (MCFT) or chiral CFT is a CFT whose n-point functions are all
meromorphic as described in [2,3]. A MCFT essentially corresponds to a Vertex Operator
Algebra in the pure mathematics literature as reviewed in [46]. We briefly review some
of their basic properties. A MCFT (V, H) consists
of a Hilbert space of states H together
with a set of vertex operators V {V (, z) = nZ n znh | H} where h is the
conformal weight (see below) and where each mode n acts as a linear operator on H. The
vertex operators are local meaning that for given , H and sufficiently large N

V (, z)V (, w) V (, w)V (, z) (z w)N = 0.
(1)
H contains a distinguished vacuum state |0 such that
V (|0, z) = IdH ,
lim V (, z)|0 = .
z0
(2)
(3)
H also contains a Virasoro state with vertex operator Laurant expansion V (, z) =

n2 where L generates the Virasoro algebra of central charge C
n
nZ Ln z

1 3
C m m m,n .
(4)
12
We will consider here MCFTs of central charge C = 24 only. The vector space H is
decomposed into finite-dimensional spaces Hn with non-negative integral L0 grading n,
the conformal weight. Taking these various properties together, the vertex operators then
satisfy the Operator Product Expansion (OPE) for |z| > |w|

V (, z)V (, w) = V V (, z w), w
(5)

C V (, w)(z w)nh h
=
(6)
[Lm , Ln ] = (m n)Lm+n +
n0 Hn
438
with h n () = Hn C and where the sum is taken over some basis for Hn [2,3].
The automorphism group Aut(V) of V is the group of linear transformations g : H H
which preserves the Virasoro state and where
gV (, z)g 1 = V (g, z).
(7)
The OPE (5) is then invariant under Aut(V). In the case of the Moonshine Module V ,
Aut(V ) = M the Monster group, which has a unitary action on V . We will assume that
Aut(V) has a unitary action from now on (which is expected physically from unitarity).
A representation (U, K) of a MCFT (V, H) consists of a vector space (module)
K together with a set of local vertex operators U {U (, z)| H} where the modes
of U (, z) act as linear operators on K with U (|0, z) = IdK . These operators satisfy the
OPE

U (, z)U (, w) = U V (, z w), w , |z| > |w|.
(8)
(U, K) is an irreducible representation if K contains no non-trivial submodule invariant
under the modes of {U (, z)}. (V, H) is said to be a self-dual MCFT or a holomorphic
VOA if (V, H) is the unique irreducible representation for itself. We assume that K is
decomposed into Verma modules of the Virasoro algebra of V with non-negative integral
L0 grading, i.e., we consider unitary Virasoro representations. Let K0 K denote the
subspace of lowest L0 grading. Then for an irreducible representation, K is generated by
the action of the modes of {U (, z)} on K0 .
The automorphism group Aut(U) of U is the group of linear transformations g : K K
of the form
gU
(, z)g 1 = U (g, z)
(9)
for some g Aut(V). A general element of Aut(V) may give rise to a mapping between
different representations and so we denote by AutK (V) the subgroup of Aut(V) associated
with (9). Clearly AutK (V) acts projectively on K with a natural homomorphism from
Aut(U) to AutK (V) whose kernel consists of phase multipliers assuming that Aut(U) acts
unitarily on K, i.e., Aut(U) = U (1).AutK (V). For an irreducible representation, where K0
is one-dimensional, then the U (1) subgroup of Aut(U) has the same action on all elements
of K so that
Aut(U) = U (1) AutK (V),
(10)
if dim(K0 ) = 1.
Let us now assume that V is self-dual. The characteristic function (or genus one partition
function) for V of central charge C = 24 is given by

Z( ) = TrH q L01 , q = e2i ,
(11)
where H, the upper half complex plane, is the usual elliptic modular parameter. V has
integral grading so that Z( ) is invariant under T : + 1. Self-duality implies that
Z( ) is invariant under S : 1/ so that Z( ) is invariant under the modular group
SL(2, Z), generated by S, T [15]. Hence Z( ) is uniquely determined up to an additive
439
constant by J ( ), the hauptmodul for SL(2, Z) [16]

Z( ) = J ( ) + N0 ,
E43
1
744 = + 0 + 196884q + ,
J ( ) = 24
(12)
q

where ( ) = q 1/24 n>0 (1 q n ) is the Dedekind eta function, En ( ) is the Eisenstein
form of weight n [16] and N0 is the number of conformal weight 1 operators in V.
Examples of such theories are lattice models, which we denote by V , where is
one of the Niemeier even self-dual 24-dimensional lattices. Then Z( ) = /24 with

2
= q /2 the lattice theta function for . In this paper we will be particularly
concerned with the Leech lattice for which N0 = 24 and the Moonshine Module V for
which N0 = 0.
2.2. Twisted sectors of a self-dual MCFT
In this section we briefly review relevant aspects of twisted representation of a MCFT.
Let G = Aut(V), denote the automorphism group (7) for V. Consider g G of finite order
o(g) = n and define the trace function

g
Z
( ) TrH gq L0 1 .
(13)
1

Clearly Z 11 ( ) Z( ) of (11). In the case where V = V , the Moonshine Module,
(13) is the Thompson series (see below). Let H(j ) denote the eigenspace of H for g
j
with eigenvalue n for n exp(2i/n). The twisted representation (Vg , Hg ) of (V, H)
consists of a set of vertex operators Vg {Vg (, z)| H} with mode expansion

Vg (, z) =
(14)
m zm1 , H(j ) ,
mZ+j/n
whose modes m are linear operators on Hg . The twisting property corresponds to

the monodromy relation Vg (, e2i z) = Vg (g 1 , z). Furthermore the twisted vertex
operators satisfy the twisted non-meromorphic OPE

j b
Vg (, z)Vg (, w) = n Vg V (, z w), w ,
(15)
for H(j ) and where b = 0, 1, 2, . . . , n 1 labels the sheet for the branched n-fold
covering for (w/z)1/n . For (V, H) self-dual, (Vg , Hg ) always exists and is unique up to
isomorphism [9]. Furthermore, under conjugation by any element x G then x(Vg )x 1 is
isomorphic to Vxgx 1 .
Considering (15) for , H(0) we see that (Vg , Hg ) forms a reducible representation
(0)
for (V (0) , H(0) ) as in (8) where V (0) are the vertex
operators for H . Hg can therefore

be decomposed into L0 eigenspaces Hg = m=0 Hg,m where Hg,m has rational L0
g
eigenvalue E0 + 1 + m/n. The subspace {ga } for a = 1, 2, . . . , Dg of lowest grade
g
g
E0 + 1 is called the g-twisted vacuum space. E0 is called the vacuum energy and Dg
is called the vacuum degeneracy.
440
The automorphism group, Aut(Vg ), preserving (15) can be defined in a way similar
to (9) as an extension of Cg = {h G | gh = hg}, the centraliser of g in G. Since Vg is
unique for a self-dual MCFT we have Aut(Vg ) = U (1).Cg . If the twisted vacuum is unique
(Dg = 1) then Aut(Vg ) = U (1) Cg from (10).
Let g Aut(Vg ) denote the lifting of g with action on the twisted vacuum as follows

g
g
ga = exp 2iE0 ga ,
(16)
g
where exp(2iE0 ) is a U (1) phase. Then (14) implies that in general

g
g = exp(2ihg )g ,
(17)
where g Hg has Virasoro grading hg . Clearly n|o(g),

where o(g)
is the order of g,
g
We say that g is
since g n is a lifting of the identity element of Cg so that E0 Z/o(g).
g
a normal element of G if nE0 Z so that g is of order n, otherwise we say that g is an
anomalous element of G.
Let g be a normal element of G and consider the twisted spaces (Vg k , Hg k ) for
(j )
k = 1, 2, . . . , n 1. Let Hg k denote the eigenspace of Hg k with g eigenvalue n where

(j )
(j )
(j )
(j )
Hg k can be further decomposed into L0 eigenspaces Hg k ,m . Then {(Vg k , Hg k )} comprises

the n2 irreducible representations for the MCFT (V (0) , H(0)) [9]. Clearly Aut(V (0)) Gg
(j )
(j )
where Gg = Cg /g so that Aut(Vg ) = U (1).Gg . Note that Aut(Vg ) depends on j in
(j
)
general. In particular, if the twisted vacuum is unique then g1 Hg 0 for some j0 and
(j )
Aut(Vg 0 ) = U (1) Gg from (10) where n0 = exp(2iE0 ).
(j )
We next define the trace function for Hg = n1
m=0 Hg,m as follows
j =0

1
Z
( ) TrHg q L0 1
g
=
n1

j =0
(18)

g
TrH(j) q L0 1 = Dg q E0 + ,
(19)

(j )
TrH(j) q L0 1 = q j/n
Dg,m q m ,
(20)
m=0
(j )
(j )
(j )
where the coefficient Dg,m is the dimension of the representation g,m of Aut(Vg ) defined
(j )
(j )
by Hg,m . Then Dg is the dimension of g0 , the representation of Aut(Vg ) acting on the
twisted vacuum.
We can similarly define the general trace function for h Aut(Vg ) lifted from h Cg
by

L 1
h
0
Z
(21)
( ) TrHg hq
g
=
n1

j =0
q j/n

m=0
(j ) m
E0 + ,
g,m (h)q
= g0 (h)q
(22)
441
(j )
(j )
(j )
where g,m (h)
= Tr(g,m (h))
denotes a character of h Aut(Vg ) and g0 is the character
for g0 . We assume below that a particular choice for h can be made which resolves the

ambiguity inherent in the notation Z hg denoting the trace (21). When Dg = 1, then
(j )
0
g,m (h)/
g (h) is a character for Cg for j = j0 and is a character for Gg for j = j0 where
j
n0 = exp(2iE0 ).
For general commuting elements g, h, the trace function (21) transforms under a
modular transformation with respect to = ac db SL(2, Z) in the following way

h
h
Z
(23)
( ),
( ) = (g, h; )Z
g
g

ha g b
+b h
where = a
hc g d and where (g, h; ) is a phase multiplier [1,9]. We will
c +d , g
assume that if all elements in g, h, the group generated by g and h, are normal then we
may choose a lifting of ha g b such that the phase multiplier (g, h; ) = 1, i.e., there are
no global phase anomalies [17]. We will employ the abbreviation h for the chosen lifting
h Aut(Vg ) of h Cg from now on. In general, the absence of this phase multiplier is an
essential ingredient in the orbifolding procedure that we discuss below.
Suppose that the g twisted vacuum is one-dimensional, i.e., Dg = 1. Let g (h) denote
the lifting of h in its action on this twisted vacuum (and hence giving the extension of h on
g
all of Hg ) where in particular g (1) = 1 and g (g) = exp(2iE0 ). We conjecture that
a
b
g (h) has the following properties where g h is a normal class for all a, b:

g g a hb = g (g)a g (h)b ,
(24)
g (h) n .
(25)
hb
g 1 hb
Eq. (24) follows from modular invariance with Z g ( + 1) = Z g ( ) using (17)
a hb = (g)
b for normal
so that g (g a hb ) = g (g)a g (hb ) and the assumption that g
a (h)
a
b
classes g h . We will prove (25) assuming (24) for the specific examples of Generalised
Moonshine Functions that we consider later on.
2.3. Orbifolding a MCFT
Assume that all elements of g Zn , the abelian group of order n generated by
g, are normal elements of Aut(V). Then (V (0) , H(0)) has n2 irreducible representations
(j )
(j )
g
(Vg k , Hg k ) for j, k = 0, . . . , n 1. The g orbifold MCFT Vorb is the MCFT with
g
(0)
(0) with
Hilbert space Horb n1
k=0 Hg k . We assume that we can augment the operators V
appropriate local operators so that the OPE (5) is satisfied. The characteristic function of
gl

g
g
Vorb is Zorb = n1 n1
l,k=0 Z g k which is a modular invariant from (23) since (g, h; ) 1.
g
g
g
g
Hence Vorb is a self-dual MCFT and so Zorb ( ) = J ( ) + N0 as in (12) where N0 is

g
the number of conformal weight 1 operators in Vorb .
g,h
We can similarly consider the orbifold MCFT Vorb found by orbifolding with respect
to the abelian group g, h of order |g, h| generated by two commuting elements g, h
442

g,h
where all the elements of g, h are assumed to be normal. Vorb has Hilbert space
g,h
1
Horb = Pg,h ( vg,h Hv ) where Pg,h |g,h|
vg,h v denotes the projection
with respect to the group g, h. Again we assume that we may augment the MCFT Pg,h V
g,h
with appropriate local vertex operators. Vorb then has modular invariant characteristic

g,h
g,h
u
1
is therefore a self-dual MCFT and so
function Zorb = |g,h|
u,vg,h Z v . Vorb
g,h
g,h
Zorb ( ) = J ( ) + N0 .
We assume that these orbifold MCFTs further can be considered as
various embeddings
in a larger non-meromorphic CFT (V , H ) with Hilbert space H = vg,h Hv . In this
CFT, all twisted states are created by vertex operators satisfying some non-meromorphic
OPE of the generic form

C V (, w)(z w)h h h ,
V (, z)V (, w) =
(26)
g,h
for Hg , Hh and Hgh and similarly for all commuting pairs in g, h. Vorb
then consists of all V operators invariant under g, h. A rigorous discussion in the case of
the reflection automorphism orbifolding of the Leech lattice theory appears in [18].
Consider independent generators g, h of g, h, i.e., g A = hB for all A = 1, . . . , o(g)1
and B = 1, . . . , o(h) 1 where o(g) is the order of g, etc. Then |g, h| = o(g)o(h) and
1 o(g)1 k
Pg,h = Pg Ph , where Pg = o(g)
g denotes the projection operator with respect
k=0
g,h
to g. The orbifold Vorb can then be considered as a composition of orbifoldings for any
independent generators g, h where
o(g)1 o(h)1

g,h
Ph
Vg k hl
Vorb = Pg
k=0
l=0
o(g)1

h
h g
Vorb g k = Vorb orb ,
= Pg
(27)
k=0
using the uniqueness of the twisted sectors for a self-dual MCFT and the assumed
g,h
g
h
embedding of Vorb , Vorb and Vorb in V . Later on we consider the consistency of (27)
under the various possible independent choices for the generators of g, h in proving
Theorems 3.1 and 3.2 in Section 3.
2.4. The Moonshine Module and Monstrous Moonshine
The Moonshine Module V is historically the first example of a self-dual orbifold MCFT
[10] and is constructed as a Z2 orbifolding of V , which denotes the Leech lattice MCFT
from now on. The Z2 automorphism r Aut(V ) is a lifting of the lattice reflection
symmetry chosen so that Pr H contains no Virasoro level one states. The r-twisted
space Hr has vacuum energy E0r = 1/2 > 0 (r is a normal element of Aut(V )) and
r
hence contains no Virasoro level one states. The resulting orbifold MCFT, V (V )orb ,

therefore has characteristic function J ( ) (12) with Aut(V ) = M, the Monster group [4,
10]. We can identify a dual automorphism r M where Pr (H ) (respectively, Pr (Hr ))
443
is even (respectively, odd) under r [8]. This is an obvious automorphism of the nonmeromorphic OPE (26) for V for g, h r. Then orbifolding V with respect to r we
recover V . Furthermore, one obtains the Monster centraliser Cr = 21+24
.Co1 where Co1
+
denotes the Conway simple group and 21+24
is
an
extra-special
2-group
[4,10,19].
+
It is conjectured that V is characterised (up to isomorphism) as the unique selfdual C = 24 MCFT with characteristic function J ( ) [4]. We may consider other Zn
orbifoldings of V with characteristic function J ( ) which should reproduce V according
to this conjecture. In general, we can classify all automorphisms a Aut(V ) lifted from
automorphisms a Co0 the Leech lattice automorphism group for which Va can be
explicitly constructed satisfying the following constraints [8]:
(i) Pa H contains no Virasoro level one states, i.e., a is fixed point free.
(ii) Ha is non-tachyonic (i.e., E0a 0) and furthermore contains no Virasoro level one
states so that
E0a > 0.
(28)
(iii) a is a normal element of Aut(V ).

There are 38 classes of Co0 obeying these constraints including the 5 prime ordered
cases considered by Dong and Mason [20].
a
For each of these 38 classes, we expect that a self-dual MCFT Vorb with characteristic
function J ( ) exists. Furthermore, we can identify a dual automorphism a of order n so
a a
that V = ((V )orb )orb and where the a centraliser agrees with a corresponding Monster
a
centraliser in all known cases [8]. All of this provides evidence that (V )orb V in each
construction lending weight to the uniqueness conjecture.
Let us now define the Thompson series Tg ( ) for each g M

L 1
g
0
Tg ( ) TrH gq
(29)
=Z
( )
1
1
= + 0 + [1 + A (g)]q + ,
(30)
q
where A (g) is the character of the 196883-dimensional adjoint representation for M. The
Thompson series for the identity element is J ( ) of (12), which is the hauptmodul for the
genus zero modular group SL(2, Z) as already stated.
Conway and Norton [11] conjectured and Borcherds [12] proved that Tg ( ) is the
hauptmodul for some genus zero fixing modular group g . This remarkable property is
known as Monstrous Moonshine. In general, for g of order n, Tg ( ) is found to be 0 (n)
invariant up to hth roots of unity where h is an integer with h|n and h|24 (see Appendix A
for the definition of various standard modular groups). g is a normal element of M if
and only if h = 1, otherwise g is anomalous. Tg ( ) is fixed by some g 0 (N) which
is contained in the normalizer of 0 (N) in SL(2, R) where N = nh [11]. This normalizer
contains the Fricke involution WN : 1/N . All classes of M can therefore be divided
into Fricke and non-Fricke type according to whether or not Tg ( ) is invariant under the
Fricke involution. There are a total of 51 non-Fricke classes of which 38 are normal and
there are a total of 120 Fricke classes of which 82 are normal. We now briefly describe
444
how the genus zero properties of Monstrous Moonshine can be understood be using the
orbifold ideas reviewed in the last section [7,8].
For each of the 38 Leech lattice automorphisms a satisfying the conditions (i)(iii)
above we can compute the dual automorphism Thompson series Ta . This agrees precisely
with the genus zero series for the 38 non-Fricke normal classes of the Monster where [8]

Ta ( ) = TrH aq L01 a1
(31)
1
a1 ,
(32)
a ( )

with a ( ) = k|n (k )ak where a is a lifting of a Co0 with characteristic equation

{ak } are called the Framedet(x a)
= k|n (x k 1)ak and where n = o(a) = o(a).
shape parameters of a.
We can also identify the other 13 non-Fricke classes which are
anomalous and find the corresponding correct genus zero Thompson series [8]. This is
a
further evidence for the assertion that (V )orb V implied by the uniqueness conjecture
for V which we will now assume from now on.
Consider next f M, a Fricke element of order n. For normal elements we orbifold V
f
with respect to f to obtain a self-dual MCFT (V )orb . Assuming Tf ( ) is a hauptmodul
f

then (V )orb V for every normal Fricke element [8]. The converse is also true, where
f
given that (V )orb V for some f M then Tf is the hauptmodul for a genus zero
modular group containing the Fricke involution [8]. In general, assuming the uniqueness
conjecture, for all normal elements
=
a
a
f
V V Ta , Tf are hauptmoduls,
(33)
where the arrows represent an orbifolding with respect to the indicated group.
For any normal Fricke element f M of order n (33) is equivalent to the following
properties for the twisted sector Vf :
f
(i) The Vf vacuum is unique so that Df = 1 and has negative vacuum energy E0 =
1/n. Vf is then said to be tachyonic.
(ii) If f r is Fricke then f s is also Fricke where (o(f r ), o(f s )) = 1 and n = o(f r )o(f s ).
These conditions are then sufficient to supply all the poles and residues of Tf so that Tf
is a hauptmodul for a genus zero fixing group which includes the Fricke involution [7,8].
The genus zero property for an anomalous class of M, which corresponds to the Harmonic
formula of [11], is described in [8].
The orbifold method can also be employed [7] to explain the Conway Norton Power
Map Formula, which is a property of the Thompson series, independent of the hauptmodul
property:
Power map formula. Suppose Tg is invariant under 0 (n|h) + e1 , e2 , . . . . Then for any
d, Tg d is invariant under 0 (n |h ) + e1 , e2 , . . . , where n = n/(n, d), h = h/(h, d) and
e1 , e2 , . . . , are the divisors of n / h amongst the numbers e1 , e2 , . . . .
445
3. Generalised Moonshine
3.1. Properties of Generalised Moonshine Functions
We now consider Generalised Moonshine Functions (GMFs) which are generalised
Thompson series depending on two commuting Monster elements of the form

h
Z
(34)
( ) = TrH hq L0 1 ,
g
g

for h Cg where the action of the lifting of h on Hg is also denoted by h as discussed in

Section 2. Norton has conjectured that [13]:
Generalised Moonshine. The GMF (34) is either constant or is a hauptmodul for some
genus zero fixing group h,g .
Note. We denote the fixing group for Z
to h,g .
h
g
(o(g) ) by h,g which is obviously conjugate
The properties of (34) that follow from the previous section can be summarised as
follows:
(i) When all elements of g, h are normal elements of M then for SL(2, Z)

d b
h
h g
Z
(35)
( ) = Z c a ( ).
h g
g
Hence h,g (o(h), o(g)) defined in Appendix A. In practice (o(h), o(g)) h,g . In
particular, since and act equally we have

1
h
h
Z
(36)
= Z 1 ,
g
g
which property is known as charge conjugation invariance.
(ii) Given the uniqueness of the twisted sectors for V , under conjugation by any element

x M then x(Vg )x 1 is isomorphic to Vxgx 1 so that

h
xhx 1
,
Z
= (g, h, x)Z
xgx 1
g
g (h)
(g, h, x) =
xgx 1 (xhx 1 )
(37)
(38)
with g of (24) and (25).

(iii) For a normal Fricke element f of order n the twisted sector Hf is unique with
f
vacuum energy E0 = 1/n. Hence

f (f ) = n
(39)
446
from (16). Recall that the extension determined by the phase f (h) for each h Cf is
chosen in order to comply with (35). If h and hf are in the same Cf conjugacy class
where h = x(hf )x 1 for some x Cf , then using (35) for = T , (36) and (37) we find

that Z fh = (f, h, x)Z hf
f where (f, h, x) = n . In general, conjugating with respect
to elements of Cf we find that the GMF is a class function up to such an nth root of unity,
i.e., (f, h, x) n for all x Cf . If, on the other hand, h and hf are not in the same
Cf conjugacy class then f can be chosen so that

f xhx 1 = f (h)
(40)
for all x Cf .
From (22) for h Cf we have

h
(1)
Z
( ) = q 1/n
f,m (h)q m +
f
m=0
= f (h) q 1/n + 0 +
n1
q j/n
j =0,j =1
(j )
f,m (h)q m
m=1

af,s (h)q s/n .
(41)
s=1
(j )
The coefficient af,s (h) = f,m (h)/f (h) is called a head character for the given GMF
where s = mn j . Then af,s (h) is a character for Gf where Gf = Cf /f for s =
1 mod n and otherwise is a character for Cf . For example, if we choose f = 2+ (the
2A Monster element) then Cf = 2.B and Gf = B, the Baby Monster. Then af,s (h) is a
character for B for odd s and is a character for 2.B for even s [21]. This implies that the
decomposition of (41) into the irreducible characters of Cg involves only those characters
obeying such conditions. This observation is confirmed in Appendix B for f = 2+ and 3+
where explicit character expansions are given.
(iv) For a normal non-Fricke element a of order n we find the S transformation of Ta
of (32) results in

1
Z ( ) = a1 + O q 1/n ,
(42)
a

i.e., Ha has vacuum energy E0a = 0. Furthermore, from (16)

a0 (a ) = 1
with vacuum degeneracy Da = dim(a0 ) = a1 . Furthermore if (a )B is
B then from (32) we find that o(a )||n, i.e., o(a )|n and (n, o(a )) = 1.
(43)
Fricke for some
(v) The value of (34) at any parabolic cusp a/c with (a, c) = 1 is determined by the
vacuum energy of the g a hc twisted sector from (i). If g a hc is Fricke then (34) is
singular at a/c [13] with residue g a hc (hd g b ). (34) is holomorphic at all other points
on H. Once these singularities are known, then (34) can be analysed to check whether it is
constant or is a hauptmodul for an appropriate genus zero modulargroup.
If all elements

of g, h are non-Fricke then there are no singular cusps so that Z hg is holomorphic on
H/ (o(h), o(g)) and hence is constant. This accounts for the constant GMFs referred to
in the Generalised Moonshine Conjecture above. We therefore assume from now on that at
447
least one element of g, h is Fricke

which we chose to be g without loss of generality. The
singularities and residues of Z hg are then constrained by certain orbifolding constraints
which we discuss below in terms of two consistency theorems.
(vi) The operators {L0 , L1 , L1 } generate an sl(2, C) sub-algebra acting on Hg =
n1
(j )
(j )
(j )
(j )
(j )
m=0 Hg,m where L1 (Hg,m ) Hg,m+1 and L1 (Hg,m ) Hg,m1 so that
j =0
(j )
Hg,m = kerH(j) L1 imH(j)

g,m
g,m1
L1 .
(44)
(j )
(j )
Hence the corresponding representations for Aut(Vg ) are related where g,m = g,m1 +
(j )
(j )
g,m where g,m is the representation formed by kerH(j) L1 . Therefore every character
g,m
(j )
g,m (h) for h Aut(Vg ) obeys the property

(j )
(j )
(j )
g,m (h) = g,m1 (h) + g,m (h).
(45)
In Appendix B we observe this property for the first ten head characters ag,mnj (h) =
(j )
g,m (h)/g (h) for g = p+ and p = 2 and 3. For p = 5 and 7 this property can be observed
in [22,23].
3.2. Two consistency theorems
In this section we prove two theorems which simplify our analysis of GMFs. Their
purpose is to identify the phases hc g a (hd g b ) for hc g a Fricke appearing in (35). This
will be achieved by considering the consistency of orbifolding V with respect to g, h
under various choices of independent generators.
Theorem 3.1. Let g, h M be independent commuting elements where both g and h are
Fricke such that g (h) = 1 and where all elements of g, h are normal. Let u, v be any
independent generators for g, h. Then:
g,h
(i) Vorb V .
(ii) If u is non-Fricke then uA v B (u) = 1 for all Fricke elements uA v B except possibly
when o(v B )||o(v) with (B, o(v)) = 1.
(iii) If u is Fricke then there is a unique A mod o(u) such that uA v is Fricke with
uA v (u) = 1 and o(uA v) = o(v).
h
Proof. (i) Since h is Fricke and normal we have Vorb V from (33). By assumption
h

g (h) = 1 and hence (Vorb )g = Ph (Vg Vgh Vgh2 ) is tachyonic, i.e., has negative
h
vacuum energy. Therefore g acts as a normal Fricke element on Vorb V and orbifolding
g,h
the latter with respect to g we find that Vorb V from (33) again.
u,v
u v
(ii) Since u, v are independent Vorb = (Vorb )orb V from (i). With u non-Fricke
u
u

then Vorb V and hence (Vorb )v B = Pu (Vv B Vuv B Vu2 v B ) is non-tachyonic

except possibly when o(v B )||o(v) and (B, o(v)) = 1. But if uA v B is Fricke, then VuA v B is
448
tachyonic so that Pu = 0 on the corresponding vacuum sector, i.e., uA v B (u) = 1 except

possibly when o(v B )||o(v) and (B, o(v)) = 1.
u
u v
(iii) With u Fricke then Vorb V . But (Vorb )orb V implies that v acts a Fricke
u
u

element of order o(v) on Vorb since u, v are independent. Hence (Vorb )v = Pu (Vv Vuv

Vu2 v ) is tachyonic which is possible iff Pu = 1 on precisely one of the tachyonic
(0)
vacuum sectors VuA v for some unique A mod o(u) where o(v) = o(uA v).
Example. Consider the orbifolding of V with respect to f for f Fricke of non-prime

order n. Let r||n and consider u = f r of order s = n/r and v = f s of order r. Then u, v are
independent generators of f and Theorem 3.1 (iii) with g = f and h = 1 implies that if
u is Fricke then there is a unique A mod s such that uA v = f Ar+s is Fricke of order r. But
(r, s) = 1 implies that A = 0 mod s so that v is Fricke. This is the AtkinLehner closure
property for Thompson series [7], i.e., if f and f r are Fricke for r||n then f s is also Fricke
for s = n/r.
Theorem 3.2. Let g, h M be independent commuting elements where g is Fricke and h
is non-Fricke such that g (h) = 1 and where all elements of g, h are normal. Let u, v be
any independent generators for g, h. Then:
g,h
(i) Vorb V .
(ii) If u is Fricke then uA v B (u) = 1 for all uA v B Fricke except possibly when o(v B )||o(v)
with (B, o(v)) = 1.
g
g,h
g
g
Proof. (i) Since g is Fricke Vorb V from (33). Then either Vorb = Ph (Vorb (Vorb )h
g,h
g
) V or Vorb V when h acts as Fricke or non-Fricke element on Vorb from (33)
g,h
again. Assume that Vorb V and consider the alternative composition of orbifoldings
g,h
h
h
h
Vorb = Pg (Vorb (Vorb )g ) where since h is non-Fricke, Vorb V . But the
h
condition g (h) = 1 implies that the vacuum energy of (Vorb )g = Ph Vg is negative
g,h
which is impossible according to (28). Therefore Vorb V .
u,v
u v
(ii) For independent u, v, Vorb = (Vorb )orb V from (i) where u Fricke implies
u
u
u

Vorb V . Hence v is non-Fricke in its action on Vorb and so (Vorb )v B = Pu (Vv B Vuv B

Vu2 v B ) is non-tachyonic except possibly when o(v B )||o(v) and (B, o(v)) = 1. But

if uA v B is Fricke, then VuA v B is tachyonic so that Pu = 0 on the corresponding vacuum

sector, i.e., uA v B (u) = 1 except possibly when o(v B )||o(v) and (B, o(v)) = 1.
Note. There is no corresponding statement to (iii) in Theorem 3.1. This is because not all
self-orbifoldings of the Leech theory are of Fricke type, e.g., the involution of Co0 with
frame shape 216 /18 has a lifting a Aut(V ) with non-tachyonic vacuum energy Ea0 = 0
a
and vacuum degeneracy Da = 16 whereas (V )orb V .
449
3.3. Symmetries of rational GMFs for g = p+

We now consider GMFs (34) where g is assumed to be Fricke and of prime order p, i.e.,
g = p+ in ConwayNorton notation (see Appendix A) for p = 2, 3, 5, . . ., 31, 41, 47, 59, 71.
If g, h = u for some u M then (34) can always be transformed to a regular Thompson
series (29) via a modular transformation (35) as follows

A
h
u
Z
( ) = Z
(46)
( ),
g
1
where g = uC , h = uD for some C, D and where A = (C, D), c = C/A, d = D/A and
a, b are chosen so that ad bc = 1. In particular, this is always possible when p and
o(h) are coprime. An example of this phenomenon is discussed later on in Section 4.4. In
these cases, the genus zero property for the GMF therefore follows from that for a regular
Thompson series (29). The GMFs with non-trivial genus zero behaviour then occur for
h Cg where o(h) = pk for k 1. These only exist for p 13. Furthermore, from now
on we restrict our analysis to k = 1 or k prime alone.
We now make the further restriction of considering rational GMFs, i.e., those with q
expansion (41) with rational (and therefore integral) coefficients. No rational GMFs for
o(h) = pk occur for p = 11 and 13 from explicit calculations of the head characters. We
therefore only consider p = 2, 3, 5 and 7 from now on. This firstly implies that the GMF
for p > 2 has leading q expansion in normalised function form [26]

h
1
Z
( ) = 1/p + 0 + O q 1/p ,
(47)
g
q
i.e., g (h) = 1. We may also assume this normalised function form for p = 2 by relabelling
as discussed below. This choice is also sufficient to ensure that g, h are independent
generators of g, h since otherwise hk = g A for some A = 0 mod p which implies the
contradictory relation (g (h))k = (g (g))A = pA from (24). Then the centraliser Cg and
sporadic finite simple group Gg = Cg /g are as given in Table 1.
Each element l Gg is the image of p distinct elements, which we denote by
(g a , l) Cg for a = 1, . . . , p where (g a1 , l1 ). (g a2 , l2 ) = (g a2 +a2 +c(l1 ,l2 ) , l1 l2 ) with nonCg
Gg
trivial cocycle c(l1 , l2 ) for p = 2, 3. If (g a1 , l1 ) (g a2 , l2 ) then clearly l1 l2 and hence

the conjugacy classes of Cg can be labelled (non-uniquely) by the conjugacy classes of Gg
as in the ATLAS [24]. We define the (1, Gg ) classes of Cg to be those classes displayed
Cg
in the ATLAS. For p = 2, 3 where Cg is a non-trivial extension of Gg then h gh for

Table 1
g = p+ centralisers for p = 2, 3, 5, 7
g = p+
Cg
Gg
Name
2+
3+
5+
7+
2.B
3.Fi
5 HN
7 He
B
Fi
HN
He
Baby Monster
Fischer
Harada-Norton
Held
450
some elements h Cg and the corresponding (1, Gg ) class contains elements of the form
(g a , l) for all a. Any Gg irreducible character Gg becomes a Cg irreducible character with
(g a , l) = Gg (l) for a = 1, . . . , p whereas for the remaining Cg irreducible characters we
have Cg (g a , l) = pa Cg (1, l). The irreducible characters for the (1, Gg ) classes are given
unambiguously in the ATLAS for p > 2 [24]. For p = 2, we define the (1, Gg ) classes to
be those classes with characters as given in the ATLAS for 2.B.
We will now restrict our attention to those elements h where o(h) = pk for k = 1 and k
prime and h an element of a (1, Gg ) class of Cg . We will find all the corresponding GMFs
below using orbifold considerations and demonstrate that they must all have genus zero
fixing groups. Towards this aim we firstly prove a theorem concerning the phase multipliers
g (h).
Theorem 3.3. For g = p+, p = 2, 3, 5 and 7 and o(h) = pk, k = 1 or k prime and
assuming (24) then for h normal g (h) p .
Proof. If k = 1 then the result is obvious from (24) since g (h)p = g (hp ) = 1. It is
necessary to separately consider (a) k = p and (b) k = p, k > 1.
(a) Assume k = p. For p = 2 using charge conjugation (36) we have g (h) = g (h1 )
since g 1 = g so that g (h) 1 for all normal h. However, note that with h = 4|2+,
an anomalous Monster class, we have g = h2 = 2+ and h projects down to the 2C class
of B. However in this case g (h)2 = g (h2 ) = g (g) = 1 and hence g (h) = i so that
(36) does not hold. This anomalous case is discussed later on in Section 4.5.
For p > 2, H = hp is of order p and hence g (H ) p . From (24) we have
g (H ) = g (h)p . On the other hand g (xH x 1) = g (xhx 1 )p for all x Cg . But then
(g, H, x) = (g, h, x)p = 1 and so from (38) g (H ) = g (xH x 1). For p > 2 we also
find that H must be an element of a class of type (1, Gg ). For p = 5 and 7 this is obvious
since the extension of Gg to Cg is trivial. For p = 3, all classes of order 3 in Fi lift to
classes of order 3 in 3.Fi from the ATLAS [24] and hence the same result follows. From the
ATLAS it is also clear that every class of order p and of type (1, Gg ) is conjugate to at least
one its powers. Hence g (H ) = g (H a ) for some a = 1 mod p. But since g (H ) p
it follows that g (H ) = g (H )a = 1 for p > 2. Hence g (h) p as claimed.
(b) Lastly consider k = p, k > 1. Then H = hp is of order k coprime to p with
g (H ) k since g (H )k = g (H k ) = 1. But in this case H and g are coprime and

hence the GMF Z Hg can be directly found from the Thompson series for a normal class
of order pk, i.e., either (i) H = k+ with gH = pk+ or (ii) H = k with gH = pk + p
from the power map formula.
g,H
gH
(i) H = k+ and gH = pk+. Since gH is Fricke we have Vorb = Vorb V .
H

Consider (Vorb )g = PH (Vg VgH VgH k1 ) which must be a tachyonic Fricke
H
H
twisted sector of Vorb of order p since Vorb V . However, gH a is of order p for

a = 0 mod p only so that PH (Vg ) is tachyonic and hence g (H ) = 1.

gH
(ii) H = k and gH = pk + p. Since gH is non-Fricke we have Vorb

Z gH
1 ( ) is the known Thompson series

( )(p )
Tpk+p ( ) =
(k )(pk )
24
(k1)(p+1)
24
,
(k 1)(p + 1)
451
V . Hence
(48)
from (32). This is invariant under the AtkinLehner transformation Wp (A.5) of

p 1
Appendix A which we can choose as Wp = ckp dp where dp ck = 1 for coprime p, k.
Then

gH
gH
Z
( ) = Z
Wp ( )
1
1

H
1
=Z
(49)
(p ) = + 0 +
g
q
so that g (H ) = 1 and hence g (h) p . Note that this argument can also be directly
used in (i) if we consider the Wp invariance for Tpk+ .
Corollary 3.4. With g, h as above in Theorem 3.3 then
(a) If h and gh are conjugate in Cg we can choose g (h) = 1 for p = 2, 3.
(b) If h and gh are not conjugate in Cg then for p = 3, 5, 7 we have g (h) = 1 iff h is an
element of a (1, Gg ) class of Cg .
Proof. (a) As noted in Section 3.1 (iii) above if h and gh are conjugate in Cg (and hence

h
a
both belong to a (1, Gg ) class) then Z gh
g = p Z g and similarly for any g h for all a.
a
Hence from Theorem 3.3 g (g h) = 1 for some a so that after the relabeling g a h h we
choose g (h) = 1.
(b) If h and gh are not conjugate in Cg then h is conjugate to ha for some a = 1 mod pk
iff h is a class of type (1, Gg ) from the Atlas [24]. Furthermore from (40) for p > 2 it
follows that g (h) = g (ha ) for some a = 1 mod pk iff g (h) = 1 and hence the result
follows.
Remark. For p = 2 if h and gh are not conjugate in Cg then we may choose g (h) = 1 in
conjunction with the definition of the (1, Gg ) classes for 2.B.
Corollary 3.5. For p = 3, 5, 7 then Z
a class of type (1, Gg ).
h
g
rational implies g (h) = 1 and h is a member of
Proof. From Theorem 3.3 g (h) p which is rational for g (h) = 1 only for p > 2.
Hence from Corollary 3.4 we have h is a member of a class of type (1, Gg ).
We make the following useful observations concerning the irreducible characters for the
groups of Table 1 which can be checked by inspecting the appropriate ATLAS character
tables [24].
452
Lemma 3.6. The irreducible characters for the groups Cp+ of Table 1 enjoy the
following properties for (1, Gg ) classes:
(a) Any irrationality for is quadratic, i.e., for h Cp+ , (h) = a b for some
a, b Q where b = 0 for rational (h).
(b) The number, N , of independent irreducible characters of a given dimension is
N = 1 or 2 or in the case of the Held group possibly N = 3. If N = 2 then
the two characters,
and (say), are irrational and algebraically conjugate, i.e.,
(h)
a b for all h in the notation of (a). In the case of the Held group there
are three characters of dimension 1275 where two of the characters are irrational and
algebraically conjugate and the other is rational.
Note. A character is said to be irrational if it is irrational for at least one conjugacy class.
Otherwise, it is said to be rational. Clearly if N = 1 then the irreducible character is
rational since is also an irreducible character.
Example. The Fischer group has two irreducible algebraically conjugate characters 6 , 7
of dimension 1603525 and twoinequivalent conjugacy classes 23A, 23B [24]
such that
6 (23A) = 7 (23B) = (1 + i 23)/2 and 6 (23B) = 7 (23A) = (1 i 23)/2.
Theorem 3.7. Consider g = p+ for p = 3, 5, 7 and o(h)= pk, k = 1 or k prime. Suppose
that xCg x 1 = Cg for some x M and that the GMF Z hg ( ) is rational. Then

h
xhx 1
Z
.
=Z
g
g
(50)
Proof. Since the GMF is rational from Corollary 3.5 then g (h) = 1 and h is a member
of a (1, Gg ) class. Furthermore h is conjugate to ha for some a = 1 mod pk from an
inspection of the ATLAS character tables [24]. Hence xhx 1 is conjugate to (xhx 1 )a
and is also member of a (1, Gg ) class of Cg . Therefore g (xhx 1 ) = 1 also. For any
irreducible character of Cg and for all h Cg , x (h) (xhx 1 ) is another irreducible
character of the same dimension. Furthermore xCg x 1 = Cg implies that the number of
elements in the conjugacy classes for h and xhx 1 in Cg are equal. We now show that
either x = or , the algebraic conjugate. Suppose that x = . From Lemma 3.6
then either (i) they are both irrational and algebraically conjugate with x = or else (ii)
Gg = He and is of dimension 1275 where one is irrational , say, and the other x is
the unique rational character of dimension 1275. However the character table [24] for He
reveals that (ii) cannot be true since there are no two conjugacy classes of He with the same
number of elements for which a 1275-dimensional character is irrational and the other
x is rational. Hence x = if x = .
The head character ag,mpj (h) coefficient of q mj/p in (41) is rational by assumption
and is some integer linear combination of the irreducible characters for Cg . If a given
irreducible character (h) is irrational then and must appear with the same (possibly
zero) multiplicity and + = x + ()
x whereas if (h) is rational then (h) = x (h).
453
It is therefore clear that ag,mpj (h) = ag,mpj (xhx 1 ) for each head character and hence
the result follows.
Theorem
3.8. Consider g = p+ for p = 2, 3, 5, 7 and o(h) = pk, k = 1 or k prime. If the
GMF Z hg ( ) is rational then

d
h
h
Z
(51)
=Z a ,
g
g
where (d, pk) = 1 and (a, p) = 1.
Proof. Each head character ag,mpj (h) of (41) is rational by assumption. This implies

that ag,mpj (h) = ag,mpj (hd ) for each d such that (d, pk) = 1, e.g., [25]. Hence Z hg =
d
Z hg since g (hd ) = 1. Furthermore, for p = 3, 5, 7, g = p+ is conjugate to g a in M for
all (a, p) = 1 so that g = xg a x 1 for some x M. Clearly Cg = Cg a = xCg x 1 so that
xhx 1 Cg and h is a member of a (1, Gg a ) class of Cg a . Hence Theorem 3.7 implies

(50) and g (xhx 1 ) = 1. Using (37) we then find Z gha = Z hg since g a (h) = 1.
Corollary
3.9. Consider g = p+ for p = 2, 3, 5, 7 with o(h) = pk for k = 1 or k prime. If

Z hg ( ) is rational then 00 (pk, p) h,g , i.e., 0 (p2 k) h,g .
a b
00 (pk, p) we have b = 0 mod p and c = 0 mod pk as described

d
in Appendix A with ad bc = 1. Hence the RHS of (35) gives Z hg a ( ) with (d, pk) = 1
and (a, p) = 1. The result then follows from Theorem 3.8.
Proof. For =
cd
Note. We conjecture that a more general version of Theorem 3.8 holds, namely, if a
GMF is rational then (51) holds with (d, o(h)) = 1 and (a, o(g)) = 1 so that the GMF
is 00 (o(h), o(g)) invariant.
Theorem 3.8 further implies that the singularity structure is restricted to the Fricke
classes of g, h inequivalent under the equivalence relation g A hB g aA hdB where
(d, pk) = (a, p) = 1 for (A, B) = 1. We call the equivalence classes under the relation
g A hB g aA hdB for (d, pk) = 1 and (a, p) = 1 the class structure of the commuting
pair g, h. We will see below that in all cases considered the class structure together with
Theorems 3.1, 3.2 and 3.8 uniquely determine the GMF and its genus zero property.
4. The genus zero property for some rational GMFs for g = p+
We now come to the main purpose of this paper which is to demonstrate the genus
zero property for Generalised Moonshine Functions (GMFs) (34) for rational GMFs with
g = p+ and o(h) = pk for k = 1 or k prime. We will show how the orbifold considerations
discussed in the previous sections allow us to demonstrate that either the given Fricke
singularity structure is inconsistent or else all singularities of the GMF can be identified
454
under some genus zero fixing group for which the GMF is a hauptmodul. We conjecture
that the method described can be used to demonstrate the genus zero property for general
GMFs including irrational cases. We begin by discussing four examples of possible Fricke
class structures for g, h where in the first two examples we demonstrate the genus zero
property and in the second two examples, demonstrate that the given Fricke class structure
is impossible. These examples of class structures are by no means exhaustive but are
nevertheless of general applicability.
Theorem 4.1. For g = p+ for p = 2, 3, 5, 7 with o(h) = pk for k = 1 or k prime with
g (h) = 1 then the following class structures give rise to a rational GMF with genus zero

fixing group h,g for Z hg (p ) as follows:
(I) g Fricke and all other classes non-Fricke then h,g = p2 k.
(II) g, h Fricke and all other classes non-Fricke then h,g = p2 k + p2 k.
Note. See Appendix A for the modular group notation.

Proof. (I) Here Z hg (p ) has a unique pole at the cusp = i (q = 0) which is invariant
under 0 (p2 k), from Corollary 3.9, and is therefore the hauptmodul for 0 (p2 k) which
must be of genus zero. This restricts p2 k to the possible values 4, 8, 9, 12, 18, 25.
(II) Here we have singular behaviour in both the h and g twisted sectors, corresponding
to cusps at = 0 and = i. From Theorem 3.1 (i) using g (h) = 1 it follows that
g,h
Vorb V . Using Theorem 3.1 (iii) with u = g, v = h it follows that there exists a unique
A such that g A h is Fricke of order pk where g A h (g) = 1. But g A h must be therefore
conjugate to h, i.e., A = 0 mod p and h (g) = 1. Therefore
1

g
Z
(52)
( ) = q 1/pk + 0 + O q 1/pk .
h
k : 1/k which is conjugate to the Fricke involution Wp2 k for 0 (p2 k) (see
Define W
k normalizes 0 (pk, p) and interchanges the cusps at 0 and i.
Appendix A). W
0
1

h

Let f ( ) = Z (Wk ) Z h ( ) = Z g (k ) Z h ( ) = O(q 1/p ). f ( ) is
g
00 (pk, p) invariant without poles on H/00 (pk, p) and hence is constant and equal to
k , i.e., h,g = p2 k + p2 k. This restricts p2 k to the
zero. Hence h,g = 00 (pk, p), W
possible values 4, 8, 9, 12, 18, 20, 25, 27.
Theorem 4.2. For g = p+ for p = 2, 3, 5, 7 and o(h) = pk for k = 1 or k prime with
g (h) = 1 the following class structures are impossible for k > 1:
(I) g, gh and h Fricke where g is the only Fricke class of order p.
(II) g, ghk and h Fricke where h is the only Fricke class of order pk.
g,h
Proof. (I) From Theorem 3.1 (i) Vorb V and from (iii) with u = gh Fricke and
v = hk of order p it follows that uA v = (gh)A hk is Fricke for some A with o(uA v) = p.
455
Hence uA v = g A hA+k must be conjugate to g being the only Fricke class of order p,
i.e., A = k mod pk so that uA v = g k with g k (gh) = 1. However from (39) we also
have g k (g k ) = p which implies that (g k (h))k = g k ((gh)k )g k (g k ) = p . But
Theorem 3.8 implies that since g (h) = 1 then g k (h) = 1 leading to a contradiction.
Hence this conjugacy class structure is impossible.
g,h
(II) Again Vorb V . As in Theorem 4.1 (II) we can firstly show that h (g) = 1. Then
applying Theorem 3.1 (iii) with u = ghk Fricke and v = h it follows that uA v = (ghk )A h
is Fricke for some A where o(uA v) = pk. Hence uA v = g A hkA+1 must be conjugate to h
being the only Fricke class of order pk, i.e., A = 0 mod p and uA v = h with h (ghk ) = 1.
But from (39) we have h (h) = pk so that h (g) = (h (h))k = p1 . This contradicts
our earlier statement that h (g) = 1. Hence this class structure is also impossible.
4.1. Case A. Rational GMFs for o(h) = p, k = 1
The analysis when o(g) = o(h) = p with prime p has been previously reported in [14].
The class structure of g, h is now determined by g, h and gh. The cusps of H/00 (p) are
at {i, 0, 1, 2, . . ., p 1} corresponding to the sectors twisted by g, h and g s h gh
(s = 1, 2, . . . , p 1) respectively with 00 (p) 00 (p, p) [7,14]. There are then four
possible cases that may occur.
Theorem 4.3. For g = p+ and h of order p with g (h) = 1 each possible class structure

gives rise to a genus zero fixing group h,g for rational Z hg (p ) as follows:
(I.A) g Fricke. h,g = p2 .
(II.A) g, h Fricke. h,g = p2 +.
(III.A) g, gh Fricke. h,g = p.
(IV.A) g, h and gh Fricke. h,g = p|p for p = 2, 3 or 5||5+ for p = 5.
Note. Here every element of g, h is non-Fricke unless otherwise stated. Table 3 shows
all possible examples and the GMFs actually observed. See Appendix A for the modular
group notation.
Proof. (I.A) This follows from Theorem 4.1 (I). This restricts the possible values for p to
p = 2, 3 and 5. p = 5 does not arise in practicesee Appendix B [22,23].
(II.A) This follows from Theorem 4.1 (II). This restricts the possible values for p to
p = 2, 3, 5, 7. p = 7 does not arise in practicesee Appendix B [22,23].
(III.A) Here we have singular behaviour
1 0in the g and gh twisted sectors. Consider the
SL(2, Z) transformation p ST S = 1
which normalises (p) (p, p) and is of
1
order p in (p). We show that the cusps {i, 1, 2, . . ., p 1} are identified under p .
From (35) we have

h
Z
(p ) = gh (h)q 1/p + 0 + O q 1/p .
(53)
g
456

g,h
First we prove that gh (h) = 1. According to Theorem 3.2 (i) Vorb V . From
Theorem 3.2 (ii) with u = g Fricke and v = h it follows that gh (g) = 1 so that gh (g) =
ps for some s = 0 mod p. However with u = g a h Fricke for a = 0 mod p and v = h it
also follows that gh (g a h) = 1. (If a = 1 mod p then from (39) gh (gh) = p whereas if
a = 1 mod p then we can choose A such that aA = 1 mod p and B = (a 1)A mod p,
B = 0 mod p, so that (g a h)A (h)B = gh and then apply Theorem 3.2 (ii)). Let gh (h) = pr
then gh (g a h) = pas+r = 1. If r = 0 mod p then since s = 0 mod p we can always find a
such that as = r mod p which is a contradiction. Hence gh (h) = 1.

From (53) it follows that f ( ) Z hg (p ) Z gh ( ) = O(q 1/p ) is (p) invariant
and is non-singular at q = 0. One can easily check that the other cusps of f are similarly
non-singular so that f is holomorphic on H/ (p) and is therefore zero. Hence h,g =
00 (p), p = 00 (1, p) invariant, i.e., h,g = 0 (p). This is possible for p = 2, 3, 5, 7.
Note that for p = 7 this is the only rational class.
(IV.A) In this case all twisted sectors are Fricke and all cusps are singular. For p = 2, 3
and 5 there is exactly one rational class of order p in Gp+ not already identified in cases
(I.A) to (III.A) above. We consider each p in turn.
g,h
Consider p = 2. From Theorem 3.1 (i) Vorb V and from (iii) with u = h and
v = g there exists a unique A mod 2 such that hA g is Fricke and where hA g (h) = 1,
i.e., A = 0 and gh (h) = 1. From (39) gh (gh) = 1 and so gh (g) = 1. Similarly, from
Theorem 3.1 (iii) with u = g and v = h there exists a unique A mod 2 such that g A h
is Fricke and g A h (g) = 1, i.e., A = 1 and h (g) = 1. From (39) h (h) = 1 and so
h (gh) = 1. We summarise these phases
in Table 2.

Using (35) we then find that Z hg ( ) is also invariant under ST of order 3 which
normalises (2) and so is the hauptmodul for 00 (2), ST of level two and index two
in SL(2, Z), i.e., h,g = 2|2.

For p = 3, 5 we can show that h (g) = 1. From Theorem 3.8, Z hg = Z gh1 which
1

implies after an S transformation that Z gh = Z gh so that h (g) = (h (g))1 . But

h (g) p implies h (g) = 1 for p > 2. Hence Z gh ( ) = q 1/p + 0 + O(q 1/p ). For
p = 3 we can consider Theorem 3.1 (iii) with u = g 2 h and v = g 2 . Then g 2A+2 hA (g 2 h) =
1 for a unique A. However g 2 (g 2 h) = 32 and h2 (g 2 h) = 32 and hence A = 1 so that
gh (g 2 h) = 1 and hence gh (g) = gh (h) = 32 since gh (gh) = 3 from (39). Hence
the phase residues of all the singular cusps are now known. For p = 5 we observe that
h is an element of the 5A class of Gg = HN being the only remaining available rational
class not identified in cases (I.A) to (III.A). But h (g) = 1 implies
g is an element
that
of the 5A class of the isomorphic group Gh = HN. Hence Z gh = Z hg which leads to
Table 2
Phase
g
gh
h
gh
1
1
1
1
1
1
1
1
1
457
Table 3
Case
(I.A)
(II.A)
(III.A)
(IV.A)
h
p
p+
p
p+
gh
h,g
Examples
p
p
p+
p+
0 (p 2 )
0 (p 2 )+
0 (p)
p = 2, 3, 5*
p = 2, 3, 5, 7*
p = 2, 3, 5, 7
2|2, 3|3, 5||5+
p|p or p||p+
* Indicates that no such GMF occurs. Fricke classes are in
boldface.
gh (g) = gh (h) = 53 . A similar argument can also be used for p = 3. Thus we find
(p+1)/2
gh (g) = gh (h) = p
for p = 3, 5.
For p = 3, 2 = T 1 ST is of order 2 and interchanges the cusps {, 0} {2, 1}
whereas S interchanges {, 1} {0, 2}. We then find using the given phase residues that
h,g = 00 (3), 2 , S which is of level 3 and index 3 in SL(2, Z), i.e., h,g = 3|3.
For p = 5, let 3 = T ST 3 which is of order 3 in 00 (5) and cyclically permutes the cusps
{, 0} {1, 4} {2, 3} whereas S interchanges the cusps {, 1, 2} {0, 4, 3}. Hence
h,g = 00 (5), S, 3 which is of level 5 and index 5 in SL(2, Z), i.e., h,g = 5||5+.
We summarise cases (I.A)(IV.A) in Table 3 where all possible examples are indicated.
Examples that do not arise in explicit calculations are marked with an asterix and all Fricke
classes are in boldface.
4.2. Case B. Rational GMFs for o(h) = p2 , k = p
The class structure of g, h is now described by g, h, gh and ghp where from
Theorem A.1 of Appendix A the cusps of H/00 (p2 , p) are at {i, 0, s, s/p} for s =
1, 2, . . . , p 1 corresponding to the sectors twisted by g, h, g s h gh and g s hp ghp for
s = 1, 2, . . . , p 1, respectively. Furthermore no such h Cg exists for p = 7 whereas for
p = 5 only two algebraically conjugate classes of order 25 exist with irrational characters
and irrational GMFs. Hence we consider p = 2, 3 only. There are then eight possible cases
that may occur as follows.
Theorem 4.4. For p = 2, 3 with g = p+ and h of order p2 with g (h) = 1 some possible

class structures gives rise to a genus zero fixing group h,g for rational Z gh (p ) while
others are impossible as follows:
(I.B)
(II.B)
(III.B)
(IV.B)
(V.B)
(VI.B)
(VII.B)
g Fricke. h,g = p3 .
g, h Fricke. h,g = p3 +.
g, ghp Fricke. h,g = p2 or p2 |p.
g, gh and ghp Fricke. Impossible.
g, h and gh Fricke. Impossible.
g, h and ghp Fricke. Impossible.
g and gh Fricke. h,g = 8 12 + for p = 2. Impossible for p = 3.
458
(VIII.B) g, h, gh and ghp Fricke. h,g = 4|2+ or 4|2 + 2 for p = 2, h,g = 9|3+ or
9||3+ for p = 3.
Note. Every element of g, h is non-Fricke unless otherwise stated. Table 5 shows all
possible examples and the GMFs actually observed. See Appendix A for the modular group
notation.
Proof. (I.B) This follows from Theorem 4.1 (I).
(II.B) This follows from Theorem 4.1 (II).
(III.B) We have singular behaviour in the g and ghp twisted sectors. Let ghp (h) = pb
1 b
(b)
for some b using (39). Consider p ST p ST b = p
1pb which normalises and is of
order p in (p2 , p). We have from (35) that

h (b)
p = ghp h1pb g b q 1/p + 0 + O q 1/p .
Z
(54)
g
Then ghp (h1pb g b ) = (ghp (ghp ))b ghp (h) = 1 due to (39). From (54) it follows that

Z hg (p ) Z hg ( ) = O(q 1/p ) is holomorphic on H/ (p2 , p) and is therefore zero.
(b)
Thus h,g = 0 (p2 , p), p . If b = 0 mod p then h,g = 0 (p, p) and h,g = 0 (p2 )
0
invariant of genus zero. If b = 0 mod p then h,g has index p in 00 (p, 1) = 0 (p) and we
find h,g = p2 |p of genus zero.
(IV.B) From (39) gh (gh) = p2 so that gh (hp ) = p . From Theorem 3.2 (ii) with
u = g Fricke and v = h it follows that gh (g) = 1 so that gh (g) = ps for some
s = 0 mod p since g is order p. Taking u = g a hp Fricke for any a = 0 mod p and v = h
it also follows that gh (g a hp ) = 1 (similarly to Theorem 4.3 (III.A)). Choose a such that
as = 1 mod p so that gh (g a hp ) = pas+1 = 1 which is a contradiction. Hence this class
structure is impossible.
(V.B) From Theorem 4.2 (I) this class structure is impossible.
(VI.B) From Theorem 4.2 (II) this class structure is impossible.
(VII.B) From (39) gh (gh) = p2 . From Theorem 3.2 (ii) with u = g Fricke and
v = h it follows that gh (g) = 1. For p = 2 these conditions imply gh (g) = 1 and
1 1 2 0
gh (h) = 41 . Consider 2(2) = 1
which normalizes 00 (4, 2) and is of order
2
h (2)
gh02 1
2. We then find that Z g (2 ( )) = Z gh (2 ) = q 1/2 + O(q 1/2). Hence we find that
(2)
h,g = 0 (4, 2), , i.e., h,g = 8 1 + of genus zero.
0
For p = 3 we will show that the orbifolding procedure is not consistent. Since h3 =
3
( ) 12
0
0
3
3
3 then from (32) we have Z h1 ( ) = [ (3
) ] + 12, i.e., h3 (h ) = 1 and h3 (h ) = 12
gh
from (43). Since gh = 9+ then Vorb V and the constant term of TrH(gh)A (Pgh q L0 1 )

vanishes for all A. Taking A = 3 we have 19 8A=0 h03 ((gh)A ) = 13 [12 + h03 (gh) +
h
h03 ((gh)1 ))] = 0. Hence h03 (gh) + h03 ((gh)1 ) = 12. Since h = 9 then Vorb V

and the constant term of TrHh (q L0 1 ) is 24. Hence 19 8A=0 8B=1 h0B (hA ) = 24. But
orb
h0 (h) = 1 with h0 (h) = 3 from (43). From this it follows that h03 (h) + h03 (h2 ) = 3.
459
3
From Theorem 4.3 (I.A) we have Z hg (3 ) = [( )/(9 )]3 + 3 and after an S
g 2
transformation we have Z h3 (/3) = 3 + 27[( )/(/9)]3 = 3 + O(q 1/9), i.e., h03 (g) =
h03 (g 2 ) = 3. Using this information we find an inconsistency in the orbifolding with

respect to g, h where in particular, h03 (Pg,h ) = ug,h h03 (u) = 1 which must be a
non-negative quantity being the dimension of the h3 twisted vacuum space invariant under
g, h. Hence this class structure is impossible for p = 3.
g,h
(VIII.B) We firstly consider p = 2. From Theorem 3.1 (i) it follows that Vorb V
and from (ii) with u = gh Fricke and v = g it follows that there exists a unique A such that
uA v = g A+1 hA = 2+ with g A+1 hA (gh) = 1. This implies that g A+1 hA is conjugate to
either g or gh2 , i.e., A = 0 or 2 mod 4. If A = 0 mod 4 then g (gh) = 1 which contradicts
the known values g (h) = 1 and g (g) = 1. Hence A = 2 mod 4 and gh2 (gh) = 1,
gh2 (gh2 ) = 1 so that gh2 (g) = gh2 (h) = 1.
Similarly from Theorem 3.1 (iii) with u = gh2 Fricke and v = h1 it follows that
there exists a unique A such that uA v = g A h2A1 = 4+ with g A h2A1 (gh2 ) = 1. This
implies that g A h2A1 must be conjugate to either h or gh, i.e., A = 0 or 1 mod 2.
Suppose A = 0 mod 2. Then h (gh2 ) = 1 and so h (h) = i implies that h (g) = 1.
Furthermore, from Theorem 3.1 (iii) we have gh (gh2 ) = 1 which means gh (h) = i
since gh (gh) = i. Hence gh (h) = i (since gh = 4+) and hence gh (g) = 1. If on the
other hand A = 1 mod 2 then gh (gh2 ) = 1 and so gh (gh) = i implies that gh (h) = i
and gh (g) = 1. Furthermore, from Theorem 3.1 (iii) we have h (gh2 ) = 1 (since gh2
is of order 2) which means h (g) = 1 since h (h) = i. Hence all residues of singular cusps
are determined for p = 2 for both values of A. In particular note that h (g) = 1 for
A = 0 mod 2 and h (g) = 1 for A = 1 mod 2.
Now we can apply an analysis similar to (III.B) and using the above information to
(1)
(b)
obtain 00 (4, 2), 2 invariance with b = 1, p = 2 in p of III.B. We then find that
h
w 2 : 1/2 either fixes or negates Z g ( ) depending on the sign of h (g). Hence
(1)
(1)
h,g = 00 (4, 2), 2 , w 2 or 00 (4, 2), 2 , w 2 of genus zero, i.e., h,g = 4|2+ or

4|2 + 2 where 2 indicates that the GMF is negated by w2 .
Consider p = 3. As in (IV.A) we find that the rationality of the GMF implies that
g,h
h (g) = 1. Next Vorb V and from Theorem 3.1 (ii) with u = gh Fricke and v = g it
follows that there exists a unique A such that uA v = g A+1 hA is Fricke with g A+1 hA (gh) =
1. This implies that g A+1 hA is conjugate to either g or gh3 , i.e., A = 0 mod 3. We
let b = A/3. If b = 0 mod 3 then g (gh) = 1 which contradicts the known values
g (h) = 1 and g (g) = 3 . If b = 1 mod 3 then gh6 (gh) = 1 and gh6 (gh6 ) = 3 implies
that gh6 (g) = gh6 (h1 ) = 3 so that gh3 (g) = gh3 (h) = 3 . If b = 1 mod 3 then
gh3 (gh) = 1 and gh3 (gh3 ) = 3 implies that gh3 (g) = 3 and gh3 (h) = 31 . Thus
we find gh3 (g) = 3 and gh3 (h) = 3b for b = 1 mod 3.
We now show that gh (g) = 3b and gh (h) = 913b . Firstly we have gh3 (g b h) = 1
and hence as in (IV.A) we find that the rationality of the GMF implies that g b h (gh3 ) = 1.
Conjugating as in (40) we then find that gh (gh3b ) = 1. However gh (gh) = 9 implies
460
Table 4
Phase
g
h
gh
gh3
3
1
3b
3
1
9
913b
3b
Table 5
Case
gh
ghp
h,g
Examples
I.B
II.B
III.B
IV.B
V.B
VI.B
VII.B
VIII.B
p2
p2 +
p2
p2
p2 +
p2 +
p2
p2 +
p2
p2
p2
p2 +
p2 +
p2
p2 +
p2 +
p
p
p+
p+
p
p+
p
p+
p3
p3 +
p 2 , p 2 |p
Impossible
Impossible
Impossible
8 12 +
4|2+, 4|2 + 2 , 9|3+, 9||3+
p=2
p = 2, 3
p = 2, 3
p=2
p = 2, 3
Fricke classes are in boldface.
that gh (g) = 3b and gh (h) = 913b given that b = 1 mod 3. Hence all residues of
singular cusps are now determined as in Table 4.
Similarly to (III.B) and using the Table 4 we then obtain 00 (9, 3), 3(b), w 3 invariance
(b)
with b = 1 and p = 3 for p of (III.B). This corresponds to h,g = 9|3+ for b =
1 mod 3 and h,g = 9||3+ for b = 1 mod 3 both of genus zero.
We summarise cases (I.B)(VIII.B) in Table 5 where all possible examples are
indicated.
4.3. Case C. Rational GMFs for o(h) = pk, k prime, k = p
For o(h) = pk, k prime, k = p, the class structure of g, h is described by g, h, gh,
ghp , hk and ghk . From Theorem A.1 of Appendix A the cusps of H/00 (pk, p) are at
{i, 0, s, p1 , pk , ks } for s = 1, 2, . . . , p 1 corresponding to the sectors twisted by g, h,
g s h gh, ghp , hk and g s hk ghk for s = 1, 2, . . . , p 1, respectively. hk and ghk are
determined by the power map formula for h and gh. There are then thirteen possible cases
that may occur (see Table 6).
Theorem 4.5. With g = p+ and h of order pk for k prime k = p with g (h) = 1 some

class structures give rise to a genus zero fixing group h,g for rational Z hg (p ) while
others are impossible as follows:
(I.C) g Fricke. h,g = p2 k .
(II.C) g, h Fricke. h,g = p2 k + p2 k.
(III.C) hk Fricke. h,g = p2 k + p2 .
461
Table 6
Case
gh
ghp
hk
ghk
I.C
II.C
III.C
IV.C
V.C
VI.C
VII.C
VIII.C
IX.C
X.C
XI.C
XII.C
XIII.C
pk
pk + pk
pk + p
pk + p
pk+
pk
pk + pk
pk + p
pk
pk + pk
pk + p
pk + k
pk+
pk
pk
pk
pk + k
pk + k
pk + p
pk + p
pk + p
pk + pk
pk + pk
pk + pk
pk+
pk+
pk + p
pk + p
pk + p
pk+
pk+
pk + p
pk + p
pk + p
pk + p
pk + p
pk + p
pk+
pk+
p
p
p+
p
p+
p
p
p+
p
p
p+
p
p+
p
p
p
p
p
p+
p+
p+
p
p
p
p+
p+
Possible classes with Fricke classes shown in boldface.
(IV.C)
(V.C)
(VI.C)
(VII.C)
(VIII.C)
(IX.C)
(X.C)
(XI.C)
(XII.C)
(XIII.C)
ghp Fricke. h,g = p2 k + k for p = 2, 3, 5, 7 or 4k + k for p = 2.

g, h, ghp and hk Fricke. h,g = p2 k+.
g and ghk Fricke. h,g = pk.
g, h and ghk Fricke. Impossible.
g, hk and ghk Fricke. h,g = pk|p for p = 2, 3. Impossible for p = 5, 7.
g and gh Fricke. h,g = 4k 21 + 4k when p = 2. Impossible for p = 3, 5, 7.
g, h and gh Fricke. Impossible.
g, gh and hk Fricke. h,g = [9k a] when p = 3. Impossible for p = 2, 5, 7.
g, gh, ghp and ghk Fricke. h,g = pk + k.
All Fricke. h,g = 2k|2 + k, 2k|2 + k for p = 2, h,g = pk|p+ for p = 2, 3 and
h,g = pk||p+ for p = 3, 5.
Note. Table 12 shows all possible examples and the GMFs actually observed. Every
element of g, h is non-Fricke unless otherwise stated. See Appendix A for the modular
group notation.
Proof. (I.C) This follows from Theorem 4.1 (I).
(II.C) This follows from Theorem 4.1 (II).
(III.C) Since hk is Fricke and g (hk ) = (g (h))k = 1 then from Theorem 3.1 (i) it
g,hk
follows that Vorb V . Then from (iii) with u = g Fricke and v = hk it follows that
there exists a unique A such that g A hk is Fricke and g A hk (g) = 1. But g A hk must be
conjugate to g or hk , i.e., A = 0 mod p and hk (g) = 1. Furthermore, since (k, p) = 1 and
hk (hp ) p it follows that hk (hp ) = 1.

p2 = p b , p2 d bk = 1, conjugate to Wp2 ,
Consider the SL(2, Z) transformation W
k pd

p2 )
the AtkinLehner involution for 0 (p2 k) from Appendix A. Then from (35) Z hg (W
h
0
pd
b
1/p
1/p
1/p
g ) 1)q
+ 0 + O(q ) = O(q ) is 0 (pk, p) invariant
Z g ( ) = (hk (h
462
without poles on H/00 (pk, p) and hence is constant and equal to zero. Therefore h,g =
p2 is a genus zero group, i.e., h,g = p2 k + p2 .
00 (pk, p), W
(IV.C) We have singular behaviour in the ghp and g twisted sectors corresponding

k = k pb =
to the cusps = 1/p and = i. Consider the transformation W
pk kd
k b
1 pb k 0
2

p kd 0 1 , det(Wk ) = k; kd bp = 1. Wk is conjugate to Wk = p 2 k kd the

k ) =
AtkinLehner involution for 0 (p2 k) of Appendix A. Then from (35) Z hg (W
kd
bp
1/p
1/p
kd
1/p
1/p
+ O(q ) = ghp (h )q
+ O(q ).
ghp (h g )q
If p = 2 then o(hk ) = 2 and hence gh2 (hk ) = 1. Then analogously to Theorem 4.4

k ) = Z h ( ) and h,g = 4k + k or 4k + k both
(VIII.B) we conclude that Z hg (W
g
of genus zero where k indicates that the GMF is negated by Wk . For p > 2 we will
1+ak

now show that ghp (hk ) = 1. From Theorem 3.8 we have Z hg ( ) = Z h g ( ) for
(1 + ak, pk) = 1. But for all a we have (1 + ak, k) = 1 and either (1 + k, p) = 1 or
k we then find
(1 + 2k, p) = 1 and 1 + 2k < pk, i.e., we can choose a = 1 or 2. Applying W
2 ad
kd
kd(1+ak)
k
), i.e., (ghp (h )) = 1. However from Theorem 3.3
that ghp (h ) = ghp (h

k
k ) Z h ( ) =
p
gh (h ) p and (ad, p) = 1 and so ghp (hk ) = 1. Hence Z hg (W
g
O(q 1/p ) is without poles on H/00 (pk, p) and hence is constant and equal to zero. So
k or h,g = p2 k + k of genus zero for p = 3, 5, 7.
h,g = 00 (pk, p), W
g,h
(V.C) From Theorem 3.3 (i) Vorb V since h is Fricke. From (iii) with u = hk Fricke
p
v = gh u, v = g, h and from (iii) it follows that there exists a unique A such that
ghAk+p is Fricke with ghAk+p (hk ) = 1. But then ghAk+p must be conjugate to ghp .
Therefore A = p with ghp (hk ) = 1. As in (III.C) above and Theorem 4.3 (II.A) we can
also show that hk (hp ) = 1 and h (g) = 1. Using the standard argument we find that
h,g = p2 k+ of genus zero.
(VI.C) Consider the SL(2, Z) transformation p = ST k S which is of order p in and
normalises (pk, p). From (35) we have

h
Z
(p ) = ghk (h)q 1/p + 0 + O q 1/p .
(55)
g
g,hk
First we will prove that ghk (h) = 1. From Theorem 3.2 (i) Vorb V . With u = g
Fricke and v = hk from (ii) it follows that ghk (g) = ps = 1. With u = g a hk Fricke for a =
0 mod p and v = hk it also follows that ghk (g a hk ) = 1 (Here as in Theorem 4.3 (III.A) we
can choose A and B such that uA v B = ghk for any a = 0 mod p and apply Theorem 3.2
(ii)). Let ghk (h) = pr then ghk (g a h) = pas+kr = 1. If r = 0 mod p then since s =
0 mod p we can always find a such that as = kr mod p which is a contradiction. Hence
ghk (h) = 1. Now from (55) it follows that h,g = 00 (pk, p), p 00 (k, p) of genus
zero and h,g = 0 (pk) where all cusps = i, ks , (s = 1, 2, . . . , p 1) are identified
under p .
(VII.C) This class structure is impossible as shown in Theorem 4.2 (II).
(VIII.C) From Theorem 4.3 (IV.A) we must have p = 2, 3 or 5. Let us first consider
p = 2. All Fricke classes are of order 2 and since k = p = 2 and using Table 2 of
Theorem 4.3 (IV.A) we obtain Table 7.
463
Table 7
Phase
g
ghk
hk
1
1
1
1
1
1
Table 8
Phase
hk
ghk
32
32
32 when k = 1 mod 3
3 when k = 1 mod 3
hk
3 when k = 1 mod 3
32 when k = 1 mod 3
Table 9
Phase
hk
ghk
53
53
hk
h
1
52
51
51
52
when k = 1 mod 5
when k = 2 mod 5
when k = 1 mod 5
when k = 2 mod 5
Using Table 7 and (35) it is easy to check that h,g = 00 (2k, 2), 2(k) of genus zero
a b
(k)
(k)
(k)
, a = 0 mod 2, b, c, d = 0 mod 2; det 2 = 1. 2 is of order 3 in and
with 2 = ck
d
normalises 00 (2k, 2). Then h,g = 2k|2.
Let us now examine the p = 3 case. Since g, hk and ghk are all Fricke and of order 3
from Theorem 4.3 (IV.A) we have ghk (g) = ghk (hk ) = 32 and hk (g) = 1. Then using
u (hk ) = (u (h))k for u = ghk , hk we obtain the residues for the singular cusps shown in
Table 8.
(k)
Using Table 8 and (35) it is easy to check that Z hg (3i ) = q 1/3 + 0 + O(q 1/3) for
(k)
(k)
1
2
T 3,k
, 32
T 2 3,k
, when k = 1 mod 3 and with 3,k ST k S.
i = 1, 2 with 31
(k)
3i is of order 2 in and normalises 00 (3k, 3). Following the standard argument we then
(k)
(k)
find that h,g = 00 (3k, 3), 31 , 32 of genus zero where the cusps {i, 1k , 2k , 3k } are
identified. Then h,g = 3k|3 for k = 2, 5 are the only possible such modular groups for
k prime.
With p = 5 since g, hk and ghk are all Fricke and of order 5 then from Theorem 4.3
(IV.A) and in a similar way to that shown above we obtain the residues for the singular
cusps (see Table 9).

Using Table 9 and (35) we can show that Z hg (5i(k) ) = q 1/5 + 0 + O(q 1/5) for
(k)
(k)
= T 2 5,k when k = 1 mod 5, 51
= T 1 5,k when k = 2 mod 5
i = 1, 2 with 51
464
1
1
and 52 = T 2 5,k
when k = 1 mod 5, 52 = T 1 5,k
when k = 2 mod 5 where
(k)
(k)
5,k = ST k S. 5i is of order 3 in and normalises 00 (5k, 5). Following the standard

(k)
(k)
argument we then find that h,g = 00 (5k, 5), 51
, 52
of genus zero where the cusps
1 2 3 4 5
{i, k , k , k , k , k } are identified. However no such genus zero modular group exists for k
prime.
(IX.C) In this case h is a pk element which is possible only if h = 6 or 10. So
there are four possibilities: (1) p = 3, k = 2 (2) p = 5, k = 2 (3) p = 2, k = 3 and (4)
p = 2, k = 5. Following detailed arguments similar to Theorem 4.4 (VII.B) we find that
(1) and (2) lead to the contradictory property hp (Pg,h ) < 0.
Consider p = 2 with k = 3, 5. Since h is non-Fricke then from Theorem 3.2 (ii)
0
0
with u = g Fricke and v = h we have gh
(g) = 1 and hence gh
(g) = 1. Consider

(k)
1 1k k 0
0
2 = 1 k 0 1 which normalises and is of order 2 in 0 (2k, 2). Then

h
(k)
0
(gh)k g 1 q 1/2 + 0 + O q 1/2 .
Z
(2 ) = gh
g
(k)
0
0
0
However, since gh
(gh) = 2k from (39) and gh
(g) = 1 so we have gh
((gh)k g 1 ) =
(k)
1. Hence h,g = 00 (2k, 2), 2 of genus zero or h,g = 4k 21 + 4k [22]. Thus h,g =
12 12 + 12 in case (3) and h,g = 20 12 + 20 in case (4) both of genus zero.
(X.C) From Theorem 4.2 (I) this class structure is impossible.
(XI.C) Since g, hk = p+ and ghk = p we find from Theorem 4.3 (II.A) that (1) p = 2
(2) p = 3 or (3) p = 5. Let gh (g) = 3 and gh (h) = pk . From (39) gh (gh) = pk and
g,h
so k + = 1 mod pk. Since Vorb V using Theorem 3.2 with u = g Fricke and v = h
we have gh (g) = 1 so that = 0 mod p. Furthermore, with u = hk Fricke and v = gh1k
we find gh (hk ) = 1 and therefore = 0 mod p.
(1) p = 2. Since = 0 mod 2 we have = 1 and = 1 k = 0 mod 2 which is a
contradiction to = 0 mod 2 since k = p = 2. Therefore this class structure is impossible.
(2) p = 3. Here h is of type 3k + 3 so that k = 2 or 7 only. Since hk = 3+ we find from
Theorem 4.3 (II.A) that hk (g) = 1. From (39) hk (hk ) = 3 so that hk (h) = 32 for k = 2
and hk (h) = 3 for k = 7.
For k = 2 the constraints , = 0 mod 3 imply that gh (g) = 3 , gh (h) = 61
so that gh (g 1 h2 ) = 1. Consider 4 ( ) = T ST 2 (2 ). 4 which normalises and is
of order 4 in 00 (6, 3) and acts on the cusps {i, 1, 2, 3/2} corresponding to the
{g, gh, g 2 h, h2 } Fricke-twisted sectors in the following way: i 2 3/2 1 i.

1 2

gh3
Then Z hg (4 ) = Z g ghh (2 ) = 1.q 1/3 + 0 + O(q 1/3) and Z hg (42 ) = Z h2 ( ) =
1.q 1/3 + 0 + O(q 1/3) so that in the usual way we find that h,g = 00 (6, 3), 4 of genus
zero so that h,g = [18a] 0 (18), T 1/3 W18 T 1/3 whose hauptmodul has q expansion
denoted by 18z [26] or 18a [27].
For k = 7 the constraints , = 0 mod 3 imply that gh (g) = 32 and gh (h) =
7 13
14 13
8 so that (gh7 ) = 1. Consider = 1
21
and 2 = 1 7
both of
gh
1
7 14
7
7
which normalise and are of order 2 in 00 (21, 3). 1 , 2 act on the cusps {i, 1, 2, 3/7}
corresponding to the {h, gh, g 2 h, h7 } Fricke-twisted sectors in the following way: 1 :
465
Table 10
Phases for p = 2
Phase
g
h
gh
gh2
hk
ghk
g
1
1
1
1
1
1
hk
1
2k
2k
(2k )(1k)/2
1
1
1
1
1
1
1
1
{i, 2} {1, 3/7} and 2 : {i, 1} {2, 3/7}. Then in the usual way we find that
h,g = 00 (21, 3), 1, 2 of genus zero so that h,g = [63a] the modular group whose
hauptmodul has q expansion denoted by 63a [27].
(3) p = 5. Similarly to Theorem 4.4 (VII.B) we find the contradictory property
h05 (Pg,h ) < 0 so that this class structure is impossible.
g,h
(XII.C) From Theorem 3.2 (i) Vorb V . Then from (ii) with u = g Fricke and v = h
it follows that gh (g) = ps for s = 0 mod p. Taking u = g a hk Fricke for a = 0 mod p
and v = h it also follows that gh (g a hk ) = 1. Let gh (hk ) = pr then if r = 0 mod p we
can find a such that as = r mod p so that gh (g a hk ) = 1 which is impossible. Hence
gh (hk ) = 1. Similarly, from Theorem 3.2 (ii) with u = g Fricke and v = h it follows
that ghp (g) = 1 and taking u = g a hk Fricke for a = 0 mod p and v = h it follows that
ghp (g a hk ) = 1. As above we then find that ghp (hk ) = 1. Finally, as in (VI.C), we can
also prove that ghk (h) = 1.
of genus
Together these results can be used to show that h,g = 00 (pk, p), p , W
ak pb a pb k 0 k
k

k ) = k;
zero where p = ST S as defined in (VI.C) and Wk = ck kd = c kd 0 1 , det(W
akd bcp = 1. Therefore we find that h,g = pk + k.
(XIII.C) In this case all twisted sectors are Fricke and all cusps are singular. This case is
closely related to case (VIII.C.) Since g, hk , ghk = p+ from Theorem 4.3 (IV.A) we must
have p = 2, 3 or 5. Furthermore, for each such p there is exactly one rational class of order
pk in Gp+ not already identified in cases (I.C) to (XII.C). We consider each p in turn.
When p = 2 by repeated use of Theorem 3.1 (iii) and (39) we obtain Table 10 with
residues which augment Table 7.
Using Table 10 and following the usual argument we can then show that h,g = 2k|2 + k
or 2k|2 + k where the GMF is either fixed (upper signs in Table 10) or negated (lower signs
in Table 10) by the involution w k : 1/k .
When p = 3, 5 the phase residues are presented in Table 11 which augment Tables 8
and 9.
These phase residues are determined by the use of Theorem 3.1 for p = 3, 5. For
example for p = 5 and k = 2 mod 5 we have ghk+1 (ghk ) = 5a for some a since o(ghk ) =
5. We will show that a = 0 mod 5. Let G = hk and H = g h5 where k + 5 = 1
so that g = H k and h = G H 5 . Then we have ghk+1 (ghk ) = H k (G H 5 )k+1 (H k G) =
G+1 H k+5 (GH k ). Furthermore, G (H ) = 1 using hk (g) = 1 from Theorem 4.3 (IV.A)
so that H GG . But H belongs to the same unique rational class in GG = G5+ as h
466
Table 11
Phases for p = 3, 5
Phase
hk
g
h
p
1
1
p
1
pk
gh
1k
p
1k
pk
ghp
hk
p
1
1
p
k + p = 1
k = 1 mod p
ghk
k = p+1
2 mod p
(p+1)/2
(p+1)/2
Parameters
= k mod p, p = 3
= 2k mod 5, k = 1 mod 5, p = 5
= 2k mod 5, k = 2 mod 5, p = 5

k
k
does in Gg and hence Z hg = Z H
G . This implies that ghk+1 (gh ) = g +1 hk+5 (gh ).
But k = 2 mod 5 implies that = 3 mod 5 and k + 5 = (1 + 3k) mod 5k. Hence
5a = g 1 h1+3k (ghk ) = ghk+1 (g 1 hbk ) where b(1 + 3k) = (k + 1) mod 5k using (51).
But b = 1 mod 5 so that 5a = ghk+1 (g 1 hk ) = 5a . Hence a = 0 mod 5 so that
1+k
ghk+1 (ghk ) = 1. By use of (51) and (39) we obtain gh (g) = 51 and gh (h) = 5k
.
We can similarly determine the other entries in Table 11.
For p = 3 we may follow case (VIII.C) and use Table 11 to find in the usual way that
(k)
(k)
h,g = 00 (3k, 3), 31
, 32
, w k of genus zero, i.e., h,g = 3k|3+ for k = 1 mod 3 which

exists for k = 7, 13 and h,g = 3k||3+ for k = 1 mod 3 which exists for k = 2, 5.
For p = 5 case we may follow case (VIII.C) and use Table 11 to find in the usual way
(k)
(k)
that h,g = 00 (5k, 5), 51
, 52
, w k of genus zero, i.e., h,g = 5k||5+ for k = 2 mod 5
which exists for k = 2, 3, 7. h,g does not exist for k = 1 mod 5. Table 12 summarises
the results for cases (I.C)(XIII.C).
4.4. Case D. Anomalous classes h = pk|p +
There are a number of GMFs for g = p+ for p = 2, 3 for which o(h) = pk but h is
a member of an anomalous M class, i.e., h = pk|p + . Then the hauptmodul property
for the GMF can be demonstrated as follows. The Thompson series Th ( ) (29) has the
following property [11]:
[Th (/p)]p = Thp ( ) + const
and similarly for GMF functions we have [8,14]
p
p
h
h
Z
( ) = Z
(p ) + const.
g
g
(56)
(57)
We then obtain the following examples of such GMFs:

2
(I.D) p = 2 with h = 4|2 and h2 = 2. Then Z hg (2 ) is a hauptmodul either for
0 (4) when gh2 = 2 or for 0 (2) when gh2 = 2+ from Table 3. Therefore from (57)
h,g = 8|2 or 4|2 where only the first example arises in practice.
467
Table 12
Cases
h,g
p=2
p=3
p=5
(I.C)
(II.C)
p 2 k
18
18 + 18
50 + 50*
(III.C)
(IV.C)
18 + 9
18 + 2
(V.C)
p2 k + p2
p2 k + k
4k + k (p = 2)
p 2 k+
18+
45+
50+
(VI.C)
(VII.C)
(VIII.C)
pk
Impossible
pk|p
12
12 + 12,
20 + 20
12 + 4, 20 + 4
12 + 3, 28 + 7,
12 + 3 , 20 + 5 ,
12+, 20+,
28+, 44+,
92+
6, 10
6|2, 14|2*
10
(IX.C )
4k 12 + 4k, p = 2
6|3
15|3*
(X.C)
(XI.C)
(XIII.C)
Impossible
[p 2 ka]
pk + k
(XIII.C)
2k|2 + k
2k|2 + k
pk|p+
pk||p+
p2 k + p2 k
12 12 + 12,
20 12 + 20
6 + 3, 10 + 5,
14 + 7, 22 + 11,
46 + 23
6|2 + 3
6|2 + 3
10|2 + 5
14|2 + 7
14|2 + 7
22|2 + 11
26|2 + 13
34|2 + 17
38|2 + 19
k = 2, 7
6+2
15 + 5
33 + 11
6||3+
15||3+
21|3+
39|3+
p=7
10 + 2
21 + 3
10||5+
15||5+
35||5+
* Indicates that no such GMF occurs.
2
(II.D) p = 2 with h = 4|2+ and h2 = 2+. Then Z hg (2 ) is a hauptmodul for 0 (4)+
when gh2 = 2 from Table 3. Therefore h,g = 8|2 + 4 or 8|2 + 4 . If gh2 = 2+ then
2
Z hg (2 ) is the hauptmodul for 2|2 but no such genus zero modular group commensurable
2
with SL(2, Z) exists fixing (Z hg (2 ) + const)1/2 .
3
(III.D) p = 3 with h = 3|3 and h3 = 1. Then Z hg (p ) is the hauptmodul for 0 (3)+
and therefore h,g = 9|3+.
(IV.D) p = 3 with h = 6|3 and h3 = 2. Then gh3 = 6 + 3 (since (gh3 )2 = 3+
3
and (gh3 )3 = 2) so that Z hg (p ) is the hauptmodul for 0 (6) + 3 and therefore
h,g = 18|3 + 3.
3
(V.D) p = 3 with h = 21|3+ and h3 = 7+. Then gh3 = 21+ and Z hg (p ) is the
hauptmodul for 0 (21)+ and therefore h,g = 63|3+.
3
(VI.D) p = 3 with h = 39|3+ and h3 = 13+. Then gh3 = 39+ and Z hg (p ) is the
hauptmodul for 0 (39)+ and therefore h,g = 117|3+.
468
4.5. Case E. Anomalous classes of type 4k|n + 2, . . . associated with the Baby Monster
Consider h = 4k|n + 2, . . . M for k = 1 or k prime. There are eight such classes in the
Monster. Then g = h2k = 2+ and hence h is of order 4k in Cg = 2.B but is of order 2k in
Gg = B. The corresponding GMF is then

h 2k

h
h
2k
ST S ,
Z
(58)
( ) = Z 2k ( ) = g, h; ST S Z
1
g
h
where (g, h; ST 2k S) is a phase that must be present since h is anomalous. This GMF is
directly related to a standard Thompson series and is therefore a hauptmodul. We have the
following examples: h = 4|2+, 8|4+, 12|2+, 12|2 + 2, 20|2+, 28|2+, 52|2+ and 68|2+.
5. Conclusions
We have shown how Generalised Moonshine can be understood within an abelian
orbifolding setting and have explicitly demonstrated the genus zero property for rational
Generalised Moonshine Functions (GMFs) arising in a number of non-trivial cases. We
have also discussed other aspects of Generalised Moonshine such as properties of the
head character expansion of GMFs and constraints on the Monster classes for products
of commuting Monster elements. The orbifold methods developed in this paper can in
principle be extended to analyse all GMFs towards proving the genus zero property in
general. Examples of GMFs with irrational coefficients using these methods appear in
[28].
Acknowledgements
We thank Simon Norton for providing us with invaluable information on replicable
series and for a number of useful discussions. We also thank Geoffrey Mason and Rex
Dark for useful discussions. We also acknowledge funding from Enterprise Ireland under
the Basic Research Grant Scheme.
Appendix A. Modular groups in Monstrous Moonshine

In this appendix we the describe various modular groups associated with Thompson
series and Generalised Moonshine Functions (GMFs). Let = SL(2, Z) denote the full
modular group. We define the following standard subgroups of :

0 (N) ac db , c = 0 mod N ,
(A.1)
00 (m, n)
(m, n)
a b
cd
a b
cd

, b = 0 mod n, c = 0 mod m ,
a b
cd
1 mod n
0 mod n,
0 mod m 1 mod m

.
(A.2)
(A.3)
469
We also use the notation 00 (m) 00 (m, m) and (m) (m, m). Note that 0 (mn)

and 00 (m, n) are conjugate where 0 (mn) = n 00 (m, n)n1 with n 10 n0 .
The normalizer N (0 (N)) = { SL(2, R) | 0 (N) 1 = 0 (N)} is also required to
describe Monstrous Moonshine [11]. Let h be an integer where h2 |N , (h2 divides N ) and
let N = nh. Then we define the following sets of matrices.
0 (n|h): the group of matrices of the form

a hb
(A.4)
,
det = 1,
cn d
where a, b, c, d Z. For h the largest divisor of 24 for which h2 |N , 0 (n|h) forms a
subgroup of N (0 (N)). For h = 1, 0 (n|h) = 0 (n).
We : the set of matrices for a given positive integer e

ae b
,
det = e, e||N,
(A.5)
cN de
where a, b, c, d Z. e||N denotes the property that e|N , and the (e, N/e) = 1. The set
We forms a single coset of 0 (N) in N (0 (N)) with W1 = 0 (N). It is straightforward to
show (up to scale factors) that

We2 = 1 mod 0 (N) ,

We1 We2 = We2 We1 = We3 mod 0 (N) ,
(A.6)
where e3 = e1 e2 /(e1 , e2 ). The coset We is referred to as an AtkinLehner (AL) involution
for
0 10(N). The simplest example is the Fricke involution WN with coset representative
which generates 1/N and interchanges the cusp points at = i and
N 0
b
= 0. For e = n we can choose the coset representative Ne de
where ed bN/e = 1
which interchanges the cusp points at = i and = e/N .
we : the set of matrices for a given positive integer e of the form

n
ae hb
(A.7)
,
det = e, e|| ,
cn de
h
where a, b, c, d Z. The set we is called an AtkinLehner (AL) involution for 0 (n|h).
The properties (A.6) are similarly obeyed by we with 0 (N) replaced by 0 (n|h).
N (0 (N)): the Normalizer of 0 (N) in SL(2, R) is constructed by adjoining to 0 (n|h)
all its AL involutions we1 , we2 , . . . where h is the largest divisor of 24 with h2 |N and
N = nh [11].
0 (n|h) + e1 , e2 , . . .: this denotes the group obtained by adjoining to 0 (n|h) a
particular subset of AL involutions we1 , we2 , . . . and forms a subgroup of N (0 (N)).
0 (n|h)+: this denotes the group obtained by adjoining to 0 (n|h) all its AL involutions
and forms a subgroup of N (0 (N)).
Sometimes we abbreviate 0 (n|h) + e1 , e2 , . . . by n|h + e1 , e2 , . . . and 0 (n|h)+ by
n|h+ (respectively, n + e1 , e2 , . . . and n+ for h = 1).
n||h+: this denotes the modular group containing 0 (nh) having index h in h (0 ( nh )
+)h1 . This conjugate contains the transformation + m/ h for all m which takes the
corresponding hauptmodul to h distinct ones labelled by the value of m, 0 m < h [26].
470
The following theorem [22] gives us information about the cusp points of 0 (N).
Theorem A.1. For N square-free, all 0 (N) inequivalent cusps can be represented by the
rational numbers a/b such that b > 0, b|N , (a, b) = 1 and 0 < a < N/b; two cusps a/b
and a1 /b1 being 0 (N) equivalent if and only if b = b1 and a = a1 mod (b, N/b).
Appendix B. Identification of generalised Moonshine Functions

In this appendix we consider the explicit form of the head character expansion of the
GMF with g (h) = 1

1
h
Z
o(g) = + 0 +
as (h)q s
q
g
(B.1)
s=1
for g = p+ for p = 2, 3, 5, 7 (see Table 1) and with h Cp+ of order o(h) = kp for k = 1
and k prime. Here as (h) ag,s (h) of (41). From (45) we therefore expect that
as (h) = asp (h) + s (h)
(B.2)
for some character s of Cp+ .

For p = 2, 3 we give the first 10 coefficients as (h) of (B.1) in terms of the irreducible
characters of Cp+ from the ATLAS [24]. The irreducible expansion for p = 5, 7 appear in
[22,23]. Then using [22,23,26] and [27] we may identify the genus zero fixing group h,g
in each case considered.
When g = 2+ then Cp+ = 2.B the double cover of the Baby Monster B then the first
10 head characters are:
a 1 = 1 + 2 ,
a2 = 185,
a3 = 21 + 2 + 3 + 4 ,
a4 = 2185 + 186 ,
a5 = 31 + 32 + 23 + 4 + 6 + 7 ,
a6 = 4185 + 2186 + 187,
a7 = 61 + 52 + 43 + 34 + 5 + 26 + 7 + 8 + 9 + 10 ,
a8 = 8185 + 4186 + 3187 + 188 ,
a9 = 81 + 102 + 73 + 44 + 25 + 56 + 47 + 28 + 29
+ 210 + 11 + 12 + 14 + 16 + 17 ,
a10 = 14185 + 9186 + 7187 + 3188 + 189 + 192 .
(B.3)
Clearly the property (B.2) is observed. Furthermore for s odd, as is a character for B
whereas for s even, as is a character for 2.B for which g is represented by 1 as discussed
in Section 3.1 (iii).
471
When g = 3+ then Cp+ = 3.Fi, the triple cover of the Fischer group Fi, then the first
10 head characters are:
a1 = 109,
a 2 = 1 + 2 ,
a3 = 109 + 110 ,
a4 = 109 + 110 + 112 ,
a5 = 31 + 22 + 3 + 8 ,
a6 = 2109 + 2110 + 112 + 113,
a7 = 3109 + 3110 + 111 + 112 + 113 + 114 ,
a8 = 41 + 52 + 23 + 5 + 28 + 9 + 10 + 13 ,
a9 = 5109 + 5110 + 111 + 2112 + 3113 + 114 + 116 + 119 ,
a10 = 6109 + 6110 + 111 + 4112 + 4113 + 2114 + 115 + 116
+ 118 + 119 + 122 .
(B.4)
Clearly the property (B.2) is again observed. Furthermore, for s = 2, 3, as is a character for
Fi otherwise as is a character for 3.Fi, the triple cover as discussed in section in Section 3.1
(iii).
References
[1] L. Dixon, J.A. Harvey, C.C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L. Dixon, J.A. Harvey, C.C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[2] P. Goddard, in: Proceedings of the CIRM Luminy Conference, World Scientific, Singapore, 1989.
[3] L. Dolan, P. Goddard, P. Montague, Commun. Math. Phys. 179 (1996) 61.
[4] I. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster, Academic Press, New
York, 1988.
[5] V. Kac, Vertex Operator Algebras for Beginners, University Lecture Series, Vol. 10, American Mathematical
Society, Boston, 1998.
[6] A. Matsuo, K. Nagatomo, Math. Soc. Japan Memoirs 4 (1999) 1.
[7] M.P. Tuite, Commun. Math. Phys. 146 (1992) 277.
[9] C. Dong, H. Li, G. Mason, Commun. Math. Phys. 214 (2000) 1.
[10] I. Frenkel, J. Lepowsky, A. Meurman, Proc. Natl. Acad. Sci. USA 81 (1984) 3256.
[11] J.H. Conway, S.P. Norton, Bull. London Math. Soc. 11 (1979) 308.
[12] R. Borcherds, Invent. Math. 109 (1992) 405.
[13] S.P. Norton, Proc. Symp. Pure Math. 47 (1987) 208.
[14] M.P. Tuite, Contemp. Math. 193 (1996) 353.
[15] Y. Zhu, J. Amer. Math. Soc. 9 (1996) 237.
[16] J.-P. Serre, A Course in Arithmetic, Springer-Verlag, Berlin, 1978.
[17] C. Vafa, Nucl. Phys. B 273 (1986) 592.
[18] Y.-Z. Huang, Contemp. Math. 193 (1996) 123.
[19] R. Griess, Invent. Math. 68 (1982) 1.
[20] C. Dong, G. Mason, U.C. Santa Cruz Preprint, 1992.
[21] C. Dong, H. Li, G. Mason, Contemp. Math. 193 (1996) 25.
472
[22] L. Queen, Some Relations Between Finite Groups, Lie Groups and Modular Functions, Ph.D. Dissertation,
University of Cambridge, Cambridge, 1980.
[23] L. Queen, Math. Comput. 37 (1981) 547.
[24] J.H. Conway, R.T. Curtis, S.P. Norton, R.A. Parker, R.A. Wilson, An Atlas of Finite Groups, Clarendon,
Oxford, 1985.
[25] W. Ledermann, Introduction to Group Characters, Cambridge Univ. Press, Cambridge, 1989.
[26] D. Ford, J. MacKay, S.P. Norton, Commun. Algebra 22 (1994) 5175.
[27] S.P. Norton, Private communication.
[28] R. Ivanov, M.P. Tuite, Nucl. Phys. B 635 (2002) 471, next article in this issue.

Some irrational Generalised Moonshine

from orbifolds
Rossen Ivanov a,b , Michael Tuite a,c
a Department of Mathematical Physics, National University of Ireland, Galway, Ireland
b Institute for Nuclear Research and Nuclear Energy, 72 Tzarigradsko shosse, 1784 Sofia, Bulgaria
c Dublin Institute for Advanced Studies, 10 Burlington Road, Dublin 4, Ireland
Received 28 February 2002; accepted 22 April 2002
Abstract
We verify the Generalised Moonshine conjectures for some irrational modular functions for the
Monster centralisers related to the HaradaNorton, Held, M12 and L3 (3) simple groups based on
certain orbifolding constraints. We find explicitly the fixing groups of the hauptmoduls arising in
each case. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf; 02.10.De; 02.20.Bb
Keywords: Conformal fields; Modular groups; Moonshine; Orbifolds
1. Introduction
The Moonshine Module [1,2] whose automorphism group is the Monster finite sporadic
group M, is an orbifold Meromorphic Conformal Field Theory (MCFT) constructed by
orbifolding the Leech lattice MCFT with respect to the group generated by a reflection
involution. The Moonshine Module partition function is the classical elliptic J function
which is a hauptmodul for the genus zero modular group SL(2, Z). The Moonshine
Module is believed to be the unique MCFT with this partition function [2]. Orbifolding
the Moonshine Module with respect to the group generated by an element g M leads
naturally to the notion of the orbifold partition function known as the Thompson series Tg .
Monstrous Moonshine is mainly concerned with the property, conjectured by Conway and
Norton [3] and subsequently proved by Borcherds [4], that each Thompson series Tg is
E-mail addresses: rossen.ivanov@nuigalway.ie (R. Ivanov), michael.tuite@nuigalway.ie (M. Tuite).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 9 - X
474
a hauptmodul for some genus zero fixing modular group. Assuming the uniqueness of
the Moonshine Module, this genus zero property is also believed to be equivalent to the
following statement [5]: the only orbifold MCFT that can arise by orbifolding with respect
to g is either the Moonshine Module itself (for g belonging to a so-called Fricke Monster
conjugacy class) or the Leech lattice MCFT (for g belonging to a non-Fricke Monster
conjugacy class).
The Generalised Moonshine conjecture of Norton [6] is concerned with modular
functions associated with a commuting pair g, h M and asserts that each such modular
function known as a Generalised Moonshine Function (GMF) is either constant or is
a hauptmodul for some genus zero fixing group. No extension of the Borcherds approach
to Monstrous Moonshine has yet been shown to be possible for Generalised Moonshine.
The most natural setting for these conjectures is to consider orbifoldings of the Moonshine
Module with respect to the Abelian group g, h generated by g, h [7,8]. In [8] we
considered the case where g is of prime order p = 2, 3, 5 and 7 and is of Fricke type
and h is of order pk for k = 1 or k prime. There we confirm Nortons conjecture for
modular functions with rational coefficients in these cases by considering orbifold modular
properties and some consistency conditions arising from the orbifolding procedure leading
to constraints on the possible Monster conjugacy classes to which the elements of g, h
may belong. In the present paper we extend the approach based on the restrictions coming
from the orbifolding of the Moonshine Module and demonstrate the hauptmodul property
for GMFs with irrational coefficients where g and h are of the same prime order.
We begin in Section 2 with a brief review of some properties of Meromorphic
Conformal Field Theory (MCFT), the Moonshine Module and Generalised Moonshine
Functions. In Section 3 we discuss the general properties for GMFs with irrational
coefficients when g is of prime order. We then state two theorems about the conjugation
properties of the centraliser elements, and one theorem concerning constraints that arise
from the consistency of orbifolding the Moonshine Module with respect to g, h under
specific choices of generators. We then analyse the GMFs with irrational coefficients
in the cases where g is Fricke of prime order p = 5, 7, 11 and 13 and h is of same
order p. This analysis in part relies on properties of the characters of the centralisers of
the Monster related to the HaradaNorton, Held, Mathieu and L3 (3) simple groups. We
give a comprehensive analysis of the possible singularity structure of GMFs for the cases
under consideration. In each case we demonstrate that all singularities of the GMF can be
identified under some genus zero fixing group for which the GMF is a hauptmodul. Thus
we verify the Generalised Moonshine conjecture in these cases.
2. Generalised Moonshine
2.1. Self-dual meromorphic conformal field theories
Let H denote the Hilbert space of a self-dual Meromorphic Conformal Field Theory
(MCFT) [9] of central charge 24. The characteristic function (or genus one partition
function) for H is given by Z( ) = TrH (q L0 1 ), q = e2i , where H, the upper
half complex plane, is the usual elliptic modular parameter. H has integral grading
475
(the conformal weight) with respect to L0 , the Virasoro level operator so that Z( ) is
invariant under T : + 1. Furthermore, self-duality of the MCFT implies invariance
under S : 1/ so that Z( ) is invariant under the modular group SL(2, Z) generated
by S, T . Hence Z( ) is given up to an additive constant by J ( ), the hauptmodul for
SL(2, Z) [10], i.e., Z( ) = J ( ) + N0 , where J has q expansion
E43 ( )
1
744 = + 0 + 196 884 q + 21 493 760 q 2 +
(1)
24 ( )
q

and where ( ) = q 1/24 n>0 (1 q n) is the Dedekind eta function, En ( ) is the Eisenstein
modular form of weight n [10], N0 is the number of conformal weight 1 operators in H.
For the Leech lattice MCFT N0 = 24 and for the FLM Moonshine Module V with Hilbert
space H [1,2] N0 = 0. This means that the latter does not contain conformal dimension 1
operators, i.e., there is an absence of the usual KacMoody symmetry.
J ( ) =
2.2. The Moonshine Module and Monstrous Moonshine

The Monster group M, the largest finite sporadic simple group, is the automorphism
group of the Moonshine Module [1,2]. The Thompson series Tg ( ) for each g M is
defined by

Tg ( ) TrH gq L0 1
(2)
with q expansion with coefficients determined by (reducible) characters of M. The

Thompson series for the identity element is J ( ) of (1), which is the hauptmodul for the
genus zero modular group SL(2, Z) as already stated.
Conway and Norton [3] conjectured and Borcherds [4] proved that Tg ( ) is the
hauptmodul for some genus zero fixing modular group g . This remarkable property is
known as Monstrous Moonshine. In general, for o(g) = n, Tg ( ) is found to be 0 (n)
invariant up to mth roots of unity where

0 (n)
a
c
b
d

, c = 0 mod n ,
and where m is an integer with m|n and m|24. g is said to be a normal element of
M if and only if m = 1, otherwise g is said to be anomalous. Tg ( ) is fixed by some
g 0 (N) which is contained in the normalizer of 0 (N) in SL(2, R) where N = nm [3].
This normalizer contains the Fricke involution WN : 1/N . All classes of M can
therefore be divided into Fricke and non-Fricke type according to whether or not Tg ( ) is
invariant under the Fricke involution. There are a total of 51 non-Fricke classes of which
38 are normal and there are a total of 120 Fricke classes of which 82 are normal. The genus
zero properties of Monstrous Moonshine can also be understood using constraints arising
from the possible orbifolding of V with respect to the group generated by g [5,11].
476
2.3. Properties of Generalised Moonshine functions

We now consider Generalised Moonshine Functions (GMFs) which are generalised
Thompson series depending on two commuting Monster elements of the form

h
Z
(3)
( ) TrH hq L0 1 ,
g
g

for h Cg , the centralizer of g in M. Hg denotes the Hilbert space for a so-called g-twisted
sector of V [8,12]. The original Thompson series (2) then corresponds to the untwisted
sector, i.e.,

g
Tg ( ) = Z
( ).
1

Each h Cg has a central extension, acting on Hg , which we also denote by h [8]. Norton
has conjectured that [6]:
Generalised Moonshine. The GMF (3) is either constant or is a hauptmodul for some
genus zero fixing group h,g .
If g, h = u for some u M then (3) can always be transformed to a regular
Thompson series (2) [8,11,13]. In particular, this is always possible when o(g) and o(h)
are coprime. In these cases, the genus zero property for the GMF therefore follows from
that for a regular Thompson series (2). The GMFs with non-trivial genus zero behaviour
then occur for h Cg where o(h) and o(g) are not coprime.
In Section 3 we analyse GMFs with irrational coefficients in their q-expansion for
o(g) = o(h) = p, prime, where g, h Zp2 (then g, h are independent, i.e., g A = hB

for all A, B = 1, . . . , p 1). We denote the fixing group for Z hg (p ) by h,g which is
obviously conjugate to h,g where

1 0
.
h,g = = p p1 , h,g , p
0 p
Various properties of (3) arising from orbifold considerations can be summarised as
follows [8]:
a b
(i) When all elements of g, h are normal elements of M then for = c d
SL(2, Z)

d b
a + b
h
h g
.
=
Z
( ) = Z c a ( ),
(4)
g
h g
c + d
Hence (p) h,g , (in fact (p) h,g ) with

a b
a b
1 mod p
(p)
,
=
c d
c d
0 mod p
0 mod p
1 mod p

.
In particular, since and act equally we have

1
h
h
Z
= Z 1 ,
g
g
which property is known as charge conjugation invariance.
(ii) For a normal non-Fricke element g of order p

h
Z
( ) = const + O q 1/p ,
g
477
(5)
(6)
i.e., Hg has zero vacuum energy and the GMF is non-singular at q = 0 [7,11].
The value of (3) at any parabolic cusp a/c with (a, c) = 1 is determined by the vacuum
energy of the g a hc twisted sector from
(4). If all elements of g, h are non-Fricke then

there are no singular cusps so that Z hg is holomorphic on H/ (p) and hence is constant.
This accounts for the constant GMFs referred to in the Generalised Moonshine Conjecture
above. We therefore assume from now on that at least one element of g, h is Fricke which
we chose to be g without loss of generality.

(iii) For a normal Fricke element g of order p the vacuum energy of Hg is 1/p and
the GMF is singular at q = 0 [8]:

h
1/p
s/p
.
Z
(7)
+0+
ag,s (h)q
( ) = g (h) q
g
s=1
Here g (h) is a pth root of unity describing the central extension of the action of h Cg

on Hg [8]. It is further assumed that g can be chosen so that [8]

g g a hb = g (g)a g (h)b ,
(8)
g (1) = 1.
Then by considering a T transformation in (4) with h = 1 we have

2i
g (g) = p exp
.
p
(9)
Let Gg Cg /g. If Cg = g Gg then h and gh are not elements of the same Cg
conjugacy class and we assume that g can be chosen so that [8]

g xhx 1 = g (h)
(10)
for all x Cg . By inspection from the ATLAS [14] h Gg is conjugate to ha , for some
a = 0 mod p, so that (8) implies
g (h) = 1.
(11)
The coefficient af,s (h) is called a head character for the given GMF and is a reducible
character of Gg [8,13]. If g a hc is Fricke then due to (7) and (4) the GMF is singular at a/c
with residue g a hc (hd g b ). The GMF (3) is holomorphic at all other points on H. Once
these singularities are known, then (3) can be analysed to check whether it is constant or is
a hauptmodul
for an appropriate genus zero modular group. The singularities and residues

of Z hg are restricted by certain orbifolding constraints discussed in [8].
478
(iv) Given the uniqueness of the twisted sectors for H , under conjugation by any

element x M then x(Hg )x 1 is isomorphic to Hxgx 1 so that if Cg = g Gg with
h Gg [8]

h
xhx 1
Z
=Z
(12)
.
g
xgx 1
3. Irrational GMFs for g = p+
3.1. Monster centralisers and irrational characters
We now briefly describe all irrational GMFs (3) where g is Fricke of prime order p, i.e.,
g = p+ in ConwayNorton notation [3] and h of order p. A GMF is said to be irrational
if it has at least one irrational head character (7). A character is said to be irrational if it
is irrational for at least one conjugacy class. Otherwise, it is said to be rational. Rational
cases are discussed in [8].
From the ATLAS [14] we see that irrational characters occur for p = 2, 3, 5, 7, 11
and 13. From the explicit calculations for the head characters of C2+ = 2.B (p = 2) and
C3+ = 3.Fi (p = 3) [8] we observe that there are no irrational head characters for p = 2
and that there is only one pair of conjugate irrational head characters for p = 3, with h of
order 54. For p = 5, 7, 11 and 13 the centraliser Cp+ = p+ Gp+ for simple group
Gp+ is shown in Table 1.
We will focus now our analysis on the groups from Table 1. For h Gg from (11) the
GMF (7) has leading q expansion in so-called Normalised Function Form [16]:

1
h
Z
( ) = 1/p + 0 + O q 1/p .
(13)
g
q
This is sufficient to ensure that g, h are independent generators of g, h since otherwise
hB = g A for some A = 0 mod p implies the contradictory relation (g (h))B = (g (g))A =
pA from (9). We make the following useful observations concerning the irreducible
characters for the groups from Table 1, which can be checked by inspecting the appropriate
ATLAS character tables [14]. In all cases irrational characters can only occur when o(h)
is divisible by p. When p = 5, 7, 11 any
character is
irrationality for an irreducible
quadratic, i.e., for h Gp+ , (h) = a b, (h) (hd ) = a b (d = 2 when p = 5

and d = 1 when p = 7, 11, see below), for some a, b Q where b = 0 for rational (h).
The HN group contains one pair of conjugate irrational classes of orders 5, 15, 20, 25
Table 1
p+ centralisers for p = 5, 7, 11, 13
g = p+
Cp+
Gp+
Name
5+
7+
11+
13+
5 HN
7 He
11 M12
13 L3 (3)
HN
He
M12
L3 (3)
HaradaNorton
Held
Mathieu
479
and 30, and two pairs of conjugate irrational classes of order 10. The He group contains
one pair of conjugate irrational classes of orders 21 and 28, and two pairs of conjugate
irrational classes of orders 7 and 14. The M12 group contains only one pair of conjugate
irrational classes of order 11. The only non-quadratic irrationalities occur when p = 13 for
L3 (3) which has four conjugate irrational classes of order 13.
Theorem 3.1. Consider g = p+ and h Gp+ with o(h) divisible by p for p = 5, 7, 11, 13.
a
Then for integers a, b = 0 mod p, Z hg b is either equal to Z hg or one of its algebraic
conjugates.
Proof. For g = p+ we have xgx 1 = g b for some x M. For any irreducible character
of Gg and for all h Gg , x (h) (xhx 1 ) is also an irreducible character of the
same dimension. Furthermore xGg x 1 = Gg b = Gg implies that the number of elements
in the conjugacy classes in Gg for h and xhx 1 are equal. Since h Gg , h is conjugate
to ha for some a = 1 mod o(h) from an inspection of the ATLAS character tables [14].
Hence xhx 1 is conjugate to (xhx 1 )a and is also member of a Gg class. Therefore
g b (xhx 1 ) = 1 also. For p = 5, 7, 11 one can show that if x = , then x = , the
algebraic conjugate [8]. Thus xhx 1 is conjugate to a power of h in Gg . When p = 13 there
are four irrational characters that are algebraically conjugate. Then x = 2r , r = 0, 1, 2
or 3, where by definition ( k)(h) (hk ), i.e., the operation k replaces every root
of unity in the character value, prime to k by its kth power (see below). This implies
that xhx 1 is either an element of the same class of Gg as h, or of a class algebraically
Gg
conjugate to h, i.e., there exists A such that xhx 1 hA . Therefore from (12) in all cases
there exists A such that

a
h
h
Z
(14)
=Z b .
g
g
a
The latter is either equal to Z hg b or to one of its algebraic conjugates and hence the result
follows.
From now on we will assume that o(h) = p, prime. Then the set of elements

S ha Gg , a = 1, 2, . . . , p 1 ,
(15)
with irrational characters can be divided into classes characterized by the so-called
Dirichlet character of a as follows. Let Cp1 m be the cyclic group of order p 1
acting on S generated by
m : h hm ,
i.e., the Galois group for the polynomial (x p 1)/(x 1). All irrational characters of
Gg are fixed by a cyclic subgroup (mN) of Cp1 , for some N . Then S can be divided
into classes with representatives

a
h S a Cp1 (mN) CN ,
(16)
480
characterised uniquely by a Dirichlet character of order N defined as follows. A Dirichlet

character D of the Galois group Cp1 is a homomorphism:
D : Cp1 C .
(17)
Then these characters form a group under pointwise multiplication which is isomorphic
(N)
then gives an
to Cp1 , but not under any canonical map. The character of order N , D
isomorphism

(N)
D : Cp1 (mN) N
and thus characterises uniquely the representatives (16).
For p odd, the character group (17) is cyclic of even order p 1, hence it has an element

(2)
(2)
. This is the Legendre symbol: D
(a) pa a (p1)/2 mod p = 1
of order 2, D
which characterises (16) in the case of quadratically irrational characters for which N = 2.
This applies to the characters of Gp+ for p = 5, 7, 11.
There is a character of order N = 4 precisely when 4|(p 1) (e.g., p = 13). There are
then exactly two characters of order 4, because a cyclic group of order divisible by 4 has
exactly two elements of order 4. These two characters are inverse to each other and hence
are complex conjugates of each other. Their quotient will be the Legendre symbol. For
p = 13, we may take m = 2 as a generator of the Galois group C12 . The fixing group for
the irrational characters of G13+ = L3 (3) is then 3 [14]. We get a Dirichlet character
(4)
D
of order 4 by mapping 2 to 4 = i = exp(i/2). The other element of order 4 obtains
(4)
by mapping 2 to i. If a is in C12 with a = 2r mod 13 then D (a) = exp(ri/2). Thus
(4)
D
:
1, 3, 9 1,
12, 10, 4 1,
5, 2, 6 i,
8, 11, 7 i.
So 1, 12, 5, 8 are representatives of C12 /3 which map to the four different 4th roots of
(4)
unity. Thus the value of D characterises each of the four conjugate irrational classes of
order 13 in L3 (3).
3.2. The genus zero property for irrational GMFs for g = p+ and o(h) = p
We now come to the main purpose of this paper which is to demonstrate the genus zero
property for Generalised Moonshine Functions (GMFs) (3) in the irrational cases where
g = p+ and o(h) = p. Such cases occur when p = 5, 7, 11, 13. Our aim is to show that the
fixing group h,g permutes all of the singular points of the GMF. In each case we find that
h,g is a subgroup of = SL(2, Z) so that h,g / (p) is a subgroup of L2 (p) = / (p),
the group permuting all H/ (p) inequivalent cusps. Then all possible singularities of the
GMF at the cusps = a/c with (a, c) = 1 (related to the g a hc Fricke-twisted sector) are
identified under h,g and the corresponding GMF is a hauptmodul for a genus zero group.
In our analysis we use the following two theorems. The first one gives some constraints
following from the orbifolding of the Moonshine Module with respect to the Abelian group
g, h. See [8] for details.
481
Theorem 3.2. Let g, h M be independent commuting elements where both g and h are
Fricke such that g (h) = 1 and where all elements of g, h are normal. Let u, v be any
independent generators for g, h. If u is Fricke then there is a unique A mod o(u) such
that uA v is Fricke with uA v (u) = 1 and o(uA v) = o(v).
The next theorem gives the relation between the fixing groups of hauptmoduls with
algebraically conjugate coefficients [15,16]:
Theorem 3.3. Let G be a fixing group of a hauptmodul such that (p) G. Let k be any
a b
integer, coprime to p. Choose coset representatives of (p) in G, ci di , such that k|ci .

i i
Then the operation k, which replaces every pth root of unity by its kth power yields a
hauptmodul with a fixing group

ai
G k = (p),
ci /k
kbi
di

which is independent of choices made.

In terms of Generalised Moonshine the action of k on the GMF Z
hk
Z g because k takes a character of h to that of hk . Hence
hk ,g = (h,g ) k.
h
g
is to map it to
(18)
We begin with a preliminary lemma.

Lemma 3.4. For p = 5, 7, 11, 13 we have 00 (p) (p), p for p of order (p 1)/2
in (p), where

00 (p)
a
c
b
d

, b = c = 0 mod p
and

2 5
2 7
,
7 =
,
5 13
7 25

40 11
85 13
,
13 =
.
11 =
11
3
13 2
5 =
Proof. By inspection of the possible solutions for a, d mod p of ad bc = 1 where b,

c = 0 mod p one can check the validity of the statement.
Note that 0 (p2 ) and 00 (p) are conjugate where 0 (p2 ) = p 00 (p)p1 . Therefore
0 (p2 ) invariance follows from invariance under (p) and p .
482
3.2.1. Irrational GMFs for g = 5+ and o(h) = 5

The HN group has one pair of quadratically irrational classes of order 5. Then m = 2,
N = 2 in (16). The two subsets of S (15) conjugate in G5+ HN are characterised by

G5+
G5+
the Legendre symbol a5 : h h4 (i.e., ha where a5 = 1) and h2 h3 (ha where
a
M s
M
4
2 M
3
5 = 1) [14]. Therefore gh gh and gh gh . Since g0 g0 for s = 0 mod 5
for any Monster element g0 of order 5 we have the following two disjoint sets S1 , S2
of conjugate elements in M defined by

n s n
S1 gh
= 1, s = 0 mod 5 ,
5

n s n
= 1, s = 0 mod 5 ,
S2 gh
5
i.e., S1 consists of the elements gh g 2 h2 g 3 h3 g 4 h4 gh4 g 2 h3 g 3 h2 g 4 h
and S2 consists of gh2 g 2 h4 g 3 h g 4 h3 gh3 g 2 h g 3 h4 g 4 h2 , where
S1 S2 = {g A hB | A, B = 0 mod 5}. The irrational GMFs occur when the elements of
the two sets S1 , S2 are of different Fricke type, e.g., S1 is Fricke but S2 is non-Fricke.
Indeed, otherwise we have some of the cases, analysed previously in [8] which lead to
rational GMFs.
Proposition 1. For g = 5+ and h Gg HN of order 5, the following class structures

give rise to a genus zero fixing group h,g for irrational GMFs Z hg (5 ) as follows:
(i) g Fricke, h non-Fricke, S1 Fricke, S2 non-Fricke,

h,g = 0 (25), T 1/5 W25 5 T 1/5 ;
(ii) g Fricke, h non-Fricke, S1 non-Fricke, S2 Fricke,

h,g = 0 (25), T 2/5 W25 T 2/5 ,
where
T r/5 =

1 r/5
,
0 1
5 =
2 1
25 13

and W25 =

0 1
,
25 0
the Fricke involution.

Proof. Firstly we will prove that h,g contains 0 (25). Due to Lemma 3.4 it is sufficient
to demonstrate invariance with respect to 5 .
d
According to (14) we have Z hg = Z hg 2 for some d. Clearly hd Gg is of order 5 and
has irrational characters. From inspection of the ATLAS [14] we haveonly
two such classes
with algebraically conjugate irreducible characters distinguished by d5 = 1. Therefore,

there are two possibilities: either Z hg = Z gh2 or Z hg 2 . The first possibility can be ruled
g 2
1
out since it implies after an ST 1 transformation that Z ggh = Z g 2 h . This is impossible
since gh S1 and g 2 h S2 and S1 and S2 are of different Fricke/non-Fricke type for
483
both cases (i) and (ii) above. Therefore the second possibility remains which implies 5
invariance and the result follows.
Case (i). From the analysis in [8], where all rational cases with o(g) = o(h) = 5 are
g a hb
analysed, it follows that Z g 2 h3 cannot be rational for (a, b) = (0, 0) mod 5. If we choose
a, b such that g a hb Gg 2 h3 (g 2 h3 (g a hb ) = 1), this GMF must be quadratically irrational.
Since the HaradaNorton group has only one pair of irrational classes of order 5 so that
g a hb

Z g 2 h3 is either equal to Z hg or its algebraic conjugate. We may then further restrict the
g a hb

choice of a, b so that Z gh = Z g 2 h3 .
Firstly we observe that b = 0 mod 5 since g a hb should be non-Fricke since h is. If
g 2 h3+b

and
a = 0 mod 5 then Z hg = Z gh2 h3 and after a T 1 transformation Z gh
g = Z g 2 h3
therefore g 2 h3+b is also Fricke and hence b = 1, 3 mod 5. Applying a T transformation
we can also see that g 3 h2+b is Fricke and hence b = 2, 4 mod 5. Thus
a = 0 mod 5.
Since g a hb is non-Fricke we conclude that (a, b)
/ {(s, ns) | n5 = 1, s = 0 mod 5}.
a T 1 transformation we also conclude that (a, b)
/ {(2, s), (s 2, ns 3) |

Applying
n
=
1,
s
=

0
mod
5}.
Thus
either
(a,
b)
=
(1,
2)
or
(4,
3)
both leading to the same
5
2
1
4
3
GMF since (gh ) = g h and all characters are real.
gh2

Thus we have shown that Z hg ( ) = Z g 2 h3 ( ) = Z gh ( ) with T S5 T .
Furthermore we find that all the singular cusps of the GMF are identified under h,g =
00 (5), which is therefore a genus zero fixing group. Clearly h,g is a subgroup
of where 3 = 52 = (5 )2 = 1 mod (5) and so h,g / (5) = D3 which is a
(maximal) subgroup of L2 (5). The corresponding hauptmodul for h,g = 0 (25),
can
be explicitly expressed as

( )( + 2/5)( + 3/5)
h
Z
(5 ) =
+1 5
g
( + 1/5)( + 4/5)(25 )

3 5 5
1
=q +0+
q 10q 2 + 5q 3
2
2

25 25 5 5
4
+
+ (21 + 5 5 )q +
(19)
q +
2
2
which is known as the 25 b series [17].
Assuming that the GMF is replicable and based on numerical matching L. Queen
conjectures that the head characters for h G5+ HN can be expanded in terms of the
irreducible characters i (h) of HN (in ATLAS notation [14]) to give [18,19]

1
h
Z
(5 ) = + 0 + 1 (h) + 3 (h) q + 4 (h)q 2 + 1 (h) + 5 (h) q 3
g
q

+ 1 (h) + 2 (h) + 5 (h) + 6 (h) q 4

+ 1 (h) + 2 (h) + 4 (h) + 5 (h) + 11 (h) q 5 +
(20)
and therefore (19) corresponds to the head character expansion (20) for the 5C class of
HN [14].
484
Case (ii). This class structure is obtained from that in case (i) by replacing h by h2 .
Therefore the fixing group given by (18) with k = 2 is h,g = 0 (25), T 2/5 W25 T 2/5 . Its
( )( +1/5)( +4/5)
hauptmodul is then (
+2/5)( +3/5)(25 ) + 1 + 5. This is known as the 25 a series [17]
and has algebraically conjugate coefficients to those of case (i). The head character
expansion corresponds to the 5D class of HN (20).
The Held group has two pairs of quadratically irrational classes of order 7. In (16)
m = 3, N = 2. The two subsets of powers for each h of order 7 (15), conjugate in

M
M
G7+ = He are: {ha } for a7 = 1 and {ha } for a7 = 1 [14]. Therefore gh gh2 gh4
M
and gh3 gh5 gh6 . Since g0 g0s (s = 1, 2, . . . , p 1) for any Monster element g0 , we
have the following two disjoint sets S1 , S2 of conjugate elements in M defined by

s n

= 1, s = 0 mod 7 ,
S1 ghn
7

n s n
S2 gh
= 1, s = 0 mod 7 ,
7
where S1 S2 = {g A hB | A, B = 0 mod 7}.
class structures
Proposition 2. For g = 7+ and h of order 7, h G7+ He the following

give rise to a genus zero fixing group h,g for irrational GMFs Z hg (7 ):
(i) g Fricke, h non-Fricke, the set S1 Fricke, S2 non-Fricke,

3
2/7

;
h,g = 0 (49), ,
35 3
(ii) g Fricke, h non-Fricke, S1 non-Fricke, S2 Fricke,

3 2/7
h,g = 0 (49), ,

;
35 3
(iii) g Fricke, h Fricke, S1 and S2 sets Fricke. There are two possible fixing groups,
which are isomorphic and usually both denoted by 7||7+.
d
According to (14) we have Z hg = Z hg 2 = for some d. We need to show that

1
Z hg = Z gh2 ( ) Z hg (7 ). Assume that d7 = 1. Then we have Z hg = Z hg 2 =
1
Z gh4 = Z hg 8 Z hg which is impossible since h and h1 are not conjugate in Gg .
d
Therefore 7 = 1 and the statement follows.
When p = 7 there are two pairs of conjugate irrational characters in G7+ He [14].
If we consider a GMF on another Fricke twisted sector, for convenience, a g 3 h5 twisted
g a hb
sector, then for a pair (a, b) = (0, 0) mod 7 and g a hb Gg 3 h5 (g 3 h5 (g a hb ) = 1), Z g 3 h5
485

must be quadratically irrational and g, h = g 3 h5 , g a hb , so that it is either equal to Z hg
or its algebraic conjugate. By choosing an appropriate multiple of (a, b) we may further
restrict the choice of a, b so that

a b
h
g h
Z
=Z 3 5 ,
(21)
g
g h
g 3 h5+b

Case (i). Suppose a = 0 mod 7. Then T 1 transformation gives Z gh
g = Z g 3 h5
and since gh is Fricke, so is g 3 h5+b , i.e., it is in the S1 Fricke set and therefore
b = 4, 3, 6 mod 7. Since h4 h2 h and h3 h5 in the corresponding centraliser,
b = 1, 2, 5 mod 7. Thus a = 0 mod 7.

/ {(s, ns) | n7 = 1, s = 0 mod 7}.
Since g a hb S2 is non-Fricke, b = 0 mod 7, (a, b)
g 3+a h5+b

Furthermore, a T 1 transformation of (21) gives Z gh
, and since gh is
3 5
g =Z
ngh
3+a
5+b
Fricke, so is g h
thus (a, b)
/ {(s 3, ns 5), (4, s) | 7 = 1, s = 0 mod 7}.
If a given pair (a, b) is inadmissible (i.e., does
not satisfy (21)) so are the pairs leading

to conjugates in the same set: (na, nb), n7 = 1. Therefore (a, b) = (5n, 4n), n7 = 1.
3 2
g 5 h4

Hence Z gh ( ) = Z g 3 h5 ( ) = Z gh ( ) where = 5 3 and thus the GMF is
modular invariant. Furthermore we find that all the singular cusps of the GMF are identified
under h,g = 00 (7), which is therefore a genus zero fixing group. Introducing
1
3
= 7 1 = 1 2 , one can check that 3 = 2 = ()3 = 1 mod (7). Therefore

(7), 7, / (7) A4 , a subgroup of L2 (7). The hauptmodul for the genus zero fixing
group h,g = 0 (49),
is

7
( )( + 3/7)( + 5/7)( + 6/7) 1
h
Z
+ i
(7 ) =
4
g
2
2
(7 )

3
3 7 2
7
1
5
= +0+ +i
q + i
q + 2q 3
q
2
2
2
2

+ 3 i 7 q 4 3q 5 +
(22)
known as the 49 a series [17].
Assuming that the GMF is replicable and based on numerical matching L. Queen
conjectures that the head characters for h G7+ He can be expanded in terms of the
irreducible characters i (h) of He (in ATLAS notation [14]) to give [18,19]

1
h
Z
(7 ) = + 0 + 2 (h)q + 3 (h) + 4 (h) q 2 + 1 (h) + 6 (h) q 3
g
q

+ 1 (h) + 6 (h) + 11 (h) q 4

+ 1 (h) + 2 (h) + 3 (h) + 6 (h) + 14 (h) q 5 + .
(23)
Thus the expansion (22) is associated with the 7E class of He.
Case (ii). This class structure is obtained from that in case (i) by replacing h by h1 .
Therefore the fixing group is given by (18) with k = 1 mod 7. The fixing group contains
3 2
= 5 3 , i.e., h,g = 00 (7), . h is in 7D class of He, with algebraically conjugate
486
characters to those of 7E. The GMF

7
( )( + 1/7)( + 2/7)( + 4/7) 1
h
+ +i
Z
(7 ) =
g
4 (7 )
2
2
corresponding to the 49 b series [17], is a hauptmodul for the genus zero fixing group
h,g = 0 (49), . This hauptmodul has algebraically conjugate coefficients compared
to (22).
Case (iii). From earlier remarks there are two algebraically conjugate
order
7 classes
of He remaining. We will demonstrate S-invariance. Since Z hg = Z gh2 , after an
2
S-transformation Z gh = Z gh and thus h (g 1 ) = h (g 2 ), i.e., h (g) = 1. So

g Gh Gg . We have only one pair of irrational classes left, hence Z gh = Z hg , or

Z gh = Z hg , its conjugate. Assume that Z gh = Z hg , then after an ST 1 transformation
A
Z ggh = Z hgh , which implies that Z ggh = Z hgh for any A. Taking A = 2, after
1 1
1
a T 1 transformation we obtain Z gghh = Z (g ghh) . This is impossible since the

irrational characters of He are complex. Hence Z hg ( ) = Z gh ( ) = Z hg (S ).
Consider the GMF Z , algebraically conjugate to Z

1
h
h
Z
(24)
Z
.
g
g
g b h2a
g b h2a
1
From (21) and S invariance we have Z hg = Z hg = Z h2 = Z g 5 h6 = Z (g 5 h6 )2 =
g b h2a
g b h2a
Z g 3 h5 = Z g 3 h5 , i.e., if Z satisfies (21) for a pair (a, b) then Z satisfies (21) for

the pair (b, 2a) (and therefore (nb, 2na) with n7 = 1).
We can show that Z and Z cannot satisfy (21) for one and the same pair (a, b)
(otherwise they would have the same fixing group and possibly hauptmodul). Let us assume
that it is possible and therefore (a, b) = (2b, 4a) mod 7 (n = 2): i.e., (a, b) = (s, 3s),
gh3
3
s = 0 mod 7. For s = 1 we have Z hg = Z g 3 h5 . Since Z hg = Z hg 5 it follows that
5 2

Z hg = Z gghh . Applying two times the last transformation we get Z hg which is a

contradiction since clearly h5 (g 4 h5 ) = 1. Thus (a, b) = (s, 3s) for 7s = 1. For s = 3
g 3 h2
g 4 h
we have Z hg = Z g 3 h5 . Applying two times this transformation we get Z hg = Z h2

which is a contradiction since clearly h2 (g 4 h) = 1. Thus (a, b) = (s, 3s) for 7s = 1
and finally (a, b) = (s, 3s) in general.
g 3 h5
Suppose a = 0 mod 7. Then Z hg = Z hb which is impossible, because h (g) = 1,
hb (g) = 1, hb (hb ) = 7 . Thus a = 0 mod 7 and similarly b = 0 mod 7.
Since g 3 h5 (g 3 h5 ) = 7 , g a hb must not be a power of g 3 h5 , (a, b) = (3s, 5s). Applying
once again (21) we have

2
h
g a(3+b) h5a+b
.
Z
=Z
g
g 22a h12b
487
g 5b2 +4b3
Let a = 4b 2 mod 7 and b = 1 or 2 mod 7. Then Z hg = Z 2b2 12b is possible only
g
h
if g 2b2 h12b (g) = 1. However according to Theorem 3.2, since h (g) = 1 it is not possible
to have g 2b2 h12b (g) = 1 unless 2 b2 = 0 mod 7 which is not the case. Therefore
their conjugates of the form (na, nb),
b) = (4, 1), (2, 2), and we exclude
n also

(a,
n
=
1:
(a,
b)
/
{(4n,
n),
(2n,
2n)
|
=
1}.
Finally, we exclude all pairs, which
7
7
can be represented as (b, 2a) from an already excluded pair (a, b). It is due to the
fact that if (a, b) satisfies (21), (b, 2a)

n must satisfy the same relation for Z (24). The
remaining pairs are (a, b) = (5n, 4n), 7 = 1 ((5, 4), (3, 1) and (6, 2)); their Z -partners
of the form (b, 2a): (3, 4), (6, 1), (5, 2). This leads to modular invariance under
3
2
4 1
= 5 3 for Z (for (a, b) = (5, 4)), and = 5 1 for Z (for (b, 2a) = (3, 4)).
Note that is of order 2 in (7) and is of order 4 in (7), i.e., the mapping to
g 3 h5 Fricke twisted sector (21) requires modular transformations of different order in
each case. Each of the two fixing groups h,g = 00 (7), , S, 00 (7), , S matches
all the singular cusps of the GMF and therefore each group is of genus zero. These two
fixing groups are isomorphic. Indeed, introducing 1 = S, 2 = 7 S, 3 = S, we have
12 = 22 = 32 = (1 2 )3 = (2 3 )3 = (3 1 )2 = 1 mod (7), the defining relations of the
group S4 . Introducing 1 = S, 2 = S , we have 12 = 23 = (1 2 )4 = 1 mod (7), also
defining the group S4 , a maximal subgroup of L2 (7). Therefore (7), 7 , , S/ (7) S4
and (7), , S/ (7) S4 . According to the definition in [16] for both groups we use
the same notation h,g = 7||7+. The GMFs represents the hauptmoduls for these groups
and can be found from those of case (i) with symmetrization:

( )( + 3/7)( + 5/7)( + 6/7)
h
Z
(7 ) =
g
4 (7 )
(49 )( + 1/7)( + 2/7)( + 4/7)
+i 7
4 (7 )

3 7
3
= q 1 + 0 + + i
q + (1 i 7 )q 2 + 9q 3
2
2
4

+ 3 1 i 7 q + 4q 5 + ,

( )( + 1/7)( + 2/7)( + 4/7)
h
Z
(7 ) =
g
4 (7 )
(49 )( + 3/7)( + 5/7)( + 6/7)
.
+i 7
4 (7 )
Z has a series expansion with complex conjugate coefficients to Z. According to (23)

Z is the GMF when h is an element of 7A class of He, Z when h is an element of 7B
class of He.
The M12 group has one pair of quadratically irrational classes of order 11. In (16) m = 2,
N
=
two subsets
of powers of h (15) conjugate in G11+ M12 : {ha } for

a 2. There are
a
a
11 = 1 and {h } for 11 = 1 [14].
488
Proposition 3. For g = 11+, h G11+ M12 and o(h) = 11, the class structure where
all elements of the group g, h are Fricke gives rise to a pair of irrational GMFs with
isomorphic genus zero fixing groups (usually both denoted by 11||11+).

According to (14) Z hg = Z hg 4 for some d. We need to show that Z hg ( ) =

Z hg 4 ( ) Z hg (11 ). Clearly hd Gg is of order 11 and has irrational characters. From
inspection of the ATLAS [14] we have only
conjugate
such classes with algebraically
d
d two
= 1. Assume that 11
= 1. Then we
irreducible characters distinguished by 11
h(1)5
1
have Z hg = Z hg 4 = Z g 45 = Z hg which is impossible since h and h1 are not
d
= 1, and the statement follows. One can demonstrate also
conjugate in Gg . Therefore 11
that h (g) = 1 and hence S invariance follows as in Proposition 2(iii).
Let us now consider a GMF related to another Fricke twisted sector, for convenience,
say a gh2 twisted sector. When o(h) = 11 there is only one pair of elements in M12 with
g a hb
irrational characters [14]. So Z gh2 cannot be rational for (a, b) = (0, 0) mod 11. If we
choose a, b such that g a hb Ggh2 (gh2 (g a hb ) = 1), this GMF must be quadratically
irrational. Since M12 has only one pair of irrational classes of order 11 (and no rational
g a hb

ones), then Z gh2 is either equal to Z hg or its algebraic conjugate. We may then further
restrict the choice of a, b so that

a b
h
g h
Z
=Z
(25)
.
g
gh2
Using (25) and S invariance we have Z
(g 5b h2a )1
g 5b h2a
and hence
Z h2 g 5.2 = Z
gh2
h1
g
=Z
h4
g
=Z
h4.(5)
g 5
=Z
g 5
h2

5b 2a
h
g h
,
Z
Z
g
gh2
where the definition of Z is as in (24). Thus if (a, b) is an invariance for Z as in (25)

then (5b, 2a) is an invariance for Z . Using similar techniques as in Proposition 2(iii)
we can prove that (a, b) = (5b, 2a) mod 11, i.e., (a, b) = (s, 9s). Furthermore, since
gh2 (gh2 ) = 11 , (a, b) = (s, 2s).
As in Proposition 2(iii) one can show that a, b = 0 mod 11. Applying once again the
g a(1+b)h2a+b2
transformation (25) we have Z hg = Z g 2a+1 h2b+2 . Let a = 5b2 mod 11 with b2 =
g 5b2 (1+b)

1 mod 11. Then Z hg = Z 1b2 2b+2 is possible only if g 1b2 h2b+2 (g) = 1. However
g
h
according to Theorem 3.2, since h (g) = 1 it is not possible to have g 1b2 h2b+2 (g) = 1
unless b2 = 1 mod 11 which is not the case. Therefore (a, b) = (5b2, b) for b = 2,
3,
n
= 1,
4, 5 mod 11, and we exclude also their conjugates of the form (5b2n, bn), 11
b = 0, 1 mod 11. Finally we exclude all pairs which can be represented as (5b, 2a)
489
from an already excluded pair (a, b). It is due to the fact that if (a, b) satisfies (25),
(5b, 2a) must satisfy the same relation for Z .
n
= 1 and their Z -partners of the
The remaining pairs are (a, b) = (10n, 10n), 11
1 1
n
form (5b, 2a): (6n, 2n), 11 = 1. This leads to modular invariance under = 2 1
1
5
for (a, b) = (10, 10) say for Z, and = 2 9 for (a, b) = (6, 2) for Z . The two
fixing groups h,g = 00 (11), , S, 00 (11), , S are conjugate and isomorphic to
A5 mod (11) (generated by 1 = 11 S, 2 = S, 3 = , which satisfy 12 = 22 =
32 = (1 2 )3 = (1 3 )2 = (2 3 )5 = (1 2 3 )5 = 1 mod (11), the defining relations
of A5 ). A5 is a (maximal) subgroup of L2 (11). For both fixing groups we use the
notation h,g = 11||11+ [16]. All the singular cusps of the GMF are identified under the
corresponding h,g which is therefore a genus zero fixing group.
The q expansions to O(q 5) of the hauptmoduls for h,g = 11||11+ are [18,19]:

1
1
11
11 3
1
h
2
i
i
Z
(11 ) = + 0 +
q + 2q +
q
g
q
2
2
2
2

11 5
1
i
1 i 11 q 4
(26)
q + .
2
2
Assuming that the GMF is replicable the head character expansion for h G11+ M12
in terms of the irreducible characters i (h) of M12 (in ATLAS notation [14]) is [18,19]

1
h
Z
(11 ) = + 0 + 1 (h) + 4 (h) q + 1 (h) + 6 (h) q 2
g
q

+ 1 (h) + 4 (h) + 6 (h) + 7 (h) q 3

+ 1 (h) + 25 (h) + 6 (h) + 7 (h) + 13 (h) q 4

+ 21 (h) + 24 (h) + 5 (h) + 26 (h) + 27 (h)

+ 11 (h) + 12 (h) + 13 (h) q 5 + .
(27)
The upper signs in (26) according to (27) are for h an element of 11A class of M12 , the
lowerfor h an element of 11B class of M12 .
Propositions 13 lead us to the following statement:
Corollary 3.5. For g = p+, o(h) = p and p = 5, 7, 11 the irrational GMFs obey

d

ad
h
h
Z
= 1,
=Z a
iff
g
g
p
where the fixing group h,g contains 0 (p2 ).
The L3 (3) group has four conjugate irrational classes of order 13. There are four subsets
(4)
of powers of h (15) conjugate in G13+ = L3 (3) determined by D
(a) {1, i}. Since
490

M
g0 g0s (s = 1, 2, . . . , p 1) for any Monster element g0 , we have the following four

disjoint sets of conjugate elements in M defined by:
s (4)

S0 ghn D
(n) = i, s = 0 mod 13 ,
s (4)

S1 ghn D (n) = 1, s = 0 mod 13 ,
s (4)

S2 ghn D (n) = i, s = 0 mod 13 ,
s (4)

S3 ghn D (n) = 1, s = 0 mod 13 ,

where 3r=0 Sr = {g A hB | A, B = 0 mod 13}.
The fixing group for the irreducible characters of L3 (3) is 3, so we expect invariance
3 0
2 =
under 13
mod (13), which is of order 3 in (13), and therefore the order
0 9
of h,g / (13) is divisible by 3. The only possibility is h,g / (13) = A4 , a maximal
subgroup of L2 (13). Then h,g contains an element , such that 3 = 2 = ()3 =
11
23r
1 mod (13). There are four solutions of the above relation, (r) = 92.2r 67 , r =
0, 1, 2, 3, corresponding to the fixing groups related to the four irrational classes. The result
can be rigorously formulated in the following
Proposition 4. For g = 13+ and h of order 13, h G13+ L3 (3) the class structure with
g Fricke, h non-Fricke, one of the sets Sr (r = 0, 1, 2, 3) Fricke and the others non-Fricke
gives rise to an irrational GMF with a genus zero fixing group h,g = (13), , (r),
such that h,g / (13) = A4 .
Proof. Let us assume that the elements of S1 are Fricke, the others are non-Fricke. Firstly
d
we show invariance under . According to (14) for some d, Z hg = Z hg 3 . A T 1
g 3 hd

and hence g 3 hd S1 since gh S1 is Fricke.
transformation gives us Z gh
g = Z g3

(4)
Hence D
(d) = 1 so Z hg ( ) = Z gh3 ( ) Z gh ( ).
Now we demonstrate (1) invariance. To this end, as before, we consider a general
g a hb
transformation to the (gh)11 Fricke twisted sector. It is clear that Z g 11 h11 cannot
be rational for (a, b) = (0, 0) mod 13. If we choose a, b such that g a hb Gg 11 h11
(g 11 h11 (g a hb ) = 1), this GMF must be still irrational, since there are no rational characters
g a hb

of order 13. So that Z g 11 h11 is either equal to Z hg or one of its algebraic conjugates. We
may then further restrict the choice of a, b so that

a b
h
g h
Z
= Z 11 11 .
g
g h
Taking a T n transformation we have Z
g n h
g
=Z
g a2n hb2n
g 11 h11
(4)
. For D (n) = 1, g n h S1
is Fricke and hence g a2n hb2n S1 also, which leads to (a, b) = (3, 5), (9, 2), (1, 6). The
solution (9, 2) provides the modular transformation (1). Thus h,g = (11), , (1).
All the singular cusps of the GMF are identified under h,g which is therefore a genus
491
zero fixing group. Note that h,g does not contain 0 (169). The fixing groups for the other
cases follow from (18).
4. Conclusions
We have shown how irrational Generalised Moonshine can be understood from the
analysis of the class structure for a pair of Monster elements of prime order p. This follows
from constraints originating from the Abelian orbifolding of the Moonshine Module and
properties of centraliser irreducible characters. We have explicitly demonstrated the genus
zero property for irrational Generalised Moonshine Functions (GMFs) in all cases. The
methods developed in this paper and in [8] can in principle be extended to analyse other
GMFs towards proving the genus zero property in general.
Acknowledgements
We are very grateful to S. Norton for providing us with information on replicable series
and to G. Mason for many valuable discussions. We acknowledge funding from Enterprise
Ireland under the Basic Research Grant Scheme.
References
[1] I. Frenkel, J. Lepowsky, A. Meurman, Proc. Nat. Acad. Sci. USA 81 (1984) 3256.
[2] I. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster, Academic Press, New
York, 1988.
[3] J.H. Conway, S.P. Norton, Bull. London Math. Soc. 11 (1979) 308.
[4] R. Borcherds, Invent. Math. 109 (1992) 405.
[6] S.P. Norton, Proc. Symp. Pure Math. 47 (1987) 208.
[7] M.P. Tuite, Contemp. Math. 193 (1996) 353.
[8] R. Ivanov, M.P. Tuite, Nucl. Phys. B 635 (2002) 433, preceding article in this issue.
[9] P. Goddard, in: Proceedings of the CIRM Luminy Conference, World Scientific, Singapore, 1989.
[10] J.-P. Serre, A Course in Arithmetic, Springer-Verlag, Berlin, 1978.
[12] C. Dong, H. Li, G. Mason, Contemp. Math. 193 (1996) 25.
[13] C. Dong, H. Li, G. Mason, Commun. Math. Phys. 214 (2000) 1.
[14] J.H. Conway, R.T. Curtis, S.P. Norton, R.A. Parker, R.A. Wilson, An Atlas of Finite Groups, Clarendon,
Oxford, 1985.
[15] S.P. Norton, Groups, difference sets and the monster, in: Proceedings of a Special Research Quarter at the
Ohio State University, de Gruyter, New York, 1996.
[16] D. Ford, J. MacKay, S.P. Norton, Commun. Algebra 22 (1994) 5175.
[17] S.P. Norton, private communication.
[18] L. Queen, Some Relations Between Finite Groups, Lie Groups and Modular Functions, PhD Dissertation,
University of Cambridge, Cambridge, 1980.
[19] L. Queen, Math. Computation 37 (1981) 547.

New multicritical random matrix ensembles

Romuald A. Janik a,b
a The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen, Denmark
b M. Smoluchowski Institute of Physics, Jagellonian University, Reymonta 4, 30-059 Cracow, Poland
Received 23 January 2002; received in revised form 13 March 2002; accepted 3 May 2002
Abstract
In this paper we construct a class of random matrix ensembles labelled by a real parameter
(0, 1), whose eigenvalue density near zero behaves like |x| . The eigenvalue spacing near
zero scales like 1/N 1/(1+) and thus these ensembles are representatives of a continuous series
of new universality classes. We study these ensembles both in the bulk and on the scale of eigenvalue
spacing. In the former case we obtain formulas for the eigenvalue density, while in the latter case we
obtain approximate expressions for the scaling functions in the microscopic limit using a very simple
approximate method based on the location of zeroes of orthogonal polynomials. 2002 Elsevier
Science B.V. All rights reserved.
1. Introduction
Random matrix ensembles arise in an overwhelming number of diverse applications.
Their utility stems from the fact that even though one can construct a multitude of different
random matrix models, when one studies properties on the scale of eigenvalue spacing
the results become universal and independent of the detailed structure of the random
matrix measure. Hence one can extract predictions even without knowing the details of
the appropriate microscopic model of the phenomenon being studied.
The type of universality class depends in general on the symmetry properties of the
random matrix model and, of course, on the scaling properties of eigenvalue spacing.
Thus one has different bulk and edge universality regimes, as well as a discrete series of
multicritical universality regimes. Consequently only a discrete series of (rational) scaling
exponents could be obtained.
E-mail address: janik@nbi.dk (R.A. Janik).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 6 - 5
R.A. Janik / Nuclear Physics B 635 [PM] (2002) 492504
493
In this paper we would like to show that there exists a real continuous range of
universality regimes with generic real scaling exponents. We will perform an explicit
construction for ensembles with eigenvalue spacing behaving like 1/N 1/(1+) with
(0, 1).
A similar continuous series of universality classes appeared naturally in [1], for random
matrix models with noncompact eigenvalue support and power-law tails in the eigenvalue
density. There the universal behaviour appeared for large eigenvalues. In general, however,
the appropriate measures were difficult to obtain in an explicit way. This was the main
motivation for this investigation.
In this paper we would like to perform a construction of a more conventional class of
models with compact support and where the universal regime is located close to = 0.
We strongly suspect that a mapping of this behaviour to infinity would correspond to
the universal scaling for the Lvy ensembles of [1]. Another possible application of similar
random matrix models (in their chiral variant) might be to model the behaviour of low
lying eigenvalues of the QCD Dirac operator at the chiral phase transition. In this context
the exponent in () || is the inverse of the critical exponent of the chiral phase
transition [24].
The plan of this paper is as follows. In Section 2 we briefly recall the main features of
the old and new universality regimes in random matrix theory. In Section 3 we construct the
random matrix measures of the new multicritical ensembles and compute the eigenvalue
density in the bulk of the spectrum. In Section 4 we move to a discussion of the microscopic
(presumably universal) scaling behaviour and propose a simple approximate method based
on the distribution of zeroes of orthogonal polynomials. In Section 5 we compare these
approximate formulas with exact expressions calculated numerically for = 1/3. We close
the paper with a discussion.
2. Universality regimes
The universality of a correlation function in a random matrix model means that it does
not change (up to a trivial rescaling) when the probability measure is modified. Typically
the universal quantities involve properties of the eigenvalues and their correlations
considered on the scale of eigenvalue spacing in the large N limit. For a multicritical point
(see below), the modifications of the probability measure are restricted to those that do not
destroy multicriticality. However, in general, this still leaves an infinite-dimensional space
of possible deformations.
The type of universal behaviour in random matrix models depends on the part of the
spectrum that one is studying. In general the scaling of eigenvalue spacing with N , in the
vicinity of a fixed eigenvalue 0 , can be easily obtained by looking at the local behaviour
of the (bulk) eigenvalue density (normalized to N ):
N() N( 0 ) .
(1)
Consequently the number of eigenvalues between 0 and is approximately n N(

0 )+1 . Reexpressing in terms of n shows that the eigenvalue spacing in the vicinity of
0 scales like 1/N 1/(1+) .
494
In the bulk the eigenvalue spacing is of the order 1/N this is the classical Wigner
Dyson regime [5]. At the edges of the spectrum the spacing is like 1/N 2/3 and one observes
universal Airy-like oscillations [6,9,11]. When the random matrix possesses an additional
chiral structure as in applications to QCD, i.e.,

0 A
M=
(2)
,
A 0
there appears a different universal regime close to = 0 (with eigenvalue spacing 1/N )
[7].
Apart from these universal regimes, when one finetunes the potential so that ()
2m , a discrete series of multicritical points appear (with scaling 1/N 1/(2m+1) ). The
behaviour on the scale of eigenvalue spacing in this regime is very difficult to extract [8,
10,11]. A different (but also discrete) class of multicritical models appears when adding an
appropriate fixed matrix to the random matrix. This behaviour and novel scaling has been
analyzed both for ordinary [12] and chiral [4] random matrix models.
From the above discussion we see that in order to construct a class of random matrix
models with arbitrary real scaling exponents we have to find a random matrix ensemble
whose bulk eigenvalue spacing behaves like
() || ,
(3)
for real . We will perform the construction for (0, 1) but an extension to other
should not be difficult.
3. Eigenvalue density in the bulk

For a hermitian random matrix model with measure
eN tr V (M) ,
(4)
we consider the following class of potentials

g
t
|M|2p .
V (M) = |M|p +
p
2p
(5)
For hermitian matrices we may define |M|p in the following way. By a unitary transformation M can be rewritten in the form M = U U with = diag(1 , . . . , N ) a diagonal
matrix. Then we may define |M|p U ||p U where ||p = diag(|1 |p , . . . , |N |p ). In
fact we only need its trace which is given by
tr |M|p =
N
|i |p .
(6)
i=1
We note that although random matrix models with potentials of the type (5) are perfectly
well defined, they seem to lack an evident diagrammatic perturbative expansion. Potentials
with a single power-like term were first considered in the mathematical literature [13] and
495
in [14]. The parameter p will be related to by

p = 1 + .
(7)
In the bulk we solve for the eigenvalue density by standard saddle point method. We are
looking for a solution with a single cut (a, a). Later we will choose a = 1. The expression
for the Greens function is then
a
V ()
d
1 2
2
z a
.
G(z) =
(8)
2
z a 2 2
a
We note that for > 0, the derivative exists. The cut endpoint is fixed through the constraint
1
a
0
V ()
d
= 1.
a 2 2
(9)
The eigenvalue density can be extracted from the imaginary part through
1
(z) = 2 a 2 z2 P V
2
a
a
V ()
d
.
z a 2 2
(10)
Because of the nonanalytic power like structure of the potential (5), it is convenient to get
rid of the principal value and rewrite the formula as an ordinary integral. First using the
fact that the potential is even we may use
1
2
1
= 2
z z + z 2
to obtain
1
(z) = 2 a 2 z2 P V
(11)
a
0
d
V ()
.
z2 2 a 2 2
Now we use the fact that

a
d
1
PV
= 0.
z2 2 a 2 2
(12)
(13)
Therefore, we may rewrite (12) as

1
(z) = 2 a 2 z2
a
0
zV (z) V ()
1
,
2
z2 2
a 2
(14)
where we could erase the principal value as the integrand is now nonsingular.
We will now fix the endpoint of the cut to a = 1. Substitution of (5) into (9) yields
( 1+p
( 1+2p
2 )
2 )
t
= 1.
p +g
2p(p)
p( 2 )
(15)
496
Since we want to locate a multicritical point we have to require that (0) = 0. This gives
the second equation for t and g:
1
0
V ()d
= 0,
1 2
(16)
( p1
( 2p1
2 )
2 )
= 0.
+
g
p
(p)
( 2 )
(17)
i.e.,
t
The solution is
t=
g=
2 p(p 12 )( p2 )
1+p
1
(p + 12 )( p1
2 ) 2(p 2 )( 2 )
2 p( p1
2 )(p)
(18)
(19)
1+p
1
(p + 12 )( p1
2 ) 2(p 2 )( 2 )
Bulk eigenvalue density

Once the potential is fixed let us compute explicitly the bulk eigenvalue density
(z) =

1
1 z2 tp (z) + g2p (z) ,
2
(20)
where
1
p (z)
0
zp p
1
.
z2 2 1 2
This integral can be explicitly expressed in terms of hypergeometric functions

( p1
p zp1
p 3p 2
2 )
p (z) =
; z + tan
.
2 F1 1, 1 ,
p
2
2
2
2 1 z2
2( 2 )
(21)
(22)
We see that it is the last term which gives the multicritical behaviour that we wanted to
obtain (z) zp1 = z . Owing to Eq. (16) the constant terms cancel. We could have
chosen of course a different power for the second term in (5), however with this choice the
numerical construction of the relevant orthogonal polynomials is made easier.
Let us note what happens when we reach the point = 1. Then p = 2 and the potential
obtained here is the standard multicritical quartic one V (x) = 4x 2 + 4x 4 . However due to
the vanishing of the coefficient of the last term in (22) the eigenvalue density behaves like
(z) z2 and not z. When we increase the expression for the eigenvalue density ceases
to be positive and thus the construction fails. Presumably a modification of the power in
the second term of the potential might cure the problem but we will not consider that here.
To sum up, the expression (20) is nonnegative when (0, 1) as considered in this paper.
497
4. Behaviour near the originorthogonal polynomials

The ultimate interest in constructing the random matrix models with the new scaling
properties is to extract universal properties which typically occur in the microscopic
regime, i.e., on the scale of eigenvalue spacing. It is well known that all the relevant
properties are encoded in the orthogonal polynomials

dx Pn (x)Pm (x)eNV (x) = nm .
(23)
It is convenient to introduce the wavefunctions

n (x) = Pn (x)eN
V (x)
2
(24)
Then the kernel which allows for the determination of all correlation functions is given by
the expression

N (z)N1 (w) N1 (z)N (w)
(25)
,
RN
zw
where RN is the recursion coefficient entering zn (z) = Rn+1 n+1 (z) + Rn n1 (z).
The microscopic limit is defined through a rescaling
K(z, w) =
x=
y
1
(26)
N 1+
and a limit N with y kept fixed. The key quantity that determines the universal
properties is the rescaled wavefunction

y
lim CN N
(27)
= F (y),
N
N 1/(1+)
(as well as a similar expression with N1 ), CN is a normalization constant chosen in such
a way that the limit exists. In fact the existence of such a limit is a nontrivial and nongeneric
property.
In general the determination of the scaling function F (y) is a very difficult problem
cf. [8,10,11]. One can formulate a differential equation satisfied by the wavefunction
n (x), but taking the scaling limit is extremely difficult and, e.g., for the case of quartic
multicritical ensemble involves data from an auxiliary Painlev equation [8,10]. Moreover
the starting point of such considerations requires a polynomial potential, which is not the
case for our class of models. It would be interesting in the future to explore the possibility
of applying the RiemannHilbert methods of [11].
Here we would like to adopt a different approach and give a very simple but approximate
method of constructing the scaling function F (y). In Section 5 we will compare the
approximate solution with the numerically obtained exact result for = 1/3, and with
the mesoscopic approximation of [8].
498
Zeroes of the orthogonal polynomials

The starting point of our construction is the elementary fact that we may reconstruct the
orthogonal polynomial from the knowledge of its zeroes. Since in any case we normalize
the (even) wave functions through 2n (0) = 1 the appropriate formula is1

n

x2
N V (x)
2
1 2 .
2n (x) = e
(28)
i
i=1
Interestingly enough there is a theorem due to Ismail [15] which states that for an
(almost) arbitrary potential V (x) (see [15]) the zeroes of the nth orthogonal polynomial
i solve the equations

1
NV (i ) 1
An (i )
(29)
=
,
+
log
i k
2
2
Rn
1kn,i =k
i.e., these are electrostatic equilibrium positions in an external field. The additional
assumptions present in [15] serve only to prove the uniqueness of a solution to (29).
The correction term involves the function An (x), given by the formula
An (x)
=N
Rn
V (x) V (y) 2
Pn (y)eNV (y) dy.
x y
(30)
For polynomial potentials of degree m, An (x) is a polynomial of at most degree m 2.

Note that the above equations are exact and valid for any N . We see that the last term in
(29) is suppressed w.r.t. the ordinary potential term.
Now in order to develop our approximation scheme let us neglect the correction factor
(log An (x)) (this is an approximation, the An term may indeed give some contributions in
the scaling limit). From (29) we see that the zeroes are distributed with a continuum density
identical to the eigenvalue density () of the corresponding random matrix model. The
basis of our approximation scheme is the assumption that locally the zeroes are distributed
uniformly w.r.t. () (see the examples below).
Classical WignerDyson scaling functions
Before we consider the case of immediate interest to us, in order to illustrate the method
let us first rederive the formulas for the standard WignerDyson scaling functions. To this
end we assume that the eigenvalue density at = 0 is nonvanishing and N is very large.
For simplicity we take the potential to be even. It is convenient to normalize the continuum
density of zeroes by

cont ()d = N.
1 We always consider even potentials.
(31)
499
In our approximation cont () is simply given by N () where () is the bulk eigenvalue

density of the random matrix model (normalized to 1). If the number of zeroes is even then
they are distributed symmetrically around 0 and there is no zero at = 0. The assumption
of uniform distribution means that the ith zero is located at i where
i
1
cont () d = i + .
2
(32)
Since we are interested only in zeroes which lie close to = 0, in the above expression we
may substitute cont () N(0). Consequently i = (i + 1/2)/(N(0)). Therefore, the
wavefunction is

V (x)
(0)xN 2
(x) = eN 2
(33)
1
.
i + 1/2
i=0
Here we extended the upper limit of the product to infinity. After introducing the scaling
variable y = xN we obtain immediately
F (y) = cos((0)y)
(34)
the exact scaling function in the bulk. The odd case is similar but then there is a zero at
= 0 and consequently the relevant equation is
i
() d = i.
(35)
From the infinite product representation one obtains then the sine function. The kernel (25)
in the scaling limit follows immediately
sin((0)y1) cos((0)y2) cos((0)y1) sin((0)y2)
y1 y2
sin((0)(y1 y2 ))
=
.
(36)
y1 y2
The fact that we obtained here the exact scaling functions does not mean, however, that we
should expect to get exact results in general.
K(y1 , y2 )
Approximate scaling function for (0, 1)

The analysis of the scaling function for the new multicritical ensembles is very similar.
For definiteness we will just consider the even case. We start from the behaviour of the
eigenvalue density near zero:
() = c || ,
(37)
where c can be easily extracted from (20). The positions of the zeroes according to the
uniform distribution approximation are determined by
i
Nc d =
0
1
1
1
c N+1
c yi+1 = i + ,
i
+1
+1
2
(38)
500
where we introduced the rescaled variable y = N 1/(1+) . Hence,

1

1 1+
1+
i+
yi =
.
c
2
Consequently the scaling function normalized by F (0) = 1 is

t y 1+
y2
F (y) = exp
1 2 .
1+ 2
yi
i=0
(39)
(40)
In the next section we will compare this approximate result with the exact orthogonal
polynomials for = 1/3. However, because the convergence properties of the infinite
product are not very good we will truncate it at some nmax (for numerical computation
we use nmax = 200) and approximate the rest of the terms through

N

()
y2
y2
.
(41)
1 2 exp
log 1 2
exp y 2
2
y
y
i
i
i=n +1
ynmax +1
max
The result is

exp

1
y2
c
.
1 yn1+1
(42)
max
As we see when taking the nmax limit this term becomes equal to unity (because
ynmax +1 ). However we include it just because of the slow numerical convergence of
the infinite product.
5. An example = 1/3
The orthogonal polynomials for any potential can be constructed using the determinant
formula:
0 1 2 n
1 2 3 n+1
Pn (x) det 2 3 4 n+2

(43)
xn
x0 x1 x2
where i are the moments

i =
x i eNV (x) dx.
(44)
The advantage of the specific form of (5) is that the moments i can be expressed
analytically in terms of the confluent hypergeometric functions 1 F1 .
We constructed orthogonal polynomials for = 1/3 (p = 4/3) for N = 40, 120 and
N = 240. In Fig. 1 we show the wavefunctions expressed in terms of the scaling variable
501
sc (y) for = 1/3 expressed as a function of the scaling variable y for N = 40, 120
Fig. 1. The wavefunctions N
and N = 240 (thick line).
sc (y) (thick line) with the approximate scaling function F (y).

Fig. 2. Comparison of the wavefunction 240
y = xN 3/4

N
3/4
Nsc (y) N y/N 3/4 = PN y/N 3/4 e 2 V (y/N )
(45)
for the above values of N , normalized by N (0) = 1. We see convergence towards a well
defined scaling function.
In Fig. 2 we compare the exact wavefunction for N = 240 with the approximate scaling
function F (y) obtained in the previous section. There is no free parameter in F (y). The
agreement is indeed surprisingly good. Even the small bump close to y = 0 is correctly
reproduced. In fact it is difficult to judge from the numerical comparison whether the
approximation is exact or not in this case. The small deviations might be caused either by
true corrections which go beyond our approximation or by finite size effects. The reason
why such a simple approximation scheme works so well certainly deserves further study.
Finally let us say a few words about the comparison with the mesoscopic approximation of [8]. An analysis of the differential equations for N (x) for the potentials considered here is still lacking, so we will just take a suitable analytical continuation of the
mesoscopic approximation from the discrete multicritical points considered in [8] to our
case. The result is just

+1
c y
.
Fmeso (y) = cos
(46)
+1
We see that the zeroes of this function on the real axis coincide with the zeroes of our
approximation, but the analytical structure in the complex plane is certainly different.
502
sc (y) (thick line) with the mesoscopic scaling function 1.04F

Fig. 3. Comparison of the wavefunction 240
meso (y).
There are spurious cuts and zeroes coming from y 4/3 . Nevertheless this approximation
is also quite good but the substructure close to y = 0 is not captured by the mesoscopic
approximation (similarly as in the quartic multicritical case considered in [8]). The
comparison is presented in Fig. 3.
6. Discussion
In this paper we constructed a continuous class of random matrix ensembles which
exhibit a new type of universal behaviour close to the zero eigenvalue. The eigenvalue
spacing in this region behaves like 1/N 1/(+1) with a real exponent (0, 1). We
derived the bulk eigenvalue density and studied approximately the behaviour of the scaling
functions in the microscopic limit. Our method for obtaining the scaling function was
approximate and based just on the local behaviour of the bulk eigenvalue distribution.
Nevertheless the approximation seems to be very good and captures even the fine structure
of the scaling functions near the origin.
It would be very interesting to see if these new universality classes could appear in some
physical systems. We conjecture that the scaling behaviour can be directly related to the
large eigenvalue behaviour of Lvy random matrix models [1].
There are numerous open questions and directions for further study. Firstly it would
be very interesting to try to obtain, even implicitly through, e.g., a differential equation,
the exact scaling function, especially as the vast majority of techniques dealing with
multicritical random matrix ensembles requires from the outset a polynomial potential.
Perhaps the most promising approach would be the RiemannHilbert method. On the
technical side a more systematic analysis of the properties of the orthogonal polynomial
zeroes starting from (29), in particular the nature and scaling properties of corrections
would be very welcome. This is especially interesting in order to better understand the
effectiveness and limitations of the approximate method which seems to work so well here.
In general, however, this seems to be quite a formidable problem. Secondly other quantities
of interest, such as eigenvalue spacing distributions, would be important in view of possible
applications. Other possible directions are an extension of the above considerations to
chiral random matrix models (in view of possible applications to QCD), and a rigorous
treatment of universality properties.
503
Note added
An independent study of a very similar class of multicritical random matrix model with
real exponents was undertaken by Gernot Akemann and Graziano Vernizzi [16].
Acknowledgements
I would like to thank Graziano Vernizzi for interesting discussion on the bulk properties,
Poul Damgaard for discussion on [8] and Maciej A. Nowak for discussions and comments.
This work was supported in part by KBN grant 2P03B01917.
References
[1] Z. Burda, R.A. Janik, J. Jurkiewicz, M.A. Nowak, G. Papp, I. Zahed, Free random Levy matrices, Phys. Rev.
E 65 (2002) 021106, cond-mat/0011451.
[2] A.D. Jackson, J.J. Verbaarschot, A random matrix model for chiral symmetry breaking, Phys. Rev. D 53
(1996) 7223, hep-ph/9509324.
[3] R.A. Janik, M.A. Nowak, I. Zahed, Chiral random matrix models: thermodynamics, phase transitions and
universality, Phys. Lett. B 392 (1997) 155, hep-ph/9604253.
[4] R.A. Janik, M.A. Nowak, G. Papp, I. Zahed, Critical scaling at zero virtuality in QCD, Phys. Lett. B 446
(1999) 9, hep-ph/9804244.
[5] E.P. Wigner, Proc. Cambridge Philos. Soc. 47 (1951) 790;
F. Dyson, J. Math. Phys. 13 (1972) 90;
M.L. Mehta, Random Matrices and the Statistical Theory of Energy Levels, Academic Press, New York,
1967.
[6] M.J. Bowick, E. Brzin, Universal scaling of the tail of the density of eigenvalues in random matrix models,
Phys. Lett. B 268 (1991) 21;
P.J. Forrester, The spectrum edge of random matrix ensembles, Nucl. Phys. B 402 (1993) 709;
C.A. Tracy, H. Widom, Level spacing distribution and the Airy kernel, Commun. Math. Phys. 159 (1994)
151.
[7] J.J. Verbaarschot, I. Zahed, Spectral density of the QCD Dirac operator near zero virtuality, Phys. Rev.
Lett. 70 (1993) 3852, hep-th/9303012;
E.V. Shuryak, J.J. Verbaarschot, Random matrix theory and spectral sum rules for the Dirac operator in
QCD, Nucl. Phys. A 560 (1993) 306, hep-th/9212088;
G. Akemann, P.H. Damgaard, U. Magnea, S.M. Nishigaki, Universality of random matrices in the
microscopic limit and the Dirac operator spectrum, Nucl. Phys. B 487 (1997) 721, hep-th/9609174.
[8] G. Akemann, P.H. Damgaard, U. Magnea, S.M. Nishigaki, Multicritical microscopic spectral correlators of
hermitian and complex matrices, Nucl. Phys. B 519 (1998) 682, hep-th/9712006.
[9] P. Bleher, A. Its, Semiclassical asymptotics of orthogonal polynomials, RiemannHilbert problem, and
universality in the matrix model, Ann. of Math. 150 (1999) 185, math-ph/9907025.
[10] P. Bleher, A. Its, Double scaling limit in the random matrix model: the RiemannHilbert approach, mathph/0201003.
[11] P. Deift, T. Kriecherbauer, K.T.-R. McLaughin, S. Venakides, X. Zhou, Commun. Pure Appl. Math. LII
(1999) 1335.
[12] E. Brzin, S. Hikami, Universal singularity at the closure of a gap in a random matrix theory, Phys. Rev.
E 57 (1998) 4140, cond-mat/9804023.
[13] See, e.g., D.S. Lubinsky, E.B. Saff, Strong Asymptotics for Extremal Polynomials Associated with Weights
on R, Lect. Notes in Math., Vol. 1305, Springer-Verlag, Berlin, 1988.
504
[14] L.A. Pastur, On the universality of the level spacing distribution for some ensembles of random matrices,
Lett. Math. Phys. 25 (1992) 259.
[15] M.E.H. Ismail, An electrostatic model for zeroes of general orthogonal polynomials, Pacific J. of Math. 193
(2000) 355.
[16] G. Akemann, G. Vernizzi, New critical matrix models and generalized universality, hep-th/0201165.

Validity of Goldstone theorem at two loops in

noncommutative U (N) linear sigma model
Yi Liao
Institut fr Theoretische Physik, Universitt Leipzig, Augustusplatz 10/11, D-04109 Leipzig, Germany
Abstract
The scalar theory is ultraviolet (UV) quadratically divergent on ordinary spacetime. On
noncommutative (NC) spacetime, this divergence will generally induce pole-like infrared (IR)
singularities in external momenta through the UV/IR mixing. In spontaneous symmetry breaking
theory this would invalidate the Goldstone theorem which is the basis for mass generation when
symmetry is gauged. We examine this issue at two loop level in the U (N) linear model which is
known to be free of such IR singularities in the Goldstone self-energies at one loop. We analyze the
structures in the NC parameter ( ) dependence in two loop integrands of Goldstone self-energies.
We find that their coefficients are effectively once subtracted at the external momentum p = 0 due
to symmetry relations between 1PI and tadpole contributions, leaving a final result proportional to a
quadratic form in p. We then compute the leading IR terms induced by NC to be of order p2 ln( )2
and p2 ln p 2 (p = p ) which are much milder than naively expected without considering the
above cancellation. The Goldstone bosons thus keep massless and the theorem holds true at this
level. However, the limit of 0 cannot be smooth any longer as it is in the one loop Goldstone
self-energies, and this nonsmooth behaviour is not necessarily associated with the IR limit of the
external momentum as we see in the term of p2 ln( )2 . 2002 Elsevier Science B.V. All rights
reserved.
PACS: 11.30.Qc; 02.40.Gh; 11.10.Gh
Keywords: Noncommutative field theory; Spontaneous symmetry breaking; UV/IR mixing
E-mail address: liaoy@itp.uni-leipzig.de (Y. Liao).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 9 3 - 0
506
Y. Liao / Nuclear Physics B 635 [PM] (2002) 505524
1. Introduction
Quantum field theory on noncommutative (NC) spacetime may be formulated in terms
of the MoyalWeyl correspondence [1]. Namely, one still works on commutative spacetime
but replaces the usual product of functions by the star product,

i
,
(f1 f2 )(x) = exp x y f1 (x)f2 (y)
(1)
2
y=x
where x, y are the usual commutative coordinates and is a real, antisymmetric, constant
matrix characterizing the noncommutativity of spacetime, [x , x ] = i . At the classical
action level, while the star product in the bilinear terms may be identified with the usual
one for rapidly decaying functions at spacetime infinity, it does modify the interaction
terms by introducing a phase which in momentum space depends on and the momenta
of fields involved. At the quantum level the phase results in a new feature never seen in
ordinary field theory, the ultravioletinfrared (UV/IR) mixing [2]. The basic mechanism
for this occurrence may be understood as follows. When an otherwise UV divergent
loop integral is multiplied by a phase depending on both the loop momentum k and the
external momentum p, e.g., exp(i/2 k p ), it may become UV convergent due to the
rapid oscillation of the phase in the UV regime. However, this improvement of the UV
convergence is effective only for a nonvanishing external momentum (or more precisely
for a nonvanishing NC momentum p = p ). The hidden singularity from the UV loop
momentum will reappear as a new form when the external momentum goes to the IR limit.
Depending on the degree of divergence of the loop integral, this NC IR singularity may be
pole-like, logarithmic, etc.
The above NC IR singularity, especially the pole-like one, may cause serious problems
in NC field theory. It leads to a drastic modification to dispersion relation at low energy
in perturbation theory which may make the theory not well-defined in the IR. When going
beyond one loop level it may destroy or at least make unclear the renormalizability of
the theory. Indeed, most of explicit model analyses made so far are restricted to the
one loop level and their renormalization is considered for nonexceptional NC external
momenta [38]. For exceptional ones we would have to choose a different subtraction
scheme. The real 4 theory has been examined at two loops [9], but again the main
concern is with the UV regime of loop momenta for nonexceptional NC momenta.
For complex 4 theory with spontaneous symmetry breaking the IR behaviour becomes
important as it is related to the issue of whether the Goldstone theorem still holds true
on NC spacetime, namely, whether the masslessness of Goldstone bosons is stable against
radiative corrections. This is a starting point to all attempts of realistic model building
including weak interactions [10]. The complex U (N) model has been studied in this
context at one loop in Refs. [5,6], and it was found that there are no NC IR singularities at
all in the self-energy of Goldstone bosons so that their masslessness is guaranteed at this
order: both pole-like and logarithmic ones are cancelled in the mass correction due to the
delicate relations governed by the spontaneously broken symmetries as occurring in the
commutative theory. This is a surprising result since the scalar theory is UV quadratically
divergent. It would be highly desirable to investigate whether this is a special feature at one
loop, or more importantly whether NC IR singularities at higher orders, if any, endanger the
507
masslessness of Goldstone bosons making the theorem no longer valid on NC spacetime.

Naively speaking, this should not be surprising if it occurs. Beyond one loop, the wouldbe NC IR singularities for external momenta at one loop now appear as an internal part
of higher loops; it is not clear whether they persist to be cancelled. Even worsely, they
may combine with remaining massless Goldstone bosons to enhance the IR behaviour
in external momenta. It is the purpose of the present work to clarify these problems by
an explicit two loop analysis. Our main results may be summarized as follows. At two
loop level, there are no IR terms more singular than p2 ln p 2 , and individual stronger
singularities at intermediate steps are finally cancelled due to symmetry relations. The
Goldstone bosons thus keep massless and the Goldstone theorem holds valid at this order
in perturbation theory. We also point out the difference between the NC singularity in the
limit of 0 and the NC IR singularity in the limit of p 0. For the self-energies of
2 and p2 ln p 2 . While the
Goldstone bosons obtained we have the NC behaviour of p2 ln
latter is leading in the IR limit, both are singular in : we have NC singularities at higher
orders independently of external momentum configurations.
In the next section we describe the NC U (N) linear model, whose Feynman rules
are reproduced in Appendix A. Then we present a detailed two loop analysis in Section 3.
Some examples of two loop integrals involving are shown in Appendix B. We conclude
in the last section.
2. The model
We follow the same conventions as in Ref. [5] in describing the NC U (N) linear
model. The complex scalar is in the fundamental representation of U (N) with
L = ( ) + 2 .
(2)
The spontaneous symmetry breaking is triggered by the nonvanishing scalar VEV,

assuming 2 , > 0,
= + 0 ,
T

= 1 , . . . , N1 , ( + i0 )/ 2 ,
(3)
0 = (0, . . . , 0, v/ 2 )T ,

with v = 2 /. The field is the Higgs boson with mass m = 2v 2 and the 0 and
i (i = 1, . . . , N 1) fields are the real and complex Goldstone bosons. We have ignored
other possible orderings of interaction like i j i j which are problematic already
at one loop [5,6]. In terms of the shifted fields, we have
1
1
1
1
L = ( )2 m2 2 + ( 0 )2 + i i
2
2
2
2

v 2 + 02 + 2i i

1 4
+ 04 + i i j j
4

1
2 02 0 0 + 2 + 02 i i
2
508
i[, 0 ]j j ,
(4)
where we have suppressed the star notation and dropped terms which vanish upon
integration over spacetime. The perturbation theory is based on the above Lagrangian. The
one loop calculation has been done in Refs. [5,6]. We now proceed to consider two loop
contributions in the next section.
3. Two loop contributions

There are three sets of contributions at two loop level: bare two loop diagrams, one loop
diagrams with one insertion of counterterms determined at one loop, and the counterterms
determined at two loops. It is clear that the third causes no IR problem. We start with the
second which is just a one loop calculation.
3.1. One loop diagrams with one insertion of one loop counter-terms
The contributing diagrams are shown in Figs. 1 and 2 where the solid and dashed
lines are for and 0,i fields, respectively. As we are explicitly including the tadpole
Fig. 1. 1PI contributions to 0 or j self-energy.
Fig. 2. Tadpole contributions to 0 or j self-energy.
509
contributions, we shall not impose the requirement of tadpole cancellation, nor introduce a
counterterm for the VEV. The counterterms for the self-energies are, respectively,

0 , j : i p2 Z m2 ,

: i p2 m2 Z m2 .
(5)
The vertex counterterms are obtained simply by attaching a factor of Z to their Feynman
rules. The quantities Z, , , are renormalization constants whose details may be
found, e.g., in Refs. [8]. For our purpose here, it is sufficient to know that
= Z Z ,
(6)
which arises due to their different mass and renormalization.

Let us first consider the part proportional to Z . We have,

(1a) + (1b) + (1c) + (2a) + (2b) Z = (one loop result) (Z ),
(7)
which is free of NC IR singularities according to Refs. [5,6]. The remaining Z

dependence will be given below together with that of Z . Next, consider the part
proportional to Z . We have similarly,

(1e) + (1f ) + (1g) + (2c) + (2d) = (one loop result) Z ,
(8)
which again is safe. The remaining Z and Z dependence is,

(1d)Z + (1h) = m2 (Z Z )
2ij [J (0, m) + ],
for i j ,
(9)
[J (0, m) + J,p (0, m) + ], for 0 0 ,
where the dots stand for the terms which are both UV (loop momentum) and IR (external
momentum) finite. Following Ref. [5] we have introduced similar notations for integrals,

1
d 4k
d 4 k cos(2k p)
J (0) =
,
J
(0)
=
,
,p
(2)4 (k 2 )2
(2)4
(k 2 )2

d 4k
1
d 4 k cos(2k p)
J (m) =
,
J
(m)
=
,
,p
(2)4 (k 2 m2 )2
(2)4 (k 2 m2 )2

d 4k
1
d 4 k cos(2k p)
J (0, m) =
(0,
m)
=
,
J
. (10)
,p
4
2
2
2
(2) k (k m )
(2)4 k 2 (k 2 m2 )
Our manipulations will be independent of schemes used to regularize divergences in the
above integrals. Now we compute the , terms and obtain,

ij 2J (0),
(1a) + (2a) = +m2
[J (0) + J,p (0)],

ij 2J (m),
(1b) + (2b) = m2
[J (m) + J,p (m)],
ij 2[J (0, m) J (0)] + ,

(1c) = +m2
[J (0, m) J (0) + J,p (0, m) J,p (0)] + ,
ij 2[J (0, m) J (m)] + ,

(1d) = m2
(11)
[J (0, m) J (m) + J,p (0, m) J,p (m)] + .
510
Using Eq. (6), the IR singularities are cancelled in the sum of Eqs. (9) and (11) leaving
behind an IR safe result proportional to p2 .
3.2. Two loop diagrams
Now we calculate the genuine two loop contributions. The 1PI diagrams are depicted
in Figs. 3 and 4 where we only show topologically different graphs with the solid line
representing all scalar fields. The number appearing as a subscript refers to the number of
diagrams actually involved.
The 1PI tadpole is found to be,

iT 1PI = i2 v
d 4 k1
(2)4
d 4 k2
T (ki ),
(2)4
T (ki ) = Ta + Tb + Tc ,
Fig. 3. Two loop 1PI contributions to tadpole.
Fig. 4. Two loop 1PI contributions to 0 or j self-energy.
(12)
511
where, using the notations Dm (q) = (q 2 m2 )1 , D(q) = (q 2 )1 and K12 = cos(2k1

k2 ), we have,
2

Ta = + 3Dm
(k1 )Dm (k2 ) + D 2 (k1 )D(k2 ) (2 + K12 )
2

+ 3Dm
(k1 )D(k2 ) + D 2 (k1 )Dm (k2 ) (2 K12 )
2

+ (N 1) 6Dm
(k1 )D(k2 ) + 4D 2 (k1 )D(k2 ) + 2D 2 (k1 )Dm (k2 )
(13)
+ 4(N 1)ND 2 (k1 )D(k2 ),

2
2
Tb /m2 = + 27/2Dm
(k1 )Dm (k2 )Dm (k1 + k2 ) + 3/2Dm
(k1 )D(k2 )D(k1 + k2 )

+ D 2 (k1 )D(k2 )Dm (k1 + k2 ) (1 + K12 )
2

(14)
+ (N 1) 6Dm
(k1 )D(k1 + k2 ) + 4D 2 (k1 )Dm (k1 + k2 ) D(k2 ),

Tc = + 3Dm (k1 )Dm (k2 ) + D(k1 )D(k2 ) Dm (k1 + k2 )(1 + K12 )
+ 4(N 1)D(k1 )D(k2 )Dm (k1 + k2 ).
(15)
Note that we can have dependence beyond one loop even if the external momentum
vanishes since there are independent loop momenta which can combine with the
antisymmetric . The result will depend on it through 2 = , etc. As long as
we do not use the Lorentz covariance to choose a specific frame for external momenta, we
can always treat in integrals as if it were a Lorentz tensor.
Upon choosing loop momenta properly in some integrals, the 1PI self-energy of the
charged Goldstone bosons i+ j is found to be,

iij1PI (p) = i2 ij
U (ki , p) =
d 4 k1
(2)4
d 4 k2
U (ki , p),
(2)4
Ux ,
(16)
x=a
2

Ua = + Dm
(k1 )Dm (k2 ) + D 2 (k1 )D(k2 ) (2 + K12 )
2

+ Dm
(k1 )D(k2 ) + D 2 (k1 )Dm (k2 ) (2 K12 )
2

+ 2(N 1) Dm
(k1 ) + D 2 (k1 ) D(k2 )

+ 2N D 2 (k1 )Dm (k2 ) + D 2 (k1 )D(k2 )
(17)
+ 4N 2 D 2 (k1 )D(k2 ),

2
2
Ub /m2 = + 9/2Dm
(k1 )Dm (k2 )Dm (k1 + k2 ) + 1/2Dm
(k1 )D(k2 )D(k1 + k2 )

2
+ D (k1 )Dm (k2 )D(k1 + k2 ) (1 + K12 )
2
+ 2(N 1)Dm
(k1 )D(k2 )D(k1 + k2 )
+ 4ND 2 (k1 )Dm (k2 )D(k1 + k2 ),
(18)
512

2
(k1 )D(k1 + p) 9Dm (k2 )Dm (k1 + k2 )
Uc /m4 = + Dm

+ D(k2 )D(k1 + k2 ) (1 + K12 )
+ 4D 2 (k1 )Dm (k1 + p)Dm (k2 )D(k1 + k2 )
Ud /m =
2
2
+ 4(N 1)Dm
(k1 )D(k1 + p)D(k2 )D(k1 + k2 ),
(19)
2
+ 2Dm
(k1 )D(k1 + p)Dm (k2 )(2 + K12 )
2
+ 2Dm (k1 )D(k1 + p)D(k2 )(2 K12 )

+ 2D 2 (k1 )Dm (k1 + p) Dm (k2 ) + D(k2 )
2
(k1 )D(k1 + p)D(k2 )
+ 4(N 1)Dm
2
+ 4ND (k1 )Dm (k1 + p)D(k2 ),
(20)
Ue /m = +2Dm (k1 )D(k1 + p)Dm (k2 )D(k2 + p)(1 + K12 ),

Uf = + Dm (k1 )Dm (k2 ) + D(k1 )D(k2 ) D(k1 + k2 p)(1 + K12 )
2
+ 2Dm (k1 )D(k2 )D(k1 + k2 p)(1 K12 )

+ 4 (N 1)D(k1 )D(k2 )

+ D(k1 + p)D(k2 + p)K12 D(k1 + k2 + p),
(21)
(22)
Ug /m4 = + 6Dm (k1 )D(k1 + p)Dm (k2 )D(k2 + p)Dm (k1 k2 )(1 + K12 )
+ 4Dm (k1 )D(k1 + p)D(k2 + p)Dm (k2 )D(k1 + k2 + p)K12 ,

Uh /m = + Dm (k1 )D(k1 + p) 6Dm (k2 )Dm (k1 + k2 ) + 2D(k2 )D(k1 + k2 )

+ 4Dm (k2 )D(k1 + k2 + p) (1 + K12 )

+ 8Dm (k1 )D(k1 + p) (N 1)D(k2 )D(k1 + k2 )

+ D(k2 p)D(k1 + k2 p)K12 .
(23)
(24)
Although the 1PI self-energy of the neutral Goldstone boson 0 has the same set
of diagrams as the charged one, it becomes more complicated due to multiplications of
trigonometric functions involving the loop and external momenta and . To simplify our
analysis of the NC IR behaviour, it is useful to cast these products into standard forms. For
the self-energy at two loops, we have three independent momenta in the integrand, k1 , k2
and p so that we can form two independent combinations with , 2k1 k2 and 2k1 p.
(2k2 p is not independent as it can be obtained from 2k1 p by k1 k2 .) We find that
it is always possible by shifting and interchanging loop momenta properly so that the only
dependence in integrands enters through either the above K12 or K1 = cos(2k1 p). For
example, the simple-looking Fig. 4(e) in this case involves the following product,
cos(p k1 ) cos(p k2 )

2 cos(k1 k2 ) cos[(k1 + p) (k2 + p)] cos[(k1 + k2 ) p]

= 1/4 cos(2k1 k2 ) + cos[2k1 k2 + 2(k1 k2 ) p]
+ cos[2(k1 + p) k2 ] + cos[2(k2 + p) k1 ]

+ cos[2(k1 k2 ) p] cos[2(k1 + k2 ) p] ,
(25)
where the five nonstandard forms may be transformed into the standard ones, e.g.,

cos 2k1 k2 + 2(k1 k2 ) p K12 , with ki ki p,

cos 2(k1 + k2 ) p K1 , with k1 k1 k2 .
513
(26)
After this manipulation, the expression for the 1PI 0 self-energy becomes very lengthy
though it has a simpler structure in ,

d 4 k1
d 4 k2
1PI
2
V (ki , p),
i00 (p) = i
(2)4
(2)4
V (ki , p) =
Vx ,
x=a

2
Va = + Dm
(k1 ) Dm (k2 )2(2 K1 + K12 ) Dm (k2 + p)K12

2
+ Dm
(k1 ) D(k2 )2(2 K1 K12 ) + D(k2 + p)K12

+ D 2 (k1 ) Dm (k2 )2(2 + K1 K12 ) Dm (k2 + p)K12

+ D 2 (k1 ) D(k2 )2(2 + K1 + K12 ) + D(k2 + p)K12
2

+ 2(N 1) Dm
(k1 )(2 K1 ) + D 2 (k1 )(2 + K1 ) D(k2 )

+ D 2 (k1 ) Dm (k2 ) + D(k2 )
+ 4(N 1)ND 2 (k1 )D(k2 ),

2
Vb /m2 = + 9/2Dm
(k1 ) Dm (k2 )Dm (k1 + k2 )(2 K1 + 2K12 )
Dm (k2 + p)Dm (k1 + k2 + p)K12
(27)
(28)
2
+ 1/2Dm
(k1 )[same as above except m 0]

2
+ D (k1 ) Dm (k2 )D(k1 + k2 )(2 + K1 + 2K12 )

+ Dm (k2 + p)D(k1 + k2 + p)K12
2
+ 2(N 1) Dm
(k1 )D(k2 )(2 K1 )

2
+ 2D (k1 )Dm (k2 ) D(k1 + k2 ),

2
Vc /m4 = + 9/4Dm
(k1 )D(k1 + p) Dm (k2 )Dm (k1 + k2 )2(1 + K1 + K12 )

+ Dm (k2 + p)Dm (k1 + k2 + p) + (p p) K12
(29)
+ 1/2D 2 (k1 )Dm (k1 + p){same as above except m 0 in each 2nd D}

2
+ 1/4Dm
(k1 )D(k1 + p){same as above except all m 0}
2
(30)
+ 2(N 1)Dm
(k1 )D(k1 + p)D(k2 )D(k1 + k2 )(1 + K1 ),

2
2
Vd /m = + Dm (k1 )D(k1 + p) Dm (k2 )(2 + 2K1 + K12 ) + Dm (k2 + p)K12

+ D(k2 )(2 + 2K1 K12 ) D(k2 + p)K12

+ D 2 (k1 )Dm (k1 + p) Dm (k2 )(2 + 2K1 K12 ) Dm (k2 + p)K12

+ D(k2 )(2 + 2K1 + K12 ) + D(k2 + p)K12 + 2(N 1)
2

(31)
Dm
(k1 )D(k1 + p) + D 2 (k1 )Dm (k1 + p) D(k2 )(1 + K1 ),
514

Ve /m2 = + D(k1 + p)Dm (k1 ) + D(k1 )Dm (k1 + p)

D(k2 + p)Dm (k2 ) + D(k2 )Dm (k2 + p) K12 + Dm (k2 )

D(k2 + p) D(k2 p) Dm (k1 + k2 )D(k1 + k2 + p)K1 ,

Vf = + Dm (k1 + p)Dm (k2 )D(k1 + k2 )(3 4K12 )

2 Dm (k1 )Dm (k2 + p) + Dm (k1 + p)Dm (k2 ) D(k1 + k2 )K1
(32)
+ 2D(k1 )Dm (k2 )Dm (k1 + k2 p)K1

+ 2Dm (k1 )Dm (k2 ) + 2Dm (k1 p)Dm (k2 + p)

+ Dm (k1 + p)Dm (k2 + p) D(k1 + k2 + p)K12
+ 1/3{same as above except + in first two lines and all m 0}

+ 2(N 1)D(k2 ) D(k1 + p)D(k1 + k2 ) + D(k1 )D(k1 + k2 p)K1

+ 2(N 1)D(k2 ) D(k1 + p)Dm (k1 + k2 )

Dm (k1 )D(k1 + k2 p)K1 ,
(33)

4
Vg /m = + 3/2 Dm (k1 )D(k1 + p)Dm (k2 )D(k2 + p)Dm (k1 k2 )(1 + K1 + K12 )

+ Dm (k1 + p)D(k1 )Dm (k2 )D(k2 + p)Dm (k1 + k2 + p)
+ Dm (k1 )D(k1 + p)Dm (k2 + p)D(k2 )Dm (k1 + k2 + p)

+ Dm (k1 p)D(k1 )Dm (k2 p)D(k2 )Dm (k1 k2 ) K12

+ Dm (k1 )D(k1 + p)Dm (k2 )Dm (k1 k2 )

+ Dm (k1 + k2 )D(k1 + k2 + p)Dm (k2 )Dm (k1 ) D(k2 + p)K1
+ 1/2{same as above but interchanging masses in
(34)
3rd and 4th D and m 0 in 5th D},

Vh /m = + 3 Dm (k1 )D(k1 + p)Dm (k2 )Dm (k1 + k2 )2(1 + K1 + K12 )

+ Dm (k1 )D(k1 + p) Dm (k2 + p)Dm (k1 + k2 + p) + (p p) K12

Dm (k2 )D(k2 + p)Dm (k1 )Dm (k1 + k2 ) + (k2 k1 k2 ) K1

Dm (k1 p)D(k1 ) Dm (k2 )Dm (k1 + k2 p) + (k2 k2 + p) K12
2
+ {same as above except m 0 in 3rd and 4th D

and + in last two lines}

+ 2 D(k1 )Dm (k1 + p) Dm (k2 p)D(k1 + k2 p) (p p) K12

+ D(k1 p)Dm (k1 ) Dm (k2 )D(k1 + k2 p) + (k2 k2 + p) K12

+ Dm (k2 )D(k1 )Dm (k1 + k2 + p)

+ Dm (k1 )D(k2 )Dm (k2 + p) D(k1 + k2 )K1
+ 4(N 1)Dm (k1 )D(k1 + p)D(k2 )D(k1 + k2 )(1 + K1 ).
(35)
Including the tadpole contributions, we have the self-energies for Goldstone bosons,

d 4 k1
d 4 k2
iij (p) = i2 ij
U (ki , p) T (ki ) ,
4
4
(2)
(2)

i00 (p) = i2
d 4 k1
(2)4

d 4 k2
,
p)
T
(k
)
,
V
(k
i
i
(2)4
515
(36)
which have the following structure in ,

d 4 k1
d 4 k2
2
iij (p) = i ij
f0 (ki , p) + f12 (ki , p)K12 ,
4
4
(2)
(2)

4k d 4k

d
1
2
i00 (p) = i2
g0 (ki , p) + g1 (ki , p)K1 + g12 (ki , p)K12 , (37)
4
4
(2)
(2)
where all dependence resides in K factors. A crucial observation from the above explicit
expressions is that all form factors f and g vanish at p = 0 so that we are effectively
subtracting at p = 0 for each form factor when doing integrations,

d 4 k1
d 4 k2
iij (p) = i2 ij
f
(k
,
p)
+
f
(k
,
p)K
,
0
i
12
i
12
(2)4
(2)4

d 4 k1
d 4 k2
2
i00 (p) = i
g0 (ki , p) + g1 (ki , p)K1 + g12 (ki , p)K12 , (38)
(2)4
(2)4
where f0 (ki , p) = f0 (ki , p) f0 (ki , 0), g0 (ki , p) = g0 (ki , p) g0 (ki , 0), etc. This
cancellation in the independent and K12 structures between the 1PI and tadpole
contributions originates from symmetry relations among vertices summarized in the Ward
identities. The vanishing of g1 (ki , 0) is a feature of the 1PI part alone which fits in the
requirement of the Goldstone theorem to be verified here. All of this also serves as a good
test of the correctness of the calculation.
Now we proceed to consider the NC IR behaviour in the total self-energies. The f0
and g0 terms are independent of , proportional to p2 as in the commutative theory and
thus IR safe. The dependent terms are proportional to some factors quadratic in p due to
the subtraction, but in principle not necessarily to p2 . With the constant antisymmetric
tensor , we may construct symmetric and antisymmetric ones by contraction, e.g.,
2 = , which may be used to build new NC momenta like p
= 2 p . Thus the
proportionality factors can be p2 , p p = p 2 , etc. The task here is to show that negative
powers of these scalars never appear in the final results so that the IR safety is guaranteed
in the self-energies. We shall show that the above quadratic p factors will be multiplied
by a leading factor of order ln 2 or ln p 2 . (The p2 ln p2 behaviour is not new as it already
appears in commutative theory.) Thus the self-energies are IR safe but cannot go to the
commutative limit smoothly.
Generally speaking, an integral with K1 or K12 will be IR safe in the external
momentum if it is already convergent both superficially and in subgraphs without these
factors. For divergent integrals either superficially or in subgraphs, we must be careful. Let
us first consider the K12 term. We shall demonstrate our calculation by some typical terms
in f12 (ki , p). The case of g12 is similar but much more complicated due to the momentum
shifts which introduce p in many propagators. Using A1 (p)A2 (p) = A1 (p)A2 (p) +
A1 (p)A2 (0), etc., we can always subtract sequentially, so that the only complication in
g12 lies in the polynomial p dependence of Feynman parameter integrals. Concerning f12 ,
the contributions from Figs. 4(a) and (b) are cancelled by those of the tadpole. Fig. 4(g) is
516
finite superficially and in subgraphs without K12 and is thus safe. The most dangerous is
Fig. 4(d) which has a quadratically divergent subgraph, which in turn may transmute into
a pole-like singularity in the external momentum. Fortunately, this quadratic divergence is
cancelled between the and 0 contributions proportional to,

d 4 k1
(2)4

d 4 k2 2
D (k1 ) Dm (k2 ) D(k2 ) D(k1 + p)K12 .
(2)4 m
(39)
When computing loop integrals, we always work on Euclidean spacetime to simplify their
analytic property. (Now Dm (k) = (k 2 + m2 )1 .) Finishing the k2 integral using I1 in
Appendix B, we have,
integral = 2m2 (4)2
1
dx

d 4 k1 2
2 k 2 .
D
(k
)D(k
+
p)K
xm
1
1
0
m
1
(2)4
(40)
There are preferred directions defined by p and . To avoid complicated angular

integration, we have to make some simplifying assumptions which should not alter the
2 is symmetric and semi-positive definite on Euclidean
IR singularity drastically. As
space, it may be diagonalized by an orthogonal rotation. We assume that it has a four-fold
2 = 2
degenerate eigenvalue of 2 , i.e.,
with > 0 a small area scale characterizing
2
2
2
NC. Then, k1 = k1 and the angular integral is much simplified,

d4 (k1 + p)

= 4

1
d sin2 k12 + p2 + 2 k12 p2 cos
= 2
1/k12
(k12 p2 ),
1/p2
(k12 p2 ).
(41)
Including the subtraction D(k1 ) term, we are thus integrating over k12 [0, p2 ]. Note that
this seems to be a special feature of angular integrals in four dimensions. For p2 m2 , we
have for the dominant p dependent part,
integral m
(4)
(4)
dk12 k12
dx
0

2 2 2
1
1
ln
xm k1
p2 k12

p ln 2 m2 p2 ,
2 2
(42)
which indeed vanishes as p 0 but singular as 0. Integrals in Fig. 4(e) can be

similarly computed to arrive at the same conclusion.
Let us now consider the case when the two loop momenta are overlapping. For example,
Fig. 4(f) contains the following integral,

d 4 k1
(2)4
d 4 k2
D(k1 + k2 )Dm (k2 )Dm (k1 + p)K12 .
(2)4
(43)
First finish k2 integral using I2 in Appendix B,

d 4 k1
integral =
Dm (k1 + p)
(2)4
1

2
2(4)
dx K0 (1 x) k12 x + m2 k12 ,
517
(44)
where the cosine factor disappears due to k1 k1 = 0. We are interested in the small p limit,

Dm (k1 + p) = Dm (k1 ) 2k1 pDm (k1 )

2
p2 Dm (k1 ) + 4(k1 p)2 Dm
(45)
(k1 ) + O p3 .
The first term vanishes since we cannot make up a vector out of and . It also
vanishes by the k1 k1 symmetry. The third term is an integral involving k1 k1 which
2 ,
may be simplified as follows. The result of the integral is composed of the , ,
3
, . . . terms, where odd products of actually cannot appear due to symmetry. As we
are also interested in the small limit, the 2 and higher terms may be dropped; namely,
2 = 2 , the above argument becomes exact
we may replace k1 k1 by k12 /4. For
since only one structure is possible. We continue to work in this assumption so that
the angular integration may be finished explicitly. Together with the second term, we have
3 (k ). To help identify the relevant integration region, we first
Dm (k1 + p) m2 p2 Dm
1
2
rescale k1 = y so that a small will not interfere with a large k12 in K0 . We have,
integral 2(4)
4 2

dy y(y + 3)
p 3
0
1
dx K0

(1 x)(xy + 3)y ,
(46)
with 3 =
As K0 decays exponentially at a large argument and explodes at a small
one, only the latter region is important. After some calculation, we obtain the following
leading term,

integral 21 (4)4 p2 ln m4 2 ,
(47)
m2 .
which is IR finite but singular as 0. Fig. 4(b) in the neutral Goldstone case is
convergent without the K12 factor after it is once subtracted at p = 0, and thus safe in
the IR limit. Figs. 4(c), (h) are similarly done. More examples of integrals, especially those
appearing in Figs. 4(a) with many massless propagators that may cause an IR problem in
the virtual loop momentum, are given in Appendix B.
The neutral Goldstone boson self-energy has an additional contribution containing K1 .
Since K1 involves only one of the loop momenta k1 , there is a big difference between
overlapping and nonoverlapping integrals. For the latter, it is easy to identify the leading
singular terms since they are essentially a product of two one-loop integrals. Fig. 4(d)
belongs to this category. For example, consider the integral,

d 4 k2
d 4 k1
2
Dm (k2 )Dm
(k1 )D(k1 + p)K1 .
(48)
4
(2)
(2)4
518
It is quadratically divergent in k2 as in commutative theory, which is not harmful at all to the

Goldstone theorem. It is also regular and vanishes in the NC IR limit. For the overlapping
case, let us begin with an integral appearing in Fig. 4(c),

d 4 k2 2
d 4 k1
D (k1 )Dm (k2 )Dm (k1 + k2 )D(k1 + p)K1 .
(49)
(2)4
(2)4 m
Integration over k2 gives a usual logarithmic UV divergence plus a ln k12 term. However,
the resulting k1 integral is again regular and vanishes in the limit of p 0. Fig. 4(e) is
proportional to the following integral,

d 4 k2
d 4 k1
Dm (k1 + k2 )Dm (k2 )
(2)4
(2)4

D(k1 + k2 + p) D(k1 + k2 p) D(k2 + p)K1

d 4 k1
d 4 k2
=
Dm (k1 + k2 )Dm (k2 )
(2)4
(2)4

D(k2 p) D(k2 + p) D(k1 + k2 p)K1 .
(50)
It becomes, using I2 in Appendix B,
integral = 2(4)

1
dx K0

(1 x) m2 + p2 x p 2

d 4 k2
(51)
Dm (k2 ) D(k2 p) D(k2 + p) cos(k2 p),
4
(2)
which vanishes by k2 k2 .
Now consider overlapping integrals arising in Figs. 4(f), (h) which are nontrivial in the
NC IR limit. For example, Fig. 4(f) contains the following one,

d 4 k2
d 4 k1
D(k1 )Dm (k1 + k2 )Dm (k2 + p)K1
4
(2)
(2)4

d 4 k2
2
= 2(4)
Dm (k2 + p)
(2)4
1

dx K0 x m2 + (1 x)k22 p 2 cos(xk2 p).
(52)
Note that only the small k2 region is important for K0 where the cosine factor may be
ignored for the leading term. A similar calculation to Eq. (44) leads to,

integral 21 (4)4 p2 ln m2 p 2 .
(53)
And for the purely massless case, we obtain,

d 4 k1
d 4 k2
D(k1 )D(k1 + k2 )D(k2 + p)K1
(2)4
(2)4

21 (4)4 p2 ln p2 p 2 .
(54)
519
Integrals from Fig. 4(h) can be similarly computed whose results are presented in
Appendix B. Figs. 4(a), (b) have no contributions containing K1 after subtraction while
Fig. 4(g) is regular in the limit of p 0.
4. Conclusion
The scalar theory is UV quadratically divergent whether or not its symmetry is
spontaneously broken. On NC spacetime this virtual UV quadratic divergence may
transmute into a pole-like IR singularity in external momenta. If this occurrence persists
in scalar theory with spontaneous symmetry breaking, it may spoil the validity of the
Goldstone theorem which is utilized to generate mass through the Higgs mechanism when
the symmetry is gauged. On naive grounds there is no reason why this should not happen.
This is especially the case when one goes beyond one loop level where the richer structure
in the NC parameter may produce NC IR singularities not appearing at one loop [5,6]
and the singularities may even be enhanced by virtual massless Goldstone bosons in the
extra loop.
We have made a complete analysis of the above problem at two loop level by studying
the self-energies of the Goldstone bosons in the NC U (N) linear model. We found that
the integrands in loop integrals have three types of dependence, i.e., independent,
involving the two loop momenta (K12 = cos(2k1 k2 )), and involving one loop and
one external momentum (K1 = cos(2k1 p)). Our crucial observation is that the form
factors of the above structures vanish in the limit of p 0. This implies that they are
effectively once subtracted at p = 0. The subtraction arising from symmetry relations
in 1PI and tadpole contributions cancels the most singular terms that are harmful to the
theorem leaving behind a result proportional to a quadratic form in p. We have computed
in detail the leading terms in the coefficients of the above form. We observed that delicate
cancellation also occurs between the Higgs and Goldstone bosons that prevents harmful
terms in the coefficients. The masslessness of virtual Goldstone bosons is not a problem;
its IR behaviour can always be separated from the one induced by NC. The final leading IR
terms in the Goldstone self-energies induced by NC are of order p2 ln 2 and p2 ln p 2 so
that the Goldstone theorem still holds true at two loop level. Since the basic mechanism for
this mild IR behaviour originates from symmetry relations amongst vertices of the Higgs
and Goldstone bosons, it seems rather natural to expect that the theorem should also be
valid beyond two loop level. On the other hand, the limit of 0 cannot be smooth at
two loops and beyond, and this nonsmooth behaviour in is not necessarily associated
with the IR limit of the external momentum as we saw in the leading term p2 ln 2 from
the K12 part.
Acknowledgements
I would like to thank K. Sibold for many helpful discussions and for reading the
manuscript carefully.
520
Appendix A. Feynman rules

For completeness, we list below the Feynman rules for the vertices in the noncommutative U (N) linear model with the scalar field in the fundamental representation, which
were first given in Ref. [5]. All momenta are incoming and shown in the parentheses of the
corresponding particles. There are no changes in propagators
(p1 ) (p2 ) = i6v c12 ,
0 (p1 )0 (p2 ) = i2v c12 ,
i+ (p1 )j (p2 ) = i2vij e12 ,
(p1 ) (p2 ) (p3 ) (p4 ) = i2(c12c34 + c31 c24 + c23 c14 ),
0 (p1 )0 (p2 )0 (p3 )0 (p4 ) = i2(c12c34 + c31 c24 + c23 c14 ),
(p1 ) (p2 )0 (p3 )0 (p4 ) = i2(2c12c34 c13,24),
(p1 ) (p2 )i+ (p3 )j (p4 ) = i2ij c12 e34 ,
0 (p1 )0 (p2 )i+ (p3 )j (p4 ) = i2ij c12 e34 ,
(p1 )0 (p2 )i+ (p3 )j (p4 ) = i2ij s12 e34 ,
i+ (p1 )j+ (p2 )k (p3 )l (p4 ) = i2[ik j l e13 e24 + il j k e14 e23 ],
(A.1)
where the following notations are used: p q = p q /2, cij = cos(pi pj ), sij =
sin(pi pj ), cij,kl = cos(pi pj + pk pl ), sij,kl = sin(pi pj + pk pl ) and
eij = exp(ipi pj ).
Appendix B. Some examples of two loop integrals involving

We work on Euclidean spacetime where the integrals have a simpler analytic property.
We start with the one loop integrals that have been computed by many authors in the
literature

2
d nk 2
I1 2 , m = 4n
(B.1)
cos(k ),
k + m2
(2)n
where will be identified later with q with q a loop or external momentum. Using
the Schwinger parameter integral to exponentiate the denominator, completing the square
and shifting k, we have

I1 =
4n

d nk
exp k 2 + m2 + ik
n
(2)

=
0

2 4n
d nk
d exp m2
exp k 2
n
4
(2)
521

4n
2
2
d exp m
4 (4)n/2
= 2(4)
2n/2

42 1
2
2
m
K
m2 2 ,
n/22
2
m 2
(B.2)
where the remaining parameter integral has been expressed in terms of the modified Bessel
function [11],

1 t
t2
1
K (t) =
,
d
exp
2 2
4
| arg t| <
,
2
Re t 2 > 0.
(B.3)

For n = 4, we have I1 = 2(4)2 K0 ( m2 2 ) which is finite except for 2 0 since
K0 (x) ln x as x 0. This is the UV/IR mixing; the virtual UV singularity is
regularized at the cost of introducing an IR singularity in the external momentum. Consider
the case of m = 0. The virtual IR singularity in I1 may be regularized either by a small mass
or working in n dimensions. In the latter case, a similar calculation leads to

2n/2

I1 2 , 0 = (4)2 2 2
(n/2 2)

2
= (4) (n/2 2) ln 2 2 + O(n/2 2) ,
(B.4)
where the first term is the virtual IR divergence and the second is the would-be virtual UV
divergence regularized by the nonvanishing external momentum . More interesting is the
case when carries the momentum of a second loop involving massless particles so that
the virtual IR singularity may be enhanced. Using the above result we also obtain,

2
d nk
cos(k )
(k + p)2 + m2
I2 2 , m = 4n
n
(2)

= I1 2 , m cos(p ).
(B.5)
2 =
In the following we give the integrals appearing in Fig. 4(a). We shall assume
2 throughout for simplicity.

d 4 k1
d 4 k2 2
I3 (m, m) =
D (k1 )Dm (k2 + p)K12
4
(2)
(2)4 m

21 (4)4 p2 ln m4 2 ,
(B.6)
which is computed using I1 and the argument employed to simplify the k1 integral in
Eq. (43). Using I1 and Eq. (41), we have

d 4 k1
d 4 k2 2
I3 (m, 0) =
D (k1 )D(k2 + p)K12
4
(2)
(2)4 m

21 (4)4 p2 ln m2 p2 2 .
(B.7)
Now consider the integral,

d 4 k1
d 4 k2 2
I3 (0, m) =
D (k1 )Dm (k2 + p)K12 .
4
(2)
(2)4
(B.8)
522
This integral looks dangerous since there is a virtual IR singularity in k1 due to

masslessness which may be mixed up with that coming from the k2 loop to enhance the
final IR singularity in p. The masslessness may be regularized either by a small mass or in
dimensional regularization. In the first case, the result is obtained from I3 (m, m) by setting
the first m to be the small mass. It is clear that the two IR singularities are separated from
each other. In the second case, we proceed as follows. Using I1 (k22 , 0) we have

d 4 k2
I3 (0, m) =
Dm (k2 + p)
(2)4

(4)2 (n/2 2) ln 2 k22 + O(n/2 2) ,
(B.9)
where the first and second terms are, respectively, from the IR and UV regions of k1 . Only
the second one is of interest here since the first is proportional to p2 and thus does not
affect our main arguments on the Goldstone theorem. Finishing the k2 integral as before,
we have the contribution from the small k2 ,

I3 (0, m) 21 (4)4 p2 ln 2 m2 2 ,
(B.10)
the same as we get from I3 (m, m) using the small mass regularization. Without giving
further details, we have,

d 4 k1
d 4 k2 2
D (k1 )D(k2 + p)K12
I3 (0, 0) =
(B.11)
4
(2)
(2)4

21 (4)4 p2 ln 2 p2 2 + ,
where the dots are the usual terms in commutative theory that vanish in dimensional
regularization. Following are the examples of integrals appearing in Figs. 4(c), (e), (h),

d 4 k1
d 4 k2 2
I4 =
D (k1 )Dm (k2 )Dm (k1 + k2 )D(k1 + p)K12
(2)4
(2)4 m

21 (4)4 m4 p2 ln m2 p2 2 ,

d 4 k1
d 4 k2
I5 =
Dm (k1 )Dm (k2 )D(k1 + p)D(k2 + p)K12
(2)4
(2)4

(4)4 m2 p2 ln m2 p2 2 ,

d 4 k1
d 4 k2
I6 =
Dm (k1 )Dm (k2 )Dm (k1 + k2 )D(k1 + p)K12
4
(2)
(2)4

(B.12)
21 (4)4 m2 p2 ln p2 .
In the following we list integrals involving K1 that arise in Fig. 4(h) and are nontrivial
in the NC IR limit.

d 4 k2
d 4 k1
Dm (k1 )Dm (k2 )Dm (k1 + k2 )D(k2 + p)K1
(2)4
(2)4

21 (4)4 m2 p2 ln m2 p 2 ,

d 4 k1
d 4 k2
D(k1 )D(k2 )Dm (k1 + k2 )Dm (k2 + p)K1
4
(2)
(2)4

21 (4)4 m2 p2 ln m2 p 2 ,

d 4 k1
d 4 k2
Dm (k1 )D(k2 )D(k1 + k2 )Dm (k2 + p)K1
4
(2)
(2)4

21 (4)4 m2 p2 ln m2 p 2 ,

d 4 k1
d 4 k2
D(k1 )Dm (k2 )D(k1 + k2 )D(k2 + p)K1
(2)4
(2)4

21 (4)4 m2 p2 ln p2 p 2 .
523
(B.13)
References
[1] For reviews, see: M.R. Douglas, N.A. Nekrasov, Noncommutative field theory, Rev. Mod. Phys. 73 (2001)
977, hep-th/0106048;
R.J. Szabo, Quantum field theory on noncommutative spaces, hep-th/0109162.
[2] S. Minwalla, M.V. Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, J. High Energy
Phys. 02 (2000) 020, hep-th/9912072;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, Two-loop diagrams in noncommutative 44 theory, Phys. Lett.
B 476 (2000) 431, hep-th/9912075;
M.V. Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics, J. High Energy
Phys. 03 (2000) 035, hep-th/0002186;
A. Matusis, L. Susskind, N. Toumbas, The UV/IR connection in the noncommutative gauge theories, J. High
Energy Phys. 12 (2000) 002, hep-th/0002075.
[3] C.P. Martin, D. Sanchez-Ruiz, The one-loop UV divergent structure of U (1) YangMills theory on
noncommutative R 4 , Phys. Rev. Lett. 83 (1999) 476, hep-th/9903077;
T. Krajewski, R. Wulkenhaar, Perturbative quantum gauge fields on the noncommutative torus, Int. J. Mod.
Phys. A 15 (2000) 1011, hep-th/9903187;
M.M. Sheikh-Jabbari, Renormalizability of the supersymmetric YangMills theories on the noncommutative
torus, J. High Energy Phys. 06 (1999) 015, hep-th/9903107;
M. Hayakawa, Perturbative analysis on infrared aspects of noncommutative QED on R 4 , Phys. Lett. B 478
(2000) 394, hep-th/9912094;
M. Hayakawa, Perturbative analysis on infrared and ultraviolet aspects of noncommutative QED on R 4 ,
hep-th/9912167;
A. Matusis, L. Susskind, N. Toumbas, J. High Energy Phys. 12 (2000) 002, hep-th/0002075;
I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, O.A. Rychkov, Renormalizability and UV/IR mixing in
noncommutative theories with scalar fields, Phys. Lett. B 487 (2000) 357.
[4] A. Armoni, Comments on perturbative dynamics of non-commutative YangMills theory, Nucl. Phys. B 593
(2001) 229, hep-th/0005208;
L. Bonora, M. Salizzoni, Renormalization of noncommutative U (N ) gauge theories, Phys. Lett. B 504
(2001) 80, hep-th/0011088;
C.P. Martin, D. Sanchez-Ruiz, The BRS invariance of noncommutative U (N ) YangMills theory at the
one-loop level, Nucl. Phys. B 598 (2001) 348, hep-th/0012024.
[5] B.A. Campbell, K. Kaminsky, Noncommutative field theory and spontaneous symmetry breaking, Nucl.
Phys. B 581 (2000) 240, hep-th/0003137;
B.A. Campbell, K. Kaminsky, Noncommutative linear sigma models, Nucl. Phys. B 606 (2001) 613, hepth/0102022.
[6] F. Ruiz Ruiz, UV/IR mixing and the Goldstone theorem in noncommutative field theory, hep-th/0202011.
[7] F.J. Petriello, The Higgs mechanism in noncommutative gauge theories, Nucl. Phys. B 601 (2001) 169,
hep-th/0101109.
[8] Y. Liao, One loop renormalization of spontaneously broken U (2) gauge theory on noncommutative
spacetime, J. High Energy Phys. 11 (2001) 067, hep-th/0110112;
524
Y. Liao, One loop renormalizability of spontaneously broken gauge theory with a product of gauge groups
on noncommutative spacetime: the U (1) U (1) case, hep-th/0201135.
[9] I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, Phys. Lett. B 476 (2000) 431, hep-th/9912075;
A. Micu, M.M. Sheikh-Jabbari, Noncommutative 4 theory at two loops, J. High Energy Phys. 01 (2001)
025, hep-th/0008057.
[10] See, for example, M. Chaichian, P. Presnajder, M.M. Sheikh-Jabbari, A. Tureanu, Noncommutative gauge
field theories: a no-go theorem, Phys. Lett. B 526 (2002) 132, hep-th/0107037;
M. Chaichian, P. Presnajder, M.M. Sheikh-Jabbari, A. Tureanu, Noncommutative standard model: model
building, hep-th/0107055;
For discussions on the unitarity problem in the latter work, see: J.L. Hewett, F.J. Petriello, T.G. Rizzo,
Noncommutativity and unitarity violation in gauge boson scattering, hep-ph/0112003.
[11] I.S. Grashteyn, I.M. Ryzhik, Table of Integrals, Series, and Products, 5th edn., Academic Press, 1994.

Generalized zeta functions and one-loop corrections

to quantum kink masses
A. Alonso Izquierdo a , W. Garca Fuertes b , M.A. Gonzlez Len a ,
J. Mateos Guilarte c
a Departamento de Matemtica Aplicada, Universidad de Salamanca, Spain
b Departamento de Fsica, Universidad de Oviedo, Spain
c Departamento de Fsica, Universidad de Salamanca, Spain
Received 19 December 2001; accepted 15 April 2002
Abstract
A method for describing the quantum kink states in the semi-classical limit of several (1 + 1)dimensional field theoretical models is developed. We use the generalized zeta function regularization
method to compute the one-loop quantum correction to the masses of the kink in the sine-Gordon
and cubic sinh-Gordon models and another two P ()2 systems with polynomial self-interactions.
PACS: 11.15.Kc; 11.27.+d; 11.10.Gh
1. Introduction
BPS states arising both in extended supersymmetric gauge theories [1] and string/Mtheory [2] play a crucial rle in the understanding of the dualities between the different
regimes of the system. In this framework, domain walls appear as extended states in N = 1
SUSY gluodynamics and the WessZumino model [3]. This circumstance prompted the
question of whether or not these topological defects saturate the quantum Bogomolny
bound. A return to the study of quantum corrections to the masses of (1 + 1)-dimensional
solitons has thus been unavoidable. These subtle matters were first addressed in the
classical papers of Dashen, Hasslacher and Neveu [4], and Faddeev and Korepin [5], for
the purely bosonic []42 and sine-Gordon theories, and then in Ref. [6] for the supersymmetric extension of these theories. Analysis of the ultraviolet cut-off regularization
E-mail address: guilarte@usal.es (J. Mateos Guilarte).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 4 1 - 3
526
A. Alonso Izquierdo et al. / Nuclear Physics B 635 [PM] (2002) 525557
procedure in the presence of a background is the main concern in the papers of Ref. [7]: the
authors carefully distinguish between using a cut-off either in the energy or in the number
of modes. The second method leads to the same result as in the computation performed
by DHN for bosonic fluctuations. Another point of view is taken in Ref. [8], where SUSY
boundary conditions related more to infrared behaviour, are carefully chosen. On this basis,
and by using higher-derivative ultraviolet regularization (SUSY preserving), the authors
demonstrated an anomaly in the central charge that compensates for the extra (quantum)
contribution to the classical mass.
In this paper we shall confine ourselves to purely bosonic theories and leave the
treatment of fermionic fluctuations for future research. We address the quantization of nonlinear waves relying on the generalized zeta function regularization method to control the
infinite quantities arising in the quantum theory. This procedure has been used previously in
computing Casimir energies and the quantum corrections to kink masses, see [9]. We shall
develop this topic, however, in a completely general way, also offering a comparison with
other approaches. As well as obtaining exact results, we also shall explain how asymptotic
methods lead to a very good approximation of the right answer. We believe that the novel
application of the asymptotic method should be very useful in the cubic sinh-Gordon model
as well as in multi-component scalar field theory, where the traditional approach is limited
by the lack of detailed knowledge of the spectrum of the second-order fluctuation operator
(see [11,12] for extensive work on multi-component kinks and their stability).
The organization of the paper is as follows: in Section 2 the general semi-classical
picture of quantum solitons, the zeta function regularization procedure, the zero-point
energy and mass renormalization prescriptions, and the asymptotic method are described.
In Section 3, we apply the method to the loop kinks of the sine-Gordon, ()42 , and cubic
sinh-Gordon models. In the first two paradigmatic cases, it is possible to obtain an exact
result, which allows comparison with other methods. Approximate computations by means
of the asymptotic expansion of the heat function are also offered to test the goodness of
our procedure against the well known exact answers. Section 4 is devoted to the analysis of
the link kink arising in the ()62 model. Finally, Section 5 offers an outlook on further
applications of our approach.
2. Semi-classical picture of quantum soliton states

We shall consider (1 + 1)-dimensional scalar field theories whose classical dynamics is
governed by the action

1
S[] = d 2 y
U
()
.
2 y y
We choose the metric tensor in T 2 (R1,1 ) as g = diag(1, 1) and the Einstein convention
will be used throughout the paper. We shall not use a natural unit system because we
wish to keep track of h in our formulas; nevertheless, we choose the speed of light to
be c = 1. Elementary dimensional analysis tells us that [h ] = ML, [U ()] = ML1 and
[] = M 1/2 L1/2 are the dimensions of the important quantities.
527
The classical configuration space C is formed by the static configurations (y), for
which the energy functional

1 d d
+ U ()
E() = dy
2 dy dy
is finite: C = {(y)/E() < +}. In the Schrdinger picture, quantum evolution is ruled
by the functional differential Schrdinger equation:

(y), t = H (y), t .
t
The quantum Hamiltonian operator

h v 2
+ E (y) ,
H = dy
2 (y) (y)
i h
with v the mass of the fundamental quanta, acts on wave functionals [(y), t] that belong
to L2 (C).
We wish to compute the matrix element of the evolution operator in the field
representation
i

G (f ) (y), (i) (y); T = (f ) (y) e h T H (i) (y)

i
= D (y, t) exp S[]
(1)
h
for the choice
(i) (y, 0) = K (y),
(f ) (y, T ) = K (y),
where K (y) is a kink static solution of the classical field equations. We are, however,
only interested in the loop (h ) expansion of G up to the first quantum correction. Also
performing a analytic continuation to Euclidean time, t = i , T = i, this is achieved
by the steepest-descent method applied to the Feynman integral in (1):

2
E[K ]
1/2
Det
2 + P K 1 + o(h ) ,
GE K (y), K (y); = exp
h
where K is the differential operator

d 2 U
2
,
K= 2 +
y
d 2 K
and P is the projector over the strictly positive part of the spectrum of K. Note that, on
the mathematical side, the steepest-descent method provides a well defined approximation
to the Feynman integral if the spectrum of the quadratic form K is positive definite and,
on the physical side, the zero eigenvalue that appears in Spec(K) contributes to the next
order in the loop expansion: it is due to neutral equilibrium on the orbit of the kink solution
under the action of the spatial translation group. Moreover, in order to avoid the problems
that arise in connection with the existence of a continuous spectrum of K, we place the
528
system in a interval of finite but very large length L, i.e., x [ L2 , L2 ], and, eventually
after assuming periodic boundary conditions on the small fluctuations all throughout the
paperwe shall let L go to infinity.
From the spectral resolution of K,
Kn (y) = n2 n (y),
n2 Spec(K) = Spec(P K) + {0},
we write the functional determinant in the form

2
2
2
det 2 + n .
Det 2 + K =
n
All the determinants in the infinite product correspond to harmonic oscillators of frequency
in and thus, with an appropriate normalization, we obtain for large

n 1/2

E[K ]
GE K (y), K (y);
e 2 n n 1+o(h ) ,
= e h
h
n
where the eigenvalue in the kernel of K has been excluded.
Inserting eigen-energy wave functionals

Hj K (y) = j j K (y)
we have an alternative expression for GE for :

0

GE K (y), K (y);
= 0 K (y) 0 K (y) e h
and, therefore, we obtain

h
0 = E[K ] +
n + o h 2 ,
2 2
n >0

0 K (y) 2 = Det1/4 P K + o h 2 ,
2
2
h
as the kink ground state energy and wave functional up to one-loop order.
We define the generalized zeta function
1
P K (s) = Tr(P K)s =
(n2 )s
2
n >0
associated to the differential operator P K. Then,

1
h
h
Tr(P K)1/2 + o h 2 = E[K ] + P K
+ o h 2 ,
2
2
2

2

0 K (y) = h exp P K (0) exp 1 dP K (0)
4 ds
0K = E[K ] +
(2)
(3)
show how to read the energy and wave functional of the quantum kink ground state in
terms of the generalized zeta function of the projection of the second variation operator K
in the semi-classical limit.
529
2.1. The generalized zeta function regularization method: zero-point energy and mass
renormalizations
The eigen-functions of K form a basis for the quantum fluctuations around the
kink background. Therefore, the sum of the associated zero-point energies encoded in
P K (1/2) in formula (2) is infinite and we need to use some regularization procedure.
We shall regularize P K (1/2) by defining the analogous quantity P K (s) at some point in
the s complex plane where P K (s) does not have a pole. P K (s) is a meromorphic function
of s, such that its residues and poles can be derived from heat kernel methods, see [13]. If
KK (y, z; ) is the kernel of the heat equation associated with K,

+ K KK (y, z; ) = 0,
KK (y, z; 0) = (y z)
(4)
the Mellin transformation tells us that,

1
P K (s) =
(s)

d s1 hP K (),
0
where
hP K [] = Tr e
P K
= Tr e

1 = 1 +
dy KK (y, y; )
is the heat function hP K [], if K is positive semi-definite and dim Ker(K) = 1. Thus, the
regularized kink energy is in the semi-classical limit:

h
0K (s) = E[K ] + 2s+1 P K (s) + o h 2 ,
2
(5)
where is a unit of length1 dimension, introduced to make the terms in (5) homogeneous
from a dimensional point of view. The infiniteness of the bare quantum energy is seen here
in the pole that the zeta function develops for s = 1/2.
To renormalize 0K we must: (i) Subtract the regularized vacuum quantum energy.
(ii) Add counter-terms that will modify the bare masses of the fundamental quanta, also
regularized by means of the generalized zeta function. (iii) Take the appropriate limits.
(1) The quantum fluctuations around the vacuum are governed by the Schrdinger
operator:

d 2 U
d2
,
V = 2 +
dy
d 2 V
where V is a constant minimum of U []. The kernel of the heat equation

+ V KV (y, z; ) = 0,
KV (y, z; 0) = (y z)
provides the heat function hV (),

hV () = Tr e V = dy KV (y, y; ).
530
We exclude the constant mode and, through the Mellin transformation, we obtain
1
V (s) =
(s)
d s1 Tr e V .
The regularized kink energy measured with respect to the regularized vacuum energy is
thus:

K (s) = E[K ] + 1 K (s) + o h 2

h
= E[K ] + 2s+1 P K (s) V (s) + o h 2 .
2
(2) If we now go to the physical limit K = lims 1 K (s), we still obtain an infinite
2
result. The reason for this is that the physical parameters of the theory have not been
renormalized. It is well known that in (1 + 1)-dimensional scalar field theory normal
ordering takes care of all the infinities in the system: the only ultraviolet divergences that
occur in perturbation theory come from graphs that contain a closed loop consisting of a
single internal line, [14]. From Wicks theorem, adapted to contractions of two fields at
the same point in spacetime, we see that normal ordering adds the mass renormalization
counter-term,

h
d 2U
: + o h 2
dy m2 :
H m2 =
2
2
d
to the Hamiltonian up to one-loop order. To regularize

1
dk
2

m =
4 k 2 + U (V )
1
V ( 12 ), if the
we first place the system in a 1D box of length L so that m2 = 2L
constant eigen-function of V is not included in V . Then, we again use the zeta function
regularization method and define:
m2 (s) =
1 (s + 1) 2s+1
V (s + 1).
L (s)
Note that
m2 = lim m2 (s).
s 12
The criterion behind this regularization prescription is the vanishing tadpole condition,
which is shown in Appendix B of Ref. [10] to be equivalent to the heat kernel subtraction
scheme.
The one-loop correction to the kink energy due to H (m2(s)) is thus

2 K (s) = K |H m2 (s) |K V |H m2 (s) |V
h 2s+1 (s + 1)
= lim
L 2L
(s)
L/2
L/2

d 2 U
d 2 U
dy V (s + 1)
d 2 K d 2 V

h 2s+1 (s + 1)
V (s + 1)
= lim
L 2L
(s)
531
L/2
dy V (y)
(6)
L/2
because the expectation values of normal ordered operators in coherent states are the
corresponding c-number-valued functions.
(3) The renormalized kink energy is thus

K
= E[K ] + 6MK + o h 2
R

= E[K ] + lim 1 K (s) + 2 K (s) + o h 2
(7)
s 12
whereas the renormalized wave functional reads

2
dx R (K )
0
= Det 4
PK

Det
(1)
4
2 h 2
2 h 2

dV
1 dP K
(0)
(0) .
= h exp P K (0) V (0) exp
4 ds
ds
2.2. Asymptotic approximation to semi-classical kink masses

In order to use the asymptotic expansion of the generalized zeta function of the K
operator to compute the semi-classical expansion of the corresponding quantum kink mass,
it is convenient to use non-dimensional variables. We define non-dimensional spacetime
coordinates x = md y and field amplitudes (x ) = cd (y ), where md and cd are
constants with dimensions [md ] = L1 and [cd ] = M 1/2L1/2 to be determined in each
2
() = cd2 U ().
specific model. Also, we write U
md
The action and the energy can now be written in terms of their non-dimensional
counterparts:

1
1
1
2

S[] = 2 d x
U () = 2
S[],
2 x x
cd
cd

md
1 d d
md
+ U () = 2 E[].
dx
E[] = 2
2 dx dx
cd
cd
The important point is that the Hessians at the vacuum and kink configurations now read

d2
d2
2
2
2
2
2

K = md 2 + v V (x) = m2d K
V = md 2 + v = md V,
dx
dx
2
2
(x). Therefore,
where ddU2 = v 2 and ddU2 = v 2 V
V
V (s) =
V (s),
m2s
d
K (s) =
K
(s).
m2s
d
532

The asymptotic expansion is superfluous if Tr eP K and P K

(s) are susceptible of an
(x) is a potential well of the PoschTeller type, see [15], one can
exact computation. If V
and there is no need for any approximation to
completely solve the spectral problem for K

P K
(s). In general the spectrum of K is not known in full detail, specially in systems with
multi-component kinks, and we can only determine P K
(s) by means of an asymptotic
expansion. Nevertheless, we shall also compute the asymptotic expansion of P K
(s) in
the cases where the exact answer is known in order to estimate the error accepted in this
approach.
K
and v 2 and write
In the formulas (4), (5), (6) and (7) we replace V, K and v 2 by V,

the kernel of the heat equation for K in the form:

KK
x, x ; = KV
x, x ; A x, x ; ,
A(x, x ; ) is thus the solution of the PDE

x x
2

+
2 V (x) A x, x ; = 0
x x
(8)
with initial condition: A(x, x ; 0) = 1.

For < 1, we solve (8) by means of an asymptotic (high-temperature) expansion:

A x, x ; =
an x, x n .
n=0
Note that there are no half-integer powers of in this expansion because our choice of
boundary conditions with no boundary effects.
In this regime the heat function is given by:
Tr e

K
md L/2
dx KK
(x, x; )
md L/2
2
e v
=
4 n=0
md L/2
2
e v
n.
dx an (x, x) =
an (K)
4 n=0
md L/2
It is not difficult to find the coefficients an (x, x) by an iterative procedure starting from
a0 (x, x ) = 1. This procedure is explained in Appendix A, which also includes the explicit
expression of some of the lower-order coefficients.

P K
in the formula for the quantum
The use of the power expansion of hP K
[] = Tr e
kink mass is quite involved:
in the form:
(1) First, we write the generalized zeta function of V
1 md L
V (s) =
(s) 4
1
0
d s 2 e v + BV (s)
2
533
with
md L [s 12 , v 2 ]
,
BV (s) =
4 v 2s1 [s]
md L [s 12 , v 2 ]
V (s) =
+ BV (s)
4 v 2s1 (s)
and [s, v 2 ] and [s 12 , v 2 ] being, respectively, the upper and lower incomplete gamma
functions, see [16]. It follows that V (s) is a meromorphic function of s with poles at
the poles of [s 12 , v 2 ], which occur when s 12 is a negative integer or zero. BV (s),
however, is a entire function of s.
(2) Second, from the asymptotic expansion of hK
[] we estimate the generalized zeta

function of P K:

1
1

3
1
1
2

an (K)
d s+n 2 e v + bn0 ,K
P K
d s1 +
(s) =
(s)
(s)
4 n<n0
0
+ BP K
(s)
=

[s + n 12 , v 2 ]
1
1

+
an (K)
1
s (s) (s) 4 n<n
v 2(s+n 2 )
0
1
b (s) + BP K
(s),
(s) n0 ,K
where
[s + n 12 , v 2 ]
1

an (K)
bn0 ,K
(s) =
1
4 nn
v 2(s+n 2 )
0
is holomorphic for Re s > n0 + 12 , whereas

1
BP K
(s) =
(s)
d Tr eP K s1
is a entire function of s. The values of s where s + n 12 is a negative integer or zero are the
1 2
poles of P K
(s) because the poles of [s + n 2 , v ] lie at these points in the s-complex
plane.
Renormalization of the zero-point energy requires the subtraction of V (s) from P K
(s).
We find,

n0 1
[s + n 1 , v 2 ]
1
an (K)
1
2
+
P K
(s) V
(s)
1
(s)
s
4
v 2(s+n 2 )
n=1
and the error in this approximation with respect to the exact result to 1 K is:

hm
1
1
1
1
d
error1 =
bn0 ,K
+ BP K
BV
.

2
2
2
2
2
534
and hence, the

Note that the subtraction of V (s) exactly cancels the contribution of a0 (K)
1
divergence arising at s = 2 , n = 0. The quadratic ultraviolet divergences appear in this
scheme as related to the pole of V (s) at s = 12 , n = 0.
(3) Third, 1 K now reads
2 s
[s + 1 , v 2 ]
a1 (K)
h
h md
2
1 K
lim
+
=
v 2s+1
( 12 ) 2 s 12 m2d
4 (s)
+
n0 1

[n 1, v 2 ]
an (K)
hm
d
.
1
2
v 2n2
4 ( 2 )
n=2
The logarithmic ultraviolet divergences, hidden at first sight in the DHN approach, arise
1
here in connection with the pole of P K
(s) V
(s) at s = 2 , n = 1.
The surplus in energy due to the mass renormalization counter-term is,

2s+1 (s + 1)
ha
1 (K)
K
lim
V (s + 1) + o h 2
2 = lim
1
L
2L s 2 md
(s)

2s+1

[s + 12 , v 2 ]
h md
lim
+ o h 2
= a1 (K)
2s+1
v
(s)
2 4
s 21 md
and the deviation from the exact result is

h
1 .
a1 (K)B
error2 = lim
V 2
L 4L
Therefore,

K
= E[K ] + 6MK + o h 2
R

n0 1
hm
[n 1, v 2 ]
1
d

an (K)
= E[K ] 1 +
v 2n2
2
4
n=2

2
h md
2s+1 [s + 12 , v 2 ] [s + 12 , v 2 ]
lim
+ a1 (K)
+
o
h .
v 2s+1 (s)
v 2s+1 (s)
2 4
s 21 md
of the poles at s = 1 in 1 K (s) and
Note that the contributions proportional to a1 (K)
2
K
2 (s) cancel.
We are left with the very compact formula:

0 = 21 ,
h md [0 + Dn ]
6MK =
(9)
n0 1 an (K
0
) [n1,v 2 ]
.
Dn0 = n=2
2n2
8
v
In sum, there are only two contributions to semi-classical kink masses obtained by means
of the asymptotic method: (1) h md 0 is due to the subtraction of the translational mode;
(2) hm
d Dn0 comes from the partial sum of the asymptotic series up to the n0 1 order.
We stress that the merit of the asymptotic method lies in the fact that there is no need to
solve the spectral problem of K: all the information is encoded in the potential V (x).
535
3. Loop kinks
The existence of kinks is guaranteed if the minima of U () are a discrete set which is
the union of orbits of the discrete symmetry group of the system. We shall use the term
loop kinks to refer to those classical solutions that interpolate between vacua belonging
to the same orbit of the symmetry group; otherwise, the solitary waves will be referred to
as link kinks, see [18]. In this section we shall discuss three kinks of the loop type.
3.1. The quantum sine-Gordon soliton
We first treat the sine-Gordon model by considering the potential energy density:

m4
.
U y =
1 cos
m
The dimensions of the parameters m and are,respectively: [m] = L1 and [] =
M 1 L3 . Therefore, we choose md = m and cd = /m and find:

(x, t) = (1 cos ).
U
The internal symmetry group of the system is the infinite dihedral group D = Z2 Z
generated by internal reflections, , and 2 translations, + 2 . The
vacuum classical configurations V (x, t) = 2n form the orbit M = D /Z2 and there
is spontaneous symmetry breakdown of the internal translational symmetry through the
= M/D , is a single point
choice of vacuum. The moduli space of vacua, however, M
and all the equivalents kinks of the model,
K (x, t) = 4 arctan ex + 2n,
4m
2nm
K (y, y0 ) = arctan emy + ,
are loop kinks. It is easy to check that E[K ] = 8m

and E[V ] = 0.
The second order variation operator around the kink solutions is
K=

d2
+ m2 1 2 sech2 my ,
2
dy
henceforth, P K (s) =
Note that K = m2 K;
we have:
V =
d2
+ m2 ,
dy 2
2
= d + 1 2 sech2 x.
K
dx 2
1
(s).
m2s P K
2
= d + 1,
V
dx 2
Simili modo, in the vacuum sector

V = m2 V,
V (s) =
1
(s).
m2s V
3.1.1. Exact computation of the mass and the wave functional

Generalized zeta function of V.
= k 2 + 1, with
acting on functions belonging to L2 (R) is Spec V
The spectrum of V
k R a real number. There is a half-bound state fk 2 =0 (x) = constant that we shall not
The spectral
consider because it is paired with the other half-bound state in Spec(K).
mL mL
density on the interval I = [ 2 , 2 ] with periodic boundary conditions is V (k) = mL
2 .
536
The heat function is,

Tr e

V
mL
=
2
dk e(k
2 +1)
mL
e
=
4
and the generalized zeta function reads:

mL
V (s) =
(s) 4

0
3
mL (s 12 )
.
d s 2 e =
4 (s)
Therefore, V (s) (hence V (s)) is a meromorphic function of s with poles at s =

1
1
3
5
2 , 2 , 2 , 2 , . . . . The generalized zeta function of the Hessian at the vacuum is,
however, also infrared-divergent: it is linearly divergent when L even at points s C
where V (s) is regular.

Generalized zeta function of K.
2

In this case Spec K = {0} {k + 1}, k R and the spectral density on I is
1 d(k)
mL
+
2
2 dk
with phase shifts
K
(k) =
(k) = 2 arctan
1
k
is the Schrdinger operator that governs the scattering through the first of the
because K
transparent PoschTeller potentials, [15]. Thus,
Tr e

K
mL
=1+
2

dk e
(k 2 +1)
1
+
2

mL
e Erfc ,
=1+
4

dk
d(k) (k 2 +1)
e
dk
has a zero mode, the

where Erfc is the complementary error function, [16]. Note that K
2
eigen-function being the translational mode dK /dx = sech x, which must be subtracted
because it arises in connection with the breaking of the translational symmetry, x x + a,
by the kink solution and does not contribute to the kink mass up to this order in the loop
expansion. There is also a half-bound state, fk 2 =0 (x) = tanh x, that exactly cancels the
contribution of the constant half-bound state in Spec V. Therefore, we obtain

Tr eP K = Tr e K 1 = Tr e V Erfc
and
1
P K
(s) = V
(s)
(s)

0

1 (s + 12 )
.
d Erfc s1 = V (s)
s (s)
537
P K
(s) (hence P K (s)) is also a meromorphic function of s that shares all the poles with
V (s), but the residues are different except at s = 1/2, a pole where the residues of P K
(s)
and V (s) coincide. The infrared divergence, however, is identical in the kink background
and the vacuum.
We can now compute the limit of the regularized quantities that enter in the one-loop
correction formula to the kink mass:
2 s

h
1 K =
(s)
(s)
lim

PK
V
2 s 12 m2

2
()
hm
= lim
2 0 m
( 12 + )

1
1
hm
lim 2 log (1) +

+ o()
=
(10)
2 0
m
2
and
2s+1

(s + 1)
2h
lim
V (s + 1) + o h 2
2 =
1
L s 2 m
(s)
2s+1
2
(s + 12 )
h m
h m
()
= lim
= lim
(s)
s 12 m
0 m
( 12 + )

hm
1
1
=
lim
+ 2 log + (1)
+ o() + o h 2 ,
2 0
m
2
K
where (z) = (z)/ (z) is the digamma function.

The important point to notice is that the renormalization of the zero-point energy
performed by the subtraction of V (1/2) still leaves a divergence coming from the
s = 1/2 poles because the residues are different. The correction due to the mass
renormalization counter-term also has a pole. The sum of the contributions of the two
poles leaves a finite remainder and we end with the finite answer:
h m
,

8m h m

h m
K
R
+ o h 2 =
+ o h 2 .
= E[K ]
1 K + 2 K =
(11)
The one-loop quantum correction to the mass of the sine-Gordon soliton obtained by means
of the generalized zeta function procedure exactly agrees with the accepted result, see [4,
5], and, henceforth, with the outcome of the mode number regularization method [7].
The square of the modulus of the ground state wave functional up to one-loop order
2 and C = h /m2 , obviously, W (s) = C 2s (s)
is given by (3). If W = P K/C
d
PK
(W (0) = P K
(0)) and we have
dW

2s dP K
= C 2s P K
(s).
(s) log C + C
ds
ds
538
Thus,

d
dP K
1 (s + 12 )
1
1

= V
s+
(s) ,
ds
ds
2
s
s (s)

1
dV
mL (s 2 )
1
=
s
(s)
ds
2
4 (s)
(12)
and
V (0) = 0,
P K
(0) = 1,
dV
(0) = mL,
ds

dP K
1

(0) = mL + (1)
.
ds
2
Therefore,

0 K (x) 2 =

1
C
exp mL ,
2
4

0 V (x) 2 = exp 1 mL .
4
Renormalizing the wave functional with respect to the vacuum we obtain

|0 (K (x))|2
C
.
=
2
2
|0 (V (x))|
(13)
3.1.2. The asymptotic expansion and quantum corrections

In the sine-Gordon model the exact formulas for Tr eP K and P K
(s) are readily
is completely known. On
derived because the spectrum of the Schrdinger operator K
the other hand, the series expansion of the complementary error function tells us that

n n 12

2
1
mL

e 1
+
Tr eP K =
1 3 5 (2n 1)
4
n=1
coefficients can be computed from this exact expression:
and the an (K)
e

n 12 1,
Tr eP K =
an (K)
4 n=0
= mL,
a0 (K)
=
an (K)
2n+1
.
(2n 1)!!
One can check by direct calculation that indeed,

=
an (K)
mL/2
dx an (x, x),
n = 0, 1, 2, 3, . . .
mL/2
are the integrals of the functions defined in Appendix A for V

(x) =
and the an (K)
2
2 sech x.
539
In any case we see from the formula (9) that the comparison with the exact result is
satisfactory:
6MK
= 0.282095h m
n
0 1
n h m
an (K)d
vs.
n=2
6MK = 0.318309h m.
The partial sums
Dn0 =
n
0 1
n=
an (K)d
n=2
n
0 1
n=2

an (K)
[n 1, 1]
8
can be estimated with the help of Table 1. For instance, choosing n0 = 10 we find that
D10 = 0.080586 and the correction obtained by means of the asymptotic expansion is:

8m3
E[K ] + 6MK
0.362681h m + o h 2 .
=
In fact

h m
1
1
1
h m
BV
BP K
BV
+ lim

L L
2
2
2
2

h m
1 Erfc
e
d
+
= 0.044373h m
=
2 3/2
2

1
is almost the total error: 0.044372h m. The difference is:

1
hm
b10,K
106 h m.
2
4
rapidly decreases with increasing n.
Note in Table 1 that an (K)
Table 1
n

an (K)
n0 1
Dn0
2
3
4
5
6
7
8
9
2.66667
1.06667
0.30476
0.06772
0.012324
0.0018944
0.00025258
0.00002972
2
3
4
5
6
7
8
9
0.0670702
0.0782849
0.0802324
0.0805373
0.0805803
0.0805857
0.0805863
0.0805863
540
3.2. The quantum ( 4 )2 kink

We now consider the other prototype of solitary waves in relativistic (1+1)-dimensional
field theory: the kink of the ( 4 )2 model. The potential energy density is:

m2 2
2
=
.
U y

4
[(x, t)] = 1 ( 2
We choose, however, md = m/ 2 but keep cd = /m and find: U
2
1)2 .
The internal symmetry group is now the Z2 group generated by the reflections
and the orbit of vacuum classical configurations V (x, t) = 1 M gives rise to a moduli
= M/Z2 which is a single point. The kink solitary waves are thus loop
space of vacua M
kinks and read
m
my
K (y, y0) = tanh .
K (x, t) = tanh x,
2
3
The kink and vacuum solutions have classical energies of E[K ] = 43 m and E[V ] = 0,
2
respectively. The Hessian operators for the vacuum and kink solutions are

d2
m2
m2
d2
2
2 +4 =
V,
V = 2 + 2m =
2
2
dy
dx

d2
d2
m2
m2
3m2
6
K = 2 + 2m2
K
=
=
+
4
2
2
dy
dx 2
cosh2 x
cosh2 my/ 2
and the corresponding generalized zeta functions satisfy
2s
2s
2
2
P K
V (s) =
V (s).
P K (s) =
(s),
m
m
3.2.1. Exact computation of the semi-classical mass and wave functional
= d 22 + 4.
Generalized zeta function of V
dx
= {k 2 + 4}, k R, but the
Acting on the L2 (R) C Hilbert space we have that Spec V
mL
mL
spectral density on the interval I = [ , ] of eigen-functions with periodic boundary

conditions is V (k) =
easily computed:
Tr e

V
2 2 2 2
mL
.
2 2
mL
=
2 2
dk e(k
mL 1
V (s) =
8 (s)
From these data, the heat and generalized zeta functions are
2 +4)
mL 4
e
=
,
8
3
mL 1 (s 12 )
d s 2 e4 =
1
8 4s 2 (s)
(14)
and we find that V (s) has the same poles and infrared behaviour in the ( 4 )2 and the
sine-Gordon models.
541
Generalized zeta function of

2
6
= d +4
.
K
dx 2
cosh2 x
is the Schrdinger operator for the second transparent PoschTeller potential [15].
K
= {0} {3} {k 2 + 4}, k R, and the spectral density on I is
Thus, Spec K
1 d(k)
mL
,
+
K
(k) =
2
dk
2 2
3k
where the phase shifts are (k) = 2 arctan 2k
2 , if PBC are considered. Thus, we find
Tr e

P K
=e
mL
+
8

dk e
(k 2 +4)
1
+
2

dk
d(k) (k 2 +4)
e
dk

mL 4
e
=
+ e3 1 Erfc Erfc 2 .
8
The Mellin transform immediately provides the generalized zeta function:

(s + 12 )
1 3 1
2
1
1 1
,s + , , s ,
P K
(s) = V
(s) +
2 F1
2
2 2 3
4 s
(s) 3s+ 21
(15)
where 2 F1 [a, b, c, d] is the Gauss hypergeometric function, [16].

The power expansion of 2 F1 ,

1 3 1
1
,s + , ,
2 F1
2
2 2 3
=

(1)l (l + 12 ) (s + l + 12 )
( 12 ) (s + 12 ) l=0 3l l!
(l + 32 )
( 32 )
1
3
5
tells us that, besides the poles of V (s), P K
(s) has poles at s = 2 +l, 2 +l, 2 +l, . . . ,
+
l Z {0}; i.e., as in the sG soliton case, V (s) and P K
(s) share the same poles except
s = 1/2 but the residues in the ( 4 )2 model are increasingly different with larger and
larger values of | Re s|.
Applying these results to the kink mass formula, we obtain

h 22 s
1 = lim
P K
(s) V
(s)
2
1 2
m
s 2

2

3 1
1
2
1
()
2
h m
, , ,
=
lim
2 F1
1
2
2 3
( 12 + ) 3
2 2 0 m2
( 12 + )4 2 +

h m
22
3 1
3
3
1
=
lim 3 ln 2 + 2 + ln 2 F1 , 0, , + o() ,
m
4
2
2 3
2 2 0
K
542
2 s+ 1
2 [s + 1]
2
6h
lim
(s + 1)
2
L L s 1 m
[s] V
2
2

2
3h m
4 ()
+ o h 2
lim
=
2
1
( 2 + )
2 0 m

22
1
3h m
1
+ ln 2 ln 4 + (1)
+ o() + o h 2 ,
=
lim
2
m
2 2 0
2 K = lim
where 2 F1 is the derivative of the Gauss hypergeometric function with respect to the second
argument. Therefore,
3h m
h m
1 K + 2 K = ,
2 6 2
and we obtain:
K
R
= E[K ] + 6MK =

1
3
4 m3
+ h m + o h 2 ,
3 2
2 6 2
the same answer as offered by the mode-number regularization method [7].

To compute the norm of the ground state wave functionals we closely follow the
procedure applied in Section 3.1. to the sine-Gordon soliton. In the ( 4 )2 model, we
find that
d
dP K
1 (s + 12 )

= V +
ds
ds
(s)

1
1
+ ln 4 + (s) s +
4s
s
2

1
1
3
1
1
1
2s3s 2 2 F1 , s + , ,
log 3 s +
+ (s)
2
2 2 3
2

1
1
3
1
1
+ 2s3s 2 2 F1 , s + , ,
2
2 2 3
and

dV
mL 1 (s 12 )
1
=
s
(s) log 4 ,
1
ds
2
8 4s 2 (s)
from these expressions and formulas (14) and (15) one checks that
P K
V (0) = 0,
(0) = 1,
dV
dP K

(0) = 2 mL,
(0) = 2 mL + log 48.
ds
ds
We obtain

mL
C 1/2
0 K (x) 2 = 1
0 V (x) 2 = exp mL
exp ,
.
2
3
2 2
2 2
The quotient of the probability densities is

|0 (K (x))|2 1 C 1/2
=
.
|0 (V (x))|2 2
3
543
(16)
3.2.2. The asymptotic expansion and quantum corrections

In the ( 4 )2 model

2
2

d 2U
(x) = 4 and V
(x) = d U (x) d U (x) = 6 sech2 x
d 2 V
d 2 V
d 2 K
are the potentials of the Schrdinger operators that, respectively, correspond to the
Hessians at the vacuum and the kink configurations. The asymptotic expansion of the heat
function
e4
mL
Tr eP K = 1 +
4
2

dx an (x, x) n
n=0 mL

2 2
e4
= 1 +
4
n1/2
an (K)
n=0
can be either obtained as a series expansion of the exact result

1 2n (1 + 22n1 ) n 1 4
mL

P K
Tr e
2 e
+
= 1 +
(2n 1)!!
8
n=1
(x) = 6 sech2 x
or from the coefficients defined in Appendix A for V
mL
=
an (K)
2
dx an (x, x),
mL
= mL
a0 (K)
,
2
n+1
2n1 )
= 2 (1 + 2
.
an (K)
(2n 1)!!
2 2
To compare with the exact result, we apply the formula given in Appendix A and
observe that
K
R
n
0 1

4 m3
n h m + o h 2
an (K)d
= 0.199471h m
3 2
n=2
is far from the exact result

K
=
R

4 m3
0.471113h m + o h 2
3 2
before adding the contribution of the terms between n = 2 and n = n0 1 in the asymptotic
expansion to the contribution coming from the subtraction of the translational mode. The
544
partial sums
Dn0 =
n
0 1
n=
an (K)d
n=2
n
0 1
n=2
[n 1, 4]

an (K)
8 2 4n1
can be estimated up to n0 = 11 with the help of Table 2.

Choosing n0 = 11,we find that D11 = 0.271900h m and the correction obtained by
adding D11 h m is:

6MK
= 0.471371h m + o h 2
in good agreement with the exact result above. In fact

h m
1
1
1
3h m
B
BP K
B
+

V
V
2
2
2
2
2L
3

hm
e
e3 Erfc Erfc2 3e4
d 3 +
+
+
=
3
3
2 2
2 2
2 2
2 2
1
0.00032792h m
is almost the total error: 0.0002580h m. The deviation is

h m
1
b10,K
104 h m.
2
4 2
With respect to the sine-Gordon model there are two differences: (a) in the ( 4 )2 model
the error committed by using asymptotic methods is smaller, of the order of 104 h m, a 0.07
percent, as compared with 102 h m, a 6.00 percent, in the sG case; (b) the rejection of the
contributions of the n0 > 11 terms and the non-exact computation of the mass counter-term
contribution has a cost of approximately 104 h m in the ( 4 )2 model versus 106 hm
in
the sG system. Both facts have to do with the larger value of the smaller eigenvalue of the
vacuum Hessian in the ( 4 )2 model with respect to the sG system, 4 vs. 1.
Table 2
n
2
3
4
5
6
7
8
9
10

an (K)
n0 1
Dn0
24.0000
35.2000
39.3143
34.7429
25.2306
15.5208
8.27702
3.89498
1.63998
2
3
4
5
6
7
8
9
10
0.165717
0.221946
0.248281
0.261260
0.267436
0.270186
0.271317
0.271748
0.271900
545
3.3. The cubic sinh-Gordon kink

We shall now study a system of the same type where the potential energy density is:
2

m4
2
U y
1 .
=
sinh
4
m
Non-dimensional quantities are defined through the choice md = m and cd = /m; the
EulerLagrange equation is

1
(t, x) = sinh(2) sinh2 1
(17)
2
and the justification for the choice of name is clear. We find this model interesting because it
reduces to the ( 4 )2 system if |(t, x)| < 1 and is the Liouville model [17], with opposite
sign of the coupling constant, in the (t, x)
= ranges. In fact, the potential energy
density

() = 1 sinh2 1 2 ,
U
4
see Fig. 1(a), presents two minima at the classical values: V = arcsinh 1. The two
vacuum points are identified by the internal symmetry transformation and the
(x) has been applied
semi-classical vacuum moduli space is a point. For this reason, U
to the study of the quantum theory of diatomic molecules: the solutions of the associated
time-independent Schrdinger equation are a good approximation to the eigen-states of
a quantum particle that moves under the influence of two centers of force. We deal with
the = 1 and M = 3 member of the Razavy family of quasi-exactly-solvable Schrdinger
operators [23], although we are looking at it from a field-theoretical perspective.
(a)
(b)
(c)
Fig. 1. Graphic representation of (a) the potential energy density (b) the kink and (c) the Hessian potential well.
546
The solutions of the first-order equations

1
d
= sinh2 1 ,
dx
2
K (x) = arctanh
tanh(x + b)
,
(18)
see Fig. 1(b) for b = 0, are the kink solitary waves of the system. The Hessian operators at
the vacuum and kink solutions are, respectively:

d2
d2
2
2

V = 2 + 4m = m 2 + 4 = m2 V,
dy
dx
16m2
14m2
d2
2
+
2m
+
dy 2
(1 + sech2 y)2 1 + sech2 y

16
14
d2
2

=m 2 +2+
= m2 K.
dx
(1 + sech2 x)2 1 + sech2 x
K=
is an Schrdinger operator:
The mass of the fundamental mesons is thus 2m. K
2
= d +4V
(x),
K
dx 2
2
2
(x) = 2 sech x(9 + sech x)
V
(1 + sech2 x)2
where the potential well plotted in Fig. 1(c), albeit analytically very different from the sG
and ( 4 )2 kink potential wells, exhibits a similar shape.
The only thing that we need to
We shall not attempt to solve the spectral problem of K.
know in order to apply the asymptotic method is that the lowest eigen-state is the unique
zero mode:
2 2
dK
=
.
f0 (x) =
dx
(3 + cosh 2x)
Therefore, the energy of the semi-classical kink state is approximately (see formula (9))

n
0 1
3

m
1
[n
1,
4]
K

R
.
an (K)
1 3 2 arcsinh 1 hm
+
=
2
2
84n1
(19)
n=2
In Table 3 below we write the Seeleys coefficients and the partial sums
Dn0 =
n
0 1
n=2
[n 1, 4]
an (K)
84n1
up to n0 = 11: obtaining the approximate answer:

6MK
= h m[0 + D11 ]
= 0.282095h m 0.317502h m = 0.599597h m.
We cannot estimate the error but we assume that this result is as good as the answer
also starts at 4.
obtained for the ( 4 )2 kink because the continuous spectrum of K
547
Table 3
n
2
3
4
5
6
7
8
9
10

an (K)
n0 1
Dn0
29.1604
39.8523
42.1618
36.0361
25.7003
15.6633
8.3143
3.9033
1.6590
2
3
4
5
6
7
8
9
10
0.20135
0.26501
0.293253
0.306715
0.313005
0.315779
0.316917
0.317349
0.317502
4. Link kinks: the ( 6 )2 model

Finally, we consider the following potential energy density:

2 2
m2 2
2
.
U y =
4m2
The choice of md = m/ 2 and cd = /m leads to the non-dimensional potential:

(x, t) = 1 2 2 1 2 .
U
2
= M/Z2 , made out of two Z2 orbits, contains two points:
The moduli space of vacua M
V0 (x, t) = 0,
V (x, t) = 1.
Quantization around the V0 (x, t) vacuum preserves the symmetry, which is

spontaneously broken at the degenerate vacua V (x, t). The kink solitary waves of the
system
1
K (x, t) = 1 tanh(x + b),
2

m
m
1 tanh (y + b)
K y, y 0 =
2
2
interpolate between V (x, t) and V0 (x, t), or vice-versa, which are vacua belonging to
distinct Z2 orbits: these solutions are thus link kinks (Fig. 2).
3
The kink and vacuum solutions have classical energies of E[K ] = 14 m and
2
E[V0 ] = E[V ] = 0, respectively. The Hessian operators for the vacuum and kink
solutions are

d2
m2 m2
m2
d2
=
V0 = 2 +
2 +1 =
V0 ,
dy
2
2
dx
2

d2
d2
m2
m2
2 +4 =
V ,
V = 2 + 2m2 =
2
2
dy
dx
548
(a)
(b)
(c)
Fig. 2. Graphical representation of (a) the potential energy density (b) the kink and (c) the Hessian potential well.
my
15m2
d2
5m2 3m2
tanh
+
my
2
4
4
dy
2 8cosh2 2

d2
15
m2
5 3
m2
2 + tanhx
K.
=
=
2
2 2
2
dx
4cosh2 x
K=
The problem of the semi-classical quantization of these and other link kinks have been
addressed somewhat unsuccessfully in [19] due to the analytical complexity of the eigen as well as the conceptual difficulty of dealing with a QFT on the real line
functions of K
where the asymptotic states far on the left and far on the right correspond to mesons with
different masses. This issue has been analyzed in depth in [20]: the main suggestion is that
the normal-order prescription should be performed with an arbitrary mass to be fixed in
order to avoid the ambiguity induced by the step function background. We now apply the
asymptotic expansion of the heat function method in this complex circumstance to find a
very natural way of choosing the mass renormalization parameter. Moreover, we improve
the approximation obtained in the computation of the quantum kink mass by going farther
than first-order in the asymptotic expansion.
Besides the bound state,
f0 (x) =
1
,
2 cosh x 2(1 tanh x)

2
includes transmissionless scattering

the 2 = 0 translational mode, the spectrum of K
states for 1 2 4, and states with both non-zero transmission and reflection coefficients
if 2 4. In the language of QFT, the topological sectors based on link kinks are peculiar
in the sense that the N -particle asymptotic states are mesons that have different masses
at x = . If the meson energy is less than 2m2 , the bosons are reflected when coming
549
from the left/right towards the kink. More energetic mesons can either be reflected by or
pass through the kink. If the mesons are transmitted there is a conversion from kinetic to
inertial energy, or vice-versa, in such a way that the poles
of the propagators far to the
left or far to the right of the kink can only occur at p2 = m2 / 2 and p2 = 2m2 .
This is the reason why the subtraction from the Casimir energy of K , h2 P K (1/2),
of either the Casimir energy of the V0 , h2 V0 (1/2), or the V , h2 V (1/2), vacua
is hopeless, even after adding the mass renormalization counter-term to the Lagrangian.
Therefore, we cannot use the generalized zeta functions V0 (s) and V (s) to renormalize
the zero point energy in the kink sector. Instead, we will gauge the kink Casimir energy
against the Casimir energies of a family of background field configurations that satisfy:
1
5B4 (x) 4B2 (x) = (1 tanh x),
2
(20)
where R+ . The rationale behind this choice is that the limit is the background
used by Lohe, [19]: B (x) = (x). The problem with Lohes choice is that the
discontinuity at the origin poses many problems for the algorithm of the asymptotic
expansion because a nightmare of delta functions and their derivatives appears at x = 0
at orders higher than the first. Thus, we need some regularization, which is achieved by
replacing the sign function by tanh in the formula (20) above. In Figs. 3(a) and (b) the
Hessian potential wells for the backgrounds B and B1 are compared.
For any non-zero finite , B (x) interpolates smoothly between 45 and 1 when x
varies from to . The jump from 1 to 0 occurring at x = 0 in B (x) becomes
a jump from 4/5 to 0, which therefore takes place at x = !, followed by the
smooth interpolation
to 1. If = 0 the background configuration is also pathological:
B0 (x) = 2+ 513/2 , x, except at x = , where there are jumps to 0 and 1.

The Schrdinger operators
2
= d + 5 3 tanh x
B
dx 2 2 2
2
= d + 5 3 (x),
B
dx 2 2 2
(a)
2
0 = d + 5
B
dx 2 2
(b)
Fig. 3. Graphic representation of the potential well produced by (a) the B1 background (b) the B background
as functions of x.
550
govern the small fluctuations around the background B . Thus,

2 s+ 1
2

hm
d
lim
1 () =
P K
(s) B
(s)
2 s 12 md
K
is the Casimir kink energy renormalized with respect to the B background. From the
asymptotic expansion of both P K
(s) and B
(s) we obtain:

2 s+ 1
1 5
2 c (K
2
2
d
1 ) [s + 2 , 2 ]
hm
+
lim
1 K () =
2
(s)
( 12 ) s 12 5m2d
4

n
0 1
) 2 n1 [n 1, 5 ]
c n (K
2
,
+
1
5
(
)
4
2
n=2
) = an (K)
a n (B
). The deviation from the exact result is:
where cn (K

hm
1
1
1
1
d
bn0 ,K
b
+
B
error1 =

n0 ,B
PK
2
2
2
2
4

1
.
BB
2
In order to implement the mass renormalization prescription, we
assume that virtual
mesons runningon the loop of the tadpole graph have a mass of m/ 2 half of the time
and a mass of 2 m the other half-time on average. The normal-order
is thus prescribed
5
for annihilation and creation operators of mesons with M = 2 m mass; this amounts to
considering

1
1
2
B0
m =
2md L
2
as the infinite quantity associated with the single divergent graph of the system. Zeta
function regularization plus the asymptotic expansion tell us that the induced counter-term
adds

2 K () = K |H m2 |K B |H m2 |B
2 s 1
2 [s + 1 , 5 ]
h md
2
2 2

= c1 (K ) lim
2
(s)
2 4
s 21 5md
to the one-loop correction to the link kink mass, whereas the error is

h
)B 1 .
c 1 (K
error2 = lim
B0 2
L 4L
The sum of the contributions coming from the s 12 poles of 1 K () and 2 K ()
vanishes:
2 s+ 1

)
2
[s + 12 , 52 ] [s + 12 , 52 ]
h md c1 (K
2
lim
= 0.
2
(s)
(s)
4 s 21 5m2d
551
Table 4
1 )
cn (K
n0 1
Dn0 (1)
9.3750
10.9375
10.2567
7.89397
5.12392
2.86874
1.40987
0.61636
0.24186
2
3
4
5
6
7
8
9
10
0.0968454
0.0617547
0.0786049
0.0703349
0.0741904
0.0725233
0.0731872
0.0729439
0.0730259
n
2
3
4
5
6
7
8
9
10
The choice of M = 25 m as a mass renormalization parameter leads to exactly the same

result that we encountered in the more conventional systems with loop kinks and we end
with the answer:

6MK
= h m 0 + Dn0 () ,
where 0 =
1
2 2
Dn0 () =
and
n
0 1
n=2
)dn =
c n (K
n
0 1
n=2
n1
[n 1, 52 ]
2

.
cn (K )
5
8 2
The coefficients and the partial sums up to n0 = 11 for = 1 are shown in Table 4. We
find:

6MK1
= h m 0 + D11 (1)
= 0.199471h m + 0.0730259h m = 0.126445h m
as the approximation to the kink Casimir energy measured with respect to the Casimir
energy of the B1 (x) background field configuration.
The choice of = 1 is optimum in the sense that for smaller values of a tendency
of the quantum correction towards is observed whereas for greater than 1 the
tendency is toward +. In Fig. 4, = 1 is identified as the inflexion point of a family
that interpolates between two background configurations with bad features: too abrupt if
= and too smooth if = 0.
We end this section by comparing our renormalization criterion with the prescription
used in [20]. Lohe and OBrien choose a mass renormalization parameter M in such a
way that the mass counter-term exactly cancels the difference in vacuum Casimir energies
between different points in the vacuum moduli space.

1
1
hm
3m
V0
(21)
V
+ Lm 2 = 0.
2
2
2 2
2
The contribution of the tadpole graph must be considered for mesons with a suitable mass
to satisfy (21):

2
1
1
= d +M
2
V
,
V
m 2 =
2
dx 2
2 mL
552
Fig. 4. Quantum correction to the kink mass as a function of in the (0.4, 2.0) interval.
2 = 2.33, M = M
m , a very close value to M. At the L limit
and we find M
2

1
2.33
1
1
B0
V
=
log
.
2
2
4
2.50
If we had used M as the mass renormalization parameter, the result would differ by
1 )
c 1 (K
hm
2.33
K 1 M K 1 M =
log
3/2
2.50
(4)
4 2
which is a very small quantity indeed.
5. Outlook
The natural continuation of this work, and the main motivation to develop the
asymptotic method, is the computation of quantum kink masses in theories with
N -component scalar fields. Nevertheless, explorations in the supersymmetric world along
these lines are also interesting.
All the models that we have described admit a supersymmetric extension because the
potential energy density always can be written as
U () =
1 dW dW
.
2 d d
() for each model is:

In non-dimensional variables the superpotential W

3

W () = 4 cos ,
W () =
,
2
3

3
2 2
1 1

sinh 2 ,
W () =
1 .
W () = 4
2
2 2
2 4
The supersymmetric extension includes also a Majorana spinor field:

1 x
,
x =
= , = 1, 2.
2 x
553
Choosing the Majorana representation 0 = 2 , 1 = i 1 , 5 = 3 of the Clifford

algebra { , } = 2g and defining the Majorana adjoint = t 0 , the action of the
supersymmetric model is:

dW
d 2W
dW
1

.
S = 2 dx 2 + i
d d
d 2
2cd
The N = 1 supersymmetry transformation is generated on the space of classical
configurations by the Hamiltonian spinor function

dW
Q = dx 0 + i 0
.
d
The components of the Majorana spinorial charge Q close the supersymmetry algebra

1
{Q , Q } = 2 0 P 2i
(22)
T.
Their (anti)-Poisson bracket is given in (22) in terms of the momentum P and the
|.
topological central charge T = | d W
5
1
The chiral projections Q = 1
2 Q and =
2 provide a very special
combination of the supersymmetric charges:

d
dW
Q+ + Q = dx (+ )
.
( + + )
dx
d

dW
Q+ + Q is zero for the classical configurations that satisfy d
dx = d and = 0 which
are thus classical BPS states. One immediately notices that our kinks are such BPS states
and besides the small bosonic fluctuations one must take into account the small fermionic
fluctuations around the kink for computing the quantum correction to the kink mass in the
extended system. The fermionic fluctuations around the kink configuration lead to other
solutions of the field equations if the Dirac equation

d 2W
i +
(
)
F (x, t) = 0
K
d 2
is satisfied. We multiply this equation for the adjoint of the Dirac operator

d 2W
d 2W
i +
(K ) i +
(K ) F (x, t) = 0
d 2
d 2
and, due to the time-independence of the kink background, look for solutions of the form:
F (x, t) = fF (x; )eit . This is tantamount to solving the spectral problem

d 2W
d 2W
d 3W
d2
1 dW
(
(
)
(
)
i
)
(
)
fF (x; ) = 2 fF (x; ).
2+
K
K
K
K
dx
d 2
d 2
d
d 3
Projecting onto the eigen-spinors of i 1 ,
fF(1) (x; ) =
1 + i 1
1
fF (x; ) =
2
2
fF+ (x; ) fF (x; )

fF+ (x; ) + fF (x; )
554
we end with the spectral problem:

d2
d 2W
d 2W
dW
d 3W
(
2+
(
)
(
)
)
(
)
fF(1) (x; ) = KfF(1) (x; )
K
K
K
K
dx
d 2
d 2
d
d 3
(1)
= 2 fF (x; )
for the same Schrdinger operator as that governing the bosonic fluctuations.
Therefore, generalized zeta function methods can also be used in supersymmetric
models for computing the quantum corrections to the mass of BPS kinks. Great care
however, is needed in choosing the boundary conditions on the fermionic fluctuations
without spoiling supersymmetry. We look forward to extend this research in this direction.
Acknowledgements
W.G.F. acknowledges the partial financial support to his work provided by Oviedo
University under contract NP-01-514-1.
Appendix A
In this appendix we describe the iterative procedure that gives the coefficients an (x, x)
used in the text. For alternative descriptions, see [21,22]. For an interesting interpretation
of these coefficients as invariants of the Kortewegde Vries equation, see [24].
Starting from formula (8) in the text, we write the recurrence relation
an+1 (x, y)
2 an (x, y)
V (x)an (x, y) =
.
x
x 2
In order to take the limit y x properly, we introduce the notation
(n + 1)an+1 (x, y) + (x y)
(k)
(A.1)
k an (x, y)
x k
An (x) = lim
yx
and, after differentiating (A.1) k times, we find

k j

k V (x) (kj )
1
(k)
(k+2)
An (x) =
An1 (x) +
An1 (x) ,
n+k
x j
j
j =0
from this equation and

(k)
A0 (x) = lim
yx
k a0
= k0 ,
x k
all the (k) An (x) can be generated, recursively. Returning to (A.1), we finally obtain a welldefined recurrence relation

1 (2)
(x)an (x, x)
An (x) + V
an+1 (x, x) =
n+1
555
suitable for our purposes.

We give the explicit expressions of the first eight an (x, x) coefficients. The abbreviated
notation is
k n

dkV
d V
n
uk =
(x),
u
=
(x) :
k
dx k
dx k
1
1
a2 (x, x) = u20 + u2 ,
2
6
1 3 1
1 2
1
u + u2 u0 + u1 + u4 ,
6 0 6
12
60
1 4
1
1
1
1
1
1
u0 + u2 u20 + u21 u0 + u4 u0 + u22 + u1 u3 +
u6 ,
24
12
12
60
40
30
840
1 5
1
1
1
1
1
u + u2 u30 + u21 u20 +
u4 u20 + u22 u0 + u1 u3 u0
120 0 36
24
120
40
30
1
11 2
23 2
19
1
u6 u0 +
u1 u2 +
u3 +
u2 u4 +
u1 u5
+
840
360
5040
2520
280
1
u8 ,
+
15120
1 6
1
1
1
1
1
u0 +
u2 u40 + u21 u30 +
u4 u30 + u22 u20 + u1 u3 u20
720
144
72
360
80
60
11 2
1
1 4
1
61 3
u1 u2 u0 +
u1 u5 u0 +
u1 +
u8 u0 +
u
+
360
280
288
15120
15120 2
43
23
5 2
19
23 2
u1 u2 u3 +
u0 u23 +
u u4 +
u0 u2 u4 +
u
+
2520
5040
1008 1
2520
30240 4
19
1 2
11
1
1
u3 u5 +
u u6 +
u2 u6 +
u1 u7 +
u10 ,
+
15120
1680 0
15120
3780
332640
1 7
1
1 2 4
1 2 3
1
u0 +
u2 u50 +
u1 u0 +
u2 u0 +
u1 u3 u30
5040
720
288
240
180
11 2
1
1 4
61 3
u u2 u20 +
u1 u5 u20 +
u u0 +
u u0
+
720 1
560
288 1
15120 2
43
5 2
1
23 2 2
u1 u2 u3 u0 +
u u4 u0 +
u10 u0 +
u u
+
2520
1008 1
332640
10080 3 0
19
1
83 2 2
1 3
31
u2 u4 u20 +
u6 u30 +
u1 u2 +
u1 u3 +
u2 u23
+
5040
5040
10080
252
10080
1
1 4
5 2
23
1
u1 u3 u4 +
u0 u4 +
u2 u4 +
u0 u24 +
u1 u2 u5
+
280
1440
2016
30240
420
19
71
1 2
11
u0 u3 u5 +
u25 +
u1 u6 +
u0 u2 u6
+
15120
665280
2016
15120
61
1
19
1
u4 u6 +
u0 u1 u7 +
u3 u7 +
u2 u8
+
332640
3780
166320
30240 0
17
1
1
u2 u8 +
u1 u9 +
u12 ,
+
332640
66528
8648640
a1 (x, x) = u0 ,
a3 (x, x) =
a4 (x, x) =
a5 (x, x) =
a6 (x, x) =
a7 (x, x) =
556
a8 (x, x) =
1
1 2 4
1
1 4 2
1 3
u80 +
u2 u0 +
u1 u3 u40 +
u1 u0 +
u u3 u0
40320
960
720
576
252 1
1
1
31
5 2
u1 u3 u4 u0 +
u1 u2 u5 u0 +
u2 u23 u0 +
u u4 u0
+
280
420
10080
2016 2
1 2
1
23 2 2
19
u u6 u0 +
u12 u0 +
u u +
u3 u5 u20
+
2016 1
8648640
60480 4 0 30240
11
1
11 2
1
u2 u6 u20 +
u1 u7 u20 +
u u2 u30 +
u8 u30
+
30240
7560
2160 1
90720
1
1 5 2
1 6
17 4
83
u4 u50 +
u0 u1 +
u0 u2 +
u1 u2 +
u0 u21 u22
+
7200
1440
4320
8640
10080
61 2 3
1261 4
43 2
227
+
u0 u2 +
u2 +
u0 u1 u2 u3 +
u1 u22 u3
30240
1814400
5040
37800
23 3 2
659 2 2
5 2 2
19 3
u u +
u u +
u u u4 +
u u2 u4
+
30240 0 3 302400 1 3 2016 0 1
15120 0
527 2
7939 2
6353
1 3
u u2 u4 +
u u4 +
u2 u24 +
u u1 u5
+
151200 1
9979200 3
9979200
1680 0
17 3
13
3067
71
u1 u5 +
u2 u3 u5 +
u1 u4 u5 +
u0 u25
+
30240
12320
4989300
665280
1
3001 2
13
61
u40 u6 +
u2 u6 +
u1 u3 u6 +
u0 u4 u6
+
20160
9979200
29700
332640
3433
109
19
1501
u2 +
u1 u2 u7 +
u0 u3 u7 +
u5 u7
+
259459200 6 498960
166320
64864800
71
17
2003
1
u2 u8 +
u0 u2 u8 +
u4 u8 +
u0 u1 u9
+
1995840 1
332640
129729600
66528
5
1
73
1
u3 u9 +
u20 u10 +
u2 u10 +
u1 u11
+
648648
665280
25945920
1441440
1
u14 .
+
259459200
References
[1] D. Olive, E. Witten, Phys. Lett. B 78 (1978) 97;
N. Seiberg, E. Witten, Nucl. Phys. B 246 (1994) 19.
[2] M. Duff, R. Khuri, J. Lu, Phys. Rep. 259 (1995) 213.
[3] G. Dvali, M. Shifman, Nucl. Phys. B 504 (1997) 127;
G. Gibbons, P. Townsend, Phys. Rev. Lett. 83 (1999) 172.
[4] R. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 10 (1974) 4130;
R. Dashen, B. Hasslacher, A. Neveu, Phys. Rev. D 12 (1975) 3424.
[5] L.D. Faddeev, V.E. Korepin, Phys. Rep. 42C (1978) 187.
[6] A. DAdda, R. Horsley, P. Di Vecchia, Phys. Lett. B 76 (1978) 298;
J. Schonfeld, Nucl. Phys. B 161 (1979) 125.
[7] A. Rebhan, P. van Nieuwenhuizen, Nucl. Phys. B 508 (1997) 449467;
H. Nastase, M. Stephanov, A. Rebhan, P. van Nieuwenhuizen, Nucl. Phys. B 542 (1999) 471514.
[8] M. Shifman, A. Vainshtein, M. Voloshin, Phys. Rev. D 59 (1999) 45016.
[9] M. Bordag, J. Phys. A 28 (1995) 755;
M. Bordag, K. Kirsten, D. Vassilevich, Phys. Rev. D 59 (1999) 085011;
557
E. Elizalde et al., Zeta Regularization Techniques with Applications, Singapore, World Scientific, 1994.
[10] M. Bordag, A. Goldhaber, P. van Nieuwenhuizen, D. Vassilevich, Heat kernels and zeta-function
regularization for the mass of the SUSY kink, hep-th/0203066.
[11] A. Alonso Izquierdo, M.A. Gonzlez Len, J. Mateos Guilarte, J. Phys. A: Math. Gen. A 31 (1998) 209;
A. Alonso Izquierdo, M.A. Gonzlez Len, J. Mateos Guilarte, Phys. Lett. B 480 (2000) 373;
A. Alonso Izquierdo, M.A. Gonzlez Len, J. Mateos Guilarte, Nonlinearity 13 (2000) 1137.
[12] J. Mateos Guilarte, Lett. Math. Phys. 14 (1987) 169;
J. Mateos Guilarte, Ann. Phys. 188 (1988) 307.
[13] P. Gilkey, Invariance Theory, the Heat Equation and the AtiyahSinger Index Theorem, Publish or Perish,
Delaware, 1984.
[14] S. Coleman, Aspects of Symmetry, Cambridge Univ. Press, Cambridge, 1985, Chapter 6.
[15] P. Drazin, R. Johnson, Solitons: an Introduction, Cambridge Univ. Press, Cambridge, 1996;
P. Morse, H. Feshbach, Methods of Theoretical Physics, McGraw Hill, New York, 1953.
[16] M. Abramowitz, I. Stegun, Handbook of Mathematical Functions with Formulas, Graphs and Mathematical
Tables, Dover, New York, 1992.
[17] J. Liouville, J. Math. Pures Appl. 18 (1853) 71.
[18] L.J. Boya, J. Casahorran, Ann. Phys. 266 (1998) 63.
[19] M.A. Lohe, Phys. Rev. D 20 (1979) 3120.
[20] M.A. Lohe, D.M. OBrien, Phys. Rev. D 23 (1981) 1771.
[21] B.S. de Witt, Dynamical Theory of Groups and Fields, Gordon and Breach, New York, 1965.
[22] M. Stone, Ann. Phys. 155 (1984) 56.
[23] M. Razavy, Am. J. Phys. 48 (1980) 285;
F. Finkel, A. Gonzalez-Lopez, M.A. Rodriguez, J. Phys. A: Math. Gen. 32 (1999) 6821.
[24] A.M. Perelomov, Y.B. Zeldovich, Quantum Mechanics: Selected Topics, World Scientific, Singapore, 1998.

CUMULATIVE AUTHOR INDEX B631B635
Ahn, Ch.
Akemann, G.
Alberghi, G.L.
Aliani, P.
Alonso Izquierdo, A.
Alvarez, O.
Antonelli, V.
Antoniadis, I.
Antoniadis, I.
Aoki, H.
Apreda, R.
Argeri, M.
Armoni, A.
Arrizabalaga, A.
Arutyunov, G.
B634 (2002) 141

B631 (2002) 471
B635 (2002) 57
B634 (2002) 393
B635 (2002) 525
B633 (2002) 309
B634 (2002) 393
B631 (2002) 3
B631 (2002) 66
B634 (2002) 71
B631 (2002) 342
B631 (2002) 388
B632 (2002) 240
B635 (2002) 255
B635 (2002) 3
Bantay, P.
Barnich, G.
Barto, E.
Basu-Mallick, B.
Becchi, C.
Becirevic, D.
Behrndt, K.
Benakli, K.
Berkovits, N.
Bertolini, M.
Bhattacharyya, T.
Bianchi, M.
Blum, J.D.
Boyanovsky, D.
Braden, H.W.
Brandt, F.
Brecher, D.
Brignole, A.
Budczies, J.
Bueno, A.
Buras, A.J.
Burkardt, M.
B633 (2002) 365

B633 (2002) 3
B632 (2002) 330
B634 (2002) 611
B633 (2002) 250
B634 (2002) 105
B635 (2002) 158
B631 (2002) 3
B635 (2002) 75
B632 (2002) 257
B634 (2002) 611
B631 (2002) 159
B634 (2002) 3
B632 (2002) 121
B633 (2002) 414
B633 (2002) 3
B634 (2002) 23
B631 (2002) 195
B635 (2002) 309
B631 (2002) 239
B631 (2002) 219
B632 (2002) 311
Cabrera-Carnero, I.
Cacciapaglia, G.
B634 (2002) 433

B634 (2002) 230
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 9 2 - 3
Cadoni, M.
Caffo, M.
Campanelli, M.
Cao, F.J.
Caravaglios, F.
Carta, P.
Chetyrkin, K.G.
Chu, C.-S.
Ciocarlie, C.
Cirelli, M.
Ciuchini, M.
Clark, T.E.
Corley, S.
Cristadoro, G.
Czarnecki, A.
Czyz, H.
B632 (2002) 383

B634 (2002) 309
B631 (2002) 239
B632 (2002) 121
B633 (2002) 193
B632 (2002) 383
B634 (2002) 413
B632 (2002) 219
B632 (2002) 303
B634 (2002) 230
B634 (2002) 105
B632 (2002) 3
B635 (2002) 57
B634 (2002) 230
B631 (2002) 219
B634 (2002) 309
Damgaard, P.H.
Degrassi, G.
Delduc, F.
Deser, S.
De Vega, H.J.
Di Vecchia, P.
Doikou, A.
Dolgov, A.D.
Dotsenko, V.S.
Dubnicka, S.
Dubnickov, A.-Z.
Dukelsky, J.
B633 (2002) 97
B631 (2002) 195
B631 (2002) 403
B631 (2002) 369
B632 (2002) 121
B631 (2002) 95
B634 (2002) 591
B632 (2002) 363
B631 (2002) 426
B632 (2002) 330
B632 (2002) 330
B634 (2002) 483
Enger, H.
Erkol, G.
B631 (2002) 95
B635 (2002) 286
Fateev, V.A.
Fjelstad, J.
Flohr, M.
Font, A.
Franco, E.
B634 (2002) 546

B633 (2002) 379
B634 (2002) 511
B634 (2002) 51
B633 (2002) 212
560
Franco, E.
Freedman, D.Z.
Freitas, A.
Friedmann, T.
Froggatt, C.D.
Frolov, S.
Fuchs, J.
Fukuda, T.
B634 (2002) 105

B631 (2002) 159
B632 (2002) 189
B635 (2002) 384
B631 (2002) 285
B632 (2002) 69
B633 (2002) 379
B635 (2002) 215
Garca Fuertes, W.
Gavai, R.V.
Gersdorff, G.V.
Giedt, J.
Gimnez, V.
Giusto, S.
Gomes, J.F.
Gomis, J.
Gonzlez Len, M.A.
Gorsky, A.
Gracey, J.A.
B635 (2002) 525

B633 (2002) 127
B634 (2002) 90
B632 (2002) 397
B634 (2002) 105
B633 (2002) 250
B634 (2002) 433
B635 (2002) 106
B635 (2002) 525
B633 (2002) 414
B634 (2002) 192
Hambye, T.
Hansen, S.H.
Harlander, R.V.
Harmark, T.
Hartnoll, S.A.
Hatsuda, M.
Hebecker, A.
Heller, U.M.
Hernndez, A.
Hollik, W.
Hosomichi, K.
Howe, P.
Hwang, S.
B633 (2002) 171

B632 (2002) 363
B634 (2002) 413
B632 (2002) 257
B631 (2002) 325
B632 (2002) 114
B632 (2002) 101
B633 (2002) 97
B634 (2002) 51
B632 (2002) 189
B635 (2002) 215
B635 (2002) 75
B633 (2002) 379
Imbimbo, C.
Imeroni, E.
Irges, N.
Iso, S.
Ivanov, R.
Ivanov, R.
B633 (2002) 250

B631 (2002) 95
B635 (2002) 127
B634 (2002) 71
B635 (2002) 435
B635 (2002) 473
Jacobsen, J.L.
Janik, R.A.
Janssen, B.
B631 (2002) 426

B635 (2002) 492
B634 (2002) 23
Kamimura, K.
Khlebnikov, S.
Khuri, R.R.
Kleinert, H.
Kniehl, B.A.
Koike, Y.
Krauss, F.
Khn, J.H.
Kuraev, E.A.
B632 (2002) 114

B631 (2002) 307
B633 (2002) 295
B632 (2002) 51
B635 (2002) 357
B632 (2002) 311
B633 (2002) 237
B634 (2002) 413
B632 (2002) 330
Laugier, A.
Lavignac, S.
Lee, P.
Lee, P.
Li, T.
Liao, Y.
Litim, D.F.
Lopez, E.
Louis, J.
Love, S.T.
Lowe, D.A.
Lozano, Y.
Lozano-Tellechea, E.
L, H.
Lubicz, V.
Lukierski, J.
B631 (2002) 3
B633 (2002) 139
B632 (2002) 283
B632 (2002) 303
B633 (2002) 83
B635 (2002) 505
B631 (2002) 128
B632 (2002) 240
B635 (2002) 395
B632 (2002) 3
B635 (2002) 57
B634 (2002) 23
B631 (2002) 95
B633 (2002) 114
B633 (2002) 212
B632 (2002) 219
Maggiore, M.
Malbouisson, A.P.C.
Malbouisson, J.M.C.
Mangano, M.L.
Martinelli, G.
Masiero, A.
Masina, I.
Mastrolia, P.
Mateos Guilarte, J.
Mescia, F.
Micu, A.
Mignemi, S.
Minasian, R.
Misiak, M.
Montani, G.
Morariu, B.
Moretti, M.
Moss, I.G.
Moss, I.G.
Mller-Kirsten, H.J.W.
B631 (2002) 342

B631 (2002) 83
B631 (2002) 83
B632 (2002) 343
B634 (2002) 105
B634 (2002) 105
B633 (2002) 139
B631 (2002) 388
B635 (2002) 525
B633 (2002) 212
B635 (2002) 395
B632 (2002) 383
B631 (2002) 43
B631 (2002) 219
B634 (2002) 370
B634 (2002) 326
B632 (2002) 343
B631 (2002) 500
B632 (2002) 173
B635 (2002) 192
Nauta, B.-J.
Navas-Concha, S.
Navelet, H.
Naylor, W.
Nepomechie, R.I.
Ng, S.
Nguyen, X.S.
Niclasen, R.
Nicolis, A.
Nielsen, H.B.
Nonnenmacher, S.
Nowling, S.R.
B635 (2002) 255

B631 (2002) 239
B634 (2002) 291
B632 (2002) 173
B631 (2002) 519
B634 (2002) 209
B631 (2002) 426
B633 (2002) 97
B631 (2002) 342
B631 (2002) 285
B635 (2002) 309
B632 (2002) 3
Obers, N.A.
Odesskii, A.
B632 (2002) 257

B633 (2002) 414
561
Singer, I.M.
Skenderis, K.
Slavich, P.
Smirnov, V.A.
Smith, J.
Soff, G.
Sokatchev, E.
Sorin, A.S.
Sotkov, G.M.
Steinhauser, M.
Stocchi, A.
Sturani, R.
Sundell, P.
Suyama, T.
Svetitsky, B.
B633 (2002) 309

B631 (2002) 159
B631 (2002) 195
B635 (2002) 357
B634 (2002) 247
B633 (2002) 237
B635 (2002) 3
B631 (2002) 403
B634 (2002) 433
B635 (2002) 357
B633 (2002) 193
B631 (2002) 66
B634 (2002) 120
B634 (2002) 71
B633 (2002) 97
Takanishi, Y.
Tamaryan, S.
Tarantino, C.
Tipunin, I.Yu.
Tomasiello, A.
Torrente-Lujan, E.
Trapletti, M.
Tseytlin, A.A.
Tuite, M.
Tuite, M.
Turan, G.
B631 (2002) 285

B635 (2002) 192
B633 (2002) 212
B633 (2002) 379
B631 (2002) 43
B634 (2002) 393
B635 (2002) 33
B632 (2002) 69
B635 (2002) 435
B635 (2002) 473
B635 (2002) 286
Urban, J.
B631 (2002) 219
Van den Bossche, B.

Van Neerven, W.L.
Vzquez-Poritz, J.F.
Verhoeven, O.
Vernizzi, G.
Von Gersdorff, G.
B632 (2002) 51
B634 (2002) 247
B633 (2002) 114
B633 (2002) 345
B631 (2002) 471
B635 (2002) 127
Waldron, A.
Walter, W.
Weiglein, G.
Westerberg, A.
Wettig, T.
Woo, K.
B631 (2002) 369

B632 (2002) 189
B632 (2002) 189
B632 (2002) 257
B632 (2002) 155
B634 (2002) 141
Yue, R.-H.
B634 (2002) 571
Zakrzewski, W.J.
Zemlyanaya, E.
Zimerman, A.H.
Zinn-Justin, P.
Zirnbauer, M.R.
Zwirner, F.
B632 (2002) 219

B632 (2002) 330
B634 (2002) 433
B634 (2002) 417
B635 (2002) 309
B631 (2002) 195
Onofri, E.
Ooguri, H.
Ooguri, H.
B634 (2002) 546

B632 (2002) 283
B635 (2002) 106
Papinutto, M.
Park, D.K.
Park, J.
Park, J.
Park, M.-I.
Pastor, S.
Pearce, P.A.
Penin, A.A.
Perry, M.
Peschanski, R.
Petcov, S.T.
Picariello, M.
Pittau, R.
Pokotilov, A.
Polychronakos, A.P.
B634 (2002) 105

B635 (2002) 192
B632 (2002) 283
B632 (2002) 303
B634 (2002) 339
B632 (2002) 363
B631 (2002) 447
B635 (2002) 357
B634 (2002) 209
B634 (2002) 291
B632 (2002) 363
B634 (2002) 393
B632 (2002) 343
B633 (2002) 295
B634 (2002) 326
Quirs, M.
Quirs, M.
B634 (2002) 90
B635 (2002) 127
Raffelt, G.G.
Ravindran, V.
Remiddi, E.
Remiddi, E.
Reyes, J.
Richard, C.
Riotto, A.
Riotto, A.
Romn, J.M.
Roudeau, P.
Rubbia, A.
Rubtsov, V.
Ruelle, P.
B632 (2002) 363

B634 (2002) 247
B631 (2002) 388
B634 (2002) 309
B634 (2002) 105
B631 (2002) 447
B631 (2002) 342
B634 (2002) 90
B634 (2002) 483
B633 (2002) 193
B631 (2002) 239
B633 (2002) 414
B633 (2002) 345
Sakaguchi, M.
Santachiara, R.
Santana, A.E.
Savoy, C.A.
Schlittgen, B.
Schlottmann, P.
Scrucca, C.A.
Semikhatov, A.M.
Semikoz, D.V.
Serone, M.
Sezgin, E.
Shnir, Ya.
Shore, G.M.
Sierra, G.
Silvestrini, L.
Simn, J.
B632 (2002) 114

B631 (2002) 426
B631 (2002) 83
B633 (2002) 139
B632 (2002) 155
B634 (2002) 571
B635 (2002) 33
B633 (2002) 379
B632 (2002) 363
B635 (2002) 33
B634 (2002) 120
B635 (2002) 309
B633 (2002) 271
B634 (2002) 483
B634 (2002) 105
B635 (2002) 175

Nucl - Phys.B v.635 PDF

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Nucl - Phys.B v.635 PDF

Enviado por

Direitos autorais:

Formatos disponíveis

Nuclear Physics B 635 (2002) 332

Implications of superconformal symmetry for

D-14476 Golm, Germany

F-74941 Annecy-le-Vieux et lUniversit de Savoie.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

2. Overview and summary of the results

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

d = 4 and J1 = J3 = 0, J2 = s for d = 6. Below we list the relevant UIRs:4

[020] [020] = [040]0 + [121]0 + [202]0 + [020]1 + [101]1

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

corresponding supergravity theory dual to any of these currents. In Section 4, by using

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

Here s, t are the conformal cross-ratios

The six tensor structures I1 I2 I3 I4 , C I1 I2 I3 I4 (and permutations) are invariant tensors of

a2 (s, t) = a2 (t, s) = a3 (s/t, 1/t),

b1 (s, t) = b3 (t, s) = b1 (s/t, 1/t),

b2 (s, t) = b2 (t, s) = b3 (s/t, 1/t).

5 We recall some details in Appendix A.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

be set to its free-theory value by evoking a dynamical mechanism. It consists in employing

b2 (s, t) = B + (1 s t)F (s, t),

where A and B are constants determined from the free N = 4 theory.

where A and B are additive integration constants, is a second-order differential operator,

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

1 2 3 4 (s, t) are defined in Appendix B.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

Section 4 we discuss however that the existence of a smooth superconformal deformation

3. General structure of the four-point amplitude

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

to imposing irreducibility under the R symmetry group. It is precisely this requirement

a3 (t, s) = a1 (s, t),

the second equation both in (7) and in (8) is automatically satisfied.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

and perform one more change of variables

In these new variables the symmetry conditions (22) become

and Eq. (19) takes the form

This equation can be integrated to give

where c( ) is an integration constant depending on . Let us introduce the function

It is even, F (, ) = F (, ) and is supposed to obey the boundary condition

where the prepotential F (s, t) is an arbitrary symmetric function.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

and h1 and h2 are arbitrary constants.

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

Formulae (27), (28) and (32) can be further simplified to give

s/t,1/t t 2 w(s, t) = t 4 s,t w(s, t)

4. Four-point amplitude from gauged supergravity

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

I are traceless symmetric and antisymmetric matrices providing bases

where k11 is the eleven-dimensional Newton constant:

obeying an on-shell conservation law. The tensors C I1 In were already introduced in

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

OI1 (x1 ) OI4 (x4 )

= 12 (C1234 C2134). We exhibit explicitly only the expression in the s-channel,

x14x23 x34 A1234 + A+

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

(ss )2 (s, t),

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

Therefore, comparing with (49) we obtain the following relation

G. Arutyunov, E. Sokatchev / Nuclear Physics B 635 (2002) 332

5. Operator product expansion

is the stress-tensor of 5 free scalars normalized as T s T s = 66 . The brackets [ ]

schematically written as Cabc